xref: /dragonfly/sys/kern/sysv_sem.c (revision 2983445f) 
1 /* $FreeBSD: src/sys/kern/sysv_sem.c,v 1.69 2004/03/17 09:37:13 cperciva Exp $ */
2 /* $DragonFly: src/sys/kern/sysv_sem.c,v 1.19 2008/01/06 16:55:51 swildner Exp $ */
3 
4 /*
5  * Implementation of SVID semaphores
6  *
7  * Author:  Daniel Boulet
8  *
9  * This software is provided ``AS IS'' without any warranties of any kind.
10  */
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/kernel.h>
18 #include <sys/proc.h>
19 #include <sys/sem.h>
20 #include <sys/sysent.h>
21 #include <sys/sysctl.h>
22 #include <sys/malloc.h>
23 #include <sys/jail.h>
24 
25 #include <sys/mplock2.h>
26 
27 static MALLOC_DEFINE(M_SEM, "sem", "SVID compatible semaphores");
28 
29 static void seminit (void *);
30 
31 static struct sem_undo *semu_alloc (struct proc *p);
32 static int semundo_adjust (struct proc *p, struct sem_undo **supptr,
33 		int semid, int semnum, int adjval);
34 static void semundo_clear (int semid, int semnum);
35 
36 /* XXX casting to (sy_call_t *) is bogus, as usual. */
37 static sy_call_t *semcalls[] = {
38 	(sy_call_t *)sys___semctl, (sy_call_t *)sys_semget,
39 	(sy_call_t *)sys_semop
40 };
41 
42 static int	semtot = 0;
43 static struct semid_ds *sema;	/* semaphore id pool */
44 static struct sem *sem;		/* semaphore pool */
45 static struct sem_undo *semu_list; /* list of active undo structures */
46 static int	*semu;		/* undo structure pool */
47 
48 struct sem {
49 	u_short	semval;		/* semaphore value */
50 	pid_t	sempid;		/* pid of last operation */
51 	u_short	semncnt;	/* # awaiting semval > cval */
52 	u_short	semzcnt;	/* # awaiting semval = 0 */
53 };
54 
55 /*
56  * Undo structure (one per process)
57  */
58 struct sem_undo {
59 	struct	sem_undo *un_next;	/* ptr to next active undo structure */
60 	struct	proc *un_proc;		/* owner of this structure */
61 	short	un_cnt;			/* # of active entries */
62 	struct undo {
63 		short	un_adjval;	/* adjust on exit values */
64 		short	un_num;		/* semaphore # */
65 		int	un_id;		/* semid */
66 	} un_ent[1];			/* undo entries */
67 };
68 
69 /*
70  * Configuration parameters
71  */
72 #ifndef SEMMNI
73 #define SEMMNI	10		/* # of semaphore identifiers */
74 #endif
75 #ifndef SEMMNS
76 #define SEMMNS	60		/* # of semaphores in system */
77 #endif
78 #ifndef SEMUME
79 #define SEMUME	10		/* max # of undo entries per process */
80 #endif
81 #ifndef SEMMNU
82 #define SEMMNU	30		/* # of undo structures in system */
83 #endif
84 
85 /* shouldn't need tuning */
86 #ifndef SEMMAP
87 #define SEMMAP	30		/* # of entries in semaphore map */
88 #endif
89 #ifndef SEMMSL
90 #define SEMMSL	SEMMNS		/* max # of semaphores per id */
91 #endif
92 #ifndef SEMOPM
93 #define SEMOPM	100		/* max # of operations per semop call */
94 #endif
95 
96 #define SEMVMX	32767		/* semaphore maximum value */
97 #define SEMAEM	16384		/* adjust on exit max value */
98 
99 /*
100  * Due to the way semaphore memory is allocated, we have to ensure that
101  * SEMUSZ is properly aligned.
