xref: /freebsd/sys/kern/sysv_sem.c (revision ee41f1b1cf5e3d4f586cb85b46123b416275862c)
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 
32 #ifndef _SYS_SYSPROTO_H_
33 struct __semctl_args;
34 int __semctl __P((struct proc *p, struct __semctl_args *uap));
35 struct semget_args;
36 int semget __P((struct proc *p, struct semget_args *uap));
37 struct semop_args;
38 int semop __P((struct proc *p, struct semop_args *uap));
39 #endif
40 
41 static struct sem_undo *semu_alloc __P((struct proc *p));
42 static int semundo_adjust __P((struct proc *p, struct sem_undo **supptr,
43 		int semid, int semnum, int adjval));
44 static void semundo_clear __P((int semid, int semnum));
45 
46 /* XXX casting to (sy_call_t *) is bogus, as usual. */
47 static sy_call_t *semcalls[] = {
48 	(sy_call_t *)__semctl, (sy_call_t *)semget,
49 	(sy_call_t *)semop
50 };
51 
52 static int	semtot = 0;
53 static struct semid_ds *sema;	/* semaphore id pool */
54 static struct sem *sem;		/* semaphore pool */
55 static struct sem_undo *semu_list; /* list of active undo structures */
56 static int	*semu;		/* undo structure pool */
57 
58 struct sem {
59 	u_short	semval;		/* semaphore value */
60 	pid_t	sempid;		/* pid of last operation */
61 	u_short	semncnt;	/* # awaiting semval > cval */
62 	u_short	semzcnt;	/* # awaiting semval = 0 */
63 };
64 
65 /*
66  * Undo structure (one per process)
67  */
68 struct sem_undo {
69 	struct	sem_undo *un_next;	/* ptr to next active undo structure */
70 	struct	proc *un_proc;		/* owner of this structure */
71 	short	un_cnt;			/* # of active entries */
72 	struct undo {
73 		short	un_adjval;	/* adjust on exit values */
74 		short	un_num;		/* semaphore # */
75 		int	un_id;		/* semid */
76 	} un_ent[1];			/* undo entries */
77 };
78 
79 /*
80  * Configuration parameters
81  */
82 #ifndef SEMMNI
83 #define SEMMNI	10		/* # of semaphore identifiers */
84 #endif
85 #ifndef SEMMNS
86 #define SEMMNS	60		/* # of semaphores in system */
87 #endif
88 #ifndef SEMUME
89 #define SEMUME	10		/* max # of undo entries per process */
90 #endif
91 #ifndef SEMMNU
92 #define SEMMNU	30		/* # of undo structures in system */
93 #endif
94 
95 /* shouldn't need tuning */
96 #ifndef SEMMAP
97 #define SEMMAP	30		/* # of entries in semaphore map */
98 #endif
99 #ifndef SEMMSL
100 #define SEMMSL	SEMMNS		/* max # of semaphores per id */
101 #endif
102 #ifndef SEMOPM
103 #define SEMOPM	100		/* max # of operations per semop call */
104 #endif
105 
106 #define SEMVMX	32767		/* semaphore maximum value */
107 #define SEMAEM	16384		/* adjust on exit max value */
108 
109 /*
110  * Due to the way semaphore memory is allocated, we have to ensure that
111  * SEMUSZ is properly aligned.
112  */
113 
114 #define SEM_ALIGN(bytes) (((bytes) + (sizeof(long) - 1)) & ~(sizeof(long) - 1))
115 
116 /* actual size of an undo structure */
117 #define SEMUSZ	SEM_ALIGN(offsetof(struct sem_undo, un_ent[SEMUME]))
118 
119 /*
120  * Macro to find a particular sem_undo vector
121  */
122 #define SEMU(ix)	((struct sem_undo *)(((intptr_t)semu)+ix * seminfo.semusz))
123 
124 /*
125  * semaphore info struct
126  */
127 struct seminfo seminfo = {
128                 SEMMAP,         /* # of entries in semaphore map */
129                 SEMMNI,         /* # of semaphore identifiers */
130                 SEMMNS,         /* # of semaphores in system */
131                 SEMMNU,         /* # of undo structures in system */
132                 SEMMSL,         /* max # of semaphores per id */
133                 SEMOPM,         /* max # of operations per semop call */
134                 SEMUME,         /* max # of undo entries per process */
135                 SEMUSZ,         /* size in bytes of undo structure */
136                 SEMVMX,         /* semaphore maximum value */
137                 SEMAEM          /* adjust on exit max value */
138 };
139 
