1 /* $NetBSD: sysv_sem.c,v 1.94 2015/05/13 01:16:15 pgoyette 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 1.94 2015/05/13 01:16:15 pgoyette Exp $"); 43 44 #ifdef _KERNEL_OPT 45 #include "opt_sysv.h" 46 #endif 47 48 #include <sys/param.h> 49 #include <sys/kernel.h> 50 #include <sys/sem.h> 51 #include <sys/sysctl.h> 52 #include <sys/kmem.h> 53 #include <sys/mount.h> /* XXX for <sys/syscallargs.h> */ 54 #include <sys/syscallargs.h> 55 #include <sys/kauth.h> 56 57 /* 58 * Memory areas: 59 * 1st: Pool of semaphore identifiers 60 * 2nd: Semaphores 61 * 3rd: Conditional variables 62 * 4th: Undo structures 63 */ 64 struct semid_ds * sema __read_mostly; 65 static struct __sem * sem __read_mostly; 66 static kcondvar_t * semcv __read_mostly; 67 static int * semu __read_mostly; 68 69 static kmutex_t semlock __cacheline_aligned; 70 static bool sem_realloc_state __read_mostly; 71 static kcondvar_t sem_realloc_cv; 72 73 /* 74 * List of active undo structures, total number of semaphores, 75 * and total number of semop waiters. 76 */ 77 static struct sem_undo *semu_list __read_mostly; 78 static u_int semtot __cacheline_aligned; 79 static u_int sem_waiters __cacheline_aligned; 80 81 /* Macro to find a particular sem_undo vector */ 82 #define SEMU(s, ix) ((struct sem_undo *)(((long)s) + ix * seminfo.semusz)) 83 84 #ifdef SEM_DEBUG 85 #define SEM_PRINTF(a) printf a 86 #else 87 #define SEM_PRINTF(a) 88 #endif 89 90 void *hook; /* cookie from exithook_establish() */ 91 92 extern int kern_has_sysvsem; 93 94 struct sem_undo *semu_alloc(struct proc *); 95 int semundo_adjust(struct proc *, struct sem_undo **, int, int, int); 96 void semundo_clear(int, int); 97 98 void 99 seminit(void) 100 { 101 int i, sz; 102 vaddr_t v; 103 104 mutex_init(&semlock, MUTEX_DEFAULT, IPL_NONE); 105 cv_init(&sem_realloc_cv, "semrealc"); 106 sem_realloc_state = false; 107 semtot = 0; 108 sem_waiters = 0; 109 110 /* Allocate the wired memory for our structures */ 111 sz = ALIGN(seminfo.semmni * sizeof(struct semid_ds)) + 112 ALIGN(seminfo.semmns * sizeof(struct __sem)) + 113 ALIGN(seminfo.semmni * sizeof(kcondvar_t)) + 114 ALIGN(seminfo.semmnu * seminfo.semusz); 115 sz = round_page(sz); 116 v = uvm_km_alloc(kernel_map, sz, 0, UVM_KMF_WIRED|UVM_KMF_ZERO); 117 if (v == 0) 118 panic("sysv_sem: cannot allocate memory"); 119 sema = (void *)v; 120 sem = (void *)((uintptr_t)sema + 121 ALIGN(seminfo.semmni * sizeof(struct semid_ds))); 122 semcv = (void *)((uintptr_t)sem + 123 ALIGN(seminfo.semmns * sizeof(struct __sem))); 124 semu = (void *)((uintptr_t)semcv + 125 ALIGN(seminfo.semmni * sizeof(kcondvar_t))); 126 127 for (i = 0; i < seminfo.semmni; i++) { 128 sema[i]._sem_base = 0; 129 sema[i].sem_perm.mode = 0; 130 cv_init(&semcv[i], "semwait"); 131 } 132 for (i = 0; i < seminfo.semmnu; i++) { 133 struct sem_undo *suptr = SEMU(semu, i); 134 suptr->un_proc = NULL; 135 } 136 semu_list = NULL; 137 hook = exithook_establish(semexit, NULL); 138 139 kern_has_sysvsem = 1; 140 141 kern_has_sysvsem = 1; 142 143 sysvipcinit(); 144 } 145 146 int 147 semfini(void) 148 { 149 int i, sz; 150 vaddr_t v = (vaddr_t)sema; 151 152 /* Don't allow module unload if we're busy */ 153 mutex_enter(&semlock); 154 if (semtot) { 155 mutex_exit(&semlock); 156 return 1; 157 } 158 159 /* Remove the exit hook */ 160 exithook_disestablish(hook); 161 162 /* Destroy all our condvars */ 163 for (i = 0; i < seminfo.semmni; i++) { 164 cv_destroy(&semcv[i]); 165 } 166 167 /* Free the wired memory that we allocated */ 168 sz = ALIGN(seminfo.semmni * sizeof(struct semid_ds)) + 169 ALIGN(seminfo.semmns * sizeof(struct __sem)) + 170 ALIGN(seminfo.semmni * sizeof(kcondvar_t)) + 171 ALIGN(seminfo.semmnu * seminfo.semusz); 172 sz = round_page(sz); 173 uvm_km_free(kernel_map, v, sz, UVM_KMF_WIRED); 174 175 /* Destroy the last cv and mutex */ 176 cv_destroy(&sem_realloc_cv); 177 mutex_exit(&semlock); 178 mutex_destroy(&semlock); 179 180 kern_has_sysvsem = 