xref: /freebsd/sys/kern/sysv_sem.c (revision 5521ff5a4d1929056e7ffc982fac3341ca54df7c)
1 /* $FreeBSD$ */
2 
3 /*
4  * Implementation of SVID semaphores
5  *
6  * Author:  Daniel Boulet
7  *
8  * This software is provided ``AS IS'' without any warranties of any kind.
9  */
10 
11 #include "opt_sysvipc.h"
12 
13 #include <sys/param.h>
14 #include <sys/systm.h>
15 #include <sys/sysproto.h>
16 #include <sys/kernel.h>
17 #include <sys/proc.h>
18 #include <sys/sem.h>
19 #include <sys/syscall.h>
20 #include <sys/sysent.h>
21 #include <sys/sysctl.h>
22 #include <sys/malloc.h>
23 #include <sys/jail.h>
24 
25 static MALLOC_DEFINE(M_SEM, "sem", "SVID compatible semaphores");
26 
27 static void seminit __P((void));
28 static int sysvsem_modload __P((struct module *, int, void *));
29 static int semunload __P((void));
30 static void semexit_myhook __P((struct proc *p));
31 static int sysctl_sema __P((SYSCTL_HANDLER_ARGS));
32 
33 #ifndef _SYS_SYSPROTO_H_
34 struct __semctl_args;
35 int __semctl __P((struct proc *p, struct __semctl_args *uap));
36 struct semget_args;
37 int semget __P((struct proc *p, struct semget_args *uap));
38 struct semop_args;
39 int semop __P((struct proc *p, struct semop_args *uap));
40 #endif
41 
42 static struct sem_undo *semu_alloc __P((struct proc *p));
43 static int semundo_adjust __P((struct proc *p, struct sem_undo **supptr,
44 		int semid, int semnum, int adjval));
45 static void semundo_clear __P((int semid, int semnum));
46 
47 /* XXX casting to (sy_call_t *) is bogus, as usual. */
48 static sy_call_t *semcalls[] = {
49 	(sy_call_t *)__semctl, (sy_call_t *)semget,
50 	(sy_call_t *)semop
51 };
52 
53 static int	semtot = 0;
54 static struct semid_ds *sema;	/* semaphore id pool */
55 static struct sem *sem;		/* semaphore pool */
56 static struct sem_undo *semu_list; /* list of active undo structures */
57 static int	*semu;		/* undo structure pool */
58 
59 struct sem {
60 	u_short	semval;		/* semaphore value */
61 	pid_t	sempid;		/* pid of last operation */
62 	u_short	semncnt;	/* # awaiting semval > cval */
63 	u_short	semzcnt;	/* # awaiting semval = 0 */
64 };
65 
66 /*
67  * Undo structure (one per process)
68  */
69 struct sem_undo {
70 	struct	sem_undo *un_next;	/* ptr to next active undo structure */
71 	struct	proc *un_proc;		/* owner of this structure */
72 	short	un_cnt;			/* # of active entries */
73 	struct undo {
74 		short	un_adjval;	/* adjust on exit values */
75 		short	un_num;		/* semaphore # */
76 		int	un_id;		/* semid */
77 	} un_ent[1];			/* undo entries */
78 };
79 
80 /*
81  * Configuration parameters
82  */
83 #ifndef SEMMNI
84 #define SEMMNI	10		/* # of semaphore identifiers */
85 #endif
86 #ifndef SEMMNS
87 #define SEMMNS	60		/* # of semaphores in system */
88 #endif
89 #ifndef SEMUME
90 #define SEMUME	10		/* max # of undo entries per process */
91 #endif
92 #ifndef SEMMNU
93 #define SEMMNU	30		/* # of undo structures in system */
94 #endif
95 
96 /* shouldn't need tuning */
97 #ifndef SEMMAP
98 #define SEMMAP	30		/* # of entries in semaphore map */
99 #endif
100 #ifndef SEMMSL
101 #define SEMMSL	SEMMNS		/* max # of semaphores per id */
102 #endif
103 #ifndef SEMOPM
104 #define SEMOPM	100		/* max # of operations per semop call */
105 #endif
106 
107 #define SEMVMX	32767		/* semaphore maximum value */
108 #define SEMAEM	16384		/* adjust on exit max value */
109 
110 /*
111  * Due to the way semaphore memory is allocated, we have to ensure that
112  * SEMUSZ is properly aligned.
