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 1.18 2006/12/23 23:47:54 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 static MALLOC_DEFINE(M_SEM, "sem", "SVID compatible semaphores"); 26 27 static void seminit (void *); 28 29 static struct sem_undo *semu_alloc (struct proc *p); 30 static int semundo_adjust (struct proc *p, struct sem_undo **supptr, 31 int semid, int semnum, int adjval); 32 static void semundo_clear (int semid, int semnum); 33 34 /* XXX casting to (sy_call_t *) is bogus, as usual. */ 35 static sy_call_t *semcalls[] = { 36 (sy_call_t *)sys___semctl, (sy_call_t *)sys_semget, 37 (sy_call_t *)sys_semop 38 }; 39 40 static int semtot = 0; 41 static struct semid_ds *sema; /* semaphore id pool */ 42 static struct sem *sem; /* semaphore pool */ 43 static struct sem_undo *semu_list; /* list of active undo structures */ 44 static int *semu; /* undo structure pool */ 45 46 struct sem { 47 u_short semval; /* semaphore value */ 48 pid_t sempid; /* pid of last operation */ 49 u_short semncnt; /* # awaiting semval > cval */ 50 u_short semzcnt; /* # awaiting semval = 0 */ 51 }; 52 53 /* 54 * Undo structure (one per process) 55 */ 56 struct sem_undo { 57 struct sem_undo *un_next; /* ptr to next active undo structure */ 58 struct proc *un_proc; /* owner of this structure */ 59 short un_cnt; /* # of active entries */ 60 struct undo { 61 short un_adjval; /* adjust on exit values */ 62 short un_num; /* semaphore # */ 63 int un_id; /* semid */ 64 } un_ent[1]; /* undo entries */ 65 }; 66 67 /* 68 * Configuration parameters 69 */ 70 #ifndef SEMMNI 71 #define SEMMNI 10 /* # of semaphore identifiers */ 72 #endif 73 #ifndef SEMMNS 74 #define SEMMNS 60 /* # of semaphores in system */ 75 #endif 76 #ifndef SEMUME 77 #define SEMUME 10 /* max # of undo entries per process */ 78 #endif 79 #ifndef SEMMNU 80 #define SEMMNU 30 /* # of undo structures in system */ 81 #endif 82 83 /* shouldn't need tuning */ 84 #ifndef SEMMAP 85 #define SEMMAP 30 /* # of entries in semaphore map */ 86 #endif 87 #ifndef SEMMSL 88 #define SEMMSL SEMMNS /* max # of semaphores per id */ 89 #endif 90 #ifndef SEMOPM 91 #define SEMOPM 100 /* max # of operations per semop call */ 92 #endif 93 94 #define SEMVMX 32767 /* semaphore maximum value */ 95 #define SEMAEM 16384 /* adjust on exit max value */ 96 97 /* 98 * Due to the way semaphore memory is allocated, we have to ensure that 99 * SEMUSZ is properly aligned. 100 */ 101 102 #define SEM_ALIGN(bytes) (((bytes) + (sizeof(long) - 1)) & ~(sizeof(long) - 1)) 103 104 /* actual size of an undo structure */ 105 #define SEMUSZ SEM_ALIGN(offsetof(struct sem_undo, un_ent[SEMUME])) 106 107 /* 108 * Macro to find a particular sem_undo vector 109 */ 110 #define SEMU(ix) ((struct sem_undo *)(((intptr_t)semu)+ix * seminfo.semusz)) 111 112 /* 113 * semaphore info struct 114 */ 115 struct seminfo seminfo = { 116 SEMMAP, /* # of entries in semaphore map */ 117 SEMMNI, /* # of semaphore identifiers */ 118 SEMMNS, /* # of semaphores in system */ 119 SEMMNU, /* # of undo structures in system */ 120 SEMMSL, /* max # of semaphores per id */ 121 SEMOPM, /* max # of operations per semop call */ 122 SEMUME, /* max # of undo entries per process */ 123 SEMUSZ, /* size in bytes of undo structure */ 124 SEMVMX, /* semaphore maximum value */ 