xref: /freebsd/sys/kern/sysv_sem.c (revision c11e094d96120a2e0e726ed9705ae0ec08db49b6)
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/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 
27 static MALLOC_DEFINE(M_SEM, "sem", "SVID compatible semaphores");
28 
29 static void seminit(void);
30 static int sysvsem_modload(struct module *, int, void *);
31 static int semunload(void);
32 static void semexit_myhook(struct proc *p);
33 static int sysctl_sema(SYSCTL_HANDLER_ARGS);
34 
35 #ifndef _SYS_SYSPROTO_H_
36 struct __semctl_args;
37 int __semctl(struct thread *td, struct __semctl_args *uap);
38 struct semget_args;
39 int semget(struct thread *td, struct semget_args *uap);
40 struct semop_args;
41 int semop(struct thread *td, struct semop_args *uap);
42 #endif
43 
44 static struct sem_undo *semu_alloc(struct thread *td);
45 static int semundo_adjust(struct thread *td, struct sem_undo **supptr,
46 		int semid, int semnum, int adjval);
47 static void semundo_clear(int semid, int semnum);
48 
49 /* XXX casting to (sy_call_t *) is bogus, as usual. */
50 static sy_call_t *semcalls[] = {
51 	(sy_call_t *)__semctl, (sy_call_t *)semget,
52 	(sy_call_t *)semop
53 };
54 
55 static int	semtot = 0;
56 static struct semid_ds *sema;	/* semaphore id pool */
57 static struct sem *sem;		/* semaphore pool */
58 static struct sem_undo *semu_list; /* list of active undo structures */
59 static int	*semu;		/* undo structure pool */
60 
61 struct 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  */
71 struct 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  */
130 struct 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 
143 SYSCTL_DECL(_kern_ipc);
144 SYSCTL_INT(_kern_ipc, OID_AUTO, semmap, CTLFLAG_RW, &seminfo.semmap, 0, "");
145 SYSCTL_INT(_kern_ipc, OID_AUTO, semmni, CTLFLAG_RD, &seminfo.semmni, 0, "");
146 SYSCTL_INT(_kern_ipc, OID_AUTO, semmns, CTLFLAG_RD, &seminfo.semmns, 0, "");
147 SYSCTL_INT(_kern_ipc, OID_AUTO, semmnu, CTLFLAG_RD, &seminfo.semmnu, 0, "");
148 SYSCTL_INT(_kern_ipc, OID_AUTO, semmsl, CTLFLAG_RW, &seminfo.semmsl, 0, "");
149 SYSCTL_INT(_kern_ipc, OID_AUTO, semopm, CTLFLAG_RD, &seminfo.semopm, 0, "");
150 SYSCTL_INT(_kern_ipc, OID_AUTO, semume, CTLFLAG_RD, &seminfo.semume, 0, "");
151 SYSCTL_INT(_kern_ipc, OID_AUTO, semusz, CTLFLAG_RD, &seminfo.semusz, 0, "");
152 SYSCTL_INT(_kern_ipc, OID_AUTO, semvmx, CTLFLAG_RW, &seminfo.semvmx, 0, "");
153 SYSCTL_INT(_kern_ipc, OID_AUTO, semaem, CTLFLAG_RW, &seminfo.semaem, 0, "");
154 SYSCTL_PROC(_kern_ipc, OID_AUTO, sema, CTLFLAG_RD,
155     NULL, 0, sysctl_sema, "", "");
156 
157 static void
158 seminit(void)
159 {
160 	register int i;
161 
162 	TUNABLE_INT_FETCH("kern.ipc.semmap", &seminfo.semmap);
163 	TUNABLE_INT_FETCH("kern.ipc.semmni", &seminfo.semmni);
164 	TUNABLE_INT_FETCH("kern.ipc.semmns", &seminfo.semmns);
165 	TUNABLE_INT_FETCH("kern.ipc.semmnu", &seminfo.semmnu);
166 	TUNABLE_INT_FETCH("kern.ipc.semmsl", &seminfo.semmsl);
167 	TUNABLE_INT_FETCH("kern.ipc.semopm", &seminfo.semopm);
168 	TUNABLE_INT_FETCH("kern.ipc.semume", &seminfo.semume);
169 	TUNABLE_INT_FETCH("kern.ipc.semusz", &seminfo.semusz);
170 	TUNABLE_INT_FETCH("kern.ipc.semvmx", &seminfo.semvmx);
171 	TUNABLE_INT_FETCH("kern.ipc.semaem", &seminfo.semaem);
172 
173 	sem = malloc(sizeof(struct sem) * seminfo.semmns, M_SEM, M_WAITOK);
174 	if (sem == NULL)
175 		panic("sem is NULL");
176 	sema = malloc(sizeof(struct semid_ds) * seminfo.semmni, M_SEM, M_WAITOK);
177 	if (sema == NULL)
178 		panic("sema is NULL");
179 	semu = malloc(seminfo.semmnu * seminfo.semusz, M_SEM, M_WAITOK);
180 	if (semu == NULL)
