xref: /dragonfly/sys/kern/sysv_sem.c (revision 678e8cc6) 
1 /* $FreeBSD: src/sys/kern/sysv_sem.c,v 1.69 2004/03/17 09:37:13 cperciva Exp $ */
2 
3 /*
4  * Implementation of SVID semaphores
5  *
6  * Author:  Daniel Boulet
7  *
8  * This software is provided ``AS IS'' without any warranties of any kind.
9  */
10 
11 #include "opt_sysvipc.h"
12 
13 #include <sys/param.h>
14 #include <sys/systm.h>
15 #include <sys/sysproto.h>
16 #include <sys/kernel.h>
17 #include <sys/proc.h>
18 #include <sys/sem.h>
19 #include <sys/sysent.h>
20 #include <sys/sysctl.h>
21 #include <sys/malloc.h>
22 #include <sys/jail.h>
23 #include <sys/thread.h>
24 
25 #include <sys/thread2.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	22		/* # of semaphore identifiers */
74 #endif
75 #ifndef SEMMNS
76 #define SEMMNS	341		/* # 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 >= NELEM(semcalls))
214 		return (EINVAL);
215 	bcopy(&uap->a2, &uap->which,
216 	      sizeof(struct semsys_args) - offsetof(struct semsys_args, a2));
217 	error = (*semcalls[which])(uap);
218 	return (error);
219 }
220 
221 /*
222  * Allocate a new sem_undo structure for a process
223  * (returns ptr to structure or NULL if no more room)
224  */
225 
226 static struct sem_undo *
227 semu_alloc(struct proc *p)
228 {
229 	int i;
230 	struct sem_undo *suptr;
231 	struct sem_undo **supptr;
232 	int attempt;
233 
234 	/*
235 	 * Try twice to allocate something.
236 	 * (we'll purge any empty structures after the first pass so
237 	 * two passes are always enough)
238 	 */
239 
240 	for (attempt = 0; attempt < 2; attempt++) {
241 		/*
242 		 * Look for a free structure.
243 		 * Fill it in and return it if we find one.
244 		 */
245 
246 		for (i = 0; i < seminfo.semmnu; i++) {
247 			suptr = SEMU(i);
248 			if (suptr->un_proc == NULL) {
249 				suptr->un_next = semu_list;
250 				semu_list = suptr;
251 				suptr->un_cnt = 0;
252 				suptr->un_proc = p;
253 				return(suptr);
254 			}
255 		}
256 
257 		/*
258 		 * We didn't find a free one, if this is the first attempt
259 		 * then try to free some structures.
260 		 */
261 
262 		if (attempt == 0) {
263 			/* All the structures are in use - try to free some */
264 			int did_something = 0;
265 
266 			supptr = &semu_list;
267 			while ((suptr = *supptr) != NULL) {
268 				if (suptr->un_cnt == 0)  {
269 					suptr->un_proc = NULL;
270 					*supptr = suptr->un_next;
271 					did_something = 1;
272 				} else
273 					supptr = &(suptr->un_next);
274 			}
275 
276 			/* If we didn't free anything then just give-up */
277 			if (!did_something)
278 				return(NULL);
279 		} else {
280 			/*
281 			 * The second pass failed even though we freed
282 			 * something after the first pass!
283 			 * This is IMPOSSIBLE!
284 			 */
285 			panic("semu_alloc - second attempt failed");
286 		}
287 	}
288 	return (NULL);
289 }
290 
291 /*
292  * Adjust a particular entry for a particular proc
293  */
294 
295 static int
296 semundo_adjust(struct proc *p, struct sem_undo **supptr, int semid, int semnum,
297 	       int adjval)
298 {
299 	struct sem_undo *suptr;
300 	struct undo *sunptr;
301 	int i;
302 
303 	/* Look for and remember the sem_undo if the caller doesn't provide
304 	   it */
305 
306 	suptr = *supptr;
307 	if (suptr == NULL) {
308 		for (suptr = semu_list; suptr != NULL;
309 		    suptr = suptr->un_next) {
310 			if (suptr->un_proc == p) {
311 				*supptr = suptr;
312 				break;
313 			}
314 		}
315 		if (suptr == NULL) {
316 			if (adjval == 0)
317 				return(0);
318 			suptr = semu_alloc(p);
319 			if (suptr == NULL)
320 				return(ENOSPC);
321 			*supptr = suptr;
322 		}
323 	}
324 
325 	/*
326 	 * Look for the requested entry and adjust it (delete if adjval becomes
327 	 * 0).
