xref: /freebsd/sys/kern/sysv_sem.c (revision eacee0ff7ec955b32e09515246bd97b6edcd2b0f)
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 __P((void));
30 static int sysvsem_modload __P((struct module *, int, void *));
31 static int semunload __P((void));
32 static void semexit_myhook __P((struct proc *p));
33 static int sysctl_sema __P((SYSCTL_HANDLER_ARGS));
34 
35 #ifndef _SYS_SYSPROTO_H_
36 struct __semctl_args;
37 int __semctl __P((struct thread *td, struct __semctl_args *uap));
38 struct semget_args;
39 int semget __P((struct thread *td, struct semget_args *uap));
40 struct semop_args;
41 int semop __P((struct thread *td, struct semop_args *uap));
42 #endif
43 
44 static struct sem_undo *semu_alloc __P((struct thread *td));
45 static int semundo_adjust __P((struct thread *td, struct sem_undo **supptr,
46 		int semid, int semnum, int adjval));
47 static void semundo_clear __P((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 	mtx_lock(&Giant);
265 	if (!jail_sysvipc_allowed && jailed(td->td_proc->p_ucred)) {
266 		error = ENOSYS;
267 		goto done2;
268 	}
269 	if (uap->which >= sizeof(semcalls)/sizeof(semcalls[0])) {
270 		error = EINVAL;
271 		goto done2;
272 	}
273 	error = (*semcalls[uap->which])(td, &uap->a2);
274 done2:
275 	mtx_unlock(&Giant);
276 	return (error);
277 }
278 
279 /*
280  * Allocate a new sem_undo structure for a process
281  * (returns ptr to structure or NULL if no more room)
282  */
283 
284 static struct sem_undo *
285 semu_alloc(td)
286 	struct thread *td;
287 {
288 	register int i;
289 	register struct sem_undo *suptr;
290 	register struct sem_undo **supptr;
291 	int attempt;
292 
293 	/*
294 	 * Try twice to allocate something.
295 	 * (we'll purge any empty structures after the first pass so
296 	 * two passes are always enough)
297 	 */
298 
299 	for (attempt = 0; attempt < 2; attempt++) {
300 		/*
301 		 * Look for a free structure.
302 		 * Fill it in and return it if we find one.
303 		 */
304 
305 		for (i = 0; i < seminfo.semmnu; i++) {
306 			suptr = SEMU(i);
307 			if (suptr->un_proc == NULL) {
308 				suptr->un_next = semu_list;
309 				semu_list = suptr;
310 				suptr->un_cnt = 0;
311 				suptr->un_proc = td->td_proc;
312 				return(suptr);
313 			}
314 		}
315 
316 		/*
317 		 * We didn't find a free one, if this is the first attempt
318 		 * then try to free some structures.
319 		 */
320 
321 		if (attempt == 0) {
322 			/* All the structures are in use - try to free some */
323 			int did_something = 0;
324 
325 			supptr = &semu_list;
326 			while ((suptr = *supptr) != NULL) {
327 				if (suptr->un_cnt == 0)  {
328 					suptr->un_proc = NULL;
329 					*supptr = suptr->un_next;
330 					did_something = 1;
331 				} else
332 					supptr = &(suptr->un_next);
333 			}
334 
335 			/* If we didn't free anything then just give-up */
336 			if (!did_something)
337 				return(NULL);
338 		} else {
339 			/*
340 			 * The second pass failed even though we freed
341 			 * something after the first pass!
342 			 * This is IMPOSSIBLE!
343 			 */
344 			panic("semu_alloc - second attempt failed");
345 		}
346 	}
347 	return (NULL);
348 }
349 
350 /*
351  * Adjust a particular entry for a particular proc
352  */
353 
354 static int
355 semundo_adjust(td, supptr, semid, semnum, adjval)
356 	register struct thread *td;
357 	struct sem_undo **supptr;
358 	int semid, semnum;
359 	int adjval;
360 {
361 	struct proc *p = td->td_proc;
362 	register struct sem_undo *suptr;
363 	register struct undo *sunptr;
364 	int i;
365 
366 	/* Look for and remember the sem_undo if the caller doesn't provide
367 	   it */
368 
369 	suptr = *supptr;
370 	if (suptr == NULL) {
371 		for (suptr = semu_list; suptr != NULL;
372 		    suptr = suptr->un_next) {
373 			if (suptr->un_proc == p) {
374 				*supptr = suptr;
375 				break;
376 			}
377 		}
378 		if (suptr == NULL) {
379 			if (adjval == 0)
380 				return(0);
381 			suptr = semu_alloc(td);
382 			if (suptr == NULL)
383 				return(ENOSPC);
384 			*supptr = suptr;
385 		}
386 	}
387 
388 	/*
389 	 * Look for the requested entry and adjust it (delete if adjval becomes
390 	 * 0).
