xref: /netbsd-src/sys/kern/sysv_sem.c (revision bdc22b2e01993381dcefeff2bc9b56ca75a4235c)
1 /*	$NetBSD: sysv_sem.c,v 1.95 2015/11/06 02:26:42 pgoyette Exp $	*/
2 
3 /*-
4  * Copyright (c) 1999, 2007 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Jason R. Thorpe of the Numerical Aerospace Simulation Facility,
9  * NASA Ames Research Center, and by Andrew Doran.
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions and the following disclaimer.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
21  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
22  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
23  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
24  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
25  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
26  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
27  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
28  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
29  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
30  * POSSIBILITY OF SUCH DAMAGE.
31  */
32 
33 /*
34  * Implementation of SVID semaphores
35  *
36  * Author: Daniel Boulet
37  *
38  * This software is provided ``AS IS'' without any warranties of any kind.
39  */
40 
41 #include <sys/cdefs.h>
42 __KERNEL_RCSID(0, "$NetBSD: sysv_sem.c,v 1.95 2015/11/06 02:26:42 pgoyette Exp $");
43 
44 #ifdef _KERNEL_OPT
45 #include "opt_sysv.h"
46 #endif
47 
48 #include <sys/param.h>
49 #include <sys/kernel.h>
50 #include <sys/sem.h>
51 #include <sys/sysctl.h>
52 #include <sys/kmem.h>
53 #include <sys/mount.h>		/* XXX for <sys/syscallargs.h> */
54 #include <sys/syscallargs.h>
55 #include <sys/kauth.h>
56 #include <sys/once.h>
57 
58 /*
59  * Memory areas:
60  *  1st: Pool of semaphore identifiers
61  *  2nd: Semaphores
62  *  3rd: Conditional variables
63  *  4th: Undo structures
64  */
65 struct semid_ds *	sema			__read_mostly;
66 static struct __sem *	sem			__read_mostly;
67 static kcondvar_t *	semcv			__read_mostly;
68 static int *		semu			__read_mostly;
69 
70 static kmutex_t		semlock			__cacheline_aligned;
71 static bool		sem_realloc_state	__read_mostly;
72 static kcondvar_t	sem_realloc_cv;
73 
74 /*
75  * List of active undo structures, total number of semaphores,
76  * and total number of semop waiters.
77  */
78 static struct sem_undo *semu_list		__read_mostly;
79 static u_int		semtot			__cacheline_aligned;
80 static u_int		sem_waiters		__cacheline_aligned;
81 
82 /* Macro to find a particular sem_undo vector */
83 #define SEMU(s, ix)	((struct sem_undo *)(((long)s) + ix * seminfo.semusz))
84 
85 #ifdef SEM_DEBUG
86 #define SEM_PRINTF(a) printf a
87 #else
88 #define SEM_PRINTF(a)
89 #endif
90 
91 void *hook;	/* cookie from exithook_establish() */
92 
93 extern int kern_has_sysvsem;
94 
95 SYSCTL_SETUP_PROTO(sysctl_ipc_sem_setup);
96 
97 struct sem_undo *semu_alloc(struct proc *);
98 int semundo_adjust(struct proc *, struct sem_undo **, int, int, int);
99 void semundo_clear(int, int);
100 
101 static ONCE_DECL(exithook_control);
102 static int seminit_exithook(void);
103 
104 void
105 seminit(struct sysctllog **clog)
106 {
107 	int i, sz;
108 	vaddr_t v;
109 
110 	mutex_init(&semlock, MUTEX_DEFAULT, IPL_NONE);
111 	cv_init(&sem_realloc_cv, "semrealc");
112 	sem_realloc_state = false;
113 	semtot = 0;
114 	sem_waiters = 0;
115 
116 	/* Allocate the wired memory for our structures */
117 	sz = ALIGN(seminfo.semmni * sizeof(struct semid_ds)) +
118 	    ALIGN(seminfo.semmns * sizeof(struct __sem)) +
119 	    ALIGN(seminfo.semmni * sizeof(kcondvar_t)) +
120 	    ALIGN(seminfo.semmnu * seminfo.semusz);
121 	sz = round_page(sz);
122 	v = uvm_km_alloc(kernel_map, sz, 0, UVM_KMF_WIRED|UVM_KMF_ZERO);
123 	if (v == 0)
124 		panic("sysv_sem: cannot allocate memory");
125 	sema = (void *)v;
126 	sem = (void *)((uintptr_t)sema +
127 	    ALIGN(seminfo.semmni * sizeof(struct semid_ds)));
128 	semcv = (void *)((uintptr_t)sem +
129 	    ALIGN(seminfo.semmns * sizeof(struct __sem)));
130 	semu = (void *)((uintptr_t)semcv +
131 	    ALIGN(seminfo.semmni * sizeof(kcondvar_t)));
132 
133 	for (i = 0; i < seminfo.semmni; i++) {
134 		sema[i]._sem_base = 0;
