xref: /freebsd/sys/kern/sysv_sem.c (revision 2357939bc239bd5334a169b62313806178dd8f30)
1 /*
2  * Implementation of SVID semaphores
3  *
4  * Author:  Daniel Boulet
5  *
6  * This software is provided ``AS IS'' without any warranties of any kind.
7  */
8 
9 #include <sys/cdefs.h>
10 __FBSDID("$FreeBSD$");
11 
12 #include "opt_sysvipc.h"
13 
14 #include <sys/param.h>
15 #include <sys/systm.h>
16 #include <sys/sysproto.h>
17 #include <sys/eventhandler.h>
18 #include <sys/kernel.h>
19 #include <sys/proc.h>
20 #include <sys/lock.h>
21 #include <sys/mutex.h>
22 #include <sys/sem.h>
23 #include <sys/syscall.h>
24 #include <sys/sysent.h>
25 #include <sys/sysctl.h>
26 #include <sys/malloc.h>
27 #include <sys/jail.h>
28 
29 static MALLOC_DEFINE(M_SEM, "sem", "SVID compatible semaphores");
30 
31 #ifdef SEM_DEBUG
32 #define DPRINTF(a)	printf a
33 #else
34 #define DPRINTF(a)
35 #endif
36 
37 static void seminit(void);
38 static int sysvsem_modload(struct module *, int, void *);
39 static int semunload(void);
40 static void semexit_myhook(void *arg, struct proc *p);
41 static int sysctl_sema(SYSCTL_HANDLER_ARGS);
42 static int semvalid(int semid, struct semid_ds *semaptr);
43 
44 #ifndef _SYS_SYSPROTO_H_
45 struct __semctl_args;
46 int __semctl(struct thread *td, struct __semctl_args *uap);
47 struct semget_args;
48 int semget(struct thread *td, struct semget_args *uap);
49 struct semop_args;
50 int semop(struct thread *td, struct semop_args *uap);
51 #endif
52 
53 static struct sem_undo *semu_alloc(struct thread *td);
54 static int semundo_adjust(struct thread *td, struct sem_undo **supptr,
55 		int semid, int semnum, int adjval);
56 static void semundo_clear(int semid, int semnum);
57 
58 /* XXX casting to (sy_call_t *) is bogus, as usual. */
59 static sy_call_t *semcalls[] = {
60 	(sy_call_t *)__semctl, (sy_call_t *)semget,
61 	(sy_call_t *)semop
62 };
63 
64 static struct mtx	sem_mtx;	/* semaphore global lock */
65 static int	semtot = 0;
66 static struct semid_ds *sema;	/* semaphore id pool */
67 static struct mtx *sema_mtx;	/* semaphore id pool mutexes*/
68 static struct sem *sem;		/* semaphore pool */
69 SLIST_HEAD(, sem_undo) semu_list;	/* list of active undo structures */
70 static int	*semu;		/* undo structure pool */
71 static eventhandler_tag semexit_tag;
72 
73 #define SEMUNDO_MTX		sem_mtx
74 #define SEMUNDO_LOCK()		mtx_lock(&SEMUNDO_MTX);
75 #define SEMUNDO_UNLOCK()	mtx_unlock(&SEMUNDO_MTX);
76 #define SEMUNDO_LOCKASSERT(how)	mtx_assert(&SEMUNDO_MTX, (how));
77 
78 struct sem {
79 	u_short	semval;		/* semaphore value */
80 	pid_t	sempid;		/* pid of last operation */
81 	u_short	semncnt;	/* # awaiting semval > cval */
82 	u_short	semzcnt;	/* # awaiting semval = 0 */
83 };
84 
85 /*
86  * Undo structure (one per process)
87  */
88 struct sem_undo {
89 	SLIST_ENTRY(sem_undo) un_next;	/* ptr to next active undo structure */
90 	struct	proc *un_proc;		/* owner of this structure */
91 	short	un_cnt;			/* # of active entries */
92 	struct undo {
93 		short	un_adjval;	/* adjust on exit values */
94 		short	un_num;		/* semaphore # */
95 		int	un_id;		/* semid */
96 	} un_ent[1];			/* undo entries */
97 };
98 
99 /*
100  * Configuration parameters
101  */
102 #ifndef SEMMNI
103 #define SEMMNI	10		/* # of semaphore identifiers */
104 #endif
105 #ifndef SEMMNS
106 #define SEMMNS	60		/* # of semaphores in system */
107 #endif
108 #ifndef SEMUME
109 #define SEMUME	10		/* max # of undo entries per process */
110 #endif
111 #ifndef SEMMNU
112 #define SEMMNU	30		/* # of undo structures in system */
113 #endif
114 
115 /* shouldn't need tuning */
116 #ifndef SEMMAP
117 #define SEMMAP	30		/* # of entries in semaphore map */
118 #endif
119 #ifndef SEMMSL
120 #define SEMMSL	SEMMNS		/* max # of semaphores per id */
121 #endif
122 #ifndef SEMOPM
123 #define SEMOPM	100		/* max # of operations per semop call */
124 #endif
125 
126 #define SEMVMX	32767		/* semaphore maximum value */
127 #define SEMAEM	16384		/* adjust on exit max value */
128 
129 /*
130  * Due to the way semaphore memory is allocated, we have to ensure that
131  * SEMUSZ is properly aligned.
