xref: /netbsd-src/sys/kern/sysv_sem.c (revision d20841bb642898112fe68f0ad3f7b26dddf56f07)
1 /*	$NetBSD: sysv_sem.c,v 1.49 2003/11/29 11:43:25 jdolecek 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 #include <sys/cdefs.h>
49 __KERNEL_RCSID(0, "$NetBSD: sysv_sem.c,v 1.49 2003/11/29 11:43:25 jdolecek Exp $");
50 
51 #define SYSVSEM
52 
53 #include <sys/param.h>
54 #include <sys/kernel.h>
55 #include <sys/sem.h>
56 #include <sys/sysctl.h>
57 #include <sys/mount.h>		/* XXX for <sys/syscallargs.h> */
58 #include <sys/sa.h>
59 #include <sys/syscallargs.h>
60 
61 static int	semtot = 0;
62 struct	semid_ds *sema;			/* semaphore id pool */
63 static struct	__sem *sem;		/* semaphore pool */
64 static struct	sem_undo *semu_list;	/* list of active undo structures */
65 static int	*semu;			/* undo structure pool */
66 
67 #ifdef SEM_DEBUG
68 #define SEM_PRINTF(a) printf a
69 #else
70 #define SEM_PRINTF(a)
71 #endif
72 
73 struct sem_undo *semu_alloc __P((struct proc *));
74 int semundo_adjust __P((struct proc *, struct sem_undo **, int, int, int));
75 void semundo_clear __P((int, int));
76 
77 /*
78  * XXXSMP Once we go MP, there needs to be a lock for the semaphore system.
79  * Until then, we're saved by being a non-preemptive kernel.
80  */
81 
82 void
83 seminit()
84 {
85 	int i, sz;
86 	vaddr_t v;
87 
88 	/* Allocate pageable memory for our structures */
89 	sz = seminfo.semmni * sizeof(struct semid_ds)
90 		+ seminfo.semmns * sizeof(struct __sem)
91 		+ seminfo.semmnu * seminfo.semusz;
92 	if ((v = uvm_km_zalloc(kernel_map, round_page(sz))) == 0)
93 		panic("sysv_sem: cannot allocate memory");
94 	sema = (void *)v;
95 	sem = (void *)(sema + seminfo.semmni);
96 	semu = (void *)(sem + seminfo.semmns);
97 
98 	for (i = 0; i < seminfo.semmni; i++) {
99 		sema[i]._sem_base = 0;
100 		sema[i].sem_perm.mode = 0;
101 	}
102 	for (i = 0; i < seminfo.semmnu; i++) {
103 		struct sem_undo *suptr = SEMU(i);
104 		suptr->un_proc = NULL;
105 	}
106 	semu_list = NULL;
107 	exithook_establish(semexit, NULL);
108 }
109 
110 /*
111  * Placebo.
112  */
113 
114 int
115 sys_semconfig(l, v, retval)
116 	struct lwp *l;
117 	void *v;
118 	register_t *retval;
119 {
120 	*retval = 0;
121 	return 0;
122 }
123 
124 /*
125  * Allocate a new sem_undo structure for a process
126  * (returns ptr to structure or NULL if no more room)
127  */
128 
129 struct sem_undo *
130 semu_alloc(p)
131 	struct proc *p;
132 {
133 	int i;
134 	struct sem_undo *suptr;
135 	struct sem_undo **supptr;
136 	int attempt;
137 
138 	/*
139 	 * Try twice to allocate something.
140 	 * (we'll purge any empty structures after the first pass so
141 	 * two passes are always enough)
142 	 */
143 
144 	for (attempt = 0; attempt < 2; attempt++) {
145 		/*
146 		 * Look for a free structure.
147 		 * Fill it in and return it if we find one.