181 		panic("semu is NULL");
182 
183 	for (i = 0; i < seminfo.semmni; i++) {
184 		sema[i].sem_base = 0;
185 		sema[i].sem_perm.mode = 0;
186 	}
187 	for (i = 0; i < seminfo.semmnu; i++) {
188 		register struct sem_undo *suptr = SEMU(i);
189 		suptr->un_proc = NULL;
190 	}
191 	semu_list = NULL;
192 	at_exit(semexit_myhook);
193 }
194 
195 static int
196 semunload(void)
197 {
198 
199 	if (semtot != 0)
200 		return (EBUSY);
201 
202 	free(sem, M_SEM);
203 	free(sema, M_SEM);
204 	free(semu, M_SEM);
205 	rm_at_exit(semexit_myhook);
206 	return (0);
207 }
208 
209 static int
210 sysvsem_modload(struct module *module, int cmd, void *arg)
211 {
212 	int error = 0;
213 
214 	switch (cmd) {
215 	case MOD_LOAD:
216 		seminit();
217 		break;
218 	case MOD_UNLOAD:
219 		error = semunload();
220 		break;
221 	case MOD_SHUTDOWN:
222 		break;
223 	default:
224 		error = EINVAL;
225 		break;
226 	}
227 	return (error);
228 }
229 
230 static moduledata_t sysvsem_mod = {
231 	"sysvsem",
232 	&sysvsem_modload,
233 	NULL
234 };
235 
236 SYSCALL_MODULE_HELPER(semsys);
237 SYSCALL_MODULE_HELPER(__semctl);
238 SYSCALL_MODULE_HELPER(semget);
239 SYSCALL_MODULE_HELPER(semop);
240 
241 DECLARE_MODULE(sysvsem, sysvsem_mod,
242 	SI_SUB_SYSV_SEM, SI_ORDER_FIRST);
243 MODULE_VERSION(sysvsem, 1);
244 
245 /*
246  * Entry point for all SEM calls
247  *
248  * MPSAFE
249  */
250 int
251 semsys(td, uap)
252 	struct thread *td;
253 	/* XXX actually varargs. */
254 	struct semsys_args /* {
255 		u_int	which;
256 		int	a2;
257 		int	a3;
258 		int	a4;
259 		int	a5;
260 	} */ *uap;
261 {
262 	int error;
263 
264 	if (!jail_sysvipc_allowed && jailed(td->td_ucred))
265 		return (ENOSYS);
266 	if (uap->which >= sizeof(semcalls)/sizeof(semcalls[0]))
267 		return (EINVAL);
268 	mtx_lock(&Giant);
269 	error = (*semcalls[uap->which])(td, &uap->a2);
270 	mtx_unlock(&Giant);
271 	return (error);
272 }
273 
274 /*
275  * Allocate a new sem_undo structure for a process
276  * (returns ptr to structure or NULL if no more room)
277  */
278 
279 static struct sem_undo *
280 semu_alloc(td)
281 	struct thread *td;
282 {
283 	register int i;
284 	register struct sem_undo *suptr;
285 	register struct sem_undo **supptr;
286 	int attempt;
287 
288 	/*
289 	 * Try twice to allocate something.
290 	 * (we'll purge any empty structures after the first pass so
291 	 * two passes are always enough)
292 	 */
293 
294 	for (attempt = 0; attempt < 2; attempt++) {
295 		/*
296 		 * Look for a free structure.
297 		 * Fill it in and return it if we find one.
298 		 */
299 
300 		for (i = 0; i < seminfo.semmnu; i++) {
301 			suptr = SEMU(i);
302 			if (suptr->un_proc == NULL) {
303 				suptr->un_next = semu_list;
304 				semu_list = suptr;
305 				suptr->un_cnt = 0;
306 				suptr->un_proc = td->td_proc;
307 				return(suptr);
308 			}
309 		}
310 
311 		/*
312 		 * We didn't find a free one, if this is the first attempt
313 		 * then try to free some structures.
314 		 */
315 
316 		if (attempt == 0) {
317 			/* All the structures are in use - try to free some */
318 			int did_something = 0;
319 
320 			supptr = &semu_list;
321 			while ((suptr = *supptr) != NULL) {
322 				if (suptr->un_cnt == 0)  {
323 					suptr->un_proc = NULL;
324 					*supptr = suptr->un_next;
325 					did_something = 1;
326 				} else
327 					supptr = &(suptr->un_next);
328 			}
329 
330 			/* If we didn't free anything then just give-up */
331 			if (!did_something)
332 				return(NULL);
333 		} else {
334 			/*
335 			 * The second pass failed even though we freed
336 			 * something after the first pass!
337 			 * This is IMPOSSIBLE!
