sysv_sem.c revision 83341 
1/* $FreeBSD: head/sys/kern/sysv_sem.c 83341 2001-09-11 12:20:24Z dd $ */
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/lock.h>
19#include <sys/mutex.h>
20#include <sys/sem.h>
21#include <sys/syscall.h>
22#include <sys/sysent.h>
23#include <sys/sysctl.h>
24#include <sys/malloc.h>
25#include <sys/jail.h>
26
27static MALLOC_DEFINE(M_SEM, "sem", "SVID compatible semaphores");
28
29static void seminit __P((void));
30static int sysvsem_modload __P((struct module *, int, void *));
31static int semunload __P((void));
32static void semexit_myhook __P((struct proc *p));
33static int sysctl_sema __P((SYSCTL_HANDLER_ARGS));
34
35#ifndef _SYS_SYSPROTO_H_
36struct __semctl_args;
37int __semctl __P((struct proc *p, struct __semctl_args *uap));
38struct semget_args;
39int semget __P((struct proc *p, struct semget_args *uap));
40struct semop_args;
41int semop __P((struct proc *p, struct semop_args *uap));
42#endif
43
44static struct sem_undo *semu_alloc __P((struct proc *p));
45static int semundo_adjust __P((struct proc *p, struct sem_undo **supptr,
46		int semid, int semnum, int adjval));
47static void semundo_clear __P((int semid, int semnum));
48
49/* XXX casting to (sy_call_t *) is bogus, as usual. */
50static sy_call_t *semcalls[] = {
51	(sy_call_t *)__semctl, (sy_call_t *)semget,
52	(sy_call_t *)semop
53};
54
55static int	semtot = 0;
56static struct semid_ds *sema;	/* semaphore id pool */
57static struct sem *sem;		/* semaphore pool */
58static struct sem_undo *semu_list; /* list of active undo structures */
59static int	*semu;		/* undo structure pool */
60
61struct sem {
62	u_short	semval;		/* semaphore value */
63	pid_t	sempid;		/* pid of last operation */
64	u_short	semncnt;	/* # awaiting semval > cval */
65	u_short	semzcnt;	/* # awaiting semval = 0 */
66};
67
68/*
69 * Undo structure (one per process)
70 */
71struct sem_undo {
72	struct	sem_undo *un_next;	/* ptr to next active undo structure */
73	struct	proc *un_proc;		/* owner of this structure */
74	short	un_cnt;			/* # of active entries */
75	struct undo {
76		short	un_adjval;	/* adjust on exit values */
77		short	un_num;		/* semaphore # */
78		int	un_id;		/* semid */
79	} un_ent[1];			/* undo entries */
80};
81
82/*
83 * Configuration parameters
84 */
85#ifndef SEMMNI
86#define SEMMNI	10		/* # of semaphore identifiers */
87#endif
88#ifndef SEMMNS
89#define SEMMNS	60		/* # of semaphores in system */
90#endif
91#ifndef SEMUME
92#define SEMUME	10		/* max # of undo entries per process */
93#endif
94#ifndef SEMMNU
95#define SEMMNU	30		/* # of undo structures in system */
96#endif
97
98/* shouldn't need tuning */
99#ifndef SEMMAP
100#define SEMMAP	30		/* # of entries in semaphore map */
101#endif
102#ifndef SEMMSL
103#define SEMMSL	SEMMNS		/* max # of semaphores per id */
104#endif
105#ifndef SEMOPM
106#define SEMOPM	100		/* max # of operations per semop call */
107#endif
108
109#define SEMVMX	32767		/* semaphore maximum value */
110#define SEMAEM	16384		/* adjust on exit max value */
111
112/*
113 * Due to the way semaphore memory is allocated, we have to ensure that
114 * SEMUSZ is properly aligned.