148 		 */
149 
150 		for (i = 0; i < seminfo.semmnu; i++) {
151 			suptr = SEMU(i);
152 			if (suptr->un_proc == NULL) {
153 				suptr->un_next = semu_list;
154 				semu_list = suptr;
155 				suptr->un_cnt = 0;
156 				suptr->un_proc = p;
157 				return(suptr);
158 			}
159 		}
160 
161 		/*
162 		 * We didn't find a free one, if this is the first attempt
163 		 * then try to free some structures.
164 		 */
165 
166 		if (attempt == 0) {
167 			/* All the structures are in use - try to free some */
168 			int did_something = 0;
169 
170 			supptr = &semu_list;
171 			while ((suptr = *supptr) != NULL) {
172 				if (suptr->un_cnt == 0)  {
173 					suptr->un_proc = NULL;
174 					*supptr = suptr->un_next;
175 					did_something = 1;
176 				} else
177 					supptr = &(suptr->un_next);
178 			}
179 
180 			/* If we didn't free anything then just give-up */
181 			if (!did_something)
182 				return(NULL);
183 		} else {
184 			/*
185 			 * The second pass failed even though we freed
186 			 * something after the first pass!
187 			 * This is IMPOSSIBLE!
188 			 */
189 			panic("semu_alloc - second attempt failed");
190 		}
191 	}
192 	return NULL;
193 }
194 
195 /*
196  * Adjust a particular entry for a particular proc
197  */
198 
199 int
200 semundo_adjust(p, supptr, semid, semnum, adjval)
201 	struct proc *p;
202 	struct sem_undo **supptr;
203 	int semid, semnum;
204 	int adjval;
205 {
206 	struct sem_undo *suptr;
207 	struct undo *sunptr;
208 	int i;
209 
210 	/* Look for and remember the sem_undo if the caller doesn't provide
211 	   it */
212 
213 	suptr = *supptr;
214 	if (suptr == NULL) {
215 		for (suptr = semu_list; suptr != NULL; suptr = suptr->un_next) {
216 			if (suptr->un_proc == p) {
217 				*supptr = suptr;
218 				break;
219 			}
220 		}
221 		if (suptr == NULL) {
222 			if (adjval == 0)
223 				return(0);
224 			suptr = semu_alloc(p);
225 			if (suptr == NULL)
226 				return(ENOSPC);
227 			*supptr = suptr;
228 		}
229 	}
230 
231 	/*
232 	 * Look for the requested entry and adjust it (delete if adjval becomes
233 	 * 0).
234 	 */
235 	sunptr = &suptr->un_ent[0];
236 	for (i = 0; i < suptr->un_cnt; i++, sunptr++) {
237 		if (sunptr->un_id != semid || sunptr->un_num != semnum)
238 			continue;
239 		if (adjval == 0)
240 			sunptr->un_adjval = 0;
241 		else
242 			sunptr->un_adjval += adjval;
243 		if (sunptr->un_adjval == 0) {
244 			suptr->un_cnt--;
245 			if (i < suptr->un_cnt)
246 				suptr->un_ent[i] =
247 				    suptr->un_ent[suptr->un_cnt];
248 		}
249 		return(0);
250 	}
251 
252 	/* Didn't find the right entry - create it */
253 	if (adjval == 0)
254 		return(0);
255 	if (suptr->un_cnt == SEMUME)
256 		return(EINVAL);
257 
258 	sunptr = &suptr->un_ent[suptr->un_cnt];
259 	suptr->un_cnt++;
260 	sunptr->un_adjval = adjval;
261 	sunptr->un_id = semid;
262 	sunptr->un_num = semnum;
263 	return(0);
264 }
265 
266 void
267 semundo_clear(semid, semnum)
268 	int semid, semnum;
269 {
270 	struct sem_undo *suptr;
271 
272 	for (suptr = semu_list; suptr != NULL; suptr = suptr->un_next) {