338 			 */
339 			panic("semu_alloc - second attempt failed");
340 		}
341 	}
342 	return (NULL);
343 }
344 
345 /*
346  * Adjust a particular entry for a particular proc
347  */
348 
349 static int
350 semundo_adjust(td, supptr, semid, semnum, adjval)
351 	register struct thread *td;
352 	struct sem_undo **supptr;
353 	int semid, semnum;
354 	int adjval;
355 {
356 	struct proc *p = td->td_proc;
357 	register struct sem_undo *suptr;
358 	register struct undo *sunptr;
359 	int i;
360 
361 	/* Look for and remember the sem_undo if the caller doesn't provide
362 	   it */
363 
364 	suptr = *supptr;
365 	if (suptr == NULL) {
366 		for (suptr = semu_list; suptr != NULL;
367 		    suptr = suptr->un_next) {
368 			if (suptr->un_proc == p) {
369 				*supptr = suptr;
370 				break;
371 			}
372 		}
373 		if (suptr == NULL) {
374 			if (adjval == 0)
375 				return(0);
376 			suptr = semu_alloc(td);
377 			if (suptr == NULL)
378 				return(ENOSPC);
379 			*supptr = suptr;
380 		}
381 	}
382 
383 	/*
384 	 * Look for the requested entry and adjust it (delete if adjval becomes
385 	 * 0).
386 	 */
387 	sunptr = &suptr->un_ent[0];
388 	for (i = 0; i < suptr->un_cnt; i++, sunptr++) {
389 		if (sunptr->un_id != semid || sunptr->un_num != semnum)
390 			continue;
391 		if (adjval != 0) {
392 			adjval += sunptr->un_adjval;
393 			if (adjval > seminfo.semaem || adjval < -seminfo.semaem)
394 				return (ERANGE);
395 		}
396 		sunptr->un_adjval = adjval;
397 		if (sunptr->un_adjval == 0) {
398 			suptr->un_cnt--;
399 			if (i < suptr->un_cnt)
400 				suptr->un_ent[i] =
401 				    suptr->un_ent[suptr->un_cnt];
402 		}
403 		return(0);
404 	}
405 
406 	/* Didn't find the right entry - create it */
407 	if (adjval == 0)
408 		return(0);
409 	if (adjval > seminfo.semaem || adjval < -seminfo.semaem)
410 		return (ERANGE);
411 	if (suptr->un_cnt != seminfo.semume) {
412 		sunptr = &suptr->un_ent[suptr->un_cnt];
413 		suptr->un_cnt++;
414 		sunptr->un_adjval = adjval;
415 		sunptr->un_id = semid; sunptr->un_num = semnum;
416 	} else
417 		return(EINVAL);
418 	return(0);
419 }
420 
421 static void
422 semundo_clear(semid, semnum)
423 	int semid, semnum;
424 {
425 	register struct sem_undo *suptr;
426 
427 	for (suptr = semu_list; suptr != NULL; suptr = suptr->un_next) {
428 		register struct undo *sunptr = &suptr->un_ent[0];
429 		register int i = 0;
430 
431 		while (i < suptr->un_cnt) {
432 			if (sunptr->un_id == semid) {
433 				if (semnum == -1 || sunptr->un_num == semnum) {
434 					suptr->un_cnt--;
435 					if (i < suptr->un_cnt) {
436 						suptr->un_ent[i] =
437 						  suptr->un_ent[suptr->un_cnt];
438 						continue;
439 					}
440 				}
441 				if (semnum != -1)
442 					break;
443 			}
444 			i++, sunptr++;
445 		}
446 	}
447 }
448 
449 /*
450  * Note that the user-mode half of this passes a union, not a pointer
451  */
452 #ifndef _SYS_SYSPROTO_H_
453 struct __semctl_args {
454 	int	semid;
455 	int	semnum;
456 	int	cmd;
457 	union	semun *arg;
458 };
459 #endif
460 
461 /*
462  * MPSAFE
463  */
464 int
465 __semctl(td, uap)
466 	struct thread *td;
467 	register struct __semctl_args *uap;
468 {
469 	int semid = uap->semid;
470 	int semnum = uap->semnum;
471 	int cmd = uap->cmd;
472 	union semun *arg = uap->arg;
473 	union semun real_arg;
474 	struct ucred *cred = td->td_ucred;
475 	int i, rval, error;
476 	struct semid_ds sbuf;
477 	register struct semid_ds *semaptr;
478 	u_short usval;
479 
480 #ifdef SEM_DEBUG
481 	printf("call to semctl(%d, %d, %d, 0x%x)\n", semid, semnum, cmd, arg);
482 #endif
483 	if (!jail_sysvipc_allowed && jailed(td->td_ucred))
484 		return (ENOSYS);
485 
486 	mtx_lock(&Giant);
487 	switch(cmd) {
488 	case SEM_STAT:
489 		if (semid < 0 || semid >= seminfo.semmni)
490 			UGAR(EINVAL);
491 		semaptr = &sema[semid];
