xref: /netbsd-src/sys/kern/sysv_sem.c (revision 21e37cc72a480a47828990a439cde7ac9ffaf0c6)
1 /*	$NetBSD: sysv_sem.c,v 1.53 2004/03/23 13:22:04 junyoung 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 1.53 2004/03/23 13:22:04 junyoung 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(struct proc *);
74 int semundo_adjust(struct proc *, struct sem_undo **, int, int, int);
75 void semundo_clear(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 
121 	*retval = 0;
122 	return 0;
123 }
124 
125 /*
126  * Allocate a new sem_undo structure for a process
127  * (returns ptr to structure or NULL if no more room)
128  */
129 
130 struct sem_undo *
131 semu_alloc(p)
132 	struct proc *p;
133 {
134 	int i;
135 	struct sem_undo *suptr;
136 	struct sem_undo **supptr;
137 	int attempt;
138 
139 	/*
140 	 * Try twice to allocate something.
141 	 * (we'll purge any empty structures after the first pass so
142 	 * two passes are always enough)
143 	 */
144 
145 	for (attempt = 0; attempt < 2; attempt++) {
146 		/*
147 		 * Look for a free structure.
148 		 * Fill it in and return it if we find one.
149 		 */
150 
151 		for (i = 0; i < seminfo.semmnu; i++) {
152 			suptr = SEMU(i);
153 			if (suptr->un_proc == NULL) {
154 				suptr->un_next = semu_list;
155 				semu_list = suptr;
156 				suptr->un_cnt = 0;
157 				suptr->un_proc = p;
158 				return (suptr);
159 			}
160 		}
161 
162 		/*
163 		 * We didn't find a free one, if this is the first attempt
164 		 * then try to free some structures.
165 		 */
166 
167 		if (attempt == 0) {
168 			/* All the structures are in use - try to free some */
169 			int did_something = 0;
170 
171 			supptr = &semu_list;
172 			while ((suptr = *supptr) != NULL) {
173 				if (suptr->un_cnt == 0)  {
174 					suptr->un_proc = NULL;
175 					*supptr = suptr->un_next;
176 					did_something = 1;
177 				} else
178 					supptr = &suptr->un_next;
179 			}
180 
181 			/* If we didn't free anything then just give-up */
182 			if (!did_something)
183 				return (NULL);
184 		} else {
185 			/*
186 			 * The second pass failed even though we freed
187 			 * something after the first pass!
188 			 * This is IMPOSSIBLE!
189 			 */
190 			panic("semu_alloc - second attempt failed");
191 		}
192 	}
193 	return NULL;
194 }
195 
196 /*
197  * Adjust a particular entry for a particular proc
198  */
199 
200 int
201 semundo_adjust(p, supptr, semid, semnum, adjval)
202 	struct proc *p;
203 	struct sem_undo **supptr;
204 	int semid, semnum;
205 	int adjval;
206 {
207 	struct sem_undo *suptr;
208 	struct undo *sunptr;
209 	int i;
210 
211 	/*
212 	 * Look for and remember the sem_undo if the caller doesn't
213 	 * provide it
214 	 */
215 
216 	suptr = *supptr;
217 	if (suptr == NULL) {
218 		for (suptr = semu_list; suptr != NULL; suptr = suptr->un_next)
219 			if (suptr->un_proc == p)
220 				break;
221 
222 		if (suptr == NULL) {
223 			suptr = semu_alloc(p);
224 			if (suptr == NULL)
225 				return (ENOSPC);
226 		}
227 		*supptr = suptr;
228 	}
229 
230 	/*
231 	 * Look for the requested entry and adjust it (delete if
232 	 * adjval becomes 0).
