1 /* $NetBSD: sysv_sem.c,v 1.86 2009/10/05 23:46:02 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 1.86 2009/10/05 23:46:02 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; 63 static struct __sem *sem; 64 static kcondvar_t *semcv; 65 static int *semu; 66 67 static kmutex_t semlock; 68 static struct sem_undo *semu_list; /* list of active undo structures */ 69 static u_int semtot = 0; /* total number of semaphores */ 70 71 static u_int sem_waiters = 0; /* total number of semop waiters */ 72 static bool sem_realloc_state; 73 static kcondvar_t sem_realloc_cv; 74 75 /* Macro to find a particular sem_undo vector */ 76 #define SEMU(s, ix) ((struct sem_undo *)(((long)s) + ix * seminfo.semusz)) 77 78 #ifdef SEM_DEBUG 79 #define SEM_PRINTF(a) printf a 80 #else 81 #define SEM_PRINTF(a) 82 #endif 83 84 struct sem_undo *semu_alloc(struct proc *); 85 int semundo_adjust(struct proc *, struct sem_undo **, int, int, int); 86 void semundo_clear(int, int); 87 88 void 89 seminit(void) 90 { 91 int i, sz; 92 vaddr_t v; 93 94 mutex_init(&semlock, MUTEX_DEFAULT, IPL_NONE); 95 cv_init(&sem_realloc_cv, "semrealc"); 96 sem_realloc_state = false; 97 98 /* Allocate the wired memory for our structures */ 99 sz = ALIGN(seminfo.semmni * sizeof(struct semid_ds)) + 100 ALIGN(seminfo.semmns * sizeof(struct __sem)) + 101 ALIGN(seminfo.semmni * sizeof(kcondvar_t)) + 102 ALIGN(seminfo.semmnu * seminfo.semusz); 103 v = uvm_km_alloc(kernel_map, round_page(sz), 0, 104 UVM_KMF_WIRED|UVM_KMF_ZERO); 105 if (v == 0) 106 panic("sysv_sem: cannot allocate memory"); 107 sema = (void *)v; 108 sem = (void *)((uintptr_t)sema + 109 ALIGN(seminfo.semmni * sizeof(struct semid_ds))); 110 semcv = (void *)((uintptr_t)sem + 111 ALIGN(seminfo.semmns * sizeof(struct __sem))); 112 semu = (void *)((uintptr_t)semcv + 113 ALIGN(seminfo.semmni * sizeof(kcondvar_t))); 114 115 for (i = 0; i < seminfo.semmni; i++) { 116 sema[i]._sem_base = 0; 117 sema[i].sem_perm.mode = 0; 118 cv_init(&semcv[i], "semwait"); 119 } 120 for (i = 0; i < seminfo.semmnu; i++) { 121 struct sem_undo *suptr = SEMU(semu, i); 122 suptr->un_proc = NULL; 123 } 124 semu_list = NULL; 125 exithook_establish(semexit, NULL); 126 } 127 128 static int 129 semrealloc(int newsemmni, int newsemmns, int newsemmnu) 130 { 131 struct semid_ds *new_sema, *old_sema; 132 struct __sem *new_sem; 133 struct sem_undo *new_semu_list, *suptr, *nsuptr; 134 int *new_semu; 135 kcondvar_t *new_semcv; 136 vaddr_t v; 137 int i, j, lsemid, nmnus, sz; 138 139 if (newsemmni < 1 || newsemmns < 1 || newsemmnu < 1) 140 return EINVAL; 141 142 /* Allocate the wired memory for our structures */ 143 sz = ALIGN(newsemmni * sizeof(struct semid_ds)) + 144 ALIGN(newsemmns * sizeof(struct __sem)) + 145 ALIGN(newsemmni * sizeof(kcondvar_t)) + 146 ALIGN(newsemmnu * seminfo.semusz); 147 v = uvm_km_alloc(kernel_map, round_page(sz), 0, 148 UVM_KMF_WIRED|UVM_KMF_ZERO); 149 if (v == 0) 150 return ENOMEM; 151 152 mutex_enter(&semlock); 153 if (sem_realloc_state) { 154 mutex_exit(&semlock); 155 uvm_km_free(kernel_map, v, sz, UVM_KMF_WIRED); 156 return EBUSY; 157 } 158 sem_realloc_state = true; 159 if (sem_waiters) { 160 /* 161 * Mark reallocation state, wake-up all waiters, 162 * and wait while they will all exit. 163 */ 164 for (i = 0; i < seminfo.semmni; i++) 165 cv_broadcast(&semcv[i]); 166 while (sem_waiters) 167 cv_wait(&sem_realloc_cv, &semlock); 168 } 169 old_sema = sema; 170 171 /* Get the number of last slot */ 172 lsemid = 0; 173 for (i = 0; i < seminfo.semmni; i++) 174 if (sema[i].sem_perm.mode & SEM_ALLOC) 175 lsemid = i; 176 177 /* Get the number of currently used undo structures */ 178 nmnus = 0; 179 for (i = 0; i < seminfo.semmnu; i++) { 180 suptr = SEMU(semu, i); 181 