1 /* $FreeBSD$ */ 2 3 /* 4 * Implementation of SVID semaphores 5 * 6 * Author: Daniel Boulet 7 * 8 * This software is provided ``AS IS'' without any warranties of any kind. 9 */ 10 11 #include "opt_sysvipc.h" 12 13 #include <sys/param.h> 14 #include <sys/systm.h> 15 #include <sys/sysproto.h> 16 #include <sys/eventhandler.h> 17 #include <sys/kernel.h> 18 #include <sys/proc.h> 19 #include <sys/lock.h> 20 #include <sys/mutex.h> 21 #include <sys/sem.h> 22 #include <sys/syscall.h> 23 #include <sys/sysent.h> 24 #include <sys/sysctl.h> 25 #include <sys/malloc.h> 26 #include <sys/jail.h> 27 28 static MALLOC_DEFINE(M_SEM, "sem", "SVID compatible semaphores"); 29 30 #ifdef SEM_DEBUG 31 #define DPRINTF(a) printf a 32 #else 33 #define DPRINTF(a) 34 #endif 35 36 static void seminit(void); 37 static int sysvsem_modload(struct module *, int, void *); 38 static int semunload(void); 39 static void semexit_myhook(void *arg, struct proc *p); 40 static int sysctl_sema(SYSCTL_HANDLER_ARGS); 41 static int semvalid(int semid, struct semid_ds *semaptr); 42 43 #ifndef _SYS_SYSPROTO_H_ 44 struct __semctl_args; 45 int __semctl(struct thread *td, struct __semctl_args *uap); 46 struct semget_args; 47 int semget(struct thread *td, struct semget_args *uap); 48 struct semop_args; 49 int semop(struct thread *td, struct semop_args *uap); 50 #endif 51 52 static struct sem_undo *semu_alloc(struct thread *td); 53 static int semundo_adjust(struct thread *td, struct sem_undo **supptr, 54 int semid, int semnum, int adjval); 55 static void semundo_clear(int semid, int semnum); 56 57 /* XXX casting to (sy_call_t *) is bogus, as usual. */ 58 static sy_call_t *semcalls[] = { 59 (sy_call_t *)__semctl, (sy_call_t *)semget, 60 (sy_call_t *)semop 61 }; 62 63 static struct mtx sem_mtx; /* semaphore global lock */ 64 static int semtot = 0; 65 static struct semid_ds *sema; /* semaphore id pool */ 66 static struct mtx *sema_mtx; /* semaphore id pool mutexes*/ 67 static struct sem *sem; /* semaphore pool */ 68 SLIST_HEAD(, sem_undo) semu_list; /* list of active undo structures */ 69 static int *semu; /* undo structure pool */ 70 static eventhandler_tag semexit_tag; 71 72 #define SEMUNDO_MTX sem_mtx 73 #define SEMUNDO_LOCK() mtx_lock(&SEMUNDO_MTX); 74 #define SEMUNDO_UNLOCK() mtx_unlock(&SEMUNDO_MTX); 75 #define SEMUNDO_LOCKASSERT(how) mtx_assert(&SEMUNDO_MTX, (how)); 76 77 struct sem { 78 u_short semval; /* semaphore value */ 79 pid_t sempid; /* pid of last operation */ 80 u_short semncnt; /* # awaiting semval > cval */ 81 u_short semzcnt; /* # awaiting semval = 0 */ 82 }; 83 84 /* 85 * Undo structure (one per process) 86 */ 87 struct sem_undo { 88 SLIST_ENTRY(sem_undo) un_next; /* ptr to next active undo structure */ 89 struct proc *un_proc; /* owner of this structure */ 90 short un_cnt; /* # of active entries */ 91 struct undo { 92 short un_adjval; /* adjust on exit values */ 93 short un_num; /* semaphore # */ 94 int un_id; /* semid */ 95 } un_ent[1]; /* undo entries */ 96 }; 97 98 /* 99 * Configuration parameters 100 */ 101 #ifndef SEMMNI 102 #define SEMMNI 10 /* # of semaphore identifiers */ 103 #endif 104 #ifndef SEMMNS 105 #define SEMMNS 60 /* # of semaphores in system */ 106 #endif 107 #ifndef SEMUME 108 #define SEMUME 10 /* max # of undo entries per process */ 109 #endif 110 #ifndef SEMMNU 111 #define SEMMNU 30 /* # of undo structures in system */ 112 #endif 113 114 /* shouldn't need tuning */ 115 #ifndef SEMMAP 116 #define SEMMAP 30 /* # of entries in semaphore map */ 117 #endif 118 #ifndef SEMMSL 119 #define SEMMSL SEMMNS /* max # of semaphores per id */ 120 #endif 121 #ifndef SEMOPM 122 #define SEMOPM 100 /* max # of operations per semop call */ 123 #endif 124 125 #define SEMVMX 32767 /* semaphore maximum value */ 126 #define SEMAEM 16384 /* adjust on exit max value */ 127 128 /* 129 * Due to the way semaphore memory is allocated, we have to ensure that 130 * SEMUSZ is properly aligned. 