132  */
133 
134 #define SEM_ALIGN(bytes) (((bytes) + (sizeof(long) - 1)) & ~(sizeof(long) - 1))
135 
136 /* actual size of an undo structure */
137 #define SEMUSZ	SEM_ALIGN(offsetof(struct sem_undo, un_ent[SEMUME]))
138 
139 /*
140  * Macro to find a particular sem_undo vector
141  */
142 #define SEMU(ix) \
143 	((struct sem_undo *)(((intptr_t)semu)+ix * seminfo.semusz))
144 
145 /*
146  * semaphore info struct
147  */
148 struct seminfo seminfo = {
149                 SEMMAP,         /* # of entries in semaphore map */
150                 SEMMNI,         /* # of semaphore identifiers */
151                 SEMMNS,         /* # of semaphores in system */
152                 SEMMNU,         /* # of undo structures in system */
153                 SEMMSL,         /* max # of semaphores per id */
154                 SEMOPM,         /* max # of operations per semop call */
155                 SEMUME,         /* max # of undo entries per process */
156                 SEMUSZ,         /* size in bytes of undo structure */
157                 SEMVMX,         /* semaphore maximum value */
158                 SEMAEM          /* adjust on exit max value */
159 };
160 
161 SYSCTL_DECL(_kern_ipc);
162 SYSCTL_INT(_kern_ipc, OID_AUTO, semmap, CTLFLAG_RW, &seminfo.semmap, 0, "");
163 SYSCTL_INT(_kern_ipc, OID_AUTO, semmni, CTLFLAG_RDTUN, &seminfo.semmni, 0, "");
164 SYSCTL_INT(_kern_ipc, OID_AUTO, semmns, CTLFLAG_RDTUN, &seminfo.semmns, 0, "");
165 SYSCTL_INT(_kern_ipc, OID_AUTO, semmnu, CTLFLAG_RDTUN, &seminfo.semmnu, 0, "");
166 SYSCTL_INT(_kern_ipc, OID_AUTO, semmsl, CTLFLAG_RW, &seminfo.semmsl, 0, "");
167 SYSCTL_INT(_kern_ipc, OID_AUTO, semopm, CTLFLAG_RDTUN, &seminfo.semopm, 0, "");
168 SYSCTL_INT(_kern_ipc, OID_AUTO, semume, CTLFLAG_RDTUN, &seminfo.semume, 0, "");
169 SYSCTL_INT(_kern_ipc, OID_AUTO, semusz, CTLFLAG_RDTUN, &seminfo.semusz, 0, "");
170 SYSCTL_INT(_kern_ipc, OID_AUTO, semvmx, CTLFLAG_RW, &seminfo.semvmx, 0, "");
171 SYSCTL_INT(_kern_ipc, OID_AUTO, semaem, CTLFLAG_RW, &seminfo.semaem, 0, "");
172 SYSCTL_PROC(_kern_ipc, OID_AUTO, sema, CTLFLAG_RD,
173     NULL, 0, sysctl_sema, "", "");
174 
175 static void
176 seminit(void)
177 {
178 	int i;
179 
180 	TUNABLE_INT_FETCH("kern.ipc.semmap", &seminfo.semmap);
181 	TUNABLE_INT_FETCH("kern.ipc.semmni", &seminfo.semmni);
182 	TUNABLE_INT_FETCH("kern.ipc.semmns", &seminfo.semmns);
183 	TUNABLE_INT_FETCH("kern.ipc.semmnu", &seminfo.semmnu);
184 	TUNABLE_INT_FETCH("kern.ipc.semmsl", &seminfo.semmsl);
185 	TUNABLE_INT_FETCH("kern.ipc.semopm", &seminfo.semopm);
186 	TUNABLE_INT_FETCH("kern.ipc.semume", &seminfo.semume);
187 	TUNABLE_INT_FETCH("kern.ipc.semusz", &seminfo.semusz);
188 	TUNABLE_INT_FETCH("kern.ipc.semvmx", &seminfo.semvmx);
189 	TUNABLE_INT_FETCH("kern.ipc.semaem", &seminfo.semaem);
190 
191 	sem = malloc(sizeof(struct sem) * seminfo.semmns, M_SEM, M_WAITOK);
192 	sema = malloc(sizeof(struct semid_ds) * seminfo.semmni, M_SEM,
193 	    M_WAITOK);
194 	sema_mtx = malloc(sizeof(struct mtx) * seminfo.semmni, M_SEM,
195 	    M_WAITOK | M_ZERO);
196 	semu = malloc(seminfo.semmnu * seminfo.semusz, M_SEM, M_WAITOK);
197 
198 	for (i = 0; i < seminfo.semmni; i++) {
199 		sema[i].sem_base = 0;
200 		sema[i].sem_perm.mode = 0;
201 		sema[i].sem_perm.seq = 0;
202 	}
203 	for (i = 0; i < seminfo.semmni; i++)
204 		mtx_init(&sema_mtx[i], "semid", NULL, MTX_DEF);
205 	for (i = 0; i < seminfo.semmnu; i++) {
206 		struct sem_undo *suptr = SEMU(i);
207 		suptr->un_proc = NULL;
208 	}
209 	SLIST_INIT(&semu_list);
210 	mtx_init(&sem_mtx, "sem", NULL, MTX_DEF);
211 	semexit_tag = EVENTHANDLER_REGISTER(process_exit, semexit_myhook, NULL,
212 	    EVENTHANDLER_PRI_ANY);
213 }
214 
215 static int
216 semunload(void)
217 {
218 	int i;
219 
220 	if (semtot != 0)
221 		return (EBUSY);
222 
223 	EVENTHANDLER_DEREGISTER(process_exit, semexit_tag);
224 	free(sem, M_SEM);
225 	free(sema, M_SEM);
226 	free(semu, M_SEM);
227 	for (i = 0; i < seminfo.semmni; i++)
228 		mtx_destroy(&sema_mtx[i]);
229 	mtx_destroy(&sem_mtx);
230 	return (0);
231 }
232 
233 static int
234 sysvsem_modload(struct module *module, int cmd, void *arg)
235 {
236 	int error = 0;
237 
238 	switch (cmd) {
239 	case MOD_LOAD:
240 		seminit();
241 		break;
242 	case MOD_UNLOAD:
243 		error = semunload();
244 		break;
245 	case MOD_SHUTDOWN:
246 		break;
247 	default:
248 		error = EINVAL;
249 		break;
250 	}
251 	return (error);
252 }
253 
254 static moduledata_t sysvsem_mod = {
255 	"sysvsem",
256 	&sysvsem_modload,
257 	NULL
258 };
259 
260 SYSCALL_MODULE_HELPER(semsys);
261 SYSCALL_MODULE_HELPER(__semctl);
262 SYSCALL_MODULE_HELPER(semget);
263 SYSCALL_MODULE_HELPER(semop);
264 
265 DECLARE_MODULE(sysvsem, sysvsem_mod,
266 	SI_SUB_SYSV_SEM, SI_ORDER_FIRST);
267 MODULE_VERSION(sysvsem, 1);
268 
269 /*
270  * Entry point for all SEM calls
271  *
272  * MPSAFE
273  */
274 int
275 semsys(td, uap)
276 	struct thread *td;
277 	/* XXX actually varargs. */
278 	struct semsys_args /* {
279 		int	which;
280 		int	a2;
281 		int	a3;
282 		int	a4;
283 		int	a5;
284 	} */ *uap;
285 {
286 	int error;
287 
288 	if (!jail_sysvipc_allowed && jailed(td->td_ucred))
289 		return (ENOSYS);
290 	if (uap->which < 0 ||
291 	    uap->which >= sizeof(semcalls)/sizeof(semcalls[0]))
292 		return (EINVAL);
293 	error = (*semcalls[uap->which])(td, &uap->a2);
294 	return (error);
295 }
296 
297 /*
298  * Allocate a new sem_undo structure for a process
299  * (returns ptr to structure or NULL if no more room)
300  */
301 
302 static struct sem_undo *
303 semu_alloc(td)
304 	struct thread *td;
305 {
306 	int i;
307 	struct sem_undo *suptr;
308 	struct sem_undo **supptr;
309 	int attempt;
310 
311 	SEMUNDO_LOCKASSERT(MA_OWNED);
312 	/*
313 	 * Try twice to allocate something.
