1 /* $NetBSD: sysv_sem.c,v 1.104 2025/05/23 09:47:34 hannken 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 1.104 2025/05/23 09:47:34 hannken 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 #include <sys/once.h> 57 58 /* 59 * Memory areas: 60 * 1st: Pool of semaphore identifiers 61 * 2nd: Semaphores 62 * 3rd: Conditional variables 63 * 4th: Undo structures 64 */ 65 struct semid_ds * sema __read_mostly; 66 static struct __sem * sem __read_mostly; 67 static kcondvar_t * semcv __read_mostly; 68 static int * semu __read_mostly; 69 70 static kmutex_t semlock __cacheline_aligned; 71 static bool sem_realloc_state __read_mostly; 72 static kcondvar_t sem_realloc_cv; 73 74 /* 75 * List of active undo structures, total number of semaphores, 76 * and total number of semop waiters. 77 */ 78 static struct sem_undo *semu_list __read_mostly; 79 static u_int semtot __cacheline_aligned; 80 static u_int sem_waiters __cacheline_aligned; 81 82 /* Macro to find a particular sem_undo vector */ 83 #define SEMU(s, ix) ((struct sem_undo *)(((long)s) + ix * seminfo.semusz)) 84 85 #ifdef SEM_DEBUG 86 #define SEM_PRINTF(a) printf a 87 #else 88 #define SEM_PRINTF(a) 89 #endif 90 91 void *hook; /* cookie from exithook_establish() */ 92 93 extern int kern_has_sysvsem; 94 95 SYSCTL_SETUP_PROTO(sysctl_ipc_sem_setup); 96 97 struct sem_undo *semu_alloc(struct proc *); 98 int semundo_adjust(struct proc *, struct sem_undo **, int, int, int); 99 void semundo_clear(int, int); 100 101 static ONCE_DECL(exithook_control); 102 static int seminit_exithook(void); 103 104 int 105 seminit(void) 106 { 107 int i, sz; 108 vaddr_t v; 109 110 mutex_init(&semlock, MUTEX_DEFAULT, IPL_NONE); 111 cv_init(&sem_realloc_cv, "semrealc"); 112 sem_realloc_state = false; 113 semtot = 0; 114 sem_waiters = 0; 115 116 /* Allocate the wired memory for our structures */ 117 sz = ALIGN(seminfo.semmni * sizeof(struct semid_ds)) + 118 ALIGN(seminfo.semmns * sizeof(struct __sem)) + 119 ALIGN(seminfo.semmni * sizeof(kcondvar_t)) + 120 ALIGN(seminfo.semmnu * seminfo.semusz); 121 sz = round_page(sz); 122 v = uvm_km_alloc(kernel_map, sz, 0, UVM_KMF_WIRED|UVM_KMF_ZERO); 123 if (v == 0) { 124 printf("sysv_sem: cannot allocate memory"); 125 return ENOMEM; 126 } 127 sema = (void *)v; 128 sem = (void *)((uintptr_t)sema + 129 ALIGN(seminfo.semmni * sizeof(struct semid_ds))); 130 semcv = (void *)((uintptr_t)sem + 131 ALIGN(seminfo.semmns * sizeof(struct __sem))); 132 semu = (void *)((uintptr_t)semcv + 133 ALIGN(seminfo.semmni * sizeof(kcondvar_t))); 134 135 for (i = 0; i < seminfo.semmni; i++) { 136 sema[i]._sem_base = 0; 137 sema[i].sem_perm.mode = 0; 138 cv_init(&semcv[i], "semwait"); 139 } 140 for (i = 0; i < seminfo.semmnu; i++) { 141 struct sem_undo *suptr = SEMU(semu, i); 142 suptr->un_proc = NULL; 143 } 144 semu_list = NULL; 145 146 kern_has_sysvsem = 1; 147 148 return 0; 149 } 150 151 static int 152 seminit_exithook(void) 153 { 154 155 hook = exithook_establish(semexit, NULL); 156 return 0; 157 } 158 159 int 160 semfini(void) 161 { 162 int i, sz; 163 vaddr_t v = (vaddr_t)sema; 164 165 /* Don't allow module unload if we're busy */ 166 mutex_enter(&semlock); 167 if (semtot) { 168 mutex_exit(&semlock); 169 return 1; 170 } 171 172 /* Remove the exit hook */ 173 if (hook) 174 exithook_disestablish(hook); 175 176 /* Destroy all our condvars */ 177 for (i = 0; i < seminfo.semmni; i++) { 178 cv_destroy(&semcv[i]); 179 } 180 181 /* Free the wired memory that we allocated */ 182 sz = ALIGN(seminfo.semmni * sizeof(struct semid_ds)) + 183 ALIGN(seminfo.semmns * sizeof(struct __sem)) + 184 ALIGN(seminfo.semmni * sizeof(kcondvar_t)) + 185 ALIGN(seminfo.semmnu * seminfo.semusz); 186 sz = round_page(sz); 187 uvm_km_free(kernel_map, v, sz, UVM_KMF_WIRED); 188 189 /* Destroy the last cv and mutex */ 190 cv_destroy(&sem_realloc_cv); 191 mutex_exit(&semlock); 192 mutex_destroy(&semlock); 193 194 kern_has_sysvsem = 0; 195 196 return 0; 197 } 198 199 static int 200 semrealloc(int newsemmni, int newsemmns, int newsemmnu) 201 { 202 struct semid_ds *new_sema, *old_sema; 203 struct __sem *new_sem; 204 struct sem_undo *new_semu_list, *suptr, *nsuptr; 205 int *new_semu; 206 kcondvar_t *new_semcv; 207 vaddr_t v; 208 int i, j, lsemid, nmnus, sz; 