115 */
116
117#define SEM_ALIGN(bytes) (((bytes) + (sizeof(long) - 1)) & ~(sizeof(long) - 1))
118
119/* actual size of an undo structure */
120#define SEMUSZ	SEM_ALIGN(offsetof(struct sem_undo, un_ent[SEMUME]))
121
122/*
123 * Macro to find a particular sem_undo vector
124 */
125#define SEMU(ix)	((struct sem_undo *)(((intptr_t)semu)+ix * seminfo.semusz))
126
127/*
128 * semaphore info struct
129 */
130struct seminfo seminfo = {
131                SEMMAP,         /* # of entries in semaphore map */
132                SEMMNI,         /* # of semaphore identifiers */
133                SEMMNS,         /* # of semaphores in system */
134                SEMMNU,         /* # of undo structures in system */
135                SEMMSL,         /* max # of semaphores per id */
136                SEMOPM,         /* max # of operations per semop call */
137                SEMUME,         /* max # of undo entries per process */
138                SEMUSZ,         /* size in bytes of undo structure */
139                SEMVMX,         /* semaphore maximum value */
140                SEMAEM          /* adjust on exit max value */
141};
142
143SYSCTL_DECL(_kern_ipc);
144SYSCTL_INT(_kern_ipc, OID_AUTO, semmap, CTLFLAG_RW, &seminfo.semmap, 0, "");
145SYSCTL_INT(_kern_ipc, OID_AUTO, semmni, CTLFLAG_RD, &seminfo.semmni, 0, "");
146SYSCTL_INT(_kern_ipc, OID_AUTO, semmns, CTLFLAG_RD, &seminfo.semmns, 0, "");
147SYSCTL_INT(_kern_ipc, OID_AUTO, semmnu, CTLFLAG_RD, &seminfo.semmnu, 0, "");
148SYSCTL_INT(_kern_ipc, OID_AUTO, semmsl, CTLFLAG_RW, &seminfo.semmsl, 0, "");
149SYSCTL_INT(_kern_ipc, OID_AUTO, semopm, CTLFLAG_RD, &seminfo.semopm, 0, "");
150SYSCTL_INT(_kern_ipc, OID_AUTO, semume, CTLFLAG_RD, &seminfo.semume, 0, "");
151SYSCTL_INT(_kern_ipc, OID_AUTO, semusz, CTLFLAG_RD, &seminfo.semusz, 0, "");
152SYSCTL_INT(_kern_ipc, OID_AUTO, semvmx, CTLFLAG_RW, &seminfo.semvmx, 0, "");
153SYSCTL_INT(_kern_ipc, OID_AUTO, semaem, CTLFLAG_RW, &seminfo.semaem, 0, "");
154SYSCTL_PROC(_kern_ipc, OID_AUTO, sema, CTLFLAG_RD,
155    NULL, 0, sysctl_sema, "", "");
156
157#if 0
158RO seminfo.semmap	/* SEMMAP unused */
159RO seminfo.semmni
160RO seminfo.semmns
161RO seminfo.semmnu	/* undo entries per system */
162RW seminfo.semmsl
163RO seminfo.semopm	/* SEMOPM unused */
164RO seminfo.semume
165RO seminfo.semusz	/* param - derived from SEMUME for per-proc sizeof */
166RO seminfo.semvmx	/* SEMVMX unused - user param */
167RO seminfo.semaem	/* SEMAEM unused - user param */
168#endif
169
170static void
171seminit(void)
172{
173	register int i;
174
175	sem = malloc(sizeof(struct sem) * seminfo.semmns, M_SEM, M_WAITOK);
176	if (sem == NULL)
177		panic("sem is NULL");
178	sema = malloc(sizeof(struct semid_ds) * seminfo.semmni, M_SEM, M_WAITOK);
179	if (sema == NULL)
180		panic("sema is NULL");
181	semu = malloc(seminfo.semmnu * seminfo.semusz, M_SEM, M_WAITOK);
182	if (semu == NULL)
183		panic("semu is NULL");
184
185	for (i = 0; i < seminfo.semmni; i++) {
186		sema[i].sem_base = 0;
187		sema[i].sem_perm.mode = 0;
188	}
189	for (i = 0; i < seminfo.semmnu; i++) {
190		register struct sem_undo *suptr = SEMU(i);
191		suptr->un_proc = NULL;
192	}
193	semu_list = NULL;
194	semexit_hook = &semexit_myhook;
195}
196
197static int
198semunload(void)
199{
200
201	if (semtot != 0)
202		return (EBUSY);
203
204	free(sem, M_SEM);
205	free(sema, M_SEM);
206	free(semu, M_SEM);
207	semexit_hook = NULL;
208	return (0);
209}
210
211static int
212sysvsem_modload(struct module *module, int cmd, void *arg)
213{
214	int error = 0;
215
216	switch (cmd) {
217	case MOD_LOAD:
218		seminit();
219		break;
220	case MOD_UNLOAD:
221		error = semunload();
222		break;
223	case MOD_SHUTDOWN:
224		break;
225	default:
226		error = EINVAL;
227		break;
228	}
229	return (error);
230}
231
232static moduledata_t sysvsem_mod = {
233	"sysvsem",
234	&sysvsem_modload,
235	NULL
236};
237
238SYSCALL_MODULE_HELPER(semsys, 5);
239SYSCALL_MODULE_HELPER(__semctl, 4);
240SYSCALL_MODULE_HELPER(semget, 3);
241SYSCALL_MODULE_HELPER(semop, 3);
242
243DECLARE_MODULE(sysvsem, sysvsem_mod,
244	SI_SUB_SYSV_SEM, SI_ORDER_FIRST);
245MODULE_VERSION(sysvsem, 1);
246
247/*
248 * Entry point for all SEM calls
249 *
250 * MPSAFE
251 */
252int
253semsys(p, uap)
254	struct proc *p;
255	/* XXX actually varargs. */
256	struct semsys_args /* {
257		u_int	which;
258		int	a2;
259		int	a3;
260		int	a4;
261		int	a5;
262	} */ *uap;
263{
264	int error;
265
266	mtx_lock(&Giant);
267
268	if (!jail_sysvipc_allowed && jailed(p->p_ucred)) {
269		error = ENOSYS;
270		goto done2;
271	}
272
273	if (uap->which >= sizeof(semcalls)/sizeof(semcalls[0])) {
274		error = EINVAL;
275		goto done2;
276	}
277	error = (*semcalls[uap->which])(p, &uap->a2);
278done2:
279	mtx_unlock(&Giant);
280	return (error);
281}
282
283/*
284 * Allocate a new sem_undo structure for a process
285 * (returns ptr to structure or NULL if no more room)
286 */
287
288static struct sem_undo *
289semu_alloc(p)
290	struct proc *p;
291{
292	register int i;
293	register struct sem_undo *suptr;
294	register struct sem_undo **supptr;
295	int attempt;
296
297	/*
298	 * Try twice to allocate something.