140 SYSCTL_DECL(_kern_ipc);
141 SYSCTL_INT(_kern_ipc, OID_AUTO, semmap, CTLFLAG_RW, &seminfo.semmap, 0, "");
142 SYSCTL_INT(_kern_ipc, OID_AUTO, semmni, CTLFLAG_RD, &seminfo.semmni, 0, "");
143 SYSCTL_INT(_kern_ipc, OID_AUTO, semmns, CTLFLAG_RD, &seminfo.semmns, 0, "");
144 SYSCTL_INT(_kern_ipc, OID_AUTO, semmnu, CTLFLAG_RD, &seminfo.semmnu, 0, "");
145 SYSCTL_INT(_kern_ipc, OID_AUTO, semmsl, CTLFLAG_RW, &seminfo.semmsl, 0, "");
146 SYSCTL_INT(_kern_ipc, OID_AUTO, semopm, CTLFLAG_RD, &seminfo.semopm, 0, "");
147 SYSCTL_INT(_kern_ipc, OID_AUTO, semume, CTLFLAG_RD, &seminfo.semume, 0, "");
148 SYSCTL_INT(_kern_ipc, OID_AUTO, semusz, CTLFLAG_RD, &seminfo.semusz, 0, "");
149 SYSCTL_INT(_kern_ipc, OID_AUTO, semvmx, CTLFLAG_RW, &seminfo.semvmx, 0, "");
150 SYSCTL_INT(_kern_ipc, OID_AUTO, semaem, CTLFLAG_RW, &seminfo.semaem, 0, "");
151 
152 #if 0
153 RO seminfo.semmap	/* SEMMAP unused */
154 RO seminfo.semmni
155 RO seminfo.semmns
156 RO seminfo.semmnu	/* undo entries per system */
157 RW seminfo.semmsl
158 RO seminfo.semopm	/* SEMOPM unused */
159 RO seminfo.semume
160 RO seminfo.semusz	/* param - derived from SEMUME for per-proc sizeof */
161 RO seminfo.semvmx	/* SEMVMX unused - user param */
162 RO seminfo.semaem	/* SEMAEM unused - user param */
163 #endif
164 
165 static void
166 seminit(void)
167 {
168 	register int i;
169 
170 	sem = malloc(sizeof(struct sem) * seminfo.semmns, M_SEM, M_WAITOK);
171 	if (sem == NULL)
172 		panic("sem is NULL");
173 	sema = malloc(sizeof(struct semid_ds) * seminfo.semmni, M_SEM, M_WAITOK);
174 	if (sema == NULL)
175 		panic("sema is NULL");
176 	semu = malloc(seminfo.semmnu * seminfo.semusz, M_SEM, M_WAITOK);
177 	if (semu == NULL)
178 		panic("semu is NULL");
179 
180 	for (i = 0; i < seminfo.semmni; i++) {
181 		sema[i].sem_base = 0;
182 		sema[i].sem_perm.mode = 0;
183 	}
184 	for (i = 0; i < seminfo.semmnu; i++) {
185 		register struct sem_undo *suptr = SEMU(i);
186 		suptr->un_proc = NULL;
187 	}
188 	semu_list = NULL;
189 	semexit_hook = &semexit_myhook;
190 }
191 
192 static int
193 semunload(void)
194 {
195 
196 	if (semtot != 0)
197 		return (EBUSY);
198 
199 	free(sem, M_SEM);
200 	free(sema, M_SEM);
201 	free(semu, M_SEM);
202 	semexit_hook = NULL;
203 	return (0);
204 }
205 
206 static int
207 sysvsem_modload(struct module *module, int cmd, void *arg)
208 {
209 	int error = 0;
210 
211 	switch (cmd) {
212 	case MOD_LOAD:
213 		seminit();
214 		break;
215 	case MOD_UNLOAD:
216 		error = semunload();
217 		break;
218 	case MOD_SHUTDOWN:
219 		break;
220 	default:
221 		error = EINVAL;
222 		break;
223 	}
224 	return (error);
225 }
226 
227 static moduledata_t sysvsem_mod = {
228 	"sysvsem",
229 	&sysvsem_modload,
230 	NULL
231 };
232 
233 SYSCALL_MODULE_HELPER(semsys, 5);
234 SYSCALL_MODULE_HELPER(__semctl, 4);
235 SYSCALL_MODULE_HELPER(semget, 3);
236 SYSCALL_MODULE_HELPER(semop, 3);
237 
238 DECLARE_MODULE(sysvsem, sysvsem_mod,
239 	SI_SUB_SYSV_SEM, SI_ORDER_FIRST);
240 MODULE_VERSION(sysvsem, 1);
241 
242 /*
243  * Entry point for all SEM calls
244  */
245 int
246 semsys(p, uap)
247 	struct proc *p;
248 	/* XXX actually varargs. */
249 	struct semsys_args /* {
250 		u_int	which;
251 		int	a2;
252 		int	a3;
253 		int	a4;
254 		int	a5;
255 	} */ *uap;
256 {
257 
258 	if (!jail_sysvipc_allowed && p->p_prison != NULL)
259 		return (ENOSYS);
260 
261 	if (uap->which >= sizeof(semcalls)/sizeof(semcalls[0]))
262 		return (EINVAL);
263 	return ((*semcalls[uap->which])(p, &uap->a2));
264 }
265 
266 /*
267  * Allocate a new sem_undo structure for a process
268  * (returns ptr to structure or NULL if no more room)
269  */
270 
271 static struct sem_undo *
272 semu_alloc(p)
273 	struct proc *p;
274 {
275 	register int i;
276 	register struct sem_undo *suptr;
277 	register struct sem_undo **supptr;
278 	int attempt;
279 
280 	/*
281 	 * Try twice to allocate something.
