xref: /dragonfly/sys/kern/sysv_sem.c (revision 62f7f702) 
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 static MALLOC_DEFINE(M_SEM, "sem", "SVID compatible semaphores");
26 
27 static void seminit (void *);
28 
29 static struct sem_undo *semu_alloc (struct proc *p);
30 static int semundo_adjust (struct proc *p, struct sem_undo **supptr,
31 		int semid, int semnum, int adjval);
32 static void semundo_clear (int semid, int semnum);
33 
34 /* XXX casting to (sy_call_t *) is bogus, as usual. */
35 static sy_call_t *semcalls[] = {
36 	(sy_call_t *)sys___semctl, (sy_call_t *)sys_semget,
37 	(sy_call_t *)sys_semop
38 };
39 
40 static int	semtot = 0;
41 static struct semid_ds *sema;	/* semaphore id pool */
42 static struct sem *sem;		/* semaphore pool */
43 static struct sem_undo *semu_list; /* list of active undo structures */
44 static int	*semu;		/* undo structure pool */
45 
46 struct sem {
47 	u_short	semval;		/* semaphore value */
48 	pid_t	sempid;		/* pid of last operation */
49 	u_short	semncnt;	/* # awaiting semval > cval */
50 	u_short	semzcnt;	/* # awaiting semval = 0 */
51 };
52 
53 /*
54  * Undo structure (one per process)
55  */
56 struct sem_undo {
57 	struct	sem_undo *un_next;	/* ptr to next active undo structure */
58 	struct	proc *un_proc;		/* owner of this structure */
59 	short	un_cnt;			/* # of active entries */
60 	struct undo {
61 		short	un_adjval;	/* adjust on exit values */
62 		short	un_num;		/* semaphore # */
63 		int	un_id;		/* semid */
64 	} un_ent[1];			/* undo entries */
65 };
66 
67 /*
68  * Configuration parameters
69  */
70 #ifndef SEMMNI
71 #define SEMMNI	10		/* # of semaphore identifiers */
72 #endif
73 #ifndef SEMMNS
74 #define SEMMNS	60		/* # of semaphores in system */
75 #endif
76 #ifndef SEMUME
77 #define SEMUME	10		/* max # of undo entries per process */
78 #endif
79 #ifndef SEMMNU
80 #define SEMMNU	30		/* # of undo structures in system */
81 #endif
82 
83 /* shouldn't need tuning */
84 #ifndef SEMMAP
85 #define SEMMAP	30		/* # of entries in semaphore map */
86 #endif
87 #ifndef SEMMSL
88 #define SEMMSL	SEMMNS		/* max # of semaphores per id */
89 #endif
90 #ifndef SEMOPM
91 #define SEMOPM	100		/* max # of operations per semop call */
92 #endif
93 
94 #define SEMVMX	32767		/* semaphore maximum value */
95 #define SEMAEM	16384		/* adjust on exit max value */
96 
97 /*
98  * Due to the way semaphore memory is allocated, we have to ensure that
99  * SEMUSZ is properly aligned.
100  */
101 
102 #define SEM_ALIGN(bytes) (((bytes) + (sizeof(long) - 1)) & ~(sizeof(long) - 1))
103 
104 /* actual size of an undo structure */
105 #define SEMUSZ	SEM_ALIGN(offsetof(struct sem_undo, un_ent[SEMUME]))
106 
107 /*
108  * Macro to find a particular sem_undo vector
109  */
110 #define SEMU(ix)	((struct sem_undo *)(((intptr_t)semu)+ix * seminfo.semusz))
111 
112 /*
113  * semaphore info struct
114  */
115 struct seminfo seminfo = {
116                 SEMMAP,         /* # of entries in semaphore map */
117                 SEMMNI,         /* # of semaphore identifiers */
118                 SEMMNS,         /* # of semaphores in system */
119                 SEMMNU,         /* # of undo structures in system */
120                 SEMMSL,         /* max # of semaphores per id */
121                 SEMOPM,         /* max # of operations per semop call */
122                 SEMUME,         /* max # of undo entries per process */
123                 SEMUSZ,         /* size in bytes of undo structure */