391 	 */
392 	sunptr = &suptr->un_ent[0];
393 	for (i = 0; i < suptr->un_cnt; i++, sunptr++) {
394 		if (sunptr->un_id != semid || sunptr->un_num != semnum)
395 			continue;
396 		if (adjval != 0) {
397 			adjval += sunptr->un_adjval;
398 			if (adjval > seminfo.semaem || adjval < -seminfo.semaem)
399 				return (ERANGE);
400 		}
401 		sunptr->un_adjval = adjval;
402 		if (sunptr->un_adjval == 0) {
403 			suptr->un_cnt--;
404 			if (i < suptr->un_cnt)
405 				suptr->un_ent[i] =
406 				    suptr->un_ent[suptr->un_cnt];
407 		}
408 		return(0);
409 	}
410 
411 	/* Didn't find the right entry - create it */
412 	if (adjval == 0)
413 		return(0);
414 	if (adjval > seminfo.semaem || adjval < -seminfo.semaem)
415 		return (ERANGE);
416 	if (suptr->un_cnt != seminfo.semume) {
417 		sunptr = &suptr->un_ent[suptr->un_cnt];
418 		suptr->un_cnt++;
419 		sunptr->un_adjval = adjval;
420 		sunptr->un_id = semid; sunptr->un_num = semnum;
421 	} else
422 		return(EINVAL);
423 	return(0);
424 }
425 
426 static void
427 semundo_clear(semid, semnum)
428 	int semid, semnum;
429 {
430 	register struct sem_undo *suptr;
431 
432 	for (suptr = semu_list; suptr != NULL; suptr = suptr->un_next) {
433 		register struct undo *sunptr = &suptr->un_ent[0];
434 		register int i = 0;
435 
436 		while (i < suptr->un_cnt) {
437 			if (sunptr->un_id == semid) {
438 				if (semnum == -1 || sunptr->un_num == semnum) {
439 					suptr->un_cnt--;
440 					if (i < suptr->un_cnt) {
441 						suptr->un_ent[i] =
442 						  suptr->un_ent[suptr->un_cnt];
443 						continue;
444 					}
445 				}
446 				if (semnum != -1)
447 					break;
448 			}
449 			i++, sunptr++;
450 		}
451 	}
452 }
453 
454 /*
455  * Note that the user-mode half of this passes a union, not a pointer
456  */
457 #ifndef _SYS_SYSPROTO_H_
458 struct __semctl_args {
459 	int	semid;
460 	int	semnum;
461 	int	cmd;
462 	union	semun *arg;
463 };
464 #endif
465 
466 /*
467  * MPSAFE
468  */
469 int
470 __semctl(td, uap)
471 	struct thread *td;
472 	register struct __semctl_args *uap;
473 {
474 	int semid = uap->semid;
475 	int semnum = uap->semnum;
476 	int cmd = uap->cmd;
477 	union semun *arg = uap->arg;
478 	union semun real_arg;
479 	struct ucred *cred = td->td_proc->p_ucred;
480 	int i, rval, error;
481 	struct semid_ds sbuf;
482 	register struct semid_ds *semaptr;
483 	u_short usval;
484 
485 #ifdef SEM_DEBUG
486 	printf("call to semctl(%d, %d, %d, 0x%x)\n", semid, semnum, cmd, arg);
487 #endif
488 	mtx_lock(&Giant);
489 	if (!jail_sysvipc_allowed && jailed(td->td_proc->p_ucred)) {
490 		error = ENOSYS;
491 		goto done2;
492 	}
493 
494 	switch(cmd) {
495 	case SEM_STAT:
496 		if (semid < 0 || semid >= seminfo.semmsl)
497 			return(EINVAL);
498 		semaptr = &sema[semid];
499 		if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0 )
500 			return(EINVAL);
501 		if ((error = ipcperm(td, &semaptr->sem_perm, IPC_R)))
502 			return(error);
503 		if ((error = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
504 			return(error);
505 		error = copyout((caddr_t)semaptr, real_arg.buf,
506 			sizeof(struct semid_ds));
507 		rval = IXSEQ_TO_IPCID(semid,semaptr->sem_perm);
508 		if (error == 0)
509 			td->td_retval[0] = rval;
510 		goto done2;
511 	}
512 
513 	semid = IPCID_TO_IX(semid);
514 	if (semid < 0 || semid >= seminfo.semmsl) {