209 210 if (newsemmni < 1 || newsemmns < 1 || newsemmnu < 1) 211 return EINVAL; 212 213 /* Allocate the wired memory for our structures */ 214 sz = ALIGN(newsemmni * sizeof(struct semid_ds)) + 215 ALIGN(newsemmns * sizeof(struct __sem)) + 216 ALIGN(newsemmni * sizeof(kcondvar_t)) + 217 ALIGN(newsemmnu * seminfo.semusz); 218 sz = round_page(sz); 219 v = uvm_km_alloc(kernel_map, sz, 0, UVM_KMF_WIRED|UVM_KMF_ZERO); 220 if (v == 0) 221 return ENOMEM; 222 223 mutex_enter(&semlock); 224 if (sem_realloc_state) { 225 mutex_exit(&semlock); 226 uvm_km_free(kernel_map, v, sz, UVM_KMF_WIRED); 227 return EBUSY; 228 } 229 sem_realloc_state = true; 230 if (sem_waiters) { 231 /* 232 * Mark reallocation state, wake-up all waiters, 233 * and wait while they will all exit. 234 */ 235 for (i = 0; i < seminfo.semmni; i++) 236 cv_broadcast(&semcv[i]); 237 while (sem_waiters) 238 cv_wait(&sem_realloc_cv, &semlock); 239 } 240 old_sema = sema; 241 242 /* Get the number of last slot */ 243 lsemid = 0; 244 for (i = 0; i < seminfo.semmni; i++) 245 if (sema[i].sem_perm.mode & SEM_ALLOC) 246 lsemid = i; 247 248 /* Get the number of currently used undo structures */ 249 nmnus = 0; 250 for (i = 0; i < seminfo.semmnu; i++) { 251 suptr = SEMU(semu, i); 252 if (suptr->un_proc == NULL) 253 continue; 254 nmnus++; 255 } 256 257 /* We cannot reallocate less memory than we use */ 258 if (lsemid >= newsemmni || semtot > newsemmns || nmnus > newsemmnu) { 259 mutex_exit(&semlock); 260 uvm_km_free(kernel_map, v, sz, UVM_KMF_WIRED); 261 return EBUSY; 262 } 263 264 new_sema = (void *)v; 265 new_sem = (void *)((uintptr_t)new_sema + 266 ALIGN(newsemmni * sizeof(struct semid_ds))); 267 new_semcv = (void *)((uintptr_t)new_sem + 268 ALIGN(newsemmns * sizeof(struct __sem))); 269 new_semu = (void *)((uintptr_t)new_semcv + 270 ALIGN(newsemmni * sizeof(kcondvar_t))); 271 272 /* Initialize all semaphore identifiers and condvars */ 273 for (i = 0; i < newsemmni; i++) { 274 new_sema[i]._sem_base = 0; 275 new_sema[i].sem_perm.mode = 0; 276 cv_init(&new_semcv[i], "semwait"); 277 } 278 for (i = 0; i < newsemmnu; i++) { 279 nsuptr = SEMU(new_semu, i); 280 nsuptr->un_proc = NULL; 281 } 282 283 /* 284 * Copy all identifiers, semaphores and list of the 285 * undo structures to the new memory allocation. 286 */ 287 j = 0; 288 for (i = 0; i <= lsemid; i++) { 289 if ((sema[i].sem_perm.mode & SEM_ALLOC) == 0) 290 continue; 291 memcpy(&new_sema[i], &sema[i], sizeof(struct semid_ds)); 292 new_sema[i]._sem_base = &new_sem[j]; 293 memcpy(new_sema[i]._sem_base, sema[i]._sem_base, 294 (sizeof(struct __sem) * sema[i].sem_nsems)); 295 j += sema[i].sem_nsems; 296 } 297 KASSERT(j == semtot); 298 299 j = 0; 300 new_semu_list = NULL; 301 for (suptr = semu_list; suptr != NULL; suptr = suptr->un_next) { 302 KASSERT(j < newsemmnu); 303 nsuptr = SEMU(new_semu, j); 304 memcpy(nsuptr, suptr, SEMUSZ); 305 nsuptr->un_next = new_semu_list; 306 new_semu_list = nsuptr; 307 j++; 308 } 309 310 for (i = 0; i < seminfo.semmni; i++) { 311 KASSERT(cv_has_waiters(&semcv[i]) == false); 312 cv_destroy(&semcv[i]); 313 } 314 315 sz = ALIGN(seminfo.semmni * sizeof(struct semid_ds)) + 316 ALIGN(seminfo.semmns * sizeof(struct __sem)) + 317 ALIGN(seminfo.semmni * sizeof(kcondvar_t)) + 318 ALIGN(seminfo.semmnu * seminfo.semusz); 319 sz = round_page(sz); 320 321 /* Set the pointers and update the new values */ 322 sema = new_sema; 323 sem = new_sem; 324 semcv = new_semcv; 325 semu = new_semu; 326 semu_list = new_semu_list; 327 328 seminfo.semmni = newsemmni; 329 seminfo.semmns = newsemmns; 330 seminfo.semmnu = newsemmnu; 331 332 /* Reallocation completed - notify all waiters, if any */ 333 sem_realloc_state = false; 334 cv_broadcast(&sem_realloc_cv); 335 mutex_exit(&semlock); 336 337 uvm_km_free(kernel_map, (vaddr_t)old_sema, sz, UVM_KMF_WIRED); 338 return 0; 339 } 340 341 /* 342 * Placebo. 343 */ 344 345 int 346 sys_semconfig(struct lwp *l, const struct sys_semconfig_args *uap, 347 register_t *retval) 348 { 349 350 RUN_ONCE(&exithook_control, seminit_exithook); 351 352 *retval = 0; 353 return 0; 354 } 355 356 /* 357 * Allocate a new sem_undo structure for a process. 358 * => Returns NULL on failure. 