131 */ 132 133 #define SEM_ALIGN(bytes) (((bytes) + (sizeof(long) - 1)) & ~(sizeof(long) - 1)) 134 135 /* actual size of an undo structure */ 136 #define SEMUSZ SEM_ALIGN(offsetof(struct sem_undo, un_ent[SEMUME])) 137 138 /* 139 * Macro to find a particular sem_undo vector 140 */ 141 #define SEMU(ix) \ 142 ((struct sem_undo *)(((intptr_t)semu)+ix * seminfo.semusz)) 143 144 /* 145 * semaphore info struct 146 */ 147 struct seminfo seminfo = { 148 SEMMAP, /* # of entries in semaphore map */ 149 SEMMNI, /* # of semaphore identifiers */ 150 SEMMNS, /* # of semaphores in system */ 151 SEMMNU, /* # of undo structures in system */ 152 SEMMSL, /* max # of semaphores per id */ 153 SEMOPM, /* max # of operations per semop call */ 154 SEMUME, /* max # of undo entries per process */ 155 SEMUSZ, /* size in bytes of undo structure */ 156 SEMVMX, /* semaphore maximum value */ 157 SEMAEM /* adjust on exit max value */ 158 }; 159 160 SYSCTL_DECL(_kern_ipc); 161 SYSCTL_INT(_kern_ipc, OID_AUTO, semmap, CTLFLAG_RW, &seminfo.semmap, 0, ""); 162 SYSCTL_INT(_kern_ipc, OID_AUTO, semmni, CTLFLAG_RD, &seminfo.semmni, 0, ""); 163 SYSCTL_INT(_kern_ipc, OID_AUTO, semmns, CTLFLAG_RD, &seminfo.semmns, 0, ""); 164 SYSCTL_INT(_kern_ipc, OID_AUTO, semmnu, CTLFLAG_RD, &seminfo.semmnu, 0, ""); 165 SYSCTL_INT(_kern_ipc, OID_AUTO, semmsl, CTLFLAG_RW, &seminfo.semmsl, 0, ""); 166 SYSCTL_INT(_kern_ipc, OID_AUTO, semopm, CTLFLAG_RD, &seminfo.semopm, 0, ""); 167 SYSCTL_INT(_kern_ipc, OID_AUTO, semume, CTLFLAG_RD, &seminfo.semume, 0, ""); 168 SYSCTL_INT(_kern_ipc, OID_AUTO, semusz, CTLFLAG_RD, &seminfo.semusz, 0, ""); 169 SYSCTL_INT(_kern_ipc, OID_AUTO, semvmx, CTLFLAG_RW, &seminfo.semvmx, 0, ""); 170 SYSCTL_INT(_kern_ipc, OID_AUTO, semaem, CTLFLAG_RW, &seminfo.semaem, 0, ""); 171 SYSCTL_PROC(_kern_ipc, OID_AUTO, sema, CTLFLAG_RD, 172 NULL, 0, sysctl_sema, "", ""); 173 174 static void 175 seminit(void) 176 { 177 int i; 178 179 TUNABLE_INT_FETCH("kern.ipc.semmap", &seminfo.semmap); 180 TUNABLE_INT_FETCH("kern.ipc.semmni", &seminfo.semmni); 181 TUNABLE_INT_FETCH("kern.ipc.semmns", &seminfo.semmns); 182 TUNABLE_INT_FETCH("kern.ipc.semmnu", &seminfo.semmnu); 183 TUNABLE_INT_FETCH("kern.ipc.semmsl", &seminfo.semmsl); 184 TUNABLE_INT_FETCH("kern.ipc.semopm", &seminfo.semopm); 185 TUNABLE_INT_FETCH("kern.ipc.semume", &seminfo.semume); 186 TUNABLE_INT_FETCH("kern.ipc.semusz", &seminfo.semusz); 187 TUNABLE_INT_FETCH("kern.ipc.semvmx", &seminfo.semvmx); 188 TUNABLE_INT_FETCH("kern.ipc.semaem", &seminfo.semaem); 189 190 sem = malloc(sizeof(struct sem) * seminfo.semmns, M_SEM, M_WAITOK); 191 sema = malloc(sizeof(struct semid_ds) * seminfo.semmni, M_SEM, 192 M_WAITOK); 193 sema_mtx = malloc(sizeof(struct mtx) * seminfo.semmni, M_SEM, 194 M_WAITOK | M_ZERO); 195 semu = malloc(seminfo.semmnu * seminfo.semusz, M_SEM, M_WAITOK); 196 197 for (i = 0; i < seminfo.semmni; i++) { 198 sema[i].sem_base = 0; 199 sema[i].sem_perm.mode = 0; 200 } 201 for (i = 0; i < seminfo.semmni; i++) 202 mtx_init(&sema_mtx[i], "semid", NULL, MTX_DEF); 203 for (i = 0; i < seminfo.semmnu; i++) { 204 struct sem_undo *suptr = SEMU(i); 205 suptr->un_proc = NULL; 206 } 207 SLIST_INIT(&semu_list); 208 mtx_init(&sem_mtx, "sem", NULL, MTX_DEF); 209 semexit_tag = EVENTHANDLER_REGISTER(process_exit, semexit_myhook, NULL, 210 EVENTHANDLER_PRI_ANY); 211 } 212 213 static int 214 semunload(void) 215 { 216 int i; 217 218 if (semtot != 0) 219 return (EBUSY); 220 221 EVENTHANDLER_DEREGISTER(process_exit, semexit_tag); 222 free(sem, M_SEM); 223 free(sema, M_SEM); 224 free(semu, M_SEM); 225 for (i = 0; i < seminfo.semmni; i++) 226 mtx_destroy(&sema_mtx[i]); 227 mtx_destroy(&sem_mtx); 228 return (0); 229 } 230 231 static int 232 sysvsem_modload(struct module *module, int cmd, void *arg) 233 { 234 int error = 0; 235 236 switch (cmd) { 237 case MOD_LOAD: 238 seminit(); 239 break; 240 case MOD_UNLOAD: 241 error = semunload(); 242 break; 243 case MOD_SHUTDOWN: 244 break; 245 default: 246 error = EINVAL; 247 break; 248 } 249 return (error); 250 } 251 252 static moduledata_t sysvsem_mod = { 253 "sysvsem", 254 &sysvsem_modload, 255 NULL 256 }; 257 258 SYSCALL_MODULE_HELPER(semsys); 259 SYSCALL_MODULE_HELPER(__semctl); 260 SYSCALL_MODULE_HELPER(semget); 261 SYSCALL_MODULE_HELPER(semop); 262 263 DECLARE_MODULE(sysvsem, sysvsem_mod, 264 SI_SUB_SYSV_SEM, SI_ORDER_FIRST); 