xref: /freebsd/sys/kern/sysv_sem.c (revision c6ec7d31830ab1c80edae95ad5e4b9dba10c47ac)
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_RW, &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_RW, &seminfo.semvmx, 0,
208     "Semaphore maximum value");
209 SYSCTL_INT(_kern_ipc, OID_AUTO, semaem, CTLFLAG_RW, &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 	TUNABLE_INT_FETCH("kern.ipc.semmni", &seminfo.semmni);
253 	TUNABLE_INT_FETCH("kern.ipc.semmns", &seminfo.semmns);
254 	TUNABLE_INT_FETCH("kern.ipc.semmnu", &seminfo.semmnu);
255 	TUNABLE_INT_FETCH("kern.ipc.semmsl", &seminfo.semmsl);
256 	TUNABLE_INT_FETCH("kern.ipc.semopm", &seminfo.semopm);
257 	TUNABLE_INT_FETCH("kern.ipc.semume", &seminfo.semume);
258 	TUNABLE_INT_FETCH("kern.ipc.semusz", &seminfo.semusz);
259 	TUNABLE_INT_FETCH("kern.ipc.semvmx", &seminfo.semvmx);
260 	TUNABLE_INT_FETCH("kern.ipc.semaem", &seminfo.semaem);
261 
262 	sem = malloc(sizeof(struct sem) * seminfo.semmns, M_SEM, M_WAITOK);
263 	sema = malloc(sizeof(struct semid_kernel) * seminfo.semmni, M_SEM,
264 	    M_WAITOK);
265 	sema_mtx = malloc(sizeof(struct mtx) * seminfo.semmni, M_SEM,
266 	    M_WAITOK | M_ZERO);
267 	semu = malloc(seminfo.semmnu * seminfo.semusz, M_SEM, M_WAITOK);
268 
269 	for (i = 0; i < seminfo.semmni; i++) {
270 		sema[i].u.sem_base = 0;
271 		sema[i].u.sem_perm.mode = 0;
272 		sema[i].u.sem_perm.seq = 0;
273 #ifdef MAC
274 		mac_sysvsem_init(&sema[i]);
275 #endif
276 	}
277 	for (i = 0; i < seminfo.semmni; i++)
278 		mtx_init(&sema_mtx[i], "semid", NULL, MTX_DEF);
279 	LIST_INIT(&semu_free_list);
280 	for (i = 0; i < seminfo.semmnu; i++) {
281 		struct sem_undo *suptr = SEMU(i);
282 		suptr->un_proc = NULL;
283 		LIST_INSERT_HEAD(&semu_free_list, suptr, un_next);
284 	}
285 	LIST_INIT(&semu_list);
286 	mtx_init(&sem_mtx, "sem", NULL, MTX_DEF);
287 	mtx_init(&sem_undo_mtx, "semu", NULL, MTX_DEF);
288 	semexit_tag = EVENTHANDLER_REGISTER(process_exit, semexit_myhook, NULL,
289 	    EVENTHANDLER_PRI_ANY);
290 
291 	error = syscall_helper_register(sem_syscalls);
292 	if (error != 0)
293 		return (error);
294 #ifdef COMPAT_FREEBSD32
295 	error = syscall32_helper_register(sem32_syscalls);
296 	if (error != 0)
297 		return (error);
298 #endif
299 	return (0);
300 }
301 
302 static int
303 semunload(void)
304 {
305 	int i;
306 
307 	/* XXXKIB */
308 	if (semtot != 0)
309 		return (EBUSY);
310 
311 #ifdef COMPAT_FREEBSD32
312 	syscall32_helper_unregister(sem32_syscalls);
313 #endif
314 	syscall_helper_unregister(sem_syscalls);
315 	EVENTHANDLER_DEREGISTER(process_exit, semexit_tag);
316 #ifdef MAC
317 	for (i = 0; i < seminfo.semmni; i++)
318 		mac_sysvsem_destroy(&sema[i]);
319 #endif
320 	free(sem, M_SEM);
321 	free(sema, M_SEM);
322 	free(semu, M_SEM);
323 	for (i = 0; i < seminfo.semmni; i++)
324 		mtx_destroy(&sema_mtx[i]);
325 	free(sema_mtx, M_SEM);
326 	mtx_destroy(&sem_mtx);
327 	mtx_destroy(&sem_undo_mtx);
328 	return (0);
329 }
330 
331 static int
332 sysvsem_modload(struct module *module, int cmd, void *arg)
333 {
334 	int error = 0;
335 
336 	switch (cmd) {
337 	case MOD_LOAD:
338 		error = seminit();
339 		if (error != 0)
340 			semunload();
341 		break;
342 	case MOD_UNLOAD:
343 		error = semunload();
344 		break;
345 	case MOD_SHUTDOWN:
346 		break;
347 	default:
348 		error = EINVAL;
349 		break;
350 	}
351 	return (error);
352 }
353 
354 static moduledata_t sysvsem_mod = {
355 	"sysvsem",
356 	&sysvsem_modload,
357 	NULL
358 };
359 
360 DECLARE_MODULE(sysvsem, sysvsem_mod, SI_SUB_SYSV_SEM, SI_ORDER_FIRST);
361 MODULE_VERSION(sysvsem, 1);
362 
363 /*
364  * Allocate a new sem_undo structure for a process
365  * (returns ptr to structure or NULL if no more room)
366  */
367 
368 static struct sem_undo *
369 semu_alloc(struct thread *td)
370 {
371 	struct sem_undo *suptr;
372 
373 	SEMUNDO_LOCKASSERT(MA_OWNED);
374 	if ((suptr = LIST_FIRST(&semu_free_list)) == NULL)
375 		return (NULL);
376 	LIST_REMOVE(suptr, un_next);
377 	LIST_INSERT_HEAD(&semu_list, suptr, un_next);
378 	suptr->un_cnt = 0;
379 	suptr->un_proc = td->td_proc;
380 	return (suptr);
381 }
382 
383 static int
384 semu_try_free(struct sem_undo *suptr)
385 {
386 
387 	SEMUNDO_LOCKASSERT(MA_OWNED);
388 
389 	if (suptr->un_cnt != 0)
390 		return (0);
391 	LIST_REMOVE(suptr, un_next);
392 	LIST_INSERT_HEAD(&semu_free_list, suptr, un_next);
393 	return (1);
394 }
395 
396 /*
397  * Adjust a particular entry for a particular proc
398  */
399 
400 static int
401 semundo_adjust(struct thread *td, struct sem_undo **supptr, int semid,
402     int semseq, int semnum, int adjval)
403 {
404 	struct proc *p = td->td_proc;
405 	struct sem_undo *suptr;
406 	struct undo *sunptr;
407 	int i;
408 
409 	SEMUNDO_LOCKASSERT(MA_OWNED);
410 	/* Look for and remember the sem_undo if the caller doesn't provide
411 	   it */
412 
413 	suptr = *supptr;
414 	if (suptr == NULL) {
415 		LIST_FOREACH(suptr, &semu_list, un_next) {
416 			if (suptr->un_proc == p) {
417 				*supptr = suptr;
418 				break;
419 			}
420 		}
421 		if (suptr == NULL) {
422 			if (adjval == 0)
423 				return(0);
424 			suptr = semu_alloc(td);
425 			if (suptr == NULL)
426 				return (ENOSPC);
427 			*supptr = suptr;
428 		}
429 	}
430 
431 	/*
432 	 * Look for the requested entry and adjust it (delete if adjval becomes
433 	 * 0).