273 		struct undo *sunptr;
274 		int i;
275 
276 		sunptr = &suptr->un_ent[0];
277 		for (i = 0; i < suptr->un_cnt; i++, sunptr++) {
278 			if (sunptr->un_id == semid) {
279 				if (semnum == -1 || sunptr->un_num == semnum) {
280 					suptr->un_cnt--;
281 					if (i < suptr->un_cnt) {
282 						suptr->un_ent[i] =
283 						  suptr->un_ent[suptr->un_cnt];
284 						i--, sunptr--;
285 					}
286 				}
287 				if (semnum != -1)
288 					break;
289 			}
290 		}
291 	}
292 }
293 
294 int
295 sys_____semctl13(l, v, retval)
296 	struct lwp *l;
297 	void *v;
298 	register_t *retval;
299 {
300 	struct sys_____semctl13_args /* {
301 		syscallarg(int) semid;
302 		syscallarg(int) semnum;
303 		syscallarg(int) cmd;
304 		syscallarg(union __semun *) arg;
305 	} */ *uap = v;
306 	struct proc *p = l->l_proc;
307 	struct semid_ds sembuf;
308 	int cmd, error;
309 	void *pass_arg;
310 	union __semun karg;
311 
312 	cmd = SCARG(uap, cmd);
313 
314 	switch (cmd) {
315 	case IPC_SET:
316 	case IPC_STAT:
317 		pass_arg = &sembuf;
318 		break;
319 
320 	case GETALL:
321 	case SETVAL:
322 	case SETALL:
323 		pass_arg = &karg;
324 		break;
325 	default:
326 		pass_arg = NULL;
327 		break;
328 	}
329 
330 	if (pass_arg) {
331 		error = copyin(SCARG(uap, arg), &karg, sizeof(karg));
332 		if (error)
333 			return error;
334 		if (cmd == IPC_SET) {
335 			error = copyin(karg.buf, &sembuf, sizeof(sembuf));
336 			if (error)
337 				return (error);
338 		}
339 	}
340 
341 	error = semctl1(p, SCARG(uap, semid), SCARG(uap, semnum), cmd,
342 	    pass_arg, retval);
343 
344 	if (error == 0 && cmd == IPC_STAT)
345 		error = copyout(&sembuf, karg.buf, sizeof(sembuf));
346 
347 	return (error);
348 }
349 
350 int
351 semctl1(p, semid, semnum, cmd, v, retval)
352 	struct proc *p;
353 	int semid, semnum, cmd;
354 	void *v;
355 	register_t *retval;
356 {
357 	struct ucred *cred = p->p_ucred;
358 	union __semun *arg = v;
359 	struct semid_ds *sembuf = v, *semaptr;
360 	int i, error, ix;
361 
362 	SEM_PRINTF(("call to semctl(%d, %d, %d, %p)\n",
363 	    semid, semnum, cmd, v));
364 
365 	ix = IPCID_TO_IX(semid);
366 	if (ix < 0 || ix >= seminfo.semmni)
367 		return (EINVAL);
368 
369 	semaptr = &sema[ix];
370 	if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0 ||
371 	    semaptr->sem_perm._seq != IPCID_TO_SEQ(semid))
372 		return (EINVAL);
373 
374 	switch (cmd) {
375 	case IPC_RMID:
376 		if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_M)) != 0)
377 			return (error);
378 		semaptr->sem_perm.cuid = cred->cr_uid;
379 		semaptr->sem_perm.uid = cred->cr_uid;
380 		semtot -= semaptr->sem_nsems;
381 		for (i = semaptr->_sem_base - sem; i < semtot; i++)
382 			sem[i] = sem[i + semaptr->sem_nsems];
383 		for (i = 0; i < seminfo.semmni; i++) {
384 			if ((sema[i].sem_perm.mode & SEM_ALLOC) &&
385 			    sema[i]._sem_base > semaptr->_sem_base)
386 				sema[i]._sem_base -= semaptr->sem_nsems;
387 		}
388 		semaptr->sem_perm.mode = 0;
389 		semundo_clear(ix, -1);