135 		sema[i].sem_perm.mode = 0;
136 		cv_init(&semcv[i], "semwait");
137 	}
138 	for (i = 0; i < seminfo.semmnu; i++) {
139 		struct sem_undo *suptr = SEMU(semu, i);
140 		suptr->un_proc = NULL;
141 	}
142 	semu_list = NULL;
143 
144 	kern_has_sysvsem = 1;
145 
146 #ifdef _MODULE
147 	if (clog)
148 		sysctl_ipc_sem_setup(clog);
149 #endif
150 }
151 
152 static int
153 seminit_exithook(void)
154 {
155 
156 	hook = exithook_establish(semexit, NULL);
157 	return 0;
158 }
159 
160 int
161 semfini(void)
162 {
163 	int i, sz;
164 	vaddr_t v = (vaddr_t)sema;
165 
166 	/* Don't allow module unload if we're busy */
167 	mutex_enter(&semlock);
168 	if (semtot) {
169 		mutex_exit(&semlock);
170 		return 1;
171 	}
172 
173 	/* Remove the exit hook */
174 	if (hook)
175 		exithook_disestablish(hook);
176 
177 	/* Destroy all our condvars */
178 	for (i = 0; i < seminfo.semmni; i++) {
179 		cv_destroy(&semcv[i]);
180 	}
181 
182 	/* Free the wired memory that we allocated */
183 	sz = ALIGN(seminfo.semmni * sizeof(struct semid_ds)) +
184 	    ALIGN(seminfo.semmns * sizeof(struct __sem)) +
185 	    ALIGN(seminfo.semmni * sizeof(kcondvar_t)) +
186 	    ALIGN(seminfo.semmnu * seminfo.semusz);
187 	sz = round_page(sz);
188 	uvm_km_free(kernel_map, v, sz, UVM_KMF_WIRED);
189 
190 	/* Destroy the last cv and mutex */
191 	cv_destroy(&sem_realloc_cv);
192 	mutex_exit(&semlock);
193 	mutex_destroy(&semlock);
194 
195 	kern_has_sysvsem = 0;
196 
197 	return 0;
198 }
199 
200 static int
201 semrealloc(int newsemmni, int newsemmns, int newsemmnu)
202 {
203 	struct semid_ds *new_sema, *old_sema;
204 	struct __sem *new_sem;
205 	struct sem_undo *new_semu_list, *suptr, *nsuptr;
206 	int *new_semu;
207 	kcondvar_t *new_semcv;
208 	vaddr_t v;
209 	int i, j, lsemid, nmnus, sz;
210 
211 	if (newsemmni < 1 || newsemmns < 1 || newsemmnu < 1)
212 		return EINVAL;
213 
214 	/* Allocate the wired memory for our structures */
215 	sz = ALIGN(newsemmni * sizeof(struct semid_ds)) +
216 	    ALIGN(newsemmns * sizeof(struct __sem)) +
217 	    ALIGN(newsemmni * sizeof(kcondvar_t)) +
218 	    ALIGN(newsemmnu * seminfo.semusz);
219 	sz = round_page(sz);
220 	v = uvm_km_alloc(kernel_map, sz, 0, UVM_KMF_WIRED|UVM_KMF_ZERO);
221 	if (v == 0)
222 		return ENOMEM;
223 
224 	mutex_enter(&semlock);
225 	if (sem_realloc_state) {
226 		mutex_exit(&semlock);
227 		uvm_km_free(kernel_map, v, sz, UVM_KMF_WIRED);
228 		return EBUSY;
229 	}
230 	sem_realloc_state = true;
231 	if (sem_waiters) {
232 		/*
233 		 * Mark reallocation state, wake-up all waiters,
234 		 * and wait while they will all exit.
235 		 */
236 		for (i = 0; i < seminfo.semmni; i++)
237 			cv_broadcast(&semcv[i]);
238 		while (sem_waiters)
239 			cv_wait(&sem_realloc_cv, &semlock);
240 	}
241 	old_sema = sema;
242 
243 	/* Get the number of last slot */
244 	lsemid = 0;
245 	for (i = 0; i < seminfo.semmni; i++)
246 		if (sema[i].sem_perm.mode & SEM_ALLOC)
247 			lsemid = i;
248 
249 	/* Get the number of currently used undo structures */
250 	nmnus = 0;
251 	for (i = 0; i < seminfo.semmnu; i++) {
252 		suptr = SEMU(semu, i);
253 		if (suptr->un_proc == NULL)
254 			continue;
255 		nmnus++;
256 	}
257 
258 	/* We cannot reallocate less memory than we use */
259 	if (lsemid >= newsemmni || semtot > newsemmns || nmnus > newsemmnu) {
260 		mutex_exit(&semlock);
261 		uvm_km_free(kernel_map, v, sz, UVM_KMF_WIRED);
262 		return EBUSY;
263 	}
264 
265 	new_sema = (void *)v;
266 	new_sem = (void *)((uintptr_t)new_sema +
267 	    ALIGN(newsemmni * sizeof(struct semid_ds)));
268 	new_semcv = (void *)((uintptr_t)new_sem +
269 	    ALIGN(newsemmns * sizeof(struct __sem)));
270 	new_semu = (void *)((uintptr_t)new_semcv +
271 	    ALIGN(newsemmni * sizeof(kcondvar_t)));
272 
273 	/* Initialize all semaphore identifiers and condvars */
274 	for (i = 0; i < newsemmni; i++) {
275 		new_sema[i]._sem_base = 0;
276 		new_sema[i].sem_perm.mode = 0;
277 		cv_init(&new_semcv[i], "semwait");
278 	}
279 	for (i = 0; i < newsemmnu; i++) {
280 		nsuptr = SEMU(new_semu, i);
281 		nsuptr->un_proc = NULL;
282 	}
283 
284 	/*
285 	 * Copy all identifiers, semaphores and list of the
286 	 * undo structures to the new memory allocation.
