xref: /freebsd/sys/kern/sysv_sem.c (revision 39beb93c)
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 	mtx_destroy(&sem_mtx);
289 	mtx_destroy(&sem_undo_mtx);
290 	return (0);
291 }
292 
293 static int
294 sysvsem_modload(struct module *module, int cmd, void *arg)
295 {
296 	int error = 0;
297 
298 	switch (cmd) {
299 	case MOD_LOAD:
300 		seminit();
301 		break;
302 	case MOD_UNLOAD:
303 		error = semunload();
304 		break;
305 	case MOD_SHUTDOWN:
306 		break;
307 	default:
308 		error = EINVAL;
309 		break;
310 	}
311 	return (error);
312 }
313 
314 static moduledata_t sysvsem_mod = {
315 	"sysvsem",
316 	&sysvsem_modload,
317 	NULL
318 };
319 
320 SYSCALL_MODULE_HELPER(semsys);
321 SYSCALL_MODULE_HELPER(__semctl);
322 SYSCALL_MODULE_HELPER(semget);
323 SYSCALL_MODULE_HELPER(semop);
324 
325 DECLARE_MODULE(sysvsem, sysvsem_mod,
326 	SI_SUB_SYSV_SEM, SI_ORDER_FIRST);
327 MODULE_VERSION(sysvsem, 1);
328 
329 /*
330  * Entry point for all SEM calls.
331  */
332 int
333 semsys(td, uap)
334 	struct thread *td;
335 	/* XXX actually varargs. */
336 	struct semsys_args /* {
337 		int	which;
338 		int	a2;
339 		int	a3;
340 		int	a4;
341 		int	a5;
342 	} */ *uap;
343 {
344 	int error;
345 
346 	if (!jail_sysvipc_allowed && jailed(td->td_ucred))
347 		return (ENOSYS);
348 	if (uap->which < 0 ||
349 	    uap->which >= sizeof(semcalls)/sizeof(semcalls[0]))
350 		return (EINVAL);
351 	error = (*semcalls[uap->which])(td, &uap->a2);
352 	return (error);
353 }
354 
355 /*
356  * Allocate a new sem_undo structure for a process
357  * (returns ptr to structure or NULL if no more room)
358  */
359 
360 static struct sem_undo *
361 semu_alloc(struct thread *td)
362 {
363 	struct sem_undo *suptr;
364 
365 	SEMUNDO_LOCKASSERT(MA_OWNED);
366 	if ((suptr = LIST_FIRST(&semu_free_list)) == NULL)
367 		return (NULL);
368 	LIST_REMOVE(suptr, un_next);
369 	LIST_INSERT_HEAD(&semu_list, suptr, un_next);
370 	suptr->un_cnt = 0;
371 	suptr->un_proc = td->td_proc;
372 	return (suptr);
373 }
374 
375 static int
376 semu_try_free(struct sem_undo *suptr)
377 {
378 
379 	SEMUNDO_LOCKASSERT(MA_OWNED);
380 
381 	if (suptr->un_cnt != 0)
382 		return (0);
383 	LIST_REMOVE(suptr, un_next);
384 	LIST_INSERT_HEAD(&semu_free_list, suptr, un_next);
385 	return (1);
386 }
387 
388 /*
389  * Adjust a particular entry for a particular proc
390  */
391 
392 static int
393 semundo_adjust(struct thread *td, struct sem_undo **supptr, int semid,
394     int semseq, int semnum, int adjval)
395 {
396 	struct proc *p = td->td_proc;
397 	struct sem_undo *suptr;
398 	struct undo *sunptr;
399 	int i;
400 
401 	SEMUNDO_LOCKASSERT(MA_OWNED);
402 	/* Look for and remember the sem_undo if the caller doesn't provide
403 	   it */
404 
405 	suptr = *supptr;
406 	if (suptr == NULL) {
407 		LIST_FOREACH(suptr, &semu_list, un_next) {
408 			if (suptr->un_proc == p) {
409 				*supptr = suptr;
410 				break;
411 			}
412 		}
413 		if (suptr == NULL) {
414 			if (adjval == 0)
415 				return(0);
416 			suptr = semu_alloc(td);
417 			if (suptr == NULL)
418 				return (ENOSPC);
419 			*supptr = suptr;
420 		}
421 	}
422 
423 	/*
424 	 * Look for the requested entry and adjust it (delete if adjval becomes
425 	 * 0).