359 */ 360 struct sem_undo * 361 semu_alloc(struct proc *p) 362 { 363 struct sem_undo *suptr, **supptr; 364 bool attempted = false; 365 int i; 366 367 KASSERT(mutex_owned(&semlock)); 368 again: 369 /* Look for a free structure. */ 370 for (i = 0; i < seminfo.semmnu; i++) { 371 suptr = SEMU(semu, i); 372 if (suptr->un_proc == NULL) { 373 /* Found. Fill it in and return. */ 374 suptr->un_next = semu_list; 375 semu_list = suptr; 376 suptr->un_cnt = 0; 377 suptr->un_proc = p; 378 return suptr; 379 } 380 } 381 382 /* Not found. Attempt to free some structures. */ 383 if (!attempted) { 384 bool freed = false; 385 386 attempted = true; 387 supptr = &semu_list; 388 while ((suptr = *supptr) != NULL) { 389 if (suptr->un_cnt == 0) { 390 suptr->un_proc = NULL; 391 *supptr = suptr->un_next; 392 freed = true; 393 } else { 394 supptr = &suptr->un_next; 395 } 396 } 397 if (freed) { 398 goto again; 399 } 400 } 401 return NULL; 402 } 403 404 /* 405 * Adjust a particular entry for a particular proc 406 */ 407 408 int 409 semundo_adjust(struct proc *p, struct sem_undo **supptr, int semid, int semnum, 410 int adjval) 411 { 412 struct sem_undo *suptr; 413 struct sem_undo_entry *sunptr; 414 int i; 415 416 KASSERT(mutex_owned(&semlock)); 417 418 /* 419 * Look for and remember the sem_undo if the caller doesn't 420 * provide it 421 */ 422 423 suptr = *supptr; 424 if (suptr == NULL) { 425 for (suptr = semu_list; suptr != NULL; suptr = suptr->un_next) 426 if (suptr->un_proc == p) 427 break; 428 429 if (suptr == NULL) { 430 suptr = semu_alloc(p); 431 if (suptr == NULL) 432 return (ENOSPC); 433 } 434 *supptr = suptr; 435 } 436 437 /* 438 * Look for the requested entry and adjust it (delete if 439 * adjval becomes 0). 440 */ 441 sunptr = &suptr->un_ent[0]; 442 for (i = 0; i < suptr->un_cnt; i++, sunptr++) { 443 if (sunptr->un_id != semid || sunptr->un_num != semnum) 444 continue; 445 sunptr->un_adjval += adjval; 446 if (sunptr->un_adjval == 0) { 447 suptr->un_cnt--; 448 if (i < suptr->un_cnt) 449 suptr->un_ent[i] = 450 suptr->un_ent[suptr->un_cnt]; 451 } 452 return (0); 453 } 454 455 /* Didn't find the right entry - create it */ 456 if (suptr->un_cnt == SEMUME) 457 return (EINVAL); 458 459 sunptr = &suptr->un_ent[suptr->un_cnt]; 460 suptr->un_cnt++; 461 sunptr->un_adjval = adjval; 462 sunptr->un_id = semid; 463 sunptr->un_num = semnum; 464 return (0); 465 } 466 467 void 468 semundo_clear(int semid, int semnum) 469 { 470 struct sem_undo *suptr; 471 struct sem_undo_entry *sunptr, *sunend; 472 473 KASSERT(mutex_owned(&semlock)); 474 475 for (suptr = semu_list; suptr != NULL; suptr = suptr->un_next) 476 for (sunptr = &suptr->un_ent[0], 477 sunend = sunptr + suptr->un_cnt; sunptr < sunend;) { 478 if (sunptr->un_id == semid) { 479 if (semnum == -1 || sunptr->un_num == semnum) { 480 suptr->un_cnt--; 481 sunend--; 482 if (sunptr != sunend) 483 *sunptr = *sunend; 484 if (semnum != -1) 485 break; 486 else 487 continue; 488 } 489 } 490 sunptr++; 491 } 492 } 493 494 int 495 sys_____semctl50(struct lwp *l, const struct sys_____semctl50_args *uap, 496 register_t *retval) 497 { 498 /* { 499 syscallarg(int) semid; 500 syscallarg(int) semnum; 501 syscallarg(int) cmd; 502 syscallarg(union __semun *) arg; 503 } */ 504 struct semid_ds sembuf; 505 int cmd, error; 506 void *pass_arg; 507 union __semun karg; 508 509 RUN_ONCE(&exithook_control, seminit_exithook); 510 511 cmd = SCARG(uap, cmd); 512 513 pass_arg = get_semctl_arg(cmd, &sembuf, &karg); 514 515 if (pass_arg) { 516 error = copyin(SCARG(uap, arg), &karg, sizeof(karg)); 517 if (error) 518 return error; 519 if (cmd == IPC_SET) { 520 error = copyin(karg.buf, &sembuf, sizeof(sembuf)); 521 if (error) 522 return (error); 523 } 524 } 525 526 error = semctl1(l, SCARG(uap, semid), SCARG(uap, semnum), cmd, 527 pass_arg, retval); 528 529 if (error == 0 && cmd == IPC_STAT) 530 error = copyout(&sembuf, karg.buf, sizeof(sembuf)); 531 532 return (error); 533 } 534 535 int 536 semctl1(struct lwp *l, int semid, int semnum, int cmd, void *v, 537 register_t *retval) 538 { 539 kauth_cred_t cred = l->l_cred; 540 union __semun *arg = v; 541 struct semid_ds *sembuf = v, *semaptr; 542 int i, error, ix; 543 544 SEM_PRINTF(("call to semctl(%d, %d, %d, %p)\n", 545 semid, semnum, cmd, v)); 546 547 mutex_enter(&semlock); 