328 	 */
329 	sunptr = &suptr->un_ent[0];
330 	for (i = 0; i < suptr->un_cnt; i++, sunptr++) {
331 		if (sunptr->un_id != semid || sunptr->un_num != semnum)
332 			continue;
333 		if (adjval == 0)
334 			sunptr->un_adjval = 0;
335 		else
336 			sunptr->un_adjval += adjval;
337 		if (sunptr->un_adjval == 0) {
338 			suptr->un_cnt--;
339 			if (i < suptr->un_cnt)
340 				suptr->un_ent[i] =
341 				    suptr->un_ent[suptr->un_cnt];
342 		}
343 		return(0);
344 	}
345 
346 	/* Didn't find the right entry - create it */
347 	if (adjval == 0)
348 		return(0);
349 	if (suptr->un_cnt != seminfo.semume) {
350 		sunptr = &suptr->un_ent[suptr->un_cnt];
351 		suptr->un_cnt++;
352 		sunptr->un_adjval = adjval;
353 		sunptr->un_id = semid; sunptr->un_num = semnum;
354 	} else
355 		return(EINVAL);
356 	return(0);
357 }
358 
359 static void
360 semundo_clear(int semid, int semnum)
361 {
362 	struct sem_undo *suptr;
363 
364 	for (suptr = semu_list; suptr != NULL; suptr = suptr->un_next) {
365 		struct undo *sunptr = &suptr->un_ent[0];
366 		int i = 0;
367 
368 		while (i < suptr->un_cnt) {
369 			if (sunptr->un_id == semid) {
370 				if (semnum == -1 || sunptr->un_num == semnum) {
371 					suptr->un_cnt--;
372 					if (i < suptr->un_cnt) {
373 						suptr->un_ent[i] =
374 						  suptr->un_ent[suptr->un_cnt];
375 						continue;
376 					}
377 				}
378 				if (semnum != -1)
379 					break;
380 			}
381 			i++, sunptr++;
382 		}
383 	}
384 }
385 
386 /*
387  * Note that the user-mode half of this passes a union, not a pointer
388  *
389  * MPALMOSTSAFE
390  */
391 int
392 sys___semctl(struct __semctl_args *uap)
393 {
394 	struct thread *td = curthread;
395 	int semid = uap->semid;
396 	int semnum = uap->semnum;
397 	int cmd = uap->cmd;
398 	union semun *arg = uap->arg;
399 	union semun real_arg;
400 	struct ucred *cred = td->td_ucred;
401 	int i, rval, eval;
402 	struct semid_ds sbuf;
403 	struct semid_ds *semaptr;
404 	struct semid_ds *semakptr;
405 
406 #ifdef SEM_DEBUG
407 	kprintf("call to semctl(%d, %d, %d, 0x%x)\n", semid, semnum, cmd, arg);
408 #endif
409 
410 	if (!jail_sysvipc_allowed && cred->cr_prison != NULL)
411 		return (ENOSYS);
412 
413 	switch (cmd) {
414 	case SEM_STAT:
415 		/*
416 		 * For this command we assume semid is an array index
417 		 * rather than an IPC id.