515 		error = EINVAL;
516 		goto done2;
517 	}
518 
519 	semaptr = &sema[semid];
520 	if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0 ||
521 	    semaptr->sem_perm.seq != IPCID_TO_SEQ(uap->semid)) {
522 		error = EINVAL;
523 		goto done2;
524 	}
525 
526 	error = 0;
527 	rval = 0;
528 
529 	switch (cmd) {
530 	case IPC_RMID:
531 		if ((error = ipcperm(td, &semaptr->sem_perm, IPC_M)))
532 			goto done2;
533 		semaptr->sem_perm.cuid = cred->cr_uid;
534 		semaptr->sem_perm.uid = cred->cr_uid;
535 		semtot -= semaptr->sem_nsems;
536 		for (i = semaptr->sem_base - sem; i < semtot; i++)
537 			sem[i] = sem[i + semaptr->sem_nsems];
538 		for (i = 0; i < seminfo.semmni; i++) {
539 			if ((sema[i].sem_perm.mode & SEM_ALLOC) &&
540 			    sema[i].sem_base > semaptr->sem_base)
541 				sema[i].sem_base -= semaptr->sem_nsems;
542 		}
543 		semaptr->sem_perm.mode = 0;
544 		semundo_clear(semid, -1);
545 		wakeup((caddr_t)semaptr);
546 		break;
547 
548 	case IPC_SET:
549 		if ((error = ipcperm(td, &semaptr->sem_perm, IPC_M)))
550 			goto done2;
551 		if ((error = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
552 			goto done2;
553 		if ((error = copyin(real_arg.buf, (caddr_t)&sbuf,
554 		    sizeof(sbuf))) != 0) {
555 			goto done2;
556 		}
557 		semaptr->sem_perm.uid = sbuf.sem_perm.uid;
558 		semaptr->sem_perm.gid = sbuf.sem_perm.gid;
559 		semaptr->sem_perm.mode = (semaptr->sem_perm.mode & ~0777) |
560 		    (sbuf.sem_perm.mode & 0777);
561 		semaptr->sem_ctime = time_second;
562 		break;
563 
564 	case IPC_STAT:
565 		if ((error = ipcperm(td, &semaptr->sem_perm, IPC_R)))
566 			goto done2;
567 		if ((error = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
568 			goto done2;
569 		error = copyout((caddr_t)semaptr, real_arg.buf,
570 				sizeof(struct semid_ds));
571 		break;
572 
573 	case GETNCNT:
574 		if ((error = ipcperm(td, &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].semncnt;
581 		break;
582 
583 	case GETPID:
584 		if ((error = ipcperm(td, &semaptr->sem_perm, IPC_R)))
585 			goto done2;
586 		if (semnum < 0 || semnum >= semaptr->sem_nsems) {
587 			error = EINVAL;
588 			goto done2;
589 		}
590 		rval = semaptr->sem_base[semnum].sempid;
591 		break;
592 
593 	case GETVAL:
594 		if ((error = ipcperm(td, &semaptr->sem_perm, IPC_R)))
595 			goto done2;
596 		if (semnum < 0 || semnum >= semaptr->sem_nsems) {
597 			error = EINVAL;
598 			goto done2;
599 		}
600 		rval = semaptr->sem_base[semnum].semval;
601 		break;
602 
603 	case GETALL:
604 		if ((error = ipcperm(td, &semaptr->sem_perm, IPC_R)))
605 			goto done2;
606 		if ((error = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
607 			goto done2;
608 		for (i = 0; i < semaptr->sem_nsems; i++) {
609 			error = copyout((caddr_t)&semaptr->sem_base[i].semval,
610 			    &real_arg.array[i], sizeof(real_arg.array[0]));
611 			if (error != 0)
612 				break;
613 		}
614 		break;
615 
616 	case GETZCNT:
617 		if ((error = ipcperm(td, &semaptr->sem_perm, IPC_R)))
618 			goto done2;
619 		if (semnum < 0 || semnum >= semaptr->sem_nsems) {
620 			error = EINVAL;
621 			goto done2;
622 		}
623 		rval = semaptr->sem_base[semnum].semzcnt;
624 		break;
625 
626 	case SETVAL:
627 		if ((error = ipcperm(td, &semaptr->sem_perm, IPC_W)))
628 			goto done2;
629 		if (semnum < 0 || semnum >= semaptr->sem_nsems) {