426 	 */
427 	sunptr = &suptr->un_ent[0];
428 	for (i = 0; i < suptr->un_cnt; i++, sunptr++) {
429 		if (sunptr->un_id != semid || sunptr->un_num != semnum)
430 			continue;
431 		if (adjval != 0) {
432 			adjval += sunptr->un_adjval;
433 			if (adjval > seminfo.semaem || adjval < -seminfo.semaem)
434 				return (ERANGE);
435 		}
436 		sunptr->un_adjval = adjval;
437 		if (sunptr->un_adjval == 0) {
438 			suptr->un_cnt--;
439 			if (i < suptr->un_cnt)
440 				suptr->un_ent[i] =
441 				    suptr->un_ent[suptr->un_cnt];
442 			if (suptr->un_cnt == 0)
443 				semu_try_free(suptr);
444 		}
445 		return (0);
446 	}
447 
448 	/* Didn't find the right entry - create it */
449 	if (adjval == 0)
450 		return (0);
451 	if (adjval > seminfo.semaem || adjval < -seminfo.semaem)
452 		return (ERANGE);
453 	if (suptr->un_cnt != seminfo.semume) {
454 		sunptr = &suptr->un_ent[suptr->un_cnt];
455 		suptr->un_cnt++;
456 		sunptr->un_adjval = adjval;
457 		sunptr->un_id = semid;
458 		sunptr->un_num = semnum;
459 		sunptr->un_seq = semseq;
460 	} else
461 		return (EINVAL);
462 	return (0);
463 }
464 
465 static void
466 semundo_clear(int semid, int semnum)
467 {
468 	struct sem_undo *suptr, *suptr1;
469 	struct undo *sunptr;
470 	int i;
471 
472 	SEMUNDO_LOCKASSERT(MA_OWNED);
473 	LIST_FOREACH_SAFE(suptr, &semu_list, un_next, suptr1) {
474 		sunptr = &suptr->un_ent[0];
475 		for (i = 0; i < suptr->un_cnt; i++, sunptr++) {
476 			if (sunptr->un_id != semid)
477 				continue;
478 			if (semnum == -1 || sunptr->un_num == semnum) {
479 				suptr->un_cnt--;
480 				if (i < suptr->un_cnt) {
481 					suptr->un_ent[i] =
482 					    suptr->un_ent[suptr->un_cnt];
483 					continue;
484 				}
485 				semu_try_free(suptr);
486 			}
487 			if (semnum != -1)
488 				break;
489 		}
490 	}
491 }
492 
493 static int
494 semvalid(int semid, struct semid_kernel *semakptr)
495 {
496 
497 	return ((semakptr->u.sem_perm.mode & SEM_ALLOC) == 0 ||
498 	    semakptr->u.sem_perm.seq != IPCID_TO_SEQ(semid) ? EINVAL : 0);
499 }
500 
501 /*
502  * Note that the user-mode half of this passes a union, not a pointer.
503  */
504 #ifndef _SYS_SYSPROTO_H_
505 struct __semctl_args {
506 	int	semid;
507 	int	semnum;
508 	int	cmd;
509 	union	semun *arg;
510 };
511 #endif
512 int
513 __semctl(struct thread *td, struct __semctl_args *uap)
514 {
515 	struct semid_ds dsbuf;
516 	union semun arg, semun;
517 	register_t rval;
518 	int error;
519 
520 	switch (uap->cmd) {
521 	case SEM_STAT:
522 	case IPC_SET:
523 	case IPC_STAT:
524 	case GETALL:
525 	case SETVAL:
526 	case SETALL:
527 		error = copyin(uap->arg, &arg, sizeof(arg));
528 		if (error)
529 			return (error);
530 		break;
531 	}
532 
533 	switch (uap->cmd) {
534 	case SEM_STAT:
535 	case IPC_STAT:
536 		semun.buf = &dsbuf;
537 		break;
538 	case IPC_SET:
539 		error = copyin(arg.buf, &dsbuf, sizeof(dsbuf));
540 		if (error)
541 			return (error);
542 		semun.buf = &dsbuf;
543 		break;
544 	case GETALL:
545 	case SETALL:
546 		semun.array = arg.array;
547 		break;
548 	case SETVAL:
549 		semun.val = arg.val;
550 		break;
551 	}
552 
553 	error = kern_semctl(td, uap->semid, uap->semnum, uap->cmd, &semun,
554 	    &rval);
555 	if (error)
556 		return (error);
557 
558 	switch (uap->cmd) {
559 	case SEM_STAT:
560 	case IPC_STAT:
561 		error = copyout(&dsbuf, arg.buf, sizeof(dsbuf));
562 		break;
563 	}
564 
565 	if (error == 0)
566 		td->td_retval[0] = rval;
567 	return (error);
568 }
569 
570 int
571 kern_semctl(struct thread *td, int semid, int semnum, int cmd,
572     union semun *arg, register_t *rval)
573 {
574 	u_short *array;
575 	struct ucred *cred = td->td_ucred;
576 	int i, error;
577 	struct semid_ds *sbuf;
578 	struct semid_kernel *semakptr;
579 	struct mtx *sema_mtxp;
580 	u_short usval, count;
581 	int semidx;
582 
583 	DPRINTF(("call to semctl(%d, %d, %d, 0x%p)\n",
584 	    semid, semnum, cmd, arg));
585 	if (!jail_sysvipc_allowed && jailed(td->td_ucred))
586 		return (ENOSYS);
587 
588 	array = NULL;
589 
590 	switch(cmd) {
591 	case SEM_STAT:
592 		/*
593 		 * For this command we assume semid is an array index
594 		 * rather than an IPC id.