0; 181 182 return 0; 183 } 184 185 static int 186 semrealloc(int newsemmni, int newsemmns, int newsemmnu) 187 { 188 struct semid_ds *new_sema, *old_sema; 189 struct __sem *new_sem; 190 struct sem_undo *new_semu_list, *suptr, *nsuptr; 191 int *new_semu; 192 kcondvar_t *new_semcv; 193 vaddr_t v; 194 int i, j, lsemid, nmnus, sz; 195 196 if (newsemmni < 1 || newsemmns < 1 || newsemmnu < 1) 197 return EINVAL; 198 199 /* Allocate the wired memory for our structures */ 200 sz = ALIGN(newsemmni * sizeof(struct semid_ds)) + 201 ALIGN(newsemmns * sizeof(struct __sem)) + 202 ALIGN(newsemmni * sizeof(kcondvar_t)) + 203 ALIGN(newsemmnu * seminfo.semusz); 204 sz = round_page(sz); 205 v = uvm_km_alloc(kernel_map, sz, 0, UVM_KMF_WIRED|UVM_KMF_ZERO); 206 if (v == 0) 207 return ENOMEM; 208 209 mutex_enter(&semlock); 210 if (sem_realloc_state) { 211 mutex_exit(&semlock); 212 uvm_km_free(kernel_map, v, sz, UVM_KMF_WIRED); 213 return EBUSY; 214 } 215 sem_realloc_state = true; 216 if (sem_waiters) { 217 /* 218 * Mark reallocation state, wake-up all waiters, 219 * and wait while they will all exit. 220 */ 221 for (i = 0; i < seminfo.semmni; i++) 222 cv_broadcast(&semcv[i]); 223 while (sem_waiters) 224 cv_wait(&sem_realloc_cv, &semlock); 225 } 226 old_sema = sema; 227 228 /* Get the number of last slot */ 229 lsemid = 0; 230 for (i = 0; i < seminfo.semmni; i++) 231 if (sema[i].sem_perm.mode & SEM_ALLOC) 232 lsemid = i; 233 234 /* Get the number of currently used undo structures */ 235 nmnus = 0; 236 for (i = 0; i < seminfo.semmnu; i++) { 237 suptr = SEMU(semu, i); 238 if (suptr->un_proc == NULL) 239 continue; 240 nmnus++; 241 } 242 243 /* We cannot reallocate less memory than we use */ 244 if (lsemid >= newsemmni || semtot > newsemmns || nmnus > newsemmnu) { 245 mutex_exit(&semlock); 246 uvm_km_free(kernel_map, v, sz, UVM_KMF_WIRED); 247 return EBUSY; 248 } 249 250 new_sema = (void *)v; 251 new_sem = (void *)((uintptr_t)new_sema + 252 ALIGN(newsemmni * sizeof(struct semid_ds))); 253 new_semcv = (void *)((uintptr_t)new_sem + 254 ALIGN(newsemmns * sizeof(struct __sem))); 255 new_semu = (void *)((uintptr_t)new_semcv + 256 ALIGN(newsemmni * sizeof(kcondvar_t))); 257 258 /* Initialize all semaphore identifiers and condvars */ 259 for (i = 0; i < newsemmni; i++) { 260 new_sema[i]._sem_base = 0; 261 new_sema[i].sem_perm.mode = 0; 262 cv_init(&new_semcv[i], "semwait"); 263 } 264 for (i = 0; i < newsemmnu; i++) { 265 nsuptr = SEMU(new_semu, i); 266 nsuptr->un_proc = NULL; 267 } 268 269 /* 270 * Copy all identifiers, semaphores and list of the 271 * undo structures to the new memory allocation. 272 */ 273 j = 0; 274 for (i = 0; i <= lsemid; i++) { 275 if ((sema[i].sem_perm.mode & SEM_ALLOC) == 0) 276 continue; 277 memcpy(&new_sema[i], &sema[i], sizeof(struct semid_ds)); 278 new_sema[i]._sem_base = &new_sem[j]; 279 memcpy(new_sema[i]._sem_base, sema[i]._sem_base, 280 (sizeof(struct __sem) * sema[i].sem_nsems)); 281 j += sema[i].sem_nsems; 282 } 283 KASSERT(j == semtot); 284 285 j = 0; 286 new_semu_list = NULL; 287 for (suptr = semu_list; suptr != NULL; suptr = suptr->un_next) { 288 KASSERT(j < newsemmnu); 289 nsuptr = SEMU(new_semu, j); 290 memcpy(nsuptr, suptr, SEMUSZ); 291 nsuptr->un_next = new_semu_list; 292 new_semu_list = nsuptr; 293 j++; 294 } 295 296 for (i = 0; i < seminfo.semmni; i++) { 297 KASSERT(cv_has_waiters(&semcv[i]) == false); 298 cv_destroy(&semcv[i]); 299 } 300 301 sz = ALIGN(seminfo.semmni * sizeof(struct semid_ds)) + 302 ALIGN(seminfo.semmns * sizeof(struct __sem)) + 303 ALIGN(seminfo.semmni * sizeof(kcondvar_t)) + 304 ALIGN(seminfo.semmnu * seminfo.semusz); 305 sz = round_page(sz); 306 307 /* Set the pointers and update the new values */ 308 sema = new_sema; 309 sem = new_sem; 310 semcv = new_semcv; 311 semu = new_semu; 312 semu_list = new_semu_list; 313 314 seminfo.semmni = newsemmni; 315 seminfo.semmns = newsemmns; 316 seminfo.semmnu = newsemmnu; 317 318 /* Reallocation completed - notify all waiters, if any */ 319 sem_realloc_state = false; 320 cv_broadcast(&sem_realloc_cv); 321 mutex_exit(&semlock); 322 323 uvm_km_free(kernel_map, (vaddr_t)old_sema, sz, UVM_KMF_WIRED); 324 return 0; 325 } 326 327 /* 328 * Placebo. 