xref: /freebsd/sys/kern/sysv_sem.c (revision 0b3105a37d7adcadcb720112fed4dc4e8040be99)
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_compat.h"
43 #include "opt_sysvipc.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/racct.h>
55 #include <sys/sem.h>
56 #include <sys/syscall.h>
57 #include <sys/syscallsubr.h>
58 #include <sys/sysent.h>
59 #include <sys/sysctl.h>
60 #include <sys/uio.h>
61 #include <sys/malloc.h>
62 #include <sys/jail.h>
63 
64 #include <security/mac/mac_framework.h>
65 
66 FEATURE(sysv_sem, "System V semaphores support");
67 
68 static MALLOC_DEFINE(M_SEM, "sem", "SVID compatible semaphores");
69 
70 #ifdef SEM_DEBUG
71 #define DPRINTF(a)	printf a
72 #else
73 #define DPRINTF(a)
74 #endif
75 
76 static int seminit(void);
77 static int sysvsem_modload(struct module *, int, void *);
78 static int semunload(void);
79 static void semexit_myhook(void *arg, struct proc *p);
80 static int sysctl_sema(SYSCTL_HANDLER_ARGS);
81 static int semvalid(int semid, struct semid_kernel *semakptr);
82 
83 #ifndef _SYS_SYSPROTO_H_
84 struct __semctl_args;
85 int __semctl(struct thread *td, struct __semctl_args *uap);
86 struct semget_args;
87 int semget(struct thread *td, struct semget_args *uap);
88 struct semop_args;
89 int semop(struct thread *td, struct semop_args *uap);
90 #endif
91 
92 static struct sem_undo *semu_alloc(struct thread *td);
93 static int semundo_adjust(struct thread *td, struct sem_undo **supptr,
94     int semid, int semseq, int semnum, int adjval);
95 static void semundo_clear(int semid, int semnum);
96 
97 static struct mtx	sem_mtx;	/* semaphore global lock */
98 static struct mtx sem_undo_mtx;
99 static int	semtot = 0;
100 static struct semid_kernel *sema;	/* semaphore id pool */
101 static struct mtx *sema_mtx;	/* semaphore id pool mutexes*/
102 static struct sem *sem;		/* semaphore pool */
103 LIST_HEAD(, sem_undo) semu_list;	/* list of active undo structures */
104 LIST_HEAD(, sem_undo) semu_free_list;	/* list of free undo structures */
105 static int	*semu;		/* undo structure pool */
106 static eventhandler_tag semexit_tag;
107 
108 #define SEMUNDO_MTX		sem_undo_mtx
109 #define SEMUNDO_LOCK()		mtx_lock(&SEMUNDO_MTX);
110 #define SEMUNDO_UNLOCK()	mtx_unlock(&SEMUNDO_MTX);
111 #define SEMUNDO_LOCKASSERT(how)	mtx_assert(&SEMUNDO_MTX, (how));
112 
113 struct sem {
114 	u_short	semval;		/* semaphore value */
115 	pid_t	sempid;		/* pid of last operation */
116 	u_short	semncnt;	/* # awaiting semval > cval */
117 	u_short	semzcnt;	/* # awaiting semval = 0 */
118 };
119 
120 /*
121  * Undo structure (one per process)
122  */
123 struct sem_undo {
124 	LIST_ENTRY(sem_undo) un_next;	/* ptr to next active undo structure */
125 	struct	proc *un_proc;		/* owner of this structure */
126 	short	un_cnt;			/* # of active entries */
127 	struct undo {
128 		short	un_adjval;	/* adjust on exit values */
129 		short	un_num;		/* semaphore # */
130 		int	un_id;		/* semid */
131 		unsigned short un_seq;
132 	} un_ent[1];			/* undo entries */
133 };
134 
135 /*
136  * Configuration parameters
137  */
138 #ifndef SEMMNI
139 #define SEMMNI	50		/* # of semaphore identifiers */
140 #endif
141 #ifndef SEMMNS
142 #define SEMMNS	340		/* # of semaphores in system */
143 #endif
144 #ifndef SEMUME
145 #define SEMUME	50		/* max # of undo entries per process */
146 #endif
147 #ifndef SEMMNU
148 #define SEMMNU	150		/* # of undo structures in system */
149 #endif
150 
151 /* shouldn't need tuning */
152 #ifndef SEMMSL
153 #define SEMMSL	SEMMNS		/* max # of semaphores per id */
154 #endif
155 #ifndef SEMOPM
156 #define SEMOPM	100		/* max # of operations per semop call */
157 #endif
158 
159 #define SEMVMX	32767		/* semaphore maximum value */
160 #define SEMAEM	16384		/* adjust on exit max value */
161 
162 /*
163  * Due to the way semaphore memory is allocated, we have to ensure that
164  * SEMUSZ is properly aligned.