if (suptr->un_proc == NULL) 182 continue; 183 nmnus++; 184 } 185 186 /* We cannot reallocate less memory than we use */ 187 if (lsemid >= newsemmni || semtot > newsemmns || nmnus > newsemmnu) { 188 mutex_exit(&semlock); 189 uvm_km_free(kernel_map, v, sz, UVM_KMF_WIRED); 190 return EBUSY; 191 } 192 193 new_sema = (void *)v; 194 new_sem = (void *)((uintptr_t)new_sema + 195 ALIGN(newsemmni * sizeof(struct semid_ds))); 196 new_semcv = (void *)((uintptr_t)new_sem + 197 ALIGN(newsemmns * sizeof(struct __sem))); 198 new_semu = (void *)((uintptr_t)new_semcv + 199 ALIGN(newsemmni * sizeof(kcondvar_t))); 200 201 /* Initialize all semaphore identifiers and condvars */ 202 for (i = 0; i < newsemmni; i++) { 203 new_sema[i]._sem_base = 0; 204 new_sema[i].sem_perm.mode = 0; 205 cv_init(&new_semcv[i], "semwait"); 206 } 207 for (i = 0; i < newsemmnu; i++) { 208 nsuptr = SEMU(new_semu, i); 209 nsuptr->un_proc = NULL; 210 } 211 212 /* 213 * Copy all identifiers, semaphores and list of the 214 * undo structures to the new memory allocation. 215 */ 216 j = 0; 217 for (i = 0; i <= lsemid; i++) { 218 if ((sema[i].sem_perm.mode & SEM_ALLOC) == 0) 219 continue; 220 memcpy(&new_sema[i], &sema[i], sizeof(struct semid_ds)); 221 new_sema[i]._sem_base = &new_sem[j]; 222 memcpy(new_sema[i]._sem_base, sema[i]._sem_base, 223 (sizeof(struct __sem) * sema[i].sem_nsems)); 224 j += sema[i].sem_nsems; 225 } 226 KASSERT(j == semtot); 227 228 j = 0; 229 new_semu_list = NULL; 230 for (suptr = semu_list; suptr != NULL; suptr = suptr->un_next) { 231 KASSERT(j < newsemmnu); 232 nsuptr = SEMU(new_semu, j); 233 memcpy(nsuptr, suptr, SEMUSZ); 234 nsuptr->un_next = new_semu_list; 235 new_semu_list = nsuptr; 236 j++; 237 } 238 239 for (i = 0; i < seminfo.semmni; i++) { 240 KASSERT(cv_has_waiters(&semcv[i]) == false); 241 cv_destroy(&semcv[i]); 242 } 243 244 sz = ALIGN(seminfo.semmni * sizeof(struct semid_ds)) + 245 ALIGN(seminfo.semmns * sizeof(struct __sem)) + 246 ALIGN(seminfo.semmni * sizeof(kcondvar_t)) + 247 ALIGN(seminfo.semmnu * seminfo.semusz); 248 249 /* Set the pointers and update the new values */ 250 sema = new_sema; 251 sem = new_sem; 252 semcv = new_semcv; 253 semu = new_semu; 254 semu_list = new_semu_list; 255 256 seminfo.semmni = newsemmni; 257 seminfo.semmns = newsemmns; 258 seminfo.semmnu = newsemmnu; 259 260 /* Reallocation completed - notify all waiters, if any */ 261 sem_realloc_state = false; 262 cv_broadcast(&sem_realloc_cv); 263 mutex_exit(&semlock); 264 265 uvm_km_free(kernel_map, (vaddr_t)old_sema, sz, UVM_KMF_WIRED); 266 return 0; 267 } 268 269 /* 270 * Placebo. 271 */ 272 273 int 274 sys_semconfig(struct lwp *l, const struct sys_semconfig_args *uap, register_t *retval) 275 { 276 277 *retval = 0; 278 return 0; 279 } 280 281 /* 282 * Allocate a new sem_undo structure for a process. 283 * => Returns NULL on failure. 284 */ 285 struct sem_undo * 286 semu_alloc(struct proc *p) 287 { 288 struct sem_undo *suptr, **supptr; 289 bool attempted = false; 290 int i; 291 292 KASSERT(mutex_owned(&semlock)); 293 again: 294 /* Look for a free structure. */ 295 for (i = 0; i < seminfo.semmnu; i++) { 296 suptr = SEMU(semu, i); 297 if (suptr->un_proc == NULL) { 298 /* Found. Fill it in and return. */ 299 suptr->un_next = semu_list; 300 semu_list = suptr; 301 suptr->un_cnt = 0; 302 suptr->un_proc = p; 303 return suptr; 304 } 305 } 306 307 /* Not found. Attempt to free some structures. */ 308 if (!attempted) { 309 bool freed = false; 310 311 attempted = true; 312 supptr = &semu_list; 313 while ((suptr = *supptr) != NULL) { 314 if (suptr->un_cnt == 0) { 315 suptr->un_proc = NULL; 316 *supptr = suptr->un_next; 317 freed = true; 318 } else { 319 supptr = &suptr->un_next; 320 } 321 } 322 if (freed) { 323 goto again; 324 } 325 } 326 return NULL; 327 } 328 329 /* 330 * Adjust a particular entry for a particular proc 331 */ 332 333 int 334 semundo_adjust(struct proc *p, struct sem_undo **supptr, int semid, int semnum, 335 int adjval) 336 { 337 struct sem_undo *suptr; 338 struct undo *sunptr; 339 int i; 340 341 KASSERT(mutex_owned(&semlock)); 342 343 /* 344 * Look for and remember the sem_undo if the caller doesn't 345 * provide it 346 */ 347 348 suptr = *supptr; 349 if (suptr == NULL) { 350 for (suptr = semu_list; suptr != NULL; suptr = suptr->un_next) 351 if (suptr->un_proc == p) 352 break; 353 354 if (suptr == NULL) { 355 suptr = semu_alloc(p); 356 if (suptr == NULL) 357 return (ENOSPC); 358 } 359 *supptr = suptr; 360 } 361 362 /* 363 * Look for the requested entry and adjust it (delete if 364 * adjval becomes 0). 365 */ 366 sunptr = &suptr->un_ent[0]; 367 for (i = 0; i < suptr->un_cnt; i++, sunptr++) { 368 if (sunptr->un_id != semid || sunptr->un_num != semnum) 369 continue; 370 sunptr->un_adjval += adjval; 371 if (sunptr->un_adjval == 0) { 372 suptr->un_cnt--; 373 if (i < suptr->un_cnt) 374 suptr->un_ent[i] = 375 suptr->un_ent[suptr->un_cnt]; 376 } 377 return (0); 378 } 379 380 /* Didn't find the right entry - create it */ 381 if (suptr->un_cnt == SEMUME) 382 return (EINVAL); 383 384 sunptr = &suptr->un_ent[suptr->un_cnt]; 385 suptr->un_cnt++; 386 sunptr->un_adjval = adjval; 387 sunptr->un_id = semid; 388 sunptr->un_num = semnum; 389 return (0); 390 } 391 392 void 393 semundo_clear(int semid, int semnum) 394 { 395 struct sem_undo *suptr; 396 struct undo *sunptr, *sunend; 397 398 KASSERT(mutex_owned(&semlock)); 399 400 for (suptr = semu_list; suptr != NULL; suptr = suptr->un_next) 401 for (sunptr = &suptr->un_ent[0], 402 sunend = sunptr + suptr->un_cnt; sunptr < sunend;) { 403 if (sunptr->un_id == semid) { 404 if (semnum == -1 || sunptr->un_num == semnum) { 405 suptr->un_cnt--; 406 sunend--; 407 if (sunptr != sunend) 408 *sunptr = *sunend; 409 if (semnum != -1) 410 break; 411 else 412 continue; 413 } 414 } 415 sunptr++; 416 } 417 } 418 419 int 420 sys_____semctl50(struct lwp *l, const struct sys_____semctl50_args *uap, 421 register_t *retval) 422 { 423 /* { 424 syscallarg(int) semid; 425 syscallarg(int) semnum; 426 syscallarg(int) cmd; 427 syscallarg(union __semun *) arg; 428 } */ 429 struct semid_ds sembuf; 430 int cmd, error; 431 void *pass_arg; 432 union __semun karg; 433 434 cmd = SCARG(uap, cmd); 435 436 pass_arg = get_semctl_arg(cmd, &sembuf, &karg); 437 438 if (pass_arg) { 439 error = copyin(SCARG(uap, arg), &karg, sizeof(karg)); 440 if (error) 441 return error; 442 if (cmd == IPC_SET) { 443 error = copyin(karg.buf, &sembuf, sizeof(sembuf)); 444 if (error) 445 return (error); 446 } 447 } 448 449 error = semctl1(l, SCARG(uap, semid), SCARG(uap, semnum), cmd, 450 pass_arg, retval); 451 452 if (error == 0 && cmd == IPC_STAT) 453 error = copyout(&sembuf, karg.buf, sizeof(sembuf)); 454 455 return (error); 456 } 457 458 int 459 semctl1(struct lwp *l, int semid, int semnum, int cmd, void *v, 460 register_t *retval) 461 { 462 kauth_cred_t cred = l->l_cred; 463 union __semun *arg = v; 464 struct semid_ds *sembuf = v, *semaptr; 465 int i, error, ix; 466 467 SEM_PRINTF(("call to semctl(%d, %d, %d, %p)\n", 468 semid, semnum, cmd, v)); 469 470 mutex_enter(&semlock); 471 472 ix = IPCID_TO_IX(semid); 473 if (ix < 0 || ix >= seminfo.semmni) { 474 mutex_exit(&semlock); 475 return (EINVAL); 476 } 477 478 semaptr = &sema[ix]; 479 if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0 || 480 semaptr->sem_perm._seq != IPCID_TO_SEQ(semid)) { 481 mutex_exit(&semlock); 482 return (EINVAL); 483 } 484 485 switch (cmd) { 486 case IPC_RMID: 487 