265 MODULE_VERSION(sysvsem, 1); 266 267 /* 268 * Entry point for all SEM calls 269 * 270 * MPSAFE 271 */ 272 int 273 semsys(td, uap) 274 struct thread *td; 275 /* XXX actually varargs. */ 276 struct semsys_args /* { 277 u_int which; 278 int a2; 279 int a3; 280 int a4; 281 int a5; 282 } */ *uap; 283 { 284 int error; 285 286 if (!jail_sysvipc_allowed && jailed(td->td_ucred)) 287 return (ENOSYS); 288 if (uap->which >= sizeof(semcalls)/sizeof(semcalls[0])) 289 return (EINVAL); 290 error = (*semcalls[uap->which])(td, &uap->a2); 291 return (error); 292 } 293 294 /* 295 * Allocate a new sem_undo structure for a process 296 * (returns ptr to structure or NULL if no more room) 297 */ 298 299 static struct sem_undo * 300 semu_alloc(td) 301 struct thread *td; 302 { 303 int i; 304 struct sem_undo *suptr; 305 struct sem_undo **supptr; 306 int attempt; 307 308 SEMUNDO_LOCKASSERT(MA_OWNED); 309 /* 310 * Try twice to allocate something. 311 * (we'll purge any empty structures after the first pass so 312 * two passes are always enough) 313 */ 314 315 for (attempt = 0; attempt < 2; attempt++) { 316 /* 317 * Look for a free structure. 318 * Fill it in and return it if we find one. 319 */ 320 321 for (i = 0; i < seminfo.semmnu; i++) { 322 suptr = SEMU(i); 323 if (suptr->un_proc == NULL) { 324 SLIST_INSERT_HEAD(&semu_list, suptr, un_next); 325 suptr->un_cnt = 0; 326 suptr->un_proc = td->td_proc; 327 return(suptr); 328 } 329 } 330 331 /* 332 * We didn't find a free one, if this is the first attempt 333 * then try to free some structures. 334 */ 335 336 if (attempt == 0) { 337 /* All the structures are in use - try to free some */ 338 int did_something = 0; 339 340 SLIST_FOREACH_PREVPTR(suptr, supptr, &semu_list, 341 un_next) { 342 if (suptr->un_cnt == 0) { 343 suptr->un_proc = NULL; 344 did_something = 1; 345 *supptr = SLIST_NEXT(suptr, un_next); 346 } 347 } 348 349 /* If we didn't free anything then just give-up */ 350 if (!did_something) 351 return(NULL); 352 } else { 353 /* 354 * The second pass failed even though we freed 355 * something after the first pass! 356 * This is IMPOSSIBLE! 357 */ 358 panic("semu_alloc - second attempt failed"); 359 } 360 } 361 return (NULL); 362 } 363 364 /* 365 * Adjust a particular entry for a particular proc 366 */ 367 368 static int 369 semundo_adjust(td, supptr, semid, semnum, adjval) 370 struct thread *td; 371 struct sem_undo **supptr; 372 int semid, semnum; 373 int adjval; 374 { 375 struct proc *p = td->td_proc; 376 struct sem_undo *suptr; 377 struct undo *sunptr; 378 int i; 379 380 SEMUNDO_LOCKASSERT(MA_OWNED); 381 /* Look for and remember the sem_undo if the caller doesn't provide 382 it */ 383 384 suptr = *supptr; 385 if (suptr == NULL) { 386 SLIST_FOREACH(suptr, &semu_list, un_next) { 387 if (suptr->un_proc == p) { 388 *supptr = suptr; 389 break; 390 } 391 } 392 if (suptr == NULL) { 393 if (adjval == 0) 394 return(0); 395 suptr = semu_alloc(td); 396 if (suptr == NULL) 397 return(ENOSPC); 398 *supptr = suptr; 399 } 400 } 401 402 /* 403 * Look for the requested entry and adjust it (delete if adjval becomes 404 * 0). 405 */ 406 sunptr = &suptr->un_ent[0]; 407 for (i = 0; i < suptr->un_cnt; i++, sunptr++) { 408 if (sunptr->un_id != semid || sunptr->un_num != semnum) 409 continue; 410 if (adjval != 0) { 411 adjval += sunptr->un_adjval; 412 if (adjval > seminfo.semaem || adjval < -seminfo.semaem) 413 return (ERANGE); 414 } 415 sunptr->un_adjval = adjval; 416 if (sunptr->un_adjval == 0) { 417 suptr->un_cnt--; 418 if (i < suptr->un_cnt) 419 suptr->un_ent[i] = 420 suptr->un_ent[suptr->un_cnt]; 421 } 422 return(0); 423 } 424 425 /* Didn't find the right entry - create it */ 426 if (adjval == 0) 427 return(0); 428 if (adjval > seminfo.semaem || adjval < -seminfo.semaem) 429 return (ERANGE); 430 if (suptr->un_cnt != seminfo.semume) { 431 sunptr = &suptr->un_ent[suptr->un_cnt]; 432 suptr->un_cnt++; 433 sunptr->un_adjval = adjval; 434 sunptr->un_id = semid; sunptr->un_num = semnum; 435 } else 436 return(EINVAL); 437 return(0); 438 } 439 440 static void 441 semundo_clear(semid, semnum) 442 int semid, semnum; 443 { 444 struct sem_undo *suptr; 