287 	 */
288 	j = 0;
289 	for (i = 0; i <= lsemid; i++) {
290 		if ((sema[i].sem_perm.mode & SEM_ALLOC) == 0)
291 			continue;
292 		memcpy(&new_sema[i], &sema[i], sizeof(struct semid_ds));
293 		new_sema[i]._sem_base = &new_sem[j];
294 		memcpy(new_sema[i]._sem_base, sema[i]._sem_base,
295 		    (sizeof(struct __sem) * sema[i].sem_nsems));
296 		j += sema[i].sem_nsems;
297 	}
298 	KASSERT(j == semtot);
299 
300 	j = 0;
301 	new_semu_list = NULL;
302 	for (suptr = semu_list; suptr != NULL; suptr = suptr->un_next) {
303 		KASSERT(j < newsemmnu);
304 		nsuptr = SEMU(new_semu, j);
305 		memcpy(nsuptr, suptr, SEMUSZ);
306 		nsuptr->un_next = new_semu_list;
307 		new_semu_list = nsuptr;
308 		j++;
309 	}
310 
311 	for (i = 0; i < seminfo.semmni; i++) {
312 		KASSERT(cv_has_waiters(&semcv[i]) == false);
313 		cv_destroy(&semcv[i]);
314 	}
315 
316 	sz = ALIGN(seminfo.semmni * sizeof(struct semid_ds)) +
317 	    ALIGN(seminfo.semmns * sizeof(struct __sem)) +
318 	    ALIGN(seminfo.semmni * sizeof(kcondvar_t)) +
319 	    ALIGN(seminfo.semmnu * seminfo.semusz);
320 	sz = round_page(sz);
321 
322 	/* Set the pointers and update the new values */
323 	sema = new_sema;
324 	sem = new_sem;
325 	semcv = new_semcv;
326 	semu = new_semu;
327 	semu_list = new_semu_list;
328 
329 	seminfo.semmni = newsemmni;
330 	seminfo.semmns = newsemmns;
331 	seminfo.semmnu = newsemmnu;
332 
333 	/* Reallocation completed - notify all waiters, if any */
334 	sem_realloc_state = false;
335 	cv_broadcast(&sem_realloc_cv);
336 	mutex_exit(&semlock);
337 
338 	uvm_km_free(kernel_map, (vaddr_t)old_sema, sz, UVM_KMF_WIRED);
339 	return 0;
340 }
341 
342 /*
343  * Placebo.
344  */
345 
346 int
347 sys_semconfig(struct lwp *l, const struct sys_semconfig_args *uap, register_t *retval)
348 {
349 
350 	RUN_ONCE(&exithook_control, seminit_exithook);
351 
352 	*retval = 0;
353 	return 0;
354 }
355 
356 /*
357  * Allocate a new sem_undo structure for a process.
358  * => Returns NULL on failure.
359  */
360 struct sem_undo *
361 semu_alloc(struct proc *p)
362 {
363 	struct sem_undo *suptr, **supptr;
364 	bool attempted = false;
365 	int i;
366 
367 	KASSERT(mutex_owned(&semlock));
368 again:
369 	/* Look for a free structure. */
370 	for (i = 0; i < seminfo.semmnu; i++) {
371 		suptr = SEMU(semu, i);
372 		if (suptr->un_proc == NULL) {
373 			/* Found.  Fill it in and return. */
374 			suptr->un_next = semu_list;
375 			semu_list = suptr;
376 			suptr->un_cnt = 0;
377 			suptr->un_proc = p;
378 			return suptr;
379 		}
380 	}
381 
382 	/* Not found.  Attempt to free some structures. */
383 	if (!attempted) {
384 		bool freed = false;
385 
386 		attempted = true;
387 		supptr = &semu_list;
388 		while ((suptr = *supptr) != NULL) {
389 			if (suptr->un_cnt == 0)  {
390 				suptr->un_proc = NULL;
391 				*supptr = suptr->un_next;
392 				freed = true;
393 			} else {
394 				supptr = &suptr->un_next;
395 			}
396 		}
397 		if (freed) {
398 			goto again;
399 		}
400 	}
401 	return NULL;
402 }
403 
404 /*
405  * Adjust a particular entry for a particular proc
406  */
407 
408 int
409 semundo_adjust(struct proc *p, struct sem_undo **supptr, int semid, int semnum,
410     int adjval)
411 {
412 	struct sem_undo *suptr;
413 	struct sem_undo_entry *sunptr;
414 	int i;
415 
416 	KASSERT(mutex_owned(&semlock));
417 
418 	/*
419 	 * Look for and remember the sem_undo if the caller doesn't
420 	 * provide it
421 	 */
422 
423 	suptr = *supptr;
424 	if (suptr == NULL) {
425 		for (suptr = semu_list; suptr != NULL; suptr = suptr->un_next)
426 			if (suptr->un_proc == p)
427 				break;
428 
429 		if (suptr == NULL) {
430 			suptr = semu_alloc(p);
431 			if (suptr == NULL)
432 				return (ENOSPC);
433 		}
434 		*supptr = suptr;
435 	}
436 
437 	/*
438 	 * Look for the requested entry and adjust it (delete if
439 	 * adjval becomes 0).