492 		if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0 )
493 			UGAR(EINVAL);
494 		if ((error = ipcperm(td, &semaptr->sem_perm, IPC_R)))
495 			UGAR(error);
496 		if ((error = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
497 			UGAR(error);
498 		error = copyout((caddr_t)semaptr, real_arg.buf,
499 			sizeof(struct semid_ds));
500 		rval = IXSEQ_TO_IPCID(semid,semaptr->sem_perm);
501 		if (error == 0)
502 			td->td_retval[0] = rval;
503 		goto done2;
504 	}
505 
506 	semid = IPCID_TO_IX(semid);
507 	if (semid < 0 || semid >= seminfo.semmni) {
508 		error = EINVAL;
509 		goto done2;
510 	}
511 
512 	semaptr = &sema[semid];
513 	if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0 ||
514 	    semaptr->sem_perm.seq != IPCID_TO_SEQ(uap->semid)) {
515 		error = EINVAL;
516 		goto done2;
517 	}
518 
519 	error = 0;
520 	rval = 0;
521 
522 	switch (cmd) {
523 	case IPC_RMID:
524 		if ((error = ipcperm(td, &semaptr->sem_perm, IPC_M)))
525 			goto done2;
526 		semaptr->sem_perm.cuid = cred->cr_uid;
527 		semaptr->sem_perm.uid = cred->cr_uid;
528 		semtot -= semaptr->sem_nsems;
529 		for (i = semaptr->sem_base - sem; i < semtot; i++)
530 			sem[i] = sem[i + semaptr->sem_nsems];
531 		for (i = 0; i < seminfo.semmni; i++) {
532 			if ((sema[i].sem_perm.mode & SEM_ALLOC) &&
533 			    sema[i].sem_base > semaptr->sem_base)
534 				sema[i].sem_base -= semaptr->sem_nsems;
535 		}
536 		semaptr->sem_perm.mode = 0;
537 		semundo_clear(semid, -1);
538 		wakeup((caddr_t)semaptr);
539 		break;
540 
541 	case IPC_SET:
542 		if ((error = ipcperm(td, &semaptr->sem_perm, IPC_M)))
543 			goto done2;
544 		if ((error = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
545 			goto done2;
546 		if ((error = copyin(real_arg.buf, (caddr_t)&sbuf,
547 		    sizeof(sbuf))) != 0) {
548 			goto done2;
549 		}
550 		semaptr->sem_perm.uid = sbuf.sem_perm.uid;
551 		semaptr->sem_perm.gid = sbuf.sem_perm.gid;
552 		semaptr->sem_perm.mode = (semaptr->sem_perm.mode & ~0777) |
553 		    (sbuf.sem_perm.mode & 0777);
554 		semaptr->sem_ctime = time_second;
555 		break;
556 
557 	case IPC_STAT:
558 		if ((error = ipcperm(td, &semaptr->sem_perm, IPC_R)))
559 			goto done2;
560 		if ((error = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
561 			goto done2;
562 		error = copyout((caddr_t)semaptr, real_arg.buf,
563 				sizeof(struct semid_ds));
564 		break;
565 
566 	case GETNCNT:
567 		if ((error = ipcperm(td, &semaptr->sem_perm, IPC_R)))
568 			goto done2;
569 		if (semnum < 0 || semnum >= semaptr->sem_nsems) {
570 			error = EINVAL;
571 			goto done2;
572 		}
573 		rval = semaptr->sem_base[semnum].semncnt;
574 		break;
575 
576 	case GETPID:
577 		if ((error = ipcperm(td, &semaptr->sem_perm, IPC_R)))
578 			goto done2;
579 		if (semnum < 0 || semnum >= semaptr->sem_nsems) {
580 			error = EINVAL;
581 			goto done2;
582 		}
583 		rval = semaptr->sem_base[semnum].sempid;
584 		break;
585 
586 	case GETVAL:
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].semval;
594 		break;
595 
596 	case GETALL:
597 		if ((error = ipcperm(td, &semaptr->sem_perm, IPC_R)))
598 			goto done2;
599 		if ((error = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
600 			goto done2;
601 		for (i = 0; i < semaptr->sem_nsems; i++) {
602 			error = copyout((caddr_t)&semaptr->sem_base[i].semval,
603 			    &real_arg.array[i], sizeof(real_arg.array[0]));
604 			if (error != 0)
605 				break;
606 		}
607 		break;
608 
609 	case GETZCNT:
610 		if ((error = ipcperm(td, &semaptr->sem_perm, IPC_R)))
611 			goto done2;
612 		if (semnum < 0 || semnum >= semaptr->sem_nsems) {
613 			error = EINVAL;
614 			goto done2;
615 		}
616 		rval = semaptr->sem_base[semnum].semzcnt;
617 		break;
618 
619 	case SETVAL:
620 		if ((error = ipcperm(td, &semaptr->sem_perm, IPC_W)))
621 			goto done2;