418 		 */
419 		if (semid < 0 || semid >= seminfo.semmni) {
420 			eval = EINVAL;
421 			break;
422 		}
423 		semakptr = &sema[semid];
424 		lwkt_getpooltoken(semakptr);
425 		if ((semakptr->sem_perm.mode & SEM_ALLOC) == 0) {
426 			eval = EINVAL;
427 			lwkt_relpooltoken(semakptr);
428 			break;
429 		}
430 		if ((eval = ipcperm(td->td_proc, &semakptr->sem_perm, IPC_R))) {
431 			lwkt_relpooltoken(semakptr);
432 			break;
433 		}
434 		bcopy(&semakptr, arg->buf, sizeof(struct semid_ds));
435 		rval = IXSEQ_TO_IPCID(semid, semakptr->sem_perm);
436 		lwkt_relpooltoken(semakptr);
437 		break;
438 	}
439 
440 	semid = IPCID_TO_IX(semid);
441 	if (semid < 0 || semid >= seminfo.semmni) {
442 		return(EINVAL);
443 	}
444 	semaptr = &sema[semid];
445 	lwkt_getpooltoken(semaptr);
446 
447 	if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0 ||
448 	    semaptr->sem_perm.seq != IPCID_TO_SEQ(uap->semid)) {
449 		lwkt_relpooltoken(semaptr);
450 		return(EINVAL);
451 	}
452 
453 	eval = 0;
454 	rval = 0;
455 
456 	switch (cmd) {
457 	case IPC_RMID:
458 		if ((eval = ipcperm(td->td_proc, &semaptr->sem_perm, IPC_M)) != 0)
459 			break;
460 		semaptr->sem_perm.cuid = cred->cr_uid;
461 		semaptr->sem_perm.uid = cred->cr_uid;
462 		semtot -= semaptr->sem_nsems;
463 		for (i = semaptr->sem_base - sem; i < semtot; i++)
464 			sem[i] = sem[i + semaptr->sem_nsems];
465 		for (i = 0; i < seminfo.semmni; i++) {
466 			if ((sema[i].sem_perm.mode & SEM_ALLOC) &&
467 			    sema[i].sem_base > semaptr->sem_base)
468 				sema[i].sem_base -= semaptr->sem_nsems;
469 		}
470 		semaptr->sem_perm.mode = 0;
471 		semundo_clear(semid, -1);
472 		wakeup((caddr_t)semaptr);
473 		break;
474 
475 	case IPC_SET:
476 		eval = ipcperm(td->td_proc, &semaptr->sem_perm, IPC_M);
477 		if (eval)
478 			break;
479 		if ((eval = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
480 			break;
481 		if ((eval = copyin(real_arg.buf, (caddr_t)&sbuf,
482 				   sizeof(sbuf))) != 0) {
483 			break;
484 		}
485 		semaptr->sem_perm.uid = sbuf.sem_perm.uid;
486 		semaptr->sem_perm.gid = sbuf.sem_perm.gid;
487 		semaptr->sem_perm.mode = (semaptr->sem_perm.mode & ~0777) |
488 					 (sbuf.sem_perm.mode & 0777);
489 		semaptr->sem_ctime = time_second;
490 		break;
491 
492 	case IPC_STAT:
493 		if ((eval = ipcperm(td->td_proc, &semaptr->sem_perm, IPC_R)))
494 			break;
495 		if ((eval = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
496 			break;
497 		eval = copyout(semaptr, real_arg.buf, sizeof(struct semid_ds));
498 		break;
499 
500 	case GETNCNT:
501 		eval = ipcperm(td->td_proc, &semaptr->sem_perm, IPC_R);
502 		if (eval)
503 			break;
504 		if (semnum < 0 || semnum >= semaptr->sem_nsems) {
505 			eval = EINVAL;
506 			break;
507 		}
508 		rval = semaptr->sem_base[semnum].semncnt;
509 		break;
510 
511 	case GETPID:
512 		eval = ipcperm(td->td_proc, &semaptr->sem_perm, IPC_R);
513 		if (eval)
514 			break;
515 		if (semnum < 0 || semnum >= semaptr->sem_nsems) {
516 			eval = EINVAL;
517 			break;
518 		}
519 		rval = semaptr->sem_base[semnum].sempid;
520 		break;
521 
522 	case GETVAL:
523 		eval = ipcperm(td->td_proc, &semaptr->sem_perm, IPC_R);
