xref: /freebsd/sys/kern/sysv_sem.c (revision 195ebc7e9e4b129de810833791a19dfb4349d6a9)
1 /*-
2  * Implementation of SVID semaphores
3  *
4  * Author:  Daniel Boulet
5  *
6  * This software is provided ``AS IS'' without any warranties of any kind.
7  */
8 /*-
9  * Copyright (c) 2003-2005 McAfee, Inc.
10  * All rights reserved.
11  *
12  * This software was developed for the FreeBSD Project in part by McAfee
13  * Research, the Security Research Division of McAfee, Inc under DARPA/SPAWAR
14  * contract N66001-01-C-8035 ("CBOSS"), as part of the DARPA CHATS research
15  * program.
16  *
17  * Redistribution and use in source and binary forms, with or without
18  * modification, are permitted provided that the following conditions
19  * are met:
20  * 1. Redistributions of source code must retain the above copyright
21  *    notice, this list of conditions and the following disclaimer.
22  * 2. Redistributions in binary form must reproduce the above copyright
23  *    notice, this list of conditions and the following disclaimer in the
24  *    documentation and/or other materials provided with the distribution.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  */
38 
39 #include <sys/cdefs.h>
40 __FBSDID("$FreeBSD$");
41 
42 #include "opt_sysvipc.h"
43 #include "opt_mac.h"
44 
45 #include <sys/param.h>
46 #include <sys/systm.h>
47 #include <sys/sysproto.h>
48 #include <sys/eventhandler.h>
49 #include <sys/kernel.h>
50 #include <sys/proc.h>
51 #include <sys/lock.h>
52 #include <sys/module.h>
53 #include <sys/mutex.h>
54 #include <sys/sem.h>
55 #include <sys/syscall.h>
56 #include <sys/syscallsubr.h>
57 #include <sys/sysent.h>
58 #include <sys/sysctl.h>
59 #include <sys/uio.h>
60 #include <sys/malloc.h>
61 #include <sys/jail.h>
62 
63 #include <security/mac/mac_framework.h>
64 
65 static MALLOC_DEFINE(M_SEM, "sem", "SVID compatible semaphores");
66 
67 #ifdef SEM_DEBUG
68 #define DPRINTF(a)	printf a
69 #else
70 #define DPRINTF(a)
71 #endif
72 
73 static void seminit(void);
74 static int sysvsem_modload(struct module *, int, void *);
75 static int semunload(void);
76 static void semexit_myhook(void *arg, struct proc *p);
77 static int sysctl_sema(SYSCTL_HANDLER_ARGS);
78 static int semvalid(int semid, struct semid_kernel *semakptr);
79 
80 #ifndef _SYS_SYSPROTO_H_
81 struct __semctl_args;
82 int __semctl(struct thread *td, struct __semctl_args *uap);
83 struct semget_args;
84 int semget(struct thread *td, struct semget_args *uap);
85 struct semop_args;
86 int semop(struct thread *td, struct semop_args *uap);
87 #endif
88 
89 static struct sem_undo *semu_alloc(struct thread *td);
90 static int semundo_adjust(struct thread *td, struct sem_undo **supptr,
91     int semid, int semseq, int semnum, int adjval);
92 static void semundo_clear(int semid, int semnum);
93 
94 /* XXX casting to (sy_call_t *) is bogus, as usual. */
95 static sy_call_t *semcalls[] = {
96 	(sy_call_t *)__semctl, (sy_call_t *)semget,
97 	(sy_call_t *)semop
98 };
99 
100 static struct mtx	sem_mtx;	/* semaphore global lock */
101 static struct mtx sem_undo_mtx;
102 static int	semtot = 0;
103 static struct semid_kernel *sema;	/* semaphore id pool */
104 static struct mtx *sema_mtx;	/* semaphore id pool mutexes*/
105 static struct sem *sem;		/* semaphore pool */
106 LIST_HEAD(, sem_undo) semu_list;	/* list of active undo structures */
107 LIST_HEAD(, sem_undo) semu_free_list;	/* list of free undo structures */
108 static int	*semu;		/* undo structure pool */
109 static eventhandler_tag semexit_tag;
110 
111 #define SEMUNDO_MTX		sem_undo_mtx
112 #define SEMUNDO_LOCK()		mtx_lock(&SEMUNDO_MTX);
113 #define SEMUNDO_UNLOCK()	mtx_unlock(&SEMUNDO_MTX);
114 #define SEMUNDO_LOCKASSERT(how)	mtx_assert(&SEMUNDO_MTX, (how));
115 
116 struct sem {
117 	u_short	semval;		/* semaphore value */
118 	pid_t	sempid;		/* pid of last operation */
119 	u_short	semncnt;	/* # awaiting semval > cval */
120 	u_short	semzcnt;	/* # awaiting semval = 0 */
121 };
122 
123 /*
124  * Undo structure (one per process)
125  */
126 struct sem_undo {
127 	LIST_ENTRY(sem_undo) un_next;	/* ptr to next active undo structure */
128 	struct	proc *un_proc;		/* owner of this structure */
129 	short	un_cnt;			/* # of active entries */
130 	struct undo {
131 		short	un_adjval;	/* adjust on exit values */
132 		short	un_num;		/* semaphore # */
133 		int	un_id;		/* semid */
134 		unsigned short un_seq;
135 	} un_ent[1];			/* undo entries */
136 };
137 
138 /*
139  * Configuration parameters
140  */
141 #ifndef SEMMNI
142 #define SEMMNI	10		/* # of semaphore identifiers */
143 #endif
144 #ifndef SEMMNS
145 #define SEMMNS	60		/* # of semaphores in system */
146 #endif
147 #ifndef SEMUME
148 #define SEMUME	10		/* max # of undo entries per process */
149 #endif
150 #ifndef SEMMNU
151 #define SEMMNU	30		/* # of undo structures in system */
152 #endif
153 
154 /* shouldn't need tuning */
155 #ifndef SEMMAP
156 #define SEMMAP	30		/* # of entries in semaphore map */
157 #endif
158 #ifndef SEMMSL
159 #define SEMMSL	SEMMNS		/* max # of semaphores per id */
160 #endif
161 #ifndef SEMOPM
162 #define SEMOPM	100		/* max # of operations per semop call */
163 #endif
164 
165 #define SEMVMX	32767		/* semaphore maximum value */
166 #define SEMAEM	16384		/* adjust on exit max value */
167 
168 /*
169  * Due to the way semaphore memory is allocated, we have to ensure that
170  * SEMUSZ is properly aligned.