233 	 */
234 	sunptr = &suptr->un_ent[0];
235 	for (i = 0; i < suptr->un_cnt; i++, sunptr++) {
236 		if (sunptr->un_id != semid || sunptr->un_num != semnum)
237 			continue;
238 		sunptr->un_adjval += adjval;
239 		if (sunptr->un_adjval == 0) {
240 			suptr->un_cnt--;
241 			if (i < suptr->un_cnt)
242 				suptr->un_ent[i] =
243 				    suptr->un_ent[suptr->un_cnt];
244 		}
245 		return (0);
246 	}
247 
248 	/* Didn't find the right entry - create it */
249 	if (suptr->un_cnt == SEMUME)
250 		return (EINVAL);
251 
252 	sunptr = &suptr->un_ent[suptr->un_cnt];
253 	suptr->un_cnt++;
254 	sunptr->un_adjval = adjval;
255 	sunptr->un_id = semid;
256 	sunptr->un_num = semnum;
257 	return (0);
258 }
259 
260 void
261 semundo_clear(semid, semnum)
262 	int semid, semnum;
263 {
264 	struct sem_undo *suptr;
265 	struct undo *sunptr, *sunend;
266 
267 	for (suptr = semu_list; suptr != NULL; suptr = suptr->un_next)
268 		for (sunptr = &suptr->un_ent[0],
269 		    sunend = sunptr + suptr->un_cnt; sunptr < sunend;) {
270 			if (sunptr->un_id == semid) {
271 				if (semnum == -1 || sunptr->un_num == semnum) {
272 					suptr->un_cnt--;
273 					sunend--;
274 					if (sunptr != sunend)
275 						*sunptr = *sunend;
276 					if (semnum != -1)
277 						break;
278 					else
279 						continue;
280 				}
281 			}
282 			sunptr++;
283 		}
284 }
285 
286 int
287 sys_____semctl13(l, v, retval)
288 	struct lwp *l;
289 	void *v;
290 	register_t *retval;
291 {
292 	struct sys_____semctl13_args /* {
293 		syscallarg(int) semid;
294 		syscallarg(int) semnum;
295 		syscallarg(int) cmd;
296 		syscallarg(union __semun *) arg;
297 	} */ *uap = v;
298 	struct proc *p = l->l_proc;
299 	struct semid_ds sembuf;
300 	int cmd, error;
301 	void *pass_arg;
302 	union __semun karg;
303 
304 	cmd = SCARG(uap, cmd);
305 
306 	switch (cmd) {
307 	case IPC_SET:
308 	case IPC_STAT:
309 		pass_arg = &sembuf;
310 		break;
311 
312 	case GETALL:
313 	case SETVAL:
314 	case SETALL:
315 		pass_arg = &karg;
316 		break;
317 	default:
318 		pass_arg = NULL;
319 		break;
320 	}
321 
322 	if (pass_arg) {
323 		error = copyin(SCARG(uap, arg), &karg, sizeof(karg));
324 		if (error)
325 			return error;
326 		if (cmd == IPC_SET) {
327 			error = copyin(karg.buf, &sembuf, sizeof(sembuf));
328 			if (error)
329 				return (error);
330 		}
331 	}
332 
333 	error = semctl1(p, SCARG(uap, semid), SCARG(uap, semnum), cmd,
334 	    pass_arg, retval);
335 
336 	if (error == 0 && cmd == IPC_STAT)
337 		error = copyout(&sembuf, karg.buf, sizeof(sembuf));
338 
339 	return (error);
340 }
341 
342 int
343 semctl1(p, semid, semnum, cmd, v, retval)
344 	struct proc *p;
345 	int semid, semnum, cmd;
346 	void *v;
347 	register_t *retval;
348 {
349 	struct ucred *cred = p->p_ucred;
350 	union __semun *arg = v;
351 	struct semid_ds *sembuf = v, *semaptr;
352 	int i, error, ix;
353 
354 	SEM_PRINTF(("call to semctl(%d, %d, %d, %p)\n",
355 	    semid, semnum, cmd, v));
356 
357 	ix = IPCID_TO_IX(semid);
358 	if (ix < 0 || ix >= seminfo.semmni)
359 		return (EINVAL);
360 
361 	semaptr = &sema[ix];
362 	if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0 ||
363 	    semaptr->sem_perm._seq != IPCID_TO_SEQ(semid))