440 	 */
441 	sunptr = &suptr->un_ent[0];
442 	for (i = 0; i < suptr->un_cnt; i++, sunptr++) {
443 		if (sunptr->un_id != semid || sunptr->un_num != semnum)
444 			continue;
445 		sunptr->un_adjval += adjval;
446 		if (sunptr->un_adjval == 0) {
447 			suptr->un_cnt--;
448 			if (i < suptr->un_cnt)
449 				suptr->un_ent[i] =
450 				    suptr->un_ent[suptr->un_cnt];
451 		}
452 		return (0);
453 	}
454 
455 	/* Didn't find the right entry - create it */
456 	if (suptr->un_cnt == SEMUME)
457 		return (EINVAL);
458 
459 	sunptr = &suptr->un_ent[suptr->un_cnt];
460 	suptr->un_cnt++;
461 	sunptr->un_adjval = adjval;
462 	sunptr->un_id = semid;
463 	sunptr->un_num = semnum;
464 	return (0);
465 }
466 
467 void
468 semundo_clear(int semid, int semnum)
469 {
470 	struct sem_undo *suptr;
471 	struct sem_undo_entry *sunptr, *sunend;
472 
473 	KASSERT(mutex_owned(&semlock));
474 
475 	for (suptr = semu_list; suptr != NULL; suptr = suptr->un_next)
476 		for (sunptr = &suptr->un_ent[0],
477 		    sunend = sunptr + suptr->un_cnt; sunptr < sunend;) {
478 			if (sunptr->un_id == semid) {
479 				if (semnum == -1 || sunptr->un_num == semnum) {
480 					suptr->un_cnt--;
481 					sunend--;
482 					if (sunptr != sunend)
483 						*sunptr = *sunend;
484 					if (semnum != -1)
485 						break;
486 					else
487 						continue;
488 				}
489 			}
490 			sunptr++;
491 		}
492 }
493 
494 int
495 sys_____semctl50(struct lwp *l, const struct sys_____semctl50_args *uap,
496     register_t *retval)
497 {
498 	/* {
499 		syscallarg(int) semid;
500 		syscallarg(int) semnum;
501 		syscallarg(int) cmd;
502 		syscallarg(union __semun *) arg;
503 	} */
504 	struct semid_ds sembuf;
505 	int cmd, error;
506 	void *pass_arg;
507 	union __semun karg;
508 
509 	RUN_ONCE(&exithook_control, seminit_exithook);
510 
511 	cmd = SCARG(uap, cmd);
512 
513 	pass_arg = get_semctl_arg(cmd, &sembuf, &karg);
514 
515 	if (pass_arg) {
516 		error = copyin(SCARG(uap, arg), &karg, sizeof(karg));
517 		if (error)
518 			return error;
519 		if (cmd == IPC_SET) {
520 			error = copyin(karg.buf, &sembuf, sizeof(sembuf));
521 			if (error)
522 				return (error);
523 		}
524 	}
525 
526 	error = semctl1(l, SCARG(uap, semid), SCARG(uap, semnum), cmd,
527 	    pass_arg, retval);
528 
529 	if (error == 0 && cmd == IPC_STAT)
530 		error = copyout(&sembuf, karg.buf, sizeof(sembuf));
531 
532 	return (error);
533 }
534 
535 int
536 semctl1(struct lwp *l, int semid, int semnum, int cmd, void *v,
537     register_t *retval)
538 {
539 	kauth_cred_t cred = l->l_cred;
540 	union __semun *arg = v;
541 	struct semid_ds *sembuf = v, *semaptr;
542 	int i, error, ix;
543 
544 	SEM_PRINTF(("call to semctl(%d, %d, %d, %p)\n",
545 	    semid, semnum, cmd, v));
546 
547 	mutex_enter(&semlock);
548 
549 	ix = IPCID_TO_IX(semid);
550 	if (ix < 0 || ix >= seminfo.semmni) {
551 		mutex_exit(&semlock);
552 		return (EINVAL);
553 	}
554 
555 	semaptr = &sema[ix];
556 	if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0 ||
557 	    semaptr->sem_perm._seq != IPCID_TO_SEQ(semid)) {
558 		mutex_exit(&semlock);
559 		return (EINVAL);
560 	}
561 
562 	switch (cmd) {
563 	case IPC_RMID:
564 		if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_M)) != 0)
565 			break;
566 		semaptr->sem_perm.cuid = kauth_cred_geteuid(cred);
567 		semaptr->sem_perm.uid = kauth_cred_geteuid(cred);
568 		semtot -= semaptr->sem_nsems;
569 		for (i = semaptr->_sem_base - sem; i < semtot; i++)
570 			sem[i] = sem[i + semaptr->sem_nsems];
571 		for (i = 0; i < seminfo.semmni; i++) {
572 			if ((sema[i].sem_perm.mode & SEM_ALLOC) &&
573 			    sema[i]._sem_base > semaptr->_sem_base)
574 				sema[i]._sem_base -= semaptr->sem_nsems;
575 		}
576 		semaptr->sem_perm.mode = 0;
577 		semundo_clear(ix, -1);
578 		cv_broadcast(&semcv[ix]);
579 		break;
580 
581 	case IPC_SET:
582 		if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_M)))
583 			break;
584 		KASSERT(sembuf != NULL);
585 		semaptr->sem_perm.uid = sembuf->sem_perm.uid;
586 		semaptr->sem_perm.gid = sembuf->sem_perm.gid;
587 		semaptr->sem_perm.mode = (semaptr->sem_perm.mode & ~0777) |
588 		    (sembuf->sem_perm.mode & 0777);
589 		semaptr->sem_ctime = time_second;
590 		break;
591 
592 	case IPC_STAT:
593 		if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_R)))
594 			break;
595 		KASSERT(sembuf != NULL);
596 		memcpy(sembuf, semaptr, sizeof(struct semid_ds));
597 		sembuf->sem_perm.mode &= 0777;
598 		break;
599 
600 	case GETNCNT:
601 		if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_R)))
602 			break;
603 		if (semnum < 0 || semnum >= semaptr->sem_nsems) {
604 			error = EINVAL;
605 			break;
606 		}
607 		*retval = semaptr->_sem_base[semnum].semncnt;
608 		break;
609 
610 	case GETPID:
611 		if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_R)))
612 			break;
613 		if (semnum < 0 || semnum >= semaptr->sem_nsems) {
614 			error = EINVAL;
615 			break;
616 		}
617 		*retval = semaptr->_sem_base[semnum].sempid;
618 		break;
619 
620 	case GETVAL:
621 		if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_R)))
622 			break;
623 		if (semnum < 0 || semnum >= semaptr->sem_nsems) {
624 			error = EINVAL;
625 			break;
626 		}
627 		*retval = semaptr->_sem_base[semnum].semval;
628 		break;
629 
630 	case GETALL:
631 		if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_R)))
632 			break;
633 		KASSERT(arg != NULL);
634 		for (i = 0; i < semaptr->sem_nsems; i++) {
635 			error = copyout(&semaptr->_sem_base[i].semval,
636 			    &arg->array[i], sizeof(arg->array[i]));
637 			if (error != 0)
638 				break;
639 		}
640 		break;
641 
642 	case GETZCNT:
643 		if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_R)))
644 			break;
645 		if (semnum < 0 || semnum >= semaptr->sem_nsems) {
646 			error = EINVAL;
647 			break;
648 		}
649 		*retval = semaptr->_sem_base[semnum].semzcnt;
650 		break;
651 
652 	case SETVAL:
653 		if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_W)))
654 			break;
655 		if (semnum < 0 || semnum >= semaptr->sem_nsems) {
656 			error = EINVAL;
657 			break;
658 		}
659 		KASSERT(arg != NULL);
660 		if ((unsigned int)arg->val > seminfo.semvmx) {
661 			error = ERANGE;
662 			break;
663 		}
664 		semaptr->_sem_base[semnum].semval = arg->val;
665 		semundo_clear(ix, semnum);
666 		cv_broadcast(&semcv[ix]);
667 		break;
668 
669 	case SETALL:
670 		if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_W)))
671 			break;
672 		KASSERT(arg != NULL);
673 		for (i = 0; i < semaptr->sem_nsems; i++) {
674 			unsigned short semval;
675 			error = copyin(&arg->array[i], &semval,
676 			    sizeof(arg->array[i]));
677 			if (error != 0)
678 				break;
679 			if ((unsigned int)semval > seminfo.semvmx) {
680 				error = ERANGE;
681 				break;
682 			}
683 			semaptr->_sem_base[i].semval = semval;
684 		}
685 		semundo_clear(ix, -1);
686 		cv_broadcast(&semcv[ix]);
687 		break;
688 
689 	default:
690 		error = EINVAL;
691 		break;
692 	}
693 
694 	mutex_exit(&semlock);
695 	return (error);
696 }
697 
698 int
699 sys_semget(struct lwp *l, const struct sys_semget_args *uap, register_t *retval)
700 {
701 	/* {
702 		syscallarg(key_t) key;
703 		syscallarg(int) nsems;
704 		syscallarg(int) semflg;
705 	} */
706 	int semid, error = 0;
707 	int key = SCARG(uap, key);
708 	int nsems = SCARG(uap, nsems);
709 	int semflg = SCARG(uap, semflg);
710 	kauth_cred_t cred = l->l_cred;
711 
712 	RUN_ONCE(&exithook_control, seminit_exithook);
713 
714 	SEM_PRINTF(("semget(0x%x, %d, 0%o)\n", key, nsems, semflg));
715 
716 	mutex_enter(&semlock);
717 
718 	if (key != IPC_PRIVATE) {
719 		for (semid = 0; semid < seminfo.semmni; semid++) {
720 			if ((sema[semid].sem_perm.mode & SEM_ALLOC) &&
721 			    sema[semid].sem_perm._key == key)
722 				break;
723 		}
724 		if (semid < seminfo.semmni) {
725 			SEM_PRINTF(("found public key\n"));
726 			if ((error = ipcperm(cred, &sema[semid].sem_perm,
727 			    semflg & 0700)))
728 			    	goto out;
729 			if (nsems > 0 && sema[semid].sem_nsems < nsems) {
730 				SEM_PRINTF(("too small\n"));
731 				error = EINVAL;
732 				goto out;
733 			}
734 			if ((semflg & IPC_CREAT) && (semflg & IPC_EXCL)) {
735 				SEM_PRINTF(("not exclusive\n"));
736 				error = EEXIST;
737 				goto out;
738 			}
739 			goto found;
740 		}
741 	}
742 
743 	SEM_PRINTF(("need to allocate the semid_ds\n"));
744 	if (key == IPC_PRIVATE || (semflg & IPC_CREAT)) {
745 		if (nsems <= 0 || nsems > seminfo.semmsl) {
746 			SEM_PRINTF(("nsems out of range (0<%d<=%d)\n", nsems,