102  */
103 
104 #define SEM_ALIGN(bytes) (((bytes) + (sizeof(long) - 1)) & ~(sizeof(long) - 1))
105 
106 /* actual size of an undo structure */
107 #define SEMUSZ	SEM_ALIGN(offsetof(struct sem_undo, un_ent[SEMUME]))
108 
109 /*
110  * Macro to find a particular sem_undo vector
111  */
112 #define SEMU(ix)	((struct sem_undo *)(((intptr_t)semu)+ix * seminfo.semusz))
113 
114 /*
115  * semaphore info struct
116  */
117 struct seminfo seminfo = {
118                 SEMMAP,         /* # of entries in semaphore map */
119                 SEMMNI,         /* # of semaphore identifiers */
120                 SEMMNS,         /* # of semaphores in system */
121                 SEMMNU,         /* # of undo structures in system */
122                 SEMMSL,         /* max # of semaphores per id */
123                 SEMOPM,         /* max # of operations per semop call */
124                 SEMUME,         /* max # of undo entries per process */
125                 SEMUSZ,         /* size in bytes of undo structure */
126                 SEMVMX,         /* semaphore maximum value */
127                 SEMAEM          /* adjust on exit max value */
128 };
129 
130 TUNABLE_INT("kern.ipc.semmap", &seminfo.semmap);
131 TUNABLE_INT("kern.ipc.semmni", &seminfo.semmni);
132 TUNABLE_INT("kern.ipc.semmns", &seminfo.semmns);
133 TUNABLE_INT("kern.ipc.semmnu", &seminfo.semmnu);
134 TUNABLE_INT("kern.ipc.semmsl", &seminfo.semmsl);
135 TUNABLE_INT("kern.ipc.semopm", &seminfo.semopm);
136 TUNABLE_INT("kern.ipc.semume", &seminfo.semume);
137 TUNABLE_INT("kern.ipc.semusz", &seminfo.semusz);
138 TUNABLE_INT("kern.ipc.semvmx", &seminfo.semvmx);
139 TUNABLE_INT("kern.ipc.semaem", &seminfo.semaem);
140 
141 SYSCTL_INT(_kern_ipc, OID_AUTO, semmap, CTLFLAG_RW, &seminfo.semmap, 0,
142     "Number of entries in semaphore map");
143 SYSCTL_INT(_kern_ipc, OID_AUTO, semmni, CTLFLAG_RD, &seminfo.semmni, 0,
144     "Number of semaphore identifiers");
145 SYSCTL_INT(_kern_ipc, OID_AUTO, semmns, CTLFLAG_RD, &seminfo.semmns, 0,
146     "Total number of semaphores");
147 SYSCTL_INT(_kern_ipc, OID_AUTO, semmnu, CTLFLAG_RD, &seminfo.semmnu, 0,
148     "Total number of undo structures");
149 SYSCTL_INT(_kern_ipc, OID_AUTO, semmsl, CTLFLAG_RW, &seminfo.semmsl, 0,
150     "Max number of semaphores per id");
151 SYSCTL_INT(_kern_ipc, OID_AUTO, semopm, CTLFLAG_RD, &seminfo.semopm, 0,
152     "Max number of operations per semop call");
153 SYSCTL_INT(_kern_ipc, OID_AUTO, semume, CTLFLAG_RD, &seminfo.semume, 0,
154     "Max number of undo entries per process");
155 SYSCTL_INT(_kern_ipc, OID_AUTO, semusz, CTLFLAG_RD, &seminfo.semusz, 0,
156     "Size in bytes of undo structure");
157 SYSCTL_INT(_kern_ipc, OID_AUTO, semvmx, CTLFLAG_RW, &seminfo.semvmx, 0,
158     "Semaphore maximum value");
159 SYSCTL_INT(_kern_ipc, OID_AUTO, semaem, CTLFLAG_RW, &seminfo.semaem, 0,
160     "Adjust on exit max value");
161 
162 #if 0
163 RO seminfo.semmap	/* SEMMAP unused */
164 RO seminfo.semmni
165 RO seminfo.semmns
166 RO seminfo.semmnu	/* undo entries per system */
167 RW seminfo.semmsl
168 RO seminfo.semopm	/* SEMOPM unused */
169 RO seminfo.semume
170 RO seminfo.semusz	/* param - derived from SEMUME for per-proc sizeof */
171 RO seminfo.semvmx	/* SEMVMX unused - user param */
172 RO seminfo.semaem	/* SEMAEM unused - user param */
173 #endif
174 
175 static void
176 seminit(void *dummy)
177 {
178 	int i;
179 
180 	sem = kmalloc(sizeof(struct sem) * seminfo.semmns, M_SEM, M_WAITOK);
181 	sema = kmalloc(sizeof(struct semid_ds) * seminfo.semmni, M_SEM, M_WAITOK);
182 	semu = kmalloc(seminfo.semmnu * seminfo.semusz, M_SEM, M_WAITOK);
183 
184 	for (i = 0; i < seminfo.semmni; i++) {
185 		sema[i].sem_base = 0;
186 		sema[i].sem_perm.mode = 0;
187 	}
188 	for (i = 0; i < seminfo.semmnu; i++) {
189 		struct sem_undo *suptr = SEMU(i);
190 		suptr->un_proc = NULL;
191 	}
192 	semu_list = NULL;
193 }
194 SYSINIT(sysv_sem, SI_SUB_SYSV_SEM, SI_ORDER_FIRST, seminit, NULL)
195 
196 /*
197  * Entry point for all SEM calls
198  *
199  * semsys_args(int which, a2, a3, ...) (VARARGS)
200  *
201  * MPALMOSTSAFE
202  */
203 int
204 sys_semsys(struct semsys_args *uap)
205 {
206 	struct thread *td = curthread;
207 	unsigned int which = (unsigned int)uap->which;
208 	int error;
209 
210 	if (!jail_sysvipc_allowed && td->td_ucred->cr_prison != NULL)
211 		return (ENOSYS);
212 
213 	if (which >= sizeof(semcalls)/sizeof(semcalls[0]))
214 		return (EINVAL);
215 	bcopy(&uap->a2, &uap->which,
216 	      sizeof(struct semsys_args) - offsetof(struct semsys_args, a2));
217 	get_mplock();
218 	error = (*semcalls[which])(uap);
219 	rel_mplock();
220 	return (error);
221 }
222 
223 /*
224  * Allocate a new sem_undo structure for a process
225  * (returns ptr to structure or NULL if no more room)
226  */
227 
228 static struct sem_undo *
229 semu_alloc(struct proc *p)
230 {
231 	int i;
232 	struct sem_undo *suptr;
233 	struct sem_undo **supptr;
234 	int attempt;
235 
236 	/*
237 	 * Try twice to allocate something.