329 */ 330 331 int 332 sys_semconfig(struct lwp *l, const struct sys_semconfig_args *uap, register_t *retval) 333 { 334 335 *retval = 0; 336 return 0; 337 } 338 339 /* 340 * Allocate a new sem_undo structure for a process. 341 * => Returns NULL on failure. 342 */ 343 struct sem_undo * 344 semu_alloc(struct proc *p) 345 { 346 struct sem_undo *suptr, **supptr; 347 bool attempted = false; 348 int i; 349 350 KASSERT(mutex_owned(&semlock)); 351 again: 352 /* Look for a free structure. */ 353 for (i = 0; i < seminfo.semmnu; i++) { 354 suptr = SEMU(semu, i); 355 if (suptr->un_proc == NULL) { 356 /* Found. Fill it in and return. */ 357 suptr->un_next = semu_list; 358 semu_list = suptr; 359 suptr->un_cnt = 0; 360 suptr->un_proc = p; 361 return suptr; 362 } 363 } 364 365 /* Not found. Attempt to free some structures. */ 366 if (!attempted) { 367 bool freed = false; 368 369 attempted = true; 370 supptr = &semu_list; 371 while ((suptr = *supptr) != NULL) { 372 if (suptr->un_cnt == 0) { 373 suptr->un_proc = NULL; 374 *supptr = suptr->un_next; 375 freed = true; 376 } else { 377 supptr = &suptr->un_next; 378 } 379 } 380 if (freed) { 381 goto again; 382 } 383 } 384 return NULL; 385 } 386 387 /* 388 * Adjust a particular entry for a particular proc 389 */ 390 391 int 392 semundo_adjust(struct proc *p, struct sem_undo **supptr, int semid, int semnum, 393 int adjval) 394 { 395 struct sem_undo *suptr; 396 struct sem_undo_entry *sunptr; 397 int i; 398 399 KASSERT(mutex_owned(&semlock)); 400 401 /* 402 * Look for and remember the sem_undo if the caller doesn't 403 * provide it 404 */ 405 406 suptr = *supptr; 407 if (suptr == NULL) { 408 for (suptr = semu_list; suptr != NULL; suptr = suptr->un_next) 409 if (suptr->un_proc == p) 410 break; 411 412 if (suptr == NULL) { 413 suptr = semu_alloc(p); 414 if (suptr == NULL) 415 return (ENOSPC); 416 } 417 *supptr = suptr; 418 } 419 420 /* 421 * Look for the requested entry and adjust it (delete if 422 * adjval becomes 0). 423 */ 424 sunptr = &suptr->un_ent[0]; 425 for (i = 0; i < suptr->un_cnt; i++, sunptr++) { 426 if (sunptr->un_id != semid || sunptr->un_num != semnum) 427 continue; 428 sunptr->un_adjval += adjval; 429 if (sunptr->un_adjval == 0) { 430 suptr->un_cnt--; 431 if (i < suptr->un_cnt) 432 suptr->un_ent[i] = 433 suptr->un_ent[suptr->un_cnt]; 434 } 435 return (0); 436 } 437 438 /* Didn't find the right entry - create it */ 439 if (suptr->un_cnt == SEMUME) 440 return (EINVAL); 441 442 sunptr = &suptr->un_ent[suptr->un_cnt]; 443 suptr->un_cnt++; 444 sunptr->un_adjval = adjval; 445 sunptr->un_id = semid; 446 sunptr->un_num = semnum; 447 return (0); 448 } 449 450 void 451 semundo_clear(int semid, int semnum) 452 { 453 struct sem_undo *suptr; 454 struct sem_undo_entry *sunptr, *sunend; 455 456 KASSERT(mutex_owned(&semlock)); 457 458 for (suptr = semu_list; suptr != NULL; suptr = suptr->un_next) 459 for (sunptr = &suptr->un_ent[0], 460 sunend = sunptr + suptr->un_cnt; sunptr < sunend;) { 461 if (sunptr->un_id == semid) { 462 if (semnum == -1 || sunptr->un_num == semnum) { 463 suptr->un_cnt--; 464 sunend--; 465 if (sunptr != sunend) 466 *sunptr = *sunend; 467 if (semnum != -1) 468 break; 469 else 470 continue; 471 } 472 } 473 sunptr++; 474 } 475 } 476 477 int 478 sys_____semctl50(struct lwp *l, const struct sys_____semctl50_args *uap, 479 register_t *retval) 480 { 481 /* { 482 syscallarg(int) semid; 483 syscallarg(int) semnum; 484 syscallarg(int) cmd; 485 syscallarg(union __semun *) arg; 486 } */ 487 struct semid_ds sembuf; 488 int cmd, error; 489 void *pass_arg; 490 union __semun karg; 491 492 cmd = SCARG(uap, cmd); 493 494 pass_arg = get_semctl_arg(cmd, &sembuf, &karg); 495 496 if (pass_arg) { 