282 	 * (we'll purge any empty structures after the first pass so
283 	 * two passes are always enough)
284 	 */
285 
286 	for (attempt = 0; attempt < 2; attempt++) {
287 		/*
288 		 * Look for a free structure.
289 		 * Fill it in and return it if we find one.
290 		 */
291 
292 		for (i = 0; i < seminfo.semmnu; i++) {
293 			suptr = SEMU(i);
294 			if (suptr->un_proc == NULL) {
295 				suptr->un_next = semu_list;
296 				semu_list = suptr;
297 				suptr->un_cnt = 0;
298 				suptr->un_proc = p;
299 				return(suptr);
300 			}
301 		}
302 
303 		/*
304 		 * We didn't find a free one, if this is the first attempt
305 		 * then try to free some structures.
306 		 */
307 
308 		if (attempt == 0) {
309 			/* All the structures are in use - try to free some */
310 			int did_something = 0;
311 
312 			supptr = &semu_list;
313 			while ((suptr = *supptr) != NULL) {
314 				if (suptr->un_cnt == 0)  {
315 					suptr->un_proc = NULL;
316 					*supptr = suptr->un_next;
317 					did_something = 1;
318 				} else
319 					supptr = &(suptr->un_next);
320 			}
321 
322 			/* If we didn't free anything then just give-up */
323 			if (!did_something)
324 				return(NULL);
325 		} else {
326 			/*
327 			 * The second pass failed even though we freed
328 			 * something after the first pass!
329 			 * This is IMPOSSIBLE!
330 			 */
331 			panic("semu_alloc - second attempt failed");
332 		}
333 	}
334 	return (NULL);
335 }
336 
337 /*
338  * Adjust a particular entry for a particular proc
339  */
340 
341 static int
342 semundo_adjust(p, supptr, semid, semnum, adjval)
343 	register struct proc *p;
344 	struct sem_undo **supptr;
345 	int semid, semnum;
346 	int adjval;
347 {
348 	register struct sem_undo *suptr;
349 	register struct undo *sunptr;
350 	int i;
351 
352 	/* Look for and remember the sem_undo if the caller doesn't provide
353 	   it */
354 
355 	suptr = *supptr;
356 	if (suptr == NULL) {
357 		for (suptr = semu_list; suptr != NULL;
358 		    suptr = suptr->un_next) {
359 			if (suptr->un_proc == p) {
360 				*supptr = suptr;
361 				break;
362 			}
363 		}
364 		if (suptr == NULL) {
365 			if (adjval == 0)
366 				return(0);
367 			suptr = semu_alloc(p);
368 			if (suptr == NULL)
369 				return(ENOSPC);
370 			*supptr = suptr;
371 		}
372 	}
373 
374 	/*
375 	 * Look for the requested entry and adjust it (delete if adjval becomes
376 	 * 0).
