sysv_sem.c revision 88633 
1/* $FreeBSD: head/sys/kern/sysv_sem.c 88633 2001-12-29 07:13:47Z alfred $ */
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 thread *td, struct __semctl_args *uap));
38struct semget_args;
39int semget __P((struct thread *td, struct semget_args *uap));
40struct semop_args;
41int semop __P((struct thread *td, struct semop_args *uap));
42#endif
43
44static struct sem_undo *semu_alloc __P((struct thread *td));
45static int semundo_adjust __P((struct thread *td, 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	TUNABLE_INT_FETCH("kern.ipc.semmap", &seminfo.semmap);
176	TUNABLE_INT_FETCH("kern.ipc.semmni", &seminfo.semmni);
177	TUNABLE_INT_FETCH("kern.ipc.semmns", &seminfo.semmns);
178	TUNABLE_INT_FETCH("kern.ipc.semmnu", &seminfo.semmnu);
179	TUNABLE_INT_FETCH("kern.ipc.semmsl", &seminfo.semmsl);
180	TUNABLE_INT_FETCH("kern.ipc.semopm", &seminfo.semopm);
181	TUNABLE_INT_FETCH("kern.ipc.semume", &seminfo.semume);
182	TUNABLE_INT_FETCH("kern.ipc.semusz", &seminfo.semusz);
183	TUNABLE_INT_FETCH("kern.ipc.semvmx", &seminfo.semvmx);
184	TUNABLE_INT_FETCH("kern.ipc.semaem", &seminfo.semaem);
185
186	sem = malloc(sizeof(struct sem) * seminfo.semmns, M_SEM, M_WAITOK);
187	if (sem == NULL)
188		panic("sem is NULL");
189	sema = malloc(sizeof(struct semid_ds) * seminfo.semmni, M_SEM, M_WAITOK);
190	if (sema == NULL)
191		panic("sema is NULL");
192	semu = malloc(seminfo.semmnu * seminfo.semusz, M_SEM, M_WAITOK);
193	if (semu == NULL)
194		panic("semu is NULL");
195
196	for (i = 0; i < seminfo.semmni; i++) {
197		sema[i].sem_base = 0;
198		sema[i].sem_perm.mode = 0;
199	}
200	for (i = 0; i < seminfo.semmnu; i++) {
201		register struct sem_undo *suptr = SEMU(i);
202		suptr->un_proc = NULL;
203	}
204	semu_list = NULL;
205	semexit_hook = &semexit_myhook;
206}
207
208static int
209semunload(void)
210{
211
212	if (semtot != 0)
213		return (EBUSY);
214
215	free(sem, M_SEM);
216	free(sema, M_SEM);
217	free(semu, M_SEM);
218	semexit_hook = NULL;
219	return (0);
220}
221
222static int
223sysvsem_modload(struct module *module, int cmd, void *arg)
224{
225	int error = 0;
226
227	switch (cmd) {
228	case MOD_LOAD:
229		seminit();
230		break;
231	case MOD_UNLOAD:
232		error = semunload();
233		break;
234	case MOD_SHUTDOWN:
235		break;
236	default:
237		error = EINVAL;
238		break;
239	}
240	return (error);
241}
242
243static moduledata_t sysvsem_mod = {
244	"sysvsem",
245	&sysvsem_modload,
246	NULL
247};
248
249SYSCALL_MODULE_HELPER(semsys);
250SYSCALL_MODULE_HELPER(__semctl);
251SYSCALL_MODULE_HELPER(semget);
252SYSCALL_MODULE_HELPER(semop);
253
254DECLARE_MODULE(sysvsem, sysvsem_mod,
255	SI_SUB_SYSV_SEM, SI_ORDER_FIRST);
256MODULE_VERSION(sysvsem, 1);
257
258/*
259 * Entry point for all SEM calls
260 *
261 * MPSAFE
262 */
263int
264semsys(td, uap)
265	struct thread *td;
266	/* XXX actually varargs. */
267	struct semsys_args /* {
268		u_int	which;
269		int	a2;
270		int	a3;
271		int	a4;
272		int	a5;
273	} */ *uap;
274{
275	int error;
276
277	mtx_lock(&Giant);
278	if (!jail_sysvipc_allowed && jailed(td->td_proc->p_ucred)) {
279		error = ENOSYS;
280		goto done2;
281	}
282	if (uap->which >= sizeof(semcalls)/sizeof(semcalls[0])) {
283		error = EINVAL;
284		goto done2;
285	}
286	error = (*semcalls[uap->which])(td, &uap->a2);
287done2:
288	mtx_unlock(&Giant);
289	return (error);
290}
291
292/*
293 * Allocate a new sem_undo structure for a process
294 * (returns ptr to structure or NULL if no more room)
295 */
296
297static struct sem_undo *
298semu_alloc(td)
299	struct thread *td;
300{
301	register int i;
302	register struct sem_undo *suptr;
303	register struct sem_undo **supptr;
304	int attempt;
305
306	/*
307	 * Try twice to allocate something.