113  */
114 
115 #define SEM_ALIGN(bytes) (((bytes) + (sizeof(long) - 1)) & ~(sizeof(long) - 1))
116 
117 /* actual size of an undo structure */
118 #define SEMUSZ	SEM_ALIGN(offsetof(struct sem_undo, un_ent[SEMUME]))
119 
120 /*
121  * Macro to find a particular sem_undo vector
122  */
123 #define SEMU(ix)	((struct sem_undo *)(((intptr_t)semu)+ix * seminfo.semusz))
124 
125 /*
126  * semaphore info struct
127  */
128 struct seminfo seminfo = {
129                 SEMMAP,         /* # of entries in semaphore map */
130                 SEMMNI,         /* # of semaphore identifiers */
131                 SEMMNS,         /* # of semaphores in system */
132                 SEMMNU,         /* # of undo structures in system */
133                 SEMMSL,         /* max # of semaphores per id */
134                 SEMOPM,         /* max # of operations per semop call */
135                 SEMUME,         /* max # of undo entries per process */
136                 SEMUSZ,         /* size in bytes of undo structure */
137                 SEMVMX,         /* semaphore maximum value */
138                 SEMAEM          /* adjust on exit max value */
139 };
140 
141 SYSCTL_DECL(_kern_ipc);
142 SYSCTL_INT(_kern_ipc, OID_AUTO, semmap, CTLFLAG_RW, &seminfo.semmap, 0, "");
143 SYSCTL_INT(_kern_ipc, OID_AUTO, semmni, CTLFLAG_RD, &seminfo.semmni, 0, "");
144 SYSCTL_INT(_kern_ipc, OID_AUTO, semmns, CTLFLAG_RD, &seminfo.semmns, 0, "");
145 SYSCTL_INT(_kern_ipc, OID_AUTO, semmnu, CTLFLAG_RD, &seminfo.semmnu, 0, "");
146 SYSCTL_INT(_kern_ipc, OID_AUTO, semmsl, CTLFLAG_RW, &seminfo.semmsl, 0, "");
147 SYSCTL_INT(_kern_ipc, OID_AUTO, semopm, CTLFLAG_RD, &seminfo.semopm, 0, "");
148 SYSCTL_INT(_kern_ipc, OID_AUTO, semume, CTLFLAG_RD, &seminfo.semume, 0, "");
149 SYSCTL_INT(_kern_ipc, OID_AUTO, semusz, CTLFLAG_RD, &seminfo.semusz, 0, "");
150 SYSCTL_INT(_kern_ipc, OID_AUTO, semvmx, CTLFLAG_RW, &seminfo.semvmx, 0, "");
151 SYSCTL_INT(_kern_ipc, OID_AUTO, semaem, CTLFLAG_RW, &seminfo.semaem, 0, "");
152 SYSCTL_PROC(_kern_ipc, OID_AUTO, sema, CTLFLAG_RD,
153     NULL, 0, sysctl_sema, "", "");
154 
155 #if 0
156 RO seminfo.semmap	/* SEMMAP unused */
157 RO seminfo.semmni
158 RO seminfo.semmns
159 RO seminfo.semmnu	/* undo entries per system */
160 RW seminfo.semmsl
161 RO seminfo.semopm	/* SEMOPM unused */
162 RO seminfo.semume
163 RO seminfo.semusz	/* param - derived from SEMUME for per-proc sizeof */
164 RO seminfo.semvmx	/* SEMVMX unused - user param */
165 RO seminfo.semaem	/* SEMAEM unused - user param */
166 #endif
167 
168 static void
169 seminit(void)
170 {
171 	register int i;
172 
173 	sem = malloc(sizeof(struct sem) * seminfo.semmns, M_SEM, M_WAITOK);
174 	if (sem == NULL)
175 		panic("sem is NULL");
176 	sema = malloc(sizeof(struct semid_ds) * seminfo.semmni, M_SEM, M_WAITOK);
177 	if (sema == NULL)
178 		panic("sema is NULL");
179 	semu = malloc(seminfo.semmnu * seminfo.semusz, M_SEM, M_WAITOK);
180 	if (semu == NULL)
181 		panic("semu is NULL");
182 
183 	for (i = 0; i < seminfo.semmni; i++) {
184 		sema[i].sem_base = 0;
185 		sema[i].sem_perm.mode = 0;
186 	}
187 	for (i = 0; i < seminfo.semmnu; i++) {
188 		register struct sem_undo *suptr = SEMU(i);
189 		suptr->un_proc = NULL;
190 	}
191 	semu_list = NULL;
192 	semexit_hook = &semexit_myhook;
193 }
194 
195 static int
196 semunload(void)
197 {
198 
199 	if (semtot != 0)
200 		return (EBUSY);
201 
202 	free(sem, M_SEM);
203 	free(sema, M_SEM);
204 	free(semu, M_SEM);
205 	semexit_hook = NULL;
206 	return (0);
207 }
208 
209 static int
210 sysvsem_modload(struct module *module, int cmd, void *arg)
211 {
212 	int error = 0;
213 
214 	switch (cmd) {
215 	case MOD_LOAD:
216 		seminit();
217 		break;
218 	case MOD_UNLOAD:
219 		error = semunload();
220 		break;
221 	case MOD_SHUTDOWN:
222 		break;
223 	default:
224 		error = EINVAL;
225 		break;
226 	}
227 	return (error);
228 }
229 
230 static moduledata_t sysvsem_mod = {
231 	"sysvsem",
232 	&sysvsem_modload,
233 	NULL
234 };
235 
236 SYSCALL_MODULE_HELPER(semsys, 5);
237 SYSCALL_MODULE_HELPER(__semctl, 4);
238 SYSCALL_MODULE_HELPER(semget, 3);
239 SYSCALL_MODULE_HELPER(semop, 3);
240 
241 DECLARE_MODULE(sysvsem, sysvsem_mod,
242 	SI_SUB_SYSV_SEM, SI_ORDER_FIRST);
243 MODULE_VERSION(sysvsem, 1);
244 
245 /*
246  * Entry point for all SEM calls
247  */
248 int
249 semsys(p, uap)
250 	struct proc *p;
251 	/* XXX actually varargs. */
252 	struct semsys_args /* {
253 		u_int	which;
254 		int	a2;
255 		int	a3;
256 		int	a4;
257 		int	a5;
258 	} */ *uap;
259 {
260 
261 	if (!jail_sysvipc_allowed && jailed(p->p_ucred))
262 		return (ENOSYS);
263 
264 	if (uap->which >= sizeof(semcalls)/sizeof(semcalls[0]))
265 		return (EINVAL);
266 	return ((*semcalls[uap->which])(p, &uap->a2));
267 }
268 
269 /*
270  * Allocate a new sem_undo structure for a process
271  * (returns ptr to structure or NULL if no more room)
272  */
273 
274 static struct sem_undo *
275 semu_alloc(p)
276 	struct proc *p;
277 {
278 	register int i;
279 	register struct sem_undo *suptr;
280 	register struct sem_undo **supptr;
281 	int attempt;
282 
283 	/*
284 	 * Try twice to allocate something.