364 		return (EINVAL);
365 
366 	switch (cmd) {
367 	case IPC_RMID:
368 		if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_M)) != 0)
369 			return (error);
370 		semaptr->sem_perm.cuid = cred->cr_uid;
371 		semaptr->sem_perm.uid = cred->cr_uid;
372 		semtot -= semaptr->sem_nsems;
373 		for (i = semaptr->_sem_base - sem; i < semtot; i++)
374 			sem[i] = sem[i + semaptr->sem_nsems];
375 		for (i = 0; i < seminfo.semmni; i++) {
376 			if ((sema[i].sem_perm.mode & SEM_ALLOC) &&
377 			    sema[i]._sem_base > semaptr->_sem_base)
378 				sema[i]._sem_base -= semaptr->sem_nsems;
379 		}
380 		semaptr->sem_perm.mode = 0;
381 		semundo_clear(ix, -1);
382 		wakeup(semaptr);
383 		break;
384 
385 	case IPC_SET:
386 		if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_M)))
387 			return (error);
388 		semaptr->sem_perm.uid = sembuf->sem_perm.uid;
389 		semaptr->sem_perm.gid = sembuf->sem_perm.gid;
390 		semaptr->sem_perm.mode = (semaptr->sem_perm.mode & ~0777) |
391 		    (sembuf->sem_perm.mode & 0777);
392 		semaptr->sem_ctime = time.tv_sec;
393 		break;
394 
395 	case IPC_STAT:
396 		if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_R)))
397 			return (error);
398 		memcpy(sembuf, semaptr, sizeof(struct semid_ds));
399 		break;
400 
401 	case GETNCNT:
402 		if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_R)))
403 			return (error);
404 		if (semnum < 0 || semnum >= semaptr->sem_nsems)
405 			return (EINVAL);
406 		*retval = semaptr->_sem_base[semnum].semncnt;
407 		break;
408 
409 	case GETPID:
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].sempid;
415 		break;
416 
417 	case GETVAL:
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].semval;
423 		break;
424 
425 	case GETALL:
426 		if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_R)))
427 			return (error);
428 		for (i = 0; i < semaptr->sem_nsems; i++) {
429 			error = copyout(&semaptr->_sem_base[i].semval,
430 			    &arg->array[i], sizeof(arg->array[i]));
431 			if (error != 0)
432 				break;
433 		}
434 		break;
435 
436 	case GETZCNT:
437 		if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_R)))
438 			return (error);
439 		if (semnum < 0 || semnum >= semaptr->sem_nsems)
440 			return (EINVAL);
441 		*retval = semaptr->_sem_base[semnum].semzcnt;
442 		break;
443 
444 	case SETVAL:
445 		if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_W)))
446 			return (error);
447 		if (semnum < 0 || semnum >= semaptr->sem_nsems)
448 			return (EINVAL);
449 		semaptr->_sem_base[semnum].semval = arg->val;
450 		semundo_clear(ix, semnum);
451 		wakeup(semaptr);
452 		break;
453 
454 	case SETALL:
455 		if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_W)))
456 			return (error);
457 		for (i = 0; i < semaptr->sem_nsems; i++) {
458 			error = copyin(&arg->array[i],
459 			    &semaptr->_sem_base[i].semval,
460 			    sizeof(arg->array[i]));
461 			if (error != 0)
462 				break;
463 		}
464 		semundo_clear(ix, -1);
465 		wakeup(semaptr);
466 		break;
467 
468 	default:
469 		return (EINVAL);
470 	}
471 
472 	return (error);
473 }
474 
475 int
476 sys_semget(l, v, retval)
477 	struct lwp *l;
478 	void *v;
479 	register_t *retval;
