1 /* $NetBSD: sysv_sem.c,v 1.85 2009/01/11 02:45:53 christos 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 1.85 2009/01/11 02:45:53 christos 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 ptr to structure or NULL if no more room) 284 */ 285 286 struct sem_undo * 287 semu_alloc(struct proc *p) 288 { 289 int i; 290 struct sem_undo *suptr; 291 struct sem_undo **supptr; 292 int attempt; 293 294 KASSERT(mutex_owned(&semlock)); 295 296 /* 297 * Try twice to allocate something. 298 * (we'll purge any empty structures after the first pass so 299 * two passes are always enough) 300 */ 301 302 for (attempt = 0; attempt < 2; attempt++) { 303 /* 304 * Look for a free structure. 305 * Fill it in and return it if we find one. 306 */ 307 308 for (i = 0; i < seminfo.semmnu; i++) { 309 suptr = SEMU(semu, i); 310 if (suptr->un_proc == NULL) { 311 suptr->un_next = semu_list; 312 semu_list = suptr; 313 suptr->un_cnt = 0; 314 suptr->un_proc = p; 315 return (suptr); 316 } 317 } 318 319 /* 320 * We didn't find a free one, if this is the first attempt 321 * then try to free some structures. 322 */ 323 324 if (attempt == 0) { 325 /* All the structures are in use - try to free some */ 326 int did_something = 0; 327 328 supptr = &semu_list; 329 while ((suptr = *supptr) != NULL) { 330 if (suptr->un_cnt == 0) { 331 suptr->un_proc = NULL; 332 *supptr = suptr->un_next; 333 did_something = 1; 334 } else 335 supptr = &suptr->un_next; 336 } 337 338 /* If we didn't free anything then just give-up */ 339 if (!did_something) 340 return (NULL); 341 } else { 342 /* 343 * The second pass failed even though we freed 344 * something after the first pass! 345 * This is IMPOSSIBLE! 346 */ 347 panic("semu_alloc - second attempt failed"); 348 } 349 } 350 return NULL; 351 } 352 353 /* 354 * Adjust a particular entry for a particular proc 355 */ 356 357 int 358 semundo_adjust(struct proc *p, struct sem_undo **supptr, int semid, int semnum, 359 int adjval) 360 { 361 struct sem_undo *suptr; 362 struct undo *sunptr; 363 int i; 364 365 KASSERT(mutex_owned(&semlock)); 366 367 /* 368 * Look for and remember the sem_undo if the caller doesn't 369 * provide it 370 */ 371 372 suptr = *supptr; 373 if (suptr == NULL) { 374 for (suptr = semu_list; suptr != NULL; suptr = suptr->un_next) 375 if (suptr->un_proc == p) 376 break; 377 378 if (suptr == NULL) { 379 suptr = semu_alloc(p); 380 if (suptr == NULL) 381 return (ENOSPC); 382 } 383 *supptr = suptr; 384 } 385 386 /* 387 * Look for the requested entry and adjust it (delete if 388 * adjval becomes 0). 389 */ 390 sunptr = &suptr->un_ent[0]; 391 for (i = 0; i < suptr->un_cnt; i++, sunptr++) { 392 if (sunptr->un_id != semid || sunptr->un_num != semnum) 393 continue; 394 sunptr->un_adjval += adjval; 395 if (sunptr->un_adjval == 0) { 396 suptr->un_cnt--; 397 if (i < suptr->un_cnt) 398 suptr->un_ent[i] = 399 suptr->un_ent[suptr->un_cnt]; 400 } 401 return (0); 402 } 403 404 /* Didn't find the right entry - create it */ 405 if (suptr->un_cnt == SEMUME) 