622 		if (semnum < 0 || semnum >= semaptr->sem_nsems) {
623 			error = EINVAL;
624 			goto done2;
625 		}
626 		if ((error = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
627 			goto done2;
628 		if (real_arg.val < 0 || real_arg.val > seminfo.semvmx) {
629 			error = ERANGE;
630 			goto done2;
631 		}
632 		semaptr->sem_base[semnum].semval = real_arg.val;
633 		semundo_clear(semid, semnum);
634 		wakeup((caddr_t)semaptr);
635 		break;
636 
637 	case SETALL:
638 		if ((error = ipcperm(td, &semaptr->sem_perm, IPC_W)))
639 			goto done2;
640 		if ((error = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
641 			goto done2;
642 		for (i = 0; i < semaptr->sem_nsems; i++) {
643 			error = copyin(&real_arg.array[i],
644 			    (caddr_t)&usval, sizeof(real_arg.array[0]));
645 			if (error != 0)
646 				break;
647 			if (usval > seminfo.semvmx) {
648 				error = ERANGE;
649 				break;
650 			}
651 			semaptr->sem_base[i].semval = usval;
652 		}
653 		semundo_clear(semid, -1);
654 		wakeup((caddr_t)semaptr);
655 		break;
656 
657 	default:
658 		error = EINVAL;
659 		break;
660 	}
661 
662 	if (error == 0)
663 		td->td_retval[0] = rval;
664 done2:
665 	mtx_unlock(&Giant);
666 	return(error);
667 }
668 
669 #ifndef _SYS_SYSPROTO_H_
670 struct semget_args {
671 	key_t	key;
672 	int	nsems;
673 	int	semflg;
674 };
675 #endif
676 
677 /*
678  * MPSAFE
679  */
680 int
681 semget(td, uap)
682 	struct thread *td;
683 	register struct semget_args *uap;
684 {
685 	int semid, error = 0;
686 	int key = uap->key;
687 	int nsems = uap->nsems;
688 	int semflg = uap->semflg;
689 	struct ucred *cred = td->td_ucred;
690 
691 #ifdef SEM_DEBUG
692 	printf("semget(0x%x, %d, 0%o)\n", key, nsems, semflg);
693 #endif
694 	if (!jail_sysvipc_allowed && jailed(td->td_ucred))
695 		return (ENOSYS);
696 
697 	mtx_lock(&Giant);
698 	if (key != IPC_PRIVATE) {
699 		for (semid = 0; semid < seminfo.semmni; semid++) {
700 			if ((sema[semid].sem_perm.mode & SEM_ALLOC) &&
701 			    sema[semid].sem_perm.key == key)
702 				break;
703 		}
704 		if (semid < seminfo.semmni) {
705 #ifdef SEM_DEBUG
706 			printf("found public key\n");
707 #endif
708 			if ((error = ipcperm(td, &sema[semid].sem_perm,
709 			    semflg & 0700))) {
710 				goto done2;
711 			}
712 			if (nsems > 0 && sema[semid].sem_nsems < nsems) {
713 #ifdef SEM_DEBUG
714 				printf("too small\n");
715 #endif
716 				error = EINVAL;
717 				goto done2;
718 			}
719 			if ((semflg & IPC_CREAT) && (semflg & IPC_EXCL)) {
720 #ifdef SEM_DEBUG
721 				printf("not exclusive\n");
722 #endif
723 				error = EEXIST;
724 				goto done2;
725 			}
726 			goto found;
727 		}
728 	}
729 
730 #ifdef SEM_DEBUG
731 	printf("need to allocate the semid_ds\n");
732 #endif
733 	if (key == IPC_PRIVATE || (semflg & IPC_CREAT)) {
734 		if (nsems <= 0 || nsems > seminfo.semmsl) {
735 #ifdef SEM_DEBUG
736 			printf("nsems out of range (0<%d<=%d)\n", nsems,
737 			    seminfo.semmsl);
738 #endif
739 			error = EINVAL;
740 			goto done2;
741 		}
742 		if (nsems > seminfo.semmns - semtot) {
743 #ifdef SEM_DEBUG
744 			printf("not enough semaphores left (need %d, got %d)\n",
745 			    nsems, seminfo.semmns - semtot);
746 #endif
747 			error = ENOSPC;
748 			goto done2;
749 		}
750 		for (semid = 0; semid < seminfo.semmni; semid++) {
751 			if ((sema[semid].sem_perm.mode & SEM_ALLOC) == 0)
752 				break;
753 		}
754 		if (semid == seminfo.semmni) {
755 #ifdef SEM_DEBUG
756 			printf("no more semid_ds's available\n");
757 #endif
758 			error = ENOSPC;
759 			goto done2;
760 		}
761 #ifdef SEM_DEBUG
762 		printf("semid %d is available\n", semid);
763 #endif
764 		sema[semid].sem_perm.key = key;
765 		sema[semid].sem_perm.cuid = cred->cr_uid;
766 		sema[semid].sem_perm.uid = cred->cr_uid;
767 		sema[semid].sem_perm.cgid = cred->cr_gid;
768 		sema[semid].sem_perm.gid = cred->cr_gid;
769 		sema[semid].sem_perm.mode = (semflg & 0777) | SEM_ALLOC;
770 		sema[semid].sem_perm.seq =
771 		    (sema[semid].sem_perm.seq + 1) & 0x7fff;
772 		sema[semid].sem_nsems = nsems;
773 		sema[semid].sem_otime = 0;
774 		sema[semid].sem_ctime = time_second;
775 		sema[semid].sem_base = &sem[semtot];
776 		semtot += nsems;
777 		bzero(sema[semid].sem_base,
778 		    sizeof(sema[semid].sem_base[0])*nsems);
779 #ifdef SEM_DEBUG
780 		printf("sembase = 0x%x, next = 0x%x\n", sema[semid].sem_base,
781 		    &sem[semtot]);
782 #endif
783 	} else {
784 #ifdef SEM_DEBUG
785 		printf("didn't find it and wasn't asked to create it\n");
786 #endif
787 		error = ENOENT;
788 		goto done2;
789 	}
790 
791 found:
792 	td->td_retval[0] = IXSEQ_TO_IPCID(semid, sema[semid].sem_perm);
793 done2:
794 	mtx_unlock(&Giant);
795 	return (error);
796 }
797 
798 #ifndef _SYS_SYSPROTO_H_
799 struct semop_args {
800 	int	semid;
801 	struct	sembuf *sops;
802 	u_int	nsops;
803 };
804 #endif
805 
806 /*
807  * MPSAFE
808  */
809 int
810 semop(td, uap)
811 	struct thread *td;
812 	register struct semop_args *uap;
813 {
814 	int semid = uap->semid;
815 	u_int nsops = uap->nsops;
816 	struct sembuf *sops = NULL;
817 	register struct semid_ds *semaptr;
818 	register struct sembuf *sopptr = 0;
819 	register struct sem *semptr = 0;
820 	struct sem_undo *suptr;
821 	int i, j, error;
822 	int do_wakeup, do_undos;
823 
824 #ifdef SEM_DEBUG
825 	printf("call to semop(%d, 0x%x, %u)\n", semid, sops, nsops);
826 #endif
827 
828 	if (!jail_sysvipc_allowed && jailed(td->td_ucred))
829 		return (ENOSYS);
830 
831 	mtx_lock(&Giant);
832 	semid = IPCID_TO_IX(semid);	/* Convert back to zero origin */
833 
834 	if (semid < 0 || semid >= seminfo.semmni) {
835 		error = EINVAL;
836 		goto done2;
837 	}
838 
839 	semaptr = &sema[semid];
840 	if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0) {
841 		error = EINVAL;
842 		goto done2;
843 	}
844 	if (semaptr->sem_perm.seq != IPCID_TO_SEQ(uap->semid)) {
845 		error = EINVAL;
846 		goto done2;
847 	}
848 	if (nsops > seminfo.semopm) {
849 #ifdef SEM_DEBUG
850 		printf("too many sops (max=%d, nsops=%d)\n", seminfo.semopm,
851 		    nsops);
852 #endif
853 		error = E2BIG;
854 		goto done2;
855 	}
856 
857 	/* Allocate memory for sem_ops */
858 	sops = malloc(nsops * sizeof(sops[0]), M_SEM, M_WAITOK);
859 	if (!sops)
860 		panic("Failed to allocate %d sem_ops", nsops);
861 
862 	if ((error = copyin(uap->sops, sops, nsops * sizeof(sops[0]))) != 0) {
863 #ifdef SEM_DEBUG
864 		printf("error = %d from copyin(%08x, %08x, %d)\n", error,
865 		    uap->sops, sops, nsops * sizeof(sops[0]));
866 #endif
867 		goto done2;
868 	}
869 
870 	/*
871 	 * Initial pass thru sops to see what permissions are needed.
872 	 * Also perform any checks that don't need repeating on each
873 	 * attempt to satisfy the request vector.
874 	 */
875 	j = 0;		/* permission needed */
876 	do_undos = 0;
877 	for (i = 0; i < nsops; i++) {
878 		sopptr = &sops[i];
879 		if (sopptr->sem_num >= semaptr->sem_nsems) {
880 			error = EFBIG;
881 			goto done2;
882 		}
883 		if (sopptr->sem_flg & SEM_UNDO && sopptr->sem_op != 0)
884 			do_undos = 1;
885 		j |= (sopptr->sem_op == 0) ? SEM_R : SEM_A;
886 	}
887 
888 	if ((error = ipcperm(td, &semaptr->sem_perm, j))) {
889 #ifdef SEM_DEBUG
890 		printf("error = %d from ipaccess\n", error);
891 #endif
892 		goto done2;
893 	}
894 
895 	/*
896 	 * Loop trying to satisfy the vector of requests.
897 	 * If we reach a point where we must wait, any requests already
898 	 * performed are rolled back and we go to sleep until some other
899 	 * process wakes us up.  At this point, we start all over again.
900 	 *
901 	 * This ensures that from the perspective of other tasks, a set
902 	 * of requests is atomic (never partially satisfied).