377 	 */
378 	sunptr = &suptr->un_ent[0];
379 	for (i = 0; i < suptr->un_cnt; i++, sunptr++) {
380 		if (sunptr->un_id != semid || sunptr->un_num != semnum)
381 			continue;
382 		if (adjval == 0)
383 			sunptr->un_adjval = 0;
384 		else
385 			sunptr->un_adjval += adjval;
386 		if (sunptr->un_adjval == 0) {
387 			suptr->un_cnt--;
388 			if (i < suptr->un_cnt)
389 				suptr->un_ent[i] =
390 				    suptr->un_ent[suptr->un_cnt];
391 		}
392 		return(0);
393 	}
394 
395 	/* Didn't find the right entry - create it */
396 	if (adjval == 0)
397 		return(0);
398 	if (suptr->un_cnt != seminfo.semume) {
399 		sunptr = &suptr->un_ent[suptr->un_cnt];
400 		suptr->un_cnt++;
401 		sunptr->un_adjval = adjval;
402 		sunptr->un_id = semid; sunptr->un_num = semnum;
403 	} else
404 		return(EINVAL);
405 	return(0);
406 }
407 
408 static void
409 semundo_clear(semid, semnum)
410 	int semid, semnum;
411 {
412 	register struct sem_undo *suptr;
413 
414 	for (suptr = semu_list; suptr != NULL; suptr = suptr->un_next) {
415 		register struct undo *sunptr = &suptr->un_ent[0];
416 		register int i = 0;
417 
418 		while (i < suptr->un_cnt) {
419 			if (sunptr->un_id == semid) {
420 				if (semnum == -1 || sunptr->un_num == semnum) {
421 					suptr->un_cnt--;
422 					if (i < suptr->un_cnt) {
423 						suptr->un_ent[i] =
424 						  suptr->un_ent[suptr->un_cnt];
425 						continue;
426 					}
427 				}
428 				if (semnum != -1)
429 					break;
430 			}
431 			i++, sunptr++;
432 		}
433 	}
434 }
435 
436 /*
437  * Note that the user-mode half of this passes a union, not a pointer
438  */
439 #ifndef _SYS_SYSPROTO_H_
440 struct __semctl_args {
441 	int	semid;
442 	int	semnum;
443 	int	cmd;
444 	union	semun *arg;
445 };
446 #endif
447 
448 int
449 __semctl(p, uap)
450 	struct proc *p;
451 	register struct __semctl_args *uap;
452 {
453 	int semid = uap->semid;
454 	int semnum = uap->semnum;
455 	int cmd = uap->cmd;
456 	union semun *arg = uap->arg;
457 	union semun real_arg;
458 	struct ucred *cred = p->p_ucred;
459 	int i, rval, eval;
460 	struct semid_ds sbuf;
461 	register struct semid_ds *semaptr;
462 
463 #ifdef SEM_DEBUG
464 	printf("call to semctl(%d, %d, %d, 0x%x)\n", semid, semnum, cmd, arg);
465 #endif
466 
467 	if (!jail_sysvipc_allowed && p->p_prison != NULL)
468 		return (ENOSYS);
469 
470 	semid = IPCID_TO_IX(semid);
471 	if (semid < 0 || semid >= seminfo.semmsl)
472 		return(EINVAL);
473 
474 	semaptr = &sema[semid];
475 	if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0 ||
476 	    semaptr->sem_perm.seq != IPCID_TO_SEQ(uap->semid))
477 		return(EINVAL);
478 
479 	eval = 0;
480 	rval = 0;
481 
482 	switch (cmd) {
483 	case IPC_RMID:
484 		if ((eval = ipcperm(p, &semaptr->sem_perm, IPC_M)))
485 			return(eval);
486 		semaptr->sem_perm.cuid = cred->cr_uid;
487 		semaptr->sem_perm.uid = cred->cr_uid;
488 		semtot -= semaptr->sem_nsems;
489 		for (i = semaptr->sem_base - sem; i < semtot; i++)
490 			sem[i] = sem[i + semaptr->sem_nsems];
491 		for (i = 0; i < seminfo.semmni; i++) {
492 			if ((sema[i].sem_perm.mode & SEM_ALLOC) &&
493 			    sema[i].sem_base > semaptr->sem_base)
494 				sema[i].sem_base -= semaptr->sem_nsems;
495 		}
496 		semaptr->sem_perm.mode = 0;
497 		semundo_clear(semid, -1);
498 		wakeup((caddr_t)semaptr);
499 		break;
500 
501 	case IPC_SET:
502 		if ((eval = ipcperm(p, &semaptr->sem_perm, IPC_M)))
503 			return(eval);
504 		if ((eval = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
505 			return(eval);
506 		if ((eval = copyin(real_arg.buf, (caddr_t)&sbuf,
507 		    sizeof(sbuf))) != 0)
508 			return(eval);
509 		semaptr->sem_perm.uid = sbuf.sem_perm.uid;
510 		semaptr->sem_perm.gid = sbuf.sem_perm.gid;
511 		semaptr->sem_perm.mode = (semaptr->sem_perm.mode & ~0777) |
512 		    (sbuf.sem_perm.mode & 0777);
513 		semaptr->sem_ctime = time_second;
514 		break;
515 
516 	case IPC_STAT:
517 		if ((eval = ipcperm(p, &semaptr->sem_perm, IPC_R)))
518 			return(eval);