406 return (EINVAL); 407 408 sunptr = &suptr->un_ent[suptr->un_cnt]; 409 suptr->un_cnt++; 410 sunptr->un_adjval = adjval; 411 sunptr->un_id = semid; 412 sunptr->un_num = semnum; 413 return (0); 414 } 415 416 void 417 semundo_clear(int semid, int semnum) 418 { 419 struct sem_undo *suptr; 420 struct undo *sunptr, *sunend; 421 422 KASSERT(mutex_owned(&semlock)); 423 424 for (suptr = semu_list; suptr != NULL; suptr = suptr->un_next) 425 for (sunptr = &suptr->un_ent[0], 426 sunend = sunptr + suptr->un_cnt; sunptr < sunend;) { 427 if (sunptr->un_id == semid) { 428 if (semnum == -1 || sunptr->un_num == semnum) { 429 suptr->un_cnt--; 430 sunend--; 431 if (sunptr != sunend) 432 *sunptr = *sunend; 433 if (semnum != -1) 434 break; 435 else 436 continue; 437 } 438 } 439 sunptr++; 440 } 441 } 442 443 int 444 sys_____semctl50(struct lwp *l, const struct sys_____semctl50_args *uap, 445 register_t *retval) 446 { 447 /* { 448 syscallarg(int) semid; 449 syscallarg(int) semnum; 450 syscallarg(int) cmd; 451 syscallarg(union __semun *) arg; 452 } */ 453 struct semid_ds sembuf; 454 int cmd, error; 455 void *pass_arg; 456 union __semun karg; 457 458 cmd = SCARG(uap, cmd); 459 460 pass_arg = get_semctl_arg(cmd, &sembuf, &karg); 461 462 if (pass_arg) { 463 error = copyin(SCARG(uap, arg), &karg, sizeof(karg)); 464 if (error) 465 return error; 466 if (cmd == IPC_SET) { 467 error = copyin(karg.buf, &sembuf, sizeof(sembuf)); 468 if (error) 469 return (error); 470 } 471 } 472 473 error = semctl1(l, SCARG(uap, semid), SCARG(uap, semnum), cmd, 474 pass_arg, retval); 475 476 if (error == 0 && cmd == IPC_STAT) 477 error = copyout(&sembuf, karg.buf, sizeof(sembuf)); 478 479 return (error); 480 } 481 482 int 483 semctl1(struct lwp *l, int semid, int semnum, int cmd, void *v, 484 register_t *retval) 485 { 486 kauth_cred_t cred = l->l_cred; 487 union __semun *arg = v; 488 struct semid_ds *sembuf = v, *semaptr; 489 int i, error, ix; 490 491 SEM_PRINTF(("call to semctl(%d, %d, %d, %p)\n", 492 semid, semnum, cmd, v)); 493 494 mutex_enter(&semlock); 495 496 ix = IPCID_TO_IX(semid); 497 if (ix < 0 || ix >= seminfo.semmni) { 498 mutex_exit(&semlock); 499 return (EINVAL); 500 } 501 502 semaptr = &sema[ix]; 503 if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0 || 504 semaptr->sem_perm._seq != IPCID_TO_SEQ(semid)) { 505 mutex_exit(&semlock); 506 return (EINVAL); 507 } 508 509 switch (cmd) { 510 case IPC_RMID: 511 if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_M)) != 0) 512 break; 513 semaptr->sem_perm.cuid = kauth_cred_geteuid(cred); 514 semaptr->sem_perm.uid = kauth_cred_geteuid(cred); 515 semtot -= semaptr->sem_nsems; 516 for (i = semaptr->_sem_base - sem; i < semtot; i++) 517 sem[i] = sem[i + semaptr->sem_nsems]; 518 for (i = 0; i < seminfo.semmni; i++) { 519 if ((sema[i].sem_perm.mode & SEM_ALLOC) && 520 sema[i]._sem_base > semaptr->_sem_base) 521 sema[i]._sem_base -= semaptr->sem_nsems; 522 } 523 semaptr->sem_perm.mode = 0; 524 semundo_clear(ix, -1); 525 cv_broadcast(&semcv[ix]); 