548 549 ix = IPCID_TO_IX(semid); 550 if (ix < 0 || ix >= seminfo.semmni) { 551 mutex_exit(&semlock); 552 return (EINVAL); 553 } 554 555 semaptr = &sema[ix]; 556 if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0 || 557 semaptr->sem_perm._seq != IPCID_TO_SEQ(semid)) { 558 mutex_exit(&semlock); 559 return (EINVAL); 560 } 561 562 switch (cmd) { 563 case IPC_RMID: 564 if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_M)) != 0) 565 break; 566 semaptr->sem_perm.cuid = kauth_cred_geteuid(cred); 567 semaptr->sem_perm.uid = kauth_cred_geteuid(cred); 568 semtot -= semaptr->sem_nsems; 569 for (i = semaptr->_sem_base - sem; i < semtot; i++) 570 sem[i] = sem[i + semaptr->sem_nsems]; 571 for (i = 0; i < seminfo.semmni; i++) { 572 if ((sema[i].sem_perm.mode & SEM_ALLOC) && 573 sema[i]._sem_base > semaptr->_sem_base) 574 sema[i]._sem_base -= semaptr->sem_nsems; 575 } 576 semaptr->sem_perm.mode = 0; 577 semundo_clear(ix, -1); 578 cv_broadcast(&semcv[ix]); 579 break; 580 581 case IPC_SET: 582 if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_M))) 583 break; 584 KASSERT(sembuf != NULL); 585 semaptr->sem_perm.uid = sembuf->sem_perm.uid; 586 semaptr->sem_perm.gid = sembuf->sem_perm.gid; 587 semaptr->sem_perm.mode = (semaptr->sem_perm.mode & ~0777) | 588 (sembuf->sem_perm.mode & 0777); 589 semaptr->sem_ctime = time_second; 590 break; 591 592 case IPC_STAT: 593 if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_R))) 594 break; 595 KASSERT(sembuf != NULL); 596 memset(sembuf, 0, sizeof *sembuf); 597 sembuf->sem_perm = semaptr->sem_perm; 598 sembuf->sem_perm.mode &= 0777; 599 sembuf->sem_nsems = semaptr->sem_nsems; 600 sembuf->sem_otime = semaptr->sem_otime; 601 sembuf->sem_ctime = semaptr->sem_ctime; 602 break; 603 604 case GETNCNT: 605 if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_R))) 606 break; 607 if (semnum < 0 || semnum >= semaptr->sem_nsems) { 608 error = EINVAL; 609 break; 610 } 611 *retval = semaptr->_sem_base[semnum].semncnt; 612 break; 613 614 case GETPID: 615 if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_R))) 616 break; 617 if (semnum < 0 || semnum >= semaptr->sem_nsems) { 618 error = EINVAL; 619 break; 620 } 621 *retval = semaptr->_sem_base[semnum].sempid; 622 break; 623 624 case GETVAL: 625 if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_R))) 626 break; 627 if (semnum < 0 || semnum >= semaptr->sem_nsems) { 628 error = EINVAL; 629 break; 630 } 631 *retval = semaptr->_sem_base[semnum].semval; 632 break; 633 634 case GETALL: 635 if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_R))) 636 break; 637 KASSERT(arg != NULL); 638 for (i = 0; i < semaptr->sem_nsems; i++) { 639 error = copyout(&semaptr->_sem_base[i].semval, 640 &arg->array[i], sizeof(arg->array[i])); 641 if (error != 0) 642 break; 643 } 644 break; 645 646 case GETZCNT: 647 if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_R))) 648 break; 649 if (semnum < 0 || semnum >= semaptr->sem_nsems) { 650 error = EINVAL; 651 break; 652 } 653 *retval = semaptr->_sem_base[semnum].semzcnt; 654 break; 655 656 case SETVAL: 657 if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_W))) 658 break; 659 if (semnum < 0 || semnum >= semaptr->sem_nsems) { 660 error = EINVAL; 661 break; 662 } 663 KASSERT(arg != NULL); 664 if ((unsigned int)arg->val > seminfo.semvmx) { 665 error = ERANGE; 666 break; 667 } 668 semaptr->_sem_base[semnum].semval = arg->val; 669 semundo_clear(ix, semnum); 670 cv_broadcast(&semcv[ix]); 671 break; 672 673 case SETALL: 674 if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_W))) 675 break; 676 KASSERT(arg != NULL); 677 for (i = 0; i < semaptr->sem_nsems; i++) { 678 unsigned short semval; 679 error = copyin(&arg->array[i], &semval, 680 sizeof(arg->array[i])); 681 if (error != 0) 682 break; 683 if ((unsigned int)semval > seminfo.semvmx) { 684 error = ERANGE; 685 break; 686 } 687 semaptr->_sem_base[i].semval = semval; 688 } 689 semundo_clear(ix, -1); 690 cv_broadcast(&semcv[ix]); 691 break; 692 693 default: 694 error = EINVAL; 695 break; 696 } 697 698 mutex_exit(&semlock); 699 return (error); 700 } 701 702 int 703 sys_semget(struct lwp *l, const struct sys_semget_args *uap, 704 register_t *retval) 705 { 706 /* { 707 