299	 * (we'll purge any empty structures after the first pass so
300	 * two passes are always enough)
301	 */
302
303	for (attempt = 0; attempt < 2; attempt++) {
304		/*
305		 * Look for a free structure.
306		 * Fill it in and return it if we find one.
307		 */
308
309		for (i = 0; i < seminfo.semmnu; i++) {
310			suptr = SEMU(i);
311			if (suptr->un_proc == NULL) {
312				suptr->un_next = semu_list;
313				semu_list = suptr;
314				suptr->un_cnt = 0;
315				suptr->un_proc = p;
316				return(suptr);
317			}
318		}
319
320		/*
321		 * We didn't find a free one, if this is the first attempt
322		 * then try to free some structures.
323		 */
324
325		if (attempt == 0) {
326			/* All the structures are in use - try to free some */
327			int did_something = 0;
328
329			supptr = &semu_list;
330			while ((suptr = *supptr) != NULL) {
331				if (suptr->un_cnt == 0)  {
332					suptr->un_proc = NULL;
333					*supptr = suptr->un_next;
334					did_something = 1;
335				} else
336					supptr = &(suptr->un_next);
337			}
338
339			/* If we didn't free anything then just give-up */
340			if (!did_something)
341				return(NULL);
342		} else {
343			/*
344			 * The second pass failed even though we freed
345			 * something after the first pass!
346			 * This is IMPOSSIBLE!
347			 */
348			panic("semu_alloc - second attempt failed");
349		}
350	}
351	return (NULL);
352}
353
354/*
355 * Adjust a particular entry for a particular proc
356 */
357
358static int
359semundo_adjust(p, supptr, semid, semnum, adjval)
360	register struct proc *p;
361	struct sem_undo **supptr;
362	int semid, semnum;
363	int adjval;
364{
365	register struct sem_undo *suptr;
366	register struct undo *sunptr;
367	int i;
368
369	/* Look for and remember the sem_undo if the caller doesn't provide
370	   it */
371
372	suptr = *supptr;
373	if (suptr == NULL) {
374		for (suptr = semu_list; suptr != NULL;
375		    suptr = suptr->un_next) {
376			if (suptr->un_proc == p) {
377				*supptr = suptr;
378				break;
379			}
380		}
381		if (suptr == NULL) {
382			if (adjval == 0)
383				return(0);
384			suptr = semu_alloc(p);
385			if (suptr == NULL)
386				return(ENOSPC);
387			*supptr = suptr;
388		}
389	}
390
391	/*
392	 * Look for the requested entry and adjust it (delete if adjval becomes
393	 * 0).