445 446 SEMUNDO_LOCKASSERT(MA_OWNED); 447 SLIST_FOREACH(suptr, &semu_list, un_next) { 448 struct undo *sunptr = &suptr->un_ent[0]; 449 int i = 0; 450 451 while (i < suptr->un_cnt) { 452 if (sunptr->un_id == semid) { 453 if (semnum == -1 || sunptr->un_num == semnum) { 454 suptr->un_cnt--; 455 if (i < suptr->un_cnt) { 456 suptr->un_ent[i] = 457 suptr->un_ent[suptr->un_cnt]; 458 continue; 459 } 460 } 461 if (semnum != -1) 462 break; 463 } 464 i++, sunptr++; 465 } 466 } 467 } 468 469 static int 470 semvalid(semid, semaptr) 471 int semid; 472 struct semid_ds *semaptr; 473 { 474 475 return ((semaptr->sem_perm.mode & SEM_ALLOC) == 0 || 476 semaptr->sem_perm.seq != IPCID_TO_SEQ(semid) ? EINVAL : 0); 477 } 478 479 /* 480 * Note that the user-mode half of this passes a union, not a pointer 481 */ 482 #ifndef _SYS_SYSPROTO_H_ 483 struct __semctl_args { 484 int semid; 485 int semnum; 486 int cmd; 487 union semun *arg; 488 }; 489 #endif 490 491 /* 492 * MPSAFE 493 */ 494 int 495 __semctl(td, uap) 496 struct thread *td; 497 struct __semctl_args *uap; 498 { 499 int semid = uap->semid; 500 int semnum = uap->semnum; 501 int cmd = uap->cmd; 502 u_short *array; 503 union semun *arg = uap->arg; 504 union semun real_arg; 505 struct ucred *cred = td->td_ucred; 506 int i, rval, error; 507 struct semid_ds sbuf; 508 struct semid_ds *semaptr; 509 struct mtx *sema_mtxp; 510 u_short usval, count; 511 512 DPRINTF(("call to semctl(%d, %d, %d, 0x%x)\n", 513 semid, semnum, cmd, arg)); 514 if (!jail_sysvipc_allowed && jailed(td->td_ucred)) 515 return (ENOSYS); 516 517 array = NULL; 518 519 switch(cmd) { 520 case SEM_STAT: 521 if (semid < 0 || semid >= seminfo.semmni) 522 return (EINVAL); 523 if ((error = copyin(arg, &real_arg, sizeof(real_arg))) != 0) 524 return (error); 525 semaptr = &sema[semid]; 526 sema_mtxp = &sema_mtx[semid]; 527 mtx_lock(sema_mtxp); 528 if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0 ) { 529 error = EINVAL; 530 goto done2; 531 } 532 if ((error = ipcperm(td, &semaptr->sem_perm, IPC_R))) 533 goto done2; 534 mtx_unlock(sema_mtxp); 535 error = copyout(semaptr, real_arg.buf, sizeof(struct semid_ds)); 536 rval = IXSEQ_TO_IPCID(semid,semaptr->sem_perm); 537 if (error == 0) 538 td->td_retval[0] = rval; 539 return (error); 540 } 541 542 semid = IPCID_TO_IX(semid); 543 if (semid < 0 || semid >= seminfo.semmni) 544 return (EINVAL); 545 546 semaptr = &sema[semid]; 547 sema_mtxp = &sema_mtx[semid]; 548 549 error = 0; 550 rval = 0; 551 552 switch (cmd) { 553 case IPC_RMID: 554 mtx_lock(sema_mtxp); 555 if ((error = semvalid(uap->semid, semaptr)) != 0) 556 goto done2; 557 if ((error = ipcperm(td, &semaptr->sem_perm, IPC_M))) 558 goto done2; 559 semaptr->sem_perm.cuid = cred->cr_uid; 560 semaptr->sem_perm.uid = cred->cr_uid; 561 semtot -= semaptr->sem_nsems; 562 for (i = semaptr->sem_base - sem; i < semtot; i++) 563 sem[i] = sem[i + semaptr->sem_nsems]; 564 for (i = 0; i < seminfo.semmni; i++) { 565 if ((sema[i].sem_perm.mode & SEM_ALLOC) && 566 sema[i].sem_base > semaptr->sem_base) 567 sema[i].sem_base -= semaptr->sem_nsems; 568 } 569 semaptr->sem_perm.mode = 0; 570 SEMUNDO_LOCK(); 571 semundo_clear(semid, -1); 572 SEMUNDO_UNLOCK(); 573 wakeup(semaptr); 574 break; 575 576 case IPC_SET: 577 if ((error = copyin(arg, &real_arg, sizeof(real_arg))) != 0) 578 goto done2; 579 if ((error = copyin(real_arg.buf, &sbuf, sizeof(sbuf))) != 0) 580 goto done2; 581 mtx_lock(sema_mtxp); 582 if ((error = semvalid(uap->semid, semaptr)) != 0) 583 goto done2; 584 if ((error = ipcperm(td, &semaptr->sem_perm, IPC_M))) 585 goto done2; 586 semaptr->sem_perm.uid = sbuf.sem_perm.uid; 587 semaptr->sem_perm.gid = sbuf.sem_perm.gid; 588 semaptr->sem_perm.mode = (semaptr->sem_perm.mode & ~0777) | 589 (sbuf.sem_perm.mode & 0777); 590 semaptr->sem_ctime = time_second; 591 break; 592 593 case IPC_STAT: 594 if ((error = copyin(arg, &real_arg, sizeof(real_arg))) != 0) 595 goto done2; 596 mtx_lock(sema_mtxp); 597 if ((error = semvalid(uap->semid, semaptr)) != 0) 598 goto done2; 599 if ((error = ipcperm(td, &semaptr->sem_perm, IPC_R))) 600 goto done2; 601 sbuf = *semaptr; 602 mtx_unlock(sema_mtxp); 603 