Home | History | Annotate | Line # | Download | only in kern
sysv_sem.c revision 1.33
      1 /*	$NetBSD: sysv_sem.c,v 1.33 1999/08/25 05:05:49 thorpej Exp $	*/
      2 
      3 /*-
      4  * Copyright (c) 1999 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.
     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  * 3. All advertising materials mentioning features or use of this software
     20  *    must display the following acknowledgement:
     21  *	This product includes software developed by the NetBSD
     22  *	Foundation, Inc. and its contributors.
     23  * 4. Neither the name of The NetBSD Foundation nor the names of its
     24  *    contributors may be used to endorse or promote products derived
     25  *    from this software without specific prior written permission.
     26  *
     27  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
     28  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
     29  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
     30  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
     31  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
     32  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
     33  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
     34  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
     35  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
     36  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
     37  * POSSIBILITY OF SUCH DAMAGE.
     38  */
     39 
     40 /*
     41  * Implementation of SVID semaphores
     42  *
     43  * Author: Daniel Boulet
     44  *
     45  * This software is provided ``AS IS'' without any warranties of any kind.
     46  */
     47 
     48 #define SYSVSEM
     49 
     50 #include <sys/param.h>
     51 #include <sys/systm.h>
     52 #include <sys/kernel.h>
     53 #include <sys/proc.h>
     54 #include <sys/sem.h>
     55 #include <sys/malloc.h>
     56 
     57 #include <sys/mount.h>
     58 #include <sys/syscallargs.h>
     59 
     60 int	semtot = 0;
     61 struct	proc *semlock_holder = NULL;
     62 
     63 #ifdef SEM_DEBUG
     64 #define SEM_PRINTF(a) printf a
     65 #else
     66 #define SEM_PRINTF(a)
     67 #endif
     68 
     69 void semlock __P((struct proc *));
     70 struct sem_undo *semu_alloc __P((struct proc *));
     71 int semundo_adjust __P((struct proc *, struct sem_undo **, int, int, int));
     72 void semundo_clear __P((int, int));
     73 
     74 void
     75 seminit()
     76 {
     77 	register int i;
     78 
     79 	if (sema == NULL)
     80 		panic("sema is NULL");
     81 	if (semu == NULL)
     82 		panic("semu is NULL");
     83 
     84 	for (i = 0; i < seminfo.semmni; i++) {
     85 		sema[i]._sem_base = 0;
     86 		sema[i].sem_perm.mode = 0;
     87 	}
     88 	for (i = 0; i < seminfo.semmnu; i++) {
     89 		register struct sem_undo *suptr = SEMU(i);
     90 		suptr->un_proc = NULL;
     91 	}
     92 	semu_list = NULL;
     93 }
     94 
     95 void
     96 semlock(p)
     97 	struct proc *p;
     98 {
     99 
    100 	while (semlock_holder != NULL && semlock_holder != p)
    101 		sleep((caddr_t)&semlock_holder, (PZERO - 4));
    102 }
    103 
    104 /*
    105  * Lock or unlock the entire semaphore facility.
    106  *
    107  * This will probably eventually evolve into a general purpose semaphore
    108  * facility status enquiry mechanism (I don't like the "read /dev/kmem"
    109  * approach currently taken by ipcs and the amount of info that we want
    110  * to be able to extract for ipcs is probably beyond the capability of
    111  * the getkerninfo facility.
    112  *
    113  * At the time that the current version of semconfig was written, ipcs is
    114  * the only user of the semconfig facility.  It uses it to ensure that the
    115  * semaphore facility data structures remain static while it fishes around
    116  * in /dev/kmem.
    117  */
    118 
    119 int
    120 sys_semconfig(p, v, retval)
    121 	struct proc *p;
    122 	void *v;
    123 	register_t *retval;
    124 {
    125 	struct sys_semconfig_args /* {
    126 		syscallarg(int) flag;
    127 	} */ *uap = v;
    128 	int eval = 0;
    129 
    130 	semlock(p);
    131 
    132 	switch (SCARG(uap, flag)) {
    133 	case SEM_CONFIG_FREEZE:
    134 		semlock_holder = p;
    135 		break;
    136 
    137 	case SEM_CONFIG_THAW:
    138 		semlock_holder = NULL;
    139 		wakeup((caddr_t)&semlock_holder);
    140 		break;
    141 
    142 	default:
    143 		printf(
    144 		    "semconfig: unknown flag parameter value (%d) - ignored\n",
    145 		    SCARG(uap, flag));
    146 		eval = EINVAL;
    147 		break;
    148 	}
    149 
    150 	*retval = 0;
    151 	return(eval);
    152 }
    153 
    154 /*
    155  * Allocate a new sem_undo structure for a process
    156  * (returns ptr to structure or NULL if no more room)
    157  */
    158 
    159 struct sem_undo *
    160 semu_alloc(p)
    161 	struct proc *p;
    162 {
    163 	register int i;
    164 	register struct sem_undo *suptr;
    165 	register struct sem_undo **supptr;
    166 	int attempt;
    167 
    168 	/*
    169 	 * Try twice to allocate something.