if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_M)) != 0) 488 break; 489 semaptr->sem_perm.cuid = kauth_cred_geteuid(cred); 490 semaptr->sem_perm.uid = kauth_cred_geteuid(cred); 491 semtot -= semaptr->sem_nsems; 492 for (i = semaptr->_sem_base - sem; i < semtot; i++) 493 sem[i] = sem[i + semaptr->sem_nsems]; 494 for (i = 0; i < seminfo.semmni; i++) { 495 if ((sema[i].sem_perm.mode & SEM_ALLOC) && 496 sema[i]._sem_base > semaptr->_sem_base) 497 sema[i]._sem_base -= semaptr->sem_nsems; 498 } 499 semaptr->sem_perm.mode = 0; 500 semundo_clear(ix, -1); 501 cv_broadcast(&semcv[ix]); 502 break; 503 504 case IPC_SET: 505 if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_M))) 506 break; 507 KASSERT(sembuf != NULL); 508 semaptr->sem_perm.uid = sembuf->sem_perm.uid; 509 semaptr->sem_perm.gid = sembuf->sem_perm.gid; 510 semaptr->sem_perm.mode = (semaptr->sem_perm.mode & ~0777) | 511 (sembuf->sem_perm.mode & 0777); 512 semaptr->sem_ctime = time_second; 513 break; 514 515 case IPC_STAT: 516 if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_R))) 517 break; 518 KASSERT(sembuf != NULL); 519 memcpy(sembuf, semaptr, sizeof(struct semid_ds)); 520 sembuf->sem_perm.mode &= 0777; 521 break; 522 523 case GETNCNT: 524 if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_R))) 525 break; 526 if (semnum < 0 || semnum >= semaptr->sem_nsems) { 527 error = EINVAL; 528 break; 529 } 530 *retval = semaptr->_sem_base[semnum].semncnt; 531 break; 532 533 case GETPID: 534 if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_R))) 535 break; 536 if (semnum < 0 || semnum >= semaptr->sem_nsems) { 537 error = EINVAL; 538 break; 539 } 540 *retval = semaptr->_sem_base[semnum].sempid; 541 break; 542 543 case GETVAL: 544 if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_R))) 545 break; 546 if (semnum < 0 || semnum >= semaptr->sem_nsems) { 547 error = EINVAL; 548 break; 549 } 550 *retval = semaptr->_sem_base[semnum].semval; 551 break; 552 553 case GETALL: 554 if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_R))) 555 break; 556 KASSERT(arg != NULL); 557 for (i = 0; i < semaptr->sem_nsems; i++) { 558 error = copyout(&semaptr->_sem_base[i].semval, 559 &arg->array[i], sizeof(arg->array[i])); 560 if (error != 0) 561 break; 562 } 563 break; 564 565 case GETZCNT: 566 if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_R))) 567 break; 568 if (semnum < 0 || semnum >= semaptr->sem_nsems) { 569 error = EINVAL; 570 break; 571 } 572 *retval = semaptr->_sem_base[semnum].semzcnt; 573 break; 574 575 case SETVAL: 576 if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_W))) 577 break; 578 if (semnum < 0 || semnum >= semaptr->sem_nsems) { 579 error = EINVAL; 580 break; 581 } 582 KASSERT(arg != NULL); 583 if ((unsigned int)arg->val > seminfo.semvmx) { 584 error = ERANGE; 585 break; 586 } 587 semaptr->_sem_base[semnum].semval = arg->val; 588 semundo_clear(ix, semnum); 589 cv_broadcast(&semcv[ix]); 590 break; 591 592 case SETALL: 593 if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_W))) 594 break; 595 KASSERT(arg != NULL); 596 for (i = 0; i < semaptr->sem_nsems; i++) { 597 unsigned short semval; 598 error = copyin(&arg->array[i], &semval, 599 sizeof(arg->array[i])); 600 if (error != 0) 601 break; 602 if ((unsigned int)semval > seminfo.semvmx) { 603 error = ERANGE; 604 break; 605 } 606 semaptr->_sem_base[i].semval = semval; 607 } 608 semundo_clear(ix, -1); 609 cv_broadcast(&semcv[ix]); 610 break; 611 612 default: 613 error = EINVAL; 614 break; 615 } 616 617 mutex_exit(&semlock); 618 return (error); 619 } 620 621 int 622 sys_semget(struct lwp *l, const struct sys_semget_args *uap, register_t *retval) 623 { 624 /* { 625 syscallarg(key_t) key; 626 syscallarg(int) nsems; 627 