error = copyout(semaptr, real_arg.buf, 604 sizeof(struct semid_ds)); 605 break; 606 607 case GETNCNT: 608 mtx_lock(sema_mtxp); 609 if ((error = semvalid(uap->semid, semaptr)) != 0) 610 goto done2; 611 if ((error = ipcperm(td, &semaptr->sem_perm, IPC_R))) 612 goto done2; 613 if (semnum < 0 || semnum >= semaptr->sem_nsems) { 614 error = EINVAL; 615 goto done2; 616 } 617 rval = semaptr->sem_base[semnum].semncnt; 618 break; 619 620 case GETPID: 621 mtx_lock(sema_mtxp); 622 if ((error = semvalid(uap->semid, semaptr)) != 0) 623 goto done2; 624 if ((error = ipcperm(td, &semaptr->sem_perm, IPC_R))) 625 goto done2; 626 if (semnum < 0 || semnum >= semaptr->sem_nsems) { 627 error = EINVAL; 628 goto done2; 629 } 630 rval = semaptr->sem_base[semnum].sempid; 631 break; 632 633 case GETVAL: 634 mtx_lock(sema_mtxp); 635 if ((error = semvalid(uap->semid, semaptr)) != 0) 636 goto done2; 637 if ((error = ipcperm(td, &semaptr->sem_perm, IPC_R))) 638 goto done2; 639 if (semnum < 0 || semnum >= semaptr->sem_nsems) { 640 error = EINVAL; 641 goto done2; 642 } 643 rval = semaptr->sem_base[semnum].semval; 644 break; 645 646 case GETALL: 647 if ((error = copyin(arg, &real_arg, sizeof(real_arg))) != 0) 648 goto done2; 649 array = malloc(sizeof(*array) * semaptr->sem_nsems, M_TEMP, 650 M_WAITOK); 651 mtx_lock(sema_mtxp); 652 if ((error = semvalid(uap->semid, semaptr)) != 0) 653 goto done2; 654 if ((error = ipcperm(td, &semaptr->sem_perm, IPC_R))) 655 goto done2; 656 for (i = 0; i < semaptr->sem_nsems; i++) 657 array[i] = semaptr->sem_base[i].semval; 658 mtx_unlock(sema_mtxp); 659 error = copyout(array, real_arg.array, 660 i * sizeof(real_arg.array[0])); 661 break; 662 663 case GETZCNT: 664 mtx_lock(sema_mtxp); 665 if ((error = semvalid(uap->semid, semaptr)) != 0) 666 goto done2; 667 if ((error = ipcperm(td, &semaptr->sem_perm, IPC_R))) 668 goto done2; 669 if (semnum < 0 || semnum >= semaptr->sem_nsems) { 670 error = EINVAL; 671 goto done2; 672 } 673 rval = semaptr->sem_base[semnum].semzcnt; 674 break; 675 676 case SETVAL: 677 if ((error = copyin(arg, &real_arg, sizeof(real_arg))) != 0) 678 goto done2; 679 mtx_lock(sema_mtxp); 680 if ((error = semvalid(uap->semid, semaptr)) != 0) 681 goto done2; 682 if ((error = ipcperm(td, &semaptr->sem_perm, IPC_W))) 683 goto done2; 684 if (semnum < 0 || semnum >= semaptr->sem_nsems) { 685 error = EINVAL; 686 goto done2; 687 } 688 if (real_arg.val < 0 || real_arg.val > seminfo.semvmx) { 689 error = ERANGE; 690 goto done2; 691 } 692 semaptr->sem_base[semnum].semval = real_arg.val; 693 SEMUNDO_LOCK(); 694 semundo_clear(semid, semnum); 695 SEMUNDO_UNLOCK(); 696 wakeup(semaptr); 697 break; 698 699 case SETALL: 700 mtx_lock(sema_mtxp); 701 raced: 702 if ((error = semvalid(uap->semid, semaptr)) != 0) 703 goto done2; 704 count = semaptr->sem_nsems; 705 mtx_unlock(sema_mtxp); 706 if ((error = copyin(arg, &real_arg, sizeof(real_arg))) != 0) 707 goto done2; 708 array = malloc(sizeof(*array) * count, M_TEMP, M_WAITOK); 709 copyin(real_arg.array, array, count * sizeof(*array)); 710 if (error) 711 break; 712 mtx_lock(sema_mtxp); 713 if ((error = semvalid(uap->semid, semaptr)) != 0) 714 goto done2; 715 /* we could have raced? */ 716 if (count != semaptr->sem_nsems) { 717 free(array, M_TEMP); 718 array = NULL; 719 goto raced; 720 } 721 if ((error = ipcperm(td, &semaptr->sem_perm, IPC_W))) 722 goto done2; 723 for (i = 0; i < semaptr->sem_nsems; i++) { 724 usval = array[i]; 725 if (usval > seminfo.semvmx) { 726 error = ERANGE; 727 break; 728 } 729 semaptr->sem_base[i].semval = usval; 730 } 731 SEMUNDO_LOCK(); 732 semundo_clear(semid, -1); 733 SEMUNDO_UNLOCK(); 734 wakeup(semaptr); 735 break; 736 737 default: 738 error = EINVAL; 739 break; 740 } 741 742 if (error == 0) 743 td->td_retval[0] = rval; 744 done2: 745 if (mtx_owned(sema_mtxp)) 746 mtx_unlock(sema_mtxp); 747 if (array != NULL) 748 free(array, M_TEMP); 749 return(error); 750 } 751 752 #ifndef _SYS_SYSPROTO_H_ 753 struct semget_args { 754 key_t key; 755 int nsems; 756 int semflg; 757 }; 758 #endif 759 760 /* 761 * MPSAFE 762 */ 763 int 764 semget(td, uap) 765 struct thread *td; 766 struct semget_args *uap; 