    170 	 * (we'll purge any empty structures after the first pass so
    171 	 * two passes are always enough)
    172 	 */
    173 
    174 	for (attempt = 0; attempt < 2; attempt++) {
    175 		/*
    176 		 * Look for a free structure.
    177 		 * Fill it in and return it if we find one.
    178 		 */
    179 
    180 		for (i = 0; i < seminfo.semmnu; i++) {
    181 			suptr = SEMU(i);
    182 			if (suptr->un_proc == NULL) {
    183 				suptr->un_next = semu_list;
    184 				semu_list = suptr;
    185 				suptr->un_cnt = 0;
    186 				suptr->un_proc = p;
    187 				return(suptr);
    188 			}
    189 		}
    190 
    191 		/*
    192 		 * We didn't find a free one, if this is the first attempt
    193 		 * then try to free some structures.
    194 		 */
    195 
    196 		if (attempt == 0) {
    197 			/* All the structures are in use - try to free some */
    198 			int did_something = 0;
    199 
    200 			supptr = &semu_list;
    201 			while ((suptr = *supptr) != NULL) {
    202 				if (suptr->un_cnt == 0)  {
    203 					suptr->un_proc = NULL;
    204 					*supptr = suptr->un_next;
    205 					did_something = 1;
    206 				} else
    207 					supptr = &(suptr->un_next);
    208 			}
    209 
    210 			/* If we didn't free anything then just give-up */
    211 			if (!did_something)
    212 				return(NULL);
    213 		} else {
    214 			/*
    215 			 * The second pass failed even though we freed
    216 			 * something after the first pass!
    217 			 * This is IMPOSSIBLE!
    218 			 */
    219 			panic("semu_alloc - second attempt failed");
    220 		}
    221 	}
    222 	return NULL;
    223 }
    224 
    225 /*
    226  * Adjust a particular entry for a particular proc
    227  */
    228 
    229 int
    230 semundo_adjust(p, supptr, semid, semnum, adjval)
    231 	register struct proc *p;
    232 	struct sem_undo **supptr;
    233 	int semid, semnum;
    234 	int adjval;
    235 {
    236 	register struct sem_undo *suptr;
    237 	register struct undo *sunptr;
    238 	int i;
    239 
    240 	/* Look for and remember the sem_undo if the caller doesn't provide
    241 	   it */
    242 
    243 	suptr = *supptr;
    244 	if (suptr == NULL) {
    245 		for (suptr = semu_list; suptr != NULL; suptr = suptr->un_next) {
    246 			if (suptr->un_proc == p) {
    247 				*supptr = suptr;
    248 				break;
    249 			}
    250 		}
    251 		if (suptr == NULL) {
    252 			if (adjval == 0)
    253 				return(0);
    254 			suptr = semu_alloc(p);
    255 			if (suptr == NULL)
    256 				return(ENOSPC);
    257 			*supptr = suptr;
    258 		}
    259 	}
    260 
    261 	/*
    262 	 * Look for the requested entry and adjust it (delete if adjval becomes
    263 	 * 0).