314 	 * (we'll purge an empty structure after the first pass so
315 	 * two passes are always enough)
316 	 */
317 
318 	for (attempt = 0; attempt < 2; attempt++) {
319 		/*
320 		 * Look for a free structure.
321 		 * Fill it in and return it if we find one.
322 		 */
323 
324 		for (i = 0; i < seminfo.semmnu; i++) {
325 			suptr = SEMU(i);
326 			if (suptr->un_proc == NULL) {
327 				SLIST_INSERT_HEAD(&semu_list, suptr, un_next);
328 				suptr->un_cnt = 0;
329 				suptr->un_proc = td->td_proc;
330 				return(suptr);
331 			}
332 		}
333 
334 		/*
335 		 * We didn't find a free one, if this is the first attempt
336 		 * then try to free a structure.
337 		 */
338 
339 		if (attempt == 0) {
340 			/* All the structures are in use - try to free one */
341 			int did_something = 0;
342 
343 			SLIST_FOREACH_PREVPTR(suptr, supptr, &semu_list,
344 			    un_next) {
345 				if (suptr->un_cnt == 0) {
346 					suptr->un_proc = NULL;
347 					did_something = 1;
348 					*supptr = SLIST_NEXT(suptr, un_next);
349 					break;
350 				}
351 			}
352 
353 			/* If we didn't free anything then just give-up */
354 			if (!did_something)
355 				return(NULL);
356 		} else {
357 			/*
358 			 * The second pass failed even though we freed
359 			 * something after the first pass!
360 			 * This is IMPOSSIBLE!
361 			 */
362 			panic("semu_alloc - second attempt failed");
363 		}
364 	}
365 	return (NULL);
366 }
367 
368 /*
369  * Adjust a particular entry for a particular proc
370  */
371 
372 static int
373 semundo_adjust(td, supptr, semid, semnum, adjval)
374 	struct thread *td;
375 	struct sem_undo **supptr;
376 	int semid, semnum;
377 	int adjval;
378 {
379 	struct proc *p = td->td_proc;
380 	struct sem_undo *suptr;
381 	struct undo *sunptr;
382 	int i;
383 
384 	SEMUNDO_LOCKASSERT(MA_OWNED);
385 	/* Look for and remember the sem_undo if the caller doesn't provide
386 	   it */
387 
388 	suptr = *supptr;
389 	if (suptr == NULL) {
390 		SLIST_FOREACH(suptr, &semu_list, un_next) {
391 			if (suptr->un_proc == p) {
392 				*supptr = suptr;
393 				break;
394 			}
395 		}
396 		if (suptr == NULL) {
397 			if (adjval == 0)
398 				return(0);
399 			suptr = semu_alloc(td);
400 			if (suptr == NULL)
401 				return(ENOSPC);
402 			*supptr = suptr;
403 		}
404 	}
405 
406 	/*
407 	 * Look for the requested entry and adjust it (delete if adjval becomes
408 	 * 0).