124                 SEMVMX,         /* semaphore maximum value */
125                 SEMAEM          /* adjust on exit max value */
126 };
127 
128 TUNABLE_INT("kern.ipc.semmap", &seminfo.semmap);
129 TUNABLE_INT("kern.ipc.semmni", &seminfo.semmni);
130 TUNABLE_INT("kern.ipc.semmns", &seminfo.semmns);
131 TUNABLE_INT("kern.ipc.semmnu", &seminfo.semmnu);
132 TUNABLE_INT("kern.ipc.semmsl", &seminfo.semmsl);
133 TUNABLE_INT("kern.ipc.semopm", &seminfo.semopm);
134 TUNABLE_INT("kern.ipc.semume", &seminfo.semume);
135 TUNABLE_INT("kern.ipc.semusz", &seminfo.semusz);
136 TUNABLE_INT("kern.ipc.semvmx", &seminfo.semvmx);
137 TUNABLE_INT("kern.ipc.semaem", &seminfo.semaem);
138 
139 SYSCTL_INT(_kern_ipc, OID_AUTO, semmap, CTLFLAG_RW, &seminfo.semmap, 0, "");
140 SYSCTL_INT(_kern_ipc, OID_AUTO, semmni, CTLFLAG_RD, &seminfo.semmni, 0, "");
141 SYSCTL_INT(_kern_ipc, OID_AUTO, semmns, CTLFLAG_RD, &seminfo.semmns, 0, "");
142 SYSCTL_INT(_kern_ipc, OID_AUTO, semmnu, CTLFLAG_RD, &seminfo.semmnu, 0, "");
143 SYSCTL_INT(_kern_ipc, OID_AUTO, semmsl, CTLFLAG_RW, &seminfo.semmsl, 0, "");
144 SYSCTL_INT(_kern_ipc, OID_AUTO, semopm, CTLFLAG_RD, &seminfo.semopm, 0, "");
145 SYSCTL_INT(_kern_ipc, OID_AUTO, semume, CTLFLAG_RD, &seminfo.semume, 0, "");
146 SYSCTL_INT(_kern_ipc, OID_AUTO, semusz, CTLFLAG_RD, &seminfo.semusz, 0, "");
147 SYSCTL_INT(_kern_ipc, OID_AUTO, semvmx, CTLFLAG_RW, &seminfo.semvmx, 0, "");
148 SYSCTL_INT(_kern_ipc, OID_AUTO, semaem, CTLFLAG_RW, &seminfo.semaem, 0, "");
149 
150 #if 0
151 RO seminfo.semmap	/* SEMMAP unused */
152 RO seminfo.semmni
153 RO seminfo.semmns
154 RO seminfo.semmnu	/* undo entries per system */
155 RW seminfo.semmsl
156 RO seminfo.semopm	/* SEMOPM unused */
157 RO seminfo.semume
158 RO seminfo.semusz	/* param - derived from SEMUME for per-proc sizeof */
159 RO seminfo.semvmx	/* SEMVMX unused - user param */
160 RO seminfo.semaem	/* SEMAEM unused - user param */
161 #endif
162 
163 static void
164 seminit(void *dummy)
165 {
166 	int i;
167 
168 	sem = kmalloc(sizeof(struct sem) * seminfo.semmns, M_SEM, M_WAITOK);
169 	sema = kmalloc(sizeof(struct semid_ds) * seminfo.semmni, M_SEM, M_WAITOK);
170 	semu = kmalloc(seminfo.semmnu * seminfo.semusz, M_SEM, M_WAITOK);
171 
172 	for (i = 0; i < seminfo.semmni; i++) {
173 		sema[i].sem_base = 0;
174 		sema[i].sem_perm.mode = 0;
175 	}
176 	for (i = 0; i < seminfo.semmnu; i++) {
177 		struct sem_undo *suptr = SEMU(i);
178 		suptr->un_proc = NULL;
179 	}
180 	semu_list = NULL;
181 }
182 SYSINIT(sysv_sem, SI_SUB_SYSV_SEM, SI_ORDER_FIRST, seminit, NULL)
183 
184 /*
185  * Entry point for all SEM calls
186  *
187  * semsys_args(int which, a2, a3, ...) (VARARGS)
188  */
189 int
190 sys_semsys(struct semsys_args *uap)
191 {
192 	struct proc *p = curproc;
193 	unsigned int which = (unsigned int)uap->which;
194 
195 	if (!jail_sysvipc_allowed && p->p_ucred->cr_prison != NULL)
196 		return (ENOSYS);
197 
198 	if (which >= sizeof(semcalls)/sizeof(semcalls[0]))
199 		return (EINVAL);
200 	bcopy(&uap->a2, &uap->which,
201 	    sizeof(struct semsys_args) - offsetof(struct semsys_args, a2));
202 	return ((*semcalls[which])(uap));
203 }
204 
205 /*
206  * Allocate a new sem_undo structure for a process
207  * (returns ptr to structure or NULL if no more room)
208  */
209 
210 static struct sem_undo *
211 semu_alloc(struct proc *p)
212 {
213 	int i;
214 	struct sem_undo *suptr;
215 	struct sem_undo **supptr;
216 	int attempt;
217 
218 	/*
219 	 * Try twice to allocate something.