171  */
172 
173 #define SEM_ALIGN(bytes) (((bytes) + (sizeof(long) - 1)) & ~(sizeof(long) - 1))
174 
175 /* actual size of an undo structure */
176 #define SEMUSZ	SEM_ALIGN(offsetof(struct sem_undo, un_ent[SEMUME]))
177 
178 /*
179  * Macro to find a particular sem_undo vector
180  */
181 #define SEMU(ix) \
182 	((struct sem_undo *)(((intptr_t)semu)+ix * seminfo.semusz))
183 
184 /*
185  * semaphore info struct
186  */
187 struct seminfo seminfo = {
188                 SEMMAP,         /* # of entries in semaphore map */
189                 SEMMNI,         /* # of semaphore identifiers */
190                 SEMMNS,         /* # of semaphores in system */
191                 SEMMNU,         /* # of undo structures in system */
192                 SEMMSL,         /* max # of semaphores per id */
193                 SEMOPM,         /* max # of operations per semop call */
194                 SEMUME,         /* max # of undo entries per process */
195                 SEMUSZ,         /* size in bytes of undo structure */
196                 SEMVMX,         /* semaphore maximum value */
197                 SEMAEM          /* adjust on exit max value */
198 };
199 
200 SYSCTL_INT(_kern_ipc, OID_AUTO, semmap, CTLFLAG_RW, &seminfo.semmap, 0,
201     "Number of entries in the semaphore map");
202 SYSCTL_INT(_kern_ipc, OID_AUTO, semmni, CTLFLAG_RDTUN, &seminfo.semmni, 0,
203     "Number of semaphore identifiers");
204 SYSCTL_INT(_kern_ipc, OID_AUTO, semmns, CTLFLAG_RDTUN, &seminfo.semmns, 0,
205     "Maximum number of semaphores in the system");
206 SYSCTL_INT(_kern_ipc, OID_AUTO, semmnu, CTLFLAG_RDTUN, &seminfo.semmnu, 0,
207     "Maximum number of undo structures in the system");
208 SYSCTL_INT(_kern_ipc, OID_AUTO, semmsl, CTLFLAG_RW, &seminfo.semmsl, 0,
209     "Max semaphores per id");
210 SYSCTL_INT(_kern_ipc, OID_AUTO, semopm, CTLFLAG_RDTUN, &seminfo.semopm, 0,
211     "Max operations per semop call");
212 SYSCTL_INT(_kern_ipc, OID_AUTO, semume, CTLFLAG_RDTUN, &seminfo.semume, 0,
213     "Max undo entries per process");
214 SYSCTL_INT(_kern_ipc, OID_AUTO, semusz, CTLFLAG_RDTUN, &seminfo.semusz, 0,
215     "Size in bytes of undo structure");
216 SYSCTL_INT(_kern_ipc, OID_AUTO, semvmx, CTLFLAG_RW, &seminfo.semvmx, 0,
217     "Semaphore maximum value");
218 SYSCTL_INT(_kern_ipc, OID_AUTO, semaem, CTLFLAG_RW, &seminfo.semaem, 0,
219     "Adjust on exit max value");
220 SYSCTL_PROC(_kern_ipc, OID_AUTO, sema, CTLFLAG_RD,
221     NULL, 0, sysctl_sema, "", "");
222 
223 static void
224 seminit(void)
225 {
226 	int i;
227 
228 	TUNABLE_INT_FETCH("kern.ipc.semmap", &seminfo.semmap);
229 	TUNABLE_INT_FETCH("kern.ipc.semmni", &seminfo.semmni);
230 	TUNABLE_INT_FETCH("kern.ipc.semmns", &seminfo.semmns);
231 	TUNABLE_INT_FETCH("kern.ipc.semmnu", &seminfo.semmnu);
232 	TUNABLE_INT_FETCH("kern.ipc.semmsl", &seminfo.semmsl);
233 	TUNABLE_INT_FETCH("kern.ipc.semopm", &seminfo.semopm);
234 	TUNABLE_INT_FETCH("kern.ipc.semume", &seminfo.semume);
235 	TUNABLE_INT_FETCH("kern.ipc.semusz", &seminfo.semusz);
236 	TUNABLE_INT_FETCH("kern.ipc.semvmx", &seminfo.semvmx);
237 	TUNABLE_INT_FETCH("kern.ipc.semaem", &seminfo.semaem);
238 
239 	sem = malloc(sizeof(struct sem) * seminfo.semmns, M_SEM, M_WAITOK);
240 	sema = malloc(sizeof(struct semid_kernel) * seminfo.semmni, M_SEM,
241 	    M_WAITOK);
242 	sema_mtx = malloc(sizeof(struct mtx) * seminfo.semmni, M_SEM,
243 	    M_WAITOK | M_ZERO);
244 	semu = malloc(seminfo.semmnu * seminfo.semusz, M_SEM, M_WAITOK);
245 
246 	for (i = 0; i < seminfo.semmni; i++) {
247 		sema[i].u.sem_base = 0;
248 		sema[i].u.sem_perm.mode = 0;
249 		sema[i].u.sem_perm.seq = 0;
250 #ifdef MAC
251 		mac_sysvsem_init(&sema[i]);
252 #endif
253 	}
254 	for (i = 0; i < seminfo.semmni; i++)
255 		mtx_init(&sema_mtx[i], "semid", NULL, MTX_DEF);
256 	LIST_INIT(&semu_free_list);
257 	for (i = 0; i < seminfo.semmnu; i++) {
258 		struct sem_undo *suptr = SEMU(i);
259 		suptr->un_proc = NULL;
260 		LIST_INSERT_HEAD(&semu_free_list, suptr, un_next);
261 	}
262 	LIST_INIT(&semu_list);
263 	mtx_init(&sem_mtx, "sem", NULL, MTX_DEF);
264 	mtx_init(&sem_undo_mtx, "semu", NULL, MTX_DEF);
265 	semexit_tag = EVENTHANDLER_REGISTER(process_exit, semexit_myhook, NULL,
266 	    EVENTHANDLER_PRI_ANY);
267 }
268 
269 static int
270 semunload(void)
271 {
272 	int i;
273 
274 	/* XXXKIB */
275 	if (semtot != 0)
276 		return (EBUSY);
277 
278 	EVENTHANDLER_DEREGISTER(process_exit, semexit_tag);
279 #ifdef MAC
280 	for (i = 0; i < seminfo.semmni; i++)
281 		mac_sysvsem_destroy(&sema[i]);
282 #endif
283 	free(sem, M_SEM);
284 	free(sema, M_SEM);
285 	free(semu, M_SEM);
286 	for (i = 0; i < seminfo.semmni; i++)
287 		mtx_destroy(&sema_mtx[i]);
288 	free(sema_mtx, M_SEM);
289 	mtx_destroy(&sem_mtx);
290 	mtx_destroy(&sem_undo_mtx);
291 	return (0);
292 }
293 
294 static int
295 sysvsem_modload(struct module *module, int cmd, void *arg)
296 {
297 	int error = 0;
298 
299 	switch (cmd) {
300 	case MOD_LOAD:
301 		seminit();
302 		break;
303 	case MOD_UNLOAD:
304 		error = semunload();
305 		break;
306 	case MOD_SHUTDOWN:
307 		break;
308 	default:
309 		error = EINVAL;
310 		break;
311 	}
312 	return (error);
313 }
314 
315 static moduledata_t sysvsem_mod = {
316 	"sysvsem",
317 	&sysvsem_modload,
318 	NULL
319 };
320 
321 SYSCALL_MODULE_HELPER(semsys);
322 SYSCALL_MODULE_HELPER(__semctl);
323 SYSCALL_MODULE_HELPER(semget);
324 SYSCALL_MODULE_HELPER(semop);
325 
326 DECLARE_MODULE(sysvsem, sysvsem_mod,
327 	SI_SUB_SYSV_SEM, SI_ORDER_FIRST);
328 MODULE_VERSION(sysvsem, 1);
329 
330 /*
331  * Entry point for all SEM calls.