394	 */
395	sunptr = &suptr->un_ent[0];
396	for (i = 0; i < suptr->un_cnt; i++, sunptr++) {
397		if (sunptr->un_id != semid || sunptr->un_num != semnum)
398			continue;
399		if (adjval == 0)
400			sunptr->un_adjval = 0;
401		else
402			sunptr->un_adjval += adjval;
403		if (sunptr->un_adjval == 0) {
404			suptr->un_cnt--;
405			if (i < suptr->un_cnt)
406				suptr->un_ent[i] =
407				    suptr->un_ent[suptr->un_cnt];
408		}
409		return(0);
410	}
411
412	/* Didn't find the right entry - create it */
413	if (adjval == 0)
414		return(0);
415	if (suptr->un_cnt != seminfo.semume) {
416		sunptr = &suptr->un_ent[suptr->un_cnt];
417		suptr->un_cnt++;
418		sunptr->un_adjval = adjval;
419		sunptr->un_id = semid; sunptr->un_num = semnum;
420	} else
421		return(EINVAL);
422	return(0);
423}
424
425static void
426semundo_clear(semid, semnum)
427	int semid, semnum;
428{
429	register struct sem_undo *suptr;
430
431	for (suptr = semu_list; suptr != NULL; suptr = suptr->un_next) {
432		register struct undo *sunptr = &suptr->un_ent[0];
433		register int i = 0;
434
435		while (i < suptr->un_cnt) {
436			if (sunptr->un_id == semid) {
437				if (semnum == -1 || sunptr->un_num == semnum) {
438					suptr->un_cnt--;
439					if (i < suptr->un_cnt) {
440						suptr->un_ent[i] =
441						  suptr->un_ent[suptr->un_cnt];
442						continue;
443					}
444				}
445				if (semnum != -1)
446					break;
447			}
448			i++, sunptr++;
449		}
450	}
451}
452
453/*
454 * Note that the user-mode half of this passes a union, not a pointer
455 */
456#ifndef _SYS_SYSPROTO_H_
457struct __semctl_args {
458	int	semid;
459	int	semnum;
460	int	cmd;
461	union	semun *arg;
462};
463#endif
464
465/*
466 * MPSAFE
467 */
468int
469__semctl(p, uap)
470	struct proc *p;
471	register struct __semctl_args *uap;
472{
473	int semid = uap->semid;
474	int semnum = uap->semnum;
475	int cmd = uap->cmd;
476	union semun *arg = uap->arg;
477	union semun real_arg;
478	struct ucred *cred = p->p_ucred;
479	int i, rval, error;
480	struct semid_ds sbuf;
481	register struct semid_ds *semaptr;
482
483#ifdef SEM_DEBUG
484	printf("call to semctl(%d, %d, %d, 0x%x)\n", semid, semnum, cmd, arg);
485#endif
486	mtx_lock(&Giant);
487
488	if (!jail_sysvipc_allowed && jailed(p->p_ucred)) {
489		error = ENOSYS;
490		goto done2;
491	}
492
493	semid = IPCID_TO_IX(semid);
494	if (semid < 0 || semid >= seminfo.semmsl) {
495		error = EINVAL;
496		goto done2;
497	}
498
499	semaptr = &sema[semid];
500	if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0 ||
501	    semaptr->sem_perm.seq != IPCID_TO_SEQ(uap->semid)) {
502		error = EINVAL;
503		goto done2;
504	}
505
506	error = 0;
507	rval = 0;
508
509	switch (cmd) {
510	case IPC_RMID:
511		if ((error = ipcperm(p, &semaptr->sem_perm, IPC_M)))
512			goto done2;
513		semaptr->sem_perm.cuid = cred->cr_uid;
514		semaptr->sem_perm.uid = cred->cr_uid;
515		semtot -= semaptr->sem_nsems;
516		for (i = semaptr->sem_base - sem; i < semtot; i++)
517			sem[i] = sem[i + semaptr->sem_nsems];
518		for (i = 0; i < seminfo.semmni; i++) {
519			if ((sema[i].sem_perm.mode & SEM_ALLOC) &&
520			    sema[i].sem_base > semaptr->sem_base)
521				sema[i].sem_base -= semaptr->sem_nsems;
522		}
523		semaptr->sem_perm.mode = 0;
524		semundo_clear(semid, -1);
525		wakeup((caddr_t)semaptr);
526		break;
527
528	case IPC_SET:
529		if ((error = ipcperm(p, &semaptr->sem_perm, IPC_M)))
530			goto done2;
531		if ((error = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
532			goto done2;
533		if ((error = copyin(real_arg.buf, (caddr_t)&sbuf,
534		    sizeof(sbuf))) != 0) {
535			goto done2;
536		}
537		semaptr->sem_perm.uid = sbuf.sem_perm.uid;
538		semaptr->sem_perm.gid = sbuf.sem_perm.gid;
539		semaptr->sem_perm.mode = (semaptr->sem_perm.mode & ~0777) |
540		    (sbuf.sem_perm.mode & 0777);
541		semaptr->sem_ctime = time_second;
542		break;
543
544	case IPC_STAT:
545		if ((error = ipcperm(p, &semaptr->sem_perm, IPC_R)))
546			goto done2;
547		if ((error = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
548			goto done2;
549		error = copyout((caddr_t)semaptr, real_arg.buf,
550				sizeof(struct semid_ds));
551		break;
552
553	case GETNCNT:
554		if ((error = ipcperm(p, &semaptr->sem_perm, IPC_R)))
555			goto done2;
556		if (semnum < 0 || semnum >= semaptr->sem_nsems) {
557			error = EINVAL;
558			goto done2;
559		}
560		rval = semaptr->sem_base[semnum].semncnt;
561		break;
562
563	case GETPID:
564		if ((error = ipcperm(p, &semaptr->sem_perm, IPC_R)))
565			goto done2;
566		if (semnum < 0 || semnum >= semaptr->sem_nsems) {
567			error = EINVAL;
568			goto done2;
569		}
570		rval = semaptr->sem_base[semnum].sempid;
571		break;
572
573	case GETVAL:
574		if ((error = ipcperm(p, &semaptr->sem_perm, IPC_R)))
575			goto done2;
576		if (semnum < 0 || semnum >= semaptr->sem_nsems) {
577			error = EINVAL;
578			goto done2;
579		}
580		rval = semaptr->sem_base[semnum].semval;
581		break;
582
583	case GETALL:
584		if ((error = ipcperm(p, &semaptr->sem_perm, IPC_R)))
585			goto done2;
586		if ((error = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
587			goto done2;
588		for (i = 0; i < semaptr->sem_nsems; i++) {
589			error = copyout((caddr_t)&semaptr->sem_base[i].semval,
590			    &real_arg.array[i], sizeof(real_arg.array[0]));
591			if (error != 0)
592				break;
593		}
594		break;
595
596	case GETZCNT:
597		if ((error = ipcperm(p, &semaptr->sem_perm, IPC_R)))
598			goto done2;
599		if (semnum < 0 || semnum >= semaptr->sem_nsems) {
600			error = EINVAL;
601			goto done2;
602		}
603		rval = semaptr->sem_base[semnum].semzcnt;
604		break;
605
606	case SETVAL:
607		if ((error = ipcperm(p, &semaptr->sem_perm, IPC_W)))
608			goto done2;
609		if (semnum < 0 || semnum >= semaptr->sem_nsems) {
610			error = EINVAL;
611			goto done2;
612		}
613		if ((error = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
614			goto done2;
615		semaptr->sem_base[semnum].semval = real_arg.val;
616		semundo_clear(semid, semnum);
617		wakeup((caddr_t)semaptr);
618		break;
619
620	case SETALL:
621		if ((error = ipcperm(p, &semaptr->sem_perm, IPC_W)))
622			goto done2;
623		if ((error = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
624			goto done2;
625		for (i = 0; i < semaptr->sem_nsems; i++) {
626			error = copyin(&real_arg.array[i],
627			    (caddr_t)&semaptr->sem_base[i].semval,
628			    sizeof(real_arg.array[0]));
629			if (error != 0)
630				break;
631		}
632		semundo_clear(semid, -1);
633		wakeup((caddr_t)semaptr);
634		break;
635
636	default:
637		error = EINVAL;
638		break;
639	}
640
641	if (error == 0)
642		p->p_retval[0] = rval;
643done2:
644	mtx_unlock(&Giant);
645	return(error);
646}
647
648#ifndef _SYS_SYSPROTO_H_
649struct semget_args {
650	key_t	key;
651	int	nsems;
652	int	semflg;
653};
654#endif
655
656/*
657 * MPSAFE
658 */
659int
660semget(p, uap)
661	struct proc *p;
662	register struct semget_args *uap;
663{
664	int semid, error = 0;
665	int key = uap->key;
666	int nsems = uap->nsems;
667	int semflg = uap->semflg;
668	struct ucred *cred = p->p_ucred;
669
670#ifdef SEM_DEBUG
671	printf("semget(0x%x, %d, 0%o)\n", key, nsems, semflg);
672#endif
673	mtx_lock(&Giant);
674
675	if (!jail_sysvipc_allowed && jailed(p->p_ucred)) {
676		error = ENOSYS;
677		goto done2;
678	}
679
680	if (key != IPC_PRIVATE) {
681		for (semid = 0; semid < seminfo.semmni; semid++) {
682			if ((sema[semid].sem_perm.mode & SEM_ALLOC) &&
683			    sema[semid].sem_perm.key == key)
684				break;
685		}
686		if (semid < seminfo.semmni) {
687#ifdef SEM_DEBUG
688			printf("found public key\n");
689#endif
690			if ((error = ipcperm(p, &sema[semid].sem_perm,
691			    semflg & 0700))) {
692				goto done2;
693			}
694			if (nsems > 0 && sema[semid].sem_nsems < nsems) {
695#ifdef SEM_DEBUG
696				printf("too small\n");
697#endif
698				error = EINVAL;
699				goto done2;
700			}
701			if ((semflg & IPC_CREAT) && (semflg & IPC_EXCL)) {
702#ifdef SEM_DEBUG
703				printf("not exclusive\n");
704#endif
705				error = EEXIST;
706				goto done2;
707			}
708			goto found;
709		}
710	}