220 	 * (we'll purge any empty structures after the first pass so
221 	 * two passes are always enough)
222 	 */
223 
224 	for (attempt = 0; attempt < 2; attempt++) {
225 		/*
226 		 * Look for a free structure.
227 		 * Fill it in and return it if we find one.
228 		 */
229 
230 		for (i = 0; i < seminfo.semmnu; i++) {
231 			suptr = SEMU(i);
232 			if (suptr->un_proc == NULL) {
233 				suptr->un_next = semu_list;
234 				semu_list = suptr;
235 				suptr->un_cnt = 0;
236 				suptr->un_proc = p;
237 				return(suptr);
238 			}
239 		}
240 
241 		/*
242 		 * We didn't find a free one, if this is the first attempt
243 		 * then try to free some structures.
244 		 */
245 
246 		if (attempt == 0) {
247 			/* All the structures are in use - try to free some */
248 			int did_something = 0;
249 
250 			supptr = &semu_list;
251 			while ((suptr = *supptr) != NULL) {
252 				if (suptr->un_cnt == 0)  {
253 					suptr->un_proc = NULL;
254 					*supptr = suptr->un_next;
255 					did_something = 1;
256 				} else
257 					supptr = &(suptr->un_next);
258 			}
259 
260 			/* If we didn't free anything then just give-up */
261 			if (!did_something)
262 				return(NULL);
263 		} else {
264 			/*
265 			 * The second pass failed even though we freed
266 			 * something after the first pass!
267 			 * This is IMPOSSIBLE!
268 			 */
269 			panic("semu_alloc - second attempt failed");
270 		}
271 	}
272 	return (NULL);
273 }
274 
275 /*
276  * Adjust a particular entry for a particular proc
277  */
278 
279 static int
280 semundo_adjust(struct proc *p, struct sem_undo **supptr, int semid, int semnum,
281 	       int adjval)
282 {
283 	struct sem_undo *suptr;
284 	struct undo *sunptr;
285 	int i;
286 
287 	/* Look for and remember the sem_undo if the caller doesn't provide
288 	   it */
289 
290 	suptr = *supptr;
291 	if (suptr == NULL) {
292 		for (suptr = semu_list; suptr != NULL;
293 		    suptr = suptr->un_next) {
294 			if (suptr->un_proc == p) {
295 				*supptr = suptr;
296 				break;
297 			}
298 		}
299 		if (suptr == NULL) {
300 			if (adjval == 0)
301 				return(0);
302 			suptr = semu_alloc(p);
303 			if (suptr == NULL)
304 				return(ENOSPC);
305 			*supptr = suptr;
306 		}
307 	}
308 
309 	/*
310 	 * Look for the requested entry and adjust it (delete if adjval becomes
311 	 * 0).
312 	 */
313 	sunptr = &suptr->un_ent[0];
314 	for (i = 0; i < suptr->un_cnt; i++, sunptr++) {
315 		if (sunptr->un_id != semid || sunptr->un_num != semnum)
316 			continue;
317 		if (adjval == 0)
318 			sunptr->un_adjval = 0;
319 		else
320 			sunptr->un_adjval += adjval;
321 		if (sunptr->un_adjval == 0) {
322 			suptr->un_cnt--;
323 			if (i < suptr->un_cnt)
324 				suptr->un_ent[i] =
325 				    suptr->un_ent[suptr->un_cnt];
326 		}
327 		return(0);
328 	}
329 
330 	/* Didn't find the right entry - create it */
331 	if (adjval == 0)
332 		return(0);
333 	if (suptr->un_cnt != seminfo.semume) {
334 		sunptr = &suptr->un_ent[suptr->un_cnt];
335 		suptr->un_cnt++;
336 		sunptr->un_adjval = adjval;
337 		sunptr->un_id = semid; sunptr->un_num = semnum;
338 	} else
339 		return(EINVAL);
340 	return(0);
341 }
342 
343 static void
344 semundo_clear(int semid, int semnum)
345 {
346 	struct sem_undo *suptr;
347 
348 	for (suptr = semu_list; suptr != NULL; suptr = suptr->un_next) {
349 		struct undo *sunptr = &suptr->un_ent[0];
350 		int i = 0;
351 
352 		while (i < suptr->un_cnt) {
353 			if (sunptr->un_id == semid) {