747 			    seminfo.semmsl));
748 			error = EINVAL;
749 			goto out;
750 		}
751 		if (nsems > seminfo.semmns - semtot) {
752 			SEM_PRINTF(("not enough semaphores left "
753 			    "(need %d, got %d)\n",
754 			    nsems, seminfo.semmns - semtot));
755 			error = ENOSPC;
756 			goto out;
757 		}
758 		for (semid = 0; semid < seminfo.semmni; semid++) {
759 			if ((sema[semid].sem_perm.mode & SEM_ALLOC) == 0)
760 				break;
761 		}
762 		if (semid == seminfo.semmni) {
763 			SEM_PRINTF(("no more semid_ds's available\n"));
764 			error = ENOSPC;
765 			goto out;
766 		}
767 		SEM_PRINTF(("semid %d is available\n", semid));
768 		sema[semid].sem_perm._key = key;
769 		sema[semid].sem_perm.cuid = kauth_cred_geteuid(cred);
770 		sema[semid].sem_perm.uid = kauth_cred_geteuid(cred);
771 		sema[semid].sem_perm.cgid = kauth_cred_getegid(cred);
772 		sema[semid].sem_perm.gid = kauth_cred_getegid(cred);
773 		sema[semid].sem_perm.mode = (semflg & 0777) | SEM_ALLOC;
774 		sema[semid].sem_perm._seq =
775 		    (sema[semid].sem_perm._seq + 1) & 0x7fff;
776 		sema[semid].sem_nsems = nsems;
777 		sema[semid].sem_otime = 0;
778 		sema[semid].sem_ctime = time_second;
779 		sema[semid]._sem_base = &sem[semtot];
780 		semtot += nsems;
781 		memset(sema[semid]._sem_base, 0,
782 		    sizeof(sema[semid]._sem_base[0]) * nsems);
783 		SEM_PRINTF(("sembase = %p, next = %p\n", sema[semid]._sem_base,
784 		    &sem[semtot]));
785 	} else {
786 		SEM_PRINTF(("didn't find it and wasn't asked to create it\n"));
787 		error = ENOENT;
788 		goto out;
789 	}
790 
791  found:
792 	*retval = IXSEQ_TO_IPCID(semid, sema[semid].sem_perm);
793  out:
794 	mutex_exit(&semlock);
795 	return (error);
796 }
797 
798 #define SMALL_SOPS 8
799 
800 int
801 sys_semop(struct lwp *l, const struct sys_semop_args *uap, register_t *retval)
802 {
803 	/* {
804 		syscallarg(int) semid;
805 		syscallarg(struct sembuf *) sops;
806 		syscallarg(size_t) nsops;
807 	} */
808 	struct proc *p = l->l_proc;
809 	int semid = SCARG(uap, semid), seq;
810 	size_t nsops = SCARG(uap, nsops);
811 	struct sembuf small_sops[SMALL_SOPS];
812 	struct sembuf *sops;
813 	struct semid_ds *semaptr;
814 	struct sembuf *sopptr = NULL;
815 	struct __sem *semptr = NULL;
816 	struct sem_undo *suptr = NULL;
817 	kauth_cred_t cred = l->l_cred;
818 	int i, error;
819 	int do_wakeup, do_undos;
820 
821 	RUN_ONCE(&exithook_control, seminit_exithook);
822 
823 	SEM_PRINTF(("call to semop(%d, %p, %zd)\n", semid, SCARG(uap,sops), nsops));
824 
825 	if (__predict_false((p->p_flag & PK_SYSVSEM) == 0)) {
826 		mutex_enter(p->p_lock);
827 		p->p_flag |= PK_SYSVSEM;
828 		mutex_exit(p->p_lock);
829 	}
830 
831 restart:
832 	if (nsops <= SMALL_SOPS) {
833 		sops = small_sops;
834 	} else if (nsops <= seminfo.semopm) {
835 		sops = kmem_alloc(nsops * sizeof(*sops), KM_SLEEP);
836 	} else {
837 		SEM_PRINTF(("too many sops (max=%d, nsops=%zd)\n",
838 		    seminfo.semopm, nsops));
839 		return (E2BIG);
840 	}
841 
842 	error = copyin(SCARG(uap, sops), sops, nsops * sizeof(sops[0]));
843 	if (error) {
844 		SEM_PRINTF(("error = %d from copyin(%p, %p, %zd)\n", error,
845 		    SCARG(uap, sops), &sops, nsops * sizeof(sops[0])));
846 		if (sops != small_sops)
847 			kmem_free(sops, nsops * sizeof(*sops));
848 		return error;
849 	}
850 
851 	mutex_enter(&semlock);
852 	/* In case of reallocation, we will wait for completion */
853 	while (__predict_false(sem_realloc_state))
854 		cv_wait(&sem_realloc_cv, &semlock);
855 
856 	semid = IPCID_TO_IX(semid);	/* Convert back to zero origin */
857 	if (semid < 0 || semid >= seminfo.semmni) {
858 		error = EINVAL;
859 		goto out;
860 	}
861 
862 	semaptr = &sema[semid];
863 	seq = IPCID_TO_SEQ(SCARG(uap, semid));
864 	if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0 ||
865 	    semaptr->sem_perm._seq != seq) {
866 		error = EINVAL;
867 		goto out;
868 	}
869 
870 	if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_W))) {
871 		SEM_PRINTF(("error = %d from ipaccess\n", error));
872 		goto out;
873 	}
874 
875 	for (i = 0; i < nsops; i++)
876 		if (sops[i].sem_num >= semaptr->sem_nsems) {
877 			error = EFBIG;
878 			goto out;
879 		}
880 
881 	/*
882 	 * Loop trying to satisfy the vector of requests.