711
712#ifdef SEM_DEBUG
713	printf("need to allocate the semid_ds\n");
714#endif
715	if (key == IPC_PRIVATE || (semflg & IPC_CREAT)) {
716		if (nsems <= 0 || nsems > seminfo.semmsl) {
717#ifdef SEM_DEBUG
718			printf("nsems out of range (0<%d<=%d)\n", nsems,
719			    seminfo.semmsl);
720#endif
721			error = EINVAL;
722			goto done2;
723		}
724		if (nsems > seminfo.semmns - semtot) {
725#ifdef SEM_DEBUG
726			printf("not enough semaphores left (need %d, got %d)\n",
727			    nsems, seminfo.semmns - semtot);
728#endif
729			error = ENOSPC;
730			goto done2;
731		}
732		for (semid = 0; semid < seminfo.semmni; semid++) {
733			if ((sema[semid].sem_perm.mode & SEM_ALLOC) == 0)
734				break;
735		}
736		if (semid == seminfo.semmni) {
737#ifdef SEM_DEBUG
738			printf("no more semid_ds's available\n");
739#endif
740			error = ENOSPC;
741			goto done2;
742		}
743#ifdef SEM_DEBUG
744		printf("semid %d is available\n", semid);
745#endif
746		sema[semid].sem_perm.key = key;
747		sema[semid].sem_perm.cuid = cred->cr_uid;
748		sema[semid].sem_perm.uid = cred->cr_uid;
749		sema[semid].sem_perm.cgid = cred->cr_gid;
750		sema[semid].sem_perm.gid = cred->cr_gid;
751		sema[semid].sem_perm.mode = (semflg & 0777) | SEM_ALLOC;
752		sema[semid].sem_perm.seq =
753		    (sema[semid].sem_perm.seq + 1) & 0x7fff;
754		sema[semid].sem_nsems = nsems;
755		sema[semid].sem_otime = 0;
756		sema[semid].sem_ctime = time_second;
757		sema[semid].sem_base = &sem[semtot];
758		semtot += nsems;
759		bzero(sema[semid].sem_base,
760		    sizeof(sema[semid].sem_base[0])*nsems);
761#ifdef SEM_DEBUG
762		printf("sembase = 0x%x, next = 0x%x\n", sema[semid].sem_base,
763		    &sem[semtot]);
764#endif
765	} else {
766#ifdef SEM_DEBUG
767		printf("didn't find it and wasn't asked to create it\n");
768#endif
769		error = ENOENT;
770		goto done2;
771	}
772
773found:
774	p->p_retval[0] = IXSEQ_TO_IPCID(semid, sema[semid].sem_perm);
775done2:
776	mtx_unlock(&Giant);
777	return (error);
778}
779
780#ifndef _SYS_SYSPROTO_H_
781struct semop_args {
782	int	semid;
783	struct	sembuf *sops;
784	u_int	nsops;
785};
786#endif
787
788/*
789 * MPSAFE
790 */
791int
792semop(p, uap)
793	struct proc *p;
794	register struct semop_args *uap;
795{
796	int semid = uap->semid;
797	u_int nsops = uap->nsops;
798	struct sembuf sops[MAX_SOPS];
799	register struct semid_ds *semaptr;
800	register struct sembuf *sopptr;
801	register struct sem *semptr;
802	struct sem_undo *suptr = NULL;
803	int i, j, error = 0;
804	int do_wakeup, do_undos;
805
806#ifdef SEM_DEBUG
807	printf("call to semop(%d, 0x%x, %u)\n", semid, sops, nsops);
808#endif
809
810	mtx_lock(&Giant);
811
812	if (!jail_sysvipc_allowed && jailed(p->p_ucred)) {
813		error = ENOSYS;
814		goto done2;
815	}
816
817	semid = IPCID_TO_IX(semid);	/* Convert back to zero origin */
818
819	if (semid < 0 || semid >= seminfo.semmsl) {
820		error = EINVAL;
821		goto done2;
822	}
823
824	semaptr = &sema[semid];
825	if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0) {
826		error = EINVAL;
827		goto done2;
828	}
829	if (semaptr->sem_perm.seq != IPCID_TO_SEQ(uap->semid)) {
830		error = EINVAL;
831		goto done2;
832	}
833
834	if ((error = ipcperm(p, &semaptr->sem_perm, IPC_W))) {
835#ifdef SEM_DEBUG
836		printf("error = %d from ipcperm\n", error);
837#endif
838		goto done2;
839	}
840
841	if (nsops > MAX_SOPS) {
842#ifdef SEM_DEBUG
843		printf("too many sops (max=%d, nsops=%u)\n", MAX_SOPS, nsops);
844#endif
845		error = E2BIG;
846		goto done2;
847	}
848
849	if ((error = copyin(uap->sops, &sops, nsops * sizeof(sops[0]))) != 0) {
850#ifdef SEM_DEBUG
851		printf("error = %d from copyin(%08x, %08x, %u)\n", error,
852		    uap->sops, &sops, nsops * sizeof(sops[0]));
853#endif
854		goto done2;
855	}
856
857	/*
858	 * Loop trying to satisfy the vector of requests.
859	 * If we reach a point where we must wait, any requests already
860	 * performed are rolled back and we go to sleep until some other
861	 * process wakes us up.  At this point, we start all over again.