767 { 768 int semid, error = 0; 769 int key = uap->key; 770 int nsems = uap->nsems; 771 int semflg = uap->semflg; 772 struct ucred *cred = td->td_ucred; 773 774 DPRINTF(("semget(0x%x, %d, 0%o)\n", key, nsems, semflg)); 775 if (!jail_sysvipc_allowed && jailed(td->td_ucred)) 776 return (ENOSYS); 777 778 mtx_lock(&Giant); 779 if (key != IPC_PRIVATE) { 780 for (semid = 0; semid < seminfo.semmni; semid++) { 781 if ((sema[semid].sem_perm.mode & SEM_ALLOC) && 782 sema[semid].sem_perm.key == key) 783 break; 784 } 785 if (semid < seminfo.semmni) { 786 DPRINTF(("found public key\n")); 787 if ((error = ipcperm(td, &sema[semid].sem_perm, 788 semflg & 0700))) { 789 goto done2; 790 } 791 if (nsems > 0 && sema[semid].sem_nsems < nsems) { 792 DPRINTF(("too small\n")); 793 error = EINVAL; 794 goto done2; 795 } 796 if ((semflg & IPC_CREAT) && (semflg & IPC_EXCL)) { 797 DPRINTF(("not exclusive\n")); 798 error = EEXIST; 799 goto done2; 800 } 801 goto found; 802 } 803 } 804 805 DPRINTF(("need to allocate the semid_ds\n")); 806 if (key == IPC_PRIVATE || (semflg & IPC_CREAT)) { 807 if (nsems <= 0 || nsems > seminfo.semmsl) { 808 DPRINTF(("nsems out of range (0<%d<=%d)\n", nsems, 809 seminfo.semmsl)); 810 error = EINVAL; 811 goto done2; 812 } 813 if (nsems > seminfo.semmns - semtot) { 814 DPRINTF(( 815 "not enough semaphores left (need %d, got %d)\n", 816 nsems, seminfo.semmns - semtot)); 817 error = ENOSPC; 818 goto done2; 819 } 820 for (semid = 0; semid < seminfo.semmni; semid++) { 821 if ((sema[semid].sem_perm.mode & SEM_ALLOC) == 0) 822 break; 823 } 824 if (semid == seminfo.semmni) { 825 DPRINTF(("no more semid_ds's available\n")); 826 error = ENOSPC; 827 goto done2; 828 } 829 DPRINTF(("semid %d is available\n", semid)); 830 sema[semid].sem_perm.key = key; 831 sema[semid].sem_perm.cuid = cred->cr_uid; 832 sema[semid].sem_perm.uid = cred->cr_uid; 833 sema[semid].sem_perm.cgid = cred->cr_gid; 834 sema[semid].sem_perm.gid = cred->cr_gid; 835 sema[semid].sem_perm.mode = (semflg & 0777) | SEM_ALLOC; 836 sema[semid].sem_perm.seq = 837 (sema[semid].sem_perm.seq + 1) & 0x7fff; 838 sema[semid].sem_nsems = nsems; 839 sema[semid].sem_otime = 0; 840 sema[semid].sem_ctime = time_second; 841 sema[semid].sem_base = &sem[semtot]; 842 semtot += nsems; 843 bzero(sema[semid].sem_base, 844 sizeof(sema[semid].sem_base[0])*nsems); 845 DPRINTF(("sembase = 0x%x, next = 0x%x\n", sema[semid].sem_base, 846 &sem[semtot])); 847 } else { 848 DPRINTF(("didn't find it and wasn't asked to create it\n")); 849 error = ENOENT; 850 goto done2; 851 } 852 853 found: 854 td->td_retval[0] = IXSEQ_TO_IPCID(semid, sema[semid].sem_perm); 855 done2: 856 mtx_unlock(&Giant); 857 return (error); 858 } 859 860 #ifndef _SYS_SYSPROTO_H_ 861 struct semop_args { 862 int semid; 863 struct sembuf *sops; 864 size_t nsops; 865 }; 866 #endif 867 868 /* 869 * MPSAFE 870 */ 871 int 872 semop(td, uap) 873 struct thread *td; 874 struct semop_args *uap; 875 { 876 int semid = uap->semid; 877 size_t nsops = uap->nsops; 878 struct sembuf *sops; 879 struct semid_ds *semaptr; 880 struct sembuf *sopptr = 0; 881 struct sem *semptr = 0; 882 struct sem_undo *suptr; 883 struct mtx *sema_mtxp; 884 size_t i, j, k; 885 int error; 886 int do_wakeup, do_undos; 887 888 DPRINTF(("call to semop(%d, 0x%x, %u)\n", semid, sops, nsops)); 889 890 if (!jail_sysvipc_allowed && jailed(td->td_ucred)) 891 return (ENOSYS); 892 893 semid = IPCID_TO_IX(semid); /* Convert back to zero origin */ 894 895 if (semid < 0 || semid >= seminfo.semmni) 896 return (EINVAL); 897 898 /* Allocate memory for sem_ops */ 899 if (nsops > seminfo.semopm) { 900 DPRINTF(("too many sops (max=%d, nsops=%d)\n", seminfo.semopm, 901 nsops)); 902 return (E2BIG); 903 } 904 sops = malloc(nsops * sizeof(sops[0]), M_SEM, M_WAITOK); 905 if ((error = copyin(uap->sops, sops, nsops * sizeof(sops[0]))) != 0) { 906 DPRINTF(("error = %d from copyin(%08x, %08x, %d)\n", error, 907 uap->sops, sops, nsops * sizeof(sops[0]))); 908 free(sops, M_SEM); 909 return (error); 910 } 911 912 semaptr = &sema[semid]; 913 sema_mtxp = &sema_mtx[semid]; 914 mtx_lock(sema_mtxp); 915 if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0) { 916 error = EINVAL; 917 goto done2; 918 } 919 if (semaptr->sem_perm.seq != IPCID_TO_SEQ(uap->semid)) { 920 error = EINVAL; 921 goto done2; 922 } 923 /* 924 * Initial pass thru sops to see what permissions are needed. 