434 	 */
435 	sunptr = &suptr->un_ent[0];
436 	for (i = 0; i < suptr->un_cnt; i++, sunptr++) {
437 		if (sunptr->un_id != semid || sunptr->un_num != semnum)
438 			continue;
439 		if (adjval != 0) {
440 			adjval += sunptr->un_adjval;
441 			if (adjval > seminfo.semaem || adjval < -seminfo.semaem)
442 				return (ERANGE);
443 		}
444 		sunptr->un_adjval = adjval;
445 		if (sunptr->un_adjval == 0) {
446 			suptr->un_cnt--;
447 			if (i < suptr->un_cnt)
448 				suptr->un_ent[i] =
449 				    suptr->un_ent[suptr->un_cnt];
450 			if (suptr->un_cnt == 0)
451 				semu_try_free(suptr);
452 		}
453 		return (0);
454 	}
455 
456 	/* Didn't find the right entry - create it */
457 	if (adjval == 0)
458 		return (0);
459 	if (adjval > seminfo.semaem || adjval < -seminfo.semaem)
460 		return (ERANGE);
461 	if (suptr->un_cnt != seminfo.semume) {
462 		sunptr = &suptr->un_ent[suptr->un_cnt];
463 		suptr->un_cnt++;
464 		sunptr->un_adjval = adjval;
465 		sunptr->un_id = semid;
466 		sunptr->un_num = semnum;
467 		sunptr->un_seq = semseq;
468 	} else
469 		return (EINVAL);
470 	return (0);
471 }
472 
473 static void
474 semundo_clear(int semid, int semnum)
475 {
476 	struct sem_undo *suptr, *suptr1;
477 	struct undo *sunptr;
478 	int i;
479 
480 	SEMUNDO_LOCKASSERT(MA_OWNED);
481 	LIST_FOREACH_SAFE(suptr, &semu_list, un_next, suptr1) {
482 		sunptr = &suptr->un_ent[0];
483 		for (i = 0; i < suptr->un_cnt; i++, sunptr++) {
484 			if (sunptr->un_id != semid)
485 				continue;
486 			if (semnum == -1 || sunptr->un_num == semnum) {
487 				suptr->un_cnt--;
488 				if (i < suptr->un_cnt) {
489 					suptr->un_ent[i] =
490 					    suptr->un_ent[suptr->un_cnt];
491 					continue;
492 				}
493 				semu_try_free(suptr);
494 			}
495 			if (semnum != -1)
496 				break;
497 		}
498 	}
499 }
500 
501 static int
502 semvalid(int semid, struct semid_kernel *semakptr)
503 {
504 
505 	return ((semakptr->u.sem_perm.mode & SEM_ALLOC) == 0 ||
506 	    semakptr->u.sem_perm.seq != IPCID_TO_SEQ(semid) ? EINVAL : 0);
507 }
508 
509 /*
510  * Note that the user-mode half of this passes a union, not a pointer.
511  */
512 #ifndef _SYS_SYSPROTO_H_
513 struct __semctl_args {
514 	int	semid;
515 	int	semnum;
516 	int	cmd;
517 	union	semun *arg;
518 };
519 #endif
520 int
521 sys___semctl(struct thread *td, struct __semctl_args *uap)
522 {
523 	struct semid_ds dsbuf;
524 	union semun arg, semun;
525 	register_t rval;
526 	int error;
527 
528 	switch (uap->cmd) {
529 	case SEM_STAT:
530 	case IPC_SET:
531 	case IPC_STAT:
532 	case GETALL:
533 	case SETVAL:
534 	case SETALL:
535 		error = copyin(uap->arg, &arg, sizeof(arg));
536 		if (error)
537 			return (error);
538 		break;
539 	}
540 
541 	switch (uap->cmd) {
542 	case SEM_STAT:
543 	case IPC_STAT:
544 		semun.buf = &dsbuf;
545 		break;
546 	case IPC_SET:
547 		error = copyin(arg.buf, &dsbuf, sizeof(dsbuf));
548 		if (error)
549 			return (error);
550 		semun.buf = &dsbuf;
551 		break;
552 	case GETALL:
553 	case SETALL:
554 		semun.array = arg.array;
555 		break;
556 	case SETVAL:
557 		semun.val = arg.val;
558 		break;
559 	}
560 
561 	error = kern_semctl(td, uap->semid, uap->semnum, uap->cmd, &semun,
562 	    &rval);
563 	if (error)
564 		return (error);
565 
566 	switch (uap->cmd) {
567 	case SEM_STAT:
568 	case IPC_STAT:
569 		error = copyout(&dsbuf, arg.buf, sizeof(dsbuf));
570 		break;
571 	}
572 
573 	if (error == 0)
574 		td->td_retval[0] = rval;
575 	return (error);
576 }
577 
578 int
579 kern_semctl(struct thread *td, int semid, int semnum, int cmd,
580     union semun *arg, register_t *rval)
581 {
582 	u_short *array;
583 	struct ucred *cred = td->td_ucred;
584 	int i, error;
585 	struct semid_ds *sbuf;
586 	struct semid_kernel *semakptr;
587 	struct mtx *sema_mtxp;
588 	u_short usval, count;
589 	int semidx;
590 
591 	DPRINTF(("call to semctl(%d, %d, %d, 0x%p)\n",
592 	    semid, semnum, cmd, arg));
593 	if (!prison_allow(td->td_ucred, PR_ALLOW_SYSVIPC))
594 		return (ENOSYS);
595 
596 	array = NULL;
597 
598 	switch(cmd) {
599 	case SEM_STAT:
600 		/*
601 		 * For this command we assume semid is an array index
602 		 * rather than an IPC id.
603 		 */
604 		if (semid < 0 || semid >= seminfo.semmni)
605 			return (EINVAL);
606 		semakptr = &sema[semid];
607 		sema_mtxp = &sema_mtx[semid];
608 		mtx_lock(sema_mtxp);
609 		if ((semakptr->u.sem_perm.mode & SEM_ALLOC) == 0) {
610 			error = EINVAL;
611 			goto done2;
612 		}
613 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
614 			goto done2;
615 #ifdef MAC
616 		error = mac_sysvsem_check_semctl(cred, semakptr, cmd);
617 		if (error != 0)
618 			goto done2;
619 #endif
620 		bcopy(&semakptr->u, arg->buf, sizeof(struct semid_ds));
621 		*rval = IXSEQ_TO_IPCID(semid, semakptr->u.sem_perm);
622 		mtx_unlock(sema_mtxp);
623 		return (0);
624 	}
625 
626 	semidx = IPCID_TO_IX(semid);
627 	if (semidx < 0 || semidx >= seminfo.semmni)
628 		return (EINVAL);
629 
630 	semakptr = &sema[semidx];
631 	sema_mtxp = &sema_mtx[semidx];
632 	if (cmd == IPC_RMID)
633 		mtx_lock(&sem_mtx);
634 	mtx_lock(sema_mtxp);
635 #ifdef MAC
636 	error = mac_sysvsem_check_semctl(cred, semakptr, cmd);
637 	if (error != 0)
638 		goto done2;
639 #endif
640 
641 	error = 0;
642 	*rval = 0;
643 
644 	switch (cmd) {
645 	case IPC_RMID:
646 		if ((error = semvalid(semid, semakptr)) != 0)
647 			goto done2;
648 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_M)))
649 			goto done2;
650 		semakptr->u.sem_perm.cuid = cred->cr_uid;
651 		semakptr->u.sem_perm.uid = cred->cr_uid;
652 		semakptr->u.sem_perm.mode = 0;
653 		racct_sub_cred(semakptr->cred, RACCT_NSEM, semakptr->u.sem_nsems);
654 		crfree(semakptr->cred);
655 		semakptr->cred = NULL;
656 		SEMUNDO_LOCK();
657 		semundo_clear(semidx, -1);
658 		SEMUNDO_UNLOCK();
659 #ifdef MAC
660 		mac_sysvsem_cleanup(semakptr);
661 #endif
662 		wakeup(semakptr);
663 		for (i = 0; i < seminfo.semmni; i++) {
664 			if ((sema[i].u.sem_perm.mode & SEM_ALLOC) &&
665 			    sema[i].u.sem_base > semakptr->u.sem_base)
666 				mtx_lock_flags(&sema_mtx[i], LOP_DUPOK);
667 		}
668 		for (i = semakptr->u.sem_base - sem; i < semtot; i++)
669 			sem[i] = sem[i + semakptr->u.sem_nsems];
670 		for (i = 0; i < seminfo.semmni; i++) {
671 			if ((sema[i].u.sem_perm.mode & SEM_ALLOC) &&
672 			    sema[i].u.sem_base > semakptr->u.sem_base) {
673 				sema[i].u.sem_base -= semakptr->u.sem_nsems;
674 				mtx_unlock(&sema_mtx[i]);
675 			}
676 		}
677 		semtot -= semakptr->u.sem_nsems;
678 		break;
679 
680 	case IPC_SET:
681 		if ((error = semvalid(semid, semakptr)) != 0)
682 			goto done2;
683 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_M)))
684 			goto done2;
685 		sbuf = arg->buf;
686 		semakptr->u.sem_perm.uid = sbuf->sem_perm.uid;
687 		semakptr->u.sem_perm.gid = sbuf->sem_perm.gid;
688 		semakptr->u.sem_perm.mode = (semakptr->u.sem_perm.mode &
689 		    ~0777) | (sbuf->sem_perm.mode & 0777);
690 		semakptr->u.sem_ctime = time_second;
691 		break;
692 
693 	case IPC_STAT:
694 		if ((error = semvalid(semid, semakptr)) != 0)
695 			goto done2;
696 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
697 			goto done2;
698 		bcopy(&semakptr->u, arg->buf, sizeof(struct semid_ds));
699 		break;
700 
701 	case GETNCNT:
702 		if ((error = semvalid(semid, semakptr)) != 0)
703 			goto done2;
704 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
705 			goto done2;
706 		if (semnum < 0 || semnum >= semakptr->u.sem_nsems) {
707 			error = EINVAL;
708 			goto done2;
709 		}
710 		*rval = semakptr->u.sem_base[semnum].semncnt;
711 		break;
712 
713 	case GETPID:
714 		if ((error = semvalid(semid, semakptr)) != 0)
715 			goto done2;
716 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
717 			goto done2;
718 		if (semnum < 0 || semnum >= semakptr->u.sem_nsems) {
719 			error = EINVAL;
720 			goto done2;
721 		}
722 		*rval = semakptr->u.sem_base[semnum].sempid;
723 		break;
724 
725 	case GETVAL:
726 		if ((error = semvalid(semid, semakptr)) != 0)
727 			goto done2;
728 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
729 			goto done2;
730 		if (semnum < 0 || semnum >= semakptr->u.sem_nsems) {
731 			error = EINVAL;
732 			goto done2;
733 		}
734 		*rval = semakptr->u.sem_base[semnum].semval;
735 		break;
736 
737 	case GETALL:
738 		/*
739 		 * Unfortunately, callers of this function don't know
740 		 * in advance how many semaphores are in this set.
741 		 * While we could just allocate the maximum size array
742 		 * and pass the actual size back to the caller, that
743 		 * won't work for SETALL since we can't copyin() more
744 		 * data than the user specified as we may return a
745 		 * spurious EFAULT.
746 		 *
747 		 * Note that the number of semaphores in a set is
748 		 * fixed for the life of that set.  The only way that
749 		 * the 'count' could change while are blocked in
750 		 * malloc() is if this semaphore set were destroyed
751 		 * and a new one created with the same index.
752 		 * However, semvalid() will catch that due to the
753 		 * sequence number unless exactly 0x8000 (or a
754 		 * multiple thereof) semaphore sets for the same index
755 		 * are created and destroyed while we are in malloc!
756 		 *
757 		 */
758 		count = semakptr->u.sem_nsems;
759 		mtx_unlock(sema_mtxp);
760 		array = malloc(sizeof(*array) * count, M_TEMP, M_WAITOK);
761 		mtx_lock(sema_mtxp);
762 		if ((error = semvalid(semid, semakptr)) != 0)
763 			goto done2;
764 		KASSERT(count == semakptr->u.sem_nsems, ("nsems changed"));
765 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
766 			goto done2;
767 		for (i = 0; i < semakptr->u.sem_nsems; i++)
768 			array[i] = semakptr->u.sem_base[i].semval;
769 		mtx_unlock(sema_mtxp);
770 		error = copyout(array, arg->array, count * sizeof(*array));
771 		mtx_lock(sema_mtxp);
772 		break;
773 
774 	case GETZCNT:
775 		if ((error = semvalid(semid, semakptr)) != 0)
776 			goto done2;
777 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
778 			goto done2;
779 		if (semnum < 0 || semnum >= semakptr->u.sem_nsems) {
780 			error = EINVAL;
781 			goto done2;
782 		}
783 		*rval = semakptr->u.sem_base[semnum].semzcnt;
784 		break;
785 
786 	case SETVAL:
787 		if ((error = semvalid(semid, semakptr)) != 0)
788 			goto done2;
789 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_W)))
790 			goto done2;
791 		if (semnum < 0 || semnum >= semakptr->u.sem_nsems) {
792 			error = EINVAL;
793 			goto done2;
794 		}
795 		if (arg->val < 0 || arg->val > seminfo.semvmx) {
796 			error = ERANGE;
797 			goto done2;
798 		}
799 		semakptr->u.sem_base[semnum].semval = arg->val;
800 		SEMUNDO_LOCK();
801 		semundo_clear(semidx, semnum);
802 		SEMUNDO_UNLOCK();
803 		wakeup(semakptr);
804 		break;
805 
806 	case SETALL:
807 		/*
808 		 * See comment on GETALL for why 'count' shouldn't change
809 		 * and why we require a userland buffer.