332  */
333 int
334 semsys(td, uap)
335 	struct thread *td;
336 	/* XXX actually varargs. */
337 	struct semsys_args /* {
338 		int	which;
339 		int	a2;
340 		int	a3;
341 		int	a4;
342 		int	a5;
343 	} */ *uap;
344 {
345 	int error;
346 
347 	if (!prison_allow(td->td_ucred, PR_ALLOW_SYSVIPC))
348 		return (ENOSYS);
349 	if (uap->which < 0 ||
350 	    uap->which >= sizeof(semcalls)/sizeof(semcalls[0]))
351 		return (EINVAL);
352 	error = (*semcalls[uap->which])(td, &uap->a2);
353 	return (error);
354 }
355 
356 /*
357  * Allocate a new sem_undo structure for a process
358  * (returns ptr to structure or NULL if no more room)
359  */
360 
361 static struct sem_undo *
362 semu_alloc(struct thread *td)
363 {
364 	struct sem_undo *suptr;
365 
366 	SEMUNDO_LOCKASSERT(MA_OWNED);
367 	if ((suptr = LIST_FIRST(&semu_free_list)) == NULL)
368 		return (NULL);
369 	LIST_REMOVE(suptr, un_next);
370 	LIST_INSERT_HEAD(&semu_list, suptr, un_next);
371 	suptr->un_cnt = 0;
372 	suptr->un_proc = td->td_proc;
373 	return (suptr);
374 }
375 
376 static int
377 semu_try_free(struct sem_undo *suptr)
378 {
379 
380 	SEMUNDO_LOCKASSERT(MA_OWNED);
381 
382 	if (suptr->un_cnt != 0)
383 		return (0);
384 	LIST_REMOVE(suptr, un_next);
385 	LIST_INSERT_HEAD(&semu_free_list, suptr, un_next);
386 	return (1);
387 }
388 
389 /*
390  * Adjust a particular entry for a particular proc
391  */
392 
393 static int
394 semundo_adjust(struct thread *td, struct sem_undo **supptr, int semid,
395     int semseq, int semnum, int adjval)
396 {
397 	struct proc *p = td->td_proc;
398 	struct sem_undo *suptr;
399 	struct undo *sunptr;
400 	int i;
401 
402 	SEMUNDO_LOCKASSERT(MA_OWNED);
403 	/* Look for and remember the sem_undo if the caller doesn't provide
404 	   it */
405 
406 	suptr = *supptr;
407 	if (suptr == NULL) {
408 		LIST_FOREACH(suptr, &semu_list, un_next) {
409 			if (suptr->un_proc == p) {
410 				*supptr = suptr;
411 				break;
412 			}
413 		}
414 		if (suptr == NULL) {
415 			if (adjval == 0)
416 				return(0);
417 			suptr = semu_alloc(td);
418 			if (suptr == NULL)
419 				return (ENOSPC);
420 			*supptr = suptr;
421 		}
422 	}
423 
424 	/*
425 	 * Look for the requested entry and adjust it (delete if adjval becomes
426 	 * 0).
427 	 */
428 	sunptr = &suptr->un_ent[0];
429 	for (i = 0; i < suptr->un_cnt; i++, sunptr++) {
430 		if (sunptr->un_id != semid || sunptr->un_num != semnum)
431 			continue;
432 		if (adjval != 0) {
433 			adjval += sunptr->un_adjval;
434 			if (adjval > seminfo.semaem || adjval < -seminfo.semaem)
435 				return (ERANGE);
436 		}
437 		sunptr->un_adjval = adjval;
438 		if (sunptr->un_adjval == 0) {
439 			suptr->un_cnt--;
440 			if (i < suptr->un_cnt)
441 				suptr->un_ent[i] =
442 				    suptr->un_ent[suptr->un_cnt];
443 			if (suptr->un_cnt == 0)
444 				semu_try_free(suptr);
445 		}
446 		return (0);
447 	}
448 
449 	/* Didn't find the right entry - create it */
450 	if (adjval == 0)
451 		return (0);
452 	if (adjval > seminfo.semaem || adjval < -seminfo.semaem)
453 		return (ERANGE);
454 	if (suptr->un_cnt != seminfo.semume) {
455 		sunptr = &suptr->un_ent[suptr->un_cnt];
456 		suptr->un_cnt++;
457 		sunptr->un_adjval = adjval;
458 		sunptr->un_id = semid;
459 		sunptr->un_num = semnum;
460 		sunptr->un_seq = semseq;
461 	} else
462 		return (EINVAL);
463 	return (0);
464 }
465 
466 static void
467 semundo_clear(int semid, int semnum)
468 {
469 	struct sem_undo *suptr, *suptr1;
470 	struct undo *sunptr;
471 	int i;
472 
473 	SEMUNDO_LOCKASSERT(MA_OWNED);
474 	LIST_FOREACH_SAFE(suptr, &semu_list, un_next, suptr1) {
475 		sunptr = &suptr->un_ent[0];
476 		for (i = 0; i < suptr->un_cnt; i++, sunptr++) {
477 			if (sunptr->un_id != semid)
478 				continue;
479 			if (semnum == -1 || sunptr->un_num == semnum) {
480 				suptr->un_cnt--;
481 				if (i < suptr->un_cnt) {
482 					suptr->un_ent[i] =
483 					    suptr->un_ent[suptr->un_cnt];
484 					continue;
485 				}
486 				semu_try_free(suptr);
487 			}
488 			if (semnum != -1)
489 				break;
490 		}
491 	}
492 }
493 
494 static int
495 semvalid(int semid, struct semid_kernel *semakptr)
496 {
497 
498 	return ((semakptr->u.sem_perm.mode & SEM_ALLOC) == 0 ||
499 	    semakptr->u.sem_perm.seq != IPCID_TO_SEQ(semid) ? EINVAL : 0);
500 }
501 
502 /*
503  * Note that the user-mode half of this passes a union, not a pointer.
504  */
505 #ifndef _SYS_SYSPROTO_H_
506 struct __semctl_args {
507 	int	semid;
508 	int	semnum;
509 	int	cmd;
510 	union	semun *arg;
511 };
512 #endif
513 int
514 __semctl(struct thread *td, struct __semctl_args *uap)
515 {
516 	struct semid_ds dsbuf;
517 	union semun arg, semun;
518 	register_t rval;
519 	int error;
520 
521 	switch (uap->cmd) {
522 	case SEM_STAT:
523 	case IPC_SET:
524 	case IPC_STAT:
525 	case GETALL:
526 	case SETVAL:
527 	case SETALL:
528 		error = copyin(uap->arg, &arg, sizeof(arg));
529 		if (error)
530 			return (error);
531 		break;
532 	}
533 
534 	switch (uap->cmd) {
535 	case SEM_STAT:
536 	case IPC_STAT:
537 		semun.buf = &dsbuf;
538 		break;
539 	case IPC_SET:
540 		error = copyin(arg.buf, &dsbuf, sizeof(dsbuf));
541 		if (error)
542 			return (error);
543 		semun.buf = &dsbuf;
544 		break;
545 	case GETALL:
546 	case SETALL:
547 		semun.array = arg.array;
548 		break;
549 	case SETVAL:
550 		semun.val = arg.val;
551 		break;
552 	}
553 
554 	error = kern_semctl(td, uap->semid, uap->semnum, uap->cmd, &semun,
555 	    &rval);
556 	if (error)
557 		return (error);
558 
559 	switch (uap->cmd) {
560 	case SEM_STAT:
561 	case IPC_STAT:
562 		error = copyout(&dsbuf, arg.buf, sizeof(dsbuf));
563 		break;
564 	}
565 
566 	if (error == 0)
567 		td->td_retval[0] = rval;
568 	return (error);
569 }
570 
571 int
572 kern_semctl(struct thread *td, int semid, int semnum, int cmd,
573     union semun *arg, register_t *rval)
574 {
575 	u_short *array;
576 	struct ucred *cred = td->td_ucred;
577 	int i, error;
578 	struct semid_ds *sbuf;
579 	struct semid_kernel *semakptr;
580 	struct mtx *sema_mtxp;
581 	u_short usval, count;
582 	int semidx;
583 
584 	DPRINTF(("call to semctl(%d, %d, %d, 0x%p)\n",
585 	    semid, semnum, cmd, arg));
586 	if (!prison_allow(td->td_ucred, PR_ALLOW_SYSVIPC))
587 		return (ENOSYS);
588 
589 	array = NULL;
590 
591 	switch(cmd) {
592 	case SEM_STAT:
593 		/*
594 		 * For this command we assume semid is an array index
595 		 * rather than an IPC id.
596 		 */
597 		if (semid < 0 || semid >= seminfo.semmni)
598 			return (EINVAL);
599 		semakptr = &sema[semid];
600 		sema_mtxp = &sema_mtx[semid];
601 		mtx_lock(sema_mtxp);
602 		if ((semakptr->u.sem_perm.mode & SEM_ALLOC) == 0) {
603 			error = EINVAL;
604 			goto done2;
605 		}
606 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
607 			goto done2;
608 #ifdef MAC
609 		error = mac_sysvsem_check_semctl(cred, semakptr, cmd);
610 		if (error != 0)
611 			goto done2;
612 #endif
613 		bcopy(&semakptr->u, arg->buf, sizeof(struct semid_ds));
614 		*rval = IXSEQ_TO_IPCID(semid, semakptr->u.sem_perm);
615 		mtx_unlock(sema_mtxp);
616 		return (0);
617 	}
618 
619 	semidx = IPCID_TO_IX(semid);
620 	if (semidx < 0 || semidx >= seminfo.semmni)
621 		return (EINVAL);
622 
623 	semakptr = &sema[semidx];
624 	sema_mtxp = &sema_mtx[semidx];
625 	if (cmd == IPC_RMID)
626 		mtx_lock(&sem_mtx);
627 	mtx_lock(sema_mtxp);
628 #ifdef MAC
629 	error = mac_sysvsem_check_semctl(cred, semakptr, cmd);
630 	if (error != 0)
631 		goto done2;
632 #endif
633 
634 	error = 0;
635 	*rval = 0;
636 
637 	switch (cmd) {
638 	case IPC_RMID:
639 		if ((error = semvalid(semid, semakptr)) != 0)
640 			goto done2;
641 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_M)))
642 			goto done2;
643 		semakptr->u.sem_perm.cuid = cred->cr_uid;
644 		semakptr->u.sem_perm.uid = cred->cr_uid;
645 		semakptr->u.sem_perm.mode = 0;
646 		SEMUNDO_LOCK();
647 		semundo_clear(semidx, -1);
648 		SEMUNDO_UNLOCK();
649 #ifdef MAC
650 		mac_sysvsem_cleanup(semakptr);
651 #endif
652 		wakeup(semakptr);
653 		for (i = 0; i < seminfo.semmni; i++) {
654 			if ((sema[i].u.sem_perm.mode & SEM_ALLOC) &&
655 			    sema[i].u.sem_base > semakptr->u.sem_base)
656 				mtx_lock_flags(&sema_mtx[i], LOP_DUPOK);
657 		}
658 		for (i = semakptr->u.sem_base - sem; i < semtot; i++)
659 			sem[i] = sem[i + semakptr->u.sem_nsems];
660 		for (i = 0; i < seminfo.semmni; i++) {
661 			if ((sema[i].u.sem_perm.mode & SEM_ALLOC) &&
662 			    sema[i].u.sem_base > semakptr->u.sem_base) {
663 				sema[i].u.sem_base -= semakptr->u.sem_nsems;
664 				mtx_unlock(&sema_mtx[i]);
665 			}
666 		}
667 		semtot -= semakptr->u.sem_nsems;
668 		break;
669 
670 	case IPC_SET:
671 		if ((error = semvalid(semid, semakptr)) != 0)
672 			goto done2;
673 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_M)))
674 			goto done2;
675 		sbuf = arg->buf;
676 		semakptr->u.sem_perm.uid = sbuf->sem_perm.uid;
677 		semakptr->u.sem_perm.gid = sbuf->sem_perm.gid;
678 		semakptr->u.sem_perm.mode = (semakptr->u.sem_perm.mode &
679 		    ~0777) | (sbuf->sem_perm.mode & 0777);
680 		semakptr->u.sem_ctime = time_second;
681 		break;
682 
683 	case IPC_STAT:
684 		if ((error = semvalid(semid, semakptr)) != 0)
685 			goto done2;
686 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
687 			goto done2;
688 		bcopy(&semakptr->u, arg->buf, sizeof(struct semid_ds));
689 		break;
690 
691 	case GETNCNT:
692 		if ((error = semvalid(semid, semakptr)) != 0)
693 			goto done2;
694 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
695 			goto done2;
696 		if (semnum < 0 || semnum >= semakptr->u.sem_nsems) {
697 			error = EINVAL;
698 			goto done2;
699 		}
700 		*rval = semakptr->u.sem_base[semnum].semncnt;
701 		break;
702 
703 	case GETPID:
704 		if ((error = semvalid(semid, semakptr)) != 0)
705 			goto done2;
706 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
707 			goto done2;
708 		if (semnum < 0 || semnum >= semakptr->u.sem_nsems) {
709 			error = EINVAL;
710 			goto done2;
711 		}
712 		*rval = semakptr->u.sem_base[semnum].sempid;
713 		break;
714 
715 	case GETVAL:
716 		if ((error = semvalid(semid, semakptr)) != 0)
717 			goto done2;
718 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
719 			goto done2;
720 		if (semnum < 0 || semnum >= semakptr->u.sem_nsems) {
721 			error = EINVAL;
722 			goto done2;
723 		}
724 		*rval = semakptr->u.sem_base[semnum].semval;
725 		break;
726 
727 	case GETALL:
728 		/*
729 		 * Unfortunately, callers of this function don't know
730 		 * in advance how many semaphores are in this set.