165  */
166 
167 #define SEM_ALIGN(bytes) (((bytes) + (sizeof(long) - 1)) & ~(sizeof(long) - 1))
168 
169 /* actual size of an undo structure */
170 #define SEMUSZ	SEM_ALIGN(offsetof(struct sem_undo, un_ent[SEMUME]))
171 
172 /*
173  * Macro to find a particular sem_undo vector
174  */
175 #define SEMU(ix) \
176 	((struct sem_undo *)(((intptr_t)semu)+ix * seminfo.semusz))
177 
178 /*
179  * semaphore info struct
180  */
181 struct seminfo seminfo = {
182                 SEMMNI,         /* # of semaphore identifiers */
183                 SEMMNS,         /* # of semaphores in system */
184                 SEMMNU,         /* # of undo structures in system */
185                 SEMMSL,         /* max # of semaphores per id */
186                 SEMOPM,         /* max # of operations per semop call */
187                 SEMUME,         /* max # of undo entries per process */
188                 SEMUSZ,         /* size in bytes of undo structure */
189                 SEMVMX,         /* semaphore maximum value */
190                 SEMAEM          /* adjust on exit max value */
191 };
192 
193 SYSCTL_INT(_kern_ipc, OID_AUTO, semmni, CTLFLAG_RDTUN, &seminfo.semmni, 0,
194     "Number of semaphore identifiers");
195 SYSCTL_INT(_kern_ipc, OID_AUTO, semmns, CTLFLAG_RDTUN, &seminfo.semmns, 0,
196     "Maximum number of semaphores in the system");
197 SYSCTL_INT(_kern_ipc, OID_AUTO, semmnu, CTLFLAG_RDTUN, &seminfo.semmnu, 0,
198     "Maximum number of undo structures in the system");
199 SYSCTL_INT(_kern_ipc, OID_AUTO, semmsl, CTLFLAG_RWTUN, &seminfo.semmsl, 0,
200     "Max semaphores per id");
201 SYSCTL_INT(_kern_ipc, OID_AUTO, semopm, CTLFLAG_RDTUN, &seminfo.semopm, 0,
202     "Max operations per semop call");
203 SYSCTL_INT(_kern_ipc, OID_AUTO, semume, CTLFLAG_RDTUN, &seminfo.semume, 0,
204     "Max undo entries per process");
205 SYSCTL_INT(_kern_ipc, OID_AUTO, semusz, CTLFLAG_RDTUN, &seminfo.semusz, 0,
206     "Size in bytes of undo structure");
207 SYSCTL_INT(_kern_ipc, OID_AUTO, semvmx, CTLFLAG_RWTUN, &seminfo.semvmx, 0,
208     "Semaphore maximum value");
209 SYSCTL_INT(_kern_ipc, OID_AUTO, semaem, CTLFLAG_RWTUN, &seminfo.semaem, 0,
210     "Adjust on exit max value");
211 SYSCTL_PROC(_kern_ipc, OID_AUTO, sema, CTLTYPE_OPAQUE | CTLFLAG_RD,
212     NULL, 0, sysctl_sema, "", "Semaphore id pool");
213 
214 static struct syscall_helper_data sem_syscalls[] = {
215 	SYSCALL_INIT_HELPER(__semctl),
216 	SYSCALL_INIT_HELPER(semget),
217 	SYSCALL_INIT_HELPER(semop),
218 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
219     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
220 	SYSCALL_INIT_HELPER(semsys),
221 	SYSCALL_INIT_HELPER_COMPAT(freebsd7___semctl),
222 #endif
223 	SYSCALL_INIT_LAST
224 };
225 
226 #ifdef COMPAT_FREEBSD32
227 #include <compat/freebsd32/freebsd32.h>
228 #include <compat/freebsd32/freebsd32_ipc.h>
229 #include <compat/freebsd32/freebsd32_proto.h>
230 #include <compat/freebsd32/freebsd32_signal.h>
231 #include <compat/freebsd32/freebsd32_syscall.h>
232 #include <compat/freebsd32/freebsd32_util.h>
233 
234 static struct syscall_helper_data sem32_syscalls[] = {
235 	SYSCALL32_INIT_HELPER(freebsd32_semctl),
236 	SYSCALL32_INIT_HELPER_COMPAT(semget),
237 	SYSCALL32_INIT_HELPER_COMPAT(semop),
238 	SYSCALL32_INIT_HELPER(freebsd32_semsys),
239 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
240     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
241 	SYSCALL32_INIT_HELPER(freebsd7_freebsd32_semctl),
242 #endif
243 	SYSCALL_INIT_LAST
244 };
245 #endif
246 
247 static int
248 seminit(void)
249 {
250 	int i, error;
251 
252 	sem = malloc(sizeof(struct sem) * seminfo.semmns, M_SEM, M_WAITOK);
253 	sema = malloc(sizeof(struct semid_kernel) * seminfo.semmni, M_SEM,
254 	    M_WAITOK);
255 	sema_mtx = malloc(sizeof(struct mtx) * seminfo.semmni, M_SEM,
256 	    M_WAITOK | M_ZERO);
257 	semu = malloc(seminfo.semmnu * seminfo.semusz, M_SEM, M_WAITOK);
258 
259 	for (i = 0; i < seminfo.semmni; i++) {
260 		sema[i].u.sem_base = 0;
261 		sema[i].u.sem_perm.mode = 0;
262 		sema[i].u.sem_perm.seq = 0;
263 #ifdef MAC
264 		mac_sysvsem_init(&sema[i]);
265 #endif
266 	}
267 	for (i = 0; i < seminfo.semmni; i++)
268 		mtx_init(&sema_mtx[i], "semid", NULL, MTX_DEF);
269 	LIST_INIT(&semu_free_list);
270 	for (i = 0; i < seminfo.semmnu; i++) {
271 		struct sem_undo *suptr = SEMU(i);
272 		suptr->un_proc = NULL;
273 		LIST_INSERT_HEAD(&semu_free_list, suptr, un_next);
274 	}
275 	LIST_INIT(&semu_list);
276 	mtx_init(&sem_mtx, "sem", NULL, MTX_DEF);
277 	mtx_init(&sem_undo_mtx, "semu", NULL, MTX_DEF);
278 	semexit_tag = EVENTHANDLER_REGISTER(process_exit, semexit_myhook, NULL,
279 	    EVENTHANDLER_PRI_ANY);
280 
281 	error = syscall_helper_register(sem_syscalls, SY_THR_STATIC_KLD);
282 	if (error != 0)
283 		return (error);
284 #ifdef COMPAT_FREEBSD32
285 	error = syscall32_helper_register(sem32_syscalls, SY_THR_STATIC_KLD);
286 	if (error != 0)
287 		return (error);
288 #endif
289 	return (0);
290 }
291 
292 static int
293 semunload(void)
294 {
295 	int i;
296 
297 	/* XXXKIB */