497 error = copyin(SCARG(uap, arg), &karg, sizeof(karg)); 498 if (error) 499 return error; 500 if (cmd == IPC_SET) { 501 error = copyin(karg.buf, &sembuf, sizeof(sembuf)); 502 if (error) 503 return (error); 504 } 505 } 506 507 error = semctl1(l, SCARG(uap, semid), SCARG(uap, semnum), cmd, 508 pass_arg, retval); 509 510 if (error == 0 && cmd == IPC_STAT) 511 error = copyout(&sembuf, karg.buf, sizeof(sembuf)); 512 513 return (error); 514 } 515 516 int 517 semctl1(struct lwp *l, int semid, int semnum, int cmd, void *v, 518 register_t *retval) 519 { 520 kauth_cred_t cred = l->l_cred; 521 union __semun *arg = v; 522 struct semid_ds *sembuf = v, *semaptr; 523 int i, error, ix; 524 525 SEM_PRINTF(("call to semctl(%d, %d, %d, %p)\n", 526 semid, semnum, cmd, v)); 527 528 mutex_enter(&semlock); 529 530 ix = IPCID_TO_IX(semid); 531 if (ix < 0 || ix >= seminfo.semmni) { 532 mutex_exit(&semlock); 533 return (EINVAL); 534 } 535 536 semaptr = &sema[ix]; 537 if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0 || 538 semaptr->sem_perm._seq != IPCID_TO_SEQ(semid)) { 539 mutex_exit(&semlock); 540 return (EINVAL); 541 } 542 543 switch (cmd) { 544 case IPC_RMID: 545 if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_M)) != 0) 546 break; 547 semaptr->sem_perm.cuid = kauth_cred_geteuid(cred); 548 semaptr->sem_perm.uid = kauth_cred_geteuid(cred); 549 semtot -= semaptr->sem_nsems; 550 for (i = semaptr->_sem_base - sem; i < semtot; i++) 551 sem[i] = sem[i + semaptr->sem_nsems]; 552 for (i = 0; i < seminfo.semmni; i++) { 553 if ((sema[i].sem_perm.mode & SEM_ALLOC) && 554 sema[i]._sem_base > semaptr->_sem_base) 555 sema[i]._sem_base -= semaptr->sem_nsems; 556 } 557 semaptr->sem_perm.mode = 0; 558 semundo_clear(ix, -1); 559 cv_broadcast(&semcv[ix]); 560 break; 561 562 case IPC_SET: 563 if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_M))) 564 break; 565 KASSERT(sembuf != NULL); 566 semaptr->sem_perm.uid = sembuf->sem_perm.uid; 567 semaptr->sem_perm.gid = sembuf->sem_perm.gid; 568 semaptr->sem_perm.mode = (semaptr->sem_perm.mode & ~0777) | 569 (sembuf->sem_perm.mode & 0777); 570 semaptr->sem_ctime = time_second; 571 break; 572 573 case IPC_STAT: 574 if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_R))) 575 break; 576 KASSERT(sembuf != NULL); 577 memcpy(sembuf, semaptr, sizeof(struct semid_ds)); 578 sembuf->sem_perm.mode &= 0777; 579 break; 580 581 case GETNCNT: 582 if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_R))) 583 break; 584 if (semnum < 0 || semnum >= semaptr->sem_nsems) { 585 error = EINVAL; 586 break; 587 } 588 *retval = semaptr->_sem_base[semnum].semncnt; 589 break; 590 591 case GETPID: 592 if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_R))) 593 break; 594 if (semnum < 0 || semnum >= semaptr->sem_nsems) { 595 error = EINVAL; 596 break; 597 } 598 *retval = semaptr->_sem_base[semnum].sempid; 599 break; 600 601 case GETVAL: 602 if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_R))) 603 break; 604 if (semnum < 0 || semnum >= semaptr->sem_nsems) { 605 error = EINVAL; 606 break; 607 } 608 *retval = semaptr->_sem_base[semnum].semval; 609 break; 610 611 case GETALL: 612 if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_R))) 613 break; 614 KASSERT(arg != NULL); 615 for (i = 0; i < semaptr->sem_nsems; i++) { 616 error = copyout(&semaptr->_sem_base[i].semval, 617 &arg->array[i], sizeof(arg->array[i])); 618 if (error != 0) 619 break; 620 } 621 break; 622 623 case GETZCNT: 624 if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_R))) 625 break; 626 if (semnum < 0 || semnum >= semaptr->sem_nsems) { 627 error = EINVAL; 628 break; 629 } 630 *retval = semaptr->_sem_base[semnum].semzcnt; 631 break; 632 633 case SETVAL: 634 if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_W))) 635 break; 636 if (semnum < 0 || semnum >= semaptr->sem_nsems) { 637 error = EINVAL; 638 break; 639 } 640 KASSERT(arg != NULL); 641 if ((unsigned int)arg->val > seminfo.semvmx) { 642 error = ERANGE; 643 break; 644 } 645 semaptr->_sem_base[semnum].semval = arg->val; 646 semundo_clear(ix, semnum); 647 cv_broadcast(&semcv[ix]); 648 break; 649 650 case SETALL: 651 if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_W))) 652 break; 653 KASSERT(arg != NULL); 654 for (i = 0; i < semaptr->sem_nsems; i++) { 655 unsigned short semval; 656 error = copyin(&arg->array[i], &semval, 657 sizeof(arg->array[i])); 658 if (error != 0) 659 break; 660 if ((unsigned int)semval > seminfo.semvmx) { 661 error = ERANGE; 662 break; 663 } 664 semaptr->_sem_base[i].semval = semval; 665 } 666 semundo_clear(ix, -1); 667 cv_broadcast(&semcv[ix]); 668 break; 669 670 default: 671 error = EINVAL; 672 break; 673 } 674 675 mutex_exit(&semlock); 676 return (error); 677 } 678 679 int 680 sys_semget(struct lwp *l, const struct sys_semget_args *uap, register_t *retval) 681 { 682 /* { 683 syscallarg(key_t) key; 684 syscallarg(int) nsems; 685 syscallarg(int) semflg; 686 } */ 687 int semid, error = 0; 688 int key = SCARG(uap, key); 689 int nsems = SCARG(uap, nsems); 690 int semflg = SCARG(uap, semflg); 691 kauth_cred_t cred = l->l_cred; 692 693 SEM_PRINTF(("semget(0x%x, %d, 0%o)\n", key, nsems, semflg)); 694 695 mutex_enter(&semlock); 696 697 if (key != IPC_PRIVATE) { 698 for (semid = 0; semid < seminfo.semmni; semid++) { 699 if ((sema[semid].sem_perm.mode & SEM_ALLOC) && 700 sema[semid].sem_perm._key == key) 701 break; 702 } 703 if (semid < seminfo.semmni) { 704 SEM_PRINTF(("found public key\n")); 705 if ((error = ipcperm(cred, &sema[semid].sem_perm, 706 semflg & 0700))) 707 goto out; 708 if (nsems > 0 && sema[semid].sem_nsems < nsems) { 709 SEM_PRINTF(("too small\n")); 710 error = EINVAL; 711 goto out; 712 } 713 if ((semflg & IPC_CREAT) && (semflg & IPC_EXCL)) { 714 SEM_PRINTF(("not exclusive\n")); 715 error = EEXIST; 716 goto out; 717 } 718 goto found; 719 } 720 } 721 722 SEM_PRINTF(("need to allocate the semid_ds\n")); 723 if (key == IPC_PRIVATE || (semflg & IPC_CREAT)) { 724 if (nsems <= 0 || nsems > seminfo.semmsl) { 725 SEM_PRINTF(("nsems out of range (0<%d<=%d)\n", nsems, 726 seminfo.semmsl)); 727 error = EINVAL; 728 goto out; 729 } 730 if (nsems > seminfo.semmns - semtot) { 731 SEM_PRINTF(("not enough semaphores left " 732 "(need %d, got %d)\n", 733 nsems, seminfo.semmns - semtot)); 734 error = ENOSPC; 735 goto out; 736 } 737 for (semid = 0; semid < seminfo.semmni; semid++) { 738 if ((sema[semid].sem_perm.mode & SEM_ALLOC) == 0) 739 break; 740 } 741 if (semid == seminfo.semmni) { 742 SEM_PRINTF(("no more semid_ds's available\n")); 743 error = ENOSPC; 744 goto out; 745 } 746 SEM_PRINTF(("semid %d is available\n", semid)); 747 sema[semid].sem_perm._key = key; 748 sema[semid].sem_perm.cuid = kauth_cred_geteuid(cred); 749 sema[semid].sem_perm.uid = kauth_cred_geteuid(cred); 750 sema[semid].sem_perm.cgid = kauth_cred_getegid(cred); 751 sema[semid].sem_perm.gid = kauth_cred_getegid(cred); 752 sema[semid].sem_perm.mode = (semflg & 0777) | SEM_ALLOC; 753 sema[semid].sem_perm._seq = 754 (sema[semid].sem_perm._seq + 1) & 0x7fff; 755 sema[semid].sem_nsems = nsems; 756 sema[semid].sem_otime = 0; 757 sema[semid].sem_ctime = time_second; 758 sema[semid]._sem_base = &sem[semtot]; 759 semtot += nsems; 760 memset(sema[semid]._sem_base, 0, 761 sizeof(sema[semid]._sem_base[0]) * nsems); 762 SEM_PRINTF(("sembase = %p, next = %p\n", sema[semid]._sem_base, 763 &sem[semtot])); 764 } else { 765 SEM_PRINTF(("didn't find it and wasn't asked to create it\n")); 766 error = ENOENT; 767 goto out; 768 } 769 770 found: 771 *retval = IXSEQ_TO_IPCID(semid, sema[semid].sem_perm); 772 out: 773 mutex_exit(&semlock); 774 return (error); 775 } 776 777 #define SMALL_SOPS 8 778 779 int 780 sys_semop(struct lwp *l, const struct sys_semop_args *uap, register_t *retval) 781 { 782 /* { 783 syscallarg(int) semid; 784 syscallarg(struct sembuf *) sops; 785 syscallarg(size_t) nsops; 786 } */ 787 struct