526 break; 527 528 case IPC_SET: 529 if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_M))) 530 break; 531 KASSERT(sembuf != NULL); 532 semaptr->sem_perm.uid = sembuf->sem_perm.uid; 533 semaptr->sem_perm.gid = sembuf->sem_perm.gid; 534 semaptr->sem_perm.mode = (semaptr->sem_perm.mode & ~0777) | 535 (sembuf->sem_perm.mode & 0777); 536 semaptr->sem_ctime = time_second; 537 break; 538 539 case IPC_STAT: 540 if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_R))) 541 break; 542 KASSERT(sembuf != NULL); 543 memcpy(sembuf, semaptr, sizeof(struct semid_ds)); 544 sembuf->sem_perm.mode &= 0777; 545 break; 546 547 case GETNCNT: 548 if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_R))) 549 break; 550 if (semnum < 0 || semnum >= semaptr->sem_nsems) { 551 error = EINVAL; 552 break; 553 } 554 *retval = semaptr->_sem_base[semnum].semncnt; 555 break; 556 557 case GETPID: 558 if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_R))) 559 break; 560 if (semnum < 0 || semnum >= semaptr->sem_nsems) { 561 error = EINVAL; 562 break; 563 } 564 *retval = semaptr->_sem_base[semnum].sempid; 565 break; 566 567 case GETVAL: 568 if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_R))) 569 break; 570 if (semnum < 0 || semnum >= semaptr->sem_nsems) { 571 error = EINVAL; 572 break; 573 } 574 *retval = semaptr->_sem_base[semnum].semval; 575 break; 576 577 case GETALL: 578 if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_R))) 579 break; 580 KASSERT(arg != NULL); 581 for (i = 0; i < semaptr->sem_nsems; i++) { 582 error = copyout(&semaptr->_sem_base[i].semval, 583 &arg->array[i], sizeof(arg->array[i])); 584 if (error != 0) 585 break; 586 } 587 break; 588 589 case GETZCNT: 590 if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_R))) 591 break; 592 if (semnum < 0 || semnum >= semaptr->sem_nsems) { 593 error = EINVAL; 594 break; 595 } 596 *retval = semaptr->_sem_base[semnum].semzcnt; 597 break; 598 599 case SETVAL: 600 if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_W))) 601 break; 602 if (semnum < 0 || semnum >= semaptr->sem_nsems) { 603 error = EINVAL; 604 break; 605 } 606 KASSERT(arg != NULL); 607 if ((unsigned int)arg->val > seminfo.semvmx) { 608 error = ERANGE; 609 break; 610 } 611 semaptr->_sem_base[semnum].semval = arg->val; 612 semundo_clear(ix, semnum); 613 cv_broadcast(&semcv[ix]); 614 break; 615 616 case SETALL: 617 if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_W))) 618 break; 619 KASSERT(arg != NULL); 620 for (i = 0; i < semaptr->sem_nsems; i++) { 621 unsigned short semval; 622 error = copyin(&arg->array[i], &semval, 623 sizeof(arg->array[i])); 624 if (error != 0) 625 break; 626 if ((unsigned int)semval > seminfo.semvmx) { 627 error = ERANGE; 628 break; 629 } 630 semaptr->_sem_base[i].semval = semval; 631 } 632 semundo_clear(ix, -1); 633 cv_broadcast(&semcv[ix]); 634 break; 635 636 default: 637 error = EINVAL; 638 break; 639 } 640 641 mutex_exit(&semlock); 642 return (error); 643 } 644 645 int 646 sys_semget(struct lwp *l, const struct sys_semget_args *uap, register_t *retval) 647 { 648 /* { 649 syscallarg(key_t) key; 650 syscallarg(int) nsems; 651 syscallarg(int) semflg; 652 } */ 653 int semid, error = 0; 654 int key = SCARG(uap, key); 655 int nsems = SCARG(uap, nsems); 656 int semflg = SCARG(uap, semflg); 657 kauth_cred_t cred = l->l_cred; 658 659 SEM_PRINTF(("semget(0x%x, %d, 0%o)\n", key, nsems, semflg)); 660 661 mutex_enter(&semlock); 662 663 if (key != IPC_PRIVATE) { 664 for (semid = 0; semid < seminfo.semmni; semid++) { 665 if ((sema[semid].sem_perm.mode & SEM_ALLOC) && 666 sema[semid].sem_perm._key == key) 667 break; 668 } 669 if (semid < seminfo.semmni) { 670 SEM_PRINTF(("found public key\n")); 671 if ((error = ipcperm(cred, &sema[semid].sem_perm, 672 semflg & 0700))) 673 goto out; 674 if (nsems > 0 && sema[semid].sem_nsems < nsems) { 675 SEM_PRINTF(("too small\n")); 676 error = EINVAL; 677 goto out; 678 } 679 if ((semflg & IPC_CREAT) && (semflg & IPC_EXCL)) { 680 SEM_PRINTF(("not exclusive\n")); 681 error = EEXIST; 682 goto out; 683 } 684 goto found; 685 } 686 } 687 688 SEM_PRINTF(("need to allocate the semid_ds\n")); 689 if (key == IPC_PRIVATE || (semflg & IPC_CREAT)) { 690 if (nsems <= 0 || nsems > seminfo.semmsl) { 691 SEM_PRINTF(("nsems out of range (0<%d<=%d)\n", nsems, 692 seminfo.semmsl)); 693 error = EINVAL; 694 goto out; 695 } 696 if (nsems > seminfo.semmns - semtot) { 697 SEM_PRINTF(("not enough semaphores left " 698 "(need %d, got %d)\n", 699 nsems, seminfo.semmns - semtot)); 700 error = ENOSPC; 701 goto out; 702 } 703 for (semid = 0; semid < seminfo.semmni; semid++) { 704 if ((sema[semid].sem_perm.mode & SEM_ALLOC) == 0) 705 break; 706 } 707 if (semid == seminfo.semmni) { 708 SEM_PRINTF(("no more semid_ds's available\n")); 709 error = ENOSPC; 710 goto out; 711 } 712 SEM_PRINTF(("semid %d is available\n", semid)); 713 sema[semid].sem_perm._key = key; 714 sema[semid].sem_perm.cuid = kauth_cred_geteuid(cred); 715 sema[semid].sem_perm.uid = kauth_cred_geteuid(cred); 716 sema[semid].sem_perm.cgid = kauth_cred_getegid(cred); 717 sema[semid].sem_perm.gid = kauth_cred_getegid(cred); 718 sema[semid].sem_perm.mode = (semflg & 0777) | SEM_ALLOC; 719 sema[semid].sem_perm._seq = 720 (sema[semid].sem_perm._seq + 1) & 0x7fff; 721 sema[semid].sem_nsems = nsems; 722 sema[semid].sem_otime = 0; 723 sema[semid].sem_ctime = time_second; 724 sema[semid]._sem_base = &sem[semtot]; 725 semtot += nsems; 726 memset(sema[semid]._sem_base, 0, 727 sizeof(sema[semid]._sem_base[0]) * nsems); 728 SEM_PRINTF(("sembase = %p, next = %p\n", sema[semid]._sem_base, 729 &sem[semtot])); 730 } else { 731 SEM_PRINTF(("didn't find it and wasn't asked to create it\n")); 732 error = ENOENT; 733 goto out; 734 } 735 736 found: 737 *retval = IXSEQ_TO_IPCID(semid, sema[semid].sem_perm); 738 out: 739 mutex_exit(&semlock); 740 return (error); 741 } 742 743 #define SMALL_SOPS 8 744 745 int 746 sys_semop(struct lwp *l, const struct sys_semop_args *uap, register_t *retval) 747 { 748 /* { 749 