519 		if ((eval = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
520 			return(eval);
521 		eval = copyout((caddr_t)semaptr, real_arg.buf,
522 		    sizeof(struct semid_ds));
523 		break;
524 
525 	case GETNCNT:
526 		if ((eval = ipcperm(p, &semaptr->sem_perm, IPC_R)))
527 			return(eval);
528 		if (semnum < 0 || semnum >= semaptr->sem_nsems)
529 			return(EINVAL);
530 		rval = semaptr->sem_base[semnum].semncnt;
531 		break;
532 
533 	case GETPID:
534 		if ((eval = ipcperm(p, &semaptr->sem_perm, IPC_R)))
535 			return(eval);
536 		if (semnum < 0 || semnum >= semaptr->sem_nsems)
537 			return(EINVAL);
538 		rval = semaptr->sem_base[semnum].sempid;
539 		break;
540 
541 	case GETVAL:
542 		if ((eval = ipcperm(p, &semaptr->sem_perm, IPC_R)))
543 			return(eval);
544 		if (semnum < 0 || semnum >= semaptr->sem_nsems)
545 			return(EINVAL);
546 		rval = semaptr->sem_base[semnum].semval;
547 		break;
548 
549 	case GETALL:
550 		if ((eval = ipcperm(p, &semaptr->sem_perm, IPC_R)))
551 			return(eval);
552 		if ((eval = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
553 			return(eval);
554 		for (i = 0; i < semaptr->sem_nsems; i++) {
555 			eval = copyout((caddr_t)&semaptr->sem_base[i].semval,
556 			    &real_arg.array[i], sizeof(real_arg.array[0]));
557 			if (eval != 0)
558 				break;
559 		}
560 		break;
561 
562 	case GETZCNT:
563 		if ((eval = ipcperm(p, &semaptr->sem_perm, IPC_R)))
564 			return(eval);
565 		if (semnum < 0 || semnum >= semaptr->sem_nsems)
566 			return(EINVAL);
567 		rval = semaptr->sem_base[semnum].semzcnt;
568 		break;
569 
570 	case SETVAL:
571 		if ((eval = ipcperm(p, &semaptr->sem_perm, IPC_W)))
572 			return(eval);
573 		if (semnum < 0 || semnum >= semaptr->sem_nsems)
574 			return(EINVAL);
575 		if ((eval = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
576 			return(eval);
577 		semaptr->sem_base[semnum].semval = real_arg.val;
578 		semundo_clear(semid, semnum);
579 		wakeup((caddr_t)semaptr);
580 		break;
581 
582 	case SETALL:
583 		if ((eval = ipcperm(p, &semaptr->sem_perm, IPC_W)))
584 			return(eval);
585 		if ((eval = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
586 			return(eval);
587 		for (i = 0; i < semaptr->sem_nsems; i++) {
588 			eval = copyin(&real_arg.array[i],
589 			    (caddr_t)&semaptr->sem_base[i].semval,
590 			    sizeof(real_arg.array[0]));
591 			if (eval != 0)
592 				break;
593 		}
594 		semundo_clear(semid, -1);
595 		wakeup((caddr_t)semaptr);
596 		break;
597 
598 	default:
599 		return(EINVAL);
600 	}
601 
602 	if (eval == 0)
603 		p->p_retval[0] = rval;
604 	return(eval);
605 }
606 
607 #ifndef _SYS_SYSPROTO_H_
608 struct semget_args {
609 	key_t	key;
610 	int	nsems;
611 	int	semflg;
612 };
613 #endif
614 
615 int
616 semget(p, uap)
617 	struct proc *p;
618 	register struct semget_args *uap;
619 {
620 	int semid, eval;
621 	int key = uap->key;
622 	int nsems = uap->nsems;
623 	int semflg = uap->semflg;
624 	struct ucred *cred = p->p_ucred;
625 
626 #ifdef SEM_DEBUG
627 	printf("semget(0x%x, %d, 0%o)\n", key, nsems, semflg);
628 #endif
629 
630 	if (!jail_sysvipc_allowed && p->p_prison != NULL)
631 		return (ENOSYS);
632 
633 	if (key != IPC_PRIVATE) {
634 		for (semid = 0; semid < seminfo.semmni; semid++) {
635 			if ((sema[semid].sem_perm.mode & SEM_ALLOC) &&
636 			    sema[semid].sem_perm.key == key)
637 				break;
638 		}
639 		if (semid < seminfo.semmni) {
640 #ifdef SEM_DEBUG
641 			printf("found public key\n");
642 #endif
643 			if ((eval = ipcperm(p, &sema[semid].sem_perm,
644 			    semflg & 0700)))
645 				return(eval);
646 			if (nsems > 0 && sema[semid].sem_nsems < nsems) {
647 #ifdef SEM_DEBUG
648 				printf("too small\n");
649 #endif
650 				return(EINVAL);
651 			}
652 			if ((semflg & IPC_CREAT) && (semflg & IPC_EXCL)) {
653 #ifdef SEM_DEBUG
654 				printf("not exclusive\n");
655 #endif
656 				return(EEXIST);
657 			}
658 			goto found;
659 		}
660 	}
661 
662 #ifdef SEM_DEBUG