524 		if (eval)
525 			break;
526 		if (semnum < 0 || semnum >= semaptr->sem_nsems) {
527 			eval = EINVAL;
528 			break;
529 		}
530 		rval = semaptr->sem_base[semnum].semval;
531 		break;
532 
533 	case GETALL:
534 		eval = ipcperm(td->td_proc, &semaptr->sem_perm, IPC_R);
535 		if (eval)
536 			break;
537 		if ((eval = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
538 			break;
539 		for (i = 0; i < semaptr->sem_nsems; i++) {
540 			eval = copyout(&semaptr->sem_base[i].semval,
541 				       &real_arg.array[i],
542 				       sizeof(real_arg.array[0]));
543 			if (eval)
544 				break;
545 		}
546 		break;
547 
548 	case GETZCNT:
549 		eval = ipcperm(td->td_proc, &semaptr->sem_perm, IPC_R);
550 		if (eval)
551 			break;
552 		if (semnum < 0 || semnum >= semaptr->sem_nsems) {
553 			eval = EINVAL;
554 			break;
555 		}
556 		rval = semaptr->sem_base[semnum].semzcnt;
557 		break;
558 
559 	case SETVAL:
560 		eval = ipcperm(td->td_proc, &semaptr->sem_perm, IPC_W);
561 		if (eval)
562 			break;
563 		if (semnum < 0 || semnum >= semaptr->sem_nsems) {
564 			eval = EINVAL;
565 			break;
566 		}
567 		if ((eval = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
568 			break;
569 		semaptr->sem_base[semnum].semval = real_arg.val;
570 		semundo_clear(semid, semnum);
571 		wakeup((caddr_t)semaptr);
572 		break;
573 
574 	case SETALL:
575 		eval = ipcperm(td->td_proc, &semaptr->sem_perm, IPC_W);
576 		if (eval)
577 			break;
578 		if ((eval = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
579 			break;
580 		for (i = 0; i < semaptr->sem_nsems; i++) {
581 			eval = copyin(&real_arg.array[i],
582 				      (caddr_t)&semaptr->sem_base[i].semval,
583 				      sizeof(real_arg.array[0]));
584 			if (eval != 0)
585 				break;
586 		}
587 		semundo_clear(semid, -1);
588 		wakeup((caddr_t)semaptr);
589 		break;
590 
591 	default:
592 		eval = EINVAL;
593 		break;
594 	}
595 	lwkt_relpooltoken(semaptr);
596 
597 	if (eval == 0)
598 		uap->sysmsg_result = rval;
599 	return(eval);
600 }
601 
602 /*
603  * MPALMOSTSAFE
604  */
605 int
606 sys_semget(struct semget_args *uap)
607 {
608 	struct thread *td = curthread;
609 	int semid, eval;
610 	int key = uap->key;
611 	int nsems = uap->nsems;
612 	int semflg = uap->semflg;
613 	struct ucred *cred = td->td_ucred;
614 
615 #ifdef SEM_DEBUG
616 	kprintf("semget(0x%x, %d, 0%o)\n", key, nsems, semflg);
617 #endif
618 
619 	if (!jail_sysvipc_allowed && cred->cr_prison != NULL)
620 		return (ENOSYS);
621 
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) == 0 ||
627 			    sema[semid].sem_perm.key != key) {
628 				continue;
629 			}
630 			lwkt_getpooltoken(&sema[semid]);
631 			if ((sema[semid].sem_perm.mode & SEM_ALLOC) == 0 ||
632 			    sema[semid].sem_perm.key != key) {
633 				lwkt_relpooltoken(&sema[semid]);
634 				continue;
635 			}
636 			break;
637 		}
638 		if (semid < seminfo.semmni) {
639 #ifdef SEM_DEBUG
640 			kprintf("found public key\n");
641 #endif
642 			if ((eval = ipcperm(td->td_proc,
643 					    &sema[semid].sem_perm,
644 					    semflg & 0700))) {
645 				lwkt_relpooltoken(&sema[semid]);