810 		 */
811 		count = semakptr->u.sem_nsems;
812 		mtx_unlock(sema_mtxp);
813 		array = malloc(sizeof(*array) * count, M_TEMP, M_WAITOK);
814 		error = copyin(arg->array, array, count * sizeof(*array));
815 		mtx_lock(sema_mtxp);
816 		if (error)
817 			break;
818 		if ((error = semvalid(semid, semakptr)) != 0)
819 			goto done2;
820 		KASSERT(count == semakptr->u.sem_nsems, ("nsems changed"));
821 		if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_W)))
822 			goto done2;
823 		for (i = 0; i < semakptr->u.sem_nsems; i++) {
824 			usval = array[i];
825 			if (usval > seminfo.semvmx) {
826 				error = ERANGE;
827 				break;
828 			}
829 			semakptr->u.sem_base[i].semval = usval;
830 		}
831 		SEMUNDO_LOCK();
832 		semundo_clear(semidx, -1);
833 		SEMUNDO_UNLOCK();
834 		wakeup(semakptr);
835 		break;
836 
837 	default:
838 		error = EINVAL;
839 		break;
840 	}
841 
842 done2:
843 	mtx_unlock(sema_mtxp);
844 	if (cmd == IPC_RMID)
845 		mtx_unlock(&sem_mtx);
846 	if (array != NULL)
847 		free(array, M_TEMP);
848 	return(error);
849 }
850 
851 #ifndef _SYS_SYSPROTO_H_
852 struct semget_args {
853 	key_t	key;
854 	int	nsems;
855 	int	semflg;
856 };
857 #endif
858 int
859 sys_semget(struct thread *td, struct semget_args *uap)
860 {
861 	int semid, error = 0;
862 	int key = uap->key;
863 	int nsems = uap->nsems;
864 	int semflg = uap->semflg;
865 	struct ucred *cred = td->td_ucred;
866 
867 	DPRINTF(("semget(0x%x, %d, 0%o)\n", key, nsems, semflg));
868 	if (!prison_allow(td->td_ucred, PR_ALLOW_SYSVIPC))
869 		return (ENOSYS);
870 
871 	mtx_lock(&sem_mtx);
872 	if (key != IPC_PRIVATE) {
873 		for (semid = 0; semid < seminfo.semmni; semid++) {
874 			if ((sema[semid].u.sem_perm.mode & SEM_ALLOC) &&
875 			    sema[semid].u.sem_perm.key == key)
876 				break;
877 		}
878 		if (semid < seminfo.semmni) {
879 			DPRINTF(("found public key\n"));
880 			if ((error = ipcperm(td, &sema[semid].u.sem_perm,
881 			    semflg & 0700))) {
882 				goto done2;
883 			}
884 			if (nsems > 0 && sema[semid].u.sem_nsems < nsems) {
885 				DPRINTF(("too small\n"));
886 				error = EINVAL;
887 				goto done2;
888 			}
889 			if ((semflg & IPC_CREAT) && (semflg & IPC_EXCL)) {
890 				DPRINTF(("not exclusive\n"));
891 				error = EEXIST;
892 				goto done2;
893 			}
894 #ifdef MAC
895 			error = mac_sysvsem_check_semget(cred, &sema[semid]);
896 			if (error != 0)
897 				goto done2;
898 #endif
899 			goto found;
900 		}
901 	}
902 
903 	DPRINTF(("need to allocate the semid_kernel\n"));
904 	if (key == IPC_PRIVATE || (semflg & IPC_CREAT)) {
905 		if (nsems <= 0 || nsems > seminfo.semmsl) {
906 			DPRINTF(("nsems out of range (0<%d<=%d)\n", nsems,
907 			    seminfo.semmsl));
908 			error = EINVAL;
909 			goto done2;
910 		}
911 		if (nsems > seminfo.semmns - semtot) {
912 			DPRINTF((
913 			    "not enough semaphores left (need %d, got %d)\n",
914 			    nsems, seminfo.semmns - semtot));
915 			error = ENOSPC;
916 			goto done2;
917 		}
918 		for (semid = 0; semid < seminfo.semmni; semid++) {
919 			if ((sema[semid].u.sem_perm.mode & SEM_ALLOC) == 0)
920 				break;
921 		}
922 		if (semid == seminfo.semmni) {
923 			DPRINTF(("no more semid_kernel's available\n"));
924 			error = ENOSPC;
925 			goto done2;
926 		}
927 #ifdef RACCT
928 		PROC_LOCK(td->td_proc);
929 		error = racct_add(td->td_proc, RACCT_NSEM, nsems);
930 		PROC_UNLOCK(td->td_proc);
931 		if (error != 0) {
932 			error = ENOSPC;
933 			goto done2;
934 		}
935 #endif
936 		DPRINTF(("semid %d is available\n", semid));
937 		mtx_lock(&sema_mtx[semid]);
938 		KASSERT((sema[semid].u.sem_perm.mode & SEM_ALLOC) == 0,
939 		    ("Lost semaphore %d", semid));
940 		sema[semid].u.sem_perm.key = key;
941 		sema[semid].u.sem_perm.cuid = cred->cr_uid;
942 		sema[semid].u.sem_perm.uid = cred->cr_uid;
943 		sema[semid].u.sem_perm.cgid = cred->cr_gid;
944 		sema[semid].u.sem_perm.gid = cred->cr_gid;
945 		sema[semid].u.sem_perm.mode = (semflg & 0777) | SEM_ALLOC;
946 		sema[semid].cred = crhold(cred);
947 		sema[semid].u.sem_perm.seq =
948 		    (sema[semid].u.sem_perm.seq + 1) & 0x7fff;
949 		sema[semid].u.sem_nsems = nsems;
950 		sema[semid].u.sem_otime = 0;
951 		sema[semid].u.sem_ctime = time_second;
952 		sema[semid].u.sem_base = &sem[semtot];
953 		semtot += nsems;
954 		bzero(sema[semid].u.sem_base,
955 		    sizeof(sema[semid].u.sem_base[0])*nsems);
956 #ifdef MAC
957 		mac_sysvsem_create(cred, &sema[semid]);
958 #endif
959 		mtx_unlock(&sema_mtx[semid]);
960 		DPRINTF(("sembase = %p, next = %p\n",
961 		    sema[semid].u.sem_base, &sem[semtot]));
962 	} else {
963 		DPRINTF(("didn't find it and wasn't asked to create it\n"));
964 		error = ENOENT;
965 		goto done2;
966 	}
967 
968 found:
969 	td->td_retval[0] = IXSEQ_TO_IPCID(semid, sema[semid].u.sem_perm);
970 done2:
971 	mtx_unlock(&sem_mtx);
972 	return (error);
973 }
974 
975 #ifndef _SYS_SYSPROTO_H_
976 struct semop_args {
977 	int	semid;
978 	struct	sembuf *sops;
979 	size_t	nsops;
980 };
981 #endif
982 int
983 sys_semop(struct thread *td, struct semop_args *uap)
984 {
985 #define SMALL_SOPS	8
986 	struct sembuf small_sops[SMALL_SOPS];
987 	int semid = uap->semid;
988 	size_t nsops = uap->nsops;
989 	struct sembuf *sops;
990 	struct semid_kernel *semakptr;
991 	struct sembuf *sopptr = 0;
992 	struct sem *semptr = 0;
993 	struct sem_undo *suptr;
994 	struct mtx *sema_mtxp;
995 	size_t i, j, k;
996 	int error;
997 	int do_wakeup, do_undos;
998 	unsigned short seq;
999 
1000 #ifdef SEM_DEBUG
1001 	sops = NULL;
1002 #endif
1003 	DPRINTF(("call to semop(%d, %p, %u)\n", semid, sops, nsops));
1004 
1005 	if (!prison_allow(td->td_ucred, PR_ALLOW_SYSVIPC))
1006 		return (ENOSYS);
1007 
1008 	semid = IPCID_TO_IX(semid);	/* Convert back to zero origin */
1009 
1010 	if (semid < 0 || semid >= seminfo.semmni)
1011 		return (EINVAL);
1012 
1013 	/* Allocate memory for sem_ops */
1014 	if (nsops <= SMALL_SOPS)
1015 		sops = small_sops;
1016 	else if (nsops > seminfo.semopm) {
1017 		DPRINTF(("too many sops (max=%d, nsops=%d)\n", seminfo.semopm,
1018 		    nsops));
1019 		return (E2BIG);
1020 	} else {
1021 #ifdef RACCT
1022 		PROC_LOCK(td->td_proc);
1023 		if (nsops > racct_get_available(td->td_proc, RACCT_NSEMOP)) {
1024 			PROC_UNLOCK(td->td_proc);
1025 			return (E2BIG);
1026 		}
1027 		PROC_UNLOCK(td->td_proc);
1028 #endif
1029 
1030 		sops = malloc(nsops * sizeof(*sops), M_TEMP, M_WAITOK);
1031 	}
1032 	if ((error = copyin(uap->sops, sops, nsops * sizeof(sops[0]))) != 0) {
1033 		DPRINTF(("error = %d from copyin(%p, %p, %d)\n", error,
1034 		    uap->sops, sops, nsops * sizeof(sops[0])));
1035 		if (sops != small_sops)
1036 			free(sops, M_SEM);
1037 		return (error);
1038 	}
1039 
1040 	semakptr = &sema[semid];
1041 	sema_mtxp = &sema_mtx[semid];
1042 	mtx_lock(sema_mtxp);
1043 	if ((semakptr->u.sem_perm.mode & SEM_ALLOC) == 0) {
1044 		error = EINVAL;
1045 		goto done2;
1046 	}
1047 	seq = semakptr->u.sem_perm.seq;
1048 	if (seq != IPCID_TO_SEQ(uap->semid)) {
1049 		error = EINVAL;
1050 		goto done2;
1051 	}
1052 	/*
1053 	 * Initial pass thru sops to see what permissions are needed.
1054 	 * Also perform any checks that don't need repeating on each
1055 	 * attempt to satisfy the request vector.
1056 	 */
1057 	j = 0;		/* permission needed */
1058 	do_undos = 0;
1059 	for (i = 0; i < nsops; i++) {
1060 		sopptr = &sops[i];
1061 		if (sopptr->sem_num >= semakptr->u.sem_nsems) {
1062 			error = EFBIG;
1063 			goto done2;
1064 		}
1065 		if (sopptr->sem_flg & SEM_UNDO && sopptr->sem_op != 0)
1066 			do_undos = 1;
1067 		j |= (sopptr->sem_op == 0) ? SEM_R : SEM_A;
1068 	}
1069 
1070 	if ((error = ipcperm(td, &semakptr->u.sem_perm, j))) {
1071 		DPRINTF(("error = %d from ipaccess\n", error));
1072 		goto done2;
1073 	}
1074 #ifdef MAC
1075 	error = mac_sysvsem_check_semop(td->td_ucred, semakptr, j);
1076 	if (error != 0)
1077 		goto done2;
1078 #endif
1079 
1080 	/*
1081 	 * Loop trying to satisfy the vector of requests.
1082 	 * If we reach a point where we must wait, any requests already
1083 	 * performed are rolled back and we go to sleep until some other
1084 	 * process wakes us up.  At this point, we start all over again.
1085 	 *
1086 	 * This ensures that from the perspective of other tasks, a set
1087 	 * of requests is atomic (never partially satisfied).
1088 	 */
1089 	for (;;) {
1090 		do_wakeup = 0;
1091 		error = 0;	/* error return if necessary */
1092 
1093 		for (i = 0; i < nsops; i++) {
1094 			sopptr = &sops[i];
1095 			semptr = &semakptr->u.sem_base[sopptr->sem_num];
1096 
1097 			DPRINTF((
1098 			    "semop:  semakptr=%p, sem_base=%p, "
1099 			    "semptr=%p, sem[%d]=%d : op=%d, flag=%s\n",
1100 			    semakptr, semakptr->u.sem_base, semptr,
1101 			    sopptr->sem_num, semptr->semval, sopptr->sem_op,
1102 			    (sopptr->sem_flg & IPC_NOWAIT) ?
1103 			    "nowait" : "wait"));
1104 
1105 			if (sopptr->sem_op < 0) {
1106 				if (semptr->semval + sopptr->sem_op < 0) {
1107 					DPRINTF(("semop:  can't do it now\n"));
1108 					break;
1109 				} else {
1110 					semptr->semval += sopptr->sem_op;
1111 					if (semptr->semval == 0 &&
1112 					    semptr->semzcnt > 0)
1113 						do_wakeup = 1;
1114 				}
1115 			} else if (sopptr->sem_op == 0) {
1116 				if (semptr->semval != 0) {
1117 					DPRINTF(("semop:  not zero now\n"));
1118 					break;
1119 				}
1120 			} else if (semptr->semval + sopptr->sem_op >
1121 			    seminfo.semvmx) {
1122 				error = ERANGE;
1123 				break;
1124 			} else {
1125 				if (semptr->semncnt > 0)
1126 					do_wakeup = 1;
1127 				semptr->semval += sopptr->sem_op;
1128 			}
1129 		}
1130 
1131 		/*
1132 		 * Did we get through the entire vector?