proc *p = l->l_proc; 788 int semid = SCARG(uap, semid), seq; 789 size_t nsops = SCARG(uap, nsops); 790 struct sembuf small_sops[SMALL_SOPS]; 791 struct sembuf *sops; 792 struct semid_ds *semaptr; 793 struct sembuf *sopptr = NULL; 794 struct __sem *semptr = NULL; 795 struct sem_undo *suptr = NULL; 796 kauth_cred_t cred = l->l_cred; 797 int i, error; 798 int do_wakeup, do_undos; 799 800 SEM_PRINTF(("call to semop(%d, %p, %zd)\n", semid, SCARG(uap,sops), nsops)); 801 802 if (__predict_false((p->p_flag & PK_SYSVSEM) == 0)) { 803 mutex_enter(p->p_lock); 804 p->p_flag |= PK_SYSVSEM; 805 mutex_exit(p->p_lock); 806 } 807 808 restart: 809 if (nsops <= SMALL_SOPS) { 810 sops = small_sops; 811 } else if (nsops <= seminfo.semopm) { 812 sops = kmem_alloc(nsops * sizeof(*sops), KM_SLEEP); 813 } else { 814 SEM_PRINTF(("too many sops (max=%d, nsops=%zd)\n", 815 seminfo.semopm, nsops)); 816 return (E2BIG); 817 } 818 819 error = copyin(SCARG(uap, sops), sops, nsops * sizeof(sops[0])); 820 if (error) { 821 SEM_PRINTF(("error = %d from copyin(%p, %p, %zd)\n", error, 822 SCARG(uap, sops), &sops, nsops * sizeof(sops[0]))); 823 if (sops != small_sops) 824 kmem_free(sops, nsops * sizeof(*sops)); 825 return error; 826 } 827 828 mutex_enter(&semlock); 829 /* In case of reallocation, we will wait for completion */ 830 while (__predict_false(sem_realloc_state)) 831 cv_wait(&sem_realloc_cv, &semlock); 832 833 semid = IPCID_TO_IX(semid); /* Convert back to zero origin */ 834 if (semid < 0 || semid >= seminfo.semmni) { 835 error = EINVAL; 836 goto out; 837 } 838 839 semaptr = &sema[semid]; 840 seq = IPCID_TO_SEQ(SCARG(uap, semid)); 841 if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0 || 842 semaptr->sem_perm._seq != seq) { 843 error = EINVAL; 844 goto out; 845 } 846 847 if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_W))) { 848 SEM_PRINTF(("error = %d from ipaccess\n", error)); 849 goto out; 850 } 851 852 for (i = 0; i < nsops; i++) 853 if (sops[i].sem_num >= semaptr->sem_nsems) { 854 error = EFBIG; 855 goto out; 856 } 857 858 /* 859 * Loop trying to satisfy the vector of requests. 860 * If we reach a point where we must wait, any requests already 861 * performed are rolled back and we go to sleep until some other 862 * process wakes us up. At this point, we start all over again. 863 * 864 * This ensures that from the perspective of other tasks, a set 865 * of requests is atomic (never partially satisfied). 866 */ 867 do_undos = 0; 868 869 for (;;) { 870 do_wakeup = 0; 871 872 for (i = 0; i < nsops; i++) { 873 sopptr = &sops[i]; 874 semptr = &semaptr->_sem_base[sopptr->sem_num]; 875 876 SEM_PRINTF(("semop: semaptr=%p, sem_base=%p, " 877 "semptr=%p, sem[%d]=%d : op=%d, flag=%s\n", 878 semaptr, semaptr->_sem_base, semptr, 879 sopptr->sem_num, semptr->semval, sopptr->sem_op, 880 (sopptr->sem_flg & IPC_NOWAIT) ? 881 "nowait" : "wait")); 882 883 if (sopptr->sem_op < 0) { 884 if ((int)(semptr->semval + 885 sopptr->sem_op) < 0) { 886 SEM_PRINTF(("semop: " 887 "can't do it now\n")); 888 break; 889 } else { 890 semptr->semval += sopptr->sem_op; 891 if (semptr->semval == 0 && 892 semptr->semzcnt > 0) 893 do_wakeup = 1; 894 } 895 if (sopptr->sem_flg & SEM_UNDO) 896 do_undos = 1; 897 } else if (sopptr->sem_op == 0) { 898 if (semptr->semval > 0) { 899 SEM_PRINTF(("semop: not zero now\n")); 900 break; 901 } 902 } else { 903 if (semptr->semncnt > 0) 904 do_wakeup = 1; 905 semptr->semval += sopptr->sem_op; 906 if (sopptr->sem_flg & SEM_UNDO) 907 do_undos = 1; 908 } 909 } 910 911 /* 912 * Did we get through the entire vector? 913 */ 914 if (i >= nsops) 915 goto done; 916 917 /* 918 * No ... rollback anything that we've already done 919 */ 920 SEM_PRINTF(("semop: rollback 0 through %d\n", i - 1)); 921 while (i-- > 0) 922 semaptr->_sem_base[sops[i].sem_num].semval -= 923 sops[i].sem_op; 924 925 /* 926 * If the request that we couldn't satisfy has the 927 * NOWAIT flag set then return with EAGAIN. 