409 	 */
410 	sunptr = &suptr->un_ent[0];
411 	for (i = 0; i < suptr->un_cnt; i++, sunptr++) {
412 		if (sunptr->un_id != semid || sunptr->un_num != semnum)
413 			continue;
414 		if (adjval != 0) {
415 			adjval += sunptr->un_adjval;
416 			if (adjval > seminfo.semaem || adjval < -seminfo.semaem)
417 				return (ERANGE);
418 		}
419 		sunptr->un_adjval = adjval;
420 		if (sunptr->un_adjval == 0) {
421 			suptr->un_cnt--;
422 			if (i < suptr->un_cnt)
423 				suptr->un_ent[i] =
424 				    suptr->un_ent[suptr->un_cnt];
425 		}
426 		return(0);
427 	}
428 
429 	/* Didn't find the right entry - create it */
430 	if (adjval == 0)
431 		return(0);
432 	if (adjval > seminfo.semaem || adjval < -seminfo.semaem)
433 		return (ERANGE);
434 	if (suptr->un_cnt != seminfo.semume) {
435 		sunptr = &suptr->un_ent[suptr->un_cnt];
436 		suptr->un_cnt++;
437 		sunptr->un_adjval = adjval;
438 		sunptr->un_id = semid; sunptr->un_num = semnum;
439 	} else
440 		return(EINVAL);
441 	return(0);
442 }
443 
444 static void
445 semundo_clear(semid, semnum)
446 	int semid, semnum;
447 {
448 	struct sem_undo *suptr;
449 
450 	SEMUNDO_LOCKASSERT(MA_OWNED);
451 	SLIST_FOREACH(suptr, &semu_list, un_next) {
452 		struct undo *sunptr = &suptr->un_ent[0];
453 		int i = 0;
454 
455 		while (i < suptr->un_cnt) {
456 			if (sunptr->un_id == semid) {
457 				if (semnum == -1 || sunptr->un_num == semnum) {
458 					suptr->un_cnt--;
459 					if (i < suptr->un_cnt) {
460 						suptr->un_ent[i] =
461 						  suptr->un_ent[suptr->un_cnt];
462 						continue;
463 					}
464 				}
465 				if (semnum != -1)
466 					break;
467 			}
468 			i++, sunptr++;
469 		}
470 	}
471 }
472 
473 static int
474 semvalid(semid, semaptr)
475 	int semid;
476 	struct semid_ds *semaptr;
477 {
478 
479 	return ((semaptr->sem_perm.mode & SEM_ALLOC) == 0 ||
480 	    semaptr->sem_perm.seq != IPCID_TO_SEQ(semid) ? EINVAL : 0);
481 }
482 
483 /*
484  * Note that the user-mode half of this passes a union, not a pointer
485  */
486 #ifndef _SYS_SYSPROTO_H_
487 struct __semctl_args {
488 	int	semid;
489 	int	semnum;
490 	int	cmd;
491 	union	semun *arg;
492 };
493 #endif
494 
495 /*
496  * MPSAFE
497  */
498 int
499 __semctl(td, uap)
500 	struct thread *td;
501 	struct __semctl_args *uap;
502 {
503 	int semid = uap->semid;
504 	int semnum = uap->semnum;
505 	int cmd = uap->cmd;
506 	u_short *array;
507 	union semun *arg = uap->arg;
508 	union semun real_arg;
509 	struct ucred *cred = td->td_ucred;
510 	int i, rval, error;
511 	struct semid_ds sbuf;
512 	struct semid_ds *semaptr;
513 	struct mtx *sema_mtxp;
514 	u_short usval, count;
515 
516 	DPRINTF(("call to semctl(%d, %d, %d, 0x%x)\n",
517 	    semid, semnum, cmd, arg));
518 	if (!jail_sysvipc_allowed && jailed(td->td_ucred))
519 		return (ENOSYS);
520 
521 	array = NULL;
522 
523 	switch(cmd) {
524 	case SEM_STAT:
525 		if (semid < 0 || semid >= seminfo.semmni)
526 			return (EINVAL);
527 		if ((error = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
528 			return (error);
529 		semaptr = &sema[semid];
530 		sema_mtxp = &sema_mtx[semid];
531 		mtx_lock(sema_mtxp);
532 		if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0) {
533 			error = EINVAL;
534 			goto done2;
535 		}
536 		if ((error = ipcperm(td, &semaptr->sem_perm, IPC_R)))
537 			goto done2;
538 		mtx_unlock(sema_mtxp);
539 		error = copyout(semaptr, real_arg.buf, sizeof(struct semid_ds));
540 		rval = IXSEQ_TO_IPCID(semid,semaptr->sem_perm);
541 		if (error == 0)
542 			td->td_retval[0] = rval;
543 		return (error);
544 	}
545 
546 	semid = IPCID_TO_IX(semid);
547 	if (semid < 0 || semid >= seminfo.semmni)
548 		return (EINVAL);
549 
550 	semaptr = &sema[semid];
551 	sema_mtxp = &sema_mtx[semid];
552 
553 	error = 0;
554 	rval = 0;
555 
556 	switch (cmd) {
557 	case IPC_RMID:
558 		mtx_lock(sema_mtxp);
559 		if ((error = semvalid(uap->semid, semaptr)) != 0)
560 			goto done2;
561 		if ((error = ipcperm(td, &semaptr->sem_perm, IPC_M)))
562 			goto done2;
563 		semaptr->sem_perm.cuid = cred->cr_uid;
564 		semaptr->sem_perm.uid = cred->cr_uid;
565 		semtot -= semaptr->sem_nsems;
566 		for (i = semaptr->sem_base - sem; i < semtot; i++)
567 			sem[i] = sem[i + semaptr->sem_nsems];
568 		for (i = 0; i < seminfo.semmni; i++) {
569 			if ((sema[i].sem_perm.mode & SEM_ALLOC) &&
570 			    sema[i].sem_base > semaptr->sem_base)
571 				sema[i].sem_base -= semaptr->sem_nsems;
572 		}
573 		semaptr->sem_perm.mode = 0;
574 		SEMUNDO_LOCK();
575 		semundo_clear(semid, -1);
576 		SEMUNDO_UNLOCK();
577 		wakeup(semaptr);
578 		break;
579 
580 	case IPC_SET:
581 		if ((error = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
582 			goto done2;
583 		if ((error = copyin(real_arg.buf, &sbuf, sizeof(sbuf))) != 0)
584 			goto done2;
585 		mtx_lock(sema_mtxp);
586 		if ((error = semvalid(uap->semid, semaptr)) != 0)
587 			goto done2;
588 		if ((error = ipcperm(td, &semaptr->sem_perm, IPC_M)))
589 			goto done2;