1133 		 */
1134 		if (i >= nsops)
1135 			goto done;
1136 
1137 		/*
1138 		 * No ... rollback anything that we've already done
1139 		 */
1140 		DPRINTF(("semop:  rollback 0 through %d\n", i-1));
1141 		for (j = 0; j < i; j++)
1142 			semakptr->u.sem_base[sops[j].sem_num].semval -=
1143 			    sops[j].sem_op;
1144 
1145 		/* If we detected an error, return it */
1146 		if (error != 0)
1147 			goto done2;
1148 
1149 		/*
1150 		 * If the request that we couldn't satisfy has the
1151 		 * NOWAIT flag set then return with EAGAIN.
1152 		 */
1153 		if (sopptr->sem_flg & IPC_NOWAIT) {
1154 			error = EAGAIN;
1155 			goto done2;
1156 		}
1157 
1158 		if (sopptr->sem_op == 0)
1159 			semptr->semzcnt++;
1160 		else
1161 			semptr->semncnt++;
1162 
1163 		DPRINTF(("semop:  good night!\n"));
1164 		error = msleep(semakptr, sema_mtxp, (PZERO - 4) | PCATCH,
1165 		    "semwait", 0);
1166 		DPRINTF(("semop:  good morning (error=%d)!\n", error));
1167 		/* return code is checked below, after sem[nz]cnt-- */
1168 
1169 		/*
1170 		 * Make sure that the semaphore still exists
1171 		 */
1172 		seq = semakptr->u.sem_perm.seq;
1173 		if ((semakptr->u.sem_perm.mode & SEM_ALLOC) == 0 ||
1174 		    seq != IPCID_TO_SEQ(uap->semid)) {
1175 			error = EIDRM;
1176 			goto done2;
1177 		}
1178 
1179 		/*
1180 		 * Renew the semaphore's pointer after wakeup since
1181 		 * during msleep sem_base may have been modified and semptr
1182 		 * is not valid any more
1183 		 */
1184 		semptr = &semakptr->u.sem_base[sopptr->sem_num];
1185 
1186 		/*
1187 		 * The semaphore is still alive.  Readjust the count of
1188 		 * waiting processes.
1189 		 */
1190 		if (sopptr->sem_op == 0)
1191 			semptr->semzcnt--;
1192 		else
1193 			semptr->semncnt--;
1194 
1195 		/*
1196 		 * Is it really morning, or was our sleep interrupted?
1197 		 * (Delayed check of msleep() return code because we
1198 		 * need to decrement sem[nz]cnt either way.)
1199 		 */
1200 		if (error != 0) {
1201 			error = EINTR;
1202 			goto done2;
1203 		}
1204 		DPRINTF(("semop:  good morning!\n"));
1205 	}
1206 
1207 done:
1208 	/*
1209 	 * Process any SEM_UNDO requests.
1210 	 */
1211 	if (do_undos) {
1212 		SEMUNDO_LOCK();
1213 		suptr = NULL;
1214 		for (i = 0; i < nsops; i++) {
1215 			/*
1216 			 * We only need to deal with SEM_UNDO's for non-zero
1217 			 * op's.
1218 			 */
1219 			int adjval;
1220 
1221 			if ((sops[i].sem_flg & SEM_UNDO) == 0)
1222 				continue;
1223 			adjval = sops[i].sem_op;
1224 			if (adjval == 0)
1225 				continue;
1226 			error = semundo_adjust(td, &suptr, semid, seq,
1227 			    sops[i].sem_num, -adjval);
1228 			if (error == 0)
1229 				continue;
1230 
1231 			/*
1232 			 * Oh-Oh!  We ran out of either sem_undo's or undo's.
1233 			 * Rollback the adjustments to this point and then
1234 			 * rollback the semaphore ups and down so we can return
1235 			 * with an error with all structures restored.  We
1236 			 * rollback the undo's in the exact reverse order that
1237 			 * we applied them.  This guarantees that we won't run
1238 			 * out of space as we roll things back out.
1239 			 */
1240 			for (j = 0; j < i; j++) {
1241 				k = i - j - 1;
1242 				if ((sops[k].sem_flg & SEM_UNDO) == 0)
1243 					continue;
1244 				adjval = sops[k].sem_op;
1245 				if (adjval == 0)
1246 					continue;
1247 				if (semundo_adjust(td, &suptr, semid, seq,
1248 				    sops[k].sem_num, adjval) != 0)
1249 					panic("semop - can't undo undos");
1250 			}
1251 
1252 			for (j = 0; j < nsops; j++)
1253 				semakptr->u.sem_base[sops[j].sem_num].semval -=
1254 				    sops[j].sem_op;
1255 
1256 			DPRINTF(("error = %d from semundo_adjust\n", error));
1257 			SEMUNDO_UNLOCK();
1258 			goto done2;
1259 		} /* loop through the sops */
1260 		SEMUNDO_UNLOCK();
1261 	} /* if (do_undos) */
1262 
1263 	/* We're definitely done - set the sempid's and time */
1264 	for (i = 0; i < nsops; i++) {
1265 		sopptr = &sops[i];
1266 		semptr = &semakptr->u.sem_base[sopptr->sem_num];
1267 		semptr->sempid = td->td_proc->p_pid;
1268 	}
1269 	semakptr->u.sem_otime = time_second;
1270 
1271 	/*
1272 	 * Do a wakeup if any semaphore was up'd whilst something was
1273 	 * sleeping on it.
1274 	 */
1275 	if (do_wakeup) {
1276 		DPRINTF(("semop:  doing wakeup\n"));
1277 		wakeup(semakptr);
1278 		DPRINTF(("semop:  back from wakeup\n"));
1279 	}
1280 	DPRINTF(("semop:  done\n"));
1281 	td->td_retval[0] = 0;
1282 done2:
1283 	mtx_unlock(sema_mtxp);
1284 	if (sops != small_sops)
1285 		free(sops, M_SEM);
1286 	return (error);
1287 }
1288 
1289 /*
1290  * Go through the undo structures for this process and apply the adjustments to
1291  * semaphores.
1292  */
1293 static void
1294 semexit_myhook(void *arg, struct proc *p)
1295 {
1296 	struct sem_undo *suptr;
1297 	struct semid_kernel *semakptr;
1298 	struct mtx *sema_mtxp;
1299 	int semid, semnum, adjval, ix;
1300 	unsigned short seq;
1301 
1302 	/*
1303 	 * Go through the chain of undo vectors looking for one
1304 	 * associated with this process.