595 		 */
596 		if (semid < 0 || semid >= seminfo.semmni)
597 			return (EINVAL);
598 		semakptr = &sema[semid];
599 		sema_mtxp = &sema_mtx[semid];
600 		mtx_lock(sema_mtxp);
601 		if ((semakptr->u.sem_perm.mode & SEM_ALLOC) == 0) {
602 			error = EINVAL;
603 			goto done2;
604 		}
605 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
606 			goto done2;
607 #ifdef MAC
608 		error = mac_sysvsem_check_semctl(cred, semakptr, cmd);
609 		if (error != 0)
610 			goto done2;
611 #endif
612 		bcopy(&semakptr->u, arg->buf, sizeof(struct semid_ds));
613 		*rval = IXSEQ_TO_IPCID(semid, semakptr->u.sem_perm);
614 		mtx_unlock(sema_mtxp);
615 		return (0);
616 	}
617 
618 	semidx = IPCID_TO_IX(semid);
619 	if (semidx < 0 || semidx >= seminfo.semmni)
620 		return (EINVAL);
621 
622 	semakptr = &sema[semidx];
623 	sema_mtxp = &sema_mtx[semidx];
624 	if (cmd == IPC_RMID)
625 		mtx_lock(&sem_mtx);
626 	mtx_lock(sema_mtxp);
627 #ifdef MAC
628 	error = mac_sysvsem_check_semctl(cred, semakptr, cmd);
629 	if (error != 0)
630 		goto done2;
631 #endif
632 
633 	error = 0;
634 	*rval = 0;
635 
636 	switch (cmd) {
637 	case IPC_RMID:
638 		if ((error = semvalid(semid, semakptr)) != 0)
639 			goto done2;
640 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_M)))
641 			goto done2;
642 		semakptr->u.sem_perm.cuid = cred->cr_uid;
643 		semakptr->u.sem_perm.uid = cred->cr_uid;
644 		semakptr->u.sem_perm.mode = 0;
645 		SEMUNDO_LOCK();
646 		semundo_clear(semidx, -1);
647 		SEMUNDO_UNLOCK();
648 #ifdef MAC
649 		mac_sysvsem_cleanup(semakptr);
650 #endif
651 		wakeup(semakptr);
652 		for (i = 0; i < seminfo.semmni; i++) {
653 			if ((sema[i].u.sem_perm.mode & SEM_ALLOC) &&
654 			    sema[i].u.sem_base > semakptr->u.sem_base)
655 				mtx_lock_flags(&sema_mtx[i], LOP_DUPOK);
656 		}
657 		for (i = semakptr->u.sem_base - sem; i < semtot; i++)
658 			sem[i] = sem[i + semakptr->u.sem_nsems];
659 		for (i = 0; i < seminfo.semmni; i++) {
660 			if ((sema[i].u.sem_perm.mode & SEM_ALLOC) &&
661 			    sema[i].u.sem_base > semakptr->u.sem_base) {
662 				sema[i].u.sem_base -= semakptr->u.sem_nsems;
663 				mtx_unlock(&sema_mtx[i]);
664 			}
665 		}
666 		semtot -= semakptr->u.sem_nsems;
667 		break;
668 
669 	case IPC_SET:
670 		if ((error = semvalid(semid, semakptr)) != 0)
671 			goto done2;
672 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_M)))
673 			goto done2;
674 		sbuf = arg->buf;
675 		semakptr->u.sem_perm.uid = sbuf->sem_perm.uid;
676 		semakptr->u.sem_perm.gid = sbuf->sem_perm.gid;
677 		semakptr->u.sem_perm.mode = (semakptr->u.sem_perm.mode &
678 		    ~0777) | (sbuf->sem_perm.mode & 0777);
679 		semakptr->u.sem_ctime = time_second;
680 		break;
681 
682 	case IPC_STAT:
683 		if ((error = semvalid(semid, semakptr)) != 0)
684 			goto done2;
685 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
686 			goto done2;
687 		bcopy(&semakptr->u, arg->buf, sizeof(struct semid_ds));
688 		break;
689 
690 	case GETNCNT:
691 		if ((error = semvalid(semid, semakptr)) != 0)
692 			goto done2;
693 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
694 			goto done2;
695 		if (semnum < 0 || semnum >= semakptr->u.sem_nsems) {
696 			error = EINVAL;
697 			goto done2;
698 		}
699 		*rval = semakptr->u.sem_base[semnum].semncnt;
700 		break;
701 
702 	case GETPID:
703 		if ((error = semvalid(semid, semakptr)) != 0)
704 			goto done2;
705 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
706 			goto done2;
707 		if (semnum < 0 || semnum >= semakptr->u.sem_nsems) {
708 			error = EINVAL;
709 			goto done2;
710 		}
711 		*rval = semakptr->u.sem_base[semnum].sempid;
712 		break;
713 
714 	case GETVAL:
715 		if ((error = semvalid(semid, semakptr)) != 0)
716 			goto done2;
717 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
718 			goto done2;
719 		if (semnum < 0 || semnum >= semakptr->u.sem_nsems) {
720 			error = EINVAL;
721 			goto done2;
722 		}
723 		*rval = semakptr->u.sem_base[semnum].semval;
724 		break;
725 
726 	case GETALL:
727 		/*
728 		 * Unfortunately, callers of this function don't know
729 		 * in advance how many semaphores are in this set.