238 	 * (we'll purge any empty structures after the first pass so
239 	 * two passes are always enough)
240 	 */
241 
242 	for (attempt = 0; attempt < 2; attempt++) {
243 		/*
244 		 * Look for a free structure.
245 		 * Fill it in and return it if we find one.
246 		 */
247 
248 		for (i = 0; i < seminfo.semmnu; i++) {
249 			suptr = SEMU(i);
250 			if (suptr->un_proc == NULL) {
251 				suptr->un_next = semu_list;
252 				semu_list = suptr;
253 				suptr->un_cnt = 0;
254 				suptr->un_proc = p;
255 				return(suptr);
256 			}
257 		}
258 
259 		/*
260 		 * We didn't find a free one, if this is the first attempt
261 		 * then try to free some structures.
262 		 */
263 
264 		if (attempt == 0) {
265 			/* All the structures are in use - try to free some */
266 			int did_something = 0;
267 
268 			supptr = &semu_list;
269 			while ((suptr = *supptr) != NULL) {
270 				if (suptr->un_cnt == 0)  {
271 					suptr->un_proc = NULL;
272 					*supptr = suptr->un_next;
273 					did_something = 1;
274 				} else
275 					supptr = &(suptr->un_next);
276 			}
277 
278 			/* If we didn't free anything then just give-up */
279 			if (!did_something)
280 				return(NULL);
281 		} else {
282 			/*
283 			 * The second pass failed even though we freed
284 			 * something after the first pass!
285 			 * This is IMPOSSIBLE!
286 			 */
287 			panic("semu_alloc - second attempt failed");
288 		}
289 	}
290 	return (NULL);
291 }
292 
293 /*
294  * Adjust a particular entry for a particular proc
295  */
296 
297 static int
298 semundo_adjust(struct proc *p, struct sem_undo **supptr, int semid, int semnum,
299 	       int adjval)
300 {
301 	struct sem_undo *suptr;
302 	struct undo *sunptr;
303 	int i;
304 
305 	/* Look for and remember the sem_undo if the caller doesn't provide
306 	   it */
307 
308 	suptr = *supptr;
309 	if (suptr == NULL) {
310 		for (suptr = semu_list; suptr != NULL;
311 		    suptr = suptr->un_next) {
312 			if (suptr->un_proc == p) {
313 				*supptr = suptr;
314 				break;
315 			}
316 		}
317 		if (suptr == NULL) {
318 			if (adjval == 0)
319 				return(0);
320 			suptr = semu_alloc(p);
321 			if (suptr == NULL)
322 				return(ENOSPC);
323 			*supptr = suptr;
324 		}
325 	}
326 
327 	/*
328 	 * Look for the requested entry and adjust it (delete if adjval becomes
329 	 * 0).
330 	 */
331 	sunptr = &suptr->un_ent[0];
332 	for (i = 0; i < suptr->un_cnt; i++, sunptr++) {
333 		if (sunptr->un_id != semid || sunptr->un_num != semnum)
334 			continue;
335 		if (adjval == 0)
336 			sunptr->un_adjval = 0;
337 		else
338 			sunptr->un_adjval += adjval;
339 		if (sunptr->un_adjval == 0) {
340 			suptr->un_cnt--;
341 			if (i < suptr->un_cnt)
342 				suptr->un_ent[i] =
343 				    suptr->un_ent[suptr->un_cnt];
344 		}
345 		return(0);
346 	}
347 
348 	/* Didn't find the right entry - create it */
349 	if (adjval == 0)
350 		return(0);
351 	if (suptr->un_cnt != seminfo.semume) {
352 		sunptr = &suptr->un_ent[suptr->un_cnt];
353 		suptr->un_cnt++;
354 		sunptr->un_adjval = adjval;
355 		sunptr->un_id = semid; sunptr->un_num = semnum;
356 	} else
357 		return(EINVAL);
358 	return(0);
359 }
360 
361 static void
362 semundo_clear(int semid, int semnum)
363 {
364 	struct sem_undo *suptr;
365 
366 	for (suptr = semu_list; suptr != NULL; suptr = suptr->un_next) {
367 		struct undo *sunptr = &suptr->un_ent[0];
368 		int i = 0;
369 
370 		while (i < suptr->un_cnt) {
371 			if (sunptr->un_id == semid) {
372 				if (semnum == -1 || sunptr->un_num == semnum) {
373 					suptr->un_cnt--;
374 					if (i < suptr->un_cnt) {
375 						suptr->un_ent[i] =
376 						  suptr->un_ent[suptr->un_cnt];
377 						continue;
378 					}
379 				}
380 				if (semnum != -1)
381 					break;
382 			}
383 			i++, sunptr++;
384 		}
385 	}
386 }
387 
388 /*
389  * Note that the user-mode half of this passes a union, not a pointer
390  *
391  * MPALMOSTSAFE
392  */
393 int
394 sys___semctl(struct __semctl_args *uap)
395 {
396 	struct thread *td = curthread;
397 	int semid = uap->semid;
398 	int semnum = uap->semnum;
399 	int cmd = uap->cmd;
400 	union semun *arg = uap->arg;
401 	union semun real_arg;
402 	struct ucred *cred = td->td_ucred;
403 	int i, rval, eval;
404 	struct semid_ds sbuf;
405 	struct semid_ds *semaptr;
406 	struct semid_ds *semakptr;
407 
408 #ifdef SEM_DEBUG
409 	kprintf("call to semctl(%d, %d, %d, 0x%x)\n", semid, semnum, cmd, arg);
410 #endif
411 
412 	if (!jail_sysvipc_allowed && cred->cr_prison != NULL)
413 		return (ENOSYS);
414 
415 	get_mplock();
416 	switch (cmd) {
417 	case SEM_STAT:
418 		/*
419 		 * For this command we assume semid is an array index
420 		 * rather than an IPC id.