630 			error = EINVAL;
631 			goto done2;
632 		}
633 		if ((error = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
634 			goto done2;
635 		if (real_arg.val < 0 || real_arg.val > seminfo.semvmx) {
636 			error = ERANGE;
637 			goto done2;
638 		}
639 		semaptr->sem_base[semnum].semval = real_arg.val;
640 		semundo_clear(semid, semnum);
641 		wakeup((caddr_t)semaptr);
642 		break;
643 
644 	case SETALL:
645 		if ((error = ipcperm(td, &semaptr->sem_perm, IPC_W)))
646 			goto done2;
647 		if ((error = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
648 			goto done2;
649 		for (i = 0; i < semaptr->sem_nsems; i++) {
650 			error = copyin(&real_arg.array[i],
651 			    (caddr_t)&usval, sizeof(real_arg.array[0]));
652 			if (error != 0)
653 				break;
654 			if (usval > seminfo.semvmx) {
655 				error = ERANGE;
656 				break;
657 			}
658 			semaptr->sem_base[i].semval = usval;
659 		}
660 		semundo_clear(semid, -1);
661 		wakeup((caddr_t)semaptr);
662 		break;
663 
664 	default:
665 		error = EINVAL;
666 		break;
667 	}
668 
669 	if (error == 0)
670 		td->td_retval[0] = rval;
671 done2:
672 	mtx_unlock(&Giant);
673 	return(error);
674 }
675 
676 #ifndef _SYS_SYSPROTO_H_
677 struct semget_args {
678 	key_t	key;
679 	int	nsems;
680 	int	semflg;
681 };
682 #endif
683 
684 /*
685  * MPSAFE
686  */
687 int
688 semget(td, uap)
689 	struct thread *td;
690 	register struct semget_args *uap;
691 {
692 	int semid, error = 0;
693 	int key = uap->key;
694 	int nsems = uap->nsems;
695 	int semflg = uap->semflg;
696 	struct ucred *cred = td->td_proc->p_ucred;
697 
698 #ifdef SEM_DEBUG
699 	printf("semget(0x%x, %d, 0%o)\n", key, nsems, semflg);
700 #endif
701 	mtx_lock(&Giant);
702 	if (!jail_sysvipc_allowed && jailed(td->td_proc->p_ucred)) {
703 		error = ENOSYS;
704 		goto done2;
705 	}
706 
707 	if (key != IPC_PRIVATE) {
708 		for (semid = 0; semid < seminfo.semmni; semid++) {
709 			if ((sema[semid].sem_perm.mode & SEM_ALLOC) &&
710 			    sema[semid].sem_perm.key == key)
711 				break;
712 		}
713 		if (semid < seminfo.semmni) {
714 #ifdef SEM_DEBUG
715 			printf("found public key\n");
716 #endif
717 			if ((error = ipcperm(td, &sema[semid].sem_perm,
718 			    semflg & 0700))) {
719 				goto done2;
720 			}
721 			if (nsems > 0 && sema[semid].sem_nsems < nsems) {
722 #ifdef SEM_DEBUG
723 				printf("too small\n");
724 #endif
725 				error = EINVAL;
726 				goto done2;
727 			}
728 			if ((semflg & IPC_CREAT) && (semflg & IPC_EXCL)) {
729 #ifdef SEM_DEBUG
730 				printf("not exclusive\n");
731 #endif
732 				error = EEXIST;
733 				goto done2;
734 			}
735 			goto found;
736 		}
737 	}
738 
739 #ifdef SEM_DEBUG
740 	printf("need to allocate the semid_ds\n");
741 #endif
742 	if (key == IPC_PRIVATE || (semflg & IPC_CREAT)) {
743 		if (nsems <= 0 || nsems > seminfo.semmsl) {
744 #ifdef SEM_DEBUG
745 			printf("nsems out of range (0<%d<=%d)\n", nsems,
746 			    seminfo.semmsl);
747 #endif
748 			error = EINVAL;
749 			goto done2;
750 		}
751 		if (nsems > seminfo.semmns - semtot) {
752 #ifdef SEM_DEBUG
753 			printf("not enough semaphores left (need %d, got %d)\n",
754 			    nsems, seminfo.semmns - semtot);
755 #endif
756 			error = ENOSPC;
757 			goto done2;
758 		}
759 		for (semid = 0; semid < seminfo.semmni; semid++) {
760 			if ((sema[semid].sem_perm.mode & SEM_ALLOC) == 0)
761 				break;
762 		}
763 		if (semid == seminfo.semmni) {
764 #ifdef SEM_DEBUG
765 			printf("no more semid_ds's available\n");
766 #endif
767 			error = ENOSPC;
768 			goto done2;
769 		}
770 #ifdef SEM_DEBUG
771 		printf("semid %d is available\n", semid);
772 #endif
773 		sema[semid].sem_perm.key = key;
774 		sema[semid].sem_perm.cuid = cred->cr_uid;
775 		sema[semid].sem_perm.uid = cred->cr_uid;
776 		sema[semid].sem_perm.cgid = cred->cr_gid;
777 		sema[semid].sem_perm.gid = cred->cr_gid;
778 		sema[semid].sem_perm.mode = (semflg & 0777) | SEM_ALLOC;
779 		sema[semid].sem_perm.seq =
780 		    (sema[semid].sem_perm.seq + 1) & 0x7fff;
781 		sema[semid].sem_nsems = nsems;
782 		sema[semid].sem_otime = 0;
783 		sema[semid].sem_ctime = time_second;
784 		sema[semid].sem_base = &sem[semtot];
785 		semtot += nsems;
786 		bzero(sema[semid].sem_base,
787 		    sizeof(sema[semid].sem_base[0])*nsems);
788 #ifdef SEM_DEBUG
789 		printf("sembase = 0x%x, next = 0x%x\n", sema[semid].sem_base,
790 		    &sem[semtot]);
791 #endif
792 	} else {
793 #ifdef SEM_DEBUG
794 		printf("didn't find it and wasn't asked to create it\n");
795 #endif
796 		error = ENOENT;
797 		goto done2;
798 	}
799 
800 found:
801 	td->td_retval[0] = IXSEQ_TO_IPCID(semid, sema[semid].sem_perm);
802 done2:
803 	mtx_unlock(&Giant);
804 	return (error);
805 }
806 
807 #ifndef _SYS_SYSPROTO_H_
808 struct semop_args {
809 	int	semid;
810 	struct	sembuf *sops;
811 	u_int	nsops;
812 };
813 #endif
814 
815 /*
816  * MPSAFE
817  */
818 int
819 semop(td, uap)
820 	struct thread *td;
821 	register struct semop_args *uap;
822 {
823 	int semid = uap->semid;
824 	u_int nsops = uap->nsops;
825 	struct sembuf *sops = NULL;
826 	register struct semid_ds *semaptr;
827 	register struct sembuf *sopptr;
828 	register struct sem *semptr;
829 	struct sem_undo *suptr;
830 	int i, j, error;
831 	int do_wakeup, do_undos;
832 
833 #ifdef SEM_DEBUG
834 	printf("call to semop(%d, 0x%x, %u)\n", semid, sops, nsops);
835 #endif
836 
837 	mtx_lock(&Giant);
838 	if (!jail_sysvipc_allowed && jailed(td->td_proc->p_ucred)) {
839 		error = ENOSYS;
840 		goto done2;
841 	}
842 
843 	semid = IPCID_TO_IX(semid);	/* Convert back to zero origin */
844 
845 	if (semid < 0 || semid >= seminfo.semmsl) {
846 		error = EINVAL;
847 		goto done2;
848 	}
849 
850 	semaptr = &sema[semid];
851 	if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0) {
852 		error = EINVAL;
853 		goto done2;
854 	}
855 	if (semaptr->sem_perm.seq != IPCID_TO_SEQ(uap->semid)) {
856 		error = EINVAL;
857 		goto done2;
858 	}
859 	if (nsops > seminfo.semopm) {
860 #ifdef SEM_DEBUG
861 		printf("too many sops (max=%d, nsops=%d)\n", seminfo.semopm,
862 		    nsops);
863 #endif
864 		error = E2BIG;
865 		goto done2;
866 	}
867 
868 	/* Allocate memory for sem_ops */
869 	sops = malloc(nsops * sizeof(sops[0]), M_SEM, M_WAITOK);
870 	if (!sops)
871 		panic("Failed to allocate %d sem_ops", nsops);
872 
873 	if ((error = copyin(uap->sops, sops, nsops * sizeof(sops[0]))) != 0) {
874 #ifdef SEM_DEBUG
875 		printf("error = %d from copyin(%08x, %08x, %d)\n", error,
876 		    uap->sops, sops, nsops * sizeof(sops[0]));
877 #endif
878 		goto done2;
879 	}
880 
881 	/*
882 	 * Initial pass thru sops to see what permissions are needed.