925 * Also perform any checks that don't need repeating on each 926 * attempt to satisfy the request vector. 927 */ 928 j = 0; /* permission needed */ 929 do_undos = 0; 930 for (i = 0; i < nsops; i++) { 931 sopptr = &sops[i]; 932 if (sopptr->sem_num >= semaptr->sem_nsems) { 933 error = EFBIG; 934 goto done2; 935 } 936 if (sopptr->sem_flg & SEM_UNDO && sopptr->sem_op != 0) 937 do_undos = 1; 938 j |= (sopptr->sem_op == 0) ? SEM_R : SEM_A; 939 } 940 941 if ((error = ipcperm(td, &semaptr->sem_perm, j))) { 942 DPRINTF(("error = %d from ipaccess\n", error)); 943 goto done2; 944 } 945 946 /* 947 * Loop trying to satisfy the vector of requests. 948 * If we reach a point where we must wait, any requests already 949 * performed are rolled back and we go to sleep until some other 950 * process wakes us up. At this point, we start all over again. 951 * 952 * This ensures that from the perspective of other tasks, a set 953 * of requests is atomic (never partially satisfied). 954 */ 955 for (;;) { 956 do_wakeup = 0; 957 error = 0; /* error return if necessary */ 958 959 for (i = 0; i < nsops; i++) { 960 sopptr = &sops[i]; 961 semptr = &semaptr->sem_base[sopptr->sem_num]; 962 963 DPRINTF(( 964 "semop: semaptr=%x, sem_base=%x, " 965 "semptr=%x, sem[%d]=%d : op=%d, flag=%s\n", 966 semaptr, semaptr->sem_base, semptr, 967 sopptr->sem_num, semptr->semval, sopptr->sem_op, 968 (sopptr->sem_flg & IPC_NOWAIT) ? 969 "nowait" : "wait")); 970 971 if (sopptr->sem_op < 0) { 972 if (semptr->semval + sopptr->sem_op < 0) { 973 DPRINTF(("semop: can't do it now\n")); 974 break; 975 } else { 976 semptr->semval += sopptr->sem_op; 977 if (semptr->semval == 0 && 978 semptr->semzcnt > 0) 979 do_wakeup = 1; 980 } 981 } else if (sopptr->sem_op == 0) { 982 if (semptr->semval != 0) { 983 DPRINTF(("semop: not zero now\n")); 984 break; 985 } 986 } else if (semptr->semval + sopptr->sem_op > 987 seminfo.semvmx) { 988 error = ERANGE; 989 break; 990 } else { 991 if (semptr->semncnt > 0) 992 do_wakeup = 1; 993 semptr->semval += sopptr->sem_op; 994 } 995 } 996 997 /* 998 * Did we get through the entire vector? 999 */ 1000 if (i >= nsops) 1001 goto done; 1002 1003 /* 1004 * No ... rollback anything that we've already done 1005 */ 1006 DPRINTF(("semop: rollback 0 through %d\n", i-1)); 1007 for (j = 0; j < i; j++) 1008 semaptr->sem_base[sops[j].sem_num].semval -= 1009 sops[j].sem_op; 1010 1011 /* If we detected an error, return it */ 1012 if (error != 0) 1013 goto done2; 1014 1015 /* 1016 * If the request that we couldn't satisfy has the 1017 * NOWAIT flag set then return with EAGAIN. 1018 */ 1019 if (sopptr->sem_flg & IPC_NOWAIT) { 1020 error = EAGAIN; 1021 goto done2; 1022 } 1023 1024 if (sopptr->sem_op == 0) 1025 semptr->semzcnt++; 1026 else 1027 semptr->semncnt++; 1028 1029 DPRINTF(("semop: good night!\n")); 1030 error = msleep(semaptr, sema_mtxp, (PZERO - 4) | PCATCH, 1031 "semwait", 0); 1032 DPRINTF(("semop: good morning (error=%d)!\n", error)); 1033 1034 if (error != 0) { 1035 error = EINTR; 1036 goto done2; 1037 } 1038 DPRINTF(("semop: good morning!\n")); 1039 1040 /* 1041 * Make sure that the semaphore still exists 1042 */ 1043 if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0 || 1044 semaptr->sem_perm.seq != IPCID_TO_SEQ(uap->semid)) { 1045 error = EIDRM; 1046 goto done2; 1047 } 1048 1049 /* 1050 * The semaphore is still alive. Readjust the count of 1051 * waiting processes. 1052 */ 1053 if (sopptr->sem_op == 0) 1054 semptr->semzcnt--; 1055 else 1056 semptr->semncnt--; 1057 } 1058 1059 done: 1060 /* 1061 * Process any SEM_UNDO requests. 