421 		 */
422 		if (semid < 0 || semid >= seminfo.semmni) {
423 			eval = EINVAL;
424 			break;
425 		}
426 		semakptr = &sema[semid];
427 		if ((semakptr->sem_perm.mode & SEM_ALLOC) == 0) {
428 			eval = EINVAL;
429 			break;
430 		}
431 		if ((eval = ipcperm(td->td_proc, &semakptr->sem_perm, IPC_R)))
432 			break;
433 
434 		bcopy(&semakptr, arg->buf, sizeof(struct semid_ds));
435 		rval = IXSEQ_TO_IPCID(semid, semakptr->sem_perm);
436 		break;
437 	}
438 
439 	semid = IPCID_TO_IX(semid);
440 	if (semid < 0 || semid >= seminfo.semmni) {
441 		rel_mplock();
442 		return(EINVAL);
443 	}
444 
445 	semaptr = &sema[semid];
446 	if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0 ||
447 	    semaptr->sem_perm.seq != IPCID_TO_SEQ(uap->semid)) {
448 		rel_mplock();
449 		return(EINVAL);
450 	}
451 
452 	eval = 0;
453 	rval = 0;
454 
455 	switch (cmd) {
456 	case IPC_RMID:
457 		if ((eval = ipcperm(td->td_proc, &semaptr->sem_perm, IPC_M)) != 0)
458 			break;
459 		semaptr->sem_perm.cuid = cred->cr_uid;
460 		semaptr->sem_perm.uid = cred->cr_uid;
461 		semtot -= semaptr->sem_nsems;
462 		for (i = semaptr->sem_base - sem; i < semtot; i++)
463 			sem[i] = sem[i + semaptr->sem_nsems];
464 		for (i = 0; i < seminfo.semmni; i++) {
465 			if ((sema[i].sem_perm.mode & SEM_ALLOC) &&
466 			    sema[i].sem_base > semaptr->sem_base)
467 				sema[i].sem_base -= semaptr->sem_nsems;
468 		}
469 		semaptr->sem_perm.mode = 0;
470 		semundo_clear(semid, -1);
471 		wakeup((caddr_t)semaptr);
472 		break;
473 
474 	case IPC_SET:
475 		eval = ipcperm(td->td_proc, &semaptr->sem_perm, IPC_M);
476 		if (eval)
477 			break;
478 		if ((eval = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
479 			break;
480 		if ((eval = copyin(real_arg.buf, (caddr_t)&sbuf,
481 				   sizeof(sbuf))) != 0) {
482 			break;
483 		}
484 		semaptr->sem_perm.uid = sbuf.sem_perm.uid;
485 		semaptr->sem_perm.gid = sbuf.sem_perm.gid;
486 		semaptr->sem_perm.mode = (semaptr->sem_perm.mode & ~0777) |
487 					 (sbuf.sem_perm.mode & 0777);
488 		semaptr->sem_ctime = time_second;
489 		break;
490 
491 	case IPC_STAT:
492 		if ((eval = ipcperm(td->td_proc, &semaptr->sem_perm, IPC_R)))
493 			break;
494 		if ((eval = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
495 			break;
496 		eval = copyout(semaptr, real_arg.buf, sizeof(struct semid_ds));
497 		break;
498 
499 	case GETNCNT:
500 		eval = ipcperm(td->td_proc, &semaptr->sem_perm, IPC_R);
501 		if (eval)
502 			break;
503 		if (semnum < 0 || semnum >= semaptr->sem_nsems) {
504 			eval = EINVAL;
505 			break;
506 		}
507 		rval = semaptr->sem_base[semnum].semncnt;
508 		break;
509 
510 	case GETPID:
511 		eval = ipcperm(td->td_proc, &semaptr->sem_perm, IPC_R);
512 		if (eval)
513 			break;
514 		if (semnum < 0 || semnum >= semaptr->sem_nsems) {
515 			eval = EINVAL;
516 			break;
517 		}
518 		rval = semaptr->sem_base[semnum].sempid;
519 		break;
520 
521 	case GETVAL:
522 		eval = ipcperm(td->td_proc, &semaptr->sem_perm, IPC_R);
523 		if (eval)
524 			break;
525 		if (semnum < 0 || semnum >= semaptr->sem_nsems) {
526 			eval = EINVAL;
527 			break;
528 		}
529 		rval = semaptr->sem_base[semnum].semval;
530 		break;
531 
532 	case GETALL:
533 		eval = ipcperm(td->td_proc, &semaptr->sem_perm, IPC_R);
534 		if (eval)
535 			break;
536 		if ((eval = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
537 			break;
538 		for (i = 0; i < semaptr->sem_nsems; i++) {
539 			eval = copyout(&semaptr->sem_base[i].semval,
540 				       &real_arg.array[i],
541 				       sizeof(real_arg.array[0]));
542 			if (eval)
543 				break;
544 		}
545 		break;
546 
547 	case GETZCNT:
548 		eval = ipcperm(td->td_proc, &semaptr->sem_perm, IPC_R);
549 		if (eval)
550 			break;
551 		if (semnum < 0 || semnum >= semaptr->sem_nsems) {
552 			eval = EINVAL;
553 			break;
554 		}
555 		rval = semaptr->sem_base[semnum].semzcnt;
556 		break;
557 
558 	case SETVAL:
559 		eval = ipcperm(td->td_proc, &semaptr->sem_perm, IPC_W);
560 		if (eval)
561 			break;
562 		if (semnum < 0 || semnum >= semaptr->sem_nsems) {
563 			eval = EINVAL;
564 			break;
565 		}
566 		if ((eval = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
567 			break;
568 		semaptr->sem_base[semnum].semval = real_arg.val;
569 		semundo_clear(semid, semnum);
570 		wakeup((caddr_t)semaptr);
571 		break;
572 
573 	case SETALL:
574 		eval = ipcperm(td->td_proc, &semaptr->sem_perm, IPC_W);
575 		if (eval)
576 			break;
577 		if ((eval = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
578 			break;
579 		for (i = 0; i < semaptr->sem_nsems; i++) {
580 			eval = copyin(&real_arg.array[i],
581 			    (caddr_t)&semaptr->sem_base[i].semval,
582 			    sizeof(real_arg.array[0]));
583 			if (eval != 0)
584 				break;
585 		}
586 		semundo_clear(semid, -1);
587 		wakeup((caddr_t)semaptr);
588 		break;
589 
590 	default:
591 		eval = EINVAL;
592 		break;
593 	}
594 	rel_mplock();
595 
596 	if (eval == 0)
597 		uap->sysmsg_result = rval;
598 	return(eval);
599 }
600 
601 /*
602  * MPALMOSTSAFE
603  */
604 int
605 sys_semget(struct semget_args *uap)
606 {
607 	struct thread *td = curthread;
608 	int semid, eval;
609 	int key = uap->key;
610 	int nsems = uap->nsems;
611 	int semflg = uap->semflg;
612 	struct ucred *cred = td->td_ucred;
613 
614 #ifdef SEM_DEBUG
615 	kprintf("semget(0x%x, %d, 0%o)\n", key, nsems, semflg);
616 #endif
617 
618 	if (!jail_sysvipc_allowed && cred->cr_prison != NULL)
619 		return (ENOSYS);
620 
621 	get_mplock();
622 	eval = 0;
623 
624 	if (key != IPC_PRIVATE) {
625 		for (semid = 0; semid < seminfo.semmni; semid++) {
626 			if ((sema[semid].sem_perm.mode & SEM_ALLOC) &&
627 			    sema[semid].sem_perm.key == key)
628 				break;
629 		}
630 		if (semid < seminfo.semmni) {
631 #ifdef SEM_DEBUG
632 			kprintf("found public key\n");
633 #endif
634 			if ((eval = ipcperm(td->td_proc,
635 					    &sema[semid].sem_perm,
636 					    semflg & 0700))) {
637 				goto done;
638 			}
639 			if (nsems > 0 && sema[semid].sem_nsems < nsems) {
640 #ifdef SEM_DEBUG
641 				kprintf("too small\n");
642 #endif
643 				eval = EINVAL;
644 				goto done;
645 			}
646 			if ((semflg & IPC_CREAT) && (semflg & IPC_EXCL)) {
647 #ifdef SEM_DEBUG
648 				kprintf("not exclusive\n");
649 #endif
650 				eval = EEXIST;
651 				goto done;
652 			}
653 			goto done;
654 		}
655 	}
656 
657 #ifdef SEM_DEBUG
658 	kprintf("need to allocate the semid_ds\n");
659 #endif
660 	if (key == IPC_PRIVATE || (semflg & IPC_CREAT)) {
661 		if (nsems <= 0 || nsems > seminfo.semmsl) {
662 #ifdef SEM_DEBUG
663 			kprintf("nsems out of range (0<%d<=%d)\n", nsems,
664 			    seminfo.semmsl);
665 #endif
666 			eval = EINVAL;
667 			goto done;
668 		}
669 		if (nsems > seminfo.semmns - semtot) {
670 #ifdef SEM_DEBUG
671 			kprintf("not enough semaphores left (need %d, got %d)\n",
672 			    nsems, seminfo.semmns - semtot);
673 #endif
674 			eval = ENOSPC;
675 			goto done;
676 		}
677 		for (semid = 0; semid < seminfo.semmni; semid++) {