730 		 * While we could just allocate the maximum size array
731 		 * and pass the actual size back to the caller, that
732 		 * won't work for SETALL since we can't copyin() more
733 		 * data than the user specified as we may return a
734 		 * spurious EFAULT.
735 		 *
736 		 * Note that the number of semaphores in a set is
737 		 * fixed for the life of that set.  The only way that
738 		 * the 'count' could change while are blocked in
739 		 * malloc() is if this semaphore set were destroyed
740 		 * and a new one created with the same index.
741 		 * However, semvalid() will catch that due to the
742 		 * sequence number unless exactly 0x8000 (or a
743 		 * multiple thereof) semaphore sets for the same index
744 		 * are created and destroyed while we are in malloc!
745 		 *
746 		 */
747 		count = semakptr->u.sem_nsems;
748 		mtx_unlock(sema_mtxp);
749 		array = malloc(sizeof(*array) * count, M_TEMP, M_WAITOK);
750 		mtx_lock(sema_mtxp);
751 		if ((error = semvalid(semid, semakptr)) != 0)
752 			goto done2;
753 		KASSERT(count == semakptr->u.sem_nsems, ("nsems changed"));
754 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
755 			goto done2;
756 		for (i = 0; i < semakptr->u.sem_nsems; i++)
757 			array[i] = semakptr->u.sem_base[i].semval;
758 		mtx_unlock(sema_mtxp);
759 		error = copyout(array, arg->array, count * sizeof(*array));
760 		mtx_lock(sema_mtxp);
761 		break;
762 
763 	case GETZCNT:
764 		if ((error = semvalid(semid, semakptr)) != 0)
765 			goto done2;
766 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
767 			goto done2;
768 		if (semnum < 0 || semnum >= semakptr->u.sem_nsems) {
769 			error = EINVAL;
770 			goto done2;
771 		}
772 		*rval = semakptr->u.sem_base[semnum].semzcnt;
773 		break;
774 
775 	case SETVAL:
776 		if ((error = semvalid(semid, semakptr)) != 0)
777 			goto done2;
778 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_W)))
779 			goto done2;
780 		if (semnum < 0 || semnum >= semakptr->u.sem_nsems) {
781 			error = EINVAL;
782 			goto done2;
783 		}
784 		if (arg->val < 0 || arg->val > seminfo.semvmx) {
785 			error = ERANGE;
786 			goto done2;
787 		}
788 		semakptr->u.sem_base[semnum].semval = arg->val;
789 		SEMUNDO_LOCK();
790 		semundo_clear(semidx, semnum);
791 		SEMUNDO_UNLOCK();
792 		wakeup(semakptr);
793 		break;
794 
795 	case SETALL:
796 		/*
797 		 * See comment on GETALL for why 'count' shouldn't change
798 		 * and why we require a userland buffer.
799 		 */
800 		count = semakptr->u.sem_nsems;
801 		mtx_unlock(sema_mtxp);
802 		array = malloc(sizeof(*array) * count, M_TEMP, M_WAITOK);
803 		error = copyin(arg->array, array, count * sizeof(*array));
804 		mtx_lock(sema_mtxp);
805 		if (error)
806 			break;
807 		if ((error = semvalid(semid, semakptr)) != 0)
808 			goto done2;
809 		KASSERT(count == semakptr->u.sem_nsems, ("nsems changed"));
810 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_W)))
811 			goto done2;
812 		for (i = 0; i < semakptr->u.sem_nsems; i++) {
813 			usval = array[i];
814 			if (usval > seminfo.semvmx) {
815 				error = ERANGE;
816 				break;
817 			}
818 			semakptr->u.sem_base[i].semval = usval;
819 		}
820 		SEMUNDO_LOCK();
821 		semundo_clear(semidx, -1);
822 		SEMUNDO_UNLOCK();
823 		wakeup(semakptr);
824 		break;
825 
826 	default:
827 		error = EINVAL;
828 		break;
829 	}
830 
831 done2:
832 	mtx_unlock(sema_mtxp);
833 	if (cmd == IPC_RMID)
834 		mtx_unlock(&sem_mtx);
835 	if (array != NULL)
836 		free(array, M_TEMP);
837 	return(error);
838 }
839 
840 #ifndef _SYS_SYSPROTO_H_
841 struct semget_args {
842 	key_t	key;
843 	int	nsems;
844 	int	semflg;
845 };
846 #endif
847 int
848 semget(struct thread *td, struct semget_args *uap)
849 {
850 	int semid, error = 0;
851 	int key = uap->key;
852 	int nsems = uap->nsems;
853 	int semflg = uap->semflg;
854 	struct ucred *cred = td->td_ucred;
855 