354 				if (semnum == -1 || sunptr->un_num == semnum) {
355 					suptr->un_cnt--;
356 					if (i < suptr->un_cnt) {
357 						suptr->un_ent[i] =
358 						  suptr->un_ent[suptr->un_cnt];
359 						continue;
360 					}
361 				}
362 				if (semnum != -1)
363 					break;
364 			}
365 			i++, sunptr++;
366 		}
367 	}
368 }
369 
370 /*
371  * Note that the user-mode half of this passes a union, not a pointer
372  */
373 
374 int
375 sys___semctl(struct __semctl_args *uap)
376 {
377 	struct proc *p = curproc;
378 	int semid = uap->semid;
379 	int semnum = uap->semnum;
380 	int cmd = uap->cmd;
381 	union semun *arg = uap->arg;
382 	union semun real_arg;
383 	struct ucred *cred = p->p_ucred;
384 	int i, rval, eval;
385 	struct semid_ds sbuf;
386 	struct semid_ds *semaptr;
387 
388 #ifdef SEM_DEBUG
389 	kprintf("call to semctl(%d, %d, %d, 0x%x)\n", semid, semnum, cmd, arg);
390 #endif
391 
392 	if (!jail_sysvipc_allowed && p->p_ucred->cr_prison != NULL)
393 		return (ENOSYS);
394 
395 	semid = IPCID_TO_IX(semid);
396 	if (semid < 0 || semid >= seminfo.semmni)
397 		return(EINVAL);
398 
399 	semaptr = &sema[semid];
400 	if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0 ||
401 	    semaptr->sem_perm.seq != IPCID_TO_SEQ(uap->semid))
402 		return(EINVAL);
403 
404 	eval = 0;
405 	rval = 0;
406 
407 	switch (cmd) {
408 	case IPC_RMID:
409 		if ((eval = ipcperm(p, &semaptr->sem_perm, IPC_M)))
410 			return(eval);
411 		semaptr->sem_perm.cuid = cred->cr_uid;
412 		semaptr->sem_perm.uid = cred->cr_uid;
413 		semtot -= semaptr->sem_nsems;
414 		for (i = semaptr->sem_base - sem; i < semtot; i++)
415 			sem[i] = sem[i + semaptr->sem_nsems];
416 		for (i = 0; i < seminfo.semmni; i++) {
417 			if ((sema[i].sem_perm.mode & SEM_ALLOC) &&
418 			    sema[i].sem_base > semaptr->sem_base)
419 				sema[i].sem_base -= semaptr->sem_nsems;
420 		}
421 		semaptr->sem_perm.mode = 0;
422 		semundo_clear(semid, -1);
423 		wakeup((caddr_t)semaptr);
424 		break;
425 
426 	case IPC_SET:
427 		if ((eval = ipcperm(p, &semaptr->sem_perm, IPC_M)))
428 			return(eval);
429 		if ((eval = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
430 			return(eval);
431 		if ((eval = copyin(real_arg.buf, (caddr_t)&sbuf,
432 		    sizeof(sbuf))) != 0)
433 			return(eval);
434 		semaptr->sem_perm.uid = sbuf.sem_perm.uid;
435 		semaptr->sem_perm.gid = sbuf.sem_perm.gid;
436 		semaptr->sem_perm.mode = (semaptr->sem_perm.mode & ~0777) |
437 		    (sbuf.sem_perm.mode & 0777);
438 		semaptr->sem_ctime = time_second;
439 		break;
440 
441 	case IPC_STAT:
442 		if ((eval = ipcperm(p, &semaptr->sem_perm, IPC_R)))
443 			return(eval);
444 		if ((eval = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
445 			return(eval);
446 		eval = copyout((caddr_t)semaptr, real_arg.buf,
447 		    sizeof(struct semid_ds));
448 		break;
449 
450 	case GETNCNT:
451 		if ((eval = ipcperm(p, &semaptr->sem_perm, IPC_R)))
452 			return(eval);
453 		if (semnum < 0 || semnum >= semaptr->sem_nsems)
454 			return(EINVAL);
455 		rval = semaptr->sem_base[semnum].semncnt;
456 		break;
457 
458 	case GETPID:
459 		if ((eval = ipcperm(p, &semaptr->sem_perm, IPC_R)))
460 			return(eval);
461 		if (semnum < 0 || semnum >= semaptr->sem_nsems)
462 			return(EINVAL);
463 		rval = semaptr->sem_base[semnum].sempid;
464 		break;
465 
466 	case GETVAL:
467 		if ((eval = ipcperm(p, &semaptr->sem_perm, IPC_R)))
468 			return(eval);