syscallarg(int) semflg; 628 } */ 629 int semid, error = 0; 630 int key = SCARG(uap, key); 631 int nsems = SCARG(uap, nsems); 632 int semflg = SCARG(uap, semflg); 633 kauth_cred_t cred = l->l_cred; 634 635 SEM_PRINTF(("semget(0x%x, %d, 0%o)\n", key, nsems, semflg)); 636 637 mutex_enter(&semlock); 638 639 if (key != IPC_PRIVATE) { 640 for (semid = 0; semid < seminfo.semmni; semid++) { 641 if ((sema[semid].sem_perm.mode & SEM_ALLOC) && 642 sema[semid].sem_perm._key == key) 643 break; 644 } 645 if (semid < seminfo.semmni) { 646 SEM_PRINTF(("found public key\n")); 647 if ((error = ipcperm(cred, &sema[semid].sem_perm, 648 semflg & 0700))) 649 goto out; 650 if (nsems > 0 && sema[semid].sem_nsems < nsems) { 651 SEM_PRINTF(("too small\n")); 652 error = EINVAL; 653 goto out; 654 } 655 if ((semflg & IPC_CREAT) && (semflg & IPC_EXCL)) { 656 SEM_PRINTF(("not exclusive\n")); 657 error = EEXIST; 658 goto out; 659 } 660 goto found; 661 } 662 } 663 664 SEM_PRINTF(("need to allocate the semid_ds\n")); 665 if (key == IPC_PRIVATE || (semflg & IPC_CREAT)) { 666 if (nsems <= 0 || nsems > seminfo.semmsl) { 667 SEM_PRINTF(("nsems out of range (0<%d<=%d)\n", nsems, 668 seminfo.semmsl)); 669 error = EINVAL; 670 goto out; 671 } 672 if (nsems > seminfo.semmns - semtot) { 673 SEM_PRINTF(("not enough semaphores left " 674 "(need %d, got %d)\n", 675 nsems, seminfo.semmns - semtot)); 676 error = ENOSPC; 677 goto out; 678 } 679 for (semid = 0; semid < seminfo.semmni; semid++) { 680 if ((sema[semid].sem_perm.mode & SEM_ALLOC) == 0) 681 break; 682 } 683 if (semid == seminfo.semmni) { 684 SEM_PRINTF(("no more semid_ds's available\n")); 685 error = ENOSPC; 686 goto out; 687 } 688 SEM_PRINTF(("semid %d is available\n", semid)); 689 sema[semid].sem_perm._key = key; 690 sema[semid].sem_perm.cuid = kauth_cred_geteuid(cred); 691 sema[semid].sem_perm.uid = kauth_cred_geteuid(cred); 692 sema[semid].sem_perm.cgid = kauth_cred_getegid(cred); 693 sema[semid].sem_perm.gid = kauth_cred_getegid(cred); 694 sema[semid].sem_perm.mode = (semflg & 0777) | SEM_ALLOC; 695 sema[semid].sem_perm._seq = 696 (sema[semid].sem_perm._seq + 1) & 0x7fff; 697 sema[semid].sem_nsems = nsems; 698 sema[semid].sem_otime = 0; 699 sema[semid].sem_ctime = time_second; 700 sema[semid]._sem_base = &sem[semtot]; 701 semtot += nsems; 702 memset(sema[semid]._sem_base, 0, 703 sizeof(sema[semid]._sem_base[0]) * nsems); 704 SEM_PRINTF(("sembase = %p, next = %p\n", sema[semid]._sem_base, 705 &sem[semtot])); 706 } else { 707 SEM_PRINTF(("didn't find it and wasn't asked to create it\n")); 708 error = ENOENT; 709 goto out; 710 } 711 712 found: 713 *retval = IXSEQ_TO_IPCID(semid, sema[semid].sem_perm); 714 out: 715 mutex_exit(&semlock); 716 return (error); 717 } 718 719 #define SMALL_SOPS 8 720 721 int 722 sys_semop(struct lwp *l, const struct sys_semop_args *uap, register_t *retval) 723 { 724 /* { 725 syscallarg(int) semid; 726 syscallarg(struct sembuf *) sops; 727 syscallarg(size_t) nsops; 728 } */ 729 struct proc *p = l->l_proc; 730 int semid = SCARG(uap, semid), seq; 731 size_t nsops = SCARG(uap, nsops); 732 struct sembuf small_sops[SMALL_SOPS]; 733 struct sembuf *sops; 734 struct semid_ds *semaptr; 735 struct sembuf *sopptr = NULL; 736 struct __sem *semptr = NULL; 737 struct sem_undo *suptr = NULL; 738 kauth_cred_t cred = l->l_cred; 739 int i, error; 740 int do_wakeup, do_undos; 741 742 SEM_PRINTF(("call to semop(%d, %p, %zd)\n", semid, SCARG(uap,sops), nsops)); 743 744 if (__predict_false((p->p_flag & PK_SYSVSEM) == 0)) { 745 mutex_enter(p->p_lock); 746 p->p_flag |= PK_SYSVSEM; 747 mutex_exit(p->p_lock); 748 } 749 750 restart: 751 if (nsops <= SMALL_SOPS) { 752 sops = small_sops; 753 } else if (nsops <= seminfo.semopm) { 754 sops = kmem_alloc(nsops * sizeof(*sops), KM_SLEEP); 755 } else { 756 SEM_PRINTF(("too many sops (max=%d, nsops=%zd)\n", 757 seminfo.semopm, nsops)); 758 return (E2BIG); 759 } 760 761 error = copyin(SCARG(uap, sops), sops, nsops * sizeof(sops[0])); 762 if (error) { 763 SEM_PRINTF(("error = %d from copyin(%p, %p, %zd)\n", error, 764 SCARG(uap, sops), &sops, nsops * sizeof(sops[0]))); 765 if (sops != small_sops) 766 kmem_free(sops, nsops * sizeof(*sops)); 767 return error; 768 } 769 770 mutex_enter(&semlock); 771 /* In case of reallocation, we will wait for completion */ 772 while (__predict_false(sem_realloc_state)) 773 cv_wait(&sem_realloc_cv, &semlock); 774 775 semid = IPCID_TO_IX(semid); /* Convert back to zero origin */ 776 if (semid < 0 || semid >= seminfo.semmni) { 777 error = EINVAL; 778 goto out; 779 } 780 781 semaptr = &sema[semid]; 782 seq = IPCID_TO_SEQ(SCARG(uap, semid)); 783 if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0 || 784 semaptr->sem_perm._seq != seq) { 785 error = EINVAL; 786 goto out; 787 } 788 789 if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_W))) { 790 SEM_PRINTF(("error = %d from ipaccess\n", error)); 791 goto out; 792 } 793 794 for (i = 0; i < nsops; i++) 795 if (sops[i].sem_num >= semaptr->sem_nsems) { 796 error = EFBIG; 797 goto out; 798 } 799 800 /* 801 * Loop trying to satisfy the vector of requests. 802 * If we reach a point where we must wait, any requests already 803 * performed are rolled back and we go to sleep until some other 804 * process wakes us up. At this point, we start all over again. 805 * 806 * This ensures that from the perspective of other tasks, a set 807 * of requests is atomic (never partially satisfied). 808 */ 809 do_undos = 0; 810 811 for (;;) { 812 do_wakeup = 0; 813 814 for (i = 0; i < nsops; i++) { 815 sopptr = &sops[i]; 816 semptr = &semaptr->_sem_base[sopptr->sem_num]; 817 818 SEM_PRINTF(("semop: semaptr=%p, sem_base=%p, " 819 "semptr=%p, sem[%d]=%d : op=%d, flag=%s\n", 820 semaptr, semaptr->_sem_base, semptr, 821 sopptr->sem_num, semptr->semval, sopptr->sem_op, 822 (sopptr->sem_flg & IPC_NOWAIT) ? 823 "nowait" : "wait")); 824 825 if (sopptr->sem_op < 0) { 826 if ((int)(semptr->semval + 827 sopptr->sem_op) < 0) { 828 SEM_PRINTF(("semop: " 829 "can't do it now\n")); 830 break; 831 } else { 832 semptr->semval += sopptr->sem_op; 833 if (semptr->semval == 0 && 834 semptr->semzcnt > 0) 835 do_wakeup = 1; 836 } 837 if (sopptr->sem_flg & SEM_UNDO) 838 do_undos = 1; 839 } else if (sopptr->sem_op == 0) { 840 if (semptr->semval > 0) { 841 SEM_PRINTF(("semop: not zero now\n")); 842 break; 843 } 844 } else { 845 if (semptr->semncnt > 0) 846 do_wakeup = 1; 847 semptr->semval += sopptr->sem_op; 848 if (sopptr->sem_flg & SEM_UNDO) 849 do_undos = 1; 850 } 851 } 852 853 /* 854 * Did we get through the entire vector? 855 */ 856 if (i >= nsops) 857 goto done; 858 859 /* 860 * No ... rollback anything that we've already done 861 */ 862 SEM_PRINTF(("semop: rollback 0 through %d\n", i - 1)); 863 while (i-- > 0) 864 semaptr->_sem_base[sops[i].sem_num].semval -= 865 sops[i].sem_op; 866 867 /* 868 * If the request that we couldn't satisfy has the 869 * NOWAIT flag set then return with EAGAIN. 870 */ 871 if (sopptr->sem_flg & IPC_NOWAIT) { 872 error = EAGAIN; 873 goto out; 874 } 875 876 if (sopptr->sem_op == 0) 877 semptr->semzcnt++; 878 else 879 semptr->semncnt++; 880 881 sem_waiters++; 882 SEM_PRINTF(("semop: good night!\n")); 883 error = cv_wait_sig(&semcv[semid], &semlock); 884 SEM_PRINTF(("semop: good morning (error=%d)!