298 	if (semtot != 0)
299 		return (EBUSY);
300 
301 #ifdef COMPAT_FREEBSD32
302 	syscall32_helper_unregister(sem32_syscalls);
303 #endif
304 	syscall_helper_unregister(sem_syscalls);
305 	EVENTHANDLER_DEREGISTER(process_exit, semexit_tag);
306 #ifdef MAC
307 	for (i = 0; i < seminfo.semmni; i++)
308 		mac_sysvsem_destroy(&sema[i]);
309 #endif
310 	free(sem, M_SEM);
311 	free(sema, M_SEM);
312 	free(semu, M_SEM);
313 	for (i = 0; i < seminfo.semmni; i++)
314 		mtx_destroy(&sema_mtx[i]);
315 	free(sema_mtx, M_SEM);
316 	mtx_destroy(&sem_mtx);
317 	mtx_destroy(&sem_undo_mtx);
318 	return (0);
319 }
320 
321 static int
322 sysvsem_modload(struct module *module, int cmd, void *arg)
323 {
324 	int error = 0;
325 
326 	switch (cmd) {
327 	case MOD_LOAD:
328 		error = seminit();
329 		if (error != 0)
330 			semunload();
331 		break;
332 	case MOD_UNLOAD:
333 		error = semunload();
334 		break;
335 	case MOD_SHUTDOWN:
336 		break;
337 	default:
338 		error = EINVAL;
339 		break;
340 	}
341 	return (error);
342 }
343 
344 static moduledata_t sysvsem_mod = {
345 	"sysvsem",
346 	&sysvsem_modload,
347 	NULL
348 };
349 
350 DECLARE_MODULE(sysvsem, sysvsem_mod, SI_SUB_SYSV_SEM, SI_ORDER_FIRST);
351 MODULE_VERSION(sysvsem, 1);
352 
353 /*
354  * Allocate a new sem_undo structure for a process
355  * (returns ptr to structure or NULL if no more room)
356  */
357 
358 static struct sem_undo *
359 semu_alloc(struct thread *td)
360 {
361 	struct sem_undo *suptr;
362 
363 	SEMUNDO_LOCKASSERT(MA_OWNED);
364 	if ((suptr = LIST_FIRST(&semu_free_list)) == NULL)
365 		return (NULL);
366 	LIST_REMOVE(suptr, un_next);
367 	LIST_INSERT_HEAD(&semu_list, suptr, un_next);
368 	suptr->un_cnt = 0;
369 	suptr->un_proc = td->td_proc;
370 	return (suptr);
371 }
372 
373 static int
374 semu_try_free(struct sem_undo *suptr)
375 {
376 
377 	SEMUNDO_LOCKASSERT(MA_OWNED);
378 
379 	if (suptr->un_cnt != 0)
380 		return (0);
381 	LIST_REMOVE(suptr, un_next);
382 	LIST_INSERT_HEAD(&semu_free_list, suptr, un_next);
383 	return (1);
384 }
385 
386 /*
387  * Adjust a particular entry for a particular proc
388  */
389 
390 static int
391 semundo_adjust(struct thread *td, struct sem_undo **supptr, int semid,
392     int semseq, int semnum, int adjval)
393 {
394 	struct proc *p = td->td_proc;
395 	struct sem_undo *suptr;
396 	struct undo *sunptr;
397 	int i;
398 
399 	SEMUNDO_LOCKASSERT(MA_OWNED);
400 	/* Look for and remember the sem_undo if the caller doesn't provide
401 	   it */
402 
403 	suptr = *supptr;
404 	if (suptr == NULL) {
405 		LIST_FOREACH(suptr, &semu_list, un_next) {
406 			if (suptr->un_proc == p) {
407 				*supptr = suptr;
408 				break;
409 			}
410 		}
411 		if (suptr == NULL) {
412 			if (adjval == 0)
413 				return(0);
414 			suptr = semu_alloc(td);
415 			if (suptr == NULL)
416 				return (ENOSPC);
417 			*supptr = suptr;
418 		}
419 	}
420 
421 	/*
422 	 * Look for the requested entry and adjust it (delete if adjval becomes
423 	 * 0).
424 	 */
425 	sunptr = &suptr->un_ent[0];
426 	for (i = 0; i < suptr->un_cnt; i++, sunptr++) {
427 		if (sunptr->un_id != semid || sunptr->un_num != semnum)
428 			continue;
429 		if (adjval != 0) {
430 			adjval += sunptr->un_adjval;
431 			if (adjval > seminfo.semaem || adjval < -seminfo.semaem)
432 				return (ERANGE);
433 		}
434 		sunptr->un_adjval = adjval;
435 		if (sunptr->un_adjval == 0) {
436 			suptr->un_cnt--;
437 			if (i < suptr->un_cnt)
438 				suptr->un_ent[i] =
439 				    suptr->un_ent[suptr->un_cnt];
440 			if (suptr->un_cnt == 0)
441 				semu_try_free(suptr);
442 		}
443 		return (0);
444 	}
445 
446 	/* Didn't find the right entry - create it */
447 	if (adjval == 0)
448 		return (0);
449 	if (adjval > seminfo.semaem || adjval < -seminfo.semaem)
450 		return (ERANGE);
451 	if (suptr->un_cnt != seminfo.semume) {
452 		sunptr = &suptr->un_ent[suptr->un_cnt];
453 		suptr->un_cnt++;
454 		sunptr->un_adjval = adjval;
455 		sunptr->un_id = semid;
456 		sunptr->un_num = semnum;
457 		sunptr->un_seq = semseq;
458 	} else
459 		return (EINVAL);
460 	return (0);
461 }
462 
463 static void
464 semundo_clear(int semid, int semnum)
465 {
466 	struct sem_undo *suptr, *suptr1;
467 	struct undo *sunptr;
468 	int i;
469 
470 	SEMUNDO_LOCKASSERT(MA_OWNED);
471 	LIST_FOREACH_SAFE(suptr, &semu_list, un_next, suptr1) {
472 		sunptr = &suptr->un_ent[0];
473 		for (i = 0; i < suptr->un_cnt; i++, sunptr++) {
474 			if (sunptr->un_id != semid)
475 				continue;
476 			if (semnum == -1 || sunptr->un_num == semnum) {
477 				suptr->un_cnt--;
478 				if (i < suptr->un_cnt) {
479 					suptr->un_ent[i] =
480 					    suptr->un_ent[suptr->un_cnt];
481 					continue;
482 				}
483 				semu_try_free(suptr);
484 			}
485 			if (semnum != -1)
486 				break;
487 		}
488 	}
489 }
490 
491 static int
492 semvalid(int semid, struct semid_kernel *semakptr)
493 {
494 
495 	return ((semakptr->u.sem_perm.mode & SEM_ALLOC) == 0 ||
496 	    semakptr->u.sem_perm.seq != IPCID_TO_SEQ(semid) ? EINVAL : 0);
497 }
498 
499 /*
500  * Note that the user-mode half of this passes a union, not a pointer.
501  */
502 #ifndef _SYS_SYSPROTO_H_
503 struct __semctl_args {
504 	int	semid;
505 	int	semnum;
506 	int	cmd;
507 	union	semun *arg;
508 };
509 #endif
510 int
511 sys___semctl(struct thread *td, struct __semctl_args *uap)
512 {
513 	struct semid_ds dsbuf;
514 	union semun arg, semun;
515 	register_t rval;
516 	int error;
517 
518 	switch (uap->cmd) {
519 	case SEM_STAT:
520 	case IPC_SET:
521 	case IPC_STAT:
522 	case GETALL:
523 	case SETVAL:
524 	case SETALL:
525 		error = copyin(uap->arg, &arg, sizeof(arg));
526 		if (error)
527 			return (error);
528 		break;
529 	}
530 
531 	switch (uap->cmd) {
532 	case SEM_STAT:
533 	case IPC_STAT:
534 		semun.buf = &dsbuf;
535 		break;
536 	case IPC_SET:
537 		error = copyin(arg.buf, &dsbuf, sizeof(dsbuf));
538 		if (error)
539 			return (error);
540 		semun.buf = &dsbuf;
541 		break;
542 	case GETALL:
543 	case SETALL:
544 		semun.array = arg.array;
545 		break;
546 	case SETVAL:
547 		semun.val = arg.val;
548 		break;
549 	}
550 
551 	error = kern_semctl(td, uap->semid, uap->semnum, uap->cmd, &semun,
552 	    &rval);
553 	if (error)
554 		return (error);
555 
556 	switch (uap->cmd) {
557 	case SEM_STAT:
558 	case IPC_STAT:
559 		error = copyout(&dsbuf, arg.buf, sizeof(dsbuf));
560 		break;
561 	}
562 
563 	if (error == 0)
564 		td->td_retval[0] = rval;
565 	return (error);
566 }
567 
568 int
569 kern_semctl(struct thread *td, int semid, int semnum, int cmd,
570     union semun *arg, register_t *rval)
571 {
572 	u_short *array;
573 	struct ucred *cred = td->td_ucred;
574 	int i, error;
575 	struct semid_ds *sbuf;
576 	struct semid_kernel *semakptr;
577 	struct mtx *sema_mtxp;
578 	u_short usval, count;
579 	int semidx;
580 
581 	DPRINTF(("call to semctl(%d, %d, %d, 0x%p)\n",
582 	    semid, semnum, cmd, arg));
583 	if (!prison_allow(td->td_ucred, PR_ALLOW_SYSVIPC))
584 		return (ENOSYS);
585 
586 	array = NULL;
587 
588 	switch(cmd) {
589 	case SEM_STAT:
590 		/*
591 		 * For this command we assume semid is an array index
592 		 * rather than an IPC id.
593 		 */
594 		if (semid < 0 || semid >= seminfo.semmni)
595 			return (EINVAL);
596 		semakptr = &sema[semid];
597 		sema_mtxp = &sema_mtx[semid];
598 		mtx_lock(sema_mtxp);
599 		if ((semakptr->u.sem_perm.mode & SEM_ALLOC) == 0) {
600 			error = EINVAL;
601 			goto done2;
602 		}
603 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
604 			goto done2;
605 #ifdef MAC
606 		error = mac_sysvsem_check_semctl(cred, semakptr, cmd);
607 		if (error != 0)
608 			goto done2;
609 #endif
610 		bcopy(&semakptr->u, arg->buf, sizeof(struct semid_ds));
611 		*rval = IXSEQ_TO_IPCID(semid, semakptr->u.sem_perm);
612 		mtx_unlock(sema_mtxp);
613 		return (0);
614 	}
615 
616 	semidx = IPCID_TO_IX(semid);
617 	if (semidx < 0 || semidx >= seminfo.semmni)
618 		return (EINVAL);
619 
620 	semakptr = &sema[semidx];
621 	sema_mtxp = &sema_mtx[semidx];
622 	if (cmd == IPC_RMID)
623 		mtx_lock(&sem_mtx);
624 	mtx_lock(sema_mtxp);
625 #ifdef MAC
626 	error = mac_sysvsem_check_semctl(cred, semakptr, cmd);
627 	if (error != 0)
628 		goto done2;
629 #endif
630 
631 	error = 0;
632 	*rval = 0;
633 
634 	switch (cmd) {
635 	case IPC_RMID:
636 		if ((error = semvalid(semid, semakptr)) != 0)
637 			goto done2;
638 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_M)))
639 			goto done2;
640 		semakptr->u.sem_perm.cuid = cred->cr_uid;
641 		semakptr->u.sem_perm.uid = cred->cr_uid;
642 		semakptr->u.sem_perm.mode = 0;
643 		racct_sub_cred(semakptr->cred, RACCT_NSEM, semakptr->u.sem_nsems);
644 		crfree(semakptr->cred);
645 		semakptr->cred = NULL;
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 sys_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 ((semflg & IPC_CREAT) && (semflg & IPC_EXCL)) {
871 				DPRINTF(("not exclusive\n"));
872 				error = EEXIST;
873 				goto done2;
874 			}
875 			if ((error = ipcperm(td, &sema[semid].u.sem_perm,
876 			    semflg & 0700))) {
877 				goto done2;
878 			}
879 			if (nsems > 0 && sema[semid].u.sem_nsems < nsems) {
880 				DPRINTF(("too small\n"));
881 				error = EINVAL;
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 #ifdef RACCT
918 		if (racct_enable) {
919 			PROC_LOCK(td->td_proc);
920 			error = racct_add(td->td_proc, RACCT_NSEM, nsems);
921 			PROC_UNLOCK(td->td_proc);
922 			if (error != 0) {
923 				error = ENOSPC;
924 				goto done2;
925 			}
926 		}
927 #endif
928 		DPRINTF(("semid %d is available\n", semid));
929 		mtx_lock(&sema_mtx[semid]);
930 		KASSERT((sema[semid].u.sem_perm.mode & SEM_ALLOC) == 0,
931 		    ("Lost semaphore %d", semid));
932 		sema[semid].u.sem_perm.key = key;
933 		sema[semid].u.sem_perm.cuid = cred->cr_uid;
934 		sema[semid].u.sem_perm.uid = cred->cr_uid;
935 		sema[semid].u.sem_perm.cgid = cred->cr_gid;
936 		sema[semid].u.sem_perm.gid = cred->cr_gid;
937 		sema[semid].u.sem_perm.mode = (semflg & 0777) | SEM_ALLOC;
938 		sema[semid].cred = crhold(cred);
939 		sema[semid].u.sem_perm.seq =
940 		    (sema[semid].u.sem_perm.seq + 1) & 0x7fff;
941 		sema[semid].u.sem_nsems = nsems;
942 		sema[semid].u.sem_otime = 0;
943 		sema[semid].u.sem_ctime = time_second;
944 		sema[semid].u.sem_base = &sem[semtot];
945 		semtot += nsems;
946 		bzero(sema[semid].u.sem_base,
947 		    sizeof(sema[semid].u.sem_base[0])*nsems);
948 #ifdef MAC