856 	DPRINTF(("semget(0x%x, %d, 0%o)\n", key, nsems, semflg));
857 	if (!jail_sysvipc_allowed && jailed(td->td_ucred))
858 		return (ENOSYS);
859 
860 	mtx_lock(&sem_mtx);
861 	if (key != IPC_PRIVATE) {
862 		for (semid = 0; semid < seminfo.semmni; semid++) {
863 			if ((sema[semid].u.sem_perm.mode & SEM_ALLOC) &&
864 			    sema[semid].u.sem_perm.key == key)
865 				break;
866 		}
867 		if (semid < seminfo.semmni) {
868 			DPRINTF(("found public key\n"));
869 			if ((error = ipcperm(td, &sema[semid].u.sem_perm,
870 			    semflg & 0700))) {
871 				goto done2;
872 			}
873 			if (nsems > 0 && sema[semid].u.sem_nsems < nsems) {
874 				DPRINTF(("too small\n"));
875 				error = EINVAL;
876 				goto done2;
877 			}
878 			if ((semflg & IPC_CREAT) && (semflg & IPC_EXCL)) {
879 				DPRINTF(("not exclusive\n"));
880 				error = EEXIST;
881 				goto done2;
882 			}
883 #ifdef MAC
884 			error = mac_sysvsem_check_semget(cred, &sema[semid]);
885 			if (error != 0)
886 				goto done2;
887 #endif
888 			goto found;
889 		}
890 	}
891 
892 	DPRINTF(("need to allocate the semid_kernel\n"));
893 	if (key == IPC_PRIVATE || (semflg & IPC_CREAT)) {
894 		if (nsems <= 0 || nsems > seminfo.semmsl) {
895 			DPRINTF(("nsems out of range (0<%d<=%d)\n", nsems,
896 			    seminfo.semmsl));
897 			error = EINVAL;
898 			goto done2;
899 		}
900 		if (nsems > seminfo.semmns - semtot) {
901 			DPRINTF((
902 			    "not enough semaphores left (need %d, got %d)\n",
903 			    nsems, seminfo.semmns - semtot));
904 			error = ENOSPC;
905 			goto done2;
906 		}
907 		for (semid = 0; semid < seminfo.semmni; semid++) {
908 			if ((sema[semid].u.sem_perm.mode & SEM_ALLOC) == 0)
909 				break;
910 		}
911 		if (semid == seminfo.semmni) {
912 			DPRINTF(("no more semid_kernel's available\n"));
913 			error = ENOSPC;
914 			goto done2;
915 		}
916 		DPRINTF(("semid %d is available\n", semid));
917 		mtx_lock(&sema_mtx[semid]);
918 		KASSERT((sema[semid].u.sem_perm.mode & SEM_ALLOC) == 0,
919 		    ("Lost semaphore %d", semid));
920 		sema[semid].u.sem_perm.key = key;
921 		sema[semid].u.sem_perm.cuid = cred->cr_uid;
922 		sema[semid].u.sem_perm.uid = cred->cr_uid;
923 		sema[semid].u.sem_perm.cgid = cred->cr_gid;
924 		sema[semid].u.sem_perm.gid = cred->cr_gid;
925 		sema[semid].u.sem_perm.mode = (semflg & 0777) | SEM_ALLOC;
926 		sema[semid].u.sem_perm.seq =
927 		    (sema[semid].u.sem_perm.seq + 1) & 0x7fff;
928 		sema[semid].u.sem_nsems = nsems;
929 		sema[semid].u.sem_otime = 0;
930 		sema[semid].u.sem_ctime = time_second;
931 		sema[semid].u.sem_base = &sem[semtot];
932 		semtot += nsems;
933 		bzero(sema[semid].u.sem_base,
934 		    sizeof(sema[semid].u.sem_base[0])*nsems);
935 #ifdef MAC
936 		mac_sysvsem_create(cred, &sema[semid]);
937 #endif
938 		mtx_unlock(&sema_mtx[semid]);
939 		DPRINTF(("sembase = %p, next = %p\n",
940 		    sema[semid].u.sem_base, &sem[semtot]));
941 	} else {
942 		DPRINTF(("didn't find it and wasn't asked to create it\n"));
943 		error = ENOENT;
944 		goto done2;
945 	}
946 
947 found:
948 	td->td_retval[0] = IXSEQ_TO_IPCID(semid, sema[semid].u.sem_perm);
949 done2:
950 	mtx_unlock(&sem_mtx);
951 	return (error);
952 }
953 
954 #ifndef _SYS_SYSPROTO_H_
955 struct semop_args {
956 	int	semid;
957 	struct	sembuf *sops;
958 	size_t	nsops;
959 };
960 #endif
961 int
962 semop(struct thread *td, struct semop_args *uap)
963 {
964 #define SMALL_SOPS	8
965 	struct sembuf small_sops[SMALL_SOPS];
966 	int semid = uap->semid;
967 	size_t nsops = uap->nsops;
968 	struct sembuf *sops;
969 	struct semid_kernel *semakptr;
970 	struct sembuf *sopptr = 0;
971 	struct sem *semptr = 0;
972 	struct sem_undo *suptr;
973 	struct mtx *sema_mtxp;
974 	size_t i, j, k;
975 	int error;
976 	int do_wakeup, do_undos;
977 	unsigned short seq;
978 
979 #ifdef SEM_DEBUG
980 	sops = NULL;
981 #endif
982 	DPRINTF(("call to semop(%d, %p, %u)\n", semid, sops, nsops));
983 
984 	if (!jail_sysvipc_allowed && jailed(td->td_ucred))
985 		return (ENOSYS);
986 
987 	semid = IPCID_TO_IX(semid);	/* Convert back to zero origin */
988 
989 	if (semid < 0 || semid >= seminfo.semmni)
990 		return (EINVAL);
991 
992 	/* Allocate memory for sem_ops */
993 	if (nsops <= SMALL_SOPS)
994 		sops = small_sops;
995 	else if (nsops <= seminfo.semopm)
996 		sops = malloc(nsops * sizeof(*sops), M_TEMP, M_WAITOK);
997 	else {
998 		DPRINTF(("too many sops (max=%d, nsops=%d)\n", seminfo.semopm,
999 		    nsops));
1000 		return (E2BIG);
1001 	}
1002 	if ((error = copyin(uap->sops, sops, nsops * sizeof(sops[0]))) != 0) {
1003 		DPRINTF(("error = %d from copyin(%p, %p, %d)\n", error,
1004 		    uap->sops, sops, nsops * sizeof(sops[0])));
1005 		if (sops != small_sops)
1006 			free(sops, M_SEM);
1007 		return (error);
1008 	}
1009 
1010 	semakptr = &sema[semid];
1011 	sema_mtxp = &sema_mtx[semid];
1012 	mtx_lock(sema_mtxp);
1013 	if ((semakptr->u.sem_perm.mode & SEM_ALLOC) == 0) {
1014 		error = EINVAL;
1015 		goto done2;
1016 	}
1017 	seq = semakptr->u.sem_perm.seq;
1018 	if (seq != IPCID_TO_SEQ(uap->semid)) {
1019 		error = EINVAL;
1020 		goto done2;
1021 	}
1022 	/*
1023 	 * Initial pass thru sops to see what permissions are needed.
1024 	 * Also perform any checks that don't need repeating on each
1025 	 * attempt to satisfy the request vector.
1026 	 */
1027 	j = 0;		/* permission needed */
1028 	do_undos = 0;
1029 	for (i = 0; i < nsops; i++) {
1030 		sopptr = &sops[i];
1031 		if (sopptr->sem_num >= semakptr->u.sem_nsems) {
1032 			error = EFBIG;
1033 			goto done2;
1034 		}
1035 		if (sopptr->sem_flg & SEM_UNDO && sopptr->sem_op != 0)
1036 			do_undos = 1;
1037 		j |= (sopptr->sem_op == 0) ? SEM_R : SEM_A;
1038 	}
1039 
1040 	if ((error = ipcperm(td, &semakptr->u.sem_perm, j))) {
1041 		DPRINTF(("error = %d from ipaccess\n", error));
1042 		goto done2;
1043 	}
1044 #ifdef MAC
1045 	error = mac_sysvsem_check_semop(td->td_ucred, semakptr, j);
1046 	if (error != 0)
1047 		goto done2;
1048 #endif
1049 
1050 	/*
1051 	 * Loop trying to satisfy the vector of requests.
1052 	 * If we reach a point where we must wait, any requests already
1053 	 * performed are rolled back and we go to sleep until some other
1054 	 * process wakes us up.  At this point, we start all over again.
1055 	 *
1056 	 * This ensures that from the perspective of other tasks, a set
1057 	 * of requests is atomic (never partially satisfied).
1058 	 */
1059 	for (;;) {
1060 		do_wakeup = 0;
1061 		error = 0;	/* error return if necessary */
1062 
1063 		for (i = 0; i < nsops; i++) {
1064 			sopptr = &sops[i];
1065 			semptr = &semakptr->u.sem_base[sopptr->sem_num];
1066 
1067 			DPRINTF((
1068 			    "semop:  semakptr=%p, sem_base=%p, "
1069 			    "semptr=%p, sem[%d]=%d : op=%d, flag=%s\n",
1070 			    semakptr, semakptr->u.sem_base, semptr,
1071 			    sopptr->sem_num, semptr->semval, sopptr->sem_op,
1072 			    (sopptr->sem_flg & IPC_NOWAIT) ?
1073 			    "nowait" : "wait"));
1074 
1075 			if (sopptr->sem_op < 0) {
1076 				if (semptr->semval + sopptr->sem_op < 0) {
1077 					DPRINTF(("semop:  can't do it now\n"));
1078 					break;
1079 				} else {
1080 					semptr->semval += sopptr->sem_op;
1081 					if (semptr->semval == 0 &&
1082 					    semptr->semzcnt > 0)
1083 						do_wakeup = 1;
1084 				}
1085 			} else if (sopptr->sem_op == 0) {
1086 				if (semptr->semval != 0) {
1087 					DPRINTF(("semop:  not zero now\n"));
1088 					break;
1089 				}
1090 			} else if (semptr->semval + sopptr->sem_op >
1091 			    seminfo.semvmx) {
1092 				error = ERANGE;
1093 				break;
1094 			} else {
1095 				if (semptr->semncnt > 0)
1096 					do_wakeup = 1;
1097 				semptr->semval += sopptr->sem_op;
1098 			}
1099 		}
1100 
1101 		/*
1102 		 * Did we get through the entire vector?