    264 	 */
    265 	sunptr = &suptr->un_ent[0];
    266 	for (i = 0; i < suptr->un_cnt; i++, sunptr++) {
    267 		if (sunptr->un_id != semid || sunptr->un_num != semnum)
    268 			continue;
    269 		if (adjval == 0)
    270 			sunptr->un_adjval = 0;
    271 		else
    272 			sunptr->un_adjval += adjval;
    273 		if (sunptr->un_adjval == 0) {
    274 			suptr->un_cnt--;
    275 			if (i < suptr->un_cnt)
    276 				suptr->un_ent[i] =
    277 				    suptr->un_ent[suptr->un_cnt];
    278 		}
    279 		return(0);
    280 	}
    281 
    282 	/* Didn't find the right entry - create it */
    283 	if (adjval == 0)
    284 		return(0);
    285 	if (suptr->un_cnt == SEMUME)
    286 		return(EINVAL);
    287 
    288 	sunptr = &suptr->un_ent[suptr->un_cnt];
    289 	suptr->un_cnt++;
    290 	sunptr->un_adjval = adjval;
    291 	sunptr->un_id = semid;
    292 	sunptr->un_num = semnum;
    293 	return(0);
    294 }
    295 
    296 void
    297 semundo_clear(semid, semnum)
    298 	int semid, semnum;
    299 {
    300 	register struct sem_undo *suptr;
    301 
    302 	for (suptr = semu_list; suptr != NULL; suptr = suptr->un_next) {
    303 		register struct undo *sunptr;
    304 		register int i;
    305 
    306 		sunptr = &suptr->un_ent[0];
    307 		for (i = 0; i < suptr->un_cnt; i++, sunptr++) {
    308 			if (sunptr->un_id == semid) {
    309 				if (semnum == -1 || sunptr->un_num == semnum) {
    310 					suptr->un_cnt--;
    311 					if (i < suptr->un_cnt) {
    312 						suptr->un_ent[i] =
    313 						  suptr->un_ent[suptr->un_cnt];
    314 						i--, sunptr--;
    315 					}
    316 				}
    317 				if (semnum != -1)
    318 					break;
    319 			}
    320 		}
    321 	}
    322 }
    323 
    324 int
    325 sys___semctl13(p, v, retval)
    326 	struct proc *p;
    327 	void *v;
    328 	register_t *retval;
    329 {
    330 	struct sys___semctl13_args /* {
    331 		syscallarg(int) semid;
    332 		syscallarg(int) semnum;
    333 		syscallarg(int) cmd;
    334 		syscallarg(union __semun) arg;
    335 	} */ *uap = v;
    336 	struct semid_ds sembuf;
    337 	int cmd, error;
    338 	void *pass_arg = NULL;
    339 
    340 	cmd = SCARG(uap, cmd);
    341 
    342 	switch (cmd) {
    343 	case IPC_SET:
    344 	case IPC_STAT:
    345 		pass_arg = &sembuf;
    346 		break;
    347 
    348 	case GETALL:
    349 	case SETVAL:
    350 	case SETALL:
    351 		pass_arg = &SCARG(uap, arg);
    352 		break;
    353 	}
    354 
    355 	if (cmd == IPC_SET) {
    356 		error = copyin(SCARG(uap, arg).buf, &sembuf, sizeof(sembuf));
    357 		if (error)
    358 			return (error);
    359 	}
    360 
    361 	error = semctl1(p, SCARG(uap, semid), SCARG(uap, semnum), cmd,
    362 	    pass_arg, retval);
    363 
    364 	if (error == 0 && cmd == IPC_STAT)
    365 		error = copyout(&sembuf, SCARG(uap, arg).buf, sizeof(sembuf));
    366 
    367 	return (error);
    368 }
    369 
    370 int
    371 semctl1(p, semid, semnum, cmd, v, retval)
    372 	struct proc *p;
    373 	int semid, semnum, cmd;
    374 	void *v;
    375 	register_t *retval;
    376 {
    377 	struct ucred *cred = p->p_ucred;
    378 	union __semun *arg = v;
    379 	struct semid_ds *sembuf = v, *semaptr;
    380 	int i, error, ix;
    381 
    382 	SEM_PRINTF(("call to semctl(%d, %d, %d, %p)\n",
    383 	    semid, semnum, cmd, v));
    384 
    385 	semlock(p);
    386 
    387 	ix = IPCID_TO_IX(semid);
    388 	if (ix < 0 || ix >= seminfo.semmsl)
    389 		return (EINVAL);
    390 
    391 	semaptr = &sema[ix];
    392 	if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0 ||
    393 	    semaptr->sem_perm._seq != IPCID_TO_SEQ(semid))
    394 		return (EINVAL);
    395 
    396 	switch (cmd) {
    397 	case IPC_RMID:
    398 		if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_M)) != 0)
    399 			return (error);
    400 		semaptr->sem_perm.cuid = cred->cr_uid;
    401 		semaptr->sem_perm.uid = cred->cr_uid;
    402 		semtot -= semaptr->sem_nsems;
    403 		for (i = semaptr->_sem_base - sem; i < semtot; i++)
    404 			sem[i] = sem[i + semaptr->sem_nsems];
    405 		for (i = 0; i < seminfo.semmni; i++) {
    406 			if ((sema[i].sem_perm.mode & SEM_ALLOC) &&
    407 			    sema[i]._sem_base > semaptr->_sem_base)
    408 				sema[i]._sem_base -= semaptr->sem_nsems;
    409 		}
    410 		semaptr->sem_perm.mode = 0;
    411 		semundo_clear(ix, -1);
    412 		wakeup(semaptr);
    413 		break;
    414 
    415 	case IPC_SET:
    416 		if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_M)))
    417 			return (error);
    418 		semaptr->sem_perm.uid = sembuf->sem_perm.uid;
    419 		semaptr->sem_perm.gid = sembuf->sem_perm.gid;
    420 		semaptr->sem_perm.mode = (semaptr->sem_perm.mode & ~0777) |
    421 		    (sembuf->sem_perm.mode & 0777);
    422 		semaptr->sem_ctime = time.tv_sec;
    423 		break;
    424 
    425 	case IPC_STAT:
    426 		if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_R)))
    427 			return (error);
    428 		memcpy(sembuf, semaptr, sizeof(struct semid_ds));
    429 		break;
    430 
    431 	case GETNCNT:
    432 		if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_R)))
    433 			return (error);
    434 		if (semnum < 0 || semnum >= semaptr->sem_nsems)
    435 			return (EINVAL);
    436 		*retval = semaptr->_sem_base[semnum].semncnt;
    437 		break;
    438 
    439 	case GETPID:
    440 		if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_R)))
    441 			return (error);
    442 		if (semnum < 0 || semnum >= semaptr->sem_nsems)
    443 			return (EINVAL);
    444 		*retval = semaptr->_sem_base[semnum].sempid;
    445 		break;
    446 
    447 	case GETVAL:
    448 		if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_R)))
    449 			return (error);
    450 		if (semnum < 0 || semnum >= semaptr->sem_nsems)
    451 			return (EINVAL);
    452 		*retval = semaptr->_sem_base[semnum].semval;
    453 		break;
    454 
    455 	case GETALL:
    456 		if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_R)))
    457 			return (error);
    458 		for (i = 0; i < semaptr->sem_nsems; i++) {
    459 			error = copyout(&semaptr->_sem_base[i].semval,
    460 			    &arg->array[i], sizeof(arg->array[i]));
    461 			if (error != 0)
    462 				break;
    463 		}
    464 		break;
    465 
    466 	case GETZCNT:
    467 		if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_R)))
    468 			return (error);
    469 		if (semnum < 0 || semnum >= semaptr->sem_nsems)
    470 			return (EINVAL);
    471 		*retval = semaptr->_sem_base[semnum].semzcnt;
    472 		break;
    473 
    474 	case SETVAL:
    475 		if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_W)))
    476 			return (error);
    477 		if (semnum < 0 || semnum >= semaptr->sem_nsems)
    478 			return (EINVAL);
    479 		semaptr->_sem_base[semnum].semval = arg->val;
    480 		semundo_clear(ix, semnum);
    481 		wakeup(semaptr);
    482 		break;
    483 
    484 	case SETALL:
    485 		if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_W)))
    486 			return (error);
    487 		for (i = 0; i < semaptr->sem_nsems; i++) {
    488 			error = copyin(&arg->array[i],
    489 			    &semaptr->_sem_base[i].semval,
    490 			    sizeof(arg->array[i]));
    491 			if (error != 0)
    492 				break;
    493 		}
    494 		semundo_clear(ix, -1);
    495 		wakeup(semaptr);
    496 		break;
    497 
    498 	default:
    499 		return (EINVAL);
    500 	}
    501 
    502 	return (error);
    503 }
    504 
    505 int
    506 sys_semget(p, v, retval)
    507 	struct proc *p;
    508 	void *v;
    509 	register_t *retval;
    510 {
    511 	register struct sys_semget_args /* {
    512 		syscallarg(key_t) key;
    513 		syscallarg(int) nsems;
    514 		syscallarg(int) semflg;
    515 	} */ *uap = v;
    516 	int semid, eval;
    517 	int key = SCARG(uap, key);
    518 	int nsems = SCARG(uap, nsems);
    519 	int semflg = SCARG(uap, semflg);
    520 	struct ucred *cred = p->p_ucred;
    521 
    522 	SEM_PRINTF(("semget(0x%x, %d, 0%o)\n", key, nsems, semflg));
    523 
    524 	semlock(p);
    525 
    526 	if (key != IPC_PRIVATE) {
    527 		for (semid = 0; semid < seminfo.semmni; semid++) {
    528 			if ((sema[semid].sem_perm.mode & SEM_ALLOC) &&
    529 			    sema[semid].sem_perm._key == key)
    530 				break;
    531 		}
    532 		if (semid < seminfo.semmni) {
    533 			SEM_PRINTF(("found public key\n"));
    534 			if ((eval = ipcperm(cred, &sema[semid].sem_perm,
    535 			    semflg & 0700)))
    536 				return(eval);
    537 			if (nsems > 0 && sema[semid].sem_nsems < nsems) {
    538 				SEM_PRINTF(("too small\n"));
    539 				return(EINVAL);
    540 			}
    541 			if ((semflg & IPC_CREAT) && (semflg & IPC_EXCL)) {
    542 				SEM_PRINTF(("not exclusive\n"));
    543 				return(EEXIST);
    544 			}
    545 			goto found;
    546 		}
    547 	}
    548 
    549 	SEM_PRINTF(("need to allocate the semid_ds\n"));
    550 	if (key == IPC_PRIVATE || (semflg & IPC_CREAT)) {
    551 		if (nsems <= 0 || nsems > seminfo.semmsl) {
    552 			SEM_PRINTF(("nsems out of range (0<%d<=%d)\n", nsems,
    553 			    seminfo.semmsl));
    554 			return(EINVAL);
    555 		}
    556 		if (nsems > seminfo.semmns - semtot) {
    557 			SEM_PRINTF(("not enough semaphores left (need %d, got %d)\n",
    558 			    nsems, seminfo.semmns - semtot));
    559 			return(ENOSPC);
    560 		}
    561 		for (semid = 0; semid < seminfo.semmni; semid++) {
    562 			if ((sema[semid].sem_perm.mode & SEM_ALLOC) == 0)
    563 				break;
    564 		}
    565 		if (semid == seminfo.semmni) {
    566 			SEM_PRINTF(("no more semid_ds's available\n"));
    567 			return(ENOSPC);
    568 		}
    569 		SEM_PRINTF(("semid %d is available\n", semid));
    570 		sema[semid].sem_perm._key = key;
    571 		sema[semid].sem_perm.cuid = cred->cr_uid;
    572 		sema[semid].sem_perm.uid = cred->cr_uid;
    573 		sema[semid].sem_perm.cgid = cred->cr_gid;
    574 		sema[semid].sem_perm.gid = cred->cr_gid;
    575 		sema[semid].sem_perm.mode = (semflg & 0777) | SEM_ALLOC;
    576 		sema[semid].sem_perm._seq =
    577 		    (sema[semid].sem_perm._seq + 1) & 0x7fff;
    578 		sema[semid].sem_nsems = nsems;
    579 		sema[semid].sem_otime = 0;
    580 		sema[semid].sem_ctime = time.tv_sec;
    581 		sema[semid]._sem_base = &sem[semtot];
    582 		semtot += nsems;
    583 		memset(sema[semid]._sem_base, 0,
    584 		    sizeof(sema[semid]._sem_base[0])*nsems);
    585 		SEM_PRINTF(("sembase = %p, next = %p\n", sema[semid]._sem_base,
    586 		    &sem[semtot]));
    587 	} else {
    588 		SEM_PRINTF(("didn't find it and wasn't asked to create it\n"));
    589 		return(ENOENT);
    590 	}
    591 
    592 found:
    593 	*retval = IXSEQ_TO_IPCID(semid, sema[semid].sem_perm);
    594 	return(0);
    595 }
    596 
    597 int
    598 sys_semop(p, v, retval)
    599 	struct proc *p;
    600 	void *v;
    601 	register_t *retval;
    602 {
    603 	register struct sys_semop_args /* {
    604 		syscallarg(int) semid;
    605 		syscallarg(struct sembuf *) sops;
    606 		syscallarg(size_t) nsops;
    607 	} */ *uap = v;
    608 	int semid = SCARG(uap, semid);
    609 	int nsops = SCARG(uap, nsops);
    610 	struct sembuf sops[MAX_SOPS];
    611 	register struct semid_ds *semaptr;
    612 	register struct sembuf *sopptr = NULL;
    613 	register struct __sem *semptr = NULL;
    614 	struct sem_undo *suptr = NULL;
    615 	struct ucred *cred = p->p_ucred;
    616 	int i, j, eval;
    617 	int do_wakeup, do_undos;
    618 
    619 	SEM_PRINTF(("call to semop(%d, %p, %d)\n", semid, sops, nsops));
    620 
    621 	semlock(p);
    622 
    623 	semid = IPCID_TO_IX(semid);	/* Convert back to zero origin */
    624 
    625 	if (semid < 0 || semid >= seminfo.semmsl)
    626 		return(EINVAL);
    627 
    628 	semaptr = &sema[semid];
    629 	if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0 ||
    630 	    semaptr->sem_perm._seq != IPCID_TO_SEQ(SCARG(uap, semid)))
    631 		return(EINVAL);
    632 
    633 	if ((eval = ipcperm(cred, &semaptr->sem_perm, IPC_W))) {
    634 		SEM_PRINTF(("eval = %d from ipaccess\n", eval));
    635 		return(eval);
    636 	}
    637 
    638 	if (nsops > MAX_SOPS) {
    639 		SEM_PRINTF(("too many sops (max=%d, nsops=%d)\n", MAX_SOPS, nsops));
    640 		return(E2BIG);
    641 	}
    642 
    643 	if ((eval = copyin(SCARG(uap, sops), sops, nsops * sizeof(sops[0])))
    644 	    != 0) {
    645 		SEM_PRINTF(("eval = %d from copyin(%p, %p, %d)\n", eval,
    646 		    SCARG(uap, sops), &sops, nsops * sizeof(sops[0])));
    647 		return(eval);
    648 	}
    649 
    650 	/*
    651 	 * Loop trying to satisfy the vector of requests.
    652 	 * If we reach a point where we must wait, any requests already
    653 	 * performed are rolled back and we go to sleep until some other
    654 	 * process wakes us up.  At this point, we start all over again.
    655 	 *
    656 	 * This ensures that from the perspective of other tasks, a set
    657 	 * of requests is atomic (never partially satisfied).
    658 	 */
    659 	do_undos = 0;
    660 
    661 	for (;;) {
    662 		do_wakeup = 0;
    663 
    664 		for (i = 0; i < nsops; i++) {
    665 			sopptr = &sops[i];
    666 
    667 			if (sopptr->sem_num >= semaptr->sem_nsems)
    668 				return(EFBIG);
    669 
    670 			semptr = &semaptr->_sem_base[sopptr->sem_num];
    671 
    672 			SEM_PRINTF(("semop:  semaptr=%x, sem_base=%x, semptr=%x, sem[%d]=%d : op=%d, flag=%s\n",
    673 			    semaptr, semaptr->_sem_base, semptr,
    674 			    sopptr->sem_num, semptr->semval, sopptr->sem_op,
    675 			    (sopptr->sem_flg & IPC_NOWAIT) ? "nowait" : "wait"));
    676 
    677 			if (sopptr->sem_op < 0) {
    678 				if ((int)(semptr->semval +
    679 					  sopptr->sem_op) < 0) {
    680 					SEM_PRINTF(("semop:  can't do it now\n"));
    681 					break;
    682 				} else {
    683 					semptr->semval += sopptr->sem_op;
    684 					if (semptr->semval == 0 &&
    685 					    semptr->semzcnt > 0)
    686 						do_wakeup = 1;
    687 				}
    688 				if (sopptr->sem_flg & SEM_UNDO)
    689 					do_undos = 1;
    690 			} else if (sopptr->sem_op == 0) {
    691 				if (semptr->semval > 0) {
    692 					SEM_PRINTF(("semop:  not zero now\n"));
    693 					break;
    694 				}
    695 			} else {
    696 				if (semptr->semncnt > 0)
    697 					do_wakeup = 1;
    698 				semptr->semval += sopptr->sem_op;
    699 				if (sopptr->sem_flg & SEM_UNDO)
    700 					do_undos = 1;
    701 			}
    702 		}
    703 
    704 		/*
    705 		 * Did we get through the entire vector?
    706 		 */
    707 		if (i >= nsops)
    708 			goto done;
    709 
    710 		/*
    711 		 * No ... rollback anything that we've already done
    712 		 */
    713 		SEM_PRINTF(("semop:  rollback 0 through %d\n", i-1));
    714 		for (j = 0; j < i; j++)
    715 			semaptr->_sem_base[sops[j].sem_num].semval -=
    716 			    sops[j].sem_op;
    717 
    718 		/*
    719 		 * If the request that we couldn't satisfy has the
    720 		 * NOWAIT flag set then return with EAGAIN.
    721 		 */
    722 		if (sopptr->sem_flg & IPC_NOWAIT)
    723 			return(EAGAIN);
    724 
    725 		if (sopptr->sem_op == 0)
    726 			semptr->semzcnt++;
    727 		else
    728 			semptr->semncnt++;
    729 
    730 		SEM_PRINTF(("semop:  good night!\n"));
    731 		eval = tsleep((caddr_t)semaptr, (PZERO - 4) | PCATCH,
    732 		    "semwait", 0);
    733 		SEM_PRINTF(("semop:  good morning (eval=%d)!\n", eval));
    734 
    735 		suptr = NULL;	/* sem_undo may have been reallocated */
    736 
    737 		if (eval != 0)
    738 			return(EINTR);
    739 		SEM_PRINTF(("semop:  good morning!\n"));
    740 
    741 		/*
    742 		 * Make sure that the semaphore still exists
    743 		 */
    744 		if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0 ||
    745 		    semaptr->sem_perm._seq != IPCID_TO_SEQ(SCARG(uap, semid))) {
    746 			/* The man page says to return EIDRM. */
    747 			/* Unfortunately, BSD doesn't define that code! */
    748 #ifdef EIDRM
    749 			return(EIDRM);
    750 #else
    751 			return(EINVAL);
    752 #endif
    753 		}
    754 
    755 		/*
    756 		 * The semaphore is still alive.  Readjust the count of
    757 		 * waiting processes.
    758 		 */
    759 		if (sopptr->sem_op == 0)
    760 			semptr->semzcnt--;
    761 		else
    762 			semptr->semncnt--;
    763 	}
    764 
    765 done:
    766 	/*
    767 	 * Process any SEM_UNDO requests.
    768 	 */
    769 	if (do_undos) {
    770 		for (i = 0; i < nsops; i++) {
    771 			/*
    772 			 * We only need to deal with SEM_UNDO's for non-zero
    773 			 * op's.
    774 			 */
    775 			int adjval;
    776 
    777 			if ((sops[i].sem_flg & SEM_UNDO) == 0)
    778 				continue;
    779 			adjval = sops[i].sem_op;
    780 			if (adjval == 0)
    781 				continue;
    782 			eval = semundo_adjust(p, &suptr, semid,
    783 			    sops[i].sem_num, -adjval);
    784 			if (eval == 0)
    785 				continue;
    786 
    787 			/*
    788 			 * Oh-Oh!  We ran out of either sem_undo's or undo's.
    789 			 * Rollback the adjustments to this point and then
    790 			 * rollback the semaphore ups and down so we can return
    791 			 * with an error with all structures restored.  We
    792 			 * rollback the undo's in the exact reverse order that
    793 			 * we applied them.  This guarantees that we won't run
    794 			 * out of space as we roll things back out.
    795 			 */
    796 			for (j = i - 1; j >= 0; j--) {
    797 				if ((sops[j].sem_flg & SEM_UNDO) == 0)
    798 					continue;
    799 				adjval = sops[j].sem_op;
    800 				if (adjval == 0)
    801 					continue;
    802 				if (semundo_adjust(p, &suptr, semid,
    803 				    sops[j].sem_num, adjval) != 0)
    804 					panic("semop - can't undo undos");
    805 			}
    806 
    807 			for (j = 0; j < nsops; j++)
    808 				semaptr->_sem_base[sops[j].sem_num].semval -=
    809 				    sops[j].sem_op;
    810 
    811 			SEM_PRINTF(("eval = %d from semundo_adjust\n", eval));
    812 			return(eval);
    813 		} /* loop through the sops */
    814 	} /* if (do_undos) */
    815 
    816 	/* We're definitely done - set the sempid's */
    817 	for (i = 0; i < nsops; i++) {
    818 		sopptr = &sops[i];
    819 		semptr = &semaptr->_sem_base[sopptr->sem_num];
    820 		semptr->sempid = p->p_pid;
    821 	}
    822 
    823 	/* Do a wakeup if any semaphore was up'd. */
    824 	if (do_wakeup) {
    825 		SEM_PRINTF(("semop:  doing wakeup\n"));
    826 #ifdef SEM_WAKEUP
    827 		sem_wakeup((caddr_t)semaptr);
    828 #else
    829 		wakeup((caddr_t)semaptr);
    830 #endif
    831 		SEM_PRINTF(("semop:  back from wakeup\n"));
    832 	}
    833 	SEM_PRINTF(("semop:  done\n"));
    834 	*retval = 0;
    835 	return(0);
    836 }
    837 
    838 /*
    839  * Go through the undo structures for this process and apply the adjustments to
    840  * semaphores.
    841  */
    842 void
    843 semexit(p)
    844 	struct proc *p;
    845 {
    846 	register struct sem_undo *suptr;
    847 	register struct sem_undo **supptr;
    848 
    849 	/*
    850 	 * Go through the chain of undo vectors looking for one associated with
    851 	 * this process.
    852 	 */
    853 
    854 	for (supptr = &semu_list; (suptr = *supptr) != NULL;
    855 	    supptr = &suptr->un_next) {
    856 		if (suptr->un_proc == p)
    857 			break;
    858 	}
    859 
    860 	/*
    861 	 * There are a few possibilities to consider here ...
    862 	 *
    863 	 * 1) The semaphore facility isn't currently locked.  In this case,
    864 	 *    this call should proceed normally.
    865 	 * 2) The semaphore facility is locked by this process (i.e. the one
    866 	 *    that is exiting).  In this case, this call should proceed as
    867 	 *    usual and the facility should be unlocked at the end of this
    868 	 *    routine (since the locker is exiting).
    869 	 * 3) The semaphore facility is locked by some other process and this
    870 	 *    process doesn't have an undo structure allocated for it.  In this
    871 	 *    case, this call should proceed normally (i.e. not accomplish
    872 	 *    anything and, most importantly, not block since that is
    873 	 *    unnecessary and could result in a LOT of processes blocking in
    874 	 *    here if the facility is locked for a long time).
    875 	 * 4) The semaphore facility is locked by some other process and this
    876 	 *    process has an undo structure allocated for it.  In this case,
    877 	 *    this call should block until the facility has been unlocked since
    878 	 *    the holder of the lock may be examining this process's proc entry
    879 	 *    (the ipcs utility does this when printing out the information
    880 	 *    from the allocated sem undo elements).
    881 	 *
    882 	 * This leads to the conclusion that we should not block unless we
    883 	 * discover that the someone else has the semaphore facility locked and
    884 	 * this process has an undo structure.  Let's do that...
    885 	 *
    886 	 * Note that we do this in a separate pass from the one that processes
    887 	 * any existing undo structure since we don't want to risk blocking at
    888 	 * that time (it would make the actual unlinking of the element from
    889 	 * the chain of allocated undo structures rather messy).