928 */ 929 if (sopptr->sem_flg & IPC_NOWAIT) { 930 error = EAGAIN; 931 goto out; 932 } 933 934 if (sopptr->sem_op == 0) 935 semptr->semzcnt++; 936 else 937 semptr->semncnt++; 938 939 sem_waiters++; 940 SEM_PRINTF(("semop: good night!\n")); 941 error = cv_wait_sig(&semcv[semid], &semlock); 942 SEM_PRINTF(("semop: good morning (error=%d)!\n", error)); 943 sem_waiters--; 944 945 /* Notify reallocator, if it is waiting */ 946 cv_broadcast(&sem_realloc_cv); 947 948 /* 949 * Make sure that the semaphore still exists 950 */ 951 if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0 || 952 semaptr->sem_perm._seq != seq) { 953 error = EIDRM; 954 goto out; 955 } 956 957 /* 958 * The semaphore is still alive. Readjust the count of 959 * waiting processes. 960 */ 961 semptr = &semaptr->_sem_base[sopptr->sem_num]; 962 if (sopptr->sem_op == 0) 963 semptr->semzcnt--; 964 else 965 semptr->semncnt--; 966 967 /* In case of such state, restart the call */ 968 if (sem_realloc_state) { 969 mutex_exit(&semlock); 970 goto restart; 971 } 972 973 /* Is it really morning, or was our sleep interrupted? */ 974 if (error != 0) { 975 error = EINTR; 976 goto out; 977 } 978 SEM_PRINTF(("semop: good morning!\n")); 979 } 980 981 done: 982 /* 983 * Process any SEM_UNDO requests. 984 */ 985 if (do_undos) { 986 for (i = 0; i < nsops; i++) { 987 /* 988 * We only need to deal with SEM_UNDO's for non-zero 989 * op's. 990 */ 991 int adjval; 992 993 if ((sops[i].sem_flg & SEM_UNDO) == 0) 994 continue; 995 adjval = sops[i].sem_op; 996 if (adjval == 0) 997 continue; 998 error = semundo_adjust(p, &suptr, semid, 999 sops[i].sem_num, -adjval); 1000 if (error == 0) 1001 continue; 1002 1003 /* 1004 * Oh-Oh! We ran out of either sem_undo's or undo's. 1005 * Rollback the adjustments to this point and then 1006 * rollback the semaphore ups and down so we can return 1007 * with an error with all structures restored. We 1008 * rollback the undo's in the exact reverse order that 1009 * we applied them. This guarantees that we won't run 1010 * out of space as we roll things back out. 1011 */ 1012 while (i-- > 0) { 1013 if ((sops[i].sem_flg & SEM_UNDO) == 0) 1014 continue; 1015 adjval = sops[i].sem_op; 1016 if (adjval == 0) 1017 continue; 1018 if (semundo_adjust(p, &suptr, semid, 1019 sops[i].sem_num, adjval) != 0) 1020 panic("semop - can't undo undos"); 1021 } 1022 1023 for (i = 0; i < nsops; i++) 1024 semaptr->_sem_base[sops[i].sem_num].semval -= 1025 sops[i].sem_op; 1026 1027 SEM_PRINTF(("error = %d from semundo_adjust\n", error)); 1028 goto out; 1029 } /* loop through the sops */ 1030 } /* if (do_undos) */ 1031 1032 /* We're definitely done - set the sempid's */ 1033 for (i = 0; i < nsops; i++) { 1034 sopptr = &sops[i]; 1035 semptr = &semaptr->_sem_base[sopptr->sem_num]; 1036 semptr->sempid = p->p_pid; 1037 } 1038 1039 /* Update sem_otime */ 1040 semaptr->sem_otime = time_second; 1041 1042 /* Do a wakeup if any semaphore was up'd. */ 1043 if (do_wakeup) { 1044 SEM_PRINTF(("semop: doing wakeup\n")); 1045 cv_broadcast(&semcv[semid]); 1046 SEM_PRINTF(("semop: back from wakeup\n")); 1047 } 1048 SEM_PRINTF(("semop: done\n")); 1049 *retval = 0; 1050 1051 out: 1052 mutex_exit(&semlock); 1053 if (sops != small_sops) 1054 kmem_free(sops, nsops * sizeof(*sops)); 1055 return error; 1056 } 1057 1058 /* 1059 * Go through the undo structures for this process and apply the 1060 * adjustments to semaphores. 1061 */ 1062 /*ARGSUSED*/ 1063 void 1064 semexit(struct proc *p, void *v) 1065 { 1066 struct sem_undo *suptr; 1067 struct sem_undo **supptr; 1068 1069 if ((p->p_flag & PK_SYSVSEM) == 0) 1070 return; 1071 1072 mutex_enter(&semlock); 1073 1074 /* 1075 * Go through the chain of undo vectors looking for one 1076 * associated with this process. 