678 			if ((sema[semid].sem_perm.mode & SEM_ALLOC) == 0)
679 				break;
680 		}
681 		if (semid == seminfo.semmni) {
682 #ifdef SEM_DEBUG
683 			kprintf("no more semid_ds's available\n");
684 #endif
685 			eval = ENOSPC;
686 			goto done;
687 		}
688 #ifdef SEM_DEBUG
689 		kprintf("semid %d is available\n", semid);
690 #endif
691 		sema[semid].sem_perm.key = key;
692 		sema[semid].sem_perm.cuid = cred->cr_uid;
693 		sema[semid].sem_perm.uid = cred->cr_uid;
694 		sema[semid].sem_perm.cgid = cred->cr_gid;
695 		sema[semid].sem_perm.gid = cred->cr_gid;
696 		sema[semid].sem_perm.mode = (semflg & 0777) | SEM_ALLOC;
697 		sema[semid].sem_perm.seq =
698 		    (sema[semid].sem_perm.seq + 1) & 0x7fff;
699 		sema[semid].sem_nsems = nsems;
700 		sema[semid].sem_otime = 0;
701 		sema[semid].sem_ctime = time_second;
702 		sema[semid].sem_base = &sem[semtot];
703 		semtot += nsems;
704 		bzero(sema[semid].sem_base,
705 		    sizeof(sema[semid].sem_base[0])*nsems);
706 #ifdef SEM_DEBUG
707 		kprintf("sembase = 0x%x, next = 0x%x\n", sema[semid].sem_base,
708 		    &sem[semtot]);
709 #endif
710 	} else {
711 #ifdef SEM_DEBUG
712 		kprintf("didn't find it and wasn't asked to create it\n");
713 #endif
714 		eval = ENOENT;
715 	}
716 
717 done:
718 	if (eval == 0) {
719 		uap->sysmsg_result = IXSEQ_TO_IPCID(semid,
720 						    sema[semid].sem_perm);
721 	}
722 	rel_mplock();
723 	return(eval);
724 }
725 
726 /*
727  * MPALMOSTSAFE
728  */
729 int
730 sys_semop(struct semop_args *uap)
731 {
732 	struct thread *td = curthread;
733 	int semid = uap->semid;
734 	u_int nsops = uap->nsops;
735 	struct sembuf sops[MAX_SOPS];
736 	struct semid_ds *semaptr;
737 	struct sembuf *sopptr;
738 	struct sem *semptr;
739 	struct sem_undo *suptr = NULL;
740 	int i, j, eval;
741 	int do_wakeup, do_undos;
742 
743 #ifdef SEM_DEBUG
744 	kprintf("call to semop(%d, 0x%x, %u)\n", semid, sops, nsops);
745 #endif
746 
747 	if (!jail_sysvipc_allowed && td->td_ucred->cr_prison != NULL)
748 		return (ENOSYS);
749 
750 	get_mplock();
751 	semid = IPCID_TO_IX(semid);	/* Convert back to zero origin */
752 
753 	if (semid < 0 || semid >= seminfo.semmni) {
754 		eval = EINVAL;
755 		goto done;
756 	}
757 
758 	semaptr = &sema[semid];
759 	if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0) {
760 		eval = EINVAL;
761 		goto done;
762 	}
763 	if (semaptr->sem_perm.seq != IPCID_TO_SEQ(uap->semid)) {
764 		eval = EINVAL;
765 		goto done;
766 	}
767 
768 	if ((eval = ipcperm(td->td_proc, &semaptr->sem_perm, IPC_W))) {
769 #ifdef SEM_DEBUG
770 		kprintf("eval = %d from ipaccess\n", eval);
771 #endif
772 		goto done;
773 	}
774 
775 	if (nsops > MAX_SOPS) {
776 #ifdef SEM_DEBUG
777 		kprintf("too many sops (max=%d, nsops=%u)\n", MAX_SOPS, nsops);
778 #endif
779 		eval = E2BIG;
780 		goto done;
781 	}
782 
783 	if ((eval = copyin(uap->sops, &sops, nsops * sizeof(sops[0]))) != 0) {
784 #ifdef SEM_DEBUG
785 		kprintf("eval = %d from copyin(%08x, %08x, %u)\n", eval,
786 		    uap->sops, &sops, nsops * sizeof(sops[0]));
787 #endif
788 		goto done;
789 	}
790 
791 	/*
792 	 * Loop trying to satisfy the vector of requests.
793 	 * If we reach a point where we must wait, any requests already
794 	 * performed are rolled back and we go to sleep until some other
795 	 * process wakes us up.  At this point, we start all over again.
796 	 *
797 	 * This ensures that from the perspective of other tasks, a set
798 	 * of requests is atomic (never partially satisfied).