663 	printf("need to allocate the semid_ds\n");
664 #endif
665 	if (key == IPC_PRIVATE || (semflg & IPC_CREAT)) {
666 		if (nsems <= 0 || nsems > seminfo.semmsl) {
667 #ifdef SEM_DEBUG
668 			printf("nsems out of range (0<%d<=%d)\n", nsems,
669 			    seminfo.semmsl);
670 #endif
671 			return(EINVAL);
672 		}
673 		if (nsems > seminfo.semmns - semtot) {
674 #ifdef SEM_DEBUG
675 			printf("not enough semaphores left (need %d, got %d)\n",
676 			    nsems, seminfo.semmns - semtot);
677 #endif
678 			return(ENOSPC);
679 		}
680 		for (semid = 0; semid < seminfo.semmni; semid++) {
681 			if ((sema[semid].sem_perm.mode & SEM_ALLOC) == 0)
682 				break;
683 		}
684 		if (semid == seminfo.semmni) {
685 #ifdef SEM_DEBUG
686 			printf("no more semid_ds's available\n");
687 #endif
688 			return(ENOSPC);
689 		}
690 #ifdef SEM_DEBUG
691 		printf("semid %d is available\n", semid);
692 #endif
693 		sema[semid].sem_perm.key = key;
694 		sema[semid].sem_perm.cuid = cred->cr_uid;
695 		sema[semid].sem_perm.uid = cred->cr_uid;
696 		sema[semid].sem_perm.cgid = cred->cr_gid;
697 		sema[semid].sem_perm.gid = cred->cr_gid;
698 		sema[semid].sem_perm.mode = (semflg & 0777) | SEM_ALLOC;
699 		sema[semid].sem_perm.seq =
700 		    (sema[semid].sem_perm.seq + 1) & 0x7fff;
701 		sema[semid].sem_nsems = nsems;
702 		sema[semid].sem_otime = 0;
703 		sema[semid].sem_ctime = time_second;
704 		sema[semid].sem_base = &sem[semtot];
705 		semtot += nsems;
706 		bzero(sema[semid].sem_base,
707 		    sizeof(sema[semid].sem_base[0])*nsems);
708 #ifdef SEM_DEBUG
709 		printf("sembase = 0x%x, next = 0x%x\n", sema[semid].sem_base,
710 		    &sem[semtot]);
711 #endif
712 	} else {
713 #ifdef SEM_DEBUG
714 		printf("didn't find it and wasn't asked to create it\n");
715 #endif
716 		return(ENOENT);
717 	}
718 
719 found:
720 	p->p_retval[0] = IXSEQ_TO_IPCID(semid, sema[semid].sem_perm);
721 	return(0);
722 }
723 
724 #ifndef _SYS_SYSPROTO_H_
725 struct semop_args {
726 	int	semid;
727 	struct	sembuf *sops;
728 	int	nsops;
729 };
730 #endif
731 
732 int
733 semop(p, uap)
734 	struct proc *p;
735 	register struct semop_args *uap;
736 {
737 	int semid = uap->semid;
738 	int nsops = uap->nsops;
739 	struct sembuf sops[MAX_SOPS];
740 	register struct semid_ds *semaptr;
741 	register struct sembuf *sopptr;
742 	register struct sem *semptr;
743 	struct sem_undo *suptr = NULL;
744 	int i, j, eval;
745 	int do_wakeup, do_undos;
746 
747 #ifdef SEM_DEBUG
748 	printf("call to semop(%d, 0x%x, %d)\n", semid, sops, nsops);
749 #endif
750 
751 	if (!jail_sysvipc_allowed && p->p_prison != NULL)
752 		return (ENOSYS);
753 
754 	semid = IPCID_TO_IX(semid);	/* Convert back to zero origin */
755 
756 	if (semid < 0 || semid >= seminfo.semmsl)
757 		return(EINVAL);
758 
759 	semaptr = &sema[semid];
760 	if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0)
761 		return(EINVAL);
762 	if (semaptr->sem_perm.seq != IPCID_TO_SEQ(uap->semid))
763 		return(EINVAL);
764 
765 	if ((eval = ipcperm(p, &semaptr->sem_perm, IPC_W))) {
766 #ifdef SEM_DEBUG
767 		printf("eval = %d from ipaccess\n", eval);
768 #endif
769 		return(eval);
770 	}
771 
772 	if (nsops > MAX_SOPS) {
773 #ifdef SEM_DEBUG
774 		printf("too many sops (max=%d, nsops=%d)\n", MAX_SOPS, nsops);
775 #endif
776 		return(E2BIG);
777 	}
778 
779 	if ((eval = copyin(uap->sops, &sops, nsops * sizeof(sops[0]))) != 0) {
780 #ifdef SEM_DEBUG
781 		printf("eval = %d from copyin(%08x, %08x, %d)\n", eval,
782 		    uap->sops, &sops, nsops * sizeof(sops[0]));
783 #endif
784 		return(eval);
785 	}
786 
787 	/*
788 	 * Loop trying to satisfy the vector of requests.
789 	 * If we reach a point where we must wait, any requests already
790 	 * performed are rolled back and we go to sleep until some other
791 	 * process wakes us up.  At this point, we start all over again.
792 	 *
793 	 * This ensures that from the perspective of other tasks, a set
794 	 * of requests is atomic (never partially satisfied).