646 				goto done;
647 			}
648 			if (nsems > 0 && sema[semid].sem_nsems < nsems) {
649 #ifdef SEM_DEBUG
650 				kprintf("too small\n");
651 #endif
652 				eval = EINVAL;
653 				lwkt_relpooltoken(&sema[semid]);
654 				goto done;
655 			}
656 			if ((semflg & IPC_CREAT) && (semflg & IPC_EXCL)) {
657 #ifdef SEM_DEBUG
658 				kprintf("not exclusive\n");
659 #endif
660 				eval = EEXIST;
661 				lwkt_relpooltoken(&sema[semid]);
662 				goto done;
663 			}
664 			goto done;
665 		}
666 	}
667 
668 #ifdef SEM_DEBUG
669 	kprintf("need to allocate the semid_ds\n");
670 #endif
671 	if (key == IPC_PRIVATE || (semflg & IPC_CREAT)) {
672 		if (nsems <= 0 || nsems > seminfo.semmsl) {
673 #ifdef SEM_DEBUG
674 			kprintf("nsems out of range (0<%d<=%d)\n",
675 				nsems, seminfo.semmsl);
676 #endif
677 			eval = EINVAL;
678 			goto done;
679 		}
680 		if (nsems > seminfo.semmns - semtot) {
681 #ifdef SEM_DEBUG
682 			kprintf("not enough semaphores left "
683 				"(need %d, got %d)\n",
684 				nsems, seminfo.semmns - semtot);
685 #endif
686 			eval = ENOSPC;
687 			goto done;
688 		}
689 		for (semid = 0; semid < seminfo.semmni; semid++) {
690 			if (sema[semid].sem_perm.mode & SEM_ALLOC)
691 				continue;
692 			lwkt_getpooltoken(&sema[semid]);
693 			if (sema[semid].sem_perm.mode & SEM_ALLOC) {
694 				lwkt_relpooltoken(&sema[semid]);
695 				continue;
696 			}
697 			break;
698 		}
699 		if (semid == seminfo.semmni) {
700 #ifdef SEM_DEBUG
701 			kprintf("no more semid_ds's available\n");
702 #endif
703 			eval = ENOSPC;
704 			goto done;
705 		}
706 #ifdef SEM_DEBUG
707 		kprintf("semid %d is available\n", semid);
708 #endif
709 		sema[semid].sem_perm.key = key;
710 		sema[semid].sem_perm.cuid = cred->cr_uid;
711 		sema[semid].sem_perm.uid = cred->cr_uid;
712 		sema[semid].sem_perm.cgid = cred->cr_gid;
713 		sema[semid].sem_perm.gid = cred->cr_gid;
714 		sema[semid].sem_perm.mode = (semflg & 0777) | SEM_ALLOC;
715 		sema[semid].sem_perm.seq =
716 		    (sema[semid].sem_perm.seq + 1) & 0x7fff;
717 		sema[semid].sem_nsems = nsems;
718 		sema[semid].sem_otime = 0;
719 		sema[semid].sem_ctime = time_second;
720 		sema[semid].sem_base = &sem[semtot];
721 		semtot += nsems;
722 		bzero(sema[semid].sem_base,
723 		    sizeof(sema[semid].sem_base[0])*nsems);
724 #ifdef SEM_DEBUG
725 		kprintf("sembase = 0x%x, next = 0x%x\n",
726 			sema[semid].sem_base, &sem[semtot]);
727 #endif
728 		/* eval == 0 */
729 	} else {
730 #ifdef SEM_DEBUG
731 		kprintf("didn't find it and wasn't asked to create it\n");
732 #endif
733 		eval = ENOENT;
734 	}
735 
736 done:
737 	if (eval == 0) {
738 		uap->sysmsg_result =
739 			IXSEQ_TO_IPCID(semid, sema[semid].sem_perm);
740 		lwkt_relpooltoken(&sema[semid]);
741 	}
742 	return(eval);
743 }
744 
745 /*
746  * MPALMOSTSAFE
747  */
748 int
749 sys_semop(struct semop_args *uap)
750 {
751 	struct thread *td = curthread;
752 	int semid = uap->semid;
753 	u_int nsops = uap->nsops;
754 	struct sembuf sops[MAX_SOPS];
755 	struct semid_ds *semaptr;
756 	struct sembuf *sopptr;
757 	struct sem *semptr;
758 	struct sem_undo *suptr = NULL;
759 	int i, j, eval;
760 	int do_wakeup, do_undos;