syscallarg(key_t) key; 708 syscallarg(int) nsems; 709 syscallarg(int) semflg; 710 } */ 711 int semid, error = 0; 712 int key = SCARG(uap, key); 713 int nsems = SCARG(uap, nsems); 714 int semflg = SCARG(uap, semflg); 715 kauth_cred_t cred = l->l_cred; 716 717 RUN_ONCE(&exithook_control, seminit_exithook); 718 719 SEM_PRINTF(("semget(0x%x, %d, 0%o)\n", key, nsems, semflg)); 720 721 mutex_enter(&semlock); 722 723 if (key != IPC_PRIVATE) { 724 for (semid = 0; semid < seminfo.semmni; semid++) { 725 if ((sema[semid].sem_perm.mode & SEM_ALLOC) && 726 sema[semid].sem_perm._key == key) 727 break; 728 } 729 if (semid < seminfo.semmni) { 730 SEM_PRINTF(("found public key\n")); 731 if ((error = ipcperm(cred, &sema[semid].sem_perm, 732 semflg & 0700))) 733 goto out; 734 if (nsems > 0 && sema[semid].sem_nsems < nsems) { 735 SEM_PRINTF(("too small\n")); 736 error = EINVAL; 737 goto out; 738 } 739 if ((semflg & IPC_CREAT) && (semflg & IPC_EXCL)) { 740 SEM_PRINTF(("not exclusive\n")); 741 error = EEXIST; 742 goto out; 743 } 744 goto found; 745 } 746 } 747 748 SEM_PRINTF(("need to allocate the semid_ds\n")); 749 if (key == IPC_PRIVATE || (semflg & IPC_CREAT)) { 750 if (nsems <= 0 || nsems > seminfo.semmsl) { 751 SEM_PRINTF(("nsems out of range (0<%d<=%d)\n", nsems, 752 seminfo.semmsl)); 753 error = EINVAL; 754 goto out; 755 } 756 if (nsems > seminfo.semmns - semtot) { 757 SEM_PRINTF(("not enough semaphores left " 758 "(need %d, got %d)\n", 759 nsems, seminfo.semmns - semtot)); 760 error = ENOSPC; 761 goto out; 762 } 763 for (semid = 0; semid < seminfo.semmni; semid++) { 764 if ((sema[semid].sem_perm.mode & SEM_ALLOC) == 0) 765 break; 766 } 767 if (semid == seminfo.semmni) { 768 SEM_PRINTF(("no more semid_ds's available\n")); 769 error = ENOSPC; 770 goto out; 771 } 772 SEM_PRINTF(("semid %d is available\n", semid)); 773 sema[semid].sem_perm._key = key; 774 sema[semid].sem_perm.cuid = kauth_cred_geteuid(cred); 775 sema[semid].sem_perm.uid = kauth_cred_geteuid(cred); 776 sema[semid].sem_perm.cgid = kauth_cred_getegid(cred); 777 sema[semid].sem_perm.gid = kauth_cred_getegid(cred); 778 sema[semid].sem_perm.mode = (semflg & 0777) | SEM_ALLOC; 779 sema[semid].sem_perm._seq = 780 (sema[semid].sem_perm._seq + 1) & 0x7fff; 781 sema[semid].sem_nsems = nsems; 782 sema[semid].sem_otime = 0; 783 sema[semid].sem_ctime = time_second; 784 sema[semid]._sem_base = &sem[semtot]; 785 semtot += nsems; 786 memset(sema[semid]._sem_base, 0, 787 sizeof(sema[semid]._sem_base[0]) * nsems); 788 SEM_PRINTF(("sembase = %p, next = %p\n", sema[semid]._sem_base, 789 &sem[semtot])); 790 } else { 791 SEM_PRINTF(("didn't find it and wasn't asked to create it\n")); 792 error = ENOENT; 793 goto out; 794 } 795 796 found: 797 *retval = IXSEQ_TO_IPCID(semid, sema[semid].sem_perm); 798 out: 799 mutex_exit(&semlock); 800 return (error); 801 } 802 803 #define SMALL_SOPS 8 804 805 void 806 do_semop_init(void) 807 { 808 809 RUN_ONCE(&exithook_control, seminit_exithook); 810 } 811 812 /* all pointers already in kernel space */ 813 int 814 do_semop1(struct lwp *l, int usemid, struct sembuf *sops, 815 size_t nsops, struct timespec *timeout, register_t *retval) 816 { 817 struct proc *p = l->l_proc; 818 int semid, seq; 819 struct semid_ds *semaptr; 820 struct sembuf *sopptr = NULL; 821 struct __sem *semptr = NULL; 822 struct sem_undo *suptr = NULL; 823 kauth_cred_t cred = l->l_cred; 824 int timo = 0; 825 int i, error; 826 int do_wakeup, do_undos; 827 828 RUN_ONCE(&exithook_control, seminit_exithook); 829 830 SEM_PRINTF(("do_semop1(%d, %p, %zu)\n", usemid, sops, nsops)); 831 832 if (__predict_false((p->p_flag & PK_SYSVSEM) == 0)) { 833 mutex_enter(p->p_lock); 834 p->p_flag |= PK_SYSVSEM; 835 mutex_exit(p->p_lock); 836 } 837 838 restart: 839 mutex_enter(&semlock); 840 /* In case of reallocation, we will wait for completion */ 841 while (__predict_false(sem_realloc_state)) 842 cv_wait(&sem_realloc_cv, &semlock); 843 844 semid = IPCID_TO_IX(usemid); /* Convert back to zero origin */ 845 if (semid < 0 || semid >= seminfo.semmni) { 846 error = EINVAL; 847 goto out; 848 } 849 850 if (timeout) { 851 error = ts2timo(CLOCK_MONOTONIC, TIMER_RELTIME, timeout, 852 &timo, NULL); 853 if (error) 854 return error; 855 } 856 857 semaptr = &sema[semid]; 858 seq = IPCID_TO_SEQ(usemid); 859 if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0 || 860 semaptr->sem_perm._seq != seq) { 861 error = EINVAL; 862 goto out; 863 } 864 865 if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_W))) { 866 SEM_PRINTF(("error = %d from ipaccess\n", error)); 867 goto out; 868 } 869 870 for (i = 0; i < nsops; i++) 871 if (sops[i].sem_num >= semaptr->sem_nsems) { 872 error = EFBIG; 873 goto out; 874 } 875 /* 876 * Loop trying to satisfy the vector of requests. 877 * If we reach a point where we must wait, any requests already 878 * performed are rolled back and we go to sleep until some other 879 * process wakes us up. At this point, we start all over again. 880 * 881 * This ensures that from the perspective of other tasks, a set 882 * of requests is atomic (never partially satisfied). 883 */ 884 do_undos = 0; 885 886 for (;;) { 887 do_wakeup = 0; 888 889 for (i = 0; i < nsops; i++) { 890 sopptr = &sops[i]; 891 semptr = &semaptr->_sem_base[sopptr->sem_num]; 892 893 SEM_PRINTF(("semop: semaptr=%p, sem_base=%p, " 894 "semptr=%p, sem[%d]=%d : op=%d, flag=%s\n", 895 semaptr, semaptr->_sem_base, semptr, 896 sopptr->sem_num, semptr->semval, sopptr->sem_op, 897 (sopptr->sem_flg & IPC_NOWAIT) ? 898 "nowait" : "wait")); 899 900 if (sopptr->sem_op < 0) { 901 if ((int)(semptr->semval + 902 sopptr->sem_op) < 0) { 903 SEM_PRINTF(("semop: " 904 "can't do it now\n")); 905 break; 906 } else { 907 semptr->semval += sopptr->sem_op; 908 if (semptr->semval == 0 && 909 semptr->semzcnt > 0) 910 do_wakeup = 1; 911 } 912 if (sopptr->sem_flg & SEM_UNDO) 913 do_undos = 1; 914 } else if (sopptr->sem_op == 0) { 915 if (semptr->semval > 0) { 916 SEM_PRINTF(("semop: not zero now\n")); 917 break; 918 } 919 } else { 920 if (semptr->semncnt > 0) 921 do_wakeup = 1; 922 semptr->semval += sopptr->sem_op; 923 if (sopptr->sem_flg & SEM_UNDO) 924 do_undos = 1; 925 } 926 } 927 928 /* 929 * Did we get through the entire vector? 930 */ 931 if (i >= nsops) 932 goto done; 933 934 /* 935 * No ... rollback anything that we've already done 936 */ 937 SEM_PRINTF(("semop: rollback 0 through %d\n", i - 1)); 938 while (i-- > 0) 939 semaptr->_sem_base[sops[i].sem_num].semval -= 940 sops[i].sem_op; 941 942 /* 943 * If the request that we couldn't satisfy has the 944 * NOWAIT flag set then return with EAGAIN. 945 */ 946 if (sopptr->sem_flg & IPC_NOWAIT) { 947 error = EAGAIN; 948 goto out; 949 } 950 951 if (sopptr->sem_op == 0) 952 semptr->semzcnt++; 953 else 954 semptr->semncnt++; 955 956 sem_waiters++; 957 SEM_PRINTF(("semop: good night!\n")); 958 error = cv_timedwait_sig(&semcv[semid], &semlock, timo); 959 SEM_PRINTF(("semop: good morning (error=%d)!\n", error)); 960 sem_waiters--; 961 962 /* Notify reallocator, if it is waiting */ 963 cv_broadcast(&sem_realloc_cv); 964 965 /* 966 * Make sure that the semaphore still exists 967 */ 968 if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0 || 969 semaptr->sem_perm._seq != seq) { 970 error = EIDRM; 971 goto out; 972 } 973 974 /* 975 * The semaphore is still alive. Readjust the count of 976 * waiting processes. 977 */ 978 semptr = &semaptr->_sem_base[sopptr->sem_num]; 979 if (sopptr->sem_op == 0) 980 semptr->semzcnt--; 981 else 982 semptr->semncnt--; 983 984 /* In case of such state, restart the call */ 985 if (sem_realloc_state) { 986 mutex_exit(&semlock); 987 goto restart; 988 } 989 990 /* Is it really morning, or was our sleep interrupted? */ 991 if (error != 0) { 992 if (error == ERESTART) 993 error = EINTR; // Simplify to just EINTR 994 else if (error == EWOULDBLOCK) 995 error = EAGAIN; // Convert timeout to EAGAIN 996 goto out; 997 } 998 SEM_PRINTF(("semop: good morning!\n")); 999 } 1000 done: 1001 /* 1002 * Process any SEM_UNDO requests. 