590 		semaptr->sem_perm.uid = sbuf.sem_perm.uid;
591 		semaptr->sem_perm.gid = sbuf.sem_perm.gid;
592 		semaptr->sem_perm.mode = (semaptr->sem_perm.mode & ~0777) |
593 		    (sbuf.sem_perm.mode & 0777);
594 		semaptr->sem_ctime = time_second;
595 		break;
596 
597 	case IPC_STAT:
598 		if ((error = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
599 			goto done2;
600 		mtx_lock(sema_mtxp);
601 		if ((error = semvalid(uap->semid, semaptr)) != 0)
602 			goto done2;
603 		if ((error = ipcperm(td, &semaptr->sem_perm, IPC_R)))
604 			goto done2;
605 		sbuf = *semaptr;
606 		mtx_unlock(sema_mtxp);
607 		error = copyout(semaptr, real_arg.buf,
608 				sizeof(struct semid_ds));
609 		break;
610 
611 	case GETNCNT:
612 		mtx_lock(sema_mtxp);
613 		if ((error = semvalid(uap->semid, semaptr)) != 0)
614 			goto done2;
615 		if ((error = ipcperm(td, &semaptr->sem_perm, IPC_R)))
616 			goto done2;
617 		if (semnum < 0 || semnum >= semaptr->sem_nsems) {
618 			error = EINVAL;
619 			goto done2;
620 		}
621 		rval = semaptr->sem_base[semnum].semncnt;
622 		break;
623 
624 	case GETPID:
625 		mtx_lock(sema_mtxp);
626 		if ((error = semvalid(uap->semid, semaptr)) != 0)
627 			goto done2;
628 		if ((error = ipcperm(td, &semaptr->sem_perm, IPC_R)))
629 			goto done2;
630 		if (semnum < 0 || semnum >= semaptr->sem_nsems) {
631 			error = EINVAL;
632 			goto done2;
633 		}
634 		rval = semaptr->sem_base[semnum].sempid;
635 		break;
636 
637 	case GETVAL:
638 		mtx_lock(sema_mtxp);
639 		if ((error = semvalid(uap->semid, semaptr)) != 0)
640 			goto done2;
641 		if ((error = ipcperm(td, &semaptr->sem_perm, IPC_R)))
642 			goto done2;
643 		if (semnum < 0 || semnum >= semaptr->sem_nsems) {
644 			error = EINVAL;
645 			goto done2;
646 		}
647 		rval = semaptr->sem_base[semnum].semval;
648 		break;
649 
650 	case GETALL:
651 		if ((error = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
652 			goto done2;
653 		array = malloc(sizeof(*array) * semaptr->sem_nsems, M_TEMP,
654 		    M_WAITOK);
655 		mtx_lock(sema_mtxp);
656 		if ((error = semvalid(uap->semid, semaptr)) != 0)
657 			goto done2;
658 		if ((error = ipcperm(td, &semaptr->sem_perm, IPC_R)))
659 			goto done2;
660 		for (i = 0; i < semaptr->sem_nsems; i++)
661 			array[i] = semaptr->sem_base[i].semval;
662 		mtx_unlock(sema_mtxp);
663 		error = copyout(array, real_arg.array,
664 		    i * sizeof(real_arg.array[0]));
665 		break;
666 
667 	case GETZCNT:
668 		mtx_lock(sema_mtxp);
669 		if ((error = semvalid(uap->semid, semaptr)) != 0)
670 			goto done2;
671 		if ((error = ipcperm(td, &semaptr->sem_perm, IPC_R)))
672 			goto done2;
673 		if (semnum < 0 || semnum >= semaptr->sem_nsems) {
674 			error = EINVAL;
675 			goto done2;
676 		}
677 		rval = semaptr->sem_base[semnum].semzcnt;
678 		break;
679 
680 	case SETVAL:
681 		if ((error = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
682 			goto done2;
683 		mtx_lock(sema_mtxp);
684 		if ((error = semvalid(uap->semid, semaptr)) != 0)
685 			goto done2;
686 		if ((error = ipcperm(td, &semaptr->sem_perm, IPC_W)))
687 			goto done2;
688 		if (semnum < 0 || semnum >= semaptr->sem_nsems) {
689 			error = EINVAL;
690 			goto done2;
691 		}
692 		if (real_arg.val < 0 || real_arg.val > seminfo.semvmx) {
693 			error = ERANGE;
694 			goto done2;
695 		}
696 		semaptr->sem_base[semnum].semval = real_arg.val;
697 		SEMUNDO_LOCK();
698 		semundo_clear(semid, semnum);
699 		SEMUNDO_UNLOCK();
700 		wakeup(semaptr);
701 		break;
702 
703 	case SETALL:
704 		mtx_lock(sema_mtxp);
705 raced:
706 		if ((error = semvalid(uap->semid, semaptr)) != 0)
707 			goto done2;
708 		count = semaptr->sem_nsems;
709 		mtx_unlock(sema_mtxp);
710 		if ((error = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
711 			goto done2;
712 		array = malloc(sizeof(*array) * count, M_TEMP, M_WAITOK);
713 		copyin(real_arg.array, array, count * sizeof(*array));
714 		if (error)
715 			break;
716 		mtx_lock(sema_mtxp);
717 		if ((error = semvalid(uap->semid, semaptr)) != 0)
718 			goto done2;
719 		/* we could have raced? */
720 		if (count != semaptr->sem_nsems) {
721 			free(array, M_TEMP);
722 			array = NULL;
723 			goto raced;
724 		}
725 		if ((error = ipcperm(td, &semaptr->sem_perm, IPC_W)))
726 			goto done2;
727 		for (i = 0; i < semaptr->sem_nsems; i++) {
728 			usval = array[i];
729 			if (usval > seminfo.semvmx) {
730 				error = ERANGE;
731 				break;
732 			}
733 			semaptr->sem_base[i].semval = usval;
734 		}
735 		SEMUNDO_LOCK();
736 		semundo_clear(semid, -1);
737 		SEMUNDO_UNLOCK();
738 		wakeup(semaptr);
739 		break;
740 
741 	default:
742 		error = EINVAL;
743 		break;
744 	}
745 
746 	if (error == 0)
747 		td->td_retval[0] = rval;
748 done2:
749 	if (mtx_owned(sema_mtxp))
750 		mtx_unlock(sema_mtxp);
751 	if (array != NULL)
752 		free(array, M_TEMP);
753 	return(error);
754 }
755 
756 #ifndef _SYS_SYSPROTO_H_
757 struct semget_args {
758 	key_t	key;
759 	int	nsems;
760 	int	semflg;
761 };
762 #endif
763 