390 		wakeup(semaptr);
391 		break;
392 
393 	case IPC_SET:
394 		if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_M)))
395 			return (error);
396 		semaptr->sem_perm.uid = sembuf->sem_perm.uid;
397 		semaptr->sem_perm.gid = sembuf->sem_perm.gid;
398 		semaptr->sem_perm.mode = (semaptr->sem_perm.mode & ~0777) |
399 		    (sembuf->sem_perm.mode & 0777);
400 		semaptr->sem_ctime = time.tv_sec;
401 		break;
402 
403 	case IPC_STAT:
404 		if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_R)))
405 			return (error);
406 		memcpy(sembuf, semaptr, sizeof(struct semid_ds));
407 		break;
408 
409 	case GETNCNT:
410 		if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_R)))
411 			return (error);
412 		if (semnum < 0 || semnum >= semaptr->sem_nsems)
413 			return (EINVAL);
414 		*retval = semaptr->_sem_base[semnum].semncnt;
415 		break;
416 
417 	case GETPID:
418 		if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_R)))
419 			return (error);
420 		if (semnum < 0 || semnum >= semaptr->sem_nsems)
421 			return (EINVAL);
422 		*retval = semaptr->_sem_base[semnum].sempid;
423 		break;
424 
425 	case GETVAL:
426 		if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_R)))
427 			return (error);
428 		if (semnum < 0 || semnum >= semaptr->sem_nsems)
429 			return (EINVAL);
430 		*retval = semaptr->_sem_base[semnum].semval;
431 		break;
432 
433 	case GETALL:
434 		if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_R)))
435 			return (error);
436 		for (i = 0; i < semaptr->sem_nsems; i++) {
437 			error = copyout(&semaptr->_sem_base[i].semval,
438 			    &arg->array[i], sizeof(arg->array[i]));
439 			if (error != 0)
440 				break;
441 		}
442 		break;
443 
444 	case GETZCNT:
445 		if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_R)))
446 			return (error);
447 		if (semnum < 0 || semnum >= semaptr->sem_nsems)
448 			return (EINVAL);
449 		*retval = semaptr->_sem_base[semnum].semzcnt;
450 		break;
451 
452 	case SETVAL:
453 		if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_W)))
454 			return (error);
455 		if (semnum < 0 || semnum >= semaptr->sem_nsems)
456 			return (EINVAL);
457 		semaptr->_sem_base[semnum].semval = arg->val;
458 		semundo_clear(ix, semnum);
459 		wakeup(semaptr);
460 		break;
461 
462 	case SETALL:
463 		if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_W)))
464 			return (error);
465 		for (i = 0; i < semaptr->sem_nsems; i++) {
466 			error = copyin(&arg->array[i],
467 			    &semaptr->_sem_base[i].semval,
468 			    sizeof(arg->array[i]));
469 			if (error != 0)
470 				break;
471 		}
472 		semundo_clear(ix, -1);
473 		wakeup(semaptr);
474 		break;
475 
476 	default:
477 		return (EINVAL);
478 	}
479 
480 	return (error);
481 }
482 
483 int
484 sys_semget(l, v, retval)
485 	struct lwp *l;
486 	void *v;
487 	register_t *retval;
488 {
489 	struct sys_semget_args /* {
490 		syscallarg(key_t) key;
491 		syscallarg(int) nsems;
492 		syscallarg(int) semflg;
493 	} */ *uap = v;
494 	int semid, eval;
495 	int key = SCARG(uap, key);