1077 */ 1078 1079 for (supptr = &semu_list; (suptr = *supptr) != NULL; 1080 supptr = &suptr->un_next) { 1081 if (suptr->un_proc == p) 1082 break; 1083 } 1084 1085 /* 1086 * If there is no undo vector, skip to the end. 1087 */ 1088 1089 if (suptr == NULL) { 1090 mutex_exit(&semlock); 1091 return; 1092 } 1093 1094 /* 1095 * We now have an undo vector for this process. 1096 */ 1097 1098 SEM_PRINTF(("proc @%p has undo structure with %d entries\n", p, 1099 suptr->un_cnt)); 1100 1101 /* 1102 * If there are any active undo elements then process them. 1103 */ 1104 if (suptr->un_cnt > 0) { 1105 int ix; 1106 1107 for (ix = 0; ix < suptr->un_cnt; ix++) { 1108 int semid = suptr->un_ent[ix].un_id; 1109 int semnum = suptr->un_ent[ix].un_num; 1110 int adjval = suptr->un_ent[ix].un_adjval; 1111 struct semid_ds *semaptr; 1112 1113 semaptr = &sema[semid]; 1114 if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0) 1115 panic("semexit - semid not allocated"); 1116 if (semnum >= semaptr->sem_nsems) 1117 panic("semexit - semnum out of range"); 1118 1119 SEM_PRINTF(("semexit: %p id=%d num=%d(adj=%d) ; " 1120 "sem=%d\n", 1121 suptr->un_proc, suptr->un_ent[ix].un_id, 1122 suptr->un_ent[ix].un_num, 1123 suptr->un_ent[ix].un_adjval, 1124 semaptr->_sem_base[semnum].semval)); 1125 1126 if (adjval < 0 && 1127 semaptr->_sem_base[semnum].semval < -adjval) 1128 semaptr->_sem_base[semnum].semval = 0; 1129 else 1130 semaptr->_sem_base[semnum].semval += adjval; 1131 1132 cv_broadcast(&semcv[semid]); 1133 SEM_PRINTF(("semexit: back from wakeup\n")); 1134 } 1135 } 1136 1137 /* 1138 * Deallocate the undo vector. 1139 */ 1140 SEM_PRINTF(("removing vector\n")); 1141 suptr->un_proc = NULL; 1142 *supptr = suptr->un_next; 1143 mutex_exit(&semlock); 1144 } 1145 1146 /* 1147 * Sysctl initialization and nodes. 1148 */ 1149 1150 static int 1151 sysctl_ipc_semmni(SYSCTLFN_ARGS) 1152 { 1153 int newsize, error; 1154 struct sysctlnode node; 1155 node = *rnode; 1156 node.sysctl_data = &newsize; 1157 1158 newsize = seminfo.semmni; 1159 error = sysctl_lookup(SYSCTLFN_CALL(&node)); 1160 if (error || newp == NULL) 1161 return error; 1162 1163 return semrealloc(newsize, seminfo.semmns, seminfo.semmnu); 1164 } 1165 1166 static int 1167 sysctl_ipc_semmns(SYSCTLFN_ARGS) 1168 { 1169 int newsize, error; 1170 struct sysctlnode node; 1171 node = *rnode; 1172 node.sysctl_data = &newsize; 1173 1174 newsize = seminfo.semmns; 1175 error = sysctl_lookup(SYSCTLFN_CALL(&node)); 1176 if (error || newp == NULL) 1177 return error; 1178 1179 return semrealloc(seminfo.semmni, newsize, seminfo.semmnu); 1180 } 1181 1182 static int 1183 sysctl_ipc_semmnu(SYSCTLFN_ARGS) 1184 { 1185 int newsize, error; 1186 struct sysctlnode node; 1187 node = *rnode; 1188 node.sysctl_data = &newsize; 1189 1190 newsize = seminfo.semmnu; 1191 error = sysctl_lookup(SYSCTLFN_CALL(&node)); 1192 if (error || newp == NULL) 1193 return error; 1194 1195 return semrealloc(seminfo.semmni, seminfo.semmns, newsize); 1196 } 1197 1198 SYSCTL_SETUP(sysctl_ipc_sem_setup, "sysctl kern.ipc subtree setup") 1199 { 1200 const struct sysctlnode *node = NULL; 1201 1202 sysctl_createv(clog, 0, NULL, &node, 1203 CTLFLAG_PERMANENT, 1204 CTLTYPE_NODE, "ipc", 1205 SYSCTL_DESCR("SysV IPC options"), 1206 NULL, 0, NULL, 0, 1207 CTL_KERN, KERN_SYSVIPC, CTL_EOL); 1208 1209 if (node == NULL) 1210 return; 1211 1212 sysctl_createv(clog, 0, &node, NULL, 1213 CTLFLAG_PERMANENT | CTLFLAG_READWRITE, 1214 CTLTYPE_INT, "semmni", 1215 SYSCTL_DESCR("Max number of number of semaphore identifiers"), 1216 sysctl_ipc_semmni, 0, &seminfo.semmni, 0, 1217 CTL_CREATE, CTL_EOL); 1218 sysctl_createv(clog, 0, &node, NULL, 1219 CTLFLAG_PERMANENT | CTLFLAG_READWRITE, 1220 CTLTYPE_INT, "semmns", 1221 SYSCTL_DESCR("Max number of number of semaphores in system"), 1222 sysctl_ipc_semmns, 0, &seminfo.semmns, 0, 1223 CTL_CREATE, CTL_EOL); 1224 sysctl_createv(clog, 0, &node, NULL, 1225 CTLFLAG_PERMANENT | CTLFLAG_READWRITE, 1226 CTLTYPE_INT, "semmnu", 1227 SYSCTL_DESCR("Max number of undo structures in system"), 1228 sysctl_ipc_semmnu, 0, &seminfo.semmnu, 0, 1229 CTL_CREATE, CTL_EOL); 1230 } 1231