903 	 */
904 	for (;;) {
905 		do_wakeup = 0;
906 		error = 0;	/* error return if necessary */
907 
908 		for (i = 0; i < nsops; i++) {
909 			sopptr = &sops[i];
910 			semptr = &semaptr->sem_base[sopptr->sem_num];
911 
912 #ifdef SEM_DEBUG
913 			printf("semop:  semaptr=%x, sem_base=%x, semptr=%x, sem[%d]=%d : op=%d, flag=%s\n",
914 			    semaptr, semaptr->sem_base, semptr,
915 			    sopptr->sem_num, semptr->semval, sopptr->sem_op,
916 			    (sopptr->sem_flg & IPC_NOWAIT) ? "nowait" : "wait");
917 #endif
918 
919 			if (sopptr->sem_op < 0) {
920 				if (semptr->semval + sopptr->sem_op < 0) {
921 #ifdef SEM_DEBUG
922 					printf("semop:  can't do it now\n");
923 #endif
924 					break;
925 				} else {
926 					semptr->semval += sopptr->sem_op;
927 					if (semptr->semval == 0 &&
928 					    semptr->semzcnt > 0)
929 						do_wakeup = 1;
930 				}
931 			} else if (sopptr->sem_op == 0) {
932 				if (semptr->semval != 0) {
933 #ifdef SEM_DEBUG
934 					printf("semop:  not zero now\n");
935 #endif
936 					break;
937 				}
938 			} else if (semptr->semval + sopptr->sem_op >
939 			    seminfo.semvmx) {
940 				error = ERANGE;
941 				break;
942 			} else {
943 				if (semptr->semncnt > 0)
944 					do_wakeup = 1;
945 				semptr->semval += sopptr->sem_op;
946 			}
947 		}
948 
949 		/*
950 		 * Did we get through the entire vector?
951 		 */
952 		if (i >= nsops)
953 			goto done;
954 
955 		/*
956 		 * No ... rollback anything that we've already done
957 		 */
958 #ifdef SEM_DEBUG
959 		printf("semop:  rollback 0 through %d\n", i-1);
960 #endif
961 		for (j = 0; j < i; j++)
962 			semaptr->sem_base[sops[j].sem_num].semval -=
963 			    sops[j].sem_op;
964 
965 		/* If we detected an error, return it */
966 		if (error != 0)
967 			goto done2;
968 
969 		/*
970 		 * If the request that we couldn't satisfy has the
971 		 * NOWAIT flag set then return with EAGAIN.
972 		 */
973 		if (sopptr->sem_flg & IPC_NOWAIT) {
974 			error = EAGAIN;
975 			goto done2;
976 		}
977 
978 		if (sopptr->sem_op == 0)
979 			semptr->semzcnt++;
980 		else
981 			semptr->semncnt++;
982 
983 #ifdef SEM_DEBUG
984 		printf("semop:  good night!\n");
985 #endif
986 		error = tsleep((caddr_t)semaptr, (PZERO - 4) | PCATCH,
987 		    "semwait", 0);
988 #ifdef SEM_DEBUG
989 		printf("semop:  good morning (error=%d)!\n", error);
990 #endif
991 
992 		if (error != 0) {
993 			error = EINTR;
994 			goto done2;
995 		}
996 #ifdef SEM_DEBUG
997 		printf("semop:  good morning!\n");
998 #endif
999 
1000 		/*
1001 		 * Make sure that the semaphore still exists
1002 		 */
1003 		if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0 ||
1004 		    semaptr->sem_perm.seq != IPCID_TO_SEQ(uap->semid)) {
1005 			error = EIDRM;
1006 			goto done2;
1007 		}
1008 
1009 		/*
1010 		 * The semaphore is still alive.  Readjust the count of
1011 		 * waiting processes.
1012 		 */
1013 		if (sopptr->sem_op == 0)
1014 			semptr->semzcnt--;
1015 		else
1016 			semptr->semncnt--;
1017 	}
1018 
1019 done:
1020 	/*
1021 	 * Process any SEM_UNDO requests.
1022 	 */
1023 	if (do_undos) {
1024 		suptr = NULL;
1025 		for (i = 0; i < nsops; i++) {
1026 			/*
1027 			 * We only need to deal with SEM_UNDO's for non-zero
1028 			 * op's.
1029 			 */
1030 			int adjval;
1031 
1032 			if ((sops[i].sem_flg & SEM_UNDO) == 0)
1033 				continue;
1034 			adjval = sops[i].sem_op;
1035 			if (adjval == 0)
1036 				continue;
1037 			error = semundo_adjust(td, &suptr, semid,
1038 			    sops[i].sem_num, -adjval);
1039 			if (error == 0)
1040 				continue;
1041 
1042 			/*
1043 			 * Oh-Oh!  We ran out of either sem_undo's or undo's.
1044 			 * Rollback the adjustments to this point and then
1045 			 * rollback the semaphore ups and down so we can return
1046 			 * with an error with all structures restored.  We
1047 			 * rollback the undo's in the exact reverse order that
1048 			 * we applied them.  This guarantees that we won't run
1049 			 * out of space as we roll things back out.