125 SEMAEM /* adjust on exit max value */ 126 }; 127 128 TUNABLE_INT("kern.ipc.semmap", &seminfo.semmap); 129 TUNABLE_INT("kern.ipc.semmni", &seminfo.semmni); 130 TUNABLE_INT("kern.ipc.semmns", &seminfo.semmns); 131 TUNABLE_INT("kern.ipc.semmnu", &seminfo.semmnu); 132 TUNABLE_INT("kern.ipc.semmsl", &seminfo.semmsl); 133 TUNABLE_INT("kern.ipc.semopm", &seminfo.semopm); 134 TUNABLE_INT("kern.ipc.semume", &seminfo.semume); 135 TUNABLE_INT("kern.ipc.semusz", &seminfo.semusz); 136 TUNABLE_INT("kern.ipc.semvmx", &seminfo.semvmx); 137 TUNABLE_INT("kern.ipc.semaem", &seminfo.semaem); 138 139 SYSCTL_INT(_kern_ipc, OID_AUTO, semmap, CTLFLAG_RW, &seminfo.semmap, 0, ""); 140 SYSCTL_INT(_kern_ipc, OID_AUTO, semmni, CTLFLAG_RD, &seminfo.semmni, 0, ""); 141 SYSCTL_INT(_kern_ipc, OID_AUTO, semmns, CTLFLAG_RD, &seminfo.semmns, 0, ""); 142 SYSCTL_INT(_kern_ipc, OID_AUTO, semmnu, CTLFLAG_RD, &seminfo.semmnu, 0, ""); 143 SYSCTL_INT(_kern_ipc, OID_AUTO, semmsl, CTLFLAG_RW, &seminfo.semmsl, 0, ""); 144 SYSCTL_INT(_kern_ipc, OID_AUTO, semopm, CTLFLAG_RD, &seminfo.semopm, 0, ""); 145 SYSCTL_INT(_kern_ipc, OID_AUTO, semume, CTLFLAG_RD, &seminfo.semume, 0, ""); 146 SYSCTL_INT(_kern_ipc, OID_AUTO, semusz, CTLFLAG_RD, &seminfo.semusz, 0, ""); 147 SYSCTL_INT(_kern_ipc, OID_AUTO, semvmx, CTLFLAG_RW, &seminfo.semvmx, 0, ""); 148 SYSCTL_INT(_kern_ipc, OID_AUTO, semaem, CTLFLAG_RW, &seminfo.semaem, 0, ""); 149 150 #if 0 151 RO seminfo.semmap /* SEMMAP unused */ 152 RO seminfo.semmni 153 RO seminfo.semmns 154 RO seminfo.semmnu /* undo entries per system */ 155 RW seminfo.semmsl 156 RO seminfo.semopm /* SEMOPM unused */ 157 RO seminfo.semume 158 RO seminfo.semusz /* param - derived from SEMUME for per-proc sizeof */ 159 RO seminfo.semvmx /* SEMVMX unused - user param */ 160 RO seminfo.semaem /* SEMAEM unused - user param */ 161 #endif 162 163 static void 164 seminit(void *dummy) 165 { 166 int i; 167 168 sem = kmalloc(sizeof(struct sem) * seminfo.semmns, M_SEM, M_WAITOK); 169 if (sem == NULL) 170 panic("sem is NULL"); 171 sema = kmalloc(sizeof(struct semid_ds) * seminfo.semmni, M_SEM, M_WAITOK); 172 if (sema == NULL) 173 panic("sema is NULL"); 174 semu = kmalloc(seminfo.semmnu * seminfo.semusz, M_SEM, M_WAITOK); 175 if (semu == NULL) 176 panic("semu is NULL"); 177 178 for (i = 0; i < seminfo.semmni; i++) { 179 sema[i].sem_base = 0; 180 sema[i].sem_perm.mode = 0; 181 } 182 for (i = 0; i < seminfo.semmnu; i++) { 183 struct sem_undo *suptr = SEMU(i); 184 suptr->un_proc = NULL; 185 } 186 semu_list = NULL; 187 } 188 SYSINIT(sysv_sem, SI_SUB_SYSV_SEM, SI_ORDER_FIRST, seminit, NULL) 189 190 /* 191 * Entry point for all SEM calls 192 * 193 * semsys_args(int which, a2, a3, ...) (VARARGS) 194 */ 195 int 196 sys_semsys(struct semsys_args *uap) 197 { 198 struct proc *p = curproc; 199 unsigned int which = (unsigned int)uap->which; 200 201 if (!jail_sysvipc_allowed && p->p_ucred->cr_prison != NULL) 202 return (ENOSYS); 203 204 if (which >= sizeof(semcalls)/sizeof(semcalls[0])) 205 return (EINVAL); 206 bcopy(&uap->a2, &uap->which, 207 sizeof(struct semsys_args) - offsetof(struct semsys_args, a2)); 208 return ((*semcalls[which])(uap)); 209 } 210 211 /* 212 * Allocate a new sem_undo structure for a process 213 * (returns ptr to structure or NULL if no more room) 214 */ 215 216 static struct sem_undo * 217 semu_alloc(struct proc *p) 218 { 219 int i; 220 struct sem_undo *suptr; 221 struct sem_undo **supptr; 222 int attempt; 223 224 /* 225 * Try twice to allocate something. 226 * (we'll purge any empty structures after the first pass so 227 * two passes are always enough) 228 */ 229 230 for (attempt = 0; attempt < 2; attempt++) { 231 /* 232 * Look for a free structure. 233 * Fill it in and return it if we find one. 234 */ 235 236 for (i = 0; i < seminfo.semmnu; i++) { 237 suptr = SEMU(i); 238 if (suptr->un_proc == NULL) { 239 suptr->un_next = semu_list; 240 semu_list = suptr; 241 suptr->un_cnt = 0; 242 suptr->un_proc = p; 243 return(suptr); 244 } 245 } 246 247 /* 248 * We didn't find a free one, if this is the first attempt 249 * then try to free some structures. 250 */ 251 252 if (attempt == 0) { 253 /* All the structures are in use - try to free some */ 254 int did_something = 0; 255 256 supptr = &semu_list; 257 while ((suptr = *supptr) != NULL) { 258 if (suptr->un_cnt == 0) { 259 suptr->un_proc = NULL; 260 *supptr = suptr->un_next; 261 did_something = 1; 262 } else 263 supptr = &(suptr->un_next); 264 } 265 266 /* If we didn't free anything then just give-up */ 267 if (!did_something) 268 return(NULL); 269 } else { 270 /* 271 * The second pass failed even though we freed 272 * something after the first pass! 273 * This is IMPOSSIBLE! 274 */ 275 panic("semu_alloc - second attempt failed"); 276 } 277 } 278 return (NULL); 279 } 280 281 /* 282 * Adjust a particular entry for a particular proc 283 */ 284 285 static int 286 semundo_adjust(struct proc *p, struct sem_undo **supptr, int semid, int semnum, 287 int adjval) 288 { 289 struct sem_undo *suptr; 290 struct undo *sunptr; 291 int i; 292 293 /* Look for and remember the sem_undo if the caller doesn't provide 294 it */ 295 296 suptr = *supptr; 297 if (suptr == NULL) { 298 for (suptr = semu_list; suptr != NULL; 299 suptr = suptr->un_next) { 300 if (suptr->un_proc == p) { 301 *supptr = suptr; 302 break; 303 } 304 } 305 if (suptr == NULL) { 306 if (adjval == 0) 307 return(0); 308 suptr = semu_alloc(p); 309 if (suptr == NULL) 310 return(ENOSPC); 311 *supptr = suptr; 312 } 313 } 314 315 /* 316 * Look for the requested entry and adjust it (delete if adjval becomes 317 * 0). 318 */ 319 sunptr = &suptr->un_ent[0]; 320 for (i = 0; i < suptr->un_cnt; i++, sunptr++) { 321 if (sunptr->un_id != semid || sunptr->un_num != semnum) 322 continue; 323 if (adjval == 0) 324 sunptr->un_adjval = 0; 325 else 326 sunptr->un_adjval += adjval; 327 if (sunptr->un_adjval == 0) { 328 suptr->un_cnt--; 329 if (i < suptr->un_cnt) 330 suptr->un_ent[i] = 331 suptr->un_ent[suptr->un_cnt]; 332 } 333 return(0); 334 } 335 336 /* Didn't find the right entry - create it */ 337 if (adjval == 0) 338 return(0); 339 if (suptr->un_cnt != seminfo.semume) { 340 sunptr = &suptr->un_ent[suptr->un_cnt]; 341 suptr->un_cnt++; 342 sunptr->un_adjval = adjval; 343 sunptr->un_id = semid; sunptr->un_num = semnum; 344 } else 345 return(EINVAL); 346 return(0); 347 } 348 349 static void 350 semundo_clear(int semid, int semnum) 351 { 352 struct sem_undo *suptr; 353 354 for (suptr = semu_list; suptr != NULL; suptr = suptr->un_next) { 355 struct undo *sunptr = &suptr->un_ent[0]; 356 int i = 0; 357 358 while (i < suptr->un_cnt) { 