496 	int nsems = SCARG(uap, nsems);
497 	int semflg = SCARG(uap, semflg);
498 	struct ucred *cred = l->l_proc->p_ucred;
499 
500 	SEM_PRINTF(("semget(0x%x, %d, 0%o)\n", key, nsems, semflg));
501 
502 	if (key != IPC_PRIVATE) {
503 		for (semid = 0; semid < seminfo.semmni; semid++) {
504 			if ((sema[semid].sem_perm.mode & SEM_ALLOC) &&
505 			    sema[semid].sem_perm._key == key)
506 				break;
507 		}
508 		if (semid < seminfo.semmni) {
509 			SEM_PRINTF(("found public key\n"));
510 			if ((eval = ipcperm(cred, &sema[semid].sem_perm,
511 			    semflg & 0700)))
512 				return(eval);
513 			if (nsems > 0 && sema[semid].sem_nsems < nsems) {
514 				SEM_PRINTF(("too small\n"));
515 				return(EINVAL);
516 			}
517 			if ((semflg & IPC_CREAT) && (semflg & IPC_EXCL)) {
518 				SEM_PRINTF(("not exclusive\n"));
519 				return(EEXIST);
520 			}
521 			goto found;
522 		}
523 	}
524 
525 	SEM_PRINTF(("need to allocate the semid_ds\n"));
526 	if (key == IPC_PRIVATE || (semflg & IPC_CREAT)) {
527 		if (nsems <= 0 || nsems > seminfo.semmsl) {
528 			SEM_PRINTF(("nsems out of range (0<%d<=%d)\n", nsems,
529 			    seminfo.semmsl));
530 			return(EINVAL);
531 		}
532 		if (nsems > seminfo.semmns - semtot) {
533 			SEM_PRINTF(("not enough semaphores left (need %d, got %d)\n",
534 			    nsems, seminfo.semmns - semtot));
535 			return(ENOSPC);
536 		}
537 		for (semid = 0; semid < seminfo.semmni; semid++) {
538 			if ((sema[semid].sem_perm.mode & SEM_ALLOC) == 0)
539 				break;
540 		}
541 		if (semid == seminfo.semmni) {
542 			SEM_PRINTF(("no more semid_ds's available\n"));
543 			return(ENOSPC);
544 		}
545 		SEM_PRINTF(("semid %d is available\n", semid));
546 		sema[semid].sem_perm._key = key;
547 		sema[semid].sem_perm.cuid = cred->cr_uid;
548 		sema[semid].sem_perm.uid = cred->cr_uid;
549 		sema[semid].sem_perm.cgid = cred->cr_gid;
550 		sema[semid].sem_perm.gid = cred->cr_gid;
551 		sema[semid].sem_perm.mode = (semflg & 0777) | SEM_ALLOC;
552 		sema[semid].sem_perm._seq =
553 		    (sema[semid].sem_perm._seq + 1) & 0x7fff;
554 		sema[semid].sem_nsems = nsems;
555 		sema[semid].sem_otime = 0;
556 		sema[semid].sem_ctime = time.tv_sec;
557 		sema[semid]._sem_base = &sem[semtot];
558 		semtot += nsems;
559 		memset(sema[semid]._sem_base, 0,
560 		    sizeof(sema[semid]._sem_base[0])*nsems);
561 		SEM_PRINTF(("sembase = %p, next = %p\n", sema[semid]._sem_base,
562 		    &sem[semtot]));
563 	} else {
564 		SEM_PRINTF(("didn't find it and wasn't asked to create it\n"));
565 		return(ENOENT);
566 	}
567 
568 found:
569 	*retval = IXSEQ_TO_IPCID(semid, sema[semid].sem_perm);
570 	return(0);
571 }
572 
573 int
574 sys_semop(l, v, retval)
575 	struct lwp *l;
576 	void *v;
577 	register_t *retval;
578 {
579 	struct sys_semop_args /* {
580 		syscallarg(int) semid;
581 		syscallarg(struct sembuf *) sops;
582 		syscallarg(size_t) nsops;
583 	} */ *uap = v;
584 	struct proc *p = l->l_proc;
585 	int semid = SCARG(uap, semid);
586 	size_t nsops = SCARG(uap, nsops);
587 	struct sembuf sops[MAX_SOPS];