308	 * (we'll purge any empty structures after the first pass so
309	 * two passes are always enough)
310	 */
311
312	for (attempt = 0; attempt < 2; attempt++) {
313		/*
314		 * Look for a free structure.
315		 * Fill it in and return it if we find one.
316		 */
317
318		for (i = 0; i < seminfo.semmnu; i++) {
319			suptr = SEMU(i);
320			if (suptr->un_proc == NULL) {
321				suptr->un_next = semu_list;
322				semu_list = suptr;
323				suptr->un_cnt = 0;
324				suptr->un_proc = td->td_proc;
325				return(suptr);
326			}
327		}
328
329		/*
330		 * We didn't find a free one, if this is the first attempt
331		 * then try to free some structures.
332		 */
333
334		if (attempt == 0) {
335			/* All the structures are in use - try to free some */
336			int did_something = 0;
337
338			supptr = &semu_list;
339			while ((suptr = *supptr) != NULL) {
340				if (suptr->un_cnt == 0)  {
341					suptr->un_proc = NULL;
342					*supptr = suptr->un_next;
343					did_something = 1;
344				} else
345					supptr = &(suptr->un_next);
346			}
347
348			/* If we didn't free anything then just give-up */
349			if (!did_something)
350				return(NULL);
351		} else {
352			/*
353			 * The second pass failed even though we freed
354			 * something after the first pass!
355			 * This is IMPOSSIBLE!
356			 */
357			panic("semu_alloc - second attempt failed");
358		}
359	}
360	return (NULL);
361}
362
363/*
364 * Adjust a particular entry for a particular proc
365 */
366
367static int
368semundo_adjust(td, supptr, semid, semnum, adjval)
369	register struct thread *td;
370	struct sem_undo **supptr;
371	int semid, semnum;
372	int adjval;
373{
374	struct proc *p = td->td_proc;
375	register struct sem_undo *suptr;
376	register struct undo *sunptr;
377	int i;
378
379	/* Look for and remember the sem_undo if the caller doesn't provide
380	   it */
381
382	suptr = *supptr;
383	if (suptr == NULL) {
384		for (suptr = semu_list; suptr != NULL;
385		    suptr = suptr->un_next) {
386			if (suptr->un_proc == p) {
387				*supptr = suptr;
388				break;
389			}
390		}
391		if (suptr == NULL) {
392			if (adjval == 0)
393				return(0);
394			suptr = semu_alloc(td);
395			if (suptr == NULL)
396				return(ENOSPC);
397			*supptr = suptr;
398		}
399	}
400
401	/*
402	 * Look for the requested entry and adjust it (delete if adjval becomes
403	 * 0).