syscallarg(int) semid; 750 syscallarg(struct sembuf *) sops; 751 syscallarg(size_t) nsops; 752 } */ 753 struct proc *p = l->l_proc; 754 int semid = SCARG(uap, semid), seq; 755 size_t nsops = SCARG(uap, nsops); 756 struct sembuf small_sops[SMALL_SOPS]; 757 struct sembuf *sops; 758 struct semid_ds *semaptr; 759 struct sembuf *sopptr = NULL; 760 struct __sem *semptr = NULL; 761 struct sem_undo *suptr = NULL; 762 kauth_cred_t cred = l->l_cred; 763 int i, error; 764 int do_wakeup, do_undos; 765 766 SEM_PRINTF(("call to semop(%d, %p, %zd)\n", semid, SCARG(uap,sops), nsops)); 767 768 if (__predict_false((p->p_flag & PK_SYSVSEM) == 0)) { 769 mutex_enter(p->p_lock); 770 p->p_flag |= PK_SYSVSEM; 771 mutex_exit(p->p_lock); 772 } 773 774 restart: 775 if (nsops <= SMALL_SOPS) { 776 sops = small_sops; 777 } else if (nsops <= seminfo.semopm) { 778 sops = kmem_alloc(nsops * sizeof(*sops), KM_SLEEP); 779 } else { 780 SEM_PRINTF(("too many sops (max=%d, nsops=%zd)\n", 781 seminfo.semopm, nsops)); 782 return (E2BIG); 783 } 784 785 error = copyin(SCARG(uap, sops), sops, nsops * sizeof(sops[0])); 786 if (error) { 787 SEM_PRINTF(("error = %d from copyin(%p, %p, %zd)\n", error, 788 SCARG(uap, sops), &sops, nsops * sizeof(sops[0]))); 789 if (sops != small_sops) 790 kmem_free(sops, nsops * sizeof(*sops)); 791 return error; 792 } 793 794 mutex_enter(&semlock); 795 /* In case of reallocation, we will wait for completion */ 796 while (__predict_false(sem_realloc_state)) 797 cv_wait(&sem_realloc_cv, &semlock); 798 799 semid = IPCID_TO_IX(semid); /* Convert back to zero origin */ 800 if (semid < 0 || semid >= seminfo.semmni) { 801 error = EINVAL; 802 goto out; 803 } 804 805 semaptr = &sema[semid]; 806 seq = IPCID_TO_SEQ(SCARG(uap, semid)); 807 if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0 || 808 semaptr->sem_perm._seq != seq) { 809 error = EINVAL; 810 goto out; 811 } 812 813 if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_W))) { 814 SEM_PRINTF(("error = %d from ipaccess\n", error)); 815 goto out; 816 } 817 818 for (i = 0; i < nsops; i++) 819 if (sops[i].sem_num >= semaptr->sem_nsems) { 820 error = EFBIG; 821 goto out; 822 } 823 824 /* 825 * Loop trying to satisfy the vector of requests. 826 * If we reach a point where we must wait, any requests already 827 * performed are rolled back and we go to sleep until some other 828 * process wakes us up. At this point, we start all over again. 829 * 830 * This ensures that from the perspective of other tasks, a set 831 * of requests is atomic (never partially satisfied). 832 */ 833 do_undos = 0; 834 835 for (;;) { 836 do_wakeup = 0; 837 838 for (i = 0; i < nsops; i++) { 839 sopptr = &sops[i]; 840 semptr = &semaptr->_sem_base[sopptr->sem_num]; 841 842 SEM_PRINTF(("semop: semaptr=%p, sem_base=%p, " 843 "semptr=%p, sem[%d]=%d : op=%d, flag=%s\n", 844 semaptr, semaptr->_sem_base, semptr, 845 sopptr->sem_num, semptr->semval, sopptr->sem_op, 846 (sopptr->sem_flg & IPC_NOWAIT) ? 