761 
762 #ifdef SEM_DEBUG
763 	kprintf("call to semop(%d, 0x%x, %u)\n", semid, sops, nsops);
764 #endif
765 
766 	if (!jail_sysvipc_allowed && td->td_ucred->cr_prison != NULL)
767 		return (ENOSYS);
768 
769 	semid = IPCID_TO_IX(semid);	/* Convert back to zero origin */
770 
771 	if (semid < 0 || semid >= seminfo.semmni) {
772 		eval = EINVAL;
773 		goto done2;
774 	}
775 	semaptr = &sema[semid];
776 	lwkt_getpooltoken(semaptr);
777 	if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0) {
778 		eval = EINVAL;
779 		goto done;
780 	}
781 	if (semaptr->sem_perm.seq != IPCID_TO_SEQ(uap->semid)) {
782 		eval = EINVAL;
783 		goto done;
784 	}
785 
786 	if ((eval = ipcperm(td->td_proc, &semaptr->sem_perm, IPC_W))) {
787 #ifdef SEM_DEBUG
788 		kprintf("eval = %d from ipaccess\n", eval);
789 #endif
790 		goto done;
791 	}
792 
793 	if (nsops > MAX_SOPS) {
794 #ifdef SEM_DEBUG
795 		kprintf("too many sops (max=%d, nsops=%u)\n", MAX_SOPS, nsops);
796 #endif
797 		eval = E2BIG;
798 		goto done;
799 	}
800 
801 	if ((eval = copyin(uap->sops, &sops, nsops * sizeof(sops[0]))) != 0) {
802 #ifdef SEM_DEBUG
803 		kprintf("eval = %d from copyin(%08x, %08x, %u)\n", eval,
804 		    uap->sops, &sops, nsops * sizeof(sops[0]));
805 #endif
806 		goto done;
807 	}
808 
809 	/*
810 	 * Loop trying to satisfy the vector of requests.
811 	 * If we reach a point where we must wait, any requests already
812 	 * performed are rolled back and we go to sleep until some other
813 	 * process wakes us up.  At this point, we start all over again.
814 	 *
815 	 * This ensures that from the perspective of other tasks, a set
816 	 * of requests is atomic (never partially satisfied).
817 	 */
818 	do_undos = 0;
819 
820 	for (;;) {
821 		do_wakeup = 0;
822 
823 		for (i = 0; i < nsops; i++) {
824 			sopptr = &sops[i];
825 
826 			if (sopptr->sem_num >= semaptr->sem_nsems) {
827 				eval = EFBIG;
828 				goto done;
829 			}
830 
831 			semptr = &semaptr->sem_base[sopptr->sem_num];
832 
833 #ifdef SEM_DEBUG
834 			kprintf("semop:  semaptr=%x, sem_base=%x, semptr=%x, sem[%d]=%d : op=%d, flag=%s\n",
835 			    semaptr, semaptr->sem_base, semptr,
836 			    sopptr->sem_num, semptr->semval, sopptr->sem_op,
837 			    (sopptr->sem_flg & IPC_NOWAIT) ? "nowait" : "wait");
838 #endif
839 
840 			if (sopptr->sem_op < 0) {
841 				if (semptr->semval + sopptr->sem_op < 0) {
842 #ifdef SEM_DEBUG
843 					kprintf("semop:  can't do it now\n");
844 #endif
845 					break;
846 				} else {
847 					semptr->semval += sopptr->sem_op;
848 					if (semptr->semval == 0 &&
849 					    semptr->semzcnt > 0)
850 						do_wakeup = 1;
851 				}
852 				if (sopptr->sem_flg & SEM_UNDO)
853 					do_undos = 1;
854 			} else if (sopptr->sem_op == 0) {
855 				if (semptr->semval > 0) {
856 #ifdef SEM_DEBUG
857 					kprintf("semop:  not zero now\n");
858 #endif
859 					break;
860 				}
861 			} else {
862 				if (semptr->semncnt > 0)
863 					do_wakeup = 1;
864 				semptr->semval += sopptr->sem_op;
865 				if (sopptr->sem_flg & SEM_UNDO)
866 					do_undos = 1;
867 			}
868 		}
869 
870 		/*
871 		 * Did we get through the entire vector?