404	 */
405	sunptr = &suptr->un_ent[0];
406	for (i = 0; i < suptr->un_cnt; i++, sunptr++) {
407		if (sunptr->un_id != semid || sunptr->un_num != semnum)
408			continue;
409		if (adjval != 0) {
410			adjval += sunptr->un_adjval;
411			if (adjval > seminfo.semaem || adjval < -seminfo.semaem)
412				return (ERANGE);
413		}
414		sunptr->un_adjval = adjval;
415		if (sunptr->un_adjval == 0) {
416			suptr->un_cnt--;
417			if (i < suptr->un_cnt)
418				suptr->un_ent[i] =
419				    suptr->un_ent[suptr->un_cnt];
420		}
421		return(0);
422	}
423
424	/* Didn't find the right entry - create it */
425	if (adjval == 0)
426		return(0);
427	if (adjval > seminfo.semaem || adjval < -seminfo.semaem)
428		return (ERANGE);
429	if (suptr->un_cnt != seminfo.semume) {
430		sunptr = &suptr->un_ent[suptr->un_cnt];
431		suptr->un_cnt++;
432		sunptr->un_adjval = adjval;
433		sunptr->un_id = semid; sunptr->un_num = semnum;
434	} else
435		return(EINVAL);
436	return(0);
437}
438
439static void
440semundo_clear(semid, semnum)
441	int semid, semnum;
442{
443	register struct sem_undo *suptr;
444
445	for (suptr = semu_list; suptr != NULL; suptr = suptr->un_next) {
446		register struct undo *sunptr = &suptr->un_ent[0];
447		register int i = 0;
448
449		while (i < suptr->un_cnt) {
450			if (sunptr->un_id == semid) {
451				if (semnum == -1 || sunptr->un_num == semnum) {
452					suptr->un_cnt--;
453					if (i < suptr->un_cnt) {
454						suptr->un_ent[i] =
455						  suptr->un_ent[suptr->un_cnt];
456						continue;
457					}
458				}
459				if (semnum != -1)
460					break;
461			}
462			i++, sunptr++;
463		}
464	}
465}
466
467/*
468 * Note that the user-mode half of this passes a union, not a pointer
469 */
470#ifndef _SYS_SYSPROTO_H_
471struct __semctl_args {
472	int	semid;
473	int	semnum;
474	int	cmd;
475	union	semun *arg;
476};
477#endif
478
479/*
480 * MPSAFE
481 */
482int
483__semctl(td, uap)
484	struct thread *td;
485	register struct __semctl_args *uap;
486{
487	int semid = uap->semid;
488	int semnum = uap->semnum;
489	int cmd = uap->cmd;
490	union semun *arg = uap->arg;
491	union semun real_arg;
492	struct ucred *cred = td->td_proc->p_ucred;
493	int i, rval, error;
494	struct semid_ds sbuf;
495	register struct semid_ds *semaptr;
496	u_short usval;
497
498#ifdef SEM_DEBUG
499	printf("call to semctl(%d, %d, %d, 0x%x)\n", semid, semnum, cmd, arg);
500#endif
501	mtx_lock(&Giant);
502	if (!jail_sysvipc_allowed && jailed(td->td_proc->p_ucred)) {
503		error = ENOSYS;
504		goto done2;
505	}
506
507	switch(cmd) {
508	case SEM_STAT:
509		if (semid < 0 || semid >= seminfo.semmsl)
510			return(EINVAL);
511		semaptr = &sema[semid];
512		if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0 )
513			return(EINVAL);
514		if ((error = ipcperm(td, &semaptr->sem_perm, IPC_R)))
515			return(error);
516		if ((error = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
517			return(error);
518		error = copyout((caddr_t)semaptr, real_arg.buf,
519			sizeof(struct semid_ds));
520		rval = IXSEQ_TO_IPCID(semid,semaptr->sem_perm);
521		if (error == 0)
522			td->td_retval[0] = rval;
523		goto done2;
524	}
525
526	semid = IPCID_TO_IX(semid);
527	if (semid < 0 || semid >= seminfo.semmsl) {
528		error = EINVAL;
529		goto done2;
530	}
531
532	semaptr = &sema[semid];
533	if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0 ||
534	    semaptr->sem_perm.seq != IPCID_TO_SEQ(uap->semid)) {
535		error = EINVAL;
536		goto done2;
537	}
538
539	error = 0;
540	rval = 0;
541
542	switch (cmd) {
543	case IPC_RMID:
544		if ((error = ipcperm(td, &semaptr->sem_perm, IPC_M)))
545			goto done2;
546		semaptr->sem_perm.cuid = cred->cr_uid;
547		semaptr->sem_perm.uid = cred->cr_uid;
548		semtot -= semaptr->sem_nsems;
549		for (i = semaptr->sem_base - sem; i < semtot; i++)