1103 		 */
1104 		if (i >= nsops)
1105 			goto done;
1106 
1107 		/*
1108 		 * No ... rollback anything that we've already done
1109 		 */
1110 		DPRINTF(("semop:  rollback 0 through %d\n", i-1));
1111 		for (j = 0; j < i; j++)
1112 			semakptr->u.sem_base[sops[j].sem_num].semval -=
1113 			    sops[j].sem_op;
1114 
1115 		/* If we detected an error, return it */
1116 		if (error != 0)
1117 			goto done2;
1118 
1119 		/*
1120 		 * If the request that we couldn't satisfy has the
1121 		 * NOWAIT flag set then return with EAGAIN.
1122 		 */
1123 		if (sopptr->sem_flg & IPC_NOWAIT) {
1124 			error = EAGAIN;
1125 			goto done2;
1126 		}
1127 
1128 		if (sopptr->sem_op == 0)
1129 			semptr->semzcnt++;
1130 		else
1131 			semptr->semncnt++;
1132 
1133 		DPRINTF(("semop:  good night!\n"));
1134 		error = msleep(semakptr, sema_mtxp, (PZERO - 4) | PCATCH,
1135 		    "semwait", 0);
1136 		DPRINTF(("semop:  good morning (error=%d)!\n", error));
1137 		/* return code is checked below, after sem[nz]cnt-- */
1138 
1139 		/*
1140 		 * Make sure that the semaphore still exists
1141 		 */
1142 		seq = semakptr->u.sem_perm.seq;
1143 		if ((semakptr->u.sem_perm.mode & SEM_ALLOC) == 0 ||
1144 		    seq != IPCID_TO_SEQ(uap->semid)) {
1145 			error = EIDRM;
1146 			goto done2;
1147 		}
1148 
1149 		/*
1150 		 * Renew the semaphore's pointer after wakeup since
1151 		 * during msleep sem_base may have been modified and semptr
1152 		 * is not valid any more
1153 		 */
1154 		semptr = &semakptr->u.sem_base[sopptr->sem_num];
1155 
1156 		/*
1157 		 * The semaphore is still alive.  Readjust the count of
1158 		 * waiting processes.
1159 		 */
1160 		if (sopptr->sem_op == 0)
1161 			semptr->semzcnt--;
1162 		else
1163 			semptr->semncnt--;
1164 
1165 		/*
1166 		 * Is it really morning, or was our sleep interrupted?
1167 		 * (Delayed check of msleep() return code because we
1168 		 * need to decrement sem[nz]cnt either way.)
1169 		 */
1170 		if (error != 0) {
1171 			error = EINTR;
1172 			goto done2;
1173 		}
1174 		DPRINTF(("semop:  good morning!\n"));
1175 	}
1176 
1177 done:
1178 	/*
1179 	 * Process any SEM_UNDO requests.
1180 	 */
1181 	if (do_undos) {
1182 		SEMUNDO_LOCK();
1183 		suptr = NULL;
1184 		for (i = 0; i < nsops; i++) {
1185 			/*
1186 			 * We only need to deal with SEM_UNDO's for non-zero
1187 			 * op's.
1188 			 */
1189 			int adjval;
1190 
1191 			if ((sops[i].sem_flg & SEM_UNDO) == 0)
1192 				continue;
1193 			adjval = sops[i].sem_op;
1194 			if (adjval == 0)
1195 				continue;
1196 			error = semundo_adjust(td, &suptr, semid, seq,
1197 			    sops[i].sem_num, -adjval);
1198 			if (error == 0)
1199 				continue;
1200 
1201 			/*
1202 			 * Oh-Oh!  We ran out of either sem_undo's or undo's.
1203 			 * Rollback the adjustments to this point and then
1204 			 * rollback the semaphore ups and down so we can return
1205 			 * with an error with all structures restored.  We
1206 			 * rollback the undo's in the exact reverse order that
1207 			 * we applied them.  This guarantees that we won't run
1208 			 * out of space as we roll things back out.