1003 */ 1004 if (do_undos) { 1005 for (i = 0; i < nsops; i++) { 1006 /* 1007 * We only need to deal with SEM_UNDO's for non-zero 1008 * op's. 1009 */ 1010 int adjval; 1011 1012 if ((sops[i].sem_flg & SEM_UNDO) == 0) 1013 continue; 1014 adjval = sops[i].sem_op; 1015 if (adjval == 0) 1016 continue; 1017 error = semundo_adjust(p, &suptr, semid, 1018 sops[i].sem_num, -adjval); 1019 if (error == 0) 1020 continue; 1021 1022 /* 1023 * Oh-Oh! We ran out of either sem_undo's or undo's. 1024 * Rollback the adjustments to this point and then 1025 * rollback the semaphore ups and down so we can return 1026 * with an error with all structures restored. We 1027 * rollback the undo's in the exact reverse order that 1028 * we applied them. This guarantees that we won't run 1029 * out of space as we roll things back out. 1030 */ 1031 while (i-- > 0) { 1032 if ((sops[i].sem_flg & SEM_UNDO) == 0) 1033 continue; 1034 adjval = sops[i].sem_op; 1035 if (adjval == 0) 1036 continue; 1037 if (semundo_adjust(p, &suptr, semid, 1038 sops[i].sem_num, adjval) != 0) 1039 panic("semop - can't undo undos"); 1040 } 1041 1042 for (i = 0; i < nsops; i++) 1043 semaptr->_sem_base[sops[i].sem_num].semval -= 1044 sops[i].sem_op; 1045 1046 SEM_PRINTF(("error = %d from semundo_adjust\n", 1047 error)); 1048 goto out; 1049 } /* loop through the sops */ 1050 } /* if (do_undos) */ 1051 1052 /* We're definitely done - set the sempid's */ 1053 for (i = 0; i < nsops; i++) { 1054 sopptr = &sops[i]; 1055 semptr = &semaptr->_sem_base[sopptr->sem_num]; 1056 semptr->sempid = p->p_pid; 1057 } 1058 1059 /* Update sem_otime */ 1060 semaptr->sem_otime = time_second; 1061 1062 /* Do a wakeup if any semaphore was up'd. */ 1063 if (do_wakeup) { 1064 SEM_PRINTF(("semop: doing wakeup\n")); 1065 cv_broadcast(&semcv[semid]); 1066 SEM_PRINTF(("semop: back from wakeup\n")); 1067 } 1068 SEM_PRINTF(("semop: done\n")); 1069 *retval = 0; 1070 1071 out: 1072 mutex_exit(&semlock); 1073 return error; 1074 } 1075 1076 static int 1077 do_semop(struct lwp *l, int usemid, struct sembuf *usops, 1078 size_t nsops, struct timespec *utimeout, register_t *retval) 1079 { 1080 struct sembuf small_sops[SMALL_SOPS]; 1081 struct sembuf *sops; 1082 struct timespec timeout; 1083 int error; 1084 1085 do_semop_init(); 1086 1087 SEM_PRINTF(("do_semop(%d, %p, %zu)\n", usemid, usops, nsops)); 1088 1089 if (nsops <= SMALL_SOPS) { 1090 sops = small_sops; 1091 } else if (seminfo.semopm > 0 && nsops <= (size_t)seminfo.semopm) { 1092 sops = kmem_alloc(nsops * sizeof(*sops), KM_SLEEP); 1093 } else { 1094 SEM_PRINTF(("too many sops (max=%d, nsops=%zu)\n", 1095 seminfo.semopm, nsops)); 1096 return (E2BIG); 1097 } 1098 1099 error = copyin(usops, sops, nsops * sizeof(sops[0])); 1100 if (error) { 1101 SEM_PRINTF(("error = %d from copyin(%p, %p, %zu)\n", error, 1102 usops, &sops, nsops * sizeof(sops[0]))); 1103 if (sops != small_sops) 1104 kmem_free(sops, nsops * sizeof(*sops)); 1105 return error; 1106 } 1107 1108 if (utimeout) { 1109 error = copyin(utimeout, &timeout, sizeof(timeout)); 1110 if (error) { 1111 SEM_PRINTF(("error = %d from copyin(%p, %p, %zu)\n", 1112 error, utimeout, &timeout, sizeof(timeout))); 1113 return error; 1114 } 1115 } 1116 1117 error = do_semop1(l, usemid, sops, nsops, utimeout ? &timeout : NULL, 1118 retval); 1119 1120 if (sops != small_sops) 1121 kmem_free(sops, nsops * sizeof(*sops)); 1122 1123 return error; 1124 } 1125 1126 int 1127 sys_semtimedop(struct lwp *l, const struct sys_semtimedop_args *uap, 1128 register_t *retval) 1129 { 1130 /* { 1131 syscallarg(int) semid; 1132 syscallarg(struct sembuf *) sops; 1133 syscallarg(size_t) nsops; 1134 syscallarg(struct timespec) timeout; 1135 } */ 1136 int semid = SCARG(uap, semid); 1137 struct sembuf *sops = SCARG(uap, sops); 1138 size_t nsops = SCARG(uap, nsops); 1139 struct timespec *utimeout = SCARG(uap, timeout); 1140 1141 return do_semop(l, semid, sops, nsops, utimeout, retval); 1142 } 1143 1144 int 1145 sys_semop(struct lwp *l, const struct sys_semop_args *uap, register_t *retval) 1146 { 1147 /* { 1148 syscallarg(int) semid; 1149 syscallarg(struct sembuf *) sops; 1150 syscallarg(size_t) nsops; 1151 } */ 1152 int semid = SCARG(uap, semid); 1153 struct sembuf *sops = SCARG(uap, sops); 1154 size_t nsops = SCARG(uap, nsops); 1155 1156 return do_semop(l, semid, sops, nsops, NULL, retval); 1157 } 1158 1159 /* 1160 * Go through the undo structures for this process and apply the 1161 * adjustments to semaphores. 1162 */ 1163 /*ARGSUSED*/ 1164 void 1165 semexit(struct proc *p, void *v) 1166 { 1167 struct sem_undo *suptr; 1168 struct sem_undo **supptr; 1169 1170 if ((p->p_flag & PK_SYSVSEM) == 0) 1171 return; 1172 1173 mutex_enter(&semlock); 1174 1175 /* 1176 * Go through the chain of undo vectors looking for one 1177 * associated with this process. 1178 */ 1179 1180 for (supptr = &semu_list; (suptr = *supptr) != NULL; 1181 supptr = &suptr->un_next) { 1182 if (suptr->un_proc == p) 1183 break; 1184 } 1185 1186 /* 1187 * If there is no undo vector, skip to the end. 1188 */ 1189 1190 if (suptr == NULL) { 1191 mutex_exit(&semlock); 1192 return; 1193 } 1194 1195 /* 1196 * We now have an undo vector for this process. 1197 */ 1198 1199 SEM_PRINTF(("proc @%p has undo structure with %d entries\n", p, 1200 suptr->un_cnt)); 1201 1202 /* 1203 * If there are any active undo elements then process them. 1204 */ 1205 if (suptr->un_cnt > 0) { 1206 int ix; 1207 1208 for (ix = 0; ix < suptr->un_cnt; ix++) { 1209 int semid = suptr->un_ent[ix].un_id; 1210 int semnum = suptr->un_ent[ix].un_num; 1211 int adjval = suptr->un_ent[ix].un_adjval; 1212 struct semid_ds *semaptr; 1213 1214 semaptr = &sema[semid]; 1215 if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0) 1216 if (semnum >= semaptr->sem_nsems) 1217 panic("semexit - semnum out of range"); 1218 1219 SEM_PRINTF(("semexit: %p id=%d num=%d(adj=%d) ; " 1220 "sem=%d\n", 1221 suptr->un_proc, suptr->un_ent[ix].un_id, 1222 suptr->un_ent[ix].un_num, 1223 suptr->un_ent[ix].un_adjval, 1224 semaptr->_sem_base[semnum].semval)); 1225 1226 if (adjval < 0 && 1227 semaptr->_sem_base[semnum].semval < -adjval) 1228 semaptr->_sem_base[semnum].semval = 0; 1229 else 1230 semaptr->_sem_base[semnum].semval += adjval; 1231 1232 cv_broadcast(&semcv[semid]); 1233 SEM_PRINTF(("semexit: back from wakeup\n")); 1234 } 1235 } 1236 1237 /* 1238 * Deallocate the undo vector. 1239 */ 1240 SEM_PRINTF(("removing vector\n")); 1241 suptr->un_proc = NULL; 1242 *supptr = suptr->un_next; 1243 mutex_exit(&semlock); 1244 } 1245 1246 /* 1247 * Sysctl initialization and nodes. 1248 */ 1249 1250 static int 1251 sysctl_ipc_semmni(SYSCTLFN_ARGS) 1252 { 1253 int newsize, error; 1254 struct sysctlnode node; 1255 node = *rnode; 1256 node.sysctl_data = &newsize; 1257 1258 newsize = seminfo.semmni; 1259 error = sysctl_lookup(SYSCTLFN_CALL(&node)); 1260 if (error || newp == NULL) 1261 return error; 1262 1263 return semrealloc(newsize, seminfo.semmns, seminfo.semmnu); 1264 } 1265 1266 static int 1267 sysctl_ipc_semmns(SYSCTLFN_ARGS) 1268 { 1269 int newsize, error; 1270 struct sysctlnode node; 1271 node = *rnode; 1272 node.sysctl_data = &newsize; 1273 1274 newsize = seminfo.semmns; 1275 error = sysctl_lookup(SYSCTLFN_CALL(&node)); 1276 if (error || newp == NULL) 1277 return error; 1278 1279 return semrealloc(seminfo.semmni, newsize, seminfo.semmnu); 1280 } 1281 1282 static int 1283 sysctl_ipc_semmnu(SYSCTLFN_ARGS) 1284 { 1285 int newsize, error; 1286 struct sysctlnode node; 1287 node = *rnode; 1288 node.sysctl_data = &newsize; 1289 1290 newsize = seminfo.semmnu; 1291 error = sysctl_lookup(SYSCTLFN_CALL(&node)); 1292 if (error || newp == NULL) 1293 return error; 1294 1295 return semrealloc(seminfo.semmni, seminfo.semmns, newsize); 1296 } 1297 1298 SYSCTL_SETUP(sysctl_ipc_sem_setup, "sysctl kern.ipc subtree setup") 1299 { 1300 const struct sysctlnode *node = NULL; 1301 1302 sysctl_createv(clog, 0, NULL, &node, 1303 CTLFLAG_PERMANENT, 1304 CTLTYPE_NODE, "ipc", 1305 SYSCTL_DESCR("SysV IPC options"), 1306 NULL, 0, NULL, 0, 1307 CTL_KERN, KERN_SYSVIPC, CTL_EOL); 1308 1309 if (node == NULL) 1310 return; 1311 1312 sysctl_createv(clog, 0, &node, NULL, 1313 CTLFLAG_PERMANENT | CTLFLAG_READWRITE, 1314 CTLTYPE_INT, "semmni", 1315 SYSCTL_DESCR("Max number of number of semaphore identifiers"), 1316 sysctl_ipc_semmni, 0, &seminfo.semmni, 0, 1317 CTL_CREATE, CTL_EOL); 1318 sysctl_createv(clog, 0, &node, NULL, 1319 CTLFLAG_PERMANENT | CTLFLAG_READWRITE, 1320 CTLTYPE_INT, "semmns", 1321 SYSCTL_DESCR("Max number of number of semaphores in system"), 1322 sysctl_ipc_semmns, 0, &seminfo.semmns, 0, 1323 CTL_CREATE, CTL_EOL); 1324 sysctl_createv(clog, 0, &node, NULL, 1325 CTLFLAG_PERMANENT | CTLFLAG_READWRITE, 1326 CTLTYPE_INT, "semmnu", 1327 SYSCTL_DESCR("Max number of undo structures in system"), 1328 sysctl_ipc_semmnu, 0, &seminfo.semmnu, 0, 1329 CTL_CREATE, CTL_EOL); 1330 } 1331
Indexes created Sat Sep 20 22:09:52 GMT 2025