847 "nowait" : "wait")); 848 849 if (sopptr->sem_op < 0) { 850 if ((int)(semptr->semval + 851 sopptr->sem_op) < 0) { 852 SEM_PRINTF(("semop: " 853 "can't do it now\n")); 854 break; 855 } else { 856 semptr->semval += sopptr->sem_op; 857 if (semptr->semval == 0 && 858 semptr->semzcnt > 0) 859 do_wakeup = 1; 860 } 861 if (sopptr->sem_flg & SEM_UNDO) 862 do_undos = 1; 863 } else if (sopptr->sem_op == 0) { 864 if (semptr->semval > 0) { 865 SEM_PRINTF(("semop: not zero now\n")); 866 break; 867 } 868 } else { 869 if (semptr->semncnt > 0) 870 do_wakeup = 1; 871 semptr->semval += sopptr->sem_op; 872 if (sopptr->sem_flg & SEM_UNDO) 873 do_undos = 1; 874 } 875 } 876 877 /* 878 * Did we get through the entire vector? 879 */ 880 if (i >= nsops) 881 goto done; 882 883 /* 884 * No ... rollback anything that we've already done 885 */ 886 SEM_PRINTF(("semop: rollback 0 through %d\n", i - 1)); 887 while (i-- > 0) 888 semaptr->_sem_base[sops[i].sem_num].semval -= 889 sops[i].sem_op; 890 891 /* 892 * If the request that we couldn't satisfy has the 893 * NOWAIT flag set then return with EAGAIN. 894 */ 895 if (sopptr->sem_flg & IPC_NOWAIT) { 896 error = EAGAIN; 897 goto out; 898 } 899 900 if (sopptr->sem_op == 0) 901 semptr->semzcnt++; 902 else 903 semptr->semncnt++; 904 905 sem_waiters++; 906 SEM_PRINTF(("semop: good night!\n")); 907 error = cv_wait_sig(&semcv[semid], &semlock); 908 SEM_PRINTF(("semop: good morning (error=%d)!\n", error)); 909 sem_waiters--; 910 911 /* Notify reallocator, if it is waiting */ 912 cv_broadcast(&sem_realloc_cv); 913 914 /* 915 * Make sure that the semaphore still exists 916 */ 917 if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0 || 918 semaptr->sem_perm._seq != seq) { 919 error = EIDRM; 920 goto out; 921 } 922 923 /* 924 * The semaphore is still alive. Readjust the count of 925 * waiting processes. 926 */ 927 semptr = &semaptr->_sem_base[sopptr->sem_num]; 928 if (sopptr->sem_op == 0) 929 semptr->semzcnt--; 930 else 931 semptr->semncnt--; 932 933 /* In case of such state, restart the call */ 934 if (sem_realloc_state) { 935 mutex_exit(&semlock); 936 goto restart; 937 } 938 939 /* Is it really morning, or was our sleep interrupted? */ 940 if (error != 0) { 941 error = EINTR; 942 goto out; 943 } 944 SEM_PRINTF(("semop: good morning!\n")); 945 } 946 947 done: 948 /* 949 * Process any SEM_UNDO requests. 950 */ 951 if (do_undos) { 952 for (i = 0; i < nsops; i++) { 953 /* 954 * We only need to deal with SEM_UNDO's for non-zero 955 * op's. 956 */ 957 int adjval; 958 959 if ((sops[i].sem_flg & SEM_UNDO) == 0) 960 continue; 961 adjval = sops[i].sem_op; 962 if (adjval == 0) 963 continue; 964 error = semundo_adjust(p, &suptr, semid, 965 sops[i].sem_num, -adjval); 966 if (error == 0) 967 continue; 968 969 /* 970 * Oh-Oh! We ran out of either sem_undo's or undo's. 971 * Rollback the adjustments to this point and then 972 * rollback the semaphore ups and down so we can return 973 * with an error with all structures restored. We 974 * rollback the undo's in the exact reverse order that 975 * we applied