1209 			 */
1210 			for (j = 0; j < i; j++) {
1211 				k = i - j - 1;
1212 				if ((sops[k].sem_flg & SEM_UNDO) == 0)
1213 					continue;
1214 				adjval = sops[k].sem_op;
1215 				if (adjval == 0)
1216 					continue;
1217 				if (semundo_adjust(td, &suptr, semid, seq,
1218 				    sops[k].sem_num, adjval) != 0)
1219 					panic("semop - can't undo undos");
1220 			}
1221 
1222 			for (j = 0; j < nsops; j++)
1223 				semakptr->u.sem_base[sops[j].sem_num].semval -=
1224 				    sops[j].sem_op;
1225 
1226 			DPRINTF(("error = %d from semundo_adjust\n", error));
1227 			SEMUNDO_UNLOCK();
1228 			goto done2;
1229 		} /* loop through the sops */
1230 		SEMUNDO_UNLOCK();
1231 	} /* if (do_undos) */
1232 
1233 	/* We're definitely done - set the sempid's and time */
1234 	for (i = 0; i < nsops; i++) {
1235 		sopptr = &sops[i];
1236 		semptr = &semakptr->u.sem_base[sopptr->sem_num];
1237 		semptr->sempid = td->td_proc->p_pid;
1238 	}
1239 	semakptr->u.sem_otime = time_second;
1240 
1241 	/*
1242 	 * Do a wakeup if any semaphore was up'd whilst something was
1243 	 * sleeping on it.
1244 	 */
1245 	if (do_wakeup) {
1246 		DPRINTF(("semop:  doing wakeup\n"));
1247 		wakeup(semakptr);
1248 		DPRINTF(("semop:  back from wakeup\n"));
1249 	}
1250 	DPRINTF(("semop:  done\n"));
1251 	td->td_retval[0] = 0;
1252 done2:
1253 	mtx_unlock(sema_mtxp);
1254 	if (sops != small_sops)
1255 		free(sops, M_SEM);
1256 	return (error);
1257 }
1258 
1259 /*
1260  * Go through the undo structures for this process and apply the adjustments to
1261  * semaphores.
1262  */
1263 static void
1264 semexit_myhook(void *arg, struct proc *p)
1265 {
1266 	struct sem_undo *suptr;
1267 	struct semid_kernel *semakptr;
1268 	struct mtx *sema_mtxp;
1269 	int semid, semnum, adjval, ix;
1270 	unsigned short seq;
1271 
1272 	/*
1273 	 * Go through the chain of undo vectors looking for one
1274 	 * associated with this process.
1275 	 */
1276 	SEMUNDO_LOCK();
1277 	LIST_FOREACH(suptr, &semu_list, un_next) {
1278 		if (suptr->un_proc == p)
1279 			break;
1280 	}
1281 	if (suptr == NULL) {
1282 		SEMUNDO_UNLOCK();
1283 		return;
1284 	}
1285 	LIST_REMOVE(suptr, un_next);
1286 
1287 	DPRINTF(("proc @%p has undo structure with %d entries\n", p,
1288 	    suptr->un_cnt));
1289 
1290 	/*
1291 	 * If there are any active undo elements then process them.
1292 	 */
1293 	if (suptr->un_cnt > 0) {
1294 		SEMUNDO_UNLOCK();
1295 		for (ix = 0; ix < suptr->un_cnt; ix++) {
1296 			semid = suptr->un_ent[ix].un_id;
1297 			semnum = suptr->un_ent[ix].un_num;
1298 			adjval = suptr->un_ent[ix].un_adjval;
1299 			seq = suptr->un_ent[ix].un_seq;
1300 			semakptr = &sema[semid];
1301 			sema_mtxp = &sema_mtx[semid];
1302 
1303 			mtx_lock(sema_mtxp);
1304 			if ((semakptr->u.sem_perm.mode & SEM_ALLOC) == 0 ||
1305 			    (semakptr->u.sem_perm.seq != seq)) {
1306 				mtx_unlock(sema_mtxp);
1307 				continue;
1308 			}
1309 			if (semnum >= semakptr->u.sem_nsems)
1310 				panic("semexit - semnum out of range");
1311 
1312 			DPRINTF((
1313 			    "semexit:  %p id=%d num=%d(adj=%d) ; sem=%d\n",
1314 			    suptr->un_proc, suptr->un_ent[ix].un_id,
1315 			    suptr->un_ent[ix].un_num,
1316 			    suptr->un_ent[ix].un_adjval,
1317 			    semakptr->u.sem_base[semnum].semval));
1318 
1319 			if (adjval < 0 && semakptr->u.sem_base[semnum].semval <
1320 			    -adjval)
1321 				semakptr->u.sem_base[semnum].semval = 0;
1322 			else
1323 				semakptr->u.sem_base[semnum].semval += adjval;
1324 
1325 			wakeup(semakptr);
1326 			DPRINTF(("semexit:  back from wakeup\n"));
1327 			mtx_unlock(sema_mtxp);
1328 		}
1329 		SEMUNDO_LOCK();
1330 	}
1331 
1332 	/*
1333 	 * Deallocate the undo vector.
1334 	 */
1335 	DPRINTF(("removing vector\n"));
1336 	suptr->un_proc = NULL;
1337 	suptr->un_cnt = 0;
1338 	LIST_INSERT_HEAD(&semu_free_list, suptr, un_next);
1339 	SEMUNDO_UNLOCK();
1340 }
1341 
1342 static int
1343 sysctl_sema(SYSCTL_HANDLER_ARGS)
1344 {
1345 
1346 	return (SYSCTL_OUT(req, sema,
1347 	    sizeof(struct semid_kernel) * seminfo.semmni));
1348 }
1349