550			sem[i] = sem[i + semaptr->sem_nsems];
551		for (i = 0; i < seminfo.semmni; i++) {
552			if ((sema[i].sem_perm.mode & SEM_ALLOC) &&
553			    sema[i].sem_base > semaptr->sem_base)
554				sema[i].sem_base -= semaptr->sem_nsems;
555		}
556		semaptr->sem_perm.mode = 0;
557		semundo_clear(semid, -1);
558		wakeup((caddr_t)semaptr);
559		break;
560
561	case IPC_SET:
562		if ((error = ipcperm(td, &semaptr->sem_perm, IPC_M)))
563			goto done2;
564		if ((error = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
565			goto done2;
566		if ((error = copyin(real_arg.buf, (caddr_t)&sbuf,
567		    sizeof(sbuf))) != 0) {
568			goto done2;
569		}
570		semaptr->sem_perm.uid = sbuf.sem_perm.uid;
571		semaptr->sem_perm.gid = sbuf.sem_perm.gid;
572		semaptr->sem_perm.mode = (semaptr->sem_perm.mode & ~0777) |
573		    (sbuf.sem_perm.mode & 0777);
574		semaptr->sem_ctime = time_second;
575		break;
576
577	case IPC_STAT:
578		if ((error = ipcperm(td, &semaptr->sem_perm, IPC_R)))
579			goto done2;
580		if ((error = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
581			goto done2;
582		error = copyout((caddr_t)semaptr, real_arg.buf,
583				sizeof(struct semid_ds));
584		break;
585
586	case GETNCNT:
587		if ((error = ipcperm(td, &semaptr->sem_perm, IPC_R)))
588			goto done2;
589		if (semnum < 0 || semnum >= semaptr->sem_nsems) {
590			error = EINVAL;
591			goto done2;
592		}
593		rval = semaptr->sem_base[semnum].semncnt;
594		break;
595
596	case GETPID:
597		if ((error = ipcperm(td, &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].sempid;
604		break;
605
606	case GETVAL:
607		if ((error = ipcperm(td, &semaptr->sem_perm, IPC_R)))
608			goto done2;
609		if (semnum < 0 || semnum >= semaptr->sem_nsems) {
610			error = EINVAL;
611			goto done2;
612		}
613		rval = semaptr->sem_base[semnum].semval;
614		break;
615
616	case GETALL:
617		if ((error = ipcperm(td, &semaptr->sem_perm, IPC_R)))
618			goto done2;
619		if ((error = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
620			goto done2;
621		for (i = 0; i < semaptr->sem_nsems; i++) {
622			error = copyout((caddr_t)&semaptr->sem_base[i].semval,
623			    &real_arg.array[i], sizeof(real_arg.array[0]));
624			if (error != 0)
625				break;
626		}
627		break;
628
629	case GETZCNT:
630		if ((error = ipcperm(td, &semaptr->sem_perm, IPC_R)))
631			goto done2;
632		if (semnum < 0 || semnum >= semaptr->sem_nsems) {
633			error = EINVAL;
634			goto done2;
635		}
636		rval = semaptr->sem_base[semnum].semzcnt;
637		break;
638
639	case SETVAL:
640		if ((error = ipcperm(td, &semaptr->sem_perm, IPC_W)))
641			goto done2;
642		if (semnum < 0 || semnum >= semaptr->sem_nsems) {
643			error = EINVAL;
644			goto done2;
645		}
646		if ((error = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
647			goto done2;
648		if (real_arg.val < 0 || real_arg.val > seminfo.semvmx) {
649			error = ERANGE;
650			goto done2;
651		}
652		semaptr->sem_base[semnum].semval = real_arg.val;
653		semundo_clear(semid, semnum);
654		wakeup((caddr_t)semaptr);
655		break;
656
657	case SETALL:
658		if ((error = ipcperm(td, &semaptr->sem_perm, IPC_W)))
659			goto done2;
660		if ((error = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
661			goto done2;
662		for (i = 0; i < semaptr->sem_nsems; i++) {
663			error = copyin(&real_arg.array[i],
664			    (caddr_t)&usval, sizeof(real_arg.array[0]));
665			if (error != 0)
666				break;
667			if (usval > seminfo.semvmx) {
668				error = ERANGE;
669				break;
670			}
671			semaptr->sem_base[i].semval = usval;
672		}
673		semundo_clear(semid, -1);
674		wakeup((caddr_t)semaptr);
675		break;
676
677	default:
678		error = EINVAL;
679		break;
680	}
681
682	if (error == 0)
683		td->td_retval[0] = rval;
684done2:
685	mtx_unlock(&Giant);
686	return(error);
687}
688
