1 /* $FreeBSD: src/sys/kern/sysv_sem.c,v 1.69 2004/03/17 09:37:13 cperciva Exp $ */ 2 /* $DragonFly: src/sys/kern/sysv_sem.c,v 1.19 2008/01/06 16:55:51 swildner Exp $ */ 3 4 /* 5 * Implementation of SVID semaphores 6 * 7 * Author: Daniel Boulet 8 * 9 * This software is provided ``AS IS'' without any warranties of any kind. 10 */ 11 12 #include "opt_sysvipc.h" 13 14 #include <sys/param.h> 15 #include <sys/systm.h> 16 #include <sys/sysproto.h> 17 #include <sys/kernel.h> 18 #include <sys/proc.h> 19 #include <sys/sem.h> 20 #include <sys/sysent.h> 21 #include <sys/sysctl.h> 22 #include <sys/malloc.h> 23 #include <sys/jail.h> 24 25 #include <sys/mplock2.h> 26 27 static MALLOC_DEFINE(M_SEM, "sem", "SVID compatible semaphores"); 28 29 static void seminit (void *); 30 31 static struct sem_undo *semu_alloc (struct proc *p); 32 static int semundo_adjust (struct proc *p, struct sem_undo **supptr, 33 int semid, int semnum, int adjval); 34 static void semundo_clear (int semid, int semnum); 35 36 /* XXX casting to (sy_call_t *) is bogus, as usual. */ 37 static sy_call_t *semcalls[] = { 38 (sy_call_t *)sys___semctl, (sy_call_t *)sys_semget, 39 (sy_call_t *)sys_semop 40 }; 41 42 static int semtot = 0; 43 static struct semid_ds *sema; /* semaphore id pool */ 44 static struct sem *sem; /* semaphore pool */ 45 static struct sem_undo *semu_list; /* list of active undo structures */ 46 static int *semu; /* undo structure pool */ 47 48 struct sem { 49 u_short semval; /* semaphore value */ 50 pid_t sempid; /* pid of last operation */ 51 u_short semncnt; /* # awaiting semval > cval */ 52 u_short semzcnt; /* # awaiting semval = 0 */ 53 }; 54 55 /* 56 * Undo structure (one per process) 57 */ 58 struct sem_undo { 59 struct sem_undo *un_next; /* ptr to next active undo structure */ 60 struct proc *un_proc; /* owner of this structure */ 61 short un_cnt; /* # of active entries */ 62 struct undo { 63 short un_adjval; /* adjust on exit values */ 64 short un_num; /* semaphore # */ 65 int un_id; /* semid */ 66 } un_ent[1]; /* undo entries */ 67 }; 68 69 /* 70 * Configuration parameters 71 */ 72 #ifndef SEMMNI 73 #define SEMMNI 10 /* # of semaphore identifiers */ 74 #endif 75 #ifndef SEMMNS 76 #define SEMMNS 60 /* # of semaphores in system */ 77 #endif 78 #ifndef SEMUME 79 #define SEMUME 10 /* max # of undo entries per process */ 80 #endif 81 #ifndef SEMMNU 82 #define SEMMNU 30 /* # of undo structures in system */ 83 #endif 84 85 /* shouldn't need tuning */ 86 #ifndef SEMMAP 87 #define SEMMAP 30 /* # of entries in semaphore map */ 88 #endif 89 #ifndef SEMMSL 90 #define SEMMSL SEMMNS /* max # of semaphores per id */ 91 #endif 92 #ifndef SEMOPM 93 #define SEMOPM 100 /* max # of operations per semop call */ 94 #endif 95 96 #define SEMVMX 32767 /* semaphore maximum value */ 97 #define SEMAEM 16384 /* adjust on exit max value */ 98 99 /* 100 * Due to the way semaphore memory is allocated, we have to ensure that 101 * SEMUSZ is properly aligned. 