883 	 * If we reach a point where we must wait, any requests already
884 	 * performed are rolled back and we go to sleep until some other
885 	 * process wakes us up.  At this point, we start all over again.
886 	 *
887 	 * This ensures that from the perspective of other tasks, a set
888 	 * of requests is atomic (never partially satisfied).
889 	 */
890 	do_undos = 0;
891 
892 	for (;;) {
893 		do_wakeup = 0;
894 
895 		for (i = 0; i < nsops; i++) {
896 			sopptr = &sops[i];
897 			semptr = &semaptr->_sem_base[sopptr->sem_num];
898 
899 			SEM_PRINTF(("semop:  semaptr=%p, sem_base=%p, "
900 			    "semptr=%p, sem[%d]=%d : op=%d, flag=%s\n",
901 			    semaptr, semaptr->_sem_base, semptr,
902 			    sopptr->sem_num, semptr->semval, sopptr->sem_op,
903 			    (sopptr->sem_flg & IPC_NOWAIT) ?
904 			    "nowait" : "wait"));
905 
906 			if (sopptr->sem_op < 0) {
907 				if ((int)(semptr->semval +
908 				    sopptr->sem_op) < 0) {
909 					SEM_PRINTF(("semop:  "
910 					    "can't do it now\n"));
911 					break;
912 				} else {
913 					semptr->semval += sopptr->sem_op;
914 					if (semptr->semval == 0 &&
915 					    semptr->semzcnt > 0)
916 						do_wakeup = 1;
917 				}
918 				if (sopptr->sem_flg & SEM_UNDO)
919 					do_undos = 1;
920 			} else if (sopptr->sem_op == 0) {
921 				if (semptr->semval > 0) {
922 					SEM_PRINTF(("semop:  not zero now\n"));
923 					break;
924 				}
925 			} else {
926 				if (semptr->semncnt > 0)
927 					do_wakeup = 1;
928 				semptr->semval += sopptr->sem_op;
929 				if (sopptr->sem_flg & SEM_UNDO)
930 					do_undos = 1;
931 			}
932 		}
933 
934 		/*
935 		 * Did we get through the entire vector?
936 		 */
937 		if (i >= nsops)
938 			goto done;
939 
940 		/*
941 		 * No ... rollback anything that we've already done
942 		 */
943 		SEM_PRINTF(("semop:  rollback 0 through %d\n", i - 1));
944 		while (i-- > 0)
945 			semaptr->_sem_base[sops[i].sem_num].semval -=
946 			    sops[i].sem_op;
947 
948 		/*
949 		 * If the request that we couldn't satisfy has the
950 		 * NOWAIT flag set then return with EAGAIN.
951 		 */
952 		if (sopptr->sem_flg & IPC_NOWAIT) {
953 			error = EAGAIN;
954 			goto out;
955 		}
956 
957 		if (sopptr->sem_op == 0)
958 			semptr->semzcnt++;
959 		else
960 			semptr->semncnt++;
961 
962 		sem_waiters++;
963 		SEM_PRINTF(("semop:  good night!\n"));
964 		error = cv_wait_sig(&semcv[semid], &semlock);
965 		SEM_PRINTF(("semop:  good morning (error=%d)!\n", error));
966 		sem_waiters--;
967 
968 		/* Notify reallocator, if it is waiting */
969 		cv_broadcast(&sem_realloc_cv);
970 
971 		/*
972 		 * Make sure that the semaphore still exists
973 		 */
974 		if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0 ||
975 		    semaptr->sem_perm._seq != seq) {
976 			error = EIDRM;
977 			goto out;
978 		}
979 
980 		/*
981 		 * The semaphore is still alive.  Readjust the count of
982 		 * waiting processes.
983 		 */
984 		semptr = &semaptr->_sem_base[sopptr->sem_num];
985 		if (sopptr->sem_op == 0)
986 			semptr->semzcnt--;
987 		else
988 			semptr->semncnt--;
989 
990 		/* In case of such state, restart the call */
991 		if (sem_realloc_state) {
992 			mutex_exit(&semlock);
993 			goto restart;
994 		}
995 
996 		/* Is it really morning, or was our sleep interrupted? */
997 		if (error != 0) {
998 			error = EINTR;
999 			goto out;
1000 		}
1001 		SEM_PRINTF(("semop:  good morning!\n"));
1002 	}
1003 
1004 done:
1005 	/*
1006 	 * Process any SEM_UNDO requests.
1007 	 */
1008 	if (do_undos) {
1009 		for (i = 0; i < nsops; i++) {
1010 			/*
1011 			 * We only need to deal with SEM_UNDO's for non-zero
1012 			 * op's.
1013 			 */
1014 			int adjval;
1015 
1016 			if ((sops[i].sem_flg & SEM_UNDO) == 0)
1017 				continue;
1018 			adjval = sops[i].sem_op;
1019 			if (adjval == 0)
1020 				continue;
1021 			error = semundo_adjust(p, &suptr, semid,
1022 			    sops[i].sem_num, -adjval);
1023 			if (error == 0)
1024 				continue;
1025 
1026 			/*
1027 			 * Oh-Oh!  We ran out of either sem_undo's or undo's.