799 	 */
800 	do_undos = 0;
801 
802 	for (;;) {
803 		do_wakeup = 0;
804 
805 		for (i = 0; i < nsops; i++) {
806 			sopptr = &sops[i];
807 
808 			if (sopptr->sem_num >= semaptr->sem_nsems) {
809 				eval = EFBIG;
810 				goto done;
811 			}
812 
813 			semptr = &semaptr->sem_base[sopptr->sem_num];
814 
815 #ifdef SEM_DEBUG
816 			kprintf("semop:  semaptr=%x, sem_base=%x, semptr=%x, sem[%d]=%d : op=%d, flag=%s\n",
817 			    semaptr, semaptr->sem_base, semptr,
818 			    sopptr->sem_num, semptr->semval, sopptr->sem_op,
819 			    (sopptr->sem_flg & IPC_NOWAIT) ? "nowait" : "wait");
820 #endif
821 
822 			if (sopptr->sem_op < 0) {
823 				if (semptr->semval + sopptr->sem_op < 0) {
824 #ifdef SEM_DEBUG
825 					kprintf("semop:  can't do it now\n");
826 #endif
827 					break;
828 				} else {
829 					semptr->semval += sopptr->sem_op;
830 					if (semptr->semval == 0 &&
831 					    semptr->semzcnt > 0)
832 						do_wakeup = 1;
833 				}
834 				if (sopptr->sem_flg & SEM_UNDO)
835 					do_undos = 1;
836 			} else if (sopptr->sem_op == 0) {
837 				if (semptr->semval > 0) {
838 #ifdef SEM_DEBUG
839 					kprintf("semop:  not zero now\n");
840 #endif
841 					break;
842 				}
843 			} else {
844 				if (semptr->semncnt > 0)
845 					do_wakeup = 1;
846 				semptr->semval += sopptr->sem_op;
847 				if (sopptr->sem_flg & SEM_UNDO)
848 					do_undos = 1;
849 			}
850 		}
851 
852 		/*
853 		 * Did we get through the entire vector?
854 		 */
855 		if (i >= nsops)
856 			goto donex;
857 
858 		/*
859 		 * No ... rollback anything that we've already done
860 		 */
861 #ifdef SEM_DEBUG
862 		kprintf("semop:  rollback 0 through %d\n", i-1);
863 #endif
864 		for (j = 0; j < i; j++)
865 			semaptr->sem_base[sops[j].sem_num].semval -=
866 			    sops[j].sem_op;
867 
868 		/*
869 		 * If the request that we couldn't satisfy has the
870 		 * NOWAIT flag set then return with EAGAIN.
871 		 */
872 		if (sopptr->sem_flg & IPC_NOWAIT) {
873 			eval = EAGAIN;
874 			goto done;
875 		}
876 
877 		if (sopptr->sem_op == 0)
878 			semptr->semzcnt++;
879 		else
880 			semptr->semncnt++;
881 
882 #ifdef SEM_DEBUG
883 		kprintf("semop:  good night!\n");
884 #endif
885 		eval = tsleep((caddr_t)semaptr, PCATCH, "semwait", 0);
886 #ifdef SEM_DEBUG
887 		kprintf("semop:  good morning (eval=%d)!\n", eval);
888 #endif
889 
890 		suptr = NULL;	/* sem_undo may have been reallocated */
891 
892 		/* return code is checked below, after sem[nz]cnt-- */
893 
894 		/*
895 		 * Make sure that the semaphore still exists
896 		 */
897 		if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0 ||
898 		    semaptr->sem_perm.seq != IPCID_TO_SEQ(uap->semid)) {
899 			eval = EIDRM;
900 			goto done;
901 		}
902 
903 		/*
904 		 * The semaphore is still alive.  Readjust the count of
905 		 * waiting processes.
906 		 */
907 		if (sopptr->sem_op == 0)
908 			semptr->semzcnt--;
909 		else
910 			semptr->semncnt--;
911 
912 		/*
913 		 * Is it really morning, or was our sleep interrupted?
914 		 * (Delayed check of tsleep() return code because we
915 		 * need to decrement sem[nz]cnt either way.)
916 		 */
917 		if (eval) {
918 			eval = EINTR;
919 			goto done;
920 		}
921 #ifdef SEM_DEBUG
922 		kprintf("semop:  good morning!\n");
923 #endif
924 	}
925 
926 donex:
927 	/*
928 	 * Process any SEM_UNDO requests.
929 	 */
930 	if (do_undos) {
931 		for (i = 0; i < nsops; i++) {
932 			/*
933 			 * We only need to deal with SEM_UNDO's for non-zero
934 			 * op's.
935 			 */
936 			int adjval;
937 
938 			if ((sops[i].sem_flg & SEM_UNDO) == 0)
939 				continue;
940 			adjval = sops[i].sem_op;
941 			if (adjval == 0)
942 				continue;
943 			eval = semundo_adjust(td->td_proc, &suptr, semid,
944 					      sops[i].sem_num, -adjval);
945 			if (eval == 0)
946 				continue;
947 
948 			/*
949 			 * Oh-Oh!  We ran out of either sem_undo's or undo's.