359 if (sunptr->un_id == semid) { 360 if (semnum == -1 || sunptr->un_num == semnum) { 361 suptr->un_cnt--; 362 if (i < suptr->un_cnt) { 363 suptr->un_ent[i] = 364 suptr->un_ent[suptr->un_cnt]; 365 continue; 366 } 367 } 368 if (semnum != -1) 369 break; 370 } 371 i++, sunptr++; 372 } 373 } 374 } 375 376 /* 377 * Note that the user-mode half of this passes a union, not a pointer 378 */ 379 380 int 381 sys___semctl(struct __semctl_args *uap) 382 { 383 struct proc *p = curproc; 384 int semid = uap->semid; 385 int semnum = uap->semnum; 386 int cmd = uap->cmd; 387 union semun *arg = uap->arg; 388 union semun real_arg; 389 struct ucred *cred = p->p_ucred; 390 int i, rval, eval; 391 struct semid_ds sbuf; 392 struct semid_ds *semaptr; 393 394 #ifdef SEM_DEBUG 395 kprintf("call to semctl(%d, %d, %d, 0x%x)\n", semid, semnum, cmd, arg); 396 #endif 397 398 if (!jail_sysvipc_allowed && p->p_ucred->cr_prison != NULL) 399 return (ENOSYS); 400 401 semid = IPCID_TO_IX(semid); 402 if (semid < 0 || semid >= seminfo.semmni) 403 return(EINVAL); 404 405 semaptr = &sema[semid]; 406 if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0 || 407 semaptr->sem_perm.seq != IPCID_TO_SEQ(uap->semid)) 408 return(EINVAL); 409 410 eval = 0; 411 rval = 0; 412 413 switch (cmd) { 414 case IPC_RMID: 415 if ((eval = ipcperm(p, &semaptr->sem_perm, IPC_M))) 416 return(eval); 417 semaptr->sem_perm.cuid = cred->cr_uid; 418 semaptr->sem_perm.uid = cred->cr_uid; 419 semtot -= semaptr->sem_nsems; 420 for (i = semaptr->sem_base - sem; i < semtot; i++) 421 sem[i] = sem[i + semaptr->sem_nsems]; 422 for (i = 0; i < seminfo.semmni; i++) { 423 if ((sema[i].sem_perm.mode & SEM_ALLOC) && 424 sema[i].sem_base > semaptr->sem_base) 425 sema[i].sem_base -= semaptr->sem_nsems; 426 } 427 semaptr->sem_perm.mode = 0; 428 semundo_clear(semid, -1); 429 wakeup((caddr_t)semaptr); 430 break; 431 432 case IPC_SET: 433 if ((eval = ipcperm(p, &semaptr->sem_perm, IPC_M))) 434 return(eval); 435 if ((eval = copyin(arg, &real_arg, sizeof(real_arg))) != 0) 436 return(eval); 437 if ((eval = copyin(real_arg.buf, (caddr_t)&sbuf, 438 sizeof(sbuf))) != 0) 439 return(eval); 440 semaptr->sem_perm.uid = sbuf.sem_perm.uid; 441 semaptr->sem_perm.gid = sbuf.sem_perm.gid; 442 semaptr->sem_perm.mode = (semaptr->sem_perm.mode & ~0777) | 443 (sbuf.sem_perm.mode & 0777); 444 semaptr->sem_ctime = time_second; 445 break; 446 447 case IPC_STAT: 448 if ((eval = ipcperm(p, &semaptr->sem_perm, IPC_R))) 449 return(eval); 450 if ((eval = copyin(arg, &real_arg, sizeof(real_arg))) != 0) 451 return(eval); 452 eval = copyout((caddr_t)semaptr, real_arg.buf, 453 sizeof(struct semid_ds)); 454 break; 455 456 case GETNCNT: 457 if ((eval = ipcperm(p, &semaptr->sem_perm, IPC_R))) 458 return(eval); 459 if (semnum < 0 || semnum >= semaptr->sem_nsems) 460 return(EINVAL); 461 rval = semaptr->sem_base[semnum].semncnt; 462 break; 463 464 case GETPID: 465 if ((eval = ipcperm(p, &semaptr->sem_perm, IPC_R))) 466 return(eval); 467 if (semnum < 0 || semnum >= semaptr->sem_nsems) 468 return(EINVAL); 469 rval = semaptr->sem_base[semnum].sempid; 470 break; 471 472 case GETVAL: 473 if ((eval = ipcperm(p, &semaptr->sem_perm, IPC_R))) 474 return(eval); 475 if (semnum < 0 || semnum >= semaptr->sem_nsems) 476 return(EINVAL); 477 rval = semaptr->sem_base[semnum].semval; 478 break; 479 480 case GETALL: 481 if ((eval = ipcperm(p, &semaptr->sem_perm, IPC_R))) 482 return(eval); 483 if ((eval = copyin(arg, &real_arg, sizeof(real_arg))) != 0) 484 return(eval); 485 for (i = 0; i < semaptr->sem_nsems; i++) { 486 eval = copyout((caddr_t)&semaptr->sem_base[i].semval, 487 &real_arg.array[i], sizeof(real_arg.array[0])); 488 if (eval != 0) 489 break; 490 } 491 break; 492 493 case GETZCNT: 494 if ((eval = ipcperm(p, &semaptr->sem_perm, IPC_R))) 495 return(eval); 496 if (semnum < 0 || semnum >= semaptr->sem_nsems) 497 return(EINVAL); 498 rval = semaptr->sem_base[semnum].semzcnt; 499 break; 500 501 case SETVAL: 502 if ((eval = ipcperm(p, &semaptr->sem_perm, IPC_W))) 503 return(eval); 504 if (semnum < 0 || semnum >= semaptr->sem_nsems) 505 return(EINVAL); 506 if ((eval = copyin(arg, &real_arg, sizeof(real_arg))) != 0) 507 return(eval); 508 semaptr->sem_base[semnum].semval = real_arg.val; 509 semundo_clear(semid, semnum); 510 wakeup((caddr_t)semaptr); 511 break; 512 513 case SETALL: 514 if ((eval = ipcperm(p, &semaptr->sem_perm, IPC_W))) 515 return(eval); 516 if ((eval = copyin(arg, &real_arg, sizeof(real_arg))) != 0) 517 return(eval); 518 for (i = 0; i < semaptr->sem_nsems; i++) { 519 eval = copyin(&real_arg.array[i], 520 (caddr_t)&semaptr->sem_base[i].semval, 521 sizeof(real_arg.array[0])); 522 if (eval != 0) 523 break; 524 } 525 semundo_clear(semid, -1); 526 wakeup((caddr_t)semaptr); 527 break; 528 529 default: 530 return(EINVAL); 531 } 532 533 if (eval == 0) 534 uap->sysmsg_result = rval; 535 return(eval); 536 } 537 538 int 539 sys_semget(struct semget_args *uap) 540 { 541 struct proc *p = curproc; 542 int semid, eval; 543 int key = uap->key; 544 int nsems = uap->nsems; 545 int semflg = uap->semflg; 546 struct ucred *cred = p->p_ucred; 547 548 #ifdef SEM_DEBUG 549 kprintf("semget(0x%x, %d, 0%o)\n", key, nsems, semflg); 550 #endif 551 552 if (!jail_sysvipc_allowed && p->p_ucred->cr_prison != NULL) 553 return (ENOSYS); 554 555 if (key != IPC_PRIVATE) { 556 for (semid = 0; semid < seminfo.semmni; semid++) { 557 if ((sema[semid].sem_perm.mode & SEM_ALLOC) && 558 sema[semid].sem_perm.key == key) 559 break; 560 } 561 if (semid < seminfo.semmni) { 562 #ifdef SEM_DEBUG 563 kprintf("found public key\n"); 564 #endif 565 if ((eval = ipcperm(p, &sema[semid].sem_perm, 566 semflg & 0700))) 567 return(eval); 568 if (nsems > 0 && sema[semid].sem_nsems < nsems) { 569 #ifdef SEM_DEBUG 570 kprintf("too small\n"); 571 #endif 572 return(EINVAL); 573 } 574 if ((semflg & IPC_CREAT) && (semflg & IPC_EXCL)) { 575 #ifdef SEM_DEBUG 576 kprintf("not exclusive\n"); 577 #endif 578 return(EEXIST); 579 } 580 goto found; 581 } 582 } 583 584 #ifdef SEM_DEBUG 585 kprintf("need to allocate the semid_ds\n"); 586 #endif 587 if (key == IPC_PRIVATE || (semflg & IPC_CREAT)) { 588 if (nsems <= 0 || nsems > seminfo.semmsl) { 589 #ifdef SEM_DEBUG 590 kprintf("nsems out of range (0<%d<=%d)\n", nsems, 591 seminfo.semmsl); 592 #endif 593 return(EINVAL); 594 } 595 if (nsems > seminfo.semmns - semtot) { 596 #ifdef SEM_DEBUG 597 kprintf("not enough semaphores left (need %d, got %d)\n", 598 nsems, seminfo.semmns - semtot); 599 #endif 600 return(ENOSPC); 601 } 602 for (semid = 0; semid < seminfo.semmni; semid++) { 603 if ((sema[semid].sem_perm.mode & SEM_ALLOC) == 0) 604 break; 605 } 606 if (semid == seminfo.semmni) { 607 #ifdef SEM_DEBUG 608 kprintf("no more semid_ds's available\n"); 609 #endif 610 return(ENOSPC); 611 } 612 #ifdef SEM_DEBUG 613 kprintf("semid %d is available\n", semid); 614 #endif 615 sema[semid].sem_perm.key = key; 616 sema[semid].sem_perm.cuid = cred->cr_uid; 617 sema[semid].sem_perm.uid = cred->cr_uid; 618 sema[semid].sem_perm.cgid = cred->cr_gid; 619 sema[semid].sem_perm.gid = cred->cr_gid; 620 sema[semid].sem_perm.mode = (semflg & 0777) | SEM_ALLOC; 621 sema[semid].sem_perm.seq = 622 (sema[semid].sem_perm.seq + 1) & 0x7fff; 623 sema[semid].sem_nsems = nsems; 624 sema[semid].sem_otime = 0; 625 sema[semid].sem_ctime = time_second; 626 sema[semid].sem_base = &sem[semtot]; 627 semtot += nsems; 628 bzero(sema[semid].sem_base, 629 sizeof(sema[semid].sem_base[0])*nsems); 630 #ifdef SEM_DEBUG 631 kprintf("sembase = 0x%x, next = 0x%x\n", sema[semid].sem_base, 632 &sem[semtot]); 633 #endif 634 } else { 635 #ifdef SEM_DEBUG 636 kprintf("didn't find it and wasn't asked to create it\n"); 637 #endif 638 return(ENOENT); 639 } 640 641 found: 642 uap->sysmsg_result = IXSEQ_TO_IPCID(semid, sema[semid].sem_perm); 643 return(0); 644 } 645 646 int 647 sys_semop(struct semop_args *uap) 648 { 649 struct proc *p = curproc; 650 int semid = uap->semid; 651 u_int nsops = uap->nsops; 652 struct sembuf sops[MAX_SOPS]; 653 struct semid_ds *semaptr; 654 struct sembuf *sopptr; 655 struct sem *semptr; 656 struct sem_undo *suptr = NULL; 657 int i, j, eval; 658 int do_wakeup, do_undos; 659 660 #ifdef SEM_DEBUG 661 kprintf("call to semop(%d, 0x%x, %u)\n", semid, sops, nsops); 662 #endif 663 664 if (!jail_sysvipc_allowed && p->p_ucred->cr_prison != NULL) 665 return (ENOSYS); 666 667 semid = IPCID_TO_IX(semid); /* Convert back to zero origin */ 668 669 if (semid < 0 || semid >= seminfo.semmni) 670 return(EINVAL); 671 672 semaptr = &sema[semid]; 673 if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0) 674 return(EINVAL); 675 if (semaptr->sem_perm.seq != IPCID_TO_SEQ(uap->semid)) 676 return(EINVAL); 677 678 if ((eval = ipcperm(p, &semaptr->sem_perm, IPC_W))) { 679 #ifdef SEM_DEBUG 680 kprintf("eval = %d from ipaccess\n", eval); 681 #endif 682 return(eval); 683 } 684 685 if (nsops > MAX_SOPS) { 686 #ifdef SEM_DEBUG 687 kprintf("too many sops (max=%d, nsops=%u)\n", MAX_SOPS, nsops); 688 #endif 689 return(E2BIG); 690 } 691 692 if ((eval = copyin(uap->sops, &sops, nsops * sizeof(sops[0]))) != 0) { 693 #ifdef SEM_DEBUG 694 kprintf("eval = %d from copyin(%08x, %08x, %u)\n", eval, 695 uap->sops, &sops, nsops * sizeof(sops[0])); 696 #endif 697 return(eval); 698 } 699 700 /* 701 * Loop trying to satisfy the vector of requests. 702 * If we reach a point where we must wait, any requests already 703 * performed are rolled back and we go to sleep until some other 704 * process wakes us up. At this point, we start all over again. 705 * 706 * This ensures that from the perspective of other tasks, a set 707 * of requests is atomic (never partially satisfied). 