949 		mac_sysvsem_create(cred, &sema[semid]);
950 #endif
951 		mtx_unlock(&sema_mtx[semid]);
952 		DPRINTF(("sembase = %p, next = %p\n",
953 		    sema[semid].u.sem_base, &sem[semtot]));
954 	} else {
955 		DPRINTF(("didn't find it and wasn't asked to create it\n"));
956 		error = ENOENT;
957 		goto done2;
958 	}
959 
960 found:
961 	td->td_retval[0] = IXSEQ_TO_IPCID(semid, sema[semid].u.sem_perm);
962 done2:
963 	mtx_unlock(&sem_mtx);
964 	return (error);
965 }
966 
967 #ifndef _SYS_SYSPROTO_H_
968 struct semop_args {
969 	int	semid;
970 	struct	sembuf *sops;
971 	size_t	nsops;
972 };
973 #endif
974 int
975 sys_semop(struct thread *td, struct semop_args *uap)
976 {
977 #define SMALL_SOPS	8
978 	struct sembuf small_sops[SMALL_SOPS];
979 	int semid = uap->semid;
980 	size_t nsops = uap->nsops;
981 	struct sembuf *sops;
982 	struct semid_kernel *semakptr;
983 	struct sembuf *sopptr = 0;
984 	struct sem *semptr = 0;
985 	struct sem_undo *suptr;
986 	struct mtx *sema_mtxp;
987 	size_t i, j, k;
988 	int error;
989 	int do_wakeup, do_undos;
990 	unsigned short seq;
991 
992 #ifdef SEM_DEBUG
993 	sops = NULL;
994 #endif
995 	DPRINTF(("call to semop(%d, %p, %u)\n", semid, sops, nsops));
996 
997 	if (!prison_allow(td->td_ucred, PR_ALLOW_SYSVIPC))
998 		return (ENOSYS);
999 
1000 	semid = IPCID_TO_IX(semid);	/* Convert back to zero origin */
1001 
1002 	if (semid < 0 || semid >= seminfo.semmni)
1003 		return (EINVAL);
1004 
1005 	/* Allocate memory for sem_ops */
1006 	if (nsops <= SMALL_SOPS)
1007 		sops = small_sops;
1008 	else if (nsops > seminfo.semopm) {
1009 		DPRINTF(("too many sops (max=%d, nsops=%d)\n", seminfo.semopm,
1010 		    nsops));
1011 		return (E2BIG);
1012 	} else {
1013 #ifdef RACCT
1014 		if (racct_enable) {
1015 			PROC_LOCK(td->td_proc);
1016 			if (nsops >
1017 			    racct_get_available(td->td_proc, RACCT_NSEMOP)) {
1018 				PROC_UNLOCK(td->td_proc);
1019 				return (E2BIG);
1020 			}
1021 			PROC_UNLOCK(td->td_proc);
1022 		}
1023 #endif
1024 
1025 		sops = malloc(nsops * sizeof(*sops), M_TEMP, M_WAITOK);
1026 	}
1027 	if ((error = copyin(uap->sops, sops, nsops * sizeof(sops[0]))) != 0) {
1028 		DPRINTF(("error = %d from copyin(%p, %p, %d)\n", error,
1029 		    uap->sops, sops, nsops * sizeof(sops[0])));
1030 		if (sops != small_sops)
1031 			free(sops, M_SEM);
1032 		return (error);
1033 	}
1034 
1035 	semakptr = &sema[semid];
1036 	sema_mtxp = &sema_mtx[semid];
1037 	mtx_lock(sema_mtxp);
1038 	if ((semakptr->u.sem_perm.mode & SEM_ALLOC) == 0) {
1039 		error = EINVAL;
1040 		goto done2;
1041 	}
1042 	seq = semakptr->u.sem_perm.seq;
1043 	if (seq != IPCID_TO_SEQ(uap->semid)) {
1044 		error = EINVAL;
1045 		goto done2;
1046 	}
1047 	/*
1048 	 * Initial pass thru sops to see what permissions are needed.
1049 	 * Also perform any checks that don't need repeating on each
1050 	 * attempt to satisfy the request vector.
1051 	 */
1052 	j = 0;		/* permission needed */
1053 	do_undos = 0;
1054 	for (i = 0; i < nsops; i++) {
1055 		sopptr = &sops[i];
1056 		if (sopptr->sem_num >= semakptr->u.sem_nsems) {
1057 			error = EFBIG;
1058 			goto done2;
1059 		}
1060 		if (sopptr->sem_flg & SEM_UNDO && sopptr->sem_op != 0)
1061 			do_undos = 1;
1062 		j |= (sopptr->sem_op == 0) ? SEM_R : SEM_A;
1063 	}
1064 
1065 	if ((error = ipcperm(td, &semakptr->u.sem_perm, j))) {
1066 		DPRINTF(("error = %d from ipaccess\n", error));
1067 		goto done2;
1068 	}
1069 #ifdef MAC
1070 	error = mac_sysvsem_check_semop(td->td_ucred, semakptr, j);
1071 	if (error != 0)
1072 		goto done2;
1073 #endif
1074 
1075 	/*
1076 	 * Loop trying to satisfy the vector of requests.
1077 	 * If we reach a point where we must wait, any requests already
1078 	 * performed are rolled back and we go to sleep until some other
1079 	 * process wakes us up.  At this point, we start all over again.
1080 	 *
1081 	 * This ensures that from the perspective of other tasks, a set
1082 	 * of requests is atomic (never partially satisfied).
1083 	 */
1084 	for (;;) {
1085 		do_wakeup = 0;
1086 		error = 0;	/* error return if necessary */
1087 
1088 		for (i = 0; i < nsops; i++) {
1089 			sopptr = &sops[i];
1090 			semptr = &semakptr->u.sem_base[sopptr->sem_num];
1091 
1092 			DPRINTF((
1093 			    "semop:  semakptr=%p, sem_base=%p, "
1094 			    "semptr=%p, sem[%d]=%d : op=%d, flag=%s\n",
1095 			    semakptr, semakptr->u.sem_base, semptr,
1096 			    sopptr->sem_num, semptr->semval, sopptr->sem_op,
1097 			    (sopptr->sem_flg & IPC_NOWAIT) ?
1098 			    "nowait" : "wait"));
1099 
1100 			if (sopptr->sem_op < 0) {
1101 				if (semptr->semval + sopptr->sem_op < 0) {
1102 					DPRINTF(("semop:  can't do it now\n"));
1103 					break;
1104 				} else {
1105 					semptr->semval += sopptr->sem_op;
1106 					if (semptr->semval == 0 &&
1107 					    semptr->semzcnt > 0)
1108 						do_wakeup = 1;
1109 				}
1110 			} else if (sopptr->sem_op == 0) {
1111 				if (semptr->semval != 0) {
1112 					DPRINTF(("semop:  not zero now\n"));
1113 					break;
1114 				}
1115 			} else if (semptr->semval + sopptr->sem_op >
1116 			    seminfo.semvmx) {
1117 				error = ERANGE;
1118 				break;
1119 			} else {
1120 				if (semptr->semncnt > 0)
1121 					do_wakeup = 1;
1122 				semptr->semval += sopptr->sem_op;
1123 			}
1124 		}
1125 
1126 		/*
1127 		 * Did we get through the entire vector?