883 	 * Also perform any checks that don't need repeating on each
884 	 * attempt to satisfy the request vector.
885 	 */
886 	j = 0;		/* permission needed */
887 	do_undos = 0;
888 	for (i = 0; i < nsops; i++) {
889 		sopptr = &sops[i];
890 		if (sopptr->sem_num >= semaptr->sem_nsems) {
891 			error = EFBIG;
892 			goto done2;
893 		}
894 		if (sopptr->sem_flg & SEM_UNDO && sopptr->sem_op != 0)
895 			do_undos = 1;
896 		j |= (sopptr->sem_op == 0) ? SEM_R : SEM_A;
897 	}
898 
899 	if ((error = ipcperm(td, &semaptr->sem_perm, j))) {
900 #ifdef SEM_DEBUG
901 		printf("error = %d from ipaccess\n", error);
902 #endif
903 		goto done2;
904 	}
905 
906 	/*
907 	 * Loop trying to satisfy the vector of requests.
908 	 * If we reach a point where we must wait, any requests already
909 	 * performed are rolled back and we go to sleep until some other
910 	 * process wakes us up.  At this point, we start all over again.
911 	 *
912 	 * This ensures that from the perspective of other tasks, a set
913 	 * of requests is atomic (never partially satisfied).
914 	 */
915 	for (;;) {
916 		do_wakeup = 0;
917 		error = 0;	/* error return if necessary */
918 
919 		for (i = 0; i < nsops; i++) {
920 			sopptr = &sops[i];
921 			semptr = &semaptr->sem_base[sopptr->sem_num];
922 
923 #ifdef SEM_DEBUG
924 			printf("semop:  semaptr=%x, sem_base=%x, semptr=%x, sem[%d]=%d : op=%d, flag=%s\n",
925 			    semaptr, semaptr->sem_base, semptr,
926 			    sopptr->sem_num, semptr->semval, sopptr->sem_op,
927 			    (sopptr->sem_flg & IPC_NOWAIT) ? "nowait" : "wait");
928 #endif
929 
930 			if (sopptr->sem_op < 0) {
931 				if (semptr->semval + sopptr->sem_op < 0) {
932 #ifdef SEM_DEBUG
933 					printf("semop:  can't do it now\n");
934 #endif
935 					break;
936 				} else {
937 					semptr->semval += sopptr->sem_op;
938 					if (semptr->semval == 0 &&
939 					    semptr->semzcnt > 0)
940 						do_wakeup = 1;
941 				}
942 			} else if (sopptr->sem_op == 0) {
943 				if (semptr->semval != 0) {
944 #ifdef SEM_DEBUG
945 					printf("semop:  not zero now\n");
946 #endif
947 					break;
948 				}
949 			} else if (semptr->semval + sopptr->sem_op >
950 			    seminfo.semvmx) {
951 				error = ERANGE;
952 				break;
953 			} else {
954 				if (semptr->semncnt > 0)
955 					do_wakeup = 1;
956 				semptr->semval += sopptr->sem_op;
957 			}
958 		}
959 
960 		/*
961 		 * Did we get through the entire vector?