708 */ 709 do_undos = 0; 710 711 for (;;) { 712 do_wakeup = 0; 713 714 for (i = 0; i < nsops; i++) { 715 sopptr = &sops[i]; 716 717 if (sopptr->sem_num >= semaptr->sem_nsems) 718 return(EFBIG); 719 720 semptr = &semaptr->sem_base[sopptr->sem_num]; 721 722 #ifdef SEM_DEBUG 723 kprintf("semop: semaptr=%x, sem_base=%x, semptr=%x, sem[%d]=%d : op=%d, flag=%s\n", 724 semaptr, semaptr->sem_base, semptr, 725 sopptr->sem_num, semptr->semval, sopptr->sem_op, 726 (sopptr->sem_flg & IPC_NOWAIT) ? "nowait" : "wait"); 727 #endif 728 729 if (sopptr->sem_op < 0) { 730 if (semptr->semval + sopptr->sem_op < 0) { 731 #ifdef SEM_DEBUG 732 kprintf("semop: can't do it now\n"); 733 #endif 734 break; 735 } else { 736 semptr->semval += sopptr->sem_op; 737 if (semptr->semval == 0 && 738 semptr->semzcnt > 0) 739 do_wakeup = 1; 740 } 741 if (sopptr->sem_flg & SEM_UNDO) 742 do_undos = 1; 743 } else if (sopptr->sem_op == 0) { 744 if (semptr->semval > 0) { 745 #ifdef SEM_DEBUG 746 kprintf("semop: not zero now\n"); 747 #endif 748 break; 749 } 750 } else { 751 if (semptr->semncnt > 0) 752 do_wakeup = 1; 753 semptr->semval += sopptr->sem_op; 754 if (sopptr->sem_flg & SEM_UNDO) 755 do_undos = 1; 756 } 757 } 758 759 /* 760 * Did we get through the entire vector? 761 */ 762 if (i >= nsops) 763 goto done; 764 765 /* 766 * No ... rollback anything that we've already done 767 */ 768 #ifdef SEM_DEBUG 769 kprintf("semop: rollback 0 through %d\n", i-1); 770 #endif 771 for (j = 0; j < i; j++) 772 semaptr->sem_base[sops[j].sem_num].semval -= 773 sops[j].sem_op; 774 775 /* 776 * If the request that we couldn't satisfy has the 777 * NOWAIT flag set then return with EAGAIN. 778 */ 779 if (sopptr->sem_flg & IPC_NOWAIT) 780 return(EAGAIN); 781 782 if (sopptr->sem_op == 0) 783 semptr->semzcnt++; 784 else 785 semptr->semncnt++; 786 787 #ifdef SEM_DEBUG 788 kprintf("semop: good night!\n"); 789 #endif 790 eval = tsleep((caddr_t)semaptr, PCATCH, "semwait", 0); 791 #ifdef SEM_DEBUG 792 kprintf("semop: good morning (eval=%d)!\n", eval); 793 #endif 794 795 suptr = NULL; /* sem_undo may have been reallocated */ 796 797 /* return code is checked below, after sem[nz]cnt-- */ 798 799 /* 800 * Make sure that the semaphore still exists 801 */ 802 if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0 || 803 semaptr->sem_perm.seq != IPCID_TO_SEQ(uap->semid)) 804 return(EIDRM); 805 806 /* 807 * The semaphore is still alive. Readjust the count of 808 * waiting processes. 809 */ 810 if (sopptr->sem_op == 0) 811 semptr->semzcnt--; 812 else 813 semptr->semncnt--; 814 815 /* 816 * Is it really morning, or was our sleep interrupted? 817 * (Delayed check of msleep() return code because we 818 * need to decrement sem[nz]cnt either way.) 819 */ 820 if (eval != 0) 821 return(EINTR); 822 #ifdef SEM_DEBUG 823 kprintf("semop: good morning!\n"); 824 #endif 825 } 826 827 done: 828 /* 829 * Process any SEM_UNDO requests. 830 */ 831 if (do_undos) { 832 for (i = 0; i < nsops; i++) { 833 /* 834 * We only need to deal with SEM_UNDO's for non-zero 835 * op's. 836 */ 837 int adjval; 838 839 if ((sops[i].sem_flg & SEM_UNDO) == 0) 840 continue; 841 adjval = sops[i].sem_op; 842 if (adjval == 0) 843 continue; 844 eval = semundo_adjust(p, &suptr, semid, 845 sops[i].sem_num, -adjval); 846 if (eval == 0) 847 continue; 848 849 /* 850 * Oh-Oh! We ran out of either sem_undo's or undo's. 851 * Rollback the adjustments to this point and then 852 * rollback the semaphore ups and down so we can return 853 * with an error with all structures restored. We 854 * rollback