469 		if (semnum < 0 || semnum >= semaptr->sem_nsems)
470 			return(EINVAL);
471 		rval = semaptr->sem_base[semnum].semval;
472 		break;
473 
474 	case GETALL:
475 		if ((eval = ipcperm(p, &semaptr->sem_perm, IPC_R)))
476 			return(eval);
477 		if ((eval = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
478 			return(eval);
479 		for (i = 0; i < semaptr->sem_nsems; i++) {
480 			eval = copyout((caddr_t)&semaptr->sem_base[i].semval,
481 			    &real_arg.array[i], sizeof(real_arg.array[0]));
482 			if (eval != 0)
483 				break;
484 		}
485 		break;
486 
487 	case GETZCNT:
488 		if ((eval = ipcperm(p, &semaptr->sem_perm, IPC_R)))
489 			return(eval);
490 		if (semnum < 0 || semnum >= semaptr->sem_nsems)
491 			return(EINVAL);
492 		rval = semaptr->sem_base[semnum].semzcnt;
493 		break;
494 
495 	case SETVAL:
496 		if ((eval = ipcperm(p, &semaptr->sem_perm, IPC_W)))
497 			return(eval);
498 		if (semnum < 0 || semnum >= semaptr->sem_nsems)
499 			return(EINVAL);
500 		if ((eval = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
501 			return(eval);
502 		semaptr->sem_base[semnum].semval = real_arg.val;
503 		semundo_clear(semid, semnum);
504 		wakeup((caddr_t)semaptr);
505 		break;
506 
507 	case SETALL:
508 		if ((eval = ipcperm(p, &semaptr->sem_perm, IPC_W)))
509 			return(eval);
510 		if ((eval = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
511 			return(eval);
512 		for (i = 0; i < semaptr->sem_nsems; i++) {
513 			eval = copyin(&real_arg.array[i],
514 			    (caddr_t)&semaptr->sem_base[i].semval,
515 			    sizeof(real_arg.array[0]));
516 			if (eval != 0)
517 				break;
518 		}
519 		semundo_clear(semid, -1);
520 		wakeup((caddr_t)semaptr);
521 		break;
522 
523 	default:
524 		return(EINVAL);
525 	}
526 
527 	if (eval == 0)
528 		uap->sysmsg_result = rval;
529 	return(eval);
530 }
531 
532 int
533 sys_semget(struct semget_args *uap)
534 {
535 	struct proc *p = curproc;
536 	int semid, eval;
537 	int key = uap->key;
538 	int nsems = uap->nsems;
539 	int semflg = uap->semflg;
540 	struct ucred *cred = p->p_ucred;
541 
542 #ifdef SEM_DEBUG
543 	kprintf("semget(0x%x, %d, 0%o)\n", key, nsems, semflg);
544 #endif
545 
546 	if (!jail_sysvipc_allowed && p->p_ucred->cr_prison != NULL)
547 		return (ENOSYS);
548 
549 	if (key != IPC_PRIVATE) {
550 		for (semid = 0; semid < seminfo.semmni; semid++) {
551 			if ((sema[semid].sem_perm.mode & SEM_ALLOC) &&
552 			    sema[semid].sem_perm.key == key)
553 				break;
554 		}
555 		if (semid < seminfo.semmni) {
556 #ifdef SEM_DEBUG
557 			kprintf("found public key\n");
558 #endif
559 			if ((eval = ipcperm(p, &sema[semid].sem_perm,
560 			    semflg & 0700)))
561 				return(eval);
562 			if (nsems > 0 && sema[semid].sem_nsems < nsems) {
563 #ifdef SEM_DEBUG
564 				kprintf("too small\n");
565 #endif
566 				return(EINVAL);
567 			}
568 			if ((semflg & IPC_CREAT) && (semflg & IPC_EXCL)) {
569 #ifdef SEM_DEBUG
570 				kprintf("not exclusive\n");
571 #endif
572 				return(EEXIST);
573 			}
574 			goto found;
575 		}
576 	}
577 
578 #ifdef SEM_DEBUG
579 	kprintf("need to allocate the semid_ds\n");
580 #endif
581 	if (key == IPC_PRIVATE || (semflg & IPC_CREAT)) {
582 		if (nsems <= 0 || nsems > seminfo.semmsl) {
583 #ifdef SEM_DEBUG
584 			kprintf("nsems out of range (0<%d<=%d)\n", nsems,
585 			    seminfo.semmsl);
586 #endif
587 			return(EINVAL);
588 		}
589 		if (nsems > seminfo.semmns - semtot) {
590 #ifdef SEM_DEBUG