\n", error)); 885 sem_waiters--; 886 887 /* Notify reallocator, if it is waiting */ 888 cv_broadcast(&sem_realloc_cv); 889 890 /* 891 * Make sure that the semaphore still exists 892 */ 893 if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0 || 894 semaptr->sem_perm._seq != seq) { 895 error = EIDRM; 896 goto out; 897 } 898 899 /* 900 * The semaphore is still alive. Readjust the count of 901 * waiting processes. 902 */ 903 semptr = &semaptr->_sem_base[sopptr->sem_num]; 904 if (sopptr->sem_op == 0) 905 semptr->semzcnt--; 906 else 907 semptr->semncnt--; 908 909 /* In case of such state, restart the call */ 910 if (sem_realloc_state) { 911 mutex_exit(&semlock); 912 goto restart; 913 } 914 915 /* Is it really morning, or was our sleep interrupted? */ 916 if (error != 0) { 917 error = EINTR; 918 goto out; 919 } 920 SEM_PRINTF(("semop: good morning!\n")); 921 } 922 923 done: 924 /* 925 * Process any SEM_UNDO requests. 926 */ 927 if (do_undos) { 928 for (i = 0; i < nsops; i++) { 929 /* 930 * We only need to deal with SEM_UNDO's for non-zero 931 * op's. 932 */ 933 int adjval; 934 935 if ((sops[i].sem_flg & SEM_UNDO) == 0) 936 continue; 937 adjval = sops[i].sem_op; 938 if (adjval == 0) 939 continue; 940 error = semundo_adjust(p, &suptr, semid, 941 sops[i].sem_num, -adjval); 942 if (error == 0) 943 continue; 944 945 /* 946 * Oh-Oh! We ran out of either sem_undo's or undo's. 947 * Rollback the adjustments to this point and then 948 * rollback the semaphore ups and down so we can return 949 * with an error with all structures restored. We 950 * rollback the undo's in the exact reverse order that 951 * we applied them. This guarantees that we won't run 952 * out of space as we roll things back out. 953 */ 954 while (i-- > 0) { 955 if ((sops[i].sem_flg & SEM_UNDO) == 0) 956 continue; 957 adjval = sops[i].sem_op; 958 if (adjval == 0) 959 continue; 960 if (semundo_adjust(p, &suptr, semid, 961 sops[i].sem_num, adjval) != 0) 962 panic("semop - can't undo undos"); 963 } 964 965 for (i = 0; i < nsops; i++) 966 semaptr->_sem_base[sops[i].sem_num].semval -= 967 sops[i].sem_op; 968 969 SEM_PRINTF(("error = %d from semundo_adjust\n", error)); 970 goto out; 971 } /* loop through the sops */ 972 } /* if (do_undos) */ 973 974 /* We're definitely done - set the sempid's */ 975 for (i = 0; i < nsops; i++) { 976 sopptr = &sops[i]; 977 semptr = &semaptr->_sem_base[sopptr->sem_num]; 978 semptr->sempid = p->p_pid; 979 } 980 981 /* Update sem_otime */ 982 semaptr->sem_otime = time_second; 983 984 /* Do a wakeup if any semaphore was up'd. */ 985 if (do_wakeup) { 986 SEM_PRINTF(("semop: doing wakeup\n")); 987 cv_broadcast(&semcv[semid]); 988 SEM_PRINTF(("semop: back from wakeup\n")); 989 } 990 SEM_PRINTF(("semop: done\n")); 991 *retval = 0; 992 993 out: 994 mutex_exit(&semlock); 995 if (sops != small_sops) 996 kmem_free(sops, nsops * sizeof(*sops)); 997 return error; 998 } 999 1000 /* 1001 * Go through the undo structures for this process and apply the 1002 * adjustments to semaphores. 1003 */ 1004 /*ARGSUSED*/ 1005 void 1006 semexit(struct proc *p, void *v) 1007 { 1008 struct sem_undo *suptr; 1009 struct sem_undo **supptr; 1010 1011 if ((p->p_flag & PK_SYSVSEM) == 0) 1012 return; 1013 1014 mutex_enter(&semlock); 1015 1016 /* 1017 * Go through the chain of undo vectors looking for one 1018 * associated with this process. 1019 */ 1020 1021 for (supptr = &semu_list; (suptr = *supptr) != NULL; 1022 supptr = &suptr->un_next) { 1023 if (suptr->un_proc == p) 1024 break; 1025 } 1026 1027 /* 1028 * If there is no undo vector, skip to the end. 1029 */ 1030 1031 if (suptr == NULL) { 1032 mutex_exit(&semlock); 1033 return; 1034 } 1035 1036 /* 1037 * We now have an undo vector for this process. 