795 	 */
796 	do_undos = 0;
797 
798 	for (;;) {
799 		do_wakeup = 0;
800 
801 		for (i = 0; i < nsops; i++) {
802 			sopptr = &sops[i];
803 
804 			if (sopptr->sem_num >= semaptr->sem_nsems)
805 				return(EFBIG);
806 
807 			semptr = &semaptr->sem_base[sopptr->sem_num];
808 
809 #ifdef SEM_DEBUG
810 			printf("semop:  semaptr=%x, sem_base=%x, semptr=%x, sem[%d]=%d : op=%d, flag=%s\n",
811 			    semaptr, semaptr->sem_base, semptr,
812 			    sopptr->sem_num, semptr->semval, sopptr->sem_op,
813 			    (sopptr->sem_flg & IPC_NOWAIT) ? "nowait" : "wait");
814 #endif
815 
816 			if (sopptr->sem_op < 0) {
817 				if (semptr->semval + sopptr->sem_op < 0) {
818 #ifdef SEM_DEBUG
819 					printf("semop:  can't do it now\n");
820 #endif
821 					break;
822 				} else {
823 					semptr->semval += sopptr->sem_op;
824 					if (semptr->semval == 0 &&
825 					    semptr->semzcnt > 0)
826 						do_wakeup = 1;
827 				}
828 				if (sopptr->sem_flg & SEM_UNDO)
829 					do_undos = 1;
830 			} else if (sopptr->sem_op == 0) {
831 				if (semptr->semval > 0) {
832 #ifdef SEM_DEBUG
833 					printf("semop:  not zero now\n");
834 #endif
835 					break;
836 				}
837 			} else {
838 				if (semptr->semncnt > 0)
839 					do_wakeup = 1;
840 				semptr->semval += sopptr->sem_op;
841 				if (sopptr->sem_flg & SEM_UNDO)
842 					do_undos = 1;
843 			}
844 		}
845 
846 		/*
847 		 * Did we get through the entire vector?
848 		 */
849 		if (i >= nsops)
850 			goto done;
851 
852 		/*
853 		 * No ... rollback anything that we've already done
854 		 */
855 #ifdef SEM_DEBUG
856 		printf("semop:  rollback 0 through %d\n", i-1);
857 #endif
858 		for (j = 0; j < i; j++)
859 			semaptr->sem_base[sops[j].sem_num].semval -=
860 			    sops[j].sem_op;
861 
862 		/*
863 		 * If the request that we couldn't satisfy has the
864 		 * NOWAIT flag set then return with EAGAIN.
865 		 */
866 		if (sopptr->sem_flg & IPC_NOWAIT)
867 			return(EAGAIN);
868 
869 		if (sopptr->sem_op == 0)
870 			semptr->semzcnt++;
871 		else
872 			semptr->semncnt++;
873 
874 #ifdef SEM_DEBUG
875 		printf("semop:  good night!\n");
876 #endif
877 		eval = tsleep((caddr_t)semaptr, (PZERO - 4) | PCATCH,
878 		    "semwait", 0);
879 #ifdef SEM_DEBUG
880 		printf("semop:  good morning (eval=%d)!\n", eval);
881 #endif
882 
883 		suptr = NULL;	/* sem_undo may have been reallocated */
884 
885 		if (eval != 0)
886 			return(EINTR);
887 #ifdef SEM_DEBUG
888 		printf("semop:  good morning!\n");
889 #endif
890 
891 		/*
892 		 * Make sure that the semaphore still exists
893 		 */
894 		if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0 ||
895 		    semaptr->sem_perm.seq != IPCID_TO_SEQ(uap->semid))
896 			return(EIDRM);
897 
898 		/*
899 		 * The semaphore is still alive.  Readjust the count of
900 		 * waiting processes.
901 		 */
902 		if (sopptr->sem_op == 0)
903 			semptr->semzcnt--;
904 		else
905 			semptr->semncnt--;
906 	}
907 
908 done:
909 	/*
910 	 * Process any SEM_UNDO requests.
911 	 */
912 	if (do_undos) {
913 		for (i = 0; i < nsops; i++) {
914 			/*
915 			 * We only need to deal with SEM_UNDO's for non-zero
916 			 * op's.
917 			 */
918 			int adjval;
919 
920 			if ((sops[i].sem_flg & SEM_UNDO) == 0)
921 				continue;
922 			adjval = sops[i].sem_op;
923 			if (adjval == 0)
924 				continue;
925 			eval = semundo_adjust(p, &suptr, semid,
926 			    sops[i].sem_num, -adjval);
927 			if (eval == 0)
928 				continue;
929 
930 			/*
931 			 * Oh-Oh!  We ran out of either sem_undo's or undo's.