591 			kprintf("not enough semaphores left (need %d, got %d)\n",
592 			    nsems, seminfo.semmns - semtot);
593 #endif
594 			return(ENOSPC);
595 		}
596 		for (semid = 0; semid < seminfo.semmni; semid++) {
597 			if ((sema[semid].sem_perm.mode & SEM_ALLOC) == 0)
598 				break;
599 		}
600 		if (semid == seminfo.semmni) {
601 #ifdef SEM_DEBUG
602 			kprintf("no more semid_ds's available\n");
603 #endif
604 			return(ENOSPC);
605 		}
606 #ifdef SEM_DEBUG
607 		kprintf("semid %d is available\n", semid);
608 #endif
609 		sema[semid].sem_perm.key = key;
610 		sema[semid].sem_perm.cuid = cred->cr_uid;
611 		sema[semid].sem_perm.uid = cred->cr_uid;
612 		sema[semid].sem_perm.cgid = cred->cr_gid;
613 		sema[semid].sem_perm.gid = cred->cr_gid;
614 		sema[semid].sem_perm.mode = (semflg & 0777) | SEM_ALLOC;
615 		sema[semid].sem_perm.seq =
616 		    (sema[semid].sem_perm.seq + 1) & 0x7fff;
617 		sema[semid].sem_nsems = nsems;
618 		sema[semid].sem_otime = 0;
619 		sema[semid].sem_ctime = time_second;
620 		sema[semid].sem_base = &sem[semtot];
621 		semtot += nsems;
622 		bzero(sema[semid].sem_base,
623 		    sizeof(sema[semid].sem_base[0])*nsems);
624 #ifdef SEM_DEBUG
625 		kprintf("sembase = 0x%x, next = 0x%x\n", sema[semid].sem_base,
626 		    &sem[semtot]);
627 #endif
628 	} else {
629 #ifdef SEM_DEBUG
630 		kprintf("didn't find it and wasn't asked to create it\n");
631 #endif
632 		return(ENOENT);
633 	}
634 
635 found:
636 	uap->sysmsg_result = IXSEQ_TO_IPCID(semid, sema[semid].sem_perm);
637 	return(0);
638 }
639 
640 int
641 sys_semop(struct semop_args *uap)
642 {
643 	struct proc *p = curproc;
644 	int semid = uap->semid;
645 	u_int nsops = uap->nsops;
646 	struct sembuf sops[MAX_SOPS];
647 	struct semid_ds *semaptr;
648 	struct sembuf *sopptr;
649 	struct sem *semptr;
650 	struct sem_undo *suptr = NULL;
651 	int i, j, eval;
652 	int do_wakeup, do_undos;
653 
654 #ifdef SEM_DEBUG
655 	kprintf("call to semop(%d, 0x%x, %u)\n", semid, sops, nsops);
656 #endif
657 
658 	if (!jail_sysvipc_allowed && p->p_ucred->cr_prison != NULL)
659 		return (ENOSYS);
660 
661 	semid = IPCID_TO_IX(semid);	/* Convert back to zero origin */
662 
663 	if (semid < 0 || semid >= seminfo.semmni)
664 		return(EINVAL);
665 
666 	semaptr = &sema[semid];
667 	if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0)
668 		return(EINVAL);
669 	if (semaptr->sem_perm.seq != IPCID_TO_SEQ(uap->semid))
670 		return(EINVAL);
671 
672 	if ((eval = ipcperm(p, &semaptr->sem_perm, IPC_W))) {
673 #ifdef SEM_DEBUG
674 		kprintf("eval = %d from ipaccess\n", eval);
675 #endif
676 		return(eval);
677 	}
678 
679 	if (nsops > MAX_SOPS) {
680 #ifdef SEM_DEBUG
681 		kprintf("too many sops (max=%d, nsops=%u)\n", MAX_SOPS, nsops);
682 #endif
683 		return(E2BIG);
684 	}
685 
686 	if ((eval = copyin(uap->sops, &sops, nsops * sizeof(sops[0]))) != 0) {
687 #ifdef SEM_DEBUG
688 		kprintf("eval = %d from copyin(%08x, %08x, %u)\n", eval,
689 		    uap->sops, &sops, nsops * sizeof(sops[0]));
690 #endif
691 		return(eval);
692 	}
693 
694 	/*
695 	 * Loop trying to satisfy the vector of requests.
696 	 * If we reach a point where we must wait, any requests already
697 	 * performed are rolled back and we go to sleep until some other
698 	 * process wakes us up.  At this point, we start all over again.
699 	 *
700 	 * This ensures that from the perspective of other tasks, a set
701 	 * of requests is atomic (never partially satisfied).