1305 	 */
1306 	SEMUNDO_LOCK();
1307 	LIST_FOREACH(suptr, &semu_list, un_next) {
1308 		if (suptr->un_proc == p)
1309 			break;
1310 	}
1311 	if (suptr == NULL) {
1312 		SEMUNDO_UNLOCK();
1313 		return;
1314 	}
1315 	LIST_REMOVE(suptr, un_next);
1316 
1317 	DPRINTF(("proc @%p has undo structure with %d entries\n", p,
1318 	    suptr->un_cnt));
1319 
1320 	/*
1321 	 * If there are any active undo elements then process them.
1322 	 */
1323 	if (suptr->un_cnt > 0) {
1324 		SEMUNDO_UNLOCK();
1325 		for (ix = 0; ix < suptr->un_cnt; ix++) {
1326 			semid = suptr->un_ent[ix].un_id;
1327 			semnum = suptr->un_ent[ix].un_num;
1328 			adjval = suptr->un_ent[ix].un_adjval;
1329 			seq = suptr->un_ent[ix].un_seq;
1330 			semakptr = &sema[semid];
1331 			sema_mtxp = &sema_mtx[semid];
1332 
1333 			mtx_lock(sema_mtxp);
1334 			if ((semakptr->u.sem_perm.mode & SEM_ALLOC) == 0 ||
1335 			    (semakptr->u.sem_perm.seq != seq)) {
1336 				mtx_unlock(sema_mtxp);
1337 				continue;
1338 			}
1339 			if (semnum >= semakptr->u.sem_nsems)
1340 				panic("semexit - semnum out of range");
1341 
1342 			DPRINTF((
1343 			    "semexit:  %p id=%d num=%d(adj=%d) ; sem=%d\n",
1344 			    suptr->un_proc, suptr->un_ent[ix].un_id,
1345 			    suptr->un_ent[ix].un_num,
1346 			    suptr->un_ent[ix].un_adjval,
1347 			    semakptr->u.sem_base[semnum].semval));
1348 
1349 			if (adjval < 0 && semakptr->u.sem_base[semnum].semval <
1350 			    -adjval)
1351 				semakptr->u.sem_base[semnum].semval = 0;
1352 			else
1353 				semakptr->u.sem_base[semnum].semval += adjval;
1354 
1355 			wakeup(semakptr);
1356 			DPRINTF(("semexit:  back from wakeup\n"));
1357 			mtx_unlock(sema_mtxp);
1358 		}
1359 		SEMUNDO_LOCK();
1360 	}
1361 
1362 	/*
1363 	 * Deallocate the undo vector.
1364 	 */
1365 	DPRINTF(("removing vector\n"));
1366 	suptr->un_proc = NULL;
1367 	suptr->un_cnt = 0;
1368 	LIST_INSERT_HEAD(&semu_free_list, suptr, un_next);
1369 	SEMUNDO_UNLOCK();
1370 }
1371 
1372 static int
1373 sysctl_sema(SYSCTL_HANDLER_ARGS)
1374 {
1375 
1376 	return (SYSCTL_OUT(req, sema,
1377 	    sizeof(struct semid_kernel) * seminfo.semmni));
1378 }
1379 
1380 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
1381     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
1382 
1383 /* XXX casting to (sy_call_t *) is bogus, as usual. */
1384 static sy_call_t *semcalls[] = {
1385 	(sy_call_t *)freebsd7___semctl, (sy_call_t *)sys_semget,
1386 	(sy_call_t *)sys_semop
1387 };
1388 
1389 /*
1390  * Entry point for all SEM calls.
1391  */
1392 int
1393 sys_semsys(td, uap)
1394 	struct thread *td;
1395 	/* XXX actually varargs. */
1396 	struct semsys_args /* {
1397 		int	which;
1398 		int	a2;
1399 		int	a3;
1400 		int	a4;
1401 		int	a5;
1402 	} */ *uap;
1403 {
1404 	int error;
1405 
1406 	if (!prison_allow(td->td_ucred, PR_ALLOW_SYSVIPC))
1407 		return (ENOSYS);
1408 	if (uap->which < 0 ||
1409 	    uap->which >= sizeof(semcalls)/sizeof(semcalls[0]))
1410 		return (EINVAL);
1411 	error = (*semcalls[uap->which])(td, &uap->a2);
1412 	return (error);
1413 }
1414 
1415 #ifndef CP
1416 #define CP(src, dst, fld)	do { (dst).fld = (src).fld; } while (0)
1417 #endif
1418 
1419 #ifndef _SYS_SYSPROTO_H_
1420 struct freebsd7___semctl_args {
1421 	int	semid;
1422 	int	semnum;
1423 	int	cmd;
1424 	union	semun_old *arg;
1425 };
1426 #endif
1427 int
1428 freebsd7___semctl(struct thread *td, struct freebsd7___semctl_args *uap)
1429 {
1430 	struct semid_ds_old dsold;
1431 	struct semid_ds dsbuf;
1432 	union semun_old arg;
1433 	union semun semun;
1434 	register_t rval;
1435 	int error;
1436 
1437 	switch (uap->cmd) {
1438 	case SEM_STAT:
1439 	case IPC_SET:
1440 	case IPC_STAT:
1441 	case GETALL:
1442 	case SETVAL:
1443 	case SETALL:
1444 		error = copyin(uap->arg, &arg, sizeof(arg));
1445 		if (error)
1446 			return (error);
1447 		break;
1448 	}
1449 
1450 	switch (uap->cmd) {
1451 	case SEM_STAT:
1452 	case IPC_STAT:
1453 		semun.buf = &dsbuf;
1454 		break;
1455 	case IPC_SET:
1456 		error = copyin(arg.buf, &dsold, sizeof(dsold));
1457 		if (error)
1458 			return (error);
1459 		ipcperm_old2new(&dsold.sem_perm, &dsbuf.sem_perm);
1460 		CP(dsold, dsbuf, sem_base);
1461 		CP(dsold, dsbuf, sem_nsems);
1462 		CP(dsold, dsbuf, sem_otime);
1463 		CP(dsold, dsbuf, sem_ctime);
1464 		semun.buf = &dsbuf;
1465 		break;
1466 	case GETALL:
1467 	case SETALL:
1468 		semun.array = arg.array;
1469 		break;
1470 	case SETVAL:
1471 		semun.val = arg.val;
1472 		break;
1473 	}
1474 
1475 	error = kern_semctl(td, uap->semid, uap->semnum, uap->cmd, &semun,
1476 	    &rval);
1477 	if (error)
1478 		return (error);
1479 
1480 	switch (uap->cmd) {
1481 	case SEM_STAT:
1482 	case IPC_STAT:
1483 		bzero(&dsold, sizeof(dsold));
1484 		ipcperm_new2old(&dsbuf.sem_perm, &dsold.sem_perm);