1128 		 */
1129 		if (i >= nsops)
1130 			goto done;
1131 
1132 		/*
1133 		 * No ... rollback anything that we've already done
1134 		 */
1135 		DPRINTF(("semop:  rollback 0 through %d\n", i-1));
1136 		for (j = 0; j < i; j++)
1137 			semakptr->u.sem_base[sops[j].sem_num].semval -=
1138 			    sops[j].sem_op;
1139 
1140 		/* If we detected an error, return it */
1141 		if (error != 0)
1142 			goto done2;
1143 
1144 		/*
1145 		 * If the request that we couldn't satisfy has the
1146 		 * NOWAIT flag set then return with EAGAIN.
1147 		 */
1148 		if (sopptr->sem_flg & IPC_NOWAIT) {
1149 			error = EAGAIN;
1150 			goto done2;
1151 		}
1152 
1153 		if (sopptr->sem_op == 0)
1154 			semptr->semzcnt++;
1155 		else
1156 			semptr->semncnt++;
1157 
1158 		DPRINTF(("semop:  good night!\n"));
1159 		error = msleep(semakptr, sema_mtxp, (PZERO - 4) | PCATCH,
1160 		    "semwait", 0);
1161 		DPRINTF(("semop:  good morning (error=%d)!\n", error));
1162 		/* return code is checked below, after sem[nz]cnt-- */
1163 
1164 		/*
1165 		 * Make sure that the semaphore still exists
1166 		 */
1167 		seq = semakptr->u.sem_perm.seq;
1168 		if ((semakptr->u.sem_perm.mode & SEM_ALLOC) == 0 ||
1169 		    seq != IPCID_TO_SEQ(uap->semid)) {
1170 			error = EIDRM;
1171 			goto done2;
1172 		}
1173 
1174 		/*
1175 		 * Renew the semaphore's pointer after wakeup since
1176 		 * during msleep sem_base may have been modified and semptr
1177 		 * is not valid any more
1178 		 */
1179 		semptr = &semakptr->u.sem_base[sopptr->sem_num];
1180 
1181 		/*
1182 		 * The semaphore is still alive.  Readjust the count of
1183 		 * waiting processes.
1184 		 */
1185 		if (sopptr->sem_op == 0)
1186 			semptr->semzcnt--;
1187 		else
1188 			semptr->semncnt--;
1189 
1190 		/*
1191 		 * Is it really morning, or was our sleep interrupted?
1192 		 * (Delayed check of msleep() return code because we
1193 		 * need to decrement sem[nz]cnt either way.)
1194 		 */
1195 		if (error != 0) {
1196 			error = EINTR;
1197 			goto done2;
1198 		}
1199 		DPRINTF(("semop:  good morning!\n"));
1200 	}
1201 
1202 done:
1203 	/*
1204 	 * Process any SEM_UNDO requests.
1205 	 */
1206 	if (do_undos) {
1207 		SEMUNDO_LOCK();
1208 		suptr = NULL;
1209 		for (i = 0; i < nsops; i++) {
1210 			/*
1211 			 * We only need to deal with SEM_UNDO's for non-zero
1212 			 * op's.
1213 			 */
1214 			int adjval;
1215 
1216 			if ((sops[i].sem_flg & SEM_UNDO) == 0)
1217 				continue;
1218 			adjval = sops[i].sem_op;
1219 			if (adjval == 0)
1220 				continue;
1221 			error = semundo_adjust(td, &suptr, semid, seq,
1222 			    sops[i].sem_num, -adjval);
1223 			if (error == 0)
1224 				continue;
1225 
1226 			/*
1227 			 * Oh-Oh!  We ran out of either sem_undo's or undo's.
1228 			 * Rollback the adjustments to this point and then
1229 			 * rollback the semaphore ups and down so we can return
1230 			 * with an error with all structures restored.  We
1231 			 * rollback the undo's in the exact reverse order that
1232 			 * we applied them.  This guarantees that we won't run
1233 			 * out of space as we roll things back out.
1234 			 */
1235 			for (j = 0; j < i; j++) {
1236 				k = i - j - 1;
1237 				if ((sops[k].sem_flg & SEM_UNDO) == 0)
1238 					continue;
1239 				adjval = sops[k].sem_op;
1240 				if (adjval == 0)
1241 					continue;
1242 				if (semundo_adjust(td, &suptr, semid, seq,
1243 				    sops[k].sem_num, adjval) != 0)
1244 					panic("semop - can't undo undos");
1245 			}
1246 
1247 			for (j = 0; j < nsops; j++)
1248 				semakptr->u.sem_base[sops[j].sem_num].semval -=
1249 				    sops[j].sem_op;
1250 
1251 			DPRINTF(("error = %d from semundo_adjust\n", error));
1252 			SEMUNDO_UNLOCK();
1253 			goto done2;
1254 		} /* loop through the sops */
1255 		SEMUNDO_UNLOCK();
1256 	} /* if (do_undos) */
1257 
1258 	/* We're definitely done - set the sempid's and time */
1259 	for (i = 0; i < nsops; i++) {
1260 		sopptr = &sops[i];
1261 		semptr = &semakptr->u.sem_base[sopptr->sem_num];
1262 		semptr->sempid = td->td_proc->p_pid;
1263 	}
1264 	semakptr->u.sem_otime = time_second;
1265 
1266 	/*
1267 	 * Do a wakeup if any semaphore was up'd whilst something was
1268 	 * sleeping on it.
1269 	 */
1270 	if (do_wakeup) {
1271 		DPRINTF(("semop:  doing wakeup\n"));
1272 		wakeup(semakptr);
1273 		DPRINTF(("semop:  back from wakeup\n"));
1274 	}
1275 	DPRINTF(("semop:  done\n"));
1276 	td->td_retval[0] = 0;
1277 done2:
1278 	mtx_unlock(sema_mtxp);
1279 	if (sops != small_sops)
1280 		free(sops, M_SEM);
1281 	return (error);
1282 }
1283 
1284 /*
1285  * Go through the undo structures for this process and apply the adjustments to
1286  * semaphores.
1287  */
1288 static void
1289 semexit_myhook(void *arg, struct proc *p)
1290 {
1291 	struct sem_undo *suptr;
1292 	struct semid_kernel *semakptr;
1293 	struct mtx *sema_mtxp;
1294 	int semid, semnum, adjval, ix;
1295 	unsigned short seq;
1296 
1297 	/*
1298 	 * Go through the chain of undo vectors looking for one
1299 	 * associated with this process.
1300 	 */
1301 	SEMUNDO_LOCK();
1302 	LIST_FOREACH(suptr, &semu_list, un_next) {
1303 		if (suptr->un_proc == p)
1304 			break;
1305 	}
1306 	if (suptr == NULL) {
1307 		SEMUNDO_UNLOCK();
1308 		return;
1309 	}
1310 	LIST_REMOVE(suptr, un_next);
1311 
1312 	DPRINTF(("proc @%p has undo structure with %d entries\n", p,
1313 	    suptr->un_cnt));
1314 
1315 	/*
1316 	 * If there are any active undo elements then process them.
1317 	 */
1318 	if (suptr->un_cnt > 0) {
1319 		SEMUNDO_UNLOCK();
1320 		for (ix = 0; ix < suptr->un_cnt; ix++) {
1321 			semid = suptr->un_ent[ix].un_id;
1322 			semnum = suptr->un_ent[ix].un_num;
1323 			adjval = suptr->un_ent[ix].un_adjval;
1324 			seq = suptr->un_ent[ix].un_seq;
1325 			semakptr = &sema[semid];
1326 			sema_mtxp = &sema_mtx[semid];
1327 
1328 			mtx_lock(sema_mtxp);
1329 			if ((semakptr->u.sem_perm.mode & SEM_ALLOC) == 0 ||
1330 			    (semakptr->u.sem_perm.seq != seq)) {
1331 				mtx_unlock(sema_mtxp);
1332 				continue;
1333 			}
1334 			if (semnum >= semakptr->u.sem_nsems)
1335 				panic("semexit - semnum out of range");
1336 
1337 			DPRINTF((
1338 			    "semexit:  %p id=%d num=%d(adj=%d) ; sem=%d\n",
1339 			    suptr->un_proc, suptr->un_ent[ix].un_id,
1340 			    suptr->un_ent[ix].un_num,
1341 			    suptr->un_ent[ix].un_adjval,
1342 			    semakptr->u.sem_base[semnum].semval));
1343 
1344 			if (adjval < 0 && semakptr->u.sem_base[semnum].semval <
1345 			    -adjval)
1346 				semakptr->u.sem_base[semnum].semval = 0;
1347 			else
1348 				semakptr->u.sem_base[semnum].semval += adjval;
1349 
1350 			wakeup(semakptr);
1351 			DPRINTF(("semexit:  back from wakeup\n"));
1352 			mtx_unlock(sema_mtxp);
1353 		}
1354 		SEMUNDO_LOCK();
1355 	}
1356 
1357 	/*
1358 	 * Deallocate the undo vector.
1359 	 */
1360 	DPRINTF(("removing vector\n"));
1361 	suptr->un_proc = NULL;
1362 	suptr->un_cnt = 0;
1363 	LIST_INSERT_HEAD(&semu_free_list, suptr, un_next);
1364 	SEMUNDO_UNLOCK();
1365 }
1366 
1367 static int
1368 sysctl_sema(SYSCTL_HANDLER_ARGS)
1369 {
1370 
1371 	return (SYSCTL_OUT(req, sema,
1372 	    sizeof(struct semid_kernel) * seminfo.semmni));
1373 }
1374 
1375 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
1376     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
1377 
1378 /* XXX casting to (sy_call_t *) is bogus, as usual. */
1379 static sy_call_t *semcalls[] = {
1380 	(sy_call_t *)freebsd7___semctl, (sy_call_t *)sys_semget,
1381 	(sy_call_t *)sys_semop
1382 };
1383 
1384 /*
1385  * Entry point for all SEM calls.
1386  */
1387 int
1388 sys_semsys(td, uap)
1389 	struct thread *td;
1390 	/* XXX actually varargs. */
1391 	struct semsys_args /* {
1392 		int	which;
1393 		int	a2;
1394 		int	a3;
1395 		int	a4;
1396 		int	a5;
1397 	} */ *uap;
1398 {
1399 	int error;
1400 
1401 	if (!prison_allow(td->td_ucred, PR_ALLOW_SYSVIPC))
1402 		return (ENOSYS);
1403 	if (uap->which < 0 ||
1404 	    uap->which >= sizeof(semcalls)/sizeof(semcalls[0]))
1405 		return (EINVAL);
1406 	error = (*semcalls[uap->which])(td, &uap->a2);
1407 	return (error);
1408 }
1409 
1410 #ifndef CP
1411 #define CP(src, dst, fld)	do { (dst).fld = (src).fld; } while (0)
1412 #endif
1413 
1414 #ifndef _SYS_SYSPROTO_H_
1415 struct freebsd7___semctl_args {
1416 	int	semid;
1417 	int	semnum;
1418 	int	cmd;
1419 	union	semun_old *arg;
1420 };
1421 #endif
1422 int
1423 freebsd7___semctl(struct thread *td, struct freebsd7___semctl_args *uap)
1424 {
1425 	struct semid_ds_old dsold;
1426 	struct semid_ds dsbuf;
1427 	union semun_old arg;
1428 	union semun semun;
1429 	register_t rval;
1430 	int error;
1431 
1432 	switch (uap->cmd) {
1433 	case SEM_STAT:
1434 	case IPC_SET:
1435 	case IPC_STAT:
1436 	case GETALL:
1437 	case SETVAL:
1438 	case SETALL:
1439 		error = copyin(uap->arg, &arg, sizeof(arg));
1440 		if (error)
1441 			return (error);
1442 		break;
1443 	}
1444 
1445 	switch (uap->cmd) {
1446 	case SEM_STAT:
1447 	case IPC_STAT:
1448 		semun.buf = &dsbuf;
1449 		break;
1450 	case IPC_SET:
1451 		error = copyin(arg.buf, &dsold, sizeof(dsold));
1452 		if (error)
1453 			return (error);
1454 		ipcperm_old2new(&dsold.sem_perm, &dsbuf.sem_perm);
1455 		CP(dsold, dsbuf, sem_base);
1456 		CP(dsold, dsbuf, sem_nsems);
1457 		CP(dsold, dsbuf, sem_otime);
1458 		CP(dsold, dsbuf, sem_ctime);
1459 		semun.buf = &dsbuf;
1460 		break;
1461 	case GETALL:
1462 	case SETALL:
1463 		semun.array = arg.array;
1464 		break;
1465 	case SETVAL:
1466 		semun.val = arg.val;
1467 		break;
1468 	}
1469 
1470 	error = kern_semctl(td, uap->semid, uap->semnum, uap->cmd, &semun,
1471 	    &rval);
1472 	if (error)
1473 		return (error);
1474 
1475 	switch (uap->cmd) {
1476 	case SEM_STAT:
1477 	case IPC_STAT:
1478 		bzero(&dsold, sizeof(dsold));
1479 		ipcperm_new2old(&dsbuf.sem_perm, &dsold.sem_perm);
1480 		CP(dsbuf, dsold, sem_base);
1481 		CP(dsbuf, dsold, sem_nsems);
1482 		CP(dsbuf, dsold, sem_otime);
1483 		CP(dsbuf, dsold, sem_ctime);
1484 		error = copyout(&dsold, arg.buf, sizeof(dsold));
1485 		break;
1486 	}
1487 