285 	 * (we'll purge any empty structures after the first pass so
286 	 * two passes are always enough)
287 	 */
288 
289 	for (attempt = 0; attempt < 2; attempt++) {
290 		/*
291 		 * Look for a free structure.
292 		 * Fill it in and return it if we find one.
293 		 */
294 
295 		for (i = 0; i < seminfo.semmnu; i++) {
296 			suptr = SEMU(i);
297 			if (suptr->un_proc == NULL) {
298 				suptr->un_next = semu_list;
299 				semu_list = suptr;
300 				suptr->un_cnt = 0;
301 				suptr->un_proc = p;
302 				return(suptr);
303 			}
304 		}
305 
306 		/*
307 		 * We didn't find a free one, if this is the first attempt
308 		 * then try to free some structures.
309 		 */
310 
311 		if (attempt == 0) {
312 			/* All the structures are in use - try to free some */
313 			int did_something = 0;
314 
315 			supptr = &semu_list;
316 			while ((suptr = *supptr) != NULL) {
317 				if (suptr->un_cnt == 0)  {
318 					suptr->un_proc = NULL;
319 					*supptr = suptr->un_next;
320 					did_something = 1;
321 				} else
322 					supptr = &(suptr->un_next);
323 			}
324 
325 			/* If we didn't free anything then just give-up */
326 			if (!did_something)
327 				return(NULL);
328 		} else {
329 			/*
330 			 * The second pass failed even though we freed
331 			 * something after the first pass!
332 			 * This is IMPOSSIBLE!
333 			 */
334 			panic("semu_alloc - second attempt failed");
335 		}
336 	}
337 	return (NULL);
338 }
339 
340 /*
341  * Adjust a particular entry for a particular proc
342  */
343 
344 static int
345 semundo_adjust(p, supptr, semid, semnum, adjval)
346 	register struct proc *p;
347 	struct sem_undo **supptr;
348 	int semid, semnum;
349 	int adjval;
350 {
351 	register struct sem_undo *suptr;
352 	register struct undo *sunptr;
353 	int i;
354 
355 	/* Look for and remember the sem_undo if the caller doesn't provide
356 	   it */
357 
358 	suptr = *supptr;
359 	if (suptr == NULL) {
360 		for (suptr = semu_list; suptr != NULL;
361 		    suptr = suptr->un_next) {
362 			if (suptr->un_proc == p) {
363 				*supptr = suptr;
364 				break;
365 			}
366 		}
367 		if (suptr == NULL) {
368 			if (adjval == 0)
369 				return(0);
370 			suptr = semu_alloc(p);
371 			if (suptr == NULL)
372 				return(ENOSPC);
373 			*supptr = suptr;
374 		}
375 	}
376 
377 	/*
378 	 * Look for the requested entry and adjust it (delete if adjval becomes
379 	 * 0).