764 /*
765  * MPSAFE
766  */
767 int
768 semget(td, uap)
769 	struct thread *td;
770 	struct semget_args *uap;
771 {
772 	int semid, error = 0;
773 	int key = uap->key;
774 	int nsems = uap->nsems;
775 	int semflg = uap->semflg;
776 	struct ucred *cred = td->td_ucred;
777 
778 	DPRINTF(("semget(0x%x, %d, 0%o)\n", key, nsems, semflg));
779 	if (!jail_sysvipc_allowed && jailed(td->td_ucred))
780 		return (ENOSYS);
781 
782 	mtx_lock(&Giant);
783 	if (key != IPC_PRIVATE) {
784 		for (semid = 0; semid < seminfo.semmni; semid++) {
785 			if ((sema[semid].sem_perm.mode & SEM_ALLOC) &&
786 			    sema[semid].sem_perm.key == key)
787 				break;
788 		}
789 		if (semid < seminfo.semmni) {
790 			DPRINTF(("found public key\n"));
791 			if ((error = ipcperm(td, &sema[semid].sem_perm,
792 			    semflg & 0700))) {
793 				goto done2;
794 			}
795 			if (nsems > 0 && sema[semid].sem_nsems < nsems) {
796 				DPRINTF(("too small\n"));
797 				error = EINVAL;
798 				goto done2;
799 			}
800 			if ((semflg & IPC_CREAT) && (semflg & IPC_EXCL)) {
801 				DPRINTF(("not exclusive\n"));
802 				error = EEXIST;
803 				goto done2;
804 			}
805 			goto found;
806 		}
807 	}
808 
809 	DPRINTF(("need to allocate the semid_ds\n"));
810 	if (key == IPC_PRIVATE || (semflg & IPC_CREAT)) {
811 		if (nsems <= 0 || nsems > seminfo.semmsl) {
812 			DPRINTF(("nsems out of range (0<%d<=%d)\n", nsems,
813 			    seminfo.semmsl));
814 			error = EINVAL;
815 			goto done2;
816 		}
817 		if (nsems > seminfo.semmns - semtot) {
818 			DPRINTF((
819 			    "not enough semaphores left (need %d, got %d)\n",
820 			    nsems, seminfo.semmns - semtot));
821 			error = ENOSPC;
822 			goto done2;
823 		}
824 		for (semid = 0; semid < seminfo.semmni; semid++) {
825 			if ((sema[semid].sem_perm.mode & SEM_ALLOC) == 0)
826 				break;
827 		}
828 		if (semid == seminfo.semmni) {
829 			DPRINTF(("no more semid_ds's available\n"));
830 			error = ENOSPC;
831 			goto done2;
832 		}
833 		DPRINTF(("semid %d is available\n", semid));
834 		sema[semid].sem_perm.key = key;
835 		sema[semid].sem_perm.cuid = cred->cr_uid;
836 		sema[semid].sem_perm.uid = cred->cr_uid;
837 		sema[semid].sem_perm.cgid = cred->cr_gid;
838 		sema[semid].sem_perm.gid = cred->cr_gid;
839 		sema[semid].sem_perm.mode = (semflg & 0777) | SEM_ALLOC;
840 		sema[semid].sem_perm.seq =
841 		    (sema[semid].sem_perm.seq + 1) & 0x7fff;
842 		sema[semid].sem_nsems = nsems;
843 		sema[semid].sem_otime = 0;
844 		sema[semid].sem_ctime = time_second;
845 		sema[semid].sem_base = &sem[semtot];
846 		semtot += nsems;
847 		bzero(sema[semid].sem_base,
848 		    sizeof(sema[semid].sem_base[0])*nsems);
849 		DPRINTF(("sembase = 0x%x, next = 0x%x\n", sema[semid].sem_base,
850 		    &sem[semtot]));
851 	} else {
852 		DPRINTF(("didn't find it and wasn't asked to create it\n"));
853 		error = ENOENT;
854 		goto done2;
855 	}
856 
857 found:
858 	td->td_retval[0] = IXSEQ_TO_IPCID(semid, sema[semid].sem_perm);
859 done2:
860 	mtx_unlock(&Giant);
861 	return (error);
862 }
863 
864 #ifndef _SYS_SYSPROTO_H_
865 struct semop_args {
866 	int	semid;
867 	struct	sembuf *sops;
868 	size_t	nsops;
869 };
870 #endif
871 
872 /*
873  * MPSAFE
874  */
875 int
876 semop(td, uap)
877 	struct thread *td;
878 	struct semop_args *uap;
879 {
880 #define SMALL_SOPS	8
881 	struct sembuf small_sops[SMALL_SOPS];
882 	int semid = uap->semid;
883 	size_t nsops = uap->nsops;
884 	struct sembuf *sops;
885 	struct semid_ds *semaptr;
886 	struct sembuf *sopptr = 0;
887 	struct sem *semptr = 0;
888 	struct sem_undo *suptr;
889 	struct mtx *sema_mtxp;
890 	size_t i, j, k;
891 	int error;
892 	int do_wakeup, do_undos;
893 
894 	DPRINTF(("call to semop(%d, 0x%x, %u)\n", semid, sops, nsops));
895 
896 	if (!jail_sysvipc_allowed && jailed(td->td_ucred))
897 		return (ENOSYS);
898 
899 	semid = IPCID_TO_IX(semid);	/* Convert back to zero origin */
900 
901 	if (semid < 0 || semid >= seminfo.semmni)
902 		return (EINVAL);
903 
904 	/* Allocate memory for sem_ops */
905 	if (nsops <= SMALL_SOPS)
906 		sops = small_sops;
907 	else if (nsops <= seminfo.semopm)
908 		sops = malloc(nsops * sizeof(*sops), M_TEMP, M_WAITOK);
909 	else {
910 		DPRINTF(("too many sops (max=%d, nsops=%d)\n", seminfo.semopm,
911 		    nsops));
912 		return (E2BIG);
913 	}
914 	if ((error = copyin(uap->sops, sops, nsops * sizeof(sops[0]))) != 0) {
915 		DPRINTF(("error = %d from copyin(%08x, %08x, %d)\n", error,
916 		    uap->sops, sops, nsops * sizeof(sops[0])));
917 		if (sops != small_sops)
918 			free(sops, M_SEM);
919 		return (error);
920 	}
921 
922 	semaptr = &sema[semid];
923 	sema_mtxp = &sema_mtx[semid];
924 	mtx_lock(sema_mtxp);
925 	if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0) {
926 		error = EINVAL;
927 		goto done2;
928 	}
929 	if (semaptr->sem_perm.seq != IPCID_TO_SEQ(uap->semid)) {
930 		error = EINVAL;
931 		goto done2;
932 	}
933 	/*
934 	 * Initial pass thru sops to see what permissions are needed.