480 {
481 	struct sys_semget_args /* {
482 		syscallarg(key_t) key;
483 		syscallarg(int) nsems;
484 		syscallarg(int) semflg;
485 	} */ *uap = v;
486 	int semid, eval;
487 	int key = SCARG(uap, key);
488 	int nsems = SCARG(uap, nsems);
489 	int semflg = SCARG(uap, semflg);
490 	struct ucred *cred = l->l_proc->p_ucred;
491 
492 	SEM_PRINTF(("semget(0x%x, %d, 0%o)\n", key, nsems, semflg));
493 
494 	if (key != IPC_PRIVATE) {
495 		for (semid = 0; semid < seminfo.semmni; semid++) {
496 			if ((sema[semid].sem_perm.mode & SEM_ALLOC) &&
497 			    sema[semid].sem_perm._key == key)
498 				break;
499 		}
500 		if (semid < seminfo.semmni) {
501 			SEM_PRINTF(("found public key\n"));
502 			if ((eval = ipcperm(cred, &sema[semid].sem_perm,
503 			    semflg & 0700)))
504 				return (eval);
505 			if (nsems > 0 && sema[semid].sem_nsems < nsems) {
506 				SEM_PRINTF(("too small\n"));
507 				return (EINVAL);
508 			}
509 			if ((semflg & IPC_CREAT) && (semflg & IPC_EXCL)) {
510 				SEM_PRINTF(("not exclusive\n"));
511 				return (EEXIST);
512 			}
513 			goto found;
514 		}
515 	}
516 
517 	SEM_PRINTF(("need to allocate the semid_ds\n"));
518 	if (key == IPC_PRIVATE || (semflg & IPC_CREAT)) {
519 		if (nsems <= 0 || nsems > seminfo.semmsl) {
520 			SEM_PRINTF(("nsems out of range (0<%d<=%d)\n", nsems,
521 			    seminfo.semmsl));
522 			return (EINVAL);
523 		}
524 		if (nsems > seminfo.semmns - semtot) {
525 			SEM_PRINTF(("not enough semaphores left "
526 			    "(need %d, got %d)\n",
527 			    nsems, seminfo.semmns - semtot));
528 			return (ENOSPC);
529 		}
530 		for (semid = 0; semid < seminfo.semmni; semid++) {
531 			if ((sema[semid].sem_perm.mode & SEM_ALLOC) == 0)
532 				break;
533 		}
534 		if (semid == seminfo.semmni) {
535 			SEM_PRINTF(("no more semid_ds's available\n"));
536 			return (ENOSPC);
537 		}
538 		SEM_PRINTF(("semid %d is available\n", semid));
539 		sema[semid].sem_perm._key = key;
540 		sema[semid].sem_perm.cuid = cred->cr_uid;
541 		sema[semid].sem_perm.uid = cred->cr_uid;
542 		sema[semid].sem_perm.cgid = cred->cr_gid;
543 		sema[semid].sem_perm.gid = cred->cr_gid;
544 		sema[semid].sem_perm.mode = (semflg & 0777) | SEM_ALLOC;
545 		sema[semid].sem_perm._seq =
546 		    (sema[semid].sem_perm._seq + 1) & 0x7fff;
547 		sema[semid].sem_nsems = nsems;
548 		sema[semid].sem_otime = 0;
549 		sema[semid].sem_ctime = time.tv_sec;
550 		sema[semid]._sem_base = &sem[semtot];
551 		semtot += nsems;
552 		memset(sema[semid]._sem_base, 0,
553 		    sizeof(sema[semid]._sem_base[0]) * nsems);
554 		SEM_PRINTF(("sembase = %p, next = %p\n", sema[semid]._sem_base,
555 		    &sem[semtot]));
556 	} else {
557 		SEM_PRINTF(("didn't find it and wasn't asked to create it\n"));
558 		return (ENOENT);
559 	}
560 
561 found:
562 	*retval = IXSEQ_TO_IPCID(semid, sema[semid].sem_perm);
563 	return (0);
564 }
565 
566 int
567 sys_semop(l, v, retval)
568 	struct lwp *l;
569 	void *v;
570 	register_t *retval;
571 {
572 	struct sys_semop_args /* {
573 		syscallarg(int) semid;
574 		syscallarg(struct sembuf *) sops;
575 		syscallarg(size_t) nsops;
576 	} */ *uap = v;
577 	struct proc *p = l->l_proc;
578 	int semid = SCARG(uap, semid), seq;