380 	 */
381 	sunptr = &suptr->un_ent[0];
382 	for (i = 0; i < suptr->un_cnt; i++, sunptr++) {
383 		if (sunptr->un_id != semid || sunptr->un_num != semnum)
384 			continue;
385 		if (adjval == 0)
386 			sunptr->un_adjval = 0;
387 		else
388 			sunptr->un_adjval += adjval;
389 		if (sunptr->un_adjval == 0) {
390 			suptr->un_cnt--;
391 			if (i < suptr->un_cnt)
392 				suptr->un_ent[i] =
393 				    suptr->un_ent[suptr->un_cnt];
394 		}
395 		return(0);
396 	}
397 
398 	/* Didn't find the right entry - create it */
399 	if (adjval == 0)
400 		return(0);
401 	if (suptr->un_cnt != seminfo.semume) {
402 		sunptr = &suptr->un_ent[suptr->un_cnt];
403 		suptr->un_cnt++;
404 		sunptr->un_adjval = adjval;
405 		sunptr->un_id = semid; sunptr->un_num = semnum;
406 	} else
407 		return(EINVAL);
408 	return(0);
409 }
410 
411 static void
412 semundo_clear(semid, semnum)
413 	int semid, semnum;
414 {
415 	register struct sem_undo *suptr;
416 
417 	for (suptr = semu_list; suptr != NULL; suptr = suptr->un_next) {
418 		register struct undo *sunptr = &suptr->un_ent[0];
419 		register int i = 0;
420 
421 		while (i < suptr->un_cnt) {
422 			if (sunptr->un_id == semid) {
423 				if (semnum == -1 || sunptr->un_num == semnum) {
424 					suptr->un_cnt--;
425 					if (i < suptr->un_cnt) {
426 						suptr->un_ent[i] =
427 						  suptr->un_ent[suptr->un_cnt];
428 						continue;
429 					}
430 				}
431 				if (semnum != -1)
432 					break;
433 			}
434 			i++, sunptr++;
435 		}
436 	}
437 }
438 
439 /*
440  * Note that the user-mode half of this passes a union, not a pointer
441  */
442 #ifndef _SYS_SYSPROTO_H_
443 struct __semctl_args {
444 	int	semid;
445 	int	semnum;
446 	int	cmd;
447 	union	semun *arg;
448 };
449 #endif
450 
451 int
452 __semctl(p, uap)
453 	struct proc *p;
454 	register struct __semctl_args *uap;
455 {
456 	int semid = uap->semid;
457 	int semnum = uap->semnum;
458 	int cmd = uap->cmd;
459 	union semun *arg = uap->arg;
460 	union semun real_arg;
461 	struct ucred *cred = p->p_ucred;
462 	int i, rval, eval;
463 	struct semid_ds sbuf;
464 	register struct semid_ds *semaptr;
465 
466 #ifdef SEM_DEBUG
467 	printf("call to semctl(%d, %d, %d, 0x%x)\n", semid, semnum, cmd, arg);
468 #endif
469 
470 	if (!jail_sysvipc_allowed && jailed(p->p_ucred))
471 		return (ENOSYS);
472 
473 	semid = IPCID_TO_IX(semid);
474 	if (semid < 0 || semid >= seminfo.semmsl)
475 		return(EINVAL);
476 
477 	semaptr = &sema[semid];
478 	if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0 ||
479 	    semaptr->sem_perm.seq != IPCID_TO_SEQ(uap->semid))
480 		return(EINVAL);
481 
482 	eval = 0;
483 	rval = 0;
484 
485 	switch (cmd) {
486 	case IPC_RMID:
487 		if ((eval = ipcperm(p, &semaptr->sem_perm, IPC_M)))
488 			return(eval);
489 		semaptr->sem_perm.cuid = cred->cr_uid;
490 		semaptr->sem_perm.uid = cred->cr_uid;
491 		semtot -= semaptr->sem_nsems;
492 		for (i = semaptr->sem_base - sem; i < semtot; i++)
493 			sem[i] = sem[i + semaptr->sem_nsems];
494 		for (i = 0; i < seminfo.semmni; i++) {
495 			if ((sema[i].sem_perm.mode & SEM_ALLOC) &&
496 			    sema[i].sem_base > semaptr->sem_base)
497 				sema[i].sem_base -= semaptr->sem_nsems;
498 		}
499 		semaptr->sem_perm.mode = 0;
500 		semundo_clear(semid, -1);
501 		wakeup((caddr_t)semaptr);
502 		break;
503 
504 	case IPC_SET:
505 		if ((eval = ipcperm(p, &semaptr->sem_perm, IPC_M)))
506 			return(eval);
507 		if ((eval = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
508 			return(eval);
509 		if ((eval = copyin(real_arg.buf, (caddr_t)&sbuf,
510 		    sizeof(sbuf))) != 0)
511 			return(eval);
512 		semaptr->sem_perm.uid = sbuf.sem_perm.uid;
513 		semaptr->sem_perm.gid = sbuf.sem_perm.gid;
514 		semaptr->sem_perm.mode = (semaptr->sem_perm.mode & ~0777) |
515 		    (sbuf.sem_perm.mode & 0777);
516 		semaptr->sem_ctime = time_second;
517 		break;
518 
519 	case IPC_STAT:
520 		if ((eval = ipcperm(p, &semaptr->sem_perm, IPC_R)))