689#ifndef _SYS_SYSPROTO_H_
690struct semget_args {
691	key_t	key;
692	int	nsems;
693	int	semflg;
694};
695#endif
696
697/*
698 * MPSAFE
699 */
700int
701semget(td, uap)
702	struct thread *td;
703	register struct semget_args *uap;
704{
705	int semid, error = 0;
706	int key = uap->key;
707	int nsems = uap->nsems;
708	int semflg = uap->semflg;
709	struct ucred *cred = td->td_proc->p_ucred;
710
711#ifdef SEM_DEBUG
712	printf("semget(0x%x, %d, 0%o)\n", key, nsems, semflg);
713#endif
714	mtx_lock(&Giant);
715	if (!jail_sysvipc_allowed && jailed(td->td_proc->p_ucred)) {
716		error = ENOSYS;
717		goto done2;
718	}
719
720	if (key != IPC_PRIVATE) {
721		for (semid = 0; semid < seminfo.semmni; semid++) {
722			if ((sema[semid].sem_perm.mode & SEM_ALLOC) &&
723			    sema[semid].sem_perm.key == key)
724				break;
725		}
726		if (semid < seminfo.semmni) {
727#ifdef SEM_DEBUG
728			printf("found public key\n");
729#endif
730			if ((error = ipcperm(td, &sema[semid].sem_perm,
731			    semflg & 0700))) {
732				goto done2;
733			}
734			if (nsems > 0 && sema[semid].sem_nsems < nsems) {
735#ifdef SEM_DEBUG
736				printf("too small\n");
737#endif
738				error = EINVAL;
739				goto done2;
740			}
741			if ((semflg & IPC_CREAT) && (semflg & IPC_EXCL)) {
742#ifdef SEM_DEBUG
743				printf("not exclusive\n");
744#endif
745				error = EEXIST;
746				goto done2;
747			}
748			goto found;
749		}
750	}
751
752#ifdef SEM_DEBUG
753	printf("need to allocate the semid_ds\n");
754#endif
755	if (key == IPC_PRIVATE || (semflg & IPC_CREAT)) {
756		if (nsems <= 0 || nsems > seminfo.semmsl) {
757#ifdef SEM_DEBUG
758			printf("nsems out of range (0<%d<=%d)\n", nsems,
759			    seminfo.semmsl);
760#endif
761			error = EINVAL;
762			goto done2;
763		}
764		if (nsems > seminfo.semmns - semtot) {
765#ifdef SEM_DEBUG
766			printf("not enough semaphores left (need %d, got %d)\n",
767			    nsems, seminfo.semmns - semtot);
768#endif
769			error = ENOSPC;
770			goto done2;
771		}
772		for (semid = 0; semid < seminfo.semmni; semid++) {
773			if ((sema[semid].sem_perm.mode & SEM_ALLOC) == 0)
774				break;
775		}
776		if (semid == seminfo.semmni) {
777#ifdef SEM_DEBUG
778			printf("no more semid_ds's available\n");
779#endif
780			error = ENOSPC;
781			goto done2;
782		}
783#ifdef SEM_DEBUG
784		printf("semid %d is available\n", semid);
785#endif
786		sema[semid].sem_perm.key = key;
787		sema[semid].sem_perm.cuid = cred->cr_uid;
788		sema[semid].sem_perm.uid = cred->cr_uid;
789		sema[semid].sem_perm.cgid = cred->cr_gid;
790		sema[semid].sem_perm.gid = cred->cr_gid;
791		sema[semid].sem_perm.mode = (semflg & 0777) | SEM_ALLOC;
792		sema[semid].sem_perm.seq =
793		    (sema[semid].sem_perm.seq + 1) & 0x7fff;
794		sema[semid].sem_nsems = nsems;
795		sema[semid].sem_otime = 0;
796		sema[semid].sem_ctime = time_second;
797		sema[semid].sem_base = &sem[semtot];
798		semtot += nsems;
799		bzero(sema[semid].sem_base,
800		    sizeof(sema[semid].sem_base[0])*nsems);
801#ifdef SEM_DEBUG
802		printf("sembase = 0x%x, next = 0x%x\n", sema[semid].sem_base,
803		    &sem[semtot]);
804#endif
805	} else {
806#ifdef SEM_DEBUG
807		printf("didn't find it and wasn't asked to create it\n");
808#endif
809		error = ENOENT;
810		goto done2;
811	}
812
813found:
814	td->td_retval[0] = IXSEQ_TO_IPCID(semid, sema[semid].sem_perm);
815done2:
816	mtx_unlock(&Giant);
817	return (error);
818}
819
820#ifndef _SYS_SYSPROTO_H_
821struct semop_args {
822	int	semid;
823	struct	sembuf *sops;
824	u_int	nsops;
825};
826#endif
827
828/*
829 * MPSAFE
830 */
831int
832semop(td, uap)
833	struct thread *td;
834	register struct semop_args *uap;
835{
836	int semid = uap->semid;
837	u_int nsops = uap->nsops;
838	struct sembuf *sops = NULL;
839	register struct semid_ds *semaptr;
840	register struct sembuf *sopptr;
841	register struct sem *semptr;
842	struct sem_undo *suptr;
843	int i, j, error;
844	int do_wakeup, do_undos;
845
846#ifdef SEM_DEBUG
847	printf("call to semop(%d, 0x%x, %u)\n", semid, sops, nsops);
848#endif
849
850	mtx_lock(&Giant);
851	if (!jail_sysvipc_allowed && jailed(td->td_proc->p_ucred)) {
852		error = ENOSYS;
853		goto done2;
854	}
855
856	semid = IPCID_TO_IX(semid);	/* Convert back to zero origin */
857
858	if (semid < 0 || semid >= seminfo.semmsl) {
859		error = EINVAL;
860		goto done2;
861	}
862
863	semaptr = &sema[semid];
864	if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0) {
865		error = EINVAL;
866		goto done2;
867	}
868	if (semaptr->sem_perm.seq != IPCID_TO_SEQ(uap->semid)) {
869		error = EINVAL;
870		goto done2;
871	}
872	if (nsops > seminfo.semopm) {
873#ifdef SEM_DEBUG
874		printf("too many sops (max=%d, nsops=%d)\n", seminfo.semopm,
875		    nsops);
876#endif
877		error = E2BIG;
878		goto done2;
879	}
880
881	/* Allocate memory for sem_ops */
882	sops = malloc(nsops * sizeof(sops[0]), M_SEM, M_WAITOK);
883	if (!sops)
884		panic("Failed to allocate %d sem_ops", nsops);
885
886	if ((error = copyin(uap->sops, sops, nsops * sizeof(sops[0]))) != 0) {
887#ifdef SEM_DEBUG
888		printf("error = %d from copyin(%08x, %08x, %d)\n", error,
889		    uap->sops, sops, nsops * sizeof(sops[0]));
890#endif
891		goto done2;
892	}
893
894	/*
895	 * Initial pass thru sops to see what permissions are needed.
896	 * Also perform any checks that don't need repeating on each
897	 * attempt to satisfy the request vector.
898	 */
899	j = 0;		/* permission needed */
900	do_undos = 0;
901	for (i = 0; i < nsops; i++) {
902		sopptr = &sops[i];
903		if (sopptr->sem_num >= semaptr->sem_nsems) {
904			error = EFBIG;
905			goto done2;
906		}
907		if (sopptr->sem_flg & SEM_UNDO && sopptr->sem_op != 0)
908			do_undos = 1;
909		j |= (sopptr->sem_op == 0) ? SEM_R : SEM_A;
910	}
911
912	if ((error = ipcperm(td, &semaptr->sem_perm, j))) {
913#ifdef SEM_DEBUG
914		printf("error = %d from ipaccess\n", error);
915#endif
916		goto done2;
917	}
918
919	/*
920	 * Loop trying to satisfy the vector of requests.
921	 * If we reach a point where we must wait, any requests already
922	 * performed are rolled back and we go to sleep until some other
923	 * process wakes us up.  At this point, we start all over again.
924	 *
925	 * This ensures that from the perspective of other tasks, a set
926	 * of requests is atomic (never partially satisfied).
927	 */
928	for (;;) {
929		do_wakeup = 0;
930		error = 0;	/* error return if necessary */
931
932		for (i = 0; i < nsops; i++) {
933			sopptr = &sops[i];
934			semptr = &semaptr->sem_base[sopptr->sem_num];
935
936#ifdef SEM_DEBUG
937			printf("semop:  semaptr=%x, sem_base=%x, semptr=%x, sem[%d]=%d : op=%d, flag=%s\n",
938			    semaptr, semaptr->sem_base, semptr,
939			    sopptr->sem_num, semptr->semval, sopptr->sem_op,
940			    (sopptr->sem_flg & IPC_NOWAIT) ? "nowait" : "wait");
941#endif
942
943			if (sopptr->sem_op < 0) {
944				if (semptr->semval + sopptr->sem_op < 0) {
945#ifdef SEM_DEBUG
946					printf("semop:  can't do it now\n");
947#endif
948					break;
949				} else {
950					semptr->semval += sopptr->sem_op;
951					if (semptr->semval == 0 &&
952					    semptr->semzcnt > 0)
953						do_wakeup = 1;
954				}
955			} else if (sopptr->sem_op == 0) {
956				if (semptr->semval != 0) {
957#ifdef SEM_DEBUG
958					printf("semop:  not zero now\n");
959#endif
960					break;
961				}
962			} else if (semptr->semval + sopptr->sem_op >
963			    seminfo.semvmx) {
964				error = ERANGE;
965				break;
966			} else {
967				if (semptr->semncnt > 0)
968					do_wakeup = 1;
969				semptr->semval += sopptr->sem_op;
970			}
971		}
972
973		/*
974		 * Did we get through the entire vector?
975		 */
976		if (i >= nsops)
977			goto done;
978
979		/*
980		 * No ... rollback anything that we've already done
981		 */
982#ifdef SEM_DEBUG
983		printf("semop:  rollback 0 through %d\n", i-1);
984#endif
985		for (j = 0; j < i; j++)
986			semaptr->sem_base[sops[j].sem_num].semval -=
987			    sops[j].sem_op;
988
989		/* If we detected an error, return it */
990		if (error != 0)
991			goto done2;
992
993		/*
994		 * If the request that we couldn't satisfy has the
995		 * NOWAIT flag set then return with EAGAIN.
996		 */
997		if (sopptr->sem_flg & IPC_NOWAIT) {
998			error = EAGAIN;
999			goto done2;
1000		}
1001
1002		if (sopptr->sem_op == 0)
1003			semptr->semzcnt++;
1004		else
1005			semptr->semncnt++;
1006
1007#ifdef SEM_DEBUG
1008		printf("semop:  good night!\n");
1009#endif
1010		error = tsleep((caddr_t)semaptr, (PZERO - 4) | PCATCH,
1011		    "semwait", 0);
1012#ifdef SEM_DEBUG
1013		printf("semop:  good morning (error=%d)!\n", error);
1014#endif
1015
1016		if (error != 0) {
1017			error = EINTR;
1018			goto done2;
1019		}
1020#ifdef SEM_DEBUG
1021		printf("semop:  good morning!\n");
1022#endif
1023
1024		/*
1025		 * Make sure that the semaphore still exists
1026		 */
1027		if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0 ||
1028		    semaptr->sem_perm.seq != IPCID_TO_SEQ(uap->semid)) {
1029			error = EIDRM;
1030			goto done2;
1031		}
1032
1033		/*
1034		 * The semaphore is still alive.  Readjust the count of
1035		 * waiting processes.
1036		 */
1037		if (sopptr->sem_op == 0)
1038			semptr->semzcnt--;
1039		else
1040			semptr->semncnt--;
1041	}
1042
1043done:
1044	/*
1045	 * Process any SEM_UNDO requests.
1046	 */
1047	if (do_undos) {
1048		suptr = NULL;
1049		for (i = 0; i < nsops; i++) {
1050			/*
1051			 * We only need to deal with SEM_UNDO's for non-zero
1052			 * op's.
1053			 */
1054			int adjval;
1055
1056			if ((sops[i].sem_flg & SEM_UNDO) == 0)
1057				continue;
1058			adjval = sops[i].sem_op;
1059			if (adjval == 0)
1060				continue;
1061			error = semundo_adjust(td, &suptr, semid,
1062			    sops[i].sem_num, -adjval);
1063			if (error == 0)
1064				continue;
1065
1066			/*
1067			 * Oh-Oh!  We ran out of either sem_undo's or undo's.
1068			 * Rollback the adjustments to this point and then
1069			 * rollback the semaphore ups and down so we can return
1070			 * with an error with all structures restored.  We
1071			 * rollback the undo's in the exact reverse order that
1072			 * we applied them.  This guarantees that we won't run
1073			 * out of space as we roll things back out.