588 	struct semid_ds *semaptr;
589 	struct sembuf *sopptr = NULL;
590 	struct __sem *semptr = NULL;
591 	struct sem_undo *suptr = NULL;
592 	struct ucred *cred = p->p_ucred;
593 	int i, j, eval;
594 	int do_wakeup, do_undos;
595 
596 	SEM_PRINTF(("call to semop(%d, %p, %lld)\n", semid, sops,
597 	    (long long)nsops));
598 
599 	semid = IPCID_TO_IX(semid);	/* Convert back to zero origin */
600 	if (semid < 0 || semid >= seminfo.semmni)
601 		return(EINVAL);
602 
603 	semaptr = &sema[semid];
604 	if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0 ||
605 	    semaptr->sem_perm._seq != IPCID_TO_SEQ(SCARG(uap, semid)))
606 		return(EINVAL);
607 
608 	if ((eval = ipcperm(cred, &semaptr->sem_perm, IPC_W))) {
609 		SEM_PRINTF(("eval = %d from ipaccess\n", eval));
610 		return(eval);
611 	}
612 
613 	if (nsops > MAX_SOPS) {
614 		SEM_PRINTF(("too many sops (max=%d, nsops=%lld)\n", MAX_SOPS,
615 		    (long long)nsops));
616 		return(E2BIG);
617 	}
618 
619 	if ((eval = copyin(SCARG(uap, sops), sops, nsops * sizeof(sops[0])))
620 	    != 0) {
621 		SEM_PRINTF(("eval = %d from copyin(%p, %p, %lld)\n", eval,
622 		    SCARG(uap, sops), &sops,
623 		    (long long)(nsops * sizeof(sops[0]))));
624 		return(eval);
625 	}
626 
627 	/*
628 	 * Loop trying to satisfy the vector of requests.
629 	 * If we reach a point where we must wait, any requests already
630 	 * performed are rolled back and we go to sleep until some other
631 	 * process wakes us up.  At this point, we start all over again.
632 	 *
633 	 * This ensures that from the perspective of other tasks, a set
634 	 * of requests is atomic (never partially satisfied).
635 	 */
636 	do_undos = 0;
637 
638 	for (;;) {
639 		do_wakeup = 0;
640 
641 		for (i = 0; i < nsops; i++) {
642 			sopptr = &sops[i];
643 
644 			if (sopptr->sem_num >= semaptr->sem_nsems)
645 				return(EFBIG);
646 
647 			semptr = &semaptr->_sem_base[sopptr->sem_num];
648 
649 			SEM_PRINTF(("semop:  semaptr=%p, sem_base=%p, semptr=%p, sem[%d]=%d : op=%d, flag=%s\n",
650 			    semaptr, semaptr->_sem_base, semptr,
651 			    sopptr->sem_num, semptr->semval, sopptr->sem_op,
652 			    (sopptr->sem_flg & IPC_NOWAIT) ? "nowait" : "wait"));
653 
654 			if (sopptr->sem_op < 0) {
655 				if ((int)(semptr->semval +
656 					  sopptr->sem_op) < 0) {
657 					SEM_PRINTF(("semop:  can't do it now\n"));
658 					break;
659 				} else {
660 					semptr->semval += sopptr->sem_op;
661 					if (semptr->semval == 0 &&
662 					    semptr->semzcnt > 0)
663 						do_wakeup = 1;
664 				}
665 				if (sopptr->sem_flg & SEM_UNDO)
666 					do_undos = 1;
667 			} else if (sopptr->sem_op == 0) {
668 				if (semptr->semval > 0) {
669 					SEM_PRINTF(("semop:  not zero now\n"));
670 					break;
671 				}
672 			} else {
673 				if (semptr->semncnt > 0)
674 					do_wakeup = 1;
675 				semptr->semval += sopptr->sem_op;
676 				if (sopptr->sem_flg & SEM_UNDO)
677 					do_undos = 1;
678 			}
679 		}
680 
681 		/*
682 		 * Did we get through the entire vector?