1038 */ 1039 1040 SEM_PRINTF(("proc @%p has undo structure with %d entries\n", p, 1041 suptr->un_cnt)); 1042 1043 /* 1044 * If there are any active undo elements then process them. 1045 */ 1046 if (suptr->un_cnt > 0) { 1047 int ix; 1048 1049 for (ix = 0; ix < suptr->un_cnt; ix++) { 1050 int semid = suptr->un_ent[ix].un_id; 1051 int semnum = suptr->un_ent[ix].un_num; 1052 int adjval = suptr->un_ent[ix].un_adjval; 1053 struct semid_ds *semaptr; 1054 1055 semaptr = &sema[semid]; 1056 if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0) 1057 panic("semexit - semid not allocated"); 1058 if (semnum >= semaptr->sem_nsems) 1059 panic("semexit - semnum out of range"); 1060 1061 SEM_PRINTF(("semexit: %p id=%d num=%d(adj=%d) ; " 1062 "sem=%d\n", 1063 suptr->un_proc, suptr->un_ent[ix].un_id, 1064 suptr->un_ent[ix].un_num, 1065 suptr->un_ent[ix].un_adjval, 1066 semaptr->_sem_base[semnum].semval)); 1067 1068 if (adjval < 0 && 1069 semaptr->_sem_base[semnum].semval < -adjval) 1070 semaptr->_sem_base[semnum].semval = 0; 1071 else 1072 semaptr->_sem_base[semnum].semval += adjval; 1073 1074 cv_broadcast(&semcv[semid]); 1075 SEM_PRINTF(("semexit: back from wakeup\n")); 1076 } 1077 } 1078 1079 /* 1080 * Deallocate the undo vector. 1081 */ 1082 SEM_PRINTF(("removing vector\n")); 1083 suptr->un_proc = NULL; 1084 *supptr = suptr->un_next; 1085 mutex_exit(&semlock); 1086 } 1087 1088 /* 1089 * Sysctl initialization and nodes. 1090 */ 1091 1092 static int 1093 sysctl_ipc_semmni(SYSCTLFN_ARGS) 1094 { 1095 int newsize, error; 1096 struct sysctlnode node; 1097 node = *rnode; 1098 node.sysctl_data = &newsize; 1099 1100 newsize = seminfo.semmni; 1101 error = sysctl_lookup(SYSCTLFN_CALL(&node)); 1102 if (error || newp == NULL) 1103 return error; 1104 1105 return semrealloc(newsize, seminfo.semmns, seminfo.semmnu); 1106 } 1107 1108 static int 1109 sysctl_ipc_semmns(SYSCTLFN_ARGS) 1110 { 1111 int newsize, error; 1112 struct sysctlnode node; 1113 node = *rnode; 1114 node.sysctl_data = &newsize; 1115 1116 newsize = seminfo.semmns; 1117 error = sysctl_lookup(SYSCTLFN_CALL(&node)); 1118 if (error || newp == NULL) 1119 return error; 1120 1121 return semrealloc(seminfo.semmni, newsize, seminfo.semmnu); 1122 } 1123 1124 static int 1125 sysctl_ipc_semmnu(SYSCTLFN_ARGS) 1126 { 1127 int newsize, error; 1128 struct sysctlnode node; 1129 node = *rnode; 1130 node.sysctl_data = &newsize; 1131 1132 newsize = seminfo.semmnu; 1133 error = sysctl_lookup(SYSCTLFN_CALL(&node)); 1134 if (error || newp == NULL) 1135 return error; 1136 1137 return semrealloc(seminfo.semmni, seminfo.semmns, newsize); 1138 } 1139 1140 SYSCTL_SETUP(sysctl_ipc_sem_setup, "sysctl kern.ipc subtree setup") 1141 { 1142 const struct sysctlnode *node = NULL; 1143 1144 sysctl_createv(clog, 0, NULL, NULL, 1145 CTLFLAG_PERMANENT, 1146 CTLTYPE_NODE, "kern", NULL, 1147 NULL, 0, NULL, 0, 1148 CTL_KERN, CTL_EOL); 1149 sysctl_createv(clog, 0, NULL, &node, 1150 CTLFLAG_PERMANENT, 1151 CTLTYPE_NODE, "ipc", 1152 SYSCTL_DESCR("SysV IPC options"), 1153 NULL, 0, NULL, 0, 1154 CTL_KERN, KERN_SYSVIPC, CTL_EOL); 1155 1156 if (node == NULL) 1157 return; 1158 1159 sysctl_createv(clog, 0, &node, NULL, 1160 CTLFLAG_PERMANENT | CTLFLAG_READWRITE, 1161 CTLTYPE_INT, "semmni", 1162 SYSCTL_DESCR("Max number of number of semaphore identifiers"), 1163 sysctl_ipc_semmni, 0, &seminfo.semmni, 0, 1164 CTL_CREATE, CTL_EOL); 1165 sysctl_createv(clog, 0, &node, NULL, 1166 CTLFLAG_PERMANENT | CTLFLAG_READWRITE, 1167 CTLTYPE_INT, "semmns", 1168 SYSCTL_DESCR("Max number of number of semaphores in system"), 1169 sysctl_ipc_semmns, 0, &seminfo.semmns, 0, 1170 CTL_CREATE, CTL_EOL); 1171 sysctl_createv(clog, 0, &node, NULL, 1172 CTLFLAG_PERMANENT | CTLFLAG_READWRITE, 1173 CTLTYPE_INT, "semmnu", 1174 SYSCTL_DESCR("Max number of undo structures in system"), 1175 sysctl_ipc_semmnu, 0, &seminfo.semmnu, 0, 1176 CTL_CREATE, CTL_EOL); 1177 } 1178