935 	 * Also perform any checks that don't need repeating on each
936 	 * attempt to satisfy the request vector.
937 	 */
938 	j = 0;		/* permission needed */
939 	do_undos = 0;
940 	for (i = 0; i < nsops; i++) {
941 		sopptr = &sops[i];
942 		if (sopptr->sem_num >= semaptr->sem_nsems) {
943 			error = EFBIG;
944 			goto done2;
945 		}
946 		if (sopptr->sem_flg & SEM_UNDO && sopptr->sem_op != 0)
947 			do_undos = 1;
948 		j |= (sopptr->sem_op == 0) ? SEM_R : SEM_A;
949 	}
950 
951 	if ((error = ipcperm(td, &semaptr->sem_perm, j))) {
952 		DPRINTF(("error = %d from ipaccess\n", error));
953 		goto done2;
954 	}
955 
956 	/*
957 	 * Loop trying to satisfy the vector of requests.
958 	 * If we reach a point where we must wait, any requests already
959 	 * performed are rolled back and we go to sleep until some other
960 	 * process wakes us up.  At this point, we start all over again.
961 	 *
962 	 * This ensures that from the perspective of other tasks, a set
963 	 * of requests is atomic (never partially satisfied).
964 	 */
965 	for (;;) {
966 		do_wakeup = 0;
967 		error = 0;	/* error return if necessary */
968 
969 		for (i = 0; i < nsops; i++) {
970 			sopptr = &sops[i];
971 			semptr = &semaptr->sem_base[sopptr->sem_num];
972 
973 			DPRINTF((
974 			    "semop:  semaptr=%x, sem_base=%x, "
975 			    "semptr=%x, sem[%d]=%d : op=%d, flag=%s\n",
976 			    semaptr, semaptr->sem_base, semptr,
977 			    sopptr->sem_num, semptr->semval, sopptr->sem_op,
978 			    (sopptr->sem_flg & IPC_NOWAIT) ?
979 			    "nowait" : "wait"));
980 
981 			if (sopptr->sem_op < 0) {
982 				if (semptr->semval + sopptr->sem_op < 0) {
983 					DPRINTF(("semop:  can't do it now\n"));
984 					break;
985 				} else {
986 					semptr->semval += sopptr->sem_op;
987 					if (semptr->semval == 0 &&
988 					    semptr->semzcnt > 0)
989 						do_wakeup = 1;
990 				}
991 			} else if (sopptr->sem_op == 0) {
992 				if (semptr->semval != 0) {
993 					DPRINTF(("semop:  not zero now\n"));
994 					break;
995 				}
996 			} else if (semptr->semval + sopptr->sem_op >
997 			    seminfo.semvmx) {
998 				error = ERANGE;
999 				break;
1000 			} else {
1001 				if (semptr->semncnt > 0)
1002 					do_wakeup = 1;
1003 				semptr->semval += sopptr->sem_op;
1004 			}
1005 		}
1006 
1007 		/*
1008 		 * Did we get through the entire vector?
1009 		 */
1010 		if (i >= nsops)
1011 			goto done;
1012 
1013 		/*
1014 		 * No ... rollback anything that we've already done
1015 		 */
1016 		DPRINTF(("semop:  rollback 0 through %d\n", i-1));
1017 		for (j = 0; j < i; j++)
1018 			semaptr->sem_base[sops[j].sem_num].semval -=
1019 			    sops[j].sem_op;
1020 
1021 		/* If we detected an error, return it */
1022 		if (error != 0)
1023 			goto done2;
1024 
1025 		/*
1026 		 * If the request that we couldn't satisfy has the
1027 		 * NOWAIT flag set then return with EAGAIN.
1028 		 */
1029 		if (sopptr->sem_flg & IPC_NOWAIT) {
1030 			error = EAGAIN;
1031 			goto done2;
1032 		}
1033 
1034 		if (sopptr->sem_op == 0)
1035 			semptr->semzcnt++;
1036 		else
1037 			semptr->semncnt++;
1038 
1039 		DPRINTF(("semop:  good night!\n"));
1040 		error = msleep(semaptr, sema_mtxp, (PZERO - 4) | PCATCH,
1041 		    "semwait", 0);
1042 		DPRINTF(("semop:  good morning (error=%d)!\n", error));
1043 		/* return code is checked below, after sem[nz]cnt-- */
1044 
1045 		/*
1046 		 * Make sure that the semaphore still exists
1047 		 */
1048 		if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0 ||
1049 		    semaptr->sem_perm.seq != IPCID_TO_SEQ(uap->semid)) {
1050 			error = EIDRM;
1051 			goto done2;
1052 		}
1053 
1054 		/*
1055 		 * The semaphore is still alive.  Readjust the count of
1056 		 * waiting processes.
1057 		 */
1058 		if (sopptr->sem_op == 0)
1059 			semptr->semzcnt--;
1060 		else
1061 			semptr->semncnt--;
1062 
1063 		/*
1064 		 * Is it really morning, or was our sleep interrupted?
1065 		 * (Delayed check of msleep() return code because we
1066 		 * need to decrement sem[nz]cnt either way.)
1067 		 */
1068 		if (error != 0) {
1069 			error = EINTR;
1070 			goto done2;
1071 		}
1072 		DPRINTF(("semop:  good morning!\n"));
1073 	}
1074 
1075 done:
1076 	/*
1077 	 * Process any SEM_UNDO requests.
1078 	 */
1079 	if (do_undos) {
1080 		SEMUNDO_LOCK();
1081 		suptr = NULL;
1082 		for (i = 0; i < nsops; i++) {
1083 			/*
1084 			 * We only need to deal with SEM_UNDO's for non-zero
1085 			 * op's.