862	 *
863	 * This ensures that from the perspective of other tasks, a set
864	 * of requests is atomic (never partially satisfied).
865	 */
866	do_undos = 0;
867
868	for (;;) {
869		do_wakeup = 0;
870
871		for (i = 0; i < nsops; i++) {
872			sopptr = &sops[i];
873
874			if (sopptr->sem_num >= semaptr->sem_nsems) {
875				error = EFBIG;
876				goto done2;
877			}
878
879			semptr = &semaptr->sem_base[sopptr->sem_num];
880
881#ifdef SEM_DEBUG
882			printf("semop:  semaptr=%x, sem_base=%x, semptr=%x, sem[%d]=%d : op=%d, flag=%s\n",
883			    semaptr, semaptr->sem_base, semptr,
884			    sopptr->sem_num, semptr->semval, sopptr->sem_op,
885			    (sopptr->sem_flg & IPC_NOWAIT) ? "nowait" : "wait");
886#endif
887
888			if (sopptr->sem_op < 0) {
889				if (semptr->semval + sopptr->sem_op < 0) {
890#ifdef SEM_DEBUG
891					printf("semop:  can't do it now\n");
892#endif
893					break;
894				} else {
895					semptr->semval += sopptr->sem_op;
896					if (semptr->semval == 0 &&
897					    semptr->semzcnt > 0)
898						do_wakeup = 1;
899				}
900				if (sopptr->sem_flg & SEM_UNDO)
901					do_undos = 1;
902			} else if (sopptr->sem_op == 0) {
903				if (semptr->semval > 0) {
904#ifdef SEM_DEBUG
905					printf("semop:  not zero now\n");
906#endif
907					break;
908				}
909			} else {
910				if (semptr->semncnt > 0)
911					do_wakeup = 1;
912				semptr->semval += sopptr->sem_op;
913				if (sopptr->sem_flg & SEM_UNDO)
914					do_undos = 1;
915			}
916		}
917
918		/*
919		 * Did we get through the entire vector?
920		 */
921		if (i >= nsops)
922			goto done;
923
924		/*
925		 * No ... rollback anything that we've already done
926		 */
927#ifdef SEM_DEBUG
928		printf("semop:  rollback 0 through %d\n", i-1);
929#endif
930		for (j = 0; j < i; j++)
931			semaptr->sem_base[sops[j].sem_num].semval -=
932			    sops[j].sem_op;
933
934		/*
935		 * If the request that we couldn't satisfy has the
936		 * NOWAIT flag set then return with EAGAIN.
937		 */
938		if (sopptr->sem_flg & IPC_NOWAIT) {
939			error = EAGAIN;
940			goto done2;
941		}
942
943		if (sopptr->sem_op == 0)
944			semptr->semzcnt++;
945		else
946			semptr->semncnt++;
947
948#ifdef SEM_DEBUG
949		printf("semop:  good night!\n");
950#endif
951		error = tsleep((caddr_t)semaptr, (PZERO - 4) | PCATCH,
952		    "semwait", 0);
953#ifdef SEM_DEBUG
954		printf("semop:  good morning (error=%d)!\n", error);
955#endif
956
957		suptr = NULL;	/* sem_undo may have been reallocated */
958
959		if (error != 0) {
960			error = EINTR;
961			goto done2;
962		}
963#ifdef SEM_DEBUG
964		printf("semop:  good morning!\n");
965#endif
966
967		/*
968		 * Make sure that the semaphore still exists
969		 */
970		if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0 ||
971		    semaptr->sem_perm.seq != IPCID_TO_SEQ(uap->semid)) {
972			error = EIDRM;
973			goto done2;
974		}
975
976		/*
977		 * The semaphore is still alive.  Readjust the count of
978		 * waiting processes.
979		 */
980		if (sopptr->sem_op == 0)
981			semptr->semzcnt--;
982		else
983			semptr->semncnt--;
984	}
985
986done:
987	/*
988	 * Process any SEM_UNDO requests.
989	 */
990	if (do_undos) {
991		for (i = 0; i < nsops; i++) {
992			/*
993			 * We only need to deal with SEM_UNDO's for non-zero
994			 * op's.
995			 */
996			int adjval;
997
998			if ((sops[i].sem_flg & SEM_UNDO) == 0)
999				continue;
1000			adjval = sops[i].sem_op;
1001			if (adjval == 0)
1002				continue;
1003			error = semundo_adjust(p, &suptr, semid,
1004			    sops[i].sem_num, -adjval);
1005			if (error == 0)
1006				continue;
1007
1008			/*
1009			 * Oh-Oh!  We ran out of either sem_undo's or undo's.
1010			 * Rollback the adjustments to this point and then
1011			 * rollback the semaphore ups and down so we can return
1012			 * with an error with all structures restored.  We
1013			 * rollback the undo's in the exact reverse order that
1014			 * we applied them.  This guarantees that we won't run
1015			 * out of space as we roll things back out.