the undo's in the exact reverse order that 855 * we applied them. This guarantees that we won't run 856 * out of space as we roll things back out. 857 */ 858 for (j = i - 1; j >= 0; j--) { 859 if ((sops[j].sem_flg & SEM_UNDO) == 0) 860 continue; 861 adjval = sops[j].sem_op; 862 if (adjval == 0) 863 continue; 864 if (semundo_adjust(p, &suptr, semid, 865 sops[j].sem_num, adjval) != 0) 866 panic("semop - can't undo undos"); 867 } 868 869 for (j = 0; j < nsops; j++) 870 semaptr->sem_base[sops[j].sem_num].semval -= 871 sops[j].sem_op; 872 873 #ifdef SEM_DEBUG 874 kprintf("eval = %d from semundo_adjust\n", eval); 875 #endif 876 return(eval); 877 } /* loop through the sops */ 878 } /* if (do_undos) */ 879 880 /* We're definitely done - set the sempid's */ 881 for (i = 0; i < nsops; i++) { 882 sopptr = &sops[i]; 883 semptr = &semaptr->sem_base[sopptr->sem_num]; 884 semptr->sempid = p->p_pid; 885 } 886 887 /* Do a wakeup if any semaphore was up'd. */ 888 if (do_wakeup) { 889 #ifdef SEM_DEBUG 890 kprintf("semop: doing wakeup\n"); 891 #endif 892 wakeup((caddr_t)semaptr); 893 #ifdef SEM_DEBUG 894 kprintf("semop: back from wakeup\n"); 895 #endif 896 } 897 #ifdef SEM_DEBUG 898 kprintf("semop: done\n"); 899 #endif 900 uap->sysmsg_result = 0; 901 return(0); 902 } 903 904 /* 905 * Go through the undo structures for this process and apply the adjustments to 906 * semaphores. 907 */ 908 void 909 semexit(struct proc *p) 910 { 911 struct sem_undo *suptr; 912 struct sem_undo **supptr; 913 int did_something; 914 915 did_something = 0; 916 917 /* 918 * Go through the chain of undo vectors looking for one 919 * associated with this process. 920 */ 921 922 for (supptr = &semu_list; (suptr = *supptr) != NULL; 923 supptr = &suptr->un_next) { 924 if (suptr->un_proc == p) 925 break; 926 } 927 928 if (suptr == NULL) 929 return; 930 931 #ifdef SEM_DEBUG 932 kprintf("proc @%08x has undo structure with %d entries\n", p, 933 suptr->un_cnt); 934 #endif 935 936 /* 937 * If there are any active undo elements then process them. 938 */ 939 if (suptr->un_cnt > 0) { 940 int ix; 941 942 for (ix = 0; ix < suptr->un_cnt; ix++) { 943 int semid = suptr->un_ent[ix].un_id; 944 int semnum = suptr->un_ent[ix].un_num; 945 int adjval = suptr->un_ent[ix].un_adjval; 946 struct semid_ds *semaptr; 947 948 semaptr = &sema[semid]; 949 if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0) 950 panic("semexit - semid not allocated"); 951 if (semnum >= semaptr->sem_nsems) 952 panic("semexit - semnum out of range"); 953 954 #ifdef SEM_DEBUG 955 kprintf("semexit: %08x id=%d num=%d(adj=%d) ; sem=%d\n", 956 suptr->un_proc, suptr->un_ent[ix].un_id, 957 suptr->un_ent[ix].un_num, 958 suptr->un_ent[ix].un_adjval, 959 semaptr->sem_base[semnum].semval); 960 #endif 961 962 if (adjval < 0) { 963 if (semaptr->sem_base[semnum].semval < -adjval) 964 semaptr->sem_base[semnum].semval = 0; 965 else 966 semaptr->sem_base[semnum].semval += 967 adjval; 968 } else 969 semaptr->sem_base[semnum].semval += adjval; 970 971 wakeup((caddr_t)semaptr); 972 #ifdef SEM_DEBUG 973 kprintf("semexit: back from wakeup\n"); 974 #endif 975 } 976 } 977 978 /* 979 * Deallocate the undo vector. 980 */ 981 #ifdef SEM_DEBUG 982 kprintf("removing vector\n"); 983 #endif 984 suptr->un_proc = NULL; 985 *supptr = suptr->un_next; 986 } 987