1086 			 */
1087 			int adjval;
1088 
1089 			if ((sops[i].sem_flg & SEM_UNDO) == 0)
1090 				continue;
1091 			adjval = sops[i].sem_op;
1092 			if (adjval == 0)
1093 				continue;
1094 			error = semundo_adjust(td, &suptr, semid,
1095 			    sops[i].sem_num, -adjval);
1096 			if (error == 0)
1097 				continue;
1098 
1099 			/*
1100 			 * Oh-Oh!  We ran out of either sem_undo's or undo's.
1101 			 * Rollback the adjustments to this point and then
1102 			 * rollback the semaphore ups and down so we can return
1103 			 * with an error with all structures restored.  We
1104 			 * rollback the undo's in the exact reverse order that
1105 			 * we applied them.  This guarantees that we won't run
1106 			 * out of space as we roll things back out.
1107 			 */
1108 			for (j = 0; j < i; j++) {
1109 				k = i - j - 1;
1110 				if ((sops[k].sem_flg & SEM_UNDO) == 0)
1111 					continue;
1112 				adjval = sops[k].sem_op;
1113 				if (adjval == 0)
1114 					continue;
1115 				if (semundo_adjust(td, &suptr, semid,
1116 				    sops[k].sem_num, adjval) != 0)
1117 					panic("semop - can't undo undos");
1118 			}
1119 
1120 			for (j = 0; j < nsops; j++)
1121 				semaptr->sem_base[sops[j].sem_num].semval -=
1122 				    sops[j].sem_op;
1123 
1124 			DPRINTF(("error = %d from semundo_adjust\n", error));
1125 			SEMUNDO_UNLOCK();
1126 			goto done2;
1127 		} /* loop through the sops */
1128 		SEMUNDO_UNLOCK();
1129 	} /* if (do_undos) */
1130 
1131 	/* We're definitely done - set the sempid's and time */
1132 	for (i = 0; i < nsops; i++) {
1133 		sopptr = &sops[i];
1134 		semptr = &semaptr->sem_base[sopptr->sem_num];
1135 		semptr->sempid = td->td_proc->p_pid;
1136 	}
1137 	semaptr->sem_otime = time_second;
1138 
1139 	/*
1140 	 * Do a wakeup if any semaphore was up'd whilst something was
1141 	 * sleeping on it.
1142 	 */
1143 	if (do_wakeup) {
1144 		DPRINTF(("semop:  doing wakeup\n"));
1145 		wakeup(semaptr);
1146 		DPRINTF(("semop:  back from wakeup\n"));
1147 	}
1148 	DPRINTF(("semop:  done\n"));
1149 	td->td_retval[0] = 0;
1150 done2:
1151 	mtx_unlock(sema_mtxp);
1152 	if (sops != small_sops)
1153 		free(sops, M_SEM);
1154 	return (error);
1155 }
1156 
1157 /*
1158  * Go through the undo structures for this process and apply the adjustments to
1159  * semaphores.
1160  */
1161 static void
1162 semexit_myhook(arg, p)
1163 	void *arg;
1164 	struct proc *p;
1165 {
1166 	struct sem_undo *suptr;
1167 	struct sem_undo **supptr;
1168 
1169 	/*
1170 	 * Go through the chain of undo vectors looking for one
1171 	 * associated with this process.
1172 	 */
1173 	SEMUNDO_LOCK();
1174 	SLIST_FOREACH_PREVPTR(suptr, supptr, &semu_list, un_next) {
1175 		if (suptr->un_proc == p)
1176 			break;
1177 	}
1178 	SEMUNDO_UNLOCK();
1179 
1180 	if (suptr == NULL)
1181 		return;
1182 
1183 	DPRINTF(("proc @%08x has undo structure with %d entries\n", p,
1184 	    suptr->un_cnt));
1185 
1186 	/*
1187 	 * If there are any active undo elements then process them.
1188 	 */
1189 	if (suptr->un_cnt > 0) {
1190 		int ix;
1191 
1192 		for (ix = 0; ix < suptr->un_cnt; ix++) {
1193 			int semid = suptr->un_ent[ix].un_id;
1194 			int semnum = suptr->un_ent[ix].un_num;
1195 			int adjval = suptr->un_ent[ix].un_adjval;
1196 			struct semid_ds *semaptr;
1197 			struct mtx *sema_mtxp;
1198 
1199 			semaptr = &sema[semid];
1200 			sema_mtxp = &sema_mtx[semid];
1201 			mtx_lock(sema_mtxp);
1202 			SEMUNDO_LOCK();
1203 			if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0)
1204 				panic("semexit - semid not allocated");
1205 			if (semnum >= semaptr->sem_nsems)
1206 				panic("semexit - semnum out of range");
1207 
1208 			DPRINTF((
1209 			    "semexit:  %08x id=%d num=%d(adj=%d) ; sem=%d\n",
1210 			    suptr->un_proc, suptr->un_ent[ix].un_id,
1211 			    suptr->un_ent[ix].un_num,
1212 			    suptr->un_ent[ix].un_adjval,
1213 			    semaptr->sem_base[semnum].semval));
1214 
1215 			if (adjval < 0) {
1216 				if (semaptr->sem_base[semnum].semval < -adjval)
1217 					semaptr->sem_base[semnum].semval = 0;
1218 				else
1219 					semaptr->sem_base[semnum].semval +=
1220 					    adjval;
1221 			} else
1222 				semaptr->sem_base[semnum].semval += adjval;
1223 
1224 			wakeup(semaptr);
1225 			DPRINTF(("semexit:  back from wakeup\n"));
1226 			mtx_unlock(sema_mtxp);
1227 			SEMUNDO_UNLOCK();
1228 		}
1229 	}
1230 
1231 	/*
1232 	 * Deallocate the undo vector.
1233 	 */
1234 	DPRINTF(("removing vector\n"));
1235 	suptr->un_proc = NULL;
1236 	*supptr = SLIST_NEXT(suptr, un_next);
1237 }
1238 
1239 static int
1240 sysctl_sema(SYSCTL_HANDLER_ARGS)
1241 {
1242 
1243 	return (SYSCTL_OUT(req, sema,
1244 	    sizeof(struct semid_ds) * seminfo.semmni));
1245 }
1246