1485 		CP(dsbuf, dsold, sem_base);
1486 		CP(dsbuf, dsold, sem_nsems);
1487 		CP(dsbuf, dsold, sem_otime);
1488 		CP(dsbuf, dsold, sem_ctime);
1489 		error = copyout(&dsold, arg.buf, sizeof(dsold));
1490 		break;
1491 	}
1492 
1493 	if (error == 0)
1494 		td->td_retval[0] = rval;
1495 	return (error);
1496 }
1497 
1498 #endif /* COMPAT_FREEBSD{4,5,6,7} */
1499 
1500 #ifdef COMPAT_FREEBSD32
1501 
1502 int
1503 freebsd32_semsys(struct thread *td, struct freebsd32_semsys_args *uap)
1504 {
1505 
1506 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
1507     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
1508 	switch (uap->which) {
1509 	case 0:
1510 		return (freebsd7_freebsd32_semctl(td,
1511 		    (struct freebsd7_freebsd32_semctl_args *)&uap->a2));
1512 	default:
1513 		return (sys_semsys(td, (struct semsys_args *)uap));
1514 	}
1515 #else
1516 	return (nosys(td, NULL));
1517 #endif
1518 }
1519 
1520 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
1521     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
1522 int
1523 freebsd7_freebsd32_semctl(struct thread *td,
1524     struct freebsd7_freebsd32_semctl_args *uap)
1525 {
1526 	struct semid_ds32_old dsbuf32;
1527 	struct semid_ds dsbuf;
1528 	union semun semun;
1529 	union semun32 arg;
1530 	register_t rval;
1531 	int error;
1532 
1533 	switch (uap->cmd) {
1534 	case SEM_STAT:
1535 	case IPC_SET:
1536 	case IPC_STAT:
1537 	case GETALL:
1538 	case SETVAL:
1539 	case SETALL:
1540 		error = copyin(uap->arg, &arg, sizeof(arg));
1541 		if (error)
1542 			return (error);
1543 		break;
1544 	}
1545 
1546 	switch (uap->cmd) {
1547 	case SEM_STAT:
1548 	case IPC_STAT:
1549 		semun.buf = &dsbuf;
1550 		break;
1551 	case IPC_SET:
1552 		error = copyin(PTRIN(arg.buf), &dsbuf32, sizeof(dsbuf32));
1553 		if (error)
1554 			return (error);
1555 		freebsd32_ipcperm_old_in(&dsbuf32.sem_perm, &dsbuf.sem_perm);
1556 		PTRIN_CP(dsbuf32, dsbuf, sem_base);
1557 		CP(dsbuf32, dsbuf, sem_nsems);
1558 		CP(dsbuf32, dsbuf, sem_otime);
1559 		CP(dsbuf32, dsbuf, sem_ctime);
1560 		semun.buf = &dsbuf;
1561 		break;
1562 	case GETALL:
1563 	case SETALL:
1564 		semun.array = PTRIN(arg.array);
1565 		break;
1566 	case SETVAL:
1567 		semun.val = arg.val;
1568 		break;
1569 	}
1570 
1571 	error = kern_semctl(td, uap->semid, uap->semnum, uap->cmd, &semun,
1572 	    &rval);
1573 	if (error)
1574 		return (error);
1575 
1576 	switch (uap->cmd) {
1577 	case SEM_STAT:
1578 	case IPC_STAT:
1579 		bzero(&dsbuf32, sizeof(dsbuf32));
1580 		freebsd32_ipcperm_old_out(&dsbuf.sem_perm, &dsbuf32.sem_perm);
1581 		PTROUT_CP(dsbuf, dsbuf32, sem_base);
1582 		CP(dsbuf, dsbuf32, sem_nsems);
1583 		CP(dsbuf, dsbuf32, sem_otime);
1584 		CP(dsbuf, dsbuf32, sem_ctime);
1585 		error = copyout(&dsbuf32, PTRIN(arg.buf), sizeof(dsbuf32));
1586 		break;
1587 	}
1588 
1589 	if (error == 0)
1590 		td->td_retval[0] = rval;
1591 	return (error);
1592 }
1593 #endif
1594 
1595 int
1596 freebsd32_semctl(struct thread *td, struct freebsd32_semctl_args *uap)
1597 {
1598 	struct semid_ds32 dsbuf32;
1599 	struct semid_ds dsbuf;
1600 	union semun semun;
1601 	union semun32 arg;
1602 	register_t rval;
1603 	int error;
1604 
1605 	switch (uap->cmd) {
1606 	case SEM_STAT:
1607 	case IPC_SET:
1608 	case IPC_STAT:
1609 	case GETALL:
1610 	case SETVAL:
1611 	case SETALL:
1612 		error = copyin(uap->arg, &arg, sizeof(arg));
1613 		if (error)
1614 			return (error);
1615 		break;
1616 	}
1617 
1618 	switch (uap->cmd) {
1619 	case SEM_STAT:
1620 	case IPC_STAT:
1621 		semun.buf = &dsbuf;
1622 		break;
1623 	case IPC_SET:
1624 		error = copyin(PTRIN(arg.buf), &dsbuf32, sizeof(dsbuf32));
1625 		if (error)
1626 			return (error);
1627 		freebsd32_ipcperm_in(&dsbuf32.sem_perm, &dsbuf.sem_perm);
1628 		PTRIN_CP(dsbuf32, dsbuf, sem_base);
1629 		CP(dsbuf32, dsbuf, sem_nsems);
1630 		CP(dsbuf32, dsbuf, sem_otime);
1631 		CP(dsbuf32, dsbuf, sem_ctime);
1632 		semun.buf = &dsbuf;
1633 		break;
1634 	case GETALL:
1635 	case SETALL:
1636 		semun.array = PTRIN(arg.array);
1637 		break;
1638 	case SETVAL:
1639 		semun.val = arg.val;
1640 		break;
1641 	}
1642 
1643 	error = kern_semctl(td, uap->semid, uap->semnum, uap->cmd, &semun,
1644 	    &rval);
1645 	if (error)
1646 		return (error);
1647 
1648 	switch (uap->cmd) {
1649 	case SEM_STAT:
1650 	case IPC_STAT:
1651 		bzero(&dsbuf32, sizeof(dsbuf32));
1652 		freebsd32_ipcperm_out(&dsbuf.sem_perm, &dsbuf32.sem_perm);
1653 		PTROUT_CP(dsbuf, dsbuf32, sem_base);
1654 		CP(dsbuf, dsbuf32, sem_nsems);
1655 		CP(dsbuf, dsbuf32, sem_otime);
1656 		CP(dsbuf, dsbuf32, sem_ctime);
1657 		error = copyout(&dsbuf32, PTRIN(arg.buf), sizeof(dsbuf32));
1658 		break;
1659 	}
1660 
1661 	if (error == 0)
1662 		td->td_retval[0] = rval;
1663 	return (error);
1664 }
1665 
1666 #endif /* COMPAT_FREEBSD32 */
1667