    890 	 */
    891 
    892 	/*
    893 	 * Does someone else hold the semaphore facility's lock?
    894 	 */
    895 
    896 	if (semlock_holder != NULL && semlock_holder != p) {
    897 		/*
    898 		 * Yes (i.e. we are in case 3 or 4).
    899 		 *
    900 		 * If we didn't find an undo vector associated with this
    901 		 * process than we can just return (i.e. we are in case 3).
    902 		 *
    903 		 * Note that we know that someone else is holding the lock so
    904 		 * we don't even have to see if we're holding it...
    905 		 */
    906 
    907 		if (suptr == NULL)
    908 			return;
    909 
    910 		/*
    911 		 * We are in case 4.
    912 		 *
    913 		 * Go to sleep as long as someone else is locking the semaphore
    914 		 * facility (note that we won't get here if we are holding the
    915 		 * lock so we don't need to check for that possibility).
    916 		 */
    917 
    918 		while (semlock_holder != NULL)
    919 			sleep((caddr_t)&semlock_holder, (PZERO - 4));
    920 
    921 		/*
    922 		 * Nobody is holding the facility (i.e. we are now in case 1).
    923 		 * We can proceed safely according to the argument outlined
    924 		 * above.
    925 		 *
    926 		 * We look up the undo vector again, in case the list changed
    927 		 * while we were asleep, and the parent is now different.
    928 		 */
    929 
    930 		for (supptr = &semu_list; (suptr = *supptr) != NULL;
    931 		    supptr = &suptr->un_next) {
    932 			if (suptr->un_proc == p)
    933 				break;
    934 		}
    935 
    936 		if (suptr == NULL)
    937 			panic("semexit: undo vector disappeared");
    938 	} else {
    939 		/*
    940 		 * No (i.e. we are in case 1 or 2).
    941 		 *
    942 		 * If there is no undo vector, skip to the end and unlock the
    943 		 * semaphore facility if necessary.
    944 		 */
    945 
    946 		if (suptr == NULL)
    947 			goto unlock;
    948 	}
    949 
    950 	/*
    951 	 * We are now in case 1 or 2, and we have an undo vector for this
    952 	 * process.
    953 	 */
    954 
    955 	SEM_PRINTF(("proc @%p has undo structure with %d entries\n", p,
    956 	    suptr->un_cnt));
    957 
    958 	/*
    959 	 * If there are any active undo elements then process them.
    960 	 */
    961 	if (suptr->un_cnt > 0) {
    962 		int ix;
    963 
    964 		for (ix = 0; ix < suptr->un_cnt; ix++) {
    965 			int semid = suptr->un_ent[ix].un_id;
    966 			int semnum = suptr->un_ent[ix].un_num;
    967 			int adjval = suptr->un_ent[ix].un_adjval;
    968 			struct semid_ds *semaptr;
    969 
    970 			semaptr = &sema[semid];
    971 			if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0)
    972 				panic("semexit - semid not allocated");
    973 			if (semnum >= semaptr->sem_nsems)
    974 				panic("semexit - semnum out of range");
    975 
    976 			SEM_PRINTF(("semexit:  %p id=%d num=%d(adj=%d) ; sem=%d\n",
    977 			    suptr->un_proc, suptr->un_ent[ix].un_id,
    978 			    suptr->un_ent[ix].un_num,
    979 			    suptr->un_ent[ix].un_adjval,
    980 			    semaptr->_sem_base[semnum].semval));
    981 
    982 			if (adjval < 0 &&
    983 			    semaptr->_sem_base[semnum].semval < -adjval)
    984 				semaptr->_sem_base[semnum].semval = 0;
    985 			else
    986 				semaptr->_sem_base[semnum].semval += adjval;
    987 
    988 #ifdef SEM_WAKEUP
    989 			sem_wakeup((caddr_t)semaptr);
    990 #else
    991 			wakeup((caddr_t)semaptr);
    992 #endif
    993 			SEM_PRINTF(("semexit:  back from wakeup\n"));
    994 		}
    995 	}
    996 
    997 	/*
    998 	 * Deallocate the undo vector.
    999 	 */
   1000 	SEM_PRINTF(("removing vector\n"));
   1001 	suptr->un_proc = NULL;
   1002 	*supptr = suptr->un_next;
   1003 
   1004 unlock:
   1005 	/*
   1006 	 * If the exiting process is holding the global semaphore facility
   1007 	 * lock (i.e. we are in case 2) then release it.
   1008 	 */
   1009 	if (semlock_holder == p) {
   1010 		semlock_holder = NULL;
   1011 		wakeup((caddr_t)&semlock_holder);
   1012 	}
   1013 }
   1014