579 	size_t nsops = SCARG(uap, nsops);
580 	struct sembuf sops[MAX_SOPS];
581 	struct semid_ds *semaptr;
582 	struct sembuf *sopptr = NULL;
583 	struct __sem *semptr = NULL;
584 	struct sem_undo *suptr = NULL;
585 	struct ucred *cred = p->p_ucred;
586 	int i, j, eval;
587 	int do_wakeup, do_undos;
588 
589 	SEM_PRINTF(("call to semop(%d, %p, %lld)\n", semid, sops,
590 	    (long long)nsops));
591 
592 	semid = IPCID_TO_IX(semid);	/* Convert back to zero origin */
593 	if (semid < 0 || semid >= seminfo.semmni)
594 		return (EINVAL);
595 
596 	semaptr = &sema[semid];
597 	seq = IPCID_TO_SEQ(SCARG(uap, semid));
598 	if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0 ||
599 	    semaptr->sem_perm._seq != seq)
600 		return (EINVAL);
601 
602 	if ((eval = ipcperm(cred, &semaptr->sem_perm, IPC_W))) {
603 		SEM_PRINTF(("eval = %d from ipaccess\n", eval));
604 		return (eval);
605 	}
606 
607 	if (nsops > MAX_SOPS) {
608 		SEM_PRINTF(("too many sops (max=%d, nsops=%lld)\n", MAX_SOPS,
609 		    (long long)nsops));
610 		return (E2BIG);
611 	}
612 
613 	if ((eval = copyin(SCARG(uap, sops),
614 	    sops, nsops * sizeof(sops[0]))) != 0) {
615 		SEM_PRINTF(("eval = %d from copyin(%p, %p, %lld)\n", eval,
616 		    SCARG(uap, sops), &sops,
617 		    (long long)(nsops * sizeof(sops[0]))));
618 		return (eval);
619 	}
620 
621 	for (i = 0; i < nsops; i++)
622 		if (sops[i].sem_num >= semaptr->sem_nsems)
623 			return (EFBIG);
624 
625 	/*
626 	 * Loop trying to satisfy the vector of requests.
627 	 * If we reach a point where we must wait, any requests already
628 	 * performed are rolled back and we go to sleep until some other
629 	 * process wakes us up.  At this point, we start all over again.
630 	 *
631 	 * This ensures that from the perspective of other tasks, a set
632 	 * of requests is atomic (never partially satisfied).
633 	 */
634 	do_undos = 0;
635 
636 	for (;;) {
637 		do_wakeup = 0;
638 
639 		for (i = 0; i < nsops; i++) {
640 			sopptr = &sops[i];
641 			semptr = &semaptr->_sem_base[sopptr->sem_num];
642 
643 			SEM_PRINTF(("semop:  semaptr=%p, sem_base=%p, "
644 			    "semptr=%p, sem[%d]=%d : op=%d, flag=%s\n",
645 			    semaptr, semaptr->_sem_base, semptr,
646 			    sopptr->sem_num, semptr->semval, sopptr->sem_op,
647 			    (sopptr->sem_flg & IPC_NOWAIT) ?
648 			    "nowait" : "wait"));
649 
650 			if (sopptr->sem_op < 0) {
651 				if ((int)(semptr->semval +
652 				    sopptr->sem_op) < 0) {
653 					SEM_PRINTF(("semop:  "
654 					    "can't do it now\n"));
655 					break;
656 				} else {
657 					semptr->semval += sopptr->sem_op;
658 					if (semptr->semval == 0 &&
659 					    semptr->semzcnt > 0)
660 						do_wakeup = 1;
661 				}
662 				if (sopptr->sem_flg & SEM_UNDO)
663 					do_undos = 1;
664 			} else if (sopptr->sem_op == 0) {
665 				if (semptr->semval > 0) {
666 					SEM_PRINTF(("semop:  not zero now\n"));
667 					break;
668 				}
669 			} else {
670 				if (semptr->semncnt > 0)
671 					do_wakeup = 1;
672 				semptr->semval += sopptr->sem_op;
673 				if (sopptr->sem_flg & SEM_UNDO)
674 					do_undos = 1;
675 			}
676 		}
677 
678 		/*
679 		 * Did we get through the entire vector?