102 */ 103 104 #define SEM_ALIGN(bytes) (((bytes) + (sizeof(long) - 1)) & ~(sizeof(long) - 1)) 105 106 /* actual size of an undo structure */ 107 #define SEMUSZ SEM_ALIGN(offsetof(struct sem_undo, un_ent[SEMUME])) 108 109 /* 110 * Macro to find a particular sem_undo vector 111 */ 112 #define SEMU(ix) ((struct sem_undo *)(((intptr_t)semu)+ix * seminfo.semusz)) 113 114 /* 115 * semaphore info struct 116 */ 117 struct seminfo seminfo = { 118 SEMMAP, /* # of entries in semaphore map */ 119 SEMMNI, /* # of semaphore identifiers */ 120 SEMMNS, /* # of semaphores in system */ 121 SEMMNU, /* # of undo structures in system */ 122 SEMMSL, /* max # of semaphores per id */ 123 SEMOPM, /* max # of operations per semop call */ 124 SEMUME, /* max # of undo entries per process */ 125 SEMUSZ, /* size in bytes of undo structure */ 126 SEMVMX, /* semaphore maximum value */ 127 SEMAEM /* adjust on exit max value */ 128 }; 129 130 TUNABLE_INT("kern.ipc.semmap", &seminfo.semmap); 131 TUNABLE_INT("kern.ipc.semmni", &seminfo.semmni); 132 TUNABLE_INT("kern.ipc.semmns", &seminfo.semmns); 133 TUNABLE_INT("kern.ipc.semmnu", &seminfo.semmnu); 134 TUNABLE_INT("kern.ipc.semmsl", &seminfo.semmsl); 135 TUNABLE_INT("kern.ipc.semopm", &seminfo.semopm); 136 TUNABLE_INT("kern.ipc.semume", &seminfo.semume); 137 TUNABLE_INT("kern.ipc.semusz", &seminfo.semusz); 138 TUNABLE_INT("kern.ipc.semvmx", &seminfo.semvmx); 139 TUNABLE_INT("kern.ipc.semaem", &seminfo.semaem); 140 141 SYSCTL_INT(_kern_ipc, OID_AUTO, semmap, CTLFLAG_RW, &seminfo.semmap, 0, ""); 142 SYSCTL_INT(_kern_ipc, OID_AUTO, semmni, CTLFLAG_RD, &seminfo.semmni, 0, ""); 143 SYSCTL_INT(_kern_ipc, OID_AUTO, semmns, CTLFLAG_RD, &seminfo.semmns, 0, ""); 144 SYSCTL_INT(_kern_ipc, OID_AUTO, semmnu, CTLFLAG_RD, &seminfo.semmnu, 0, ""); 145 SYSCTL_INT(_kern_ipc, OID_AUTO, semmsl, CTLFLAG_RW, &seminfo.semmsl, 0, ""); 146 SYSCTL_INT(_kern_ipc, OID_AUTO, semopm, CTLFLAG_RD, &seminfo.semopm, 0, ""); 147 SYSCTL_INT(_kern_ipc, OID_AUTO, semume, CTLFLAG_RD, &seminfo.semume, 0, ""); 148 SYSCTL_INT(_kern_ipc, OID_AUTO, semusz, CTLFLAG_RD, &seminfo.semusz, 0, ""); 149 SYSCTL_INT(_kern_ipc, OID_AUTO, semvmx, CTLFLAG_RW, &seminfo.semvmx, 0, ""); 150 SYSCTL_INT(_kern_ipc, OID_AUTO, semaem, CTLFLAG_RW, &seminfo.semaem, 0, ""); 151 152 #if 0 153 RO seminfo.semmap /* SEMMAP unused */ 154 RO seminfo.semmni 155 RO seminfo.semmns 156 RO seminfo.semmnu /* undo entries per system */ 157 RW seminfo.semmsl 158 RO seminfo.semopm /* SEMOPM unused */ 159 RO seminfo.semume 160 RO seminfo.semusz /* param - derived from SEMUME for per-proc sizeof */ 161 RO seminfo.semvmx /* SEMVMX unused - user param */ 162 RO seminfo.semaem /* SEMAEM unused - user param */ 163 #endif 164 165 static void 166 seminit(void *dummy) 167 { 168 int i; 169 170 sem = kmalloc(sizeof(struct sem) * seminfo.semmns, M_SEM, M_WAITOK); 171 sema = kmalloc(sizeof(struct semid_ds) * seminfo.semmni, M_SEM, M_WAITOK); 172 semu = kmalloc(seminfo.semmnu * seminfo.semusz, M_SEM, M_WAITOK); 173 174 for (i = 0; i < seminfo.semmni; i++) { 175 sema[i].sem_base = 0; 176 sema[i].sem_perm.mode = 0; 177 } 178 for (i = 0; i < seminfo.semmnu; i++) { 179 struct sem_undo *suptr = SEMU(i); 180 suptr->un_proc = NULL; 181 } 182 semu_list = NULL; 183 } 184 SYSINIT(sysv_sem, SI_SUB_SYSV_SEM, SI_ORDER_FIRST, seminit, NULL) 185 186 /* 187 * Entry point for all SEM calls 188 * 189 * semsys_args(int which, a2, a3, ...) (VARARGS) 190 * 191 * MPALMOSTSAFE 192 */ 193 int 194 sys_semsys(struct semsys_args *uap) 195 { 196 struct thread *td = curthread; 197 unsigned int which = (unsigned int)uap->which; 198 int error; 199 200 if (!jail_sysvipc_allowed && td->td_ucred->cr_prison != NULL) 201 return (ENOSYS); 202 203 if (which >= sizeof(semcalls)/sizeof(semcalls[0])) 204 return (EINVAL); 205 bcopy(&uap->a2, &uap->which, 206 sizeof(struct semsys_args) - offsetof(struct semsys_args, a2)); 207 get_mplock(); 208 error = (*semcalls[which])(uap); 209 rel_mplock(); 210 return (error); 211 } 212 213 /* 214 * Allocate a new sem_undo structure for a process 215 * (returns ptr to structure or NULL if no more room) 216 */ 217 218 static struct sem_undo * 219 semu_alloc(struct proc *p) 220 { 221 int i; 222 struct sem_undo *suptr; 223 struct sem_undo **supptr; 224 int attempt; 225 226 /* 227 * Try twice to allocate something. 228 * (we'll purge any empty structures after the first pass so 229 * two passes are always enough) 230 */ 231 232 for (attempt = 0; attempt < 2; attempt++) { 233 /* 234 * Look for a free structure. 235 * Fill it in and return it if we find one. 236 */ 237 238 for (i = 0; i < seminfo.semmnu; i++) { 239 suptr = SEMU(i); 240 if (suptr->un_proc == NULL) { 241 suptr->un_next = semu_list; 242 semu_list = suptr; 243 suptr->un_cnt = 0; 244 suptr->un_proc = p; 245 return(suptr); 246 } 247 } 248 249 /* 250 * We didn't find a free one, if this is the first attempt 251 * then try to free some structures. 252 */ 253 254 if (attempt == 0) { 255 /* All the structures are in use - try to free some */ 256 int did_something = 0; 257 258 supptr = &semu_list; 259 while ((suptr = *supptr) != NULL) { 260 if (suptr->un_cnt == 0) { 261 suptr->un_proc = NULL; 262 *supptr = suptr->un_next; 263 did_something = 1; 264 } else 265 supptr = &(suptr->un_next); 266 } 267 268 /* If we didn't free anything then just give-up */ 269 if (!did_something) 270 return(NULL); 271 } else { 272 /* 273 * The second pass failed even though we freed 274 * something after the first pass! 275 * This is IMPOSSIBLE! 276 */ 277 panic("semu_alloc - second attempt failed"); 278 } 279 } 280 return (NULL); 281 } 282 283 /* 284 * Adjust a particular entry for a particular proc 285 */ 286 287 static int 288 semundo_adjust(struct proc *p, struct sem_undo **supptr, int semid, int semnum, 289 int adjval) 290 { 291 struct sem_undo *suptr; 292 struct undo *sunptr; 293 int i; 294 295 /* Look for and remember the sem_undo if the caller doesn't provide 296 it */ 297 298 suptr = *supptr; 299 if (suptr == NULL) { 300 for (suptr = semu_list; suptr != NULL; 301 suptr = suptr->un_next) { 302 if (suptr->un_proc == p) { 303 *supptr = suptr; 304 break; 305 } 306 } 307 if (suptr == NULL) { 308 if (adjval == 0) 309 return(0); 310 suptr = semu_alloc(p); 311 if (suptr == NULL) 312 return(ENOSPC); 313 *supptr = suptr; 314 } 315 } 316 317 /* 318 * Look for the requested entry and adjust it (delete if adjval becomes 319 * 0). 320 */ 321 sunptr = &suptr->un_ent[0]; 322 for (i = 0; i < suptr->un_cnt; i++, sunptr++) { 323 if (sunptr->un_id != semid || sunptr->un_num != semnum) 324 continue; 325 if (adjval == 0) 326 sunptr->un_adjval = 0; 327 else 328 sunptr->un_adjval += adjval; 329 if (sunptr->un_adjval == 0) { 330 suptr->un_cnt--; 331 if (i < suptr->un_cnt) 332 suptr->un_ent[i] = 333 suptr->un_ent[suptr->un_cnt]; 334 } 335 return(0); 336 } 337 338 /* Didn't find the right entry - create it */ 339 if (adjval == 0) 340 return(0); 341 if (suptr->un_cnt != seminfo.semume) { 342 sunptr = &suptr->un_ent[suptr->un_cnt]; 343 suptr->un_cnt++; 344 sunptr->un_adjval = adjval; 345 sunptr->un_id = semid; sunptr->un_num = semnum; 346 } else 347 return(EINVAL); 348 return(0); 349 } 350 351 static void 352 semundo_clear(int semid, int semnum) 353 { 354 struct sem_undo *suptr; 355 356 for (suptr = semu_list; suptr != NULL; suptr = suptr->un_next) { 357 struct undo *sunptr = &suptr->un_ent[0]; 358 int i = 0; 359 360 while (i < suptr->un_cnt) { 361 if (sunptr->un_id == semid) { 362 if (semnum == -1 || sunptr->un_num == semnum) { 363 suptr->un_cnt--; 364 if (i < suptr->un_cnt) { 365 suptr->un_ent[i] = 366 suptr->un_ent[suptr->un_cnt]; 367 continue; 368 } 369 } 370 if (semnum != -1) 371 break; 372 } 373 i++, sunptr++; 374 } 375 } 376 } 377 378 /* 379 * Note that the user-mode half of this passes a union, not a pointer 380 * 381 * MPALMOSTSAFE 382 */ 383 int 384 sys___semctl(struct __semctl_args *uap) 385 { 386 struct thread *td = curthread; 387 int semid = uap->semid; 388 int semnum = uap->semnum; 389 int cmd = uap->cmd; 390 union semun *arg = uap->arg; 391 union semun real_arg; 392 struct ucred *cred = td->td_ucred; 393 int i, rval, eval; 394 struct semid_ds sbuf; 395 struct semid_ds *semaptr; 396 397 #ifdef SEM_DEBUG 398 kprintf("call to semctl(%d, %d, %d, 0x%x)\n", semid, semnum, cmd, arg); 399 #endif 400 401 if (!jail_sysvipc_allowed && cred->cr_prison != NULL) 402 return (ENOSYS); 403 404 semid = IPCID_TO_IX(semid); 405 if (semid < 0 || semid >= seminfo.semmni) 406 return(EINVAL); 407 408 get_mplock(); 409 semaptr = &sema[semid]; 410 if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0 || 411 semaptr->sem_perm.seq != IPCID_TO_SEQ(uap->semid)) { 412 rel_mplock(); 413 return(EINVAL); 414 } 415 416 eval = 0; 417 rval = 0; 418 419 switch (cmd) { 420 case IPC_RMID: 421 if ((eval = ipcperm(td->td_proc, &semaptr->sem_perm, IPC_M)) != 0) 422 break; 423 semaptr->sem_perm.cuid = cred->cr_uid; 424 semaptr->sem_perm.uid = cred->cr_uid; 425 semtot -= semaptr->sem_nsems; 426 for (i = semaptr->sem_base - sem; i < semtot; i++) 427 sem[i] = sem[i + semaptr->sem_nsems]; 428 for (i = 0; i < seminfo.semmni; i++) { 429 if ((sema[i].sem_perm.mode & SEM_ALLOC) && 430 sema[i].sem_base > semaptr->sem_base) 431 sema[i].sem_base -= semaptr->sem_nsems; 432 } 433 semaptr->sem_perm.mode = 0; 434 semundo_clear(semid, -1); 435 wakeup((caddr_t)semaptr); 436 break; 437 438 case IPC_SET: 439 eval = ipcperm(td->td_proc, &semaptr->sem_perm, IPC_M); 440 if (eval) 441 break; 442 if ((eval = copyin(arg, &real_arg, sizeof(real_arg))) != 0) 443 break; 444 if ((eval = copyin(real_arg.buf, (caddr_t)&sbuf, 445 sizeof(sbuf))) != 0) { 446 break; 447 } 448 semaptr->sem_perm.uid = sbuf.sem_perm.uid; 449 semaptr->sem_perm.gid = sbuf.sem_perm.gid; 450 semaptr->sem_perm.mode = (semaptr->sem_perm.mode & ~0777) | 451 (sbuf.sem_perm.mode & 0777); 452 semaptr->sem_ctime = time_second; 453 break; 454 455 case IPC_STAT: 456 if ((eval = ipcperm(td->td_proc, &semaptr->sem_perm, IPC_R))) 457 break; 458 if ((eval = copyin(arg, &real_arg, sizeof(real_arg))) != 0) 459 break; 460 eval = copyout(semaptr, real_arg.buf, sizeof(struct semid_ds)); 461 break; 462 463 case GETNCNT: 464 eval = ipcperm(td->td_proc, &semaptr->sem_perm, IPC_R); 465 if (eval) 466 break; 467 if (semnum < 0 || semnum >= semaptr->sem_nsems) { 468 eval = EINVAL; 469 break; 470 } 471 rval = semaptr->sem_base[semnum].semncnt; 472 break; 473 474 case GETPID: 475 eval = ipcperm(td->td_proc, &semaptr->sem_perm, IPC_R); 476 if (eval) 477 break; 478 if (semnum < 0 || semnum >= semaptr->sem_nsems) { 479 eval = EINVAL; 480 break; 481 } 482 rval = semaptr->sem_base[semnum].sempid; 483 break; 484 485 case GETVAL: 486 eval = ipcperm(td->td_proc, &semaptr->sem_perm, IPC_R); 487 if (eval) 488 break; 489 if (semnum < 0 || semnum >= semaptr->sem_nsems) { 490 eval = EINVAL; 491 break; 492 } 493 rval = semaptr->sem_base[semnum].semval; 494 break; 495 496 case GETALL: 497 eval = ipcperm(td->td_proc, &semaptr->sem_perm, IPC_R); 498 if (eval) 499 break; 500 if ((eval = copyin(arg, &real_arg, sizeof(real_arg))) != 0) 501 break; 502 for (i = 0; i < semaptr->sem_nsems; i++) { 503 eval = copyout(&semaptr->sem_base[i].semval, 504 &real_arg.array[i], 505 sizeof(real_arg.array[0])); 506 if (eval) 507 break; 508 } 509 break; 510 511 case GETZCNT: 512 eval = ipcperm(td->td_proc, &semaptr->sem_perm, IPC_R); 513 if (eval) 514 break; 515 if (semnum < 0 || semnum >= semaptr->sem_nsems) { 516 eval = EINVAL; 517 break; 518 } 519 rval = semaptr->sem_base[semnum].semzcnt; 520 break; 521 522 case SETVAL: 523 eval = ipcperm(td->td_proc, &semaptr->sem_perm, IPC_W); 524 if (eval) 525 break; 526 if (semnum < 0 || semnum >= semaptr->sem_nsems) { 527 eval = EINVAL; 528 break; 529 } 530 if ((eval = copyin(arg, &real_arg, sizeof(real_arg))) != 0) 531 break; 532 semaptr->sem_base[semnum].semval = real_arg.val; 533 semundo_clear(semid, semnum); 534 wakeup((caddr_t)semaptr); 535 break; 536 537 case SETALL: 538 eval = ipcperm(td->td_proc, &semaptr->sem_perm, IPC_W); 539 if (eval) 540 break; 541 if ((eval = copyin(arg, &real_arg, sizeof(real_arg))) != 0) 542 break; 543 for (i = 0; i < semaptr->sem_nsems; i++) { 544 eval = copyin(&real_arg.array[i], 545 (caddr_t)&semaptr->sem_base[i].semval, 546 sizeof(real_arg.array[0])); 547 if (eval != 0) 548 break; 549 } 550 semundo_clear(semid, -1); 551 wakeup((caddr_t)semaptr); 552 break; 553 554 default: 555 eval = EINVAL; 556 break; 557 } 558 rel_mplock(); 559 560 if (eval == 0) 561 uap->sysmsg_result = rval; 562 return(eval); 563 } 564 565 /* 566 * MPALMOSTSAFE 567 */ 568 int 569 sys_semget(struct semget_args *uap) 570 { 571 struct thread *td = curthread; 572 int semid, eval; 573 int key = uap->key; 574 int nsems = uap->nsems; 575 int semflg = uap->semflg; 576 struct ucred *cred = td->td_ucred; 577 578 #ifdef SEM_DEBUG 579 kprintf("semget(0x%x, %d, 0%o)\n", key, nsems, semflg); 580 #endif 581 582 if (!jail_sysvipc_allowed && cred->cr_prison != NULL) 583 return (ENOSYS); 584 585 get_mplock(); 586 eval = 0; 587 588 if (key != IPC_PRIVATE) { 589 for (semid = 0; semid < seminfo.semmni; semid++) { 590 if ((sema[semid].sem_perm.mode & SEM_ALLOC) && 591 sema[semid].sem_perm.key == key) 592 break; 593 } 594 if (semid < seminfo.semmni) { 595 #ifdef SEM_DEBUG 596 kprintf("found public key\n"); 597 #endif 598 if ((eval = ipcperm(td->td_proc, 599 &sema[semid].sem_perm, 600 semflg & 0700))) { 601 goto done; 602 } 603 if (nsems > 0 && sema[semid].sem_nsems < nsems) { 604 #ifdef SEM_DEBUG 605 kprintf("too small\n"); 606 #endif 607 eval = EINVAL; 608 goto done; 609 } 610 if ((semflg & IPC_CREAT) && (semflg & IPC_EXCL)) { 611 #ifdef SEM_DEBUG 612 kprintf("not exclusive\n"); 613 #endif 614 eval = EEXIST; 615 goto done; 616 } 617 goto done; 618 } 619 } 620 621 #ifdef SEM_DEBUG 622 kprintf("need to allocate the semid_ds\n"); 623 #endif 624 if (key == IPC_PRIVATE || (semflg & IPC_CREAT)) { 625 if (nsems <= 0 || nsems > seminfo.semmsl) { 626 #ifdef SEM_DEBUG 627 kprintf("nsems out of range (0<%d<=%d)\n", nsems, 628 seminfo.semmsl); 629 #endif 630 eval = EINVAL; 631 goto done; 632 } 633 if (nsems > seminfo.semmns - semtot) { 634 #ifdef SEM_DEBUG 635 kprintf("not enough semaphores left (need %d, got %d)\n", 636 nsems, seminfo.semmns - semtot); 637 #endif 638 eval = ENOSPC; 639 goto done; 640 } 641 for (semid = 0; semid < seminfo.semmni; semid++) { 642 if ((sema[semid].sem_perm.mode & SEM_ALLOC) == 0) 643 break; 644 } 645 if (semid == seminfo.semmni) { 646 #ifdef SEM_DEBUG 647 kprintf("no more semid_ds's available\n"); 648 #endif 649 eval = ENOSPC; 650 goto done; 651 } 652 #ifdef SEM_DEBUG 653 kprintf("semid %d is available\n", semid); 654 #endif 655 sema[semid].sem_perm.key = key; 656 sema[semid].sem_perm.cuid = cred->cr_uid; 657 sema[semid].sem_perm.uid = cred->cr_uid; 658 sema[semid].sem_perm.cgid = cred->cr_gid; 659 sema[semid].sem_perm.gid = cred->cr_gid; 660 sema[semid].sem_perm.mode = (semflg & 0777) | SEM_ALLOC; 661 sema[semid].sem_perm.seq = 662 (sema[semid].sem_perm.seq + 1) & 0x7fff; 663 sema[semid].sem_nsems = nsems; 664 sema[semid].sem_otime = 0; 665 sema[semid].sem_ctime = time_second; 666 sema[semid].sem_base = &sem[semtot]; 667 semtot += nsems; 668 bzero(sema[semid].sem_base, 669 sizeof(sema[semid].sem_base[0])*nsems); 670 #ifdef SEM_DEBUG 671 kprintf("sembase = 0x%x, next = 0x%x\n", sema[semid].sem_base, 672 &sem[semtot]); 673 #endif 674 } else { 675 #ifdef SEM_DEBUG 676 kprintf("didn't find it and wasn't asked to create it\n"); 677 #endif 678 eval = ENOENT; 679 } 680 681 done: 682 if (eval == 0) { 683 uap->sysmsg_result = IXSEQ_TO_IPCID(semid, 684 sema[semid].sem_perm); 685 } 686 rel_mplock(); 687 return(eval); 688 } 689 690 /* 691 * MPALMOSTSAFE 692 */ 693 int 694 sys_semop(struct semop_args *uap) 695 { 696 struct thread *td = curthread; 697 int semid = uap->semid; 698 u_int nsops = uap->nsops; 699 struct sembuf sops[MAX_SOPS]; 700 struct semid_ds *semaptr; 701 struct sembuf *sopptr; 702 struct sem *semptr; 703 struct sem_undo *suptr = NULL; 704 int i, j, eval; 705 int do_wakeup, do_undos; 706 707 #ifdef SEM_DEBUG 708 kprintf("call to semop(%d, 0x%x, %u)\n", semid, sops, nsops); 709 #endif 710 711 if (!jail_sysvipc_allowed && td->td_ucred->cr_prison != NULL) 712 return (ENOSYS); 713 714 get_mplock(); 715 semid = IPCID_TO_IX(semid); /* Convert back to zero origin */ 716 717 if (semid < 0 || semid >= seminfo.semmni) { 718 eval = EINVAL; 719 goto done; 720 } 721 722 semaptr = &sema[semid]; 723 if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0) { 724 eval = EINVAL; 725 goto done; 726 } 727 if (semaptr->sem_perm.seq != IPCID_TO_SEQ(uap->semid)) { 728 eval = EINVAL; 729 goto done; 730 } 731 732 if ((eval = ipcperm(td->td_proc, &semaptr->sem_perm, IPC_W))) { 733 #ifdef SEM_DEBUG 734 kprintf("eval = %d from ipaccess\n", eval); 735 #endif 736 goto done; 737 } 738 739 if (nsops > MAX_SOPS) { 740 #ifdef SEM_DEBUG 741 kprintf("too many sops (max=%d, nsops=%u)\n", MAX_SOPS, nsops); 742 #endif 743 eval = E2BIG; 744 goto done; 745 } 746 747 if ((eval = copyin(uap->sops, &sops, nsops * sizeof(sops[0]))) != 0) { 748 #ifdef SEM_DEBUG 749 kprintf("eval = %d from copyin(%08x, %08x, %u)\n", eval, 750 uap->sops, &sops, nsops * sizeof(sops[0])); 751 #endif 752 goto done; 753 } 754 755 /* 756 * Loop trying to satisfy the vector of requests. 757 * If we reach a point where we must wait, any requests already 758 * performed are rolled back and we go to sleep until some other 759 * process wakes us up. At this point, we start all over again. 760 * 761 * This ensures that from the perspective of other tasks, a set 762 * of requests is atomic (never partially satisfied). 763 */ 764 do_undos = 0; 765 766 for (;;) { 767 do_wakeup = 0; 768 769 for (i = 0; i < nsops; i++) { 770 sopptr = &sops[i]; 771 772 if (sopptr->sem_num >= semaptr->sem_nsems) { 773 eval = EFBIG; 774 goto done; 775 } 776 777 semptr = &semaptr->sem_base[sopptr->sem_num]; 778 779 #ifdef SEM_DEBUG 780 kprintf("semop: semaptr=%x, sem_base=%x, semptr=%x, sem[%d]=%d : op=%d, flag=%s\n", 781 semaptr, semaptr->sem_base, semptr, 782 sopptr->sem_num, semptr->semval, sopptr->sem_op, 783 (sopptr->sem_flg & IPC_NOWAIT) ? "nowait" : "wait"); 784 #endif 785 786 if (sopptr->sem_op < 0) { 787 if (semptr->semval + sopptr->sem_op < 0) { 788 #ifdef SEM_DEBUG 789 kprintf("semop: can't do it now\n"); 790 #endif 791 break; 792 } else { 793 semptr->semval += sopptr->sem_op; 794 if (semptr->semval == 0 && 795 semptr->semzcnt > 0) 796 do_wakeup = 1; 797 } 798 if (sopptr->sem_flg & SEM_UNDO) 799 do_undos = 1; 800 } else if (sopptr->sem_op == 0) { 801 if (semptr->semval > 0) { 802 #ifdef SEM_DEBUG 803 kprintf("semop: not zero now\n"); 804 #endif 805 break; 806 } 807 } else { 808 if (semptr->semncnt > 0) 809 do_wakeup = 1; 810 semptr->semval += sopptr->sem_op; 811 if (sopptr->sem_flg & SEM_UNDO) 812 do_undos = 1; 813 } 814 } 815 816 /* 817 * Did we get through the entire vector? 818 */ 819 if (i >= nsops) 820 goto donex; 821 822 /* 823 * No ... rollback anything that we've already done 824 */ 825 #ifdef SEM_DEBUG 826 kprintf("semop: rollback 0 through %d\n", i-1); 827 #endif 828 for (j = 0; j < i; j++) 829 semaptr->sem_base[sops[j].sem_num].semval -= 830 sops[j].sem_op; 831 832 /* 833 * If the request that we couldn't satisfy has the 834 * NOWAIT flag set then return with EAGAIN. 835 */ 836 if (sopptr->sem_flg & IPC_NOWAIT) { 837 eval = EAGAIN; 838 goto done; 839 } 840 841 if (sopptr->sem_op == 0) 842 semptr->semzcnt++; 843 else 844 semptr->semncnt++; 845 846 #ifdef SEM_DEBUG 847 kprintf("semop: good night!\n"); 848 #endif 849 eval = tsleep((caddr_t)semaptr, PCATCH, "semwait", 0); 850 #ifdef SEM_DEBUG 851 kprintf("semop: good morning (eval=%d)!\n", eval); 852 #endif 853 854 suptr = NULL; /* sem_undo may have been reallocated */ 855 856 /* return code is checked below, after sem[nz]cnt-- */ 857 858 /* 859 * Make sure that the semaphore still exists 860 */ 861 if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0 || 862 semaptr->sem_perm.seq != IPCID_TO_SEQ(uap->semid)) { 863 eval = EIDRM; 864 goto done; 865 } 866 867 /* 868 * The semaphore is still alive. Readjust the count of 869 * waiting processes. 870 */ 871 if (sopptr->sem_op == 0) 872 semptr->semzcnt--; 873 else 874 semptr->semncnt--; 875 876 /* 877 * Is it really morning, or was our sleep interrupted? 878 * (Delayed check of tsleep() return code because we 879 * need to decrement sem[nz]cnt either way.) 880 */ 881 if (eval) { 882 eval = EINTR; 883 goto done; 884 } 885 #ifdef SEM_DEBUG 886 kprintf("semop: good morning!\n"); 887 #endif 888 } 889 890 donex: 891 /* 892 * Process any SEM_UNDO requests. 893 */ 894 if (do_undos) { 895 for (i = 0; i < nsops; i++) { 896 /* 897 * We only need to deal with SEM_UNDO's for non-zero 898 * op's. 899 */ 900 int adjval; 901 902 if ((sops[i].sem_flg & SEM_UNDO) == 0) 903 continue; 904 adjval = sops[i].sem_op; 905 if (adjval == 0) 906 continue; 907 eval = semundo_adjust(td->td_proc, &suptr, semid, 908 sops[i].sem_num, -adjval); 909 if (eval == 0) 910 continue; 911 912 /* 913 * Oh-Oh! We ran out of either sem_undo's or undo's. 914 * Rollback the adjustments to this point and then 915 * rollback the semaphore ups and down so we can return 916 * with an error with all structures restored. We 917 * rollback the undo's in the exact reverse order that 918 * we applied them. This guarantees that we won't run 919 * out of space as we roll things back out. 920 */ 921 for (j = i - 1; j >= 0; j--) { 922 if ((sops[j].sem_flg & SEM_UNDO) == 0) 923 continue; 924 adjval = sops[j].sem_op; 925 if (adjval == 0) 926 continue; 927 if (semundo_adjust(td->td_proc, &suptr, semid, 928 sops[j].sem_num, adjval) != 0) 929 panic("semop - can't undo undos"); 930 } 931 932 for (j = 0; j < nsops; j++) 933 semaptr->sem_base[sops[j].sem_num].semval -= 934 sops[j].sem_op; 935 936 #ifdef SEM_DEBUG 937 kprintf("eval = %d from semundo_adjust\n", eval); 938 #endif 939 goto done; 940 } /* loop through the sops */ 941 } /* if (do_undos) */ 942 943 /* We're definitely done - set the sempid's */ 944 for (i = 0; i < nsops; i++) { 945 sopptr = &sops[i]; 946 semptr = &semaptr->sem_base[sopptr->sem_num]; 947 semptr->sempid = td->td_proc->p_pid; 948 } 949 950 /* Do a wakeup if any semaphore was up'd. */ 951 if (do_wakeup) { 952 #ifdef SEM_DEBUG 953 kprintf("semop: doing wakeup\n"); 954 #endif 955 wakeup((caddr_t)semaptr); 956 #ifdef SEM_DEBUG 957 kprintf("semop: back from wakeup\n"); 958 #endif 959 } 960 #ifdef SEM_DEBUG 961 kprintf("semop: done\n"); 962 #endif 963 uap->sysmsg_result = 0; 964 eval = 0; 965 done: 966 rel_mplock(); 967 return(eval); 968 } 969 970 /* 971 * Go through the undo structures for this process and apply the adjustments to 972 * semaphores. 973 */ 974 void 975 semexit(struct proc *p) 976 { 977 struct sem_undo *suptr; 978 struct sem_undo **supptr; 979 int did_something; 980 981 did_something = 0; 982 983 /* 984 * Go through the chain of undo vectors looking for one 985 * associated with this process. 986 */ 987 988 for (supptr = &semu_list; (suptr = *supptr) != NULL; 989 supptr = &suptr->un_next) { 990 if (suptr->un_proc == p) 991 break; 992 } 993 994 if (suptr == NULL) 995 return; 996 997 #ifdef SEM_DEBUG 998 kprintf("proc @%08x has undo structure with %d entries\n", p, 999 suptr->un_cnt); 1000 #endif 1001 1002 /* 1003 * If there are any active undo elements then process them. 1004 */ 1005 if (suptr->un_cnt > 0) { 1006 int ix; 1007 1008 for (ix = 0; ix < suptr->un_cnt; ix++) { 1009 int semid = suptr->un_ent[ix].un_id; 1010 int semnum = suptr->un_ent[ix].un_num; 1011 int adjval = suptr->un_ent[ix].un_adjval; 1012 struct semid_ds *semaptr; 1013 1014 semaptr = &sema[semid]; 1015 if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0) 1016 panic("semexit - semid not allocated"); 1017 if (semnum >= semaptr->sem_nsems) 1018 panic("semexit - semnum out of range"); 1019 1020 #ifdef SEM_DEBUG 1021 kprintf("semexit: %08x id=%d num=%d(adj=%d) ; sem=%d\n", 1022 suptr->un_proc, suptr->un_ent[ix].un_id, 1023 suptr->un_ent[ix].un_num, 1024 suptr->un_ent[ix].un_adjval, 1025 semaptr->sem_base[semnum].semval); 1026 #endif 1027 1028 if (adjval < 0) { 1029 if (semaptr->sem_base[semnum].semval < -adjval) 1030 semaptr->sem_base[semnum].semval = 0; 1031 else 1032 semaptr->sem_base[semnum].semval += 1033 adjval; 1034 } else 1035 semaptr->sem_base[semnum].semval += adjval; 1036 1037 wakeup((caddr_t)semaptr); 1038 #ifdef SEM_DEBUG 1039 kprintf("semexit: back from wakeup\n"); 1040 #endif 1041 } 1042 } 1043 1044 /* 1045 * Deallocate the undo vector. 1046 */ 1047 #ifdef SEM_DEBUG 1048 kprintf("removing vector\n"); 1049 #endif 1050 suptr->un_proc = NULL; 1051 *supptr = suptr->un_next; 1052 } 1053