521 			return(eval);
522 		if ((eval = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
523 			return(eval);
524 		eval = copyout((caddr_t)semaptr, real_arg.buf,
525 		    sizeof(struct semid_ds));
526 		break;
527 
528 	case GETNCNT:
529 		if ((eval = ipcperm(p, &semaptr->sem_perm, IPC_R)))
530 			return(eval);
531 		if (semnum < 0 || semnum >= semaptr->sem_nsems)
532 			return(EINVAL);
533 		rval = semaptr->sem_base[semnum].semncnt;
534 		break;
535 
536 	case GETPID:
537 		if ((eval = ipcperm(p, &semaptr->sem_perm, IPC_R)))
538 			return(eval);
539 		if (semnum < 0 || semnum >= semaptr->sem_nsems)
540 			return(EINVAL);
541 		rval = semaptr->sem_base[semnum].sempid;
542 		break;
543 
544 	case GETVAL:
545 		if ((eval = ipcperm(p, &semaptr->sem_perm, IPC_R)))
546 			return(eval);
547 		if (semnum < 0 || semnum >= semaptr->sem_nsems)
548 			return(EINVAL);
549 		rval = semaptr->sem_base[semnum].semval;
550 		break;
551 
552 	case GETALL:
553 		if ((eval = ipcperm(p, &semaptr->sem_perm, IPC_R)))
554 			return(eval);
555 		if ((eval = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
556 			return(eval);
557 		for (i = 0; i < semaptr->sem_nsems; i++) {
558 			eval = copyout((caddr_t)&semaptr->sem_base[i].semval,
559 			    &real_arg.array[i], sizeof(real_arg.array[0]));
560 			if (eval != 0)
561 				break;
562 		}
563 		break;
564 
565 	case GETZCNT:
566 		if ((eval = ipcperm(p, &semaptr->sem_perm, IPC_R)))
567 			return(eval);
568 		if (semnum < 0 || semnum >= semaptr->sem_nsems)
569 			return(EINVAL);
570 		rval = semaptr->sem_base[semnum].semzcnt;
571 		break;
572 
573 	case SETVAL:
574 		if ((eval = ipcperm(p, &semaptr->sem_perm, IPC_W)))
575 			return(eval);
576 		if (semnum < 0 || semnum >= semaptr->sem_nsems)
577 			return(EINVAL);
578 		if ((eval = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
579 			return(eval);
580 		semaptr->sem_base[semnum].semval = real_arg.val;
581 		semundo_clear(semid, semnum);
582 		wakeup((caddr_t)semaptr);
583 		break;
584 
585 	case SETALL:
586 		if ((eval = ipcperm(p, &semaptr->sem_perm, IPC_W)))
587 			return(eval);
588 		if ((eval = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
589 			return(eval);
590 		for (i = 0; i < semaptr->sem_nsems; i++) {
591 			eval = copyin(&real_arg.array[i],
592 			    (caddr_t)&semaptr->sem_base[i].semval,
593 			    sizeof(real_arg.array[0]));
594 			if (eval != 0)
595 				break;
596 		}
597 		semundo_clear(semid, -1);
598 		wakeup((caddr_t)semaptr);
599 		break;
600 
601 	default:
602 		return(EINVAL);
603 	}
604 
605 	if (eval == 0)
606 		p->p_retval[0] = rval;
607 	return(eval);
608 }
609 
610 #ifndef _SYS_SYSPROTO_H_
611 struct semget_args {
612 	key_t	key;
613 	int	nsems;
614 	int	semflg;
615 };
616 #endif
617 
618 int
619 semget(p, uap)
620 	struct proc *p;
621 	register struct semget_args *uap;
622 {
623 	int semid, eval;
624 	int key = uap->key;
625 	int nsems = uap->nsems;
626 	int semflg = uap->semflg;
627 	struct ucred *cred = p->p_ucred;
628 
629 #ifdef SEM_DEBUG
630 	printf("semget(0x%x, %d, 0%o)\n", key, nsems, semflg);
631 #endif
632 
633 	if (!jail_sysvipc_allowed && jailed(p->p_ucred))
634 		return (ENOSYS);
635 
636 	if (key != IPC_PRIVATE) {
637 		for (semid = 0; semid < seminfo.semmni; semid++) {
638 			if ((sema[semid].sem_perm.mode & SEM_ALLOC) &&
639 			    sema[semid].sem_perm.key == key)
640 				break;
641 		}
642 		if (semid < seminfo.semmni) {
643 #ifdef SEM_DEBUG
644 			printf("found public key\n");
645 #endif
646 			if ((eval = ipcperm(p, &sema[semid].sem_perm,
647 			    semflg & 0700)))
648 				return(eval);
649 			if (nsems > 0 && sema[semid].sem_nsems < nsems) {
650 #ifdef SEM_DEBUG
651 				printf("too small\n");
652 #endif
653 				return(EINVAL);
654 			}
655 			if ((semflg & IPC_CREAT) && (semflg & IPC_EXCL)) {
656 #ifdef SEM_DEBUG
657 				printf("not exclusive\n");
658 #endif