1016			 */
1017			for (j = i - 1; j >= 0; j--) {
1018				if ((sops[j].sem_flg & SEM_UNDO) == 0)
1019					continue;
1020				adjval = sops[j].sem_op;
1021				if (adjval == 0)
1022					continue;
1023				if (semundo_adjust(p, &suptr, semid,
1024				    sops[j].sem_num, adjval) != 0)
1025					panic("semop - can't undo undos");
1026			}
1027
1028			for (j = 0; j < nsops; j++)
1029				semaptr->sem_base[sops[j].sem_num].semval -=
1030				    sops[j].sem_op;
1031
1032#ifdef SEM_DEBUG
1033			printf("error = %d from semundo_adjust\n", error);
1034#endif
1035			goto done2;
1036		} /* loop through the sops */
1037	} /* if (do_undos) */
1038
1039	/* We're definitely done - set the sempid's */
1040	for (i = 0; i < nsops; i++) {
1041		sopptr = &sops[i];
1042		semptr = &semaptr->sem_base[sopptr->sem_num];
1043		semptr->sempid = p->p_pid;
1044	}
1045
1046	/* Do a wakeup if any semaphore was up'd. */
1047	if (do_wakeup) {
1048#ifdef SEM_DEBUG
1049		printf("semop:  doing wakeup\n");
1050#endif
1051		wakeup((caddr_t)semaptr);
1052#ifdef SEM_DEBUG
1053		printf("semop:  back from wakeup\n");
1054#endif
1055	}
1056#ifdef SEM_DEBUG
1057	printf("semop:  done\n");
1058#endif
1059	p->p_retval[0] = 0;
1060done2:
1061	mtx_unlock(&Giant);
1062	return (error);
1063}
1064
1065/*
1066 * Go through the undo structures for this process and apply the adjustments to
1067 * semaphores.
1068 */
1069static void
1070semexit_myhook(p)
1071	struct proc *p;
1072{
1073	register struct sem_undo *suptr;
1074	register struct sem_undo **supptr;
1075	int did_something;
1076
1077	did_something = 0;
1078
1079	/*
1080	 * Go through the chain of undo vectors looking for one
1081	 * associated with this process.
1082	 */
1083
1084	for (supptr = &semu_list; (suptr = *supptr) != NULL;
1085	    supptr = &suptr->un_next) {
1086		if (suptr->un_proc == p)
1087			break;
1088	}
1089
1090	if (suptr == NULL)
1091		return;
1092
1093#ifdef SEM_DEBUG
1094	printf("proc @%08x has undo structure with %d entries\n", p,
1095	    suptr->un_cnt);
1096#endif
1097
1098	/*
1099	 * If there are any active undo elements then process them.
1100	 */
1101	if (suptr->un_cnt > 0) {
1102		int ix;
1103
1104		for (ix = 0; ix < suptr->un_cnt; ix++) {
1105			int semid = suptr->un_ent[ix].un_id;
1106			int semnum = suptr->un_ent[ix].un_num;
1107			int adjval = suptr->un_ent[ix].un_adjval;
1108			struct semid_ds *semaptr;
1109
1110			semaptr = &sema[semid];
1111			if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0)
1112				panic("semexit - semid not allocated");
1113			if (semnum >= semaptr->sem_nsems)
1114				panic("semexit - semnum out of range");
1115
1116#ifdef SEM_DEBUG
1117			printf("semexit:  %08x id=%d num=%d(adj=%d) ; sem=%d\n",
1118			    suptr->un_proc, suptr->un_ent[ix].un_id,
1119			    suptr->un_ent[ix].un_num,
1120			    suptr->un_ent[ix].un_adjval,
1121			    semaptr->sem_base[semnum].semval);
1122#endif
1123
1124			if (adjval < 0) {
1125				if (semaptr->sem_base[semnum].semval < -adjval)
1126					semaptr->sem_base[semnum].semval = 0;
1127				else
1128					semaptr->sem_base[semnum].semval +=
1129					    adjval;
1130			} else
1131				semaptr->sem_base[semnum].semval += adjval;
1132
1133			wakeup((caddr_t)semaptr);
1134#ifdef SEM_DEBUG
1135			printf("semexit:  back from wakeup\n");
1136#endif
1137		}
1138	}
1139
1140	/*
1141	 * Deallocate the undo vector.
1142	 */
1143#ifdef SEM_DEBUG
1144	printf("removing vector\n");
1145#endif
1146	suptr->un_proc = NULL;
1147	*supptr = suptr->un_next;
1148}
1149
1150static int
1151sysctl_sema(SYSCTL_HANDLER_ARGS)
1152{
1153
1154	return (SYSCTL_OUT(req, sema,
1155	    sizeof(struct semid_ds) * seminfo.semmni));
1156}
1157