950 			 * Rollback the adjustments to this point and then
951 			 * rollback the semaphore ups and down so we can return
952 			 * with an error with all structures restored.  We
953 			 * rollback the undo's in the exact reverse order that
954 			 * we applied them.  This guarantees that we won't run
955 			 * out of space as we roll things back out.
956 			 */
957 			for (j = i - 1; j >= 0; j--) {
958 				if ((sops[j].sem_flg & SEM_UNDO) == 0)
959 					continue;
960 				adjval = sops[j].sem_op;
961 				if (adjval == 0)
962 					continue;
963 				if (semundo_adjust(td->td_proc, &suptr, semid,
964 					       sops[j].sem_num, adjval) != 0)
965 					panic("semop - can't undo undos");
966 			}
967 
968 			for (j = 0; j < nsops; j++)
969 				semaptr->sem_base[sops[j].sem_num].semval -=
970 				    sops[j].sem_op;
971 
972 #ifdef SEM_DEBUG
973 			kprintf("eval = %d from semundo_adjust\n", eval);
974 #endif
975 			goto done;
976 		} /* loop through the sops */
977 	} /* if (do_undos) */
978 
979 	/* We're definitely done - set the sempid's */
980 	for (i = 0; i < nsops; i++) {
981 		sopptr = &sops[i];
982 		semptr = &semaptr->sem_base[sopptr->sem_num];
983 		semptr->sempid = td->td_proc->p_pid;
984 	}
985 
986 	/* Do a wakeup if any semaphore was up'd. */
987 	if (do_wakeup) {
988 #ifdef SEM_DEBUG
989 		kprintf("semop:  doing wakeup\n");
990 #endif
991 		wakeup((caddr_t)semaptr);
992 #ifdef SEM_DEBUG
993 		kprintf("semop:  back from wakeup\n");
994 #endif
995 	}
996 #ifdef SEM_DEBUG
997 	kprintf("semop:  done\n");
998 #endif
999 	uap->sysmsg_result = 0;
1000 	eval = 0;
1001 done:
1002 	rel_mplock();
1003 	return(eval);
1004 }
1005 
1006 /*
1007  * Go through the undo structures for this process and apply the adjustments to
1008  * semaphores.
1009  */
1010 void
1011 semexit(struct proc *p)
1012 {
1013 	struct sem_undo *suptr;
1014 	struct sem_undo **supptr;
1015 	int did_something;
1016 
1017 	did_something = 0;
1018 
1019 	/*
1020 	 * Go through the chain of undo vectors looking for one
1021 	 * associated with this process.
1022 	 */
1023 
1024 	for (supptr = &semu_list; (suptr = *supptr) != NULL;
1025 	    supptr = &suptr->un_next) {
1026 		if (suptr->un_proc == p)
1027 			break;
1028 	}
1029 
1030 	if (suptr == NULL)
1031 		return;
1032 
1033 #ifdef SEM_DEBUG
1034 	kprintf("proc @%08x has undo structure with %d entries\n", p,
1035 	    suptr->un_cnt);
1036 #endif
1037 
1038 	/*
1039 	 * If there are any active undo elements then process them.
1040 	 */
1041 	if (suptr->un_cnt > 0) {
1042 		int ix;
1043 
1044 		for (ix = 0; ix < suptr->un_cnt; ix++) {
1045 			int semid = suptr->un_ent[ix].un_id;
1046 			int semnum = suptr->un_ent[ix].un_num;
1047 			int adjval = suptr->un_ent[ix].un_adjval;
1048 			struct semid_ds *semaptr;
1049 
1050 			semaptr = &sema[semid];
1051 			if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0)
1052 				panic("semexit - semid not allocated");
1053 			if (semnum >= semaptr->sem_nsems)
1054 				panic("semexit - semnum out of range");
1055 
1056 #ifdef SEM_DEBUG
1057 			kprintf("semexit:  %08x id=%d num=%d(adj=%d) ; sem=%d\n",
1058 			    suptr->un_proc, suptr->un_ent[ix].un_id,
1059 			    suptr->un_ent[ix].un_num,
1060 			    suptr->un_ent[ix].un_adjval,
1061 			    semaptr->sem_base[semnum].semval);
1062 #endif
1063 
1064 			if (adjval < 0) {
1065 				if (semaptr->sem_base[semnum].semval < -adjval)
1066 					semaptr->sem_base[semnum].semval = 0;
1067 				else
1068 					semaptr->sem_base[semnum].semval +=
1069 					    adjval;
1070 			} else
1071 				semaptr->sem_base[semnum].semval += adjval;
1072 
1073 			wakeup((caddr_t)semaptr);
1074 #ifdef SEM_DEBUG
1075 			kprintf("semexit:  back from wakeup\n");
1076 #endif
1077 		}
1078 	}
1079 
1080 	/*
1081 	 * Deallocate the undo vector.
1082 	 */
1083 #ifdef SEM_DEBUG
1084 	kprintf("removing vector\n");
1085 #endif
1086 	suptr->un_proc = NULL;
1087 	*supptr = suptr->un_next;
1088 }
1089