680 		 */
681 		if (i >= nsops)
682 			goto done;
683 
684 		/*
685 		 * No ... rollback anything that we've already done
686 		 */
687 		SEM_PRINTF(("semop:  rollback 0 through %d\n", i - 1));
688 		while (i-- > 0)
689 			semaptr->_sem_base[sops[i].sem_num].semval -=
690 			    sops[i].sem_op;
691 
692 		/*
693 		 * If the request that we couldn't satisfy has the
694 		 * NOWAIT flag set then return with EAGAIN.
695 		 */
696 		if (sopptr->sem_flg & IPC_NOWAIT)
697 			return (EAGAIN);
698 
699 		if (sopptr->sem_op == 0)
700 			semptr->semzcnt++;
701 		else
702 			semptr->semncnt++;
703 
704 		SEM_PRINTF(("semop:  good night!\n"));
705 		eval = tsleep((caddr_t)semaptr, (PZERO - 4) | PCATCH,
706 		    "semwait", 0);
707 		SEM_PRINTF(("semop:  good morning (eval=%d)!\n", eval));
708 
709 		/*
710 		 * Make sure that the semaphore still exists
711 		 */
712 		if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0 ||
713 		    semaptr->sem_perm._seq != seq) {
714 			/* The man page says to return EIDRM. */
715 			/* Unfortunately, BSD doesn't define that code! */
716 #ifdef EIDRM
717 			return (EIDRM);
718 #else
719 			return (EINVAL);
720 #endif
721 		}
722 
723 		/*
724 		 * The semaphore is still alive.  Readjust the count of
725 		 * waiting processes.
726 		 */
727 		semptr = &semaptr->_sem_base[sopptr->sem_num];
728 		if (sopptr->sem_op == 0)
729 			semptr->semzcnt--;
730 		else
731 			semptr->semncnt--;
732 		/*
733 		 * Is it really morning, or was our sleep interrupted?
734 		 * (Delayed check of tsleep() return code because we
735 		 * need to decrement sem[nz]cnt either way.)
736 		 */
737 		if (eval != 0)
738 			return (EINTR);
739 		SEM_PRINTF(("semop:  good morning!\n"));
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 			while (i-- > 0) {
774 				if ((sops[i].sem_flg & SEM_UNDO) == 0)
775 					continue;
776 				adjval = sops[i].sem_op;
777 				if (adjval == 0)
778 					continue;
779 				if (semundo_adjust(p, &suptr, semid,
780 				    sops[i].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
817  * adjustments to 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
830 	 * associated with 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) ; "
872 			    "sem=%d\n",
873 			    suptr->un_proc, suptr->un_ent[ix].un_id,
874 			    suptr->un_ent[ix].un_num,
875 			    suptr->un_ent[ix].un_adjval,
876 			    semaptr->_sem_base[semnum].semval));
877 
878 			if (adjval < 0 &&
879 			    semaptr->_sem_base[semnum].semval < -adjval)
880 				semaptr->_sem_base[semnum].semval = 0;
881 			else
882 				semaptr->_sem_base[semnum].semval += adjval;
883 
884 #ifdef SEM_WAKEUP
885 			sem_wakeup((caddr_t)semaptr);
886 #else
887 			wakeup((caddr_t)semaptr);
888 #endif
889 			SEM_PRINTF(("semexit:  back from wakeup\n"));
890 		}
891 	}
892 
893 	/*
894 	 * Deallocate the undo vector.
895 	 */
896 	SEM_PRINTF(("removing vector\n"));
897 	suptr->un_proc = NULL;
898 	*supptr = suptr->un_next;
899 }
900