1074			 */
1075			for (j = i - 1; j >= 0; j--) {
1076				if ((sops[j].sem_flg & SEM_UNDO) == 0)
1077					continue;
1078				adjval = sops[j].sem_op;
1079				if (adjval == 0)
1080					continue;
1081				if (semundo_adjust(td, &suptr, semid,
1082				    sops[j].sem_num, adjval) != 0)
1083					panic("semop - can't undo undos");
1084			}
1085
1086			for (j = 0; j < nsops; j++)
1087				semaptr->sem_base[sops[j].sem_num].semval -=
1088				    sops[j].sem_op;
1089
1090#ifdef SEM_DEBUG
1091			printf("error = %d from semundo_adjust\n", error);
1092#endif
1093			goto done2;
1094		} /* loop through the sops */
1095	} /* if (do_undos) */
1096
1097	/* We're definitely done - set the sempid's and time */
1098	for (i = 0; i < nsops; i++) {
1099		sopptr = &sops[i];
1100		semptr = &semaptr->sem_base[sopptr->sem_num];
1101		semptr->sempid = td->td_proc->p_pid;
1102	}
1103	semaptr->sem_otime = time_second;
1104
1105	/*
1106	 * Do a wakeup if any semaphore was up'd whilst something was
1107	 * sleeping on it.
1108	 */
1109	if (do_wakeup) {
1110#ifdef SEM_DEBUG
1111		printf("semop:  doing wakeup\n");
1112#endif
1113		wakeup((caddr_t)semaptr);
1114#ifdef SEM_DEBUG
1115		printf("semop:  back from wakeup\n");
1116#endif
1117	}
1118#ifdef SEM_DEBUG
1119	printf("semop:  done\n");
1120#endif
1121	td->td_retval[0] = 0;
1122done2:
1123	if (sops)
1124	    free(sops, M_SEM);
1125	mtx_unlock(&Giant);
1126	return (error);
1127}
1128
1129/*
1130 * Go through the undo structures for this process and apply the adjustments to
1131 * semaphores.
1132 */
1133static void
1134semexit_myhook(p)
1135	struct proc *p;
1136{
1137	register struct sem_undo *suptr;
1138	register struct sem_undo **supptr;
1139
1140	/*
1141	 * Go through the chain of undo vectors looking for one
1142	 * associated with this process.
1143	 */
1144
1145	for (supptr = &semu_list; (suptr = *supptr) != NULL;
1146	    supptr = &suptr->un_next) {
1147		if (suptr->un_proc == p)
1148			break;
1149	}
1150
1151	if (suptr == NULL)
1152		return;
1153
1154#ifdef SEM_DEBUG
1155	printf("proc @%08x has undo structure with %d entries\n", p,
1156	    suptr->un_cnt);
1157#endif
1158
1159	/*
1160	 * If there are any active undo elements then process them.
1161	 */
1162	if (suptr->un_cnt > 0) {
1163		int ix;
1164
1165		for (ix = 0; ix < suptr->un_cnt; ix++) {
1166			int semid = suptr->un_ent[ix].un_id;
1167			int semnum = suptr->un_ent[ix].un_num;
1168			int adjval = suptr->un_ent[ix].un_adjval;
1169			struct semid_ds *semaptr;
1170
1171			semaptr = &sema[semid];
1172			if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0)
1173				panic("semexit - semid not allocated");
1174			if (semnum >= semaptr->sem_nsems)
1175				panic("semexit - semnum out of range");
1176
1177#ifdef SEM_DEBUG
1178			printf("semexit:  %08x id=%d num=%d(adj=%d) ; sem=%d\n",
1179			    suptr->un_proc, suptr->un_ent[ix].un_id,
1180			    suptr->un_ent[ix].un_num,
1181			    suptr->un_ent[ix].un_adjval,
1182			    semaptr->sem_base[semnum].semval);
1183#endif
1184
1185			if (adjval < 0) {
1186				if (semaptr->sem_base[semnum].semval < -adjval)
1187					semaptr->sem_base[semnum].semval = 0;
1188				else
1189					semaptr->sem_base[semnum].semval +=
1190					    adjval;
1191			} else
1192				semaptr->sem_base[semnum].semval += adjval;
1193
1194			wakeup((caddr_t)semaptr);
1195#ifdef SEM_DEBUG
1196			printf("semexit:  back from wakeup\n");
1197#endif
1198		}
1199	}
1200
1201	/*
1202	 * Deallocate the undo vector.
1203	 */
1204#ifdef SEM_DEBUG
1205	printf("removing vector\n");
1206#endif
1207	suptr->un_proc = NULL;
1208	*supptr = suptr->un_next;
1209}
1210
1211static int
1212sysctl_sema(SYSCTL_HANDLER_ARGS)
1213{
1214
1215	return (SYSCTL_OUT(req, sema,
1216	    sizeof(struct semid_ds) * seminfo.semmni));
1217}
1218