659 				return(EEXIST);
660 			}
661 			goto found;
662 		}
663 	}
664 
665 #ifdef SEM_DEBUG
666 	printf("need to allocate the semid_ds\n");
667 #endif
668 	if (key == IPC_PRIVATE || (semflg & IPC_CREAT)) {
669 		if (nsems <= 0 || nsems > seminfo.semmsl) {
670 #ifdef SEM_DEBUG
671 			printf("nsems out of range (0<%d<=%d)\n", nsems,
672 			    seminfo.semmsl);
673 #endif
674 			return(EINVAL);
675 		}
676 		if (nsems > seminfo.semmns - semtot) {
677 #ifdef SEM_DEBUG
678 			printf("not enough semaphores left (need %d, got %d)\n",
679 			    nsems, seminfo.semmns - semtot);
680 #endif
681 			return(ENOSPC);
682 		}
683 		for (semid = 0; semid < seminfo.semmni; semid++) {
684 			if ((sema[semid].sem_perm.mode & SEM_ALLOC) == 0)
685 				break;
686 		}
687 		if (semid == seminfo.semmni) {
688 #ifdef SEM_DEBUG
689 			printf("no more semid_ds's available\n");
690 #endif
691 			return(ENOSPC);
692 		}
693 #ifdef SEM_DEBUG
694 		printf("semid %d is available\n", semid);
695 #endif
696 		sema[semid].sem_perm.key = key;
697 		sema[semid].sem_perm.cuid = cred->cr_uid;
698 		sema[semid].sem_perm.uid = cred->cr_uid;
699 		sema[semid].sem_perm.cgid = cred->cr_gid;
700 		sema[semid].sem_perm.gid = cred->cr_gid;
701 		sema[semid].sem_perm.mode = (semflg & 0777) | SEM_ALLOC;
702 		sema[semid].sem_perm.seq =
703 		    (sema[semid].sem_perm.seq + 1) & 0x7fff;
704 		sema[semid].sem_nsems = nsems;
705 		sema[semid].sem_otime = 0;
706 		sema[semid].sem_ctime = time_second;
707 		sema[semid].sem_base = &sem[semtot];
708 		semtot += nsems;
709 		bzero(sema[semid].sem_base,
710 		    sizeof(sema[semid].sem_base[0])*nsems);
711 #ifdef SEM_DEBUG
712 		printf("sembase = 0x%x, next = 0x%x\n", sema[semid].sem_base,
713 		    &sem[semtot]);
714 #endif
715 	} else {
716 #ifdef SEM_DEBUG
717 		printf("didn't find it and wasn't asked to create it\n");
718 #endif
719 		return(ENOENT);
720 	}
721 
722 found:
723 	p->p_retval[0] = IXSEQ_TO_IPCID(semid, sema[semid].sem_perm);
724 	return(0);
725 }
726 
727 #ifndef _SYS_SYSPROTO_H_
728 struct semop_args {
729 	int	semid;
730 	struct	sembuf *sops;
731 	int	nsops;
732 };
733 #endif
734 
735 int
736 semop(p, uap)
737 	struct proc *p;
738 	register struct semop_args *uap;
739 {
740 	int semid = uap->semid;
741 	int nsops = uap->nsops;
742 	struct sembuf sops[MAX_SOPS];
743 	register struct semid_ds *semaptr;
744 	register struct sembuf *sopptr;
745 	register struct sem *semptr;
746 	struct sem_undo *suptr = NULL;
747 	int i, j, eval;
748 	int do_wakeup, do_undos;
749 
750 #ifdef SEM_DEBUG
751 	printf("call to semop(%d, 0x%x, %d)\n", semid, sops, nsops);
752 #endif
753 
754 	if (!jail_sysvipc_allowed && jailed(p->p_ucred))
755 		return (ENOSYS);
756 
757 	semid = IPCID_TO_IX(semid);	/* Convert back to zero origin */
758 
759 	if (semid < 0 || semid >= seminfo.semmsl)
760 		return(EINVAL);
761 
762 	semaptr = &sema[semid];
763 	if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0)
764 		return(EINVAL);
765 	if (semaptr->sem_perm.seq != IPCID_TO_SEQ(uap->semid))
766 		return(EINVAL);
767 
768 	if ((eval = ipcperm(p, &semaptr->sem_perm, IPC_W))) {
769 #ifdef SEM_DEBUG
770 		printf("eval = %d from ipaccess\n", eval);
771 #endif
772 		return(eval);
773 	}
774 
775 	if (nsops > MAX_SOPS) {
776 #ifdef SEM_DEBUG
777 		printf("too many sops (max=%d, nsops=%d)\n", MAX_SOPS, nsops);
778 #endif
779 		return(E2BIG);
780 	}
781 
782 	if ((eval = copyin(uap->sops, &sops, nsops * sizeof(sops[0]))) != 0) {
783 #ifdef SEM_DEBUG
784 		printf("eval = %d from copyin(%08x, %08x, %d)\n", eval,
785 		    uap->sops, &sops, nsops * sizeof(sops[0]));
786 #endif
787 		return(eval);
788 	}
789 
790 	/*
791 	 * Loop trying to satisfy the vector of requests.
792 	 * If we reach a point where we must wait, any requests already
793 	 * performed are rolled back and we go to sleep until some other
794 	 * process wakes us up.  At this point, we start all over again.