683 		 */
684 		if (i >= nsops)
685 			goto done;
686 
687 		/*
688 		 * No ... rollback anything that we've already done
689 		 */
690 		SEM_PRINTF(("semop:  rollback 0 through %d\n", i-1));
691 		for (j = 0; j < i; j++)
692 			semaptr->_sem_base[sops[j].sem_num].semval -=
693 			    sops[j].sem_op;
694 
695 		/*
696 		 * If the request that we couldn't satisfy has the
697 		 * NOWAIT flag set then return with EAGAIN.
698 		 */
699 		if (sopptr->sem_flg & IPC_NOWAIT)
700 			return(EAGAIN);
701 
702 		if (sopptr->sem_op == 0)
703 			semptr->semzcnt++;
704 		else
705 			semptr->semncnt++;
706 
707 		SEM_PRINTF(("semop:  good night!\n"));
708 		eval = tsleep((caddr_t)semaptr, (PZERO - 4) | PCATCH,
709 		    "semwait", 0);
710 		SEM_PRINTF(("semop:  good morning (eval=%d)!\n", eval));
711 
712 		suptr = NULL;	/* sem_undo may have been reallocated */
713 
714 		if (eval != 0)
715 			return(EINTR);
716 		SEM_PRINTF(("semop:  good morning!\n"));
717 
718 		/*
719 		 * Make sure that the semaphore still exists
720 		 */
721 		if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0 ||
722 		    semaptr->sem_perm._seq != IPCID_TO_SEQ(SCARG(uap, semid))) {
723 			/* The man page says to return EIDRM. */
724 			/* Unfortunately, BSD doesn't define that code! */
725 #ifdef EIDRM
726 			return(EIDRM);
727 #else
728 			return(EINVAL);
729 #endif
730 		}
731 
732 		/*
733 		 * The semaphore is still alive.  Readjust the count of
734 		 * waiting processes.
735 		 */
736 		if (sopptr->sem_op == 0)
737 			semptr->semzcnt--;
738 		else
739 			semptr->semncnt--;
740 	}
741 
742 done:
743 	/*
744 	 * Process any SEM_UNDO requests.
745 	 */
746 	if (do_undos) {
747 		for (i = 0; i < nsops; i++) {
748 			/*
749 			 * We only need to deal with SEM_UNDO's for non-zero
750 			 * op's.
751 			 */
752 			int adjval;
753 
754 			if ((sops[i].sem_flg & SEM_UNDO) == 0)
755 				continue;
756 			adjval = sops[i].sem_op;
757 			if (adjval == 0)
758 				continue;
759 			eval = semundo_adjust(p, &suptr, semid,
760 			    sops[i].sem_num, -adjval);
761 			if (eval == 0)
762 				continue;
763 
764 			/*
765 			 * Oh-Oh!  We ran out of either sem_undo's or undo's.
766 			 * Rollback the adjustments to this point and then
767 			 * rollback the semaphore ups and down so we can return
768 			 * with an error with all structures restored.  We
769 			 * rollback the undo's in the exact reverse order that
770 			 * we applied them.  This guarantees that we won't run
771 			 * out of space as we roll things back out.