731 		 * While we could just allocate the maximum size array
732 		 * and pass the actual size back to the caller, that
733 		 * won't work for SETALL since we can't copyin() more
734 		 * data than the user specified as we may return a
735 		 * spurious EFAULT.
736 		 *
737 		 * Note that the number of semaphores in a set is
738 		 * fixed for the life of that set.  The only way that
739 		 * the 'count' could change while are blocked in
740 		 * malloc() is if this semaphore set were destroyed
741 		 * and a new one created with the same index.
742 		 * However, semvalid() will catch that due to the
743 		 * sequence number unless exactly 0x8000 (or a
744 		 * multiple thereof) semaphore sets for the same index
745 		 * are created and destroyed while we are in malloc!
746 		 *
747 		 */
748 		count = semakptr->u.sem_nsems;
749 		mtx_unlock(sema_mtxp);
750 		array = malloc(sizeof(*array) * count, M_TEMP, M_WAITOK);
751 		mtx_lock(sema_mtxp);
752 		if ((error = semvalid(semid, semakptr)) != 0)
753 			goto done2;
754 		KASSERT(count == semakptr->u.sem_nsems, ("nsems changed"));
755 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
756 			goto done2;
757 		for (i = 0; i < semakptr->u.sem_nsems; i++)
758 			array[i] = semakptr->u.sem_base[i].semval;
759 		mtx_unlock(sema_mtxp);
760 		error = copyout(array, arg->array, count * sizeof(*array));
761 		mtx_lock(sema_mtxp);
762 		break;
763 
764 	case GETZCNT:
765 		if ((error = semvalid(semid, semakptr)) != 0)
766 			goto done2;
767 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
768 			goto done2;
769 		if (semnum < 0 || semnum >= semakptr->u.sem_nsems) {
770 			error = EINVAL;
771 			goto done2;
772 		}
773 		*rval = semakptr->u.sem_base[semnum].semzcnt;
774 		break;
775 
776 	case SETVAL:
777 		if ((error = semvalid(semid, semakptr)) != 0)
778 			goto done2;
779 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_W)))
780 			goto done2;
781 		if (semnum < 0 || semnum >= semakptr->u.sem_nsems) {
782 			error = EINVAL;
783 			goto done2;
784 		}
785 		if (arg->val < 0 || arg->val > seminfo.semvmx) {
786 			error = ERANGE;
787 			goto done2;
788 		}
789 		semakptr->u.sem_base[semnum].semval = arg->val;
790 		SEMUNDO_LOCK();
791 		semundo_clear(semidx, semnum);
792 		SEMUNDO_UNLOCK();
793 		wakeup(semakptr);
794 		break;
795 
796 	case SETALL:
797 		/*
798 		 * See comment on GETALL for why 'count' shouldn't change
799 		 * and why we require a userland buffer.
800 		 */
801 		count = semakptr->u.sem_nsems;
802 		mtx_unlock(sema_mtxp);
803 		array = malloc(sizeof(*array) * count, M_TEMP, M_WAITOK);
804 		error = copyin(arg->array, array, count * sizeof(*array));
805 		mtx_lock(sema_mtxp);
806 		if (error)
807 			break;
808 		if ((error = semvalid(semid, semakptr)) != 0)
809 			goto done2;
810 		KASSERT(count == semakptr->u.sem_nsems, ("nsems changed"));
811 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_W)))
812 			goto done2;
813 		for (i = 0; i < semakptr->u.sem_nsems; i++) {
814 			usval = array[i];
815 			if (usval > seminfo.semvmx) {
816 				error = ERANGE;
817 				break;
818 			}
819 			semakptr->u.sem_base[i].semval = usval;
820 		}
821 		SEMUNDO_LOCK();
822 		semundo_clear(semidx, -1);
823 		SEMUNDO_UNLOCK();
824 		wakeup(semakptr);
825 		break;
826 
827 	default:
828 		error = EINVAL;
829 		break;
830 	}
831 
832 done2:
833 	mtx_unlock(sema_mtxp);
834 	if (cmd == IPC_RMID)
835 		mtx_unlock(&sem_mtx);
836 	if (array != NULL)
837 		free(array, M_TEMP);
838 	return(error);
839 }
840 
841 #ifndef _SYS_SYSPROTO_H_
842 struct semget_args {
843 	key_t	key;
844 	int	nsems;
845 	int	semflg;
846 };
847 #endif
848 int
849 semget(struct thread *td, struct semget_args *uap)
850 {
851 	int semid, error = 0;
852 	int key = uap->key;
853 	int nsems = uap->nsems;
854 	int semflg = uap->semflg;
855 	struct ucred *cred = td->td_ucred;
856 
857 	DPRINTF(("semget(0x%x, %d, 0%o)\n", key, nsems, semflg));
858 	if (!prison_allow(td->td_ucred, PR_ALLOW_SYSVIPC))
859 		return (ENOSYS);
860 
861 	mtx_lock(&sem_mtx);
862 	if (key != IPC_PRIVATE) {
863 		for (semid = 0; semid < seminfo.semmni; semid++) {
864 			if ((sema[semid].u.sem_perm.mode & SEM_ALLOC) &&
865 			    sema[semid].u.sem_perm.key == key)
866 				break;
867 		}
868 		if (semid < seminfo.semmni) {
869 			DPRINTF(("found public key\n"));
870 			if ((error = ipcperm(td, &sema[semid].u.sem_perm,
871 			    semflg & 0700))) {
872 				goto done2;
873 			}
874 			if (nsems > 0 && sema[semid].u.sem_nsems < nsems) {
875 				DPRINTF(("too small\n"));
876 				error = EINVAL;
877 				goto done2;
878 			}
879 			if ((semflg & IPC_CREAT) && (semflg & IPC_EXCL)) {
880 				DPRINTF(("not exclusive\n"));
881 				error = EEXIST;
882 				goto done2;
883 			}
884 #ifdef MAC
885 			error = mac_sysvsem_check_semget(cred, &sema[semid]);
886 			if (error != 0)
887 				goto done2;
888 #endif
889 			goto found;
890 		}
891 	}
892 
893 	DPRINTF(("need to allocate the semid_kernel\n"));
894 	if (key == IPC_PRIVATE || (semflg & IPC_CREAT)) {
895 		if (nsems <= 0 || nsems > seminfo.semmsl) {
896 			DPRINTF(("nsems out of range (0<%d<=%d)\n", nsems,
897 			    seminfo.semmsl));
898 			error = EINVAL;
899 			goto done2;
900 		}
901 		if (nsems > seminfo.semmns - semtot) {
902 			DPRINTF((
903 			    "not enough semaphores left (need %d, got %d)\n",
904 			    nsems, seminfo.semmns - semtot));
905 			error = ENOSPC;
906 			goto done2;
907 		}
908 		for (semid = 0; semid < seminfo.semmni; semid++) {
909 			if ((sema[semid].u.sem_perm.mode & SEM_ALLOC) == 0)
910 				break;
911 		}
912 		if (semid == seminfo.semmni) {
913 			DPRINTF(("no more semid_kernel's available\n"));
914 			error = ENOSPC;
915 			goto done2;
916 		}
917 		DPRINTF(("semid %d is available\n", semid));
918 		mtx_lock(&sema_mtx[semid]);
919 		KASSERT((sema[semid].u.sem_perm.mode & SEM_ALLOC) == 0,
920 		    ("Lost semaphore %d", semid));
921 		sema[semid].u.sem_perm.key = key;
922 		sema[semid].u.sem_perm.cuid = cred->cr_uid;
923 		sema[semid].u.sem_perm.uid = cred->cr_uid;
924 		sema[semid].u.sem_perm.cgid = cred->cr_gid;
925 		sema[semid].u.sem_perm.gid = cred->cr_gid;
926 		sema[semid].u.sem_perm.mode = (semflg & 0777) | SEM_ALLOC;
927 		sema[semid].u.sem_perm.seq =
928 		    (sema[semid].u.sem_perm.seq + 1) & 0x7fff;
929 		sema[semid].u.sem_nsems = nsems;
930 		sema[semid].u.sem_otime = 0;
931 		sema[semid].u.sem_ctime = time_second;
932 		sema[semid].u.sem_base = &sem[semtot];
933 		semtot += nsems;
934 		bzero(sema[semid].u.sem_base,
935 		    sizeof(sema[semid].u.sem_base[0])*nsems);
936 #ifdef MAC
937 		mac_sysvsem_create(cred, &sema[semid]);
938 #endif
939 		mtx_unlock(&sema_mtx[semid]);
940 		DPRINTF(("sembase = %p, next = %p\n",
941 		    sema[semid].u.sem_base, &sem[semtot]));
942 	} else {
943 		DPRINTF(("didn't find it and wasn't asked to create it\n"));
944 		error = ENOENT;
945 		goto done2;
946 	}
947 
948 found:
949 	td->td_retval[0] = IXSEQ_TO_IPCID(semid, sema[semid].u.sem_perm);
950 done2:
951 	mtx_unlock(&sem_mtx);
952 	return (error);
953 }
954 
955 #ifndef _SYS_SYSPROTO_H_
956 struct semop_args {
957 	int	semid;
958 	struct	sembuf *sops;
959 	size_t	nsops;
960 };
961 #endif
962 int
963 semop(struct thread *td, struct semop_args *uap)
964 {
965 #define SMALL_SOPS	8
966 	struct sembuf small_sops[SMALL_SOPS];
967 	int semid = uap->semid;
968 	size_t nsops = uap->nsops;
969 	struct sembuf *sops;
970 	struct semid_kernel *semakptr;
971 	struct sembuf *sopptr = 0;
972 	struct sem *semptr = 0;
973 	struct sem_undo *suptr;
974 	struct mtx *sema_mtxp;
975 	size_t i, j, k;
976 	int error;
977 	int do_wakeup, do_undos;
978 	unsigned short seq;
979 
980 #ifdef SEM_DEBUG
981 	sops = NULL;
982 #endif
983 	DPRINTF(("call to semop(%d, %p, %u)\n", semid, sops, nsops));
984 
985 	if (!prison_allow(td->td_ucred, PR_ALLOW_SYSVIPC))
986 		return (ENOSYS);
987 
988 	semid = IPCID_TO_IX(semid);	/* Convert back to zero origin */
989 
990 	if (semid < 0 || semid >= seminfo.semmni)
991 		return (EINVAL);
992 
993 	/* Allocate memory for sem_ops */
994 	if (nsops <= SMALL_SOPS)
995 		sops = small_sops;
996 	else if (nsops <= seminfo.semopm)
997 		sops = malloc(nsops * sizeof(*sops), M_TEMP, M_WAITOK);
998 	else {
999 		DPRINTF(("too many sops (max=%d, nsops=%d)\n", seminfo.semopm,
1000 		    nsops));
1001 		return (E2BIG);
1002 	}
1003 	if ((error = copyin(uap->sops, sops, nsops * sizeof(sops[0]))) != 0) {
1004 		DPRINTF(("error = %d from copyin(%p, %p, %d)\n", error,
1005 		    uap->sops, sops, nsops * sizeof(sops[0])));
1006 		if (sops != small_sops)
1007 			free(sops, M_SEM);
1008 		return (error);
1009 	}
1010 
1011 	semakptr = &sema[semid];
1012 	sema_mtxp = &sema_mtx[semid];
1013 	mtx_lock(sema_mtxp);
1014 	if ((semakptr->u.sem_perm.mode & SEM_ALLOC) == 0) {
1015 		error = EINVAL;
1016 		goto done2;
1017 	}
1018 	seq = semakptr->u.sem_perm.seq;
1019 	if (seq != IPCID_TO_SEQ(uap->semid)) {
1020 		error = EINVAL;
1021 		goto done2;
1022 	}
1023 	/*
1024 	 * Initial pass thru sops to see what permissions are needed.