795 	 *
796 	 * This ensures that from the perspective of other tasks, a set
797 	 * of requests is atomic (never partially satisfied).
798 	 */
799 	do_undos = 0;
800 
801 	for (;;) {
802 		do_wakeup = 0;
803 
804 		for (i = 0; i < nsops; i++) {
805 			sopptr = &sops[i];
806 
807 			if (sopptr->sem_num >= semaptr->sem_nsems)
808 				return(EFBIG);
809 
810 			semptr = &semaptr->sem_base[sopptr->sem_num];
811 
812 #ifdef SEM_DEBUG
813 			printf("semop:  semaptr=%x, sem_base=%x, semptr=%x, sem[%d]=%d : op=%d, flag=%s\n",
814 			    semaptr, semaptr->sem_base, semptr,
815 			    sopptr->sem_num, semptr->semval, sopptr->sem_op,
816 			    (sopptr->sem_flg & IPC_NOWAIT) ? "nowait" : "wait");
817 #endif
818 
819 			if (sopptr->sem_op < 0) {
820 				if (semptr->semval + sopptr->sem_op < 0) {
821 #ifdef SEM_DEBUG
822 					printf("semop:  can't do it now\n");
823 #endif
824 					break;
825 				} else {
826 					semptr->semval += sopptr->sem_op;
827 					if (semptr->semval == 0 &&
828 					    semptr->semzcnt > 0)
829 						do_wakeup = 1;
830 				}
831 				if (sopptr->sem_flg & SEM_UNDO)
832 					do_undos = 1;
833 			} else if (sopptr->sem_op == 0) {
834 				if (semptr->semval > 0) {
835 #ifdef SEM_DEBUG
836 					printf("semop:  not zero now\n");
837 #endif
838 					break;
839 				}
840 			} else {
841 				if (semptr->semncnt > 0)
842 					do_wakeup = 1;
843 				semptr->semval += sopptr->sem_op;
844 				if (sopptr->sem_flg & SEM_UNDO)
845 					do_undos = 1;
846 			}
847 		}
848 
849 		/*
850 		 * Did we get through the entire vector?
851 		 */
852 		if (i >= nsops)
853 			goto done;
854 
855 		/*
856 		 * No ... rollback anything that we've already done
857 		 */
858 #ifdef SEM_DEBUG
859 		printf("semop:  rollback 0 through %d\n", i-1);
860 #endif
861 		for (j = 0; j < i; j++)
862 			semaptr->sem_base[sops[j].sem_num].semval -=
863 			    sops[j].sem_op;
864 
865 		/*
866 		 * If the request that we couldn't satisfy has the
867 		 * NOWAIT flag set then return with EAGAIN.
868 		 */
869 		if (sopptr->sem_flg & IPC_NOWAIT)
870 			return(EAGAIN);
871 
872 		if (sopptr->sem_op == 0)
873 			semptr->semzcnt++;
874 		else
875 			semptr->semncnt++;
876 
877 #ifdef SEM_DEBUG
878 		printf("semop:  good night!\n");
879 #endif
880 		eval = tsleep((caddr_t)semaptr, (PZERO - 4) | PCATCH,
881 		    "semwait", 0);
882 #ifdef SEM_DEBUG
883 		printf("semop:  good morning (eval=%d)!\n", eval);
884 #endif
885 
886 		suptr = NULL;	/* sem_undo may have been reallocated */
887 
888 		if (eval != 0)
889 			return(EINTR);
890 #ifdef SEM_DEBUG
891 		printf("semop:  good morning!\n");
892 #endif
893 
894 		/*
895 		 * Make sure that the semaphore still exists
896 		 */
897 		if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0 ||
898 		    semaptr->sem_perm.seq != IPCID_TO_SEQ(uap->semid))
899 			return(EIDRM);
900 
901 		/*
902 		 * The semaphore is still alive.  Readjust the count of
903 		 * waiting processes.
904 		 */
905 		if (sopptr->sem_op == 0)
906 			semptr->semzcnt--;
907 		else
908 			semptr->semncnt--;
909 	}
910 
911 done:
912 	/*
913 	 * Process any SEM_UNDO requests.
914 	 */
915 	if (do_undos) {
916 		for (i = 0; i < nsops; i++) {
917 			/*
918 			 * We only need to deal with SEM_UNDO's for non-zero
919 			 * op's.
920 			 */
921 			int adjval;
922 
923 			if ((sops[i].sem_flg & SEM_UNDO) == 0)
924 				continue;
925 			adjval = sops[i].sem_op;
926 			if (adjval == 0)
927 				continue;
928 			eval = semundo_adjust(p, &suptr, semid,
929 			    sops[i].sem_num, -adjval);
930 			if (eval == 0)
931 				continue;
932 
933 			/*
934 			 * Oh-Oh!  We ran out of either sem_undo's or undo's.
935 			 * Rollback the adjustments to this point and then
936 			 * rollback the semaphore ups and down so we can return
937 			 * with an error with all structures restored.  We
938 			 * rollback the undo's in the exact reverse order that
939 			 * we applied them.  This guarantees that we won't run
940 			 * out of space as we roll things back out.