702 	 */
703 	do_undos = 0;
704 
705 	for (;;) {
706 		do_wakeup = 0;
707 
708 		for (i = 0; i < nsops; i++) {
709 			sopptr = &sops[i];
710 
711 			if (sopptr->sem_num >= semaptr->sem_nsems)
712 				return(EFBIG);
713 
714 			semptr = &semaptr->sem_base[sopptr->sem_num];
715 
716 #ifdef SEM_DEBUG
717 			kprintf("semop:  semaptr=%x, sem_base=%x, semptr=%x, sem[%d]=%d : op=%d, flag=%s\n",
718 			    semaptr, semaptr->sem_base, semptr,
719 			    sopptr->sem_num, semptr->semval, sopptr->sem_op,
720 			    (sopptr->sem_flg & IPC_NOWAIT) ? "nowait" : "wait");
721 #endif
722 
723 			if (sopptr->sem_op < 0) {
724 				if (semptr->semval + sopptr->sem_op < 0) {
725 #ifdef SEM_DEBUG
726 					kprintf("semop:  can't do it now\n");
727 #endif
728 					break;
729 				} else {
730 					semptr->semval += sopptr->sem_op;
731 					if (semptr->semval == 0 &&
732 					    semptr->semzcnt > 0)
733 						do_wakeup = 1;
734 				}
735 				if (sopptr->sem_flg & SEM_UNDO)
736 					do_undos = 1;
737 			} else if (sopptr->sem_op == 0) {
738 				if (semptr->semval > 0) {
739 #ifdef SEM_DEBUG
740 					kprintf("semop:  not zero now\n");
741 #endif
742 					break;
743 				}
744 			} else {
745 				if (semptr->semncnt > 0)
746 					do_wakeup = 1;
747 				semptr->semval += sopptr->sem_op;
748 				if (sopptr->sem_flg & SEM_UNDO)
749 					do_undos = 1;
750 			}
751 		}
752 
753 		/*
754 		 * Did we get through the entire vector?
755 		 */
756 		if (i >= nsops)
757 			goto done;
758 
759 		/*
760 		 * No ... rollback anything that we've already done
761 		 */
762 #ifdef SEM_DEBUG
763 		kprintf("semop:  rollback 0 through %d\n", i-1);
764 #endif
765 		for (j = 0; j < i; j++)
766 			semaptr->sem_base[sops[j].sem_num].semval -=
767 			    sops[j].sem_op;
768 
769 		/*
770 		 * If the request that we couldn't satisfy has the
771 		 * NOWAIT flag set then return with EAGAIN.
772 		 */
773 		if (sopptr->sem_flg & IPC_NOWAIT)
774 			return(EAGAIN);
775 
776 		if (sopptr->sem_op == 0)
777 			semptr->semzcnt++;
778 		else
779 			semptr->semncnt++;
780 
781 #ifdef SEM_DEBUG
782 		kprintf("semop:  good night!\n");
783 #endif
784 		eval = tsleep((caddr_t)semaptr, PCATCH, "semwait", 0);
785 #ifdef SEM_DEBUG
786 		kprintf("semop:  good morning (eval=%d)!\n", eval);
787 #endif
788 
789 		suptr = NULL;	/* sem_undo may have been reallocated */
790 
791 		/* return code is checked below, after sem[nz]cnt-- */
792 
793 		/*
794 		 * Make sure that the semaphore still exists
795 		 */
796 		if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0 ||
797 		    semaptr->sem_perm.seq != IPCID_TO_SEQ(uap->semid))
798 			return(EIDRM);
799 
800 		/*
801 		 * The semaphore is still alive.  Readjust the count of
802 		 * waiting processes.
803 		 */
804 		if (sopptr->sem_op == 0)
805 			semptr->semzcnt--;
806 		else
807 			semptr->semncnt--;
808 
809 		/*
810 		 * Is it really morning, or was our sleep interrupted?
811 		 * (Delayed check of msleep() return code because we
812 		 * need to decrement sem[nz]cnt either way.)
813 		 */
814 		if (eval != 0)
815 			return(EINTR);
816 #ifdef SEM_DEBUG
817 		kprintf("semop:  good morning!\n");
818 #endif
819 	}
820 
821 done:
822 	/*
823 	 * Process any SEM_UNDO requests.
824 	 */
825 	if (do_undos) {
826 		for (i = 0; i < nsops; i++) {
827 			/*
828 			 * We only need to deal with SEM_UNDO's for non-zero
829 			 * op's.
830 			 */
831 			int adjval;
832 
833 			if ((sops[i].sem_flg & SEM_UNDO) == 0)
834 				continue;
835 			adjval = sops[i].sem_op;
836 			if (adjval == 0)
837 				continue;
838 			eval = semundo_adjust(p, &suptr, semid,
839 			    sops[i].sem_num, -adjval);
840 			if (eval == 0)
841 				continue;
842 
843 			/*
844 			 * Oh-Oh!  We ran out of either sem_undo's or undo's.