1025 	 * Also perform any checks that don't need repeating on each
1026 	 * attempt to satisfy the request vector.
1027 	 */
1028 	j = 0;		/* permission needed */
1029 	do_undos = 0;
1030 	for (i = 0; i < nsops; i++) {
1031 		sopptr = &sops[i];
1032 		if (sopptr->sem_num >= semakptr->u.sem_nsems) {
1033 			error = EFBIG;
1034 			goto done2;
1035 		}
1036 		if (sopptr->sem_flg & SEM_UNDO && sopptr->sem_op != 0)
1037 			do_undos = 1;
1038 		j |= (sopptr->sem_op == 0) ? SEM_R : SEM_A;
1039 	}
1040 
1041 	if ((error = ipcperm(td, &semakptr->u.sem_perm, j))) {
1042 		DPRINTF(("error = %d from ipaccess\n", error));
1043 		goto done2;
1044 	}
1045 #ifdef MAC
1046 	error = mac_sysvsem_check_semop(td->td_ucred, semakptr, j);
1047 	if (error != 0)
1048 		goto done2;
1049 #endif
1050 
1051 	/*
1052 	 * Loop trying to satisfy the vector of requests.
1053 	 * If we reach a point where we must wait, any requests already
1054 	 * performed are rolled back and we go to sleep until some other
1055 	 * process wakes us up.  At this point, we start all over again.
1056 	 *
1057 	 * This ensures that from the perspective of other tasks, a set
1058 	 * of requests is atomic (never partially satisfied).
1059 	 */
1060 	for (;;) {
1061 		do_wakeup = 0;
1062 		error = 0;	/* error return if necessary */
1063 
1064 		for (i = 0; i < nsops; i++) {
1065 			sopptr = &sops[i];
1066 			semptr = &semakptr->u.sem_base[sopptr->sem_num];
1067 
1068 			DPRINTF((
1069 			    "semop:  semakptr=%p, sem_base=%p, "
1070 			    "semptr=%p, sem[%d]=%d : op=%d, flag=%s\n",
1071 			    semakptr, semakptr->u.sem_base, semptr,
1072 			    sopptr->sem_num, semptr->semval, sopptr->sem_op,
1073 			    (sopptr->sem_flg & IPC_NOWAIT) ?
1074 			    "nowait" : "wait"));
1075 
1076 			if (sopptr->sem_op < 0) {
1077 				if (semptr->semval + sopptr->sem_op < 0) {
1078 					DPRINTF(("semop:  can't do it now\n"));
1079 					break;
1080 				} else {
1081 					semptr->semval += sopptr->sem_op;
1082 					if (semptr->semval == 0 &&
1083 					    semptr->semzcnt > 0)
1084 						do_wakeup = 1;
1085 				}
1086 			} else if (sopptr->sem_op == 0) {
1087 				if (semptr->semval != 0) {
1088 					DPRINTF(("semop:  not zero now\n"));
1089 					break;
1090 				}
1091 			} else if (semptr->semval + sopptr->sem_op >
1092 			    seminfo.semvmx) {
1093 				error = ERANGE;
1094 				break;
1095 			} else {
1096 				if (semptr->semncnt > 0)
1097 					do_wakeup = 1;
1098 				semptr->semval += sopptr->sem_op;
1099 			}
1100 		}
1101 
1102 		/*
1103 		 * Did we get through the entire vector?
1104 		 */
1105 		if (i >= nsops)
1106 			goto done;
1107 
1108 		/*
1109 		 * No ... rollback anything that we've already done
1110 		 */
1111 		DPRINTF(("semop:  rollback 0 through %d\n", i-1));
1112 		for (j = 0; j < i; j++)
1113 			semakptr->u.sem_base[sops[j].sem_num].semval -=
1114 			    sops[j].sem_op;
1115 
1116 		/* If we detected an error, return it */
1117 		if (error != 0)
1118 			goto done2;
1119 
1120 		/*
1121 		 * If the request that we couldn't satisfy has the
1122 		 * NOWAIT flag set then return with EAGAIN.
1123 		 */
1124 		if (sopptr->sem_flg & IPC_NOWAIT) {
1125 			error = EAGAIN;
1126 			goto done2;
1127 		}
1128 
1129 		if (sopptr->sem_op == 0)
1130 			semptr->semzcnt++;
1131 		else
1132 			semptr->semncnt++;
1133 
1134 		DPRINTF(("semop:  good night!\n"));
1135 		error = msleep(semakptr, sema_mtxp, (PZERO - 4) | PCATCH,
1136 		    "semwait", 0);
1137 		DPRINTF(("semop:  good morning (error=%d)!\n", error));
1138 		/* return code is checked below, after sem[nz]cnt-- */
1139 
1140 		/*
1141 		 * Make sure that the semaphore still exists
1142 		 */
1143 		seq = semakptr->u.sem_perm.seq;
1144 		if ((semakptr->u.sem_perm.mode & SEM_ALLOC) == 0 ||
1145 		    seq != IPCID_TO_SEQ(uap->semid)) {
1146 			error = EIDRM;
1147 			goto done2;
1148 		}
1149 
1150 		/*
1151 		 * Renew the semaphore's pointer after wakeup since
1152 		 * during msleep sem_base may have been modified and semptr
1153 		 * is not valid any more
1154 		 */
1155 		semptr = &semakptr->u.sem_base[sopptr->sem_num];
1156 
1157 		/*
1158 		 * The semaphore is still alive.  Readjust the count of
1159 		 * waiting processes.
1160 		 */
1161 		if (sopptr->sem_op == 0)
1162 			semptr->semzcnt--;
1163 		else
1164 			semptr->semncnt--;
1165 
1166 		/*
1167 		 * Is it really morning, or was our sleep interrupted?
1168 		 * (Delayed check of msleep() return code because we
1169 		 * need to decrement sem[nz]cnt either way.)
1170 		 */
1171 		if (error != 0) {
1172 			error = EINTR;
1173 			goto done2;
1174 		}
1175 		DPRINTF(("semop:  good morning!\n"));
1176 	}
1177 
1178 done:
1179 	/*
1180 	 * Process any SEM_UNDO requests.
1181 	 */
1182 	if (do_undos) {
1183 		SEMUNDO_LOCK();
1184 		suptr = NULL;
1185 		for (i = 0; i < nsops; i++) {
1186 			/*
1187 			 * We only need to deal with SEM_UNDO's for non-zero
1188 			 * op's.
1189 			 */
1190 			int adjval;
1191 
1192 			if ((sops[i].sem_flg & SEM_UNDO) == 0)
1193 				continue;
1194 			adjval = sops[i].sem_op;
1195 			if (adjval == 0)
1196 				continue;
1197 			error = semundo_adjust(td, &suptr, semid, seq,
1198 			    sops[i].sem_num, -adjval);
1199 			if (error == 0)
1200 				continue;
1201 
1202 			/*
1203 			 * Oh-Oh!  We ran out of either sem_undo's or undo's.
1204 			 * Rollback the adjustments to this point and then
1205 			 * rollback the semaphore ups and down so we can return
1206 			 * with an error with all structures restored.  We
1207 			 * rollback the undo's in the exact reverse order that
1208 			 * we applied them.  This guarantees that we won't run
1209 			 * out of space as we roll things back out.
1210 			 */
1211 			for (j = 0; j < i; j++) {
1212 				k = i - j - 1;
1213 				if ((sops[k].sem_flg & SEM_UNDO) == 0)
1214 					continue;
1215 				adjval = sops[k].sem_op;
1216 				if (adjval == 0)
1217 					continue;
1218 				if (semundo_adjust(td, &suptr, semid, seq,
1219 				    sops[k].sem_num, adjval) != 0)
1220 					panic("semop - can't undo undos");
1221 			}
1222 
1223 			for (j = 0; j < nsops; j++)
1224 				semakptr->u.sem_base[sops[j].sem_num].semval -=
1225 				    sops[j].sem_op;
1226 
1227 			DPRINTF(("error = %d from semundo_adjust\n", error));
1228 			SEMUNDO_UNLOCK();
1229 			goto done2;
1230 		} /* loop through the sops */
1231 		SEMUNDO_UNLOCK();
1232 	} /* if (do_undos) */
1233 
1234 	/* We're definitely done - set the sempid's and time */
1235 	for (i = 0; i < nsops; i++) {
1236 		sopptr = &sops[i];
1237 		semptr = &semakptr->u.sem_base[sopptr->sem_num];
1238 		semptr->sempid = td->td_proc->p_pid;
1239 	}
1240 	semakptr->u.sem_otime = time_second;
1241 
1242 	/*
1243 	 * Do a wakeup if any semaphore was up'd whilst something was
1244 	 * sleeping on it.
1245 	 */
1246 	if (do_wakeup) {
1247 		DPRINTF(("semop:  doing wakeup\n"));
1248 		wakeup(semakptr);
1249 		DPRINTF(("semop:  back from wakeup\n"));
1250 	}
1251 	DPRINTF(("semop:  done\n"));
1252 	td->td_retval[0] = 0;
1253 done2:
1254 	mtx_unlock(sema_mtxp);
1255 	if (sops != small_sops)
1256 		free(sops, M_SEM);
1257 	return (error);
1258 }
1259 
1260 /*
1261  * Go through the undo structures for this process and apply the adjustments to
1262  * semaphores.
1263  */
1264 static void
1265 semexit_myhook(void *arg, struct proc *p)
1266 {
1267 	struct sem_undo *suptr;
1268 	struct semid_kernel *semakptr;
1269 	struct mtx *sema_mtxp;
1270 	int semid, semnum, adjval, ix;
1271 	unsigned short seq;
1272 
1273 	/*
1274 	 * Go through the chain of undo vectors looking for one
1275 	 * associated with this process.
1276 	 */
1277 	SEMUNDO_LOCK();
1278 	LIST_FOREACH(suptr, &semu_list, un_next) {
1279 		if (suptr->un_proc == p)
1280 			break;
1281 	}
1282 	if (suptr == NULL) {
1283 		SEMUNDO_UNLOCK();
1284 		return;
1285 	}
1286 	LIST_REMOVE(suptr, un_next);
1287 
1288 	DPRINTF(("proc @%p has undo structure with %d entries\n", p,
1289 	    suptr->un_cnt));
1290 
1291 	/*
1292 	 * If there are any active undo elements then process them.
1293 	 */
1294 	if (suptr->un_cnt > 0) {
1295 		SEMUNDO_UNLOCK();
1296 		for (ix = 0; ix < suptr->un_cnt; ix++) {
1297 			semid = suptr->un_ent[ix].un_id;
1298 			semnum = suptr->un_ent[ix].un_num;
1299 			adjval = suptr->un_ent[ix].un_adjval;
1300 			seq = suptr->un_ent[ix].un_seq;
1301 			semakptr = &sema[semid];
1302 			sema_mtxp = &sema_mtx[semid];
1303 
1304 			mtx_lock(sema_mtxp);
1305 			if ((semakptr->u.sem_perm.mode & SEM_ALLOC) == 0 ||
1306 			    (semakptr->u.sem_perm.seq != seq)) {
1307 				mtx_unlock(sema_mtxp);
1308 				continue;
1309 			}
1310 			if (semnum >= semakptr->u.sem_nsems)
1311 				panic("semexit - semnum out of range");
1312 
1313 			DPRINTF((
1314 			    "semexit:  %p id=%d num=%d(adj=%d) ; sem=%d\n",
1315 			    suptr->un_proc, suptr->un_ent[ix].un_id,
1316 			    suptr->un_ent[ix].un_num,
1317 			    suptr->un_ent[ix].un_adjval,
1318 			    semakptr->u.sem_base[semnum].semval));
1319 
1320 			if (adjval < 0 && semakptr->u.sem_base[semnum].semval <
1321 			    -adjval)
1322 				semakptr->u.sem_base[semnum].semval = 0;
1323 			else
1324 				semakptr->u.sem_base[semnum].semval += adjval;
1325 
1326 			wakeup(semakptr);
1327 			DPRINTF(("semexit:  back from wakeup\n"));
1328 			mtx_unlock(sema_mtxp);
1329 		}
1330 		SEMUNDO_LOCK();
1331 	}
1332 
1333 	/*
1334 	 * Deallocate the undo vector.
1335 	 */
1336 	DPRINTF(("removing vector\n"));
1337 	suptr->un_proc = NULL;
1338 	suptr->un_cnt = 0;
1339 	LIST_INSERT_HEAD(&semu_free_list, suptr, un_next);
1340 	SEMUNDO_UNLOCK();
1341 }
1342 
1343 static int
1344 sysctl_sema(SYSCTL_HANDLER_ARGS)
1345 {
1346 
1347 	return (SYSCTL_OUT(req, sema,
1348 	    sizeof(struct semid_kernel) * seminfo.semmni));
1349 }
1350