1062 */ 1063 if (do_undos) { 1064 SEMUNDO_LOCK(); 1065 suptr = NULL; 1066 for (i = 0; i < nsops; i++) { 1067 /* 1068 * We only need to deal with SEM_UNDO's for non-zero 1069 * op's. 1070 */ 1071 int adjval; 1072 1073 if ((sops[i].sem_flg & SEM_UNDO) == 0) 1074 continue; 1075 adjval = sops[i].sem_op; 1076 if (adjval == 0) 1077 continue; 1078 error = semundo_adjust(td, &suptr, semid, 1079 sops[i].sem_num, -adjval); 1080 if (error == 0) 1081 continue; 1082 1083 /* 1084 * Oh-Oh! We ran out of either sem_undo's or undo's. 1085 * Rollback the adjustments to this point and then 1086 * rollback the semaphore ups and down so we can return 1087 * with an error with all structures restored. We 1088 * rollback the undo's in the exact reverse order that 1089 * we applied them. This guarantees that we won't run 1090 * out of space as we roll things back out. 1091 */ 1092 for (j = 0; j < i; j++) { 1093 k = i - j - 1; 1094 if ((sops[k].sem_flg & SEM_UNDO) == 0) 1095 continue; 1096 adjval = sops[k].sem_op; 1097 if (adjval == 0) 1098 continue; 1099 if (semundo_adjust(td, &suptr, semid, 1100 sops[k].sem_num, adjval) != 0) 1101 panic("semop - can't undo undos"); 1102 } 1103 1104 for (j = 0; j < nsops; j++) 1105 semaptr->sem_base[sops[j].sem_num].semval -= 1106 sops[j].sem_op; 1107 1108 DPRINTF(("error = %d from semundo_adjust\n", error)); 1109 SEMUNDO_UNLOCK(); 1110 goto done2; 1111 } /* loop through the sops */ 1112 SEMUNDO_UNLOCK(); 1113 } /* if (do_undos) */ 1114 1115 /* We're definitely done - set the sempid's and time */ 1116 for (i = 0; i < nsops; i++) { 1117 sopptr = &sops[i]; 1118 semptr = &semaptr->sem_base[sopptr->sem_num]; 1119 semptr->sempid = td->td_proc->p_pid; 1120 } 1121 semaptr->sem_otime = time_second; 1122 1123 /* 1124 * Do a wakeup if any semaphore was up'd whilst something was 1125 * sleeping on it. 1126 */ 1127 if (do_wakeup) { 1128 DPRINTF(("semop: doing wakeup\n")); 1129 wakeup(semaptr); 1130 DPRINTF(("semop: back from wakeup\n")); 1131 } 1132 DPRINTF(("semop: done\n")); 1133 td->td_retval[0] = 0; 1134 done2: 1135 mtx_unlock(sema_mtxp); 1136 free(sops, M_SEM); 1137 return (error); 1138 } 1139 1140 /* 1141 * Go through the undo structures for this process and apply the adjustments to 1142 * semaphores. 1143 */ 1144 static void 1145 semexit_myhook(arg, p) 1146 void *arg; 1147 struct proc *p; 1148 { 1149 struct sem_undo *suptr; 1150 struct sem_undo **supptr; 1151 1152 /* 1153 * Go through the chain of undo vectors looking for one 1154 * associated with this process. 1155 */ 1156 SEMUNDO_LOCK(); 1157 SLIST_FOREACH_PREVPTR(suptr, supptr, &semu_list, un_next) { 1158 if (suptr->un_proc == p) 1159 break; 1160 } 1161 SEMUNDO_UNLOCK(); 1162 1163 if (suptr == NULL) 1164 return; 1165 1166 DPRINTF(("proc @%08x has undo structure with %d entries\n", p, 1167 suptr->un_cnt)); 1168 1169 /* 1170 * If there are any active undo elements then process them. 1171 */ 1172 if (suptr->un_cnt > 0) { 1173 int ix; 1174 1175 for (ix = 0; ix < suptr->un_cnt; ix++) { 1176 int semid = suptr->un_ent[ix].un_id; 1177 int semnum = suptr->un_ent[ix].un_num; 1178 int adjval = suptr->un_ent[ix].un_adjval; 1179 struct semid_ds *semaptr; 1180 struct mtx *sema_mtxp; 1181 1182 semaptr = &sema[semid]; 1183 sema_mtxp = &sema_mtx[semid]; 1184 mtx_lock(sema_mtxp); 1185 SEMUNDO_LOCK(); 1186 if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0) 1187 panic("semexit - semid not allocated"); 1188 if (semnum >= semaptr->sem_nsems) 1189 panic("semexit - semnum out of range"); 1190 1191 DPRINTF(( 1192 "semexit: %08x id=%d num=%d(adj=%d) ; sem=%d\n", 1193 suptr->un_proc, suptr->un_ent[ix].un_id, 1194 suptr->un_ent[ix].un_num, 1195 suptr->un_ent[ix].un_adjval, 1196 semaptr->sem_base[semnum].semval)); 1197 1198 if (adjval < 0) { 1199 if (semaptr->sem_base[semnum].semval < -adjval) 1200 semaptr->sem_base[semnum].semval = 0; 1201 else 1202 semaptr->sem_base[semnum].semval += 1203 adjval; 1204 } else 1205 semaptr->sem_base[semnum].semval += adjval; 1206 1207 wakeup(semaptr); 1208 DPRINTF(("semexit: back from wakeup\n")); 1209 mtx_unlock(sema_mtxp); 1210 SEMUNDO_UNLOCK(); 1211 } 1212 } 1213 1214 /* 1215 * Deallocate the undo vector. 1216 */ 1217 DPRINTF(("removing vector\n")); 1218 suptr->un_proc = NULL; 1219 *supptr = SLIST_NEXT(suptr, un_next); 1220 } 1221 1222 static int 1223 sysctl_sema(SYSCTL_HANDLER_ARGS) 1224 { 1225 1226 return (SYSCTL_OUT(req, sema, 1227 sizeof(struct semid_ds) * seminfo.semmni)); 1228 } 1229