1050 			 */
1051 			for (j = i - 1; j >= 0; j--) {
1052 				if ((sops[j].sem_flg & SEM_UNDO) == 0)
1053 					continue;
1054 				adjval = sops[j].sem_op;
1055 				if (adjval == 0)
1056 					continue;
1057 				if (semundo_adjust(td, &suptr, semid,
1058 				    sops[j].sem_num, adjval) != 0)
1059 					panic("semop - can't undo undos");
1060 			}
1061 
1062 			for (j = 0; j < nsops; j++)
1063 				semaptr->sem_base[sops[j].sem_num].semval -=
1064 				    sops[j].sem_op;
1065 
1066 #ifdef SEM_DEBUG
1067 			printf("error = %d from semundo_adjust\n", error);
1068 #endif
1069 			goto done2;
1070 		} /* loop through the sops */
1071 	} /* if (do_undos) */
1072 
1073 	/* We're definitely done - set the sempid's and time */
1074 	for (i = 0; i < nsops; i++) {
1075 		sopptr = &sops[i];
1076 		semptr = &semaptr->sem_base[sopptr->sem_num];
1077 		semptr->sempid = td->td_proc->p_pid;
1078 	}
1079 	semaptr->sem_otime = time_second;
1080 
1081 	/*
1082 	 * Do a wakeup if any semaphore was up'd whilst something was
1083 	 * sleeping on it.
1084 	 */
1085 	if (do_wakeup) {
1086 #ifdef SEM_DEBUG
1087 		printf("semop:  doing wakeup\n");
1088 #endif
1089 		wakeup((caddr_t)semaptr);
1090 #ifdef SEM_DEBUG
1091 		printf("semop:  back from wakeup\n");
1092 #endif
1093 	}
1094 #ifdef SEM_DEBUG
1095 	printf("semop:  done\n");
1096 #endif
1097 	td->td_retval[0] = 0;
1098 done2:
1099 	if (sops)
1100 	    free(sops, M_SEM);
1101 	mtx_unlock(&Giant);
1102 	return (error);
1103 }
1104 
1105 /*
1106  * Go through the undo structures for this process and apply the adjustments to
1107  * semaphores.
1108  */
1109 static void
1110 semexit_myhook(p)
1111 	struct proc *p;
1112 {
1113 	register struct sem_undo *suptr;
1114 	register struct sem_undo **supptr;
1115 
1116 	/*
1117 	 * Go through the chain of undo vectors looking for one
1118 	 * associated with this process.
1119 	 */
1120 
1121 	for (supptr = &semu_list; (suptr = *supptr) != NULL;
1122 	    supptr = &suptr->un_next) {
1123 		if (suptr->un_proc == p)
1124 			break;
1125 	}
1126 
1127 	if (suptr == NULL)
1128 		return;
1129 
1130 #ifdef SEM_DEBUG
1131 	printf("proc @%08x has undo structure with %d entries\n", p,
1132 	    suptr->un_cnt);
1133 #endif
1134 
1135 	/*
1136 	 * If there are any active undo elements then process them.
1137 	 */
1138 	if (suptr->un_cnt > 0) {
1139 		int ix;
1140 
1141 		for (ix = 0; ix < suptr->un_cnt; ix++) {
1142 			int semid = suptr->un_ent[ix].un_id;
1143 			int semnum = suptr->un_ent[ix].un_num;
1144 			int adjval = suptr->un_ent[ix].un_adjval;
1145 			struct semid_ds *semaptr;
1146 
1147 			semaptr = &sema[semid];
1148 			if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0)
1149 				panic("semexit - semid not allocated");
1150 			if (semnum >= semaptr->sem_nsems)
1151 				panic("semexit - semnum out of range");
1152 
1153 #ifdef SEM_DEBUG
1154 			printf("semexit:  %08x id=%d num=%d(adj=%d) ; sem=%d\n",
1155 			    suptr->un_proc, suptr->un_ent[ix].un_id,
1156 			    suptr->un_ent[ix].un_num,
1157 			    suptr->un_ent[ix].un_adjval,
1158 			    semaptr->sem_base[semnum].semval);
1159 #endif
1160 
1161 			if (adjval < 0) {
1162 				if (semaptr->sem_base[semnum].semval < -adjval)
1163 					semaptr->sem_base[semnum].semval = 0;
1164 				else
1165 					semaptr->sem_base[semnum].semval +=
1166 					    adjval;
1167 			} else
1168 				semaptr->sem_base[semnum].semval += adjval;
1169 
1170 			wakeup((caddr_t)semaptr);
1171 #ifdef SEM_DEBUG
1172 			printf("semexit:  back from wakeup\n");
1173 #endif
1174 		}
1175 	}
1176 
1177 	/*
1178 	 * Deallocate the undo vector.
1179 	 */
1180 #ifdef SEM_DEBUG
1181 	printf("removing vector\n");
1182 #endif
1183 	suptr->un_proc = NULL;
1184 	*supptr = suptr->un_next;
1185 }
1186 
1187 static int
1188 sysctl_sema(SYSCTL_HANDLER_ARGS)
1189 {
1190 
1191 	return (SYSCTL_OUT(req, sema,
1192 	    sizeof(struct semid_ds) * seminfo.semmni));
1193 }
1194