941 			 */
942 			for (j = i - 1; j >= 0; j--) {
943 				if ((sops[j].sem_flg & SEM_UNDO) == 0)
944 					continue;
945 				adjval = sops[j].sem_op;
946 				if (adjval == 0)
947 					continue;
948 				if (semundo_adjust(p, &suptr, semid,
949 				    sops[j].sem_num, adjval) != 0)
950 					panic("semop - can't undo undos");
951 			}
952 
953 			for (j = 0; j < nsops; j++)
954 				semaptr->sem_base[sops[j].sem_num].semval -=
955 				    sops[j].sem_op;
956 
957 #ifdef SEM_DEBUG
958 			printf("eval = %d from semundo_adjust\n", eval);
959 #endif
960 			return(eval);
961 		} /* loop through the sops */
962 	} /* if (do_undos) */
963 
964 	/* We're definitely done - set the sempid's */
965 	for (i = 0; i < nsops; i++) {
966 		sopptr = &sops[i];
967 		semptr = &semaptr->sem_base[sopptr->sem_num];
968 		semptr->sempid = p->p_pid;
969 	}
970 
971 	/* Do a wakeup if any semaphore was up'd. */
972 	if (do_wakeup) {
973 #ifdef SEM_DEBUG
974 		printf("semop:  doing wakeup\n");
975 #endif
976 		wakeup((caddr_t)semaptr);
977 #ifdef SEM_DEBUG
978 		printf("semop:  back from wakeup\n");
979 #endif
980 	}
981 #ifdef SEM_DEBUG
982 	printf("semop:  done\n");
983 #endif
984 	p->p_retval[0] = 0;
985 	return(0);
986 }
987 
988 /*
989  * Go through the undo structures for this process and apply the adjustments to
990  * semaphores.
991  */
992 static void
993 semexit_myhook(p)
994 	struct proc *p;
995 {
996 	register struct sem_undo *suptr;
997 	register struct sem_undo **supptr;
998 	int did_something;
999 
1000 	did_something = 0;
1001 
1002 	/*
1003 	 * Go through the chain of undo vectors looking for one
1004 	 * associated with this process.
1005 	 */
1006 
1007 	for (supptr = &semu_list; (suptr = *supptr) != NULL;
1008 	    supptr = &suptr->un_next) {
1009 		if (suptr->un_proc == p)
1010 			break;
1011 	}
1012 
1013 	if (suptr == NULL)
1014 		return;
1015 
1016 #ifdef SEM_DEBUG
1017 	printf("proc @%08x has undo structure with %d entries\n", p,
1018 	    suptr->un_cnt);
1019 #endif
1020 
1021 	/*
1022 	 * If there are any active undo elements then process them.
1023 	 */
1024 	if (suptr->un_cnt > 0) {
1025 		int ix;
1026 
1027 		for (ix = 0; ix < suptr->un_cnt; ix++) {
1028 			int semid = suptr->un_ent[ix].un_id;
1029 			int semnum = suptr->un_ent[ix].un_num;
1030 			int adjval = suptr->un_ent[ix].un_adjval;
1031 			struct semid_ds *semaptr;
1032 
1033 			semaptr = &sema[semid];
1034 			if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0)
1035 				panic("semexit - semid not allocated");
1036 			if (semnum >= semaptr->sem_nsems)
1037 				panic("semexit - semnum out of range");
1038 
1039 #ifdef SEM_DEBUG
1040 			printf("semexit:  %08x id=%d num=%d(adj=%d) ; sem=%d\n",
1041 			    suptr->un_proc, suptr->un_ent[ix].un_id,
1042 			    suptr->un_ent[ix].un_num,
1043 			    suptr->un_ent[ix].un_adjval,
1044 			    semaptr->sem_base[semnum].semval);
1045 #endif
1046 
1047 			if (adjval < 0) {
1048 				if (semaptr->sem_base[semnum].semval < -adjval)
1049 					semaptr->sem_base[semnum].semval = 0;
1050 				else
1051 					semaptr->sem_base[semnum].semval +=
1052 					    adjval;
1053 			} else
1054 				semaptr->sem_base[semnum].semval += adjval;
1055 
1056 			wakeup((caddr_t)semaptr);
1057 #ifdef SEM_DEBUG
1058 			printf("semexit:  back from wakeup\n");
1059 #endif
1060 		}
1061 	}
1062 
1063 	/*
1064 	 * Deallocate the undo vector.
1065 	 */
1066 #ifdef SEM_DEBUG
1067 	printf("removing vector\n");
1068 #endif
1069 	suptr->un_proc = NULL;
1070 	*supptr = suptr->un_next;
1071 }
1072 
1073 static int
1074 sysctl_sema(SYSCTL_HANDLER_ARGS)
1075 {
1076 
1077 	return (SYSCTL_OUT(req, sema,
1078 	    sizeof(struct semid_ds) * seminfo.semmni));
1079 }
1080