845 			 * Rollback the adjustments to this point and then
846 			 * rollback the semaphore ups and down so we can return
847 			 * with an error with all structures restored.  We
848 			 * rollback the undo's in the exact reverse order that
849 			 * we applied them.  This guarantees that we won't run
850 			 * out of space as we roll things back out.
851 			 */
852 			for (j = i - 1; j >= 0; j--) {
853 				if ((sops[j].sem_flg & SEM_UNDO) == 0)
854 					continue;
855 				adjval = sops[j].sem_op;
856 				if (adjval == 0)
857 					continue;
858 				if (semundo_adjust(p, &suptr, semid,
859 				    sops[j].sem_num, adjval) != 0)
860 					panic("semop - can't undo undos");
861 			}
862 
863 			for (j = 0; j < nsops; j++)
864 				semaptr->sem_base[sops[j].sem_num].semval -=
865 				    sops[j].sem_op;
866 
867 #ifdef SEM_DEBUG
868 			kprintf("eval = %d from semundo_adjust\n", eval);
869 #endif
870 			return(eval);
871 		} /* loop through the sops */
872 	} /* if (do_undos) */
873 
874 	/* We're definitely done - set the sempid's */
875 	for (i = 0; i < nsops; i++) {
876 		sopptr = &sops[i];
877 		semptr = &semaptr->sem_base[sopptr->sem_num];
878 		semptr->sempid = p->p_pid;
879 	}
880 
881 	/* Do a wakeup if any semaphore was up'd. */
882 	if (do_wakeup) {
883 #ifdef SEM_DEBUG
884 		kprintf("semop:  doing wakeup\n");
885 #endif
886 		wakeup((caddr_t)semaptr);
887 #ifdef SEM_DEBUG
888 		kprintf("semop:  back from wakeup\n");
889 #endif
890 	}
891 #ifdef SEM_DEBUG
892 	kprintf("semop:  done\n");
893 #endif
894 	uap->sysmsg_result = 0;
895 	return(0);
896 }
897 
898 /*
899  * Go through the undo structures for this process and apply the adjustments to
900  * semaphores.
901  */
902 void
903 semexit(struct proc *p)
904 {
905 	struct sem_undo *suptr;
906 	struct sem_undo **supptr;
907 	int did_something;
908 
909 	did_something = 0;
910 
911 	/*
912 	 * Go through the chain of undo vectors looking for one
913 	 * associated with this process.
914 	 */
915 
916 	for (supptr = &semu_list; (suptr = *supptr) != NULL;
917 	    supptr = &suptr->un_next) {
918 		if (suptr->un_proc == p)
919 			break;
920 	}
921 
922 	if (suptr == NULL)
923 		return;
924 
925 #ifdef SEM_DEBUG
926 	kprintf("proc @%08x has undo structure with %d entries\n", p,
927 	    suptr->un_cnt);
928 #endif
929 
930 	/*
931 	 * If there are any active undo elements then process them.
932 	 */
933 	if (suptr->un_cnt > 0) {
934 		int ix;
935 
936 		for (ix = 0; ix < suptr->un_cnt; ix++) {
937 			int semid = suptr->un_ent[ix].un_id;
938 			int semnum = suptr->un_ent[ix].un_num;
939 			int adjval = suptr->un_ent[ix].un_adjval;
940 			struct semid_ds *semaptr;
941 
942 			semaptr = &sema[semid];
943 			if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0)
944 				panic("semexit - semid not allocated");
945 			if (semnum >= semaptr->sem_nsems)
946 				panic("semexit - semnum out of range");
947 
948 #ifdef SEM_DEBUG
949 			kprintf("semexit:  %08x id=%d num=%d(adj=%d) ; sem=%d\n",
950 			    suptr->un_proc, suptr->un_ent[ix].un_id,
951 			    suptr->un_ent[ix].un_num,
952 			    suptr->un_ent[ix].un_adjval,
953 			    semaptr->sem_base[semnum].semval);
954 #endif
955 
956 			if (adjval < 0) {
957 				if (semaptr->sem_base[semnum].semval < -adjval)
958 					semaptr->sem_base[semnum].semval = 0;
959 				else
960 					semaptr->sem_base[semnum].semval +=
961 					    adjval;
962 			} else
963 				semaptr->sem_base[semnum].semval += adjval;
964 
965 			wakeup((caddr_t)semaptr);
966 #ifdef SEM_DEBUG
967 			kprintf("semexit:  back from wakeup\n");
968 #endif
969 		}
970 	}
971 
972 	/*
973 	 * Deallocate the undo vector.
974 	 */
975 #ifdef SEM_DEBUG
976 	kprintf("removing vector\n");
977 #endif
978 	suptr->un_proc = NULL;
979 	*supptr = suptr->un_next;
980 }
981