them. This guarantees that we won't run 976 * out of space as we roll things back out. 977 */ 978 while (i-- > 0) { 979 if ((sops[i].sem_flg & SEM_UNDO) == 0) 980 continue; 981 adjval = sops[i].sem_op; 982 if (adjval == 0) 983 continue; 984 if (semundo_adjust(p, &suptr, semid, 985 sops[i].sem_num, adjval) != 0) 986 panic("semop - can't undo undos"); 987 } 988 989 for (i = 0; i < nsops; i++) 990 semaptr->_sem_base[sops[i].sem_num].semval -= 991 sops[i].sem_op; 992 993 SEM_PRINTF(("error = %d from semundo_adjust\n", error)); 994 goto out; 995 } /* loop through the sops */ 996 } /* if (do_undos) */ 997 998 /* We're definitely done - set the sempid's */ 999 for (i = 0; i < nsops; i++) { 1000 sopptr = &sops[i]; 1001 semptr = &semaptr->_sem_base[sopptr->sem_num]; 1002 semptr->sempid = p->p_pid; 1003 } 1004 1005 /* Update sem_otime */ 1006 semaptr->sem_otime = time_second; 1007 1008 /* Do a wakeup if any semaphore was up'd. */ 1009 if (do_wakeup) { 1010 SEM_PRINTF(("semop: doing wakeup\n")); 1011 cv_broadcast(&semcv[semid]); 1012 SEM_PRINTF(("semop: back from wakeup\n")); 1013 } 1014 SEM_PRINTF(("semop: done\n")); 1015 *retval = 0; 1016 1017 out: 1018 mutex_exit(&semlock); 1019 if (sops != small_sops) 1020 kmem_free(sops, nsops * sizeof(*sops)); 1021 return error; 1022 } 1023 1024 /* 1025 * Go through the undo structures for this process and apply the 1026 * adjustments to semaphores. 1027 */ 1028 /*ARGSUSED*/ 1029 void 1030 semexit(struct proc *p, void *v) 1031 { 1032 struct sem_undo *suptr; 1033 struct sem_undo **supptr; 1034 1035 if ((p->p_flag & PK_SYSVSEM) == 0) 1036 return; 1037 1038 mutex_enter(&semlock); 1039 1040 /* 1041 * Go through the chain of undo vectors looking for one 1042 * associated with this process. 1043 */ 1044 1045 for (supptr = &semu_list; (suptr = *supptr) != NULL; 1046 supptr = &suptr->un_next) { 1047 if (suptr->un_proc == p) 1048 break; 1049 } 1050 1051 /* 1052 * If there is no undo vector, skip to the end. 1053 */ 1054 1055 if (suptr == NULL) { 1056 mutex_exit(&semlock); 1057 return; 1058 } 1059 1060 /* 1061 * We now have an undo vector for this process. 1062 */ 1063 1064 SEM_PRINTF(("proc @%p has undo structure with %d entries\n", p, 1065 suptr->un_cnt)); 1066 1067 /* 1068 * If there are any active undo elements then process them. 1069 */ 1070 if (suptr->un_cnt > 0) { 1071 int ix; 1072 1073 for (ix = 0; ix < suptr->un_cnt; ix++) { 1074 int semid = suptr->un_ent[ix].un_id; 1075 int semnum = suptr->un_ent[ix].un_num; 1076 int adjval = suptr->un_ent[ix].un_adjval; 1077 struct semid_ds *semaptr; 1078 1079 semaptr = &sema[semid]; 1080 if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0) 1081 panic("semexit - semid not allocated"); 1082 if (semnum >= semaptr->sem_nsems) 1083 panic("semexit - semnum out of range"); 1084 1085 SEM_PRINTF(("semexit: %p id=%d num=%d(adj=%d) ; " 1086 "sem=%d\n", 1087 suptr->un_proc, suptr->un_ent[ix].un_id, 1088 suptr->un_ent[ix].un_num, 1089 suptr->un_ent[ix].un_adjval, 1090 semaptr->_sem_base[semnum].semval)); 1091 1092 if (adjval < 0 && 1093 semaptr->_sem_base[semnum].semval < -adjval) 1094 semaptr->_sem_base[semnum].semval = 0; 1095 else 1096 semaptr->_sem_base[semnum].semval += adjval; 1097 1098 cv_broadcast(&semcv[semid]); 1099 SEM_PRINTF(("semexit: back from wakeup\n")); 1100 } 1101 } 1102 1103 /* 1104 * Deallocate the undo vector. 1105 */ 1106 SEM_PRINTF(("removing vector\n")); 1107 suptr->un_proc = NULL; 1108 *supptr = suptr->un_next; 1109 mutex_exit(&semlock); 1110 } 1111 1112 /* 1113 * Sysctl initialization and nodes. 1114 */ 1115 1116 static int 1117 sysctl_ipc_semmni(SYSCTLFN_ARGS) 1118 { 1119 int newsize, error; 1120 struct sysctlnode node; 1121 node = *rnode; 1122 node.sysctl_data = &newsize; 1123 1124 newsize = seminfo.semmni; 1125 error = sysctl_lookup(SYSCTLFN_CALL(&node)); 1126 if (error || newp == NULL) 1127 return error; 1128 1129 return semrealloc(newsize, seminfo.semmns, seminfo.semmnu); 1130 } 1131 1132 static int 1133 sysctl_ipc_semmns(SYSCTLFN_ARGS) 1134 { 1135 int newsize, error; 1136 struct sysctlnode node; 1137 node = *rnode; 1138 node.sysctl_data = &newsize; 1139 1140 newsize = seminfo.semmns; 1141 error = sysctl_lookup(SYSCTLFN_CALL(&node)); 1142 if (error || newp == NULL) 1143 return error; 1144 1145 return semrealloc(seminfo.semmni, newsize, seminfo.semmnu); 1146 } 1147 1148 static int 1149 sysctl_ipc_semmnu(SYSCTLFN_ARGS) 1150 { 1151 int newsize, error; 1152 struct sysctlnode node; 1153 node = *rnode; 1154 node.sysctl_data = &newsize; 1155 1156 newsize = seminfo.semmnu; 1157 error = sysctl_lookup(SYSCTLFN_CALL(&node)); 1158 if (error || newp == NULL) 1159 return error; 1160 1161 return semrealloc(seminfo.semmni, seminfo.semmns, newsize); 1162 } 1163 1164 SYSCTL_SETUP(sysctl_ipc_sem_setup, "sysctl kern.ipc subtree setup") 1165 { 1166 const struct sysctlnode *node = NULL; 1167 1168 sysctl_createv(clog, 0, NULL, NULL, 1169 CTLFLAG_PERMANENT, 1170 CTLTYPE_NODE, "kern", NULL, 1171 NULL, 0, NULL, 0, 1172 CTL_KERN, CTL_EOL); 1173 sysctl_createv(clog, 0, NULL, &node, 1174 CTLFLAG_PERMANENT, 1175 CTLTYPE_NODE, "ipc", 1176 SYSCTL_DESCR("SysV IPC options"), 1177 NULL, 0, NULL, 0, 1178 CTL_KERN, KERN_SYSVIPC, CTL_EOL); 1179 1180 if (node == NULL) 1181 return; 1182 1183 sysctl_createv(clog, 0, &node, NULL, 1184 CTLFLAG_PERMANENT | CTLFLAG_READWRITE, 1185 CTLTYPE_INT, "semmni", 1186 SYSCTL_DESCR("Max number of number of semaphore identifiers"), 1187 sysctl_ipc_semmni, 0, &seminfo.semmni, 0, 1188 CTL_CREATE, CTL_EOL); 1189 sysctl_createv(clog, 0, &node, NULL, 1190 CTLFLAG_PERMANENT | CTLFLAG_READWRITE, 1191 CTLTYPE_INT, "semmns", 1192 SYSCTL_DESCR("Max number of number of semaphores in system"), 1193 sysctl_ipc_semmns, 0, &seminfo.semmns, 0, 1194 CTL_CREATE, CTL_EOL); 1195 sysctl_createv(clog, 0, &node, NULL, 1196 CTLFLAG_PERMANENT | CTLFLAG_READWRITE, 1197 CTLTYPE_INT, "semmnu", 1198 SYSCTL_DESCR("Max number of undo structures in system"), 1199 sysctl_ipc_semmnu, 0, &seminfo.semmnu, 0, 1200 CTL_CREATE, CTL_EOL); 1201 } 1202