872 		 */
873 		if (i >= nsops)
874 			goto donex;
875 
876 		/*
877 		 * No ... rollback anything that we've already done
878 		 */
879 #ifdef SEM_DEBUG
880 		kprintf("semop:  rollback 0 through %d\n", i-1);
881 #endif
882 		for (j = 0; j < i; j++)
883 			semaptr->sem_base[sops[j].sem_num].semval -=
884 			    sops[j].sem_op;
885 
886 		/*
887 		 * If the request that we couldn't satisfy has the
888 		 * NOWAIT flag set then return with EAGAIN.
889 		 */
890 		if (sopptr->sem_flg & IPC_NOWAIT) {
891 			eval = EAGAIN;
892 			goto done;
893 		}
894 
895 		if (sopptr->sem_op == 0)
896 			semptr->semzcnt++;
897 		else
898 			semptr->semncnt++;
899 
900 #ifdef SEM_DEBUG
901 		kprintf("semop:  good night!\n");
902 #endif
903 		eval = tsleep((caddr_t)semaptr, PCATCH, "semwait", 0);
904 #ifdef SEM_DEBUG
905 		kprintf("semop:  good morning (eval=%d)!\n", eval);
906 #endif
907 
908 		suptr = NULL;	/* sem_undo may have been reallocated */
909 
910 		/* return code is checked below, after sem[nz]cnt-- */
911 
912 		/*
913 		 * Make sure that the semaphore still exists
914 		 */
915 		if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0 ||
916 		    semaptr->sem_perm.seq != IPCID_TO_SEQ(uap->semid)) {
917 			eval = EIDRM;
918 			goto done;
919 		}
920 
921 		/*
922 		 * The semaphore is still alive.  Readjust the count of
923 		 * waiting processes.
924 		 */
925 		if (sopptr->sem_op == 0)
926 			semptr->semzcnt--;
927 		else
928 			semptr->semncnt--;
929 
930 		/*
931 		 * Is it really morning, or was our sleep interrupted?
932 		 * (Delayed check of tsleep() return code because we
933 		 * need to decrement sem[nz]cnt either way.)
934 		 */
935 		if (eval) {
936 			eval = EINTR;
937 			goto done;
938 		}
939 #ifdef SEM_DEBUG
940 		kprintf("semop:  good morning!\n");
941 #endif
942 	}
943 
944 donex:
945 	/*
946 	 * Process any SEM_UNDO requests.
947 	 */
948 	if (do_undos) {
949 		for (i = 0; i < nsops; i++) {
950 			/*
951 			 * We only need to deal with SEM_UNDO's for non-zero
952 			 * op's.
953 			 */
954 			int adjval;
955 
956 			if ((sops[i].sem_flg & SEM_UNDO) == 0)
957 				continue;
958 			adjval = sops[i].sem_op;
959 			if (adjval == 0)
960 				continue;
961 			eval = semundo_adjust(td->td_proc, &suptr, semid,
962 					      sops[i].sem_num, -adjval);
963 			if (eval == 0)
964 				continue;
965 
966 			/*
967 			 * Oh-Oh!  We ran out of either sem_undo's or undo's.
968 			 * Rollback the adjustments to this point and then
969 			 * rollback the semaphore ups and down so we can return
970 			 * with an error with all structures restored.  We
971 			 * rollback the undo's in the exact reverse order that
972 			 * we applied them.  This guarantees that we won't run
973 			 * out of space as we roll things back out.
974 			 */
975 			for (j = i - 1; j >= 0; j--) {
976 				if ((sops[j].sem_flg & SEM_UNDO) == 0)
977 					continue;
978 				adjval = sops[j].sem_op;
979 				if (adjval == 0)
980 					continue;
981 				if (semundo_adjust(td->td_proc, &suptr, semid,
982 					       sops[j].sem_num, adjval) != 0)
983 					panic("semop - can't undo undos");
984 			}
985 
986 			for (j = 0; j < nsops; j++)
987 				semaptr->sem_base[sops[j].sem_num].semval -=
988 				    sops[j].sem_op;
989 
990 #ifdef SEM_DEBUG
991 			kprintf("eval = %d from semundo_adjust\n", eval);
992 #endif
993 			goto done;
994 		} /* loop through the sops */
995 	} /* if (do_undos) */
996 
997 	/* We're definitely done - set the sempid's */
998 	for (i = 0; i < nsops; i++) {
999 		sopptr = &sops[i];
1000 		semptr = &semaptr->sem_base[sopptr->sem_num];
1001 		semptr->sempid = td->td_proc->p_pid;
1002 	}
1003 
1004 	/* Do a wakeup if any semaphore was up'd. */
1005 	if (do_wakeup) {
1006 #ifdef SEM_DEBUG
1007 		kprintf("semop:  doing wakeup\n");
1008 #endif
1009 		wakeup((caddr_t)semaptr);
1010 #ifdef SEM_DEBUG
1011 		kprintf("semop:  back from wakeup\n");
1012 #endif
1013 	}
1014 #ifdef SEM_DEBUG
1015 	kprintf("semop:  done\n");
1016 #endif
1017 	uap->sysmsg_result = 0;
1018 	eval = 0;
1019 done:
1020 	lwkt_relpooltoken(semaptr);
1021 done2:
1022 	return(eval);
1023 }
1024 
1025 /*
1026  * Go through the undo structures for this process and apply the adjustments to
1027  * semaphores.
1028  */
1029 void
1030 semexit(struct proc *p)
1031 {
1032 	struct sem_undo *suptr;
1033 	struct sem_undo **supptr;
1034 	int did_something;
1035 
1036 	did_something = 0;
1037 
1038 	/*
1039 	 * Go through the chain of undo vectors looking for one
1040 	 * associated with this process.
1041 	 */
1042 
1043 	for (supptr = &semu_list; (suptr = *supptr) != NULL;
1044 	    supptr = &suptr->un_next) {
1045 		if (suptr->un_proc == p)
1046 			break;
1047 	}
1048 
1049 	if (suptr == NULL)
1050 		return;
1051 
1052 #ifdef SEM_DEBUG
1053 	kprintf("proc @%08x has undo structure with %d entries\n", p,
1054 	    suptr->un_cnt);
1055 #endif
1056 
1057 	/*
1058 	 * If there are any active undo elements then process them.
1059 	 */
1060 	if (suptr->un_cnt > 0) {
1061 		int ix;
1062 
1063 		for (ix = 0; ix < suptr->un_cnt; ix++) {
1064 			int semid = suptr->un_ent[ix].un_id;
1065 			int semnum = suptr->un_ent[ix].un_num;
1066 			int adjval = suptr->un_ent[ix].un_adjval;
1067 			struct semid_ds *semaptr;
1068 
1069 			semaptr = &sema[semid];
1070 			if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0)
1071 				panic("semexit - semid not allocated");
1072 			if (semnum >= semaptr->sem_nsems)
1073 				panic("semexit - semnum out of range");
1074 
1075 #ifdef SEM_DEBUG
1076 			kprintf("semexit:  %08x id=%d num=%d(adj=%d) ; sem=%d\n",
1077 			    suptr->un_proc, suptr->un_ent[ix].un_id,
1078 			    suptr->un_ent[ix].un_num,
1079 			    suptr->un_ent[ix].un_adjval,
1080 			    semaptr->sem_base[semnum].semval);
1081 #endif
1082 
1083 			if (adjval < 0) {
1084 				if (semaptr->sem_base[semnum].semval < -adjval)
1085 					semaptr->sem_base[semnum].semval = 0;
1086 				else
1087 					semaptr->sem_base[semnum].semval +=
1088 					    adjval;
1089 			} else
1090 				semaptr->sem_base[semnum].semval += adjval;
1091 
1092 			wakeup((caddr_t)semaptr);
1093 #ifdef SEM_DEBUG
1094 			kprintf("semexit:  back from wakeup\n");
1095 #endif
1096 		}
1097 	}
1098 
1099 	/*
1100 	 * Deallocate the undo vector.
1101 	 */
1102 #ifdef SEM_DEBUG
1103 	kprintf("removing vector\n");
1104 #endif
1105 	suptr->un_proc = NULL;
1106 	*supptr = suptr->un_next;
1107 }
1108