962 		 */
963 		if (i >= nsops)
964 			goto done;
965 
966 		/*
967 		 * No ... rollback anything that we've already done
968 		 */
969 #ifdef SEM_DEBUG
970 		printf("semop:  rollback 0 through %d\n", i-1);
971 #endif
972 		for (j = 0; j < i; j++)
973 			semaptr->sem_base[sops[j].sem_num].semval -=
974 			    sops[j].sem_op;
975 
976 		/* If we detected an error, return it */
977 		if (error != 0)
978 			goto done2;
979 
980 		/*
981 		 * If the request that we couldn't satisfy has the
982 		 * NOWAIT flag set then return with EAGAIN.
983 		 */
984 		if (sopptr->sem_flg & IPC_NOWAIT) {
985 			error = EAGAIN;
986 			goto done2;
987 		}
988 
989 		if (sopptr->sem_op == 0)
990 			semptr->semzcnt++;
991 		else
992 			semptr->semncnt++;
993 
994 #ifdef SEM_DEBUG
995 		printf("semop:  good night!\n");
996 #endif
997 		error = tsleep((caddr_t)semaptr, (PZERO - 4) | PCATCH,
998 		    "semwait", 0);
999 #ifdef SEM_DEBUG
1000 		printf("semop:  good morning (error=%d)!\n", error);
1001 #endif
1002 
1003 		if (error != 0) {
1004 			error = EINTR;
1005 			goto done2;
1006 		}
1007 #ifdef SEM_DEBUG
1008 		printf("semop:  good morning!\n");
1009 #endif
1010 
1011 		/*
1012 		 * Make sure that the semaphore still exists
1013 		 */
1014 		if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0 ||
1015 		    semaptr->sem_perm.seq != IPCID_TO_SEQ(uap->semid)) {
1016 			error = EIDRM;
1017 			goto done2;
1018 		}
1019 
1020 		/*
1021 		 * The semaphore is still alive.  Readjust the count of
1022 		 * waiting processes.
1023 		 */
1024 		if (sopptr->sem_op == 0)
1025 			semptr->semzcnt--;
1026 		else
1027 			semptr->semncnt--;
1028 	}
1029 
1030 done:
1031 	/*
1032 	 * Process any SEM_UNDO requests.
1033 	 */
1034 	if (do_undos) {
1035 		suptr = NULL;
1036 		for (i = 0; i < nsops; i++) {
1037 			/*
1038 			 * We only need to deal with SEM_UNDO's for non-zero
1039 			 * op's.
1040 			 */
1041 			int adjval;
1042 
1043 			if ((sops[i].sem_flg & SEM_UNDO) == 0)
1044 				continue;
1045 			adjval = sops[i].sem_op;
1046 			if (adjval == 0)
1047 				continue;
1048 			error = semundo_adjust(td, &suptr, semid,
1049 			    sops[i].sem_num, -adjval);
1050 			if (error == 0)
1051 				continue;
1052 
1053 			/*
1054 			 * Oh-Oh!  We ran out of either sem_undo's or undo's.
1055 			 * Rollback the adjustments to this point and then
1056 			 * rollback the semaphore ups and down so we can return
1057 			 * with an error with all structures restored.  We
1058 			 * rollback the undo's in the exact reverse order that
1059 			 * we applied them.  This guarantees that we won't run
1060 			 * out of space as we roll things back out.
1061 			 */
1062 			for (j = i - 1; j >= 0; j--) {
1063 				if ((sops[j].sem_flg & SEM_UNDO) == 0)
1064 					continue;
1065 				adjval = sops[j].sem_op;
1066 				if (adjval == 0)
1067 					continue;
1068 				if (semundo_adjust(td, &suptr, semid,
1069 				    sops[j].sem_num, adjval) != 0)
1070 					panic("semop - can't undo undos");
1071 			}
1072 
1073 			for (j = 0; j < nsops; j++)
1074 				semaptr->sem_base[sops[j].sem_num].semval -=
1075 				    sops[j].sem_op;
1076 
1077 #ifdef SEM_DEBUG
1078 			printf("error = %d from semundo_adjust\n", error);
1079 #endif
1080 			goto done2;
1081 		} /* loop through the sops */
1082 	} /* if (do_undos) */
1083 
1084 	/* We're definitely done - set the sempid's and time */
1085 	for (i = 0; i < nsops; i++) {
1086 		sopptr = &sops[i];
1087 		semptr = &semaptr->sem_base[sopptr->sem_num];
1088 		semptr->sempid = td->td_proc->p_pid;
1089 	}
1090 	semaptr->sem_otime = time_second;
1091 
1092 	/*
1093 	 * Do a wakeup if any semaphore was up'd whilst something was
1094 	 * sleeping on it.