932 			 * Rollback the adjustments to this point and then
933 			 * rollback the semaphore ups and down so we can return
934 			 * with an error with all structures restored.  We
935 			 * rollback the undo's in the exact reverse order that
936 			 * we applied them.  This guarantees that we won't run
937 			 * out of space as we roll things back out.
938 			 */
939 			for (j = i - 1; j >= 0; j--) {
940 				if ((sops[j].sem_flg & SEM_UNDO) == 0)
941 					continue;
942 				adjval = sops[j].sem_op;
943 				if (adjval == 0)
944 					continue;
945 				if (semundo_adjust(p, &suptr, semid,
946 				    sops[j].sem_num, adjval) != 0)
947 					panic("semop - can't undo undos");
948 			}
949 
950 			for (j = 0; j < nsops; j++)
951 				semaptr->sem_base[sops[j].sem_num].semval -=
952 				    sops[j].sem_op;
953 
954 #ifdef SEM_DEBUG
955 			printf("eval = %d from semundo_adjust\n", eval);
956 #endif
957 			return(eval);
958 		} /* loop through the sops */
959 	} /* if (do_undos) */
960 
961 	/* We're definitely done - set the sempid's */
962 	for (i = 0; i < nsops; i++) {
963 		sopptr = &sops[i];
964 		semptr = &semaptr->sem_base[sopptr->sem_num];
965 		semptr->sempid = p->p_pid;
966 	}
967 
968 	/* Do a wakeup if any semaphore was up'd. */
969 	if (do_wakeup) {
970 #ifdef SEM_DEBUG
971 		printf("semop:  doing wakeup\n");
972 #endif
973 		wakeup((caddr_t)semaptr);
974 #ifdef SEM_DEBUG
975 		printf("semop:  back from wakeup\n");
976 #endif
977 	}
978 #ifdef SEM_DEBUG
979 	printf("semop:  done\n");
980 #endif
981 	p->p_retval[0] = 0;
982 	return(0);
983 }
984 
985 /*
986  * Go through the undo structures for this process and apply the adjustments to
987  * semaphores.
988  */
989 static void
990 semexit_myhook(p)
991 	struct proc *p;
992 {
993 	register struct sem_undo *suptr;
994 	register struct sem_undo **supptr;
995 	int did_something;
996 
997 	did_something = 0;
998 
999 	/*
1000 	 * Go through the chain of undo vectors looking for one
1001 	 * associated with this process.
1002 	 */
1003 
1004 	for (supptr = &semu_list; (suptr = *supptr) != NULL;
1005 	    supptr = &suptr->un_next) {
1006 		if (suptr->un_proc == p)
1007 			break;
1008 	}
1009 
1010 	if (suptr == NULL)
1011 		return;
1012 
1013 #ifdef SEM_DEBUG
1014 	printf("proc @%08x has undo structure with %d entries\n", p,
1015 	    suptr->un_cnt);
1016 #endif
1017 
1018 	/*
1019 	 * If there are any active undo elements then process them.
1020 	 */
1021 	if (suptr->un_cnt > 0) {
1022 		int ix;
1023 
1024 		for (ix = 0; ix < suptr->un_cnt; ix++) {
1025 			int semid = suptr->un_ent[ix].un_id;
1026 			int semnum = suptr->un_ent[ix].un_num;
1027 			int adjval = suptr->un_ent[ix].un_adjval;
1028 			struct semid_ds *semaptr;
1029 
1030 			semaptr = &sema[semid];
1031 			if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0)
1032 				panic("semexit - semid not allocated");
1033 			if (semnum >= semaptr->sem_nsems)
1034 				panic("semexit - semnum out of range");
1035 
1036 #ifdef SEM_DEBUG
1037 			printf("semexit:  %08x id=%d num=%d(adj=%d) ; sem=%d\n",
1038 			    suptr->un_proc, suptr->un_ent[ix].un_id,
1039 			    suptr->un_ent[ix].un_num,
1040 			    suptr->un_ent[ix].un_adjval,
1041 			    semaptr->sem_base[semnum].semval);
1042 #endif
1043 
1044 			if (adjval < 0) {
1045 				if (semaptr->sem_base[semnum].semval < -adjval)
1046 					semaptr->sem_base[semnum].semval = 0;
1047 				else
1048 					semaptr->sem_base[semnum].semval +=
1049 					    adjval;
1050 			} else
1051 				semaptr->sem_base[semnum].semval += adjval;
1052 
1053 			wakeup((caddr_t)semaptr);
1054 #ifdef SEM_DEBUG
1055 			printf("semexit:  back from wakeup\n");
1056 #endif
1057 		}
1058 	}
1059 
1060 	/*
1061 	 * Deallocate the undo vector.
1062 	 */
1063 #ifdef SEM_DEBUG
1064 	printf("removing vector\n");
1065 #endif
1066 	suptr->un_proc = NULL;
1067 	*supptr = suptr->un_next;
1068 }
1069