772 			 */
773 			for (j = i - 1; j >= 0; j--) {
774 				if ((sops[j].sem_flg & SEM_UNDO) == 0)
775 					continue;
776 				adjval = sops[j].sem_op;
777 				if (adjval == 0)
778 					continue;
779 				if (semundo_adjust(p, &suptr, semid,
780 				    sops[j].sem_num, adjval) != 0)
781 					panic("semop - can't undo undos");
782 			}
783 
784 			for (j = 0; j < nsops; j++)
785 				semaptr->_sem_base[sops[j].sem_num].semval -=
786 				    sops[j].sem_op;
787 
788 			SEM_PRINTF(("eval = %d from semundo_adjust\n", eval));
789 			return(eval);
790 		} /* loop through the sops */
791 	} /* if (do_undos) */
792 
793 	/* We're definitely done - set the sempid's */
794 	for (i = 0; i < nsops; i++) {
795 		sopptr = &sops[i];
796 		semptr = &semaptr->_sem_base[sopptr->sem_num];
797 		semptr->sempid = p->p_pid;
798 	}
799 
800 	/* Do a wakeup if any semaphore was up'd. */
801 	if (do_wakeup) {
802 		SEM_PRINTF(("semop:  doing wakeup\n"));
803 #ifdef SEM_WAKEUP
804 		sem_wakeup((caddr_t)semaptr);
805 #else
806 		wakeup((caddr_t)semaptr);
807 #endif
808 		SEM_PRINTF(("semop:  back from wakeup\n"));
809 	}
810 	SEM_PRINTF(("semop:  done\n"));
811 	*retval = 0;
812 	return(0);
813 }
814 
815 /*
816  * Go through the undo structures for this process and apply the adjustments to
817  * semaphores.
818  */
819 /*ARGSUSED*/
820 void
821 semexit(p, v)
822 	struct proc *p;
823 	void *v;
824 {
825 	struct sem_undo *suptr;
826 	struct sem_undo **supptr;
827 
828 	/*
829 	 * Go through the chain of undo vectors looking for one associated with
830 	 * this process.
831 	 */
832 
833 	for (supptr = &semu_list; (suptr = *supptr) != NULL;
834 	    supptr = &suptr->un_next) {
835 		if (suptr->un_proc == p)
836 			break;
837 	}
838 
839 	/*
840 	 * If there is no undo vector, skip to the end.
841 	 */
842 
843 	if (suptr == NULL)
844 		return;
845 
846 	/*
847 	 * We now have an undo vector for this process.
848 	 */
849 
850 	SEM_PRINTF(("proc @%p has undo structure with %d entries\n", p,
851 	    suptr->un_cnt));
852 
853 	/*
854 	 * If there are any active undo elements then process them.
855 	 */
856 	if (suptr->un_cnt > 0) {
857 		int ix;
858 
859 		for (ix = 0; ix < suptr->un_cnt; ix++) {
860 			int semid = suptr->un_ent[ix].un_id;
861 			int semnum = suptr->un_ent[ix].un_num;
862 			int adjval = suptr->un_ent[ix].un_adjval;
863 			struct semid_ds *semaptr;
864 
865 			semaptr = &sema[semid];
866 			if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0)
867 				panic("semexit - semid not allocated");
868 			if (semnum >= semaptr->sem_nsems)
869 				panic("semexit - semnum out of range");
870 
871 			SEM_PRINTF(("semexit:  %p id=%d num=%d(adj=%d) ; sem=%d\n",
872 			    suptr->un_proc, suptr->un_ent[ix].un_id,
873 			    suptr->un_ent[ix].un_num,
874 			    suptr->un_ent[ix].un_adjval,
875 			    semaptr->_sem_base[semnum].semval));
876 
877 			if (adjval < 0 &&
878 			    semaptr->_sem_base[semnum].semval < -adjval)
879 				semaptr->_sem_base[semnum].semval = 0;
880 			else
881 				semaptr->_sem_base[semnum].semval += adjval;
882 
883 #ifdef SEM_WAKEUP
884 			sem_wakeup((caddr_t)semaptr);
885 #else
886 			wakeup((caddr_t)semaptr);
887 #endif
888 			SEM_PRINTF(("semexit:  back from wakeup\n"));
889 		}
890 	}
891 
892 	/*
893 	 * Deallocate the undo vector.
894 	 */
895 	SEM_PRINTF(("removing vector\n"));
896 	suptr->un_proc = NULL;
897 	*supptr = suptr->un_next;
898 }
899