1095 	 */
1096 	if (do_wakeup) {
1097 #ifdef SEM_DEBUG
1098 		printf("semop:  doing wakeup\n");
1099 #endif
1100 		wakeup((caddr_t)semaptr);
1101 #ifdef SEM_DEBUG
1102 		printf("semop:  back from wakeup\n");
1103 #endif
1104 	}
1105 #ifdef SEM_DEBUG
1106 	printf("semop:  done\n");
1107 #endif
1108 	td->td_retval[0] = 0;
1109 done2:
1110 	if (sops)
1111 	    free(sops, M_SEM);
1112 	mtx_unlock(&Giant);
1113 	return (error);
1114 }
1115 
1116 /*
1117  * Go through the undo structures for this process and apply the adjustments to
1118  * semaphores.
1119  */
1120 static void
1121 semexit_myhook(p)
1122 	struct proc *p;
1123 {
1124 	register struct sem_undo *suptr;
1125 	register struct sem_undo **supptr;
1126 
1127 	/*
1128 	 * Go through the chain of undo vectors looking for one
1129 	 * associated with this process.
1130 	 */
1131 
1132 	for (supptr = &semu_list; (suptr = *supptr) != NULL;
1133 	    supptr = &suptr->un_next) {
1134 		if (suptr->un_proc == p)
1135 			break;
1136 	}
1137 
1138 	if (suptr == NULL)
1139 		return;
1140 
1141 #ifdef SEM_DEBUG
1142 	printf("proc @%08x has undo structure with %d entries\n", p,
1143 	    suptr->un_cnt);
1144 #endif
1145 
1146 	/*
1147 	 * If there are any active undo elements then process them.
1148 	 */
1149 	if (suptr->un_cnt > 0) {
1150 		int ix;
1151 
1152 		for (ix = 0; ix < suptr->un_cnt; ix++) {
1153 			int semid = suptr->un_ent[ix].un_id;
1154 			int semnum = suptr->un_ent[ix].un_num;
1155 			int adjval = suptr->un_ent[ix].un_adjval;
1156 			struct semid_ds *semaptr;
1157 
1158 			semaptr = &sema[semid];
1159 			if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0)
1160 				panic("semexit - semid not allocated");
1161 			if (semnum >= semaptr->sem_nsems)
1162 				panic("semexit - semnum out of range");
1163 
1164 #ifdef SEM_DEBUG
1165 			printf("semexit:  %08x id=%d num=%d(adj=%d) ; sem=%d\n",
1166 			    suptr->un_proc, suptr->un_ent[ix].un_id,
1167 			    suptr->un_ent[ix].un_num,
1168 			    suptr->un_ent[ix].un_adjval,
1169 			    semaptr->sem_base[semnum].semval);
1170 #endif
1171 
1172 			if (adjval < 0) {
1173 				if (semaptr->sem_base[semnum].semval < -adjval)
1174 					semaptr->sem_base[semnum].semval = 0;
1175 				else
1176 					semaptr->sem_base[semnum].semval +=
1177 					    adjval;
1178 			} else
1179 				semaptr->sem_base[semnum].semval += adjval;
1180 
1181 			wakeup((caddr_t)semaptr);
1182 #ifdef SEM_DEBUG
1183 			printf("semexit:  back from wakeup\n");
1184 #endif
1185 		}
1186 	}
1187 
1188 	/*
1189 	 * Deallocate the undo vector.
1190 	 */
1191 #ifdef SEM_DEBUG
1192 	printf("removing vector\n");
1193 #endif
1194 	suptr->un_proc = NULL;
1195 	*supptr = suptr->un_next;
1196 }
1197 
1198 static int
1199 sysctl_sema(SYSCTL_HANDLER_ARGS)
1200 {
1201 
1202 	return (SYSCTL_OUT(req, sema,
1203 	    sizeof(struct semid_ds) * seminfo.semmni));
1204 }
1205