1488 	if (error == 0)
1489 		td->td_retval[0] = rval;
1490 	return (error);
1491 }
1492 
1493 #endif /* COMPAT_FREEBSD{4,5,6,7} */
1494 
1495 #ifdef COMPAT_FREEBSD32
1496 
1497 int
1498 freebsd32_semsys(struct thread *td, struct freebsd32_semsys_args *uap)
1499 {
1500 
1501 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
1502     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
1503 	switch (uap->which) {
1504 	case 0:
1505 		return (freebsd7_freebsd32_semctl(td,
1506 		    (struct freebsd7_freebsd32_semctl_args *)&uap->a2));
1507 	default:
1508 		return (sys_semsys(td, (struct semsys_args *)uap));
1509 	}
1510 #else
1511 	return (nosys(td, NULL));
1512 #endif
1513 }
1514 
1515 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
1516     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
1517 int
1518 freebsd7_freebsd32_semctl(struct thread *td,
1519     struct freebsd7_freebsd32_semctl_args *uap)
1520 {
1521 	struct semid_ds32_old dsbuf32;
1522 	struct semid_ds dsbuf;
1523 	union semun semun;
1524 	union semun32 arg;
1525 	register_t rval;
1526 	int error;
1527 
1528 	switch (uap->cmd) {
1529 	case SEM_STAT:
1530 	case IPC_SET:
1531 	case IPC_STAT:
1532 	case GETALL:
1533 	case SETVAL:
1534 	case SETALL:
1535 		error = copyin(uap->arg, &arg, sizeof(arg));
1536 		if (error)
1537 			return (error);
1538 		break;
1539 	}
1540 
1541 	switch (uap->cmd) {
1542 	case SEM_STAT:
1543 	case IPC_STAT:
1544 		semun.buf = &dsbuf;
1545 		break;
1546 	case IPC_SET:
1547 		error = copyin(PTRIN(arg.buf), &dsbuf32, sizeof(dsbuf32));
1548 		if (error)
1549 			return (error);
1550 		freebsd32_ipcperm_old_in(&dsbuf32.sem_perm, &dsbuf.sem_perm);
1551 		PTRIN_CP(dsbuf32, dsbuf, sem_base);
1552 		CP(dsbuf32, dsbuf, sem_nsems);
1553 		CP(dsbuf32, dsbuf, sem_otime);
1554 		CP(dsbuf32, dsbuf, sem_ctime);
1555 		semun.buf = &dsbuf;
1556 		break;
1557 	case GETALL:
1558 	case SETALL:
1559 		semun.array = PTRIN(arg.array);
1560 		break;
1561 	case SETVAL:
1562 		semun.val = arg.val;
1563 		break;
1564 	}
1565 
1566 	error = kern_semctl(td, uap->semid, uap->semnum, uap->cmd, &semun,
1567 	    &rval);
1568 	if (error)
1569 		return (error);
1570 
1571 	switch (uap->cmd) {
1572 	case SEM_STAT:
1573 	case IPC_STAT:
1574 		bzero(&dsbuf32, sizeof(dsbuf32));
1575 		freebsd32_ipcperm_old_out(&dsbuf.sem_perm, &dsbuf32.sem_perm);
1576 		PTROUT_CP(dsbuf, dsbuf32, sem_base);
1577 		CP(dsbuf, dsbuf32, sem_nsems);
1578 		CP(dsbuf, dsbuf32, sem_otime);
1579 		CP(dsbuf, dsbuf32, sem_ctime);
1580 		error = copyout(&dsbuf32, PTRIN(arg.buf), sizeof(dsbuf32));
1581 		break;
1582 	}
1583 
1584 	if (error == 0)
1585 		td->td_retval[0] = rval;
1586 	return (error);
1587 }
1588 #endif
1589 
1590 int
1591 freebsd32_semctl(struct thread *td, struct freebsd32_semctl_args *uap)
1592 {
1593 	struct semid_ds32 dsbuf32;
1594 	struct semid_ds dsbuf;
1595 	union semun semun;
1596 	union semun32 arg;
1597 	register_t rval;
1598 	int error;
1599 
1600 	switch (uap->cmd) {
1601 	case SEM_STAT:
1602 	case IPC_SET:
1603 	case IPC_STAT:
1604 	case GETALL:
1605 	case SETVAL:
1606 	case SETALL:
1607 		error = copyin(uap->arg, &arg, sizeof(arg));
1608 		if (error)
1609 			return (error);
1610 		break;
1611 	}
1612 
1613 	switch (uap->cmd) {
1614 	case SEM_STAT:
1615 	case IPC_STAT:
1616 		semun.buf = &dsbuf;
1617 		break;
1618 	case IPC_SET:
1619 		error = copyin(PTRIN(arg.buf), &dsbuf32, sizeof(dsbuf32));
1620 		if (error)
1621 			return (error);
1622 		freebsd32_ipcperm_in(&dsbuf32.sem_perm, &dsbuf.sem_perm);
1623 		PTRIN_CP(dsbuf32, dsbuf, sem_base);
1624 		CP(dsbuf32, dsbuf, sem_nsems);
1625 		CP(dsbuf32, dsbuf, sem_otime);
1626 		CP(dsbuf32, dsbuf, sem_ctime);
1627 		semun.buf = &dsbuf;
1628 		break;
1629 	case GETALL:
1630 	case SETALL:
1631 		semun.array = PTRIN(arg.array);
1632 		break;
1633 	case SETVAL:
1634 		semun.val = arg.val;
1635 		break;
1636 	}
1637 
1638 	error = kern_semctl(td, uap->semid, uap->semnum, uap->cmd, &semun,
1639 	    &rval);
1640 	if (error)
1641 		return (error);
1642 
1643 	switch (uap->cmd) {
1644 	case SEM_STAT:
1645 	case IPC_STAT:
1646 		bzero(&dsbuf32, sizeof(dsbuf32));
1647 		freebsd32_ipcperm_out(&dsbuf.sem_perm, &dsbuf32.sem_perm);
1648 		PTROUT_CP(dsbuf, dsbuf32, sem_base);
1649 		CP(dsbuf, dsbuf32, sem_nsems);
1650 		CP(dsbuf, dsbuf32, sem_otime);
1651 		CP(dsbuf, dsbuf32, sem_ctime);
1652 		error = copyout(&dsbuf32, PTRIN(arg.buf), sizeof(dsbuf32));
1653 		break;
1654 	}
1655 
1656 	if (error == 0)
1657 		td->td_retval[0] = rval;
1658 	return (error);
1659 }
1660 
1661 #endif /* COMPAT_FREEBSD32 */
1662