1028 			 * Rollback the adjustments to this point and then
1029 			 * rollback the semaphore ups and down so we can return
1030 			 * with an error with all structures restored.  We
1031 			 * rollback the undo's in the exact reverse order that
1032 			 * we applied them.  This guarantees that we won't run
1033 			 * out of space as we roll things back out.
1034 			 */
1035 			while (i-- > 0) {
1036 				if ((sops[i].sem_flg & SEM_UNDO) == 0)
1037 					continue;
1038 				adjval = sops[i].sem_op;
1039 				if (adjval == 0)
1040 					continue;
1041 				if (semundo_adjust(p, &suptr, semid,
1042 				    sops[i].sem_num, adjval) != 0)
1043 					panic("semop - can't undo undos");
1044 			}
1045 
1046 			for (i = 0; i < nsops; i++)
1047 				semaptr->_sem_base[sops[i].sem_num].semval -=
1048 				    sops[i].sem_op;
1049 
1050 			SEM_PRINTF(("error = %d from semundo_adjust\n", error));
1051 			goto out;
1052 		} /* loop through the sops */
1053 	} /* if (do_undos) */
1054 
1055 	/* We're definitely done - set the sempid's */
1056 	for (i = 0; i < nsops; i++) {
1057 		sopptr = &sops[i];
1058 		semptr = &semaptr->_sem_base[sopptr->sem_num];
1059 		semptr->sempid = p->p_pid;
1060 	}
1061 
1062 	/* Update sem_otime */
1063 	semaptr->sem_otime = time_second;
1064 
1065 	/* Do a wakeup if any semaphore was up'd. */
1066 	if (do_wakeup) {
1067 		SEM_PRINTF(("semop:  doing wakeup\n"));
1068 		cv_broadcast(&semcv[semid]);
1069 		SEM_PRINTF(("semop:  back from wakeup\n"));
1070 	}
1071 	SEM_PRINTF(("semop:  done\n"));
1072 	*retval = 0;
1073 
1074  out:
1075 	mutex_exit(&semlock);
1076 	if (sops != small_sops)
1077 		kmem_free(sops, nsops * sizeof(*sops));
1078 	return error;
1079 }
1080 
1081 /*
1082  * Go through the undo structures for this process and apply the
1083  * adjustments to semaphores.
1084  */
1085 /*ARGSUSED*/
1086 void
1087 semexit(struct proc *p, void *v)
1088 {
1089 	struct sem_undo *suptr;
1090 	struct sem_undo **supptr;
1091 
1092 	if ((p->p_flag & PK_SYSVSEM) == 0)
1093 		return;
1094 
1095 	mutex_enter(&semlock);
1096 
1097 	/*
1098 	 * Go through the chain of undo vectors looking for one
1099 	 * associated with this process.
1100 	 */
1101 
1102 	for (supptr = &semu_list; (suptr = *supptr) != NULL;
1103 	    supptr = &suptr->un_next) {
1104 		if (suptr->un_proc == p)
1105 			break;
1106 	}
1107 
1108 	/*
1109 	 * If there is no undo vector, skip to the end.
1110 	 */
1111 
1112 	if (suptr == NULL) {
1113 		mutex_exit(&semlock);
1114 		return;
1115 	}
1116 
1117 	/*
1118 	 * We now have an undo vector for this process.
1119 	 */
1120 
1121 	SEM_PRINTF(("proc @%p has undo structure with %d entries\n", p,
1122 	    suptr->un_cnt));
1123 
1124 	/*
1125 	 * If there are any active undo elements then process them.
1126 	 */
1127 	if (suptr->un_cnt > 0) {
1128 		int ix;
1129 
1130 		for (ix = 0; ix < suptr->un_cnt; ix++) {
1131 			int semid = suptr->un_ent[ix].un_id;
1132 			int semnum = suptr->un_ent[ix].un_num;
1133 			int adjval = suptr->un_ent[ix].un_adjval;
1134 			struct semid_ds *semaptr;
1135 
1136 			semaptr = &sema[semid];
1137 			if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0)
1138 			if (semnum >= semaptr->sem_nsems)
1139 				panic("semexit - semnum out of range");
1140 
1141 			SEM_PRINTF(("semexit:  %p id=%d num=%d(adj=%d) ; "
1142 			    "sem=%d\n",
1143 			    suptr->un_proc, suptr->un_ent[ix].un_id,
1144 			    suptr->un_ent[ix].un_num,
1145 			    suptr->un_ent[ix].un_adjval,
1146 			    semaptr->_sem_base[semnum].semval));
1147 
1148 			if (adjval < 0 &&
1149 			    semaptr->_sem_base[semnum].semval < -adjval)
1150 				semaptr->_sem_base[semnum].semval = 0;
1151 			else
1152 				semaptr->_sem_base[semnum].semval += adjval;
1153 
1154 			cv_broadcast(&semcv[semid]);
1155 			SEM_PRINTF(("semexit:  back from wakeup\n"));
1156 		}
1157 	}
1158 
1159 	/*
1160 	 * Deallocate the undo vector.
1161 	 */
1162 	SEM_PRINTF(("removing vector\n"));
1163 	suptr->un_proc = NULL;
1164 	*supptr = suptr->un_next;
1165 	mutex_exit(&semlock);
1166 }
1167 
1168 /*
1169  * Sysctl initialization and nodes.
1170  */
1171 
1172 static int
1173 sysctl_ipc_semmni(SYSCTLFN_ARGS)
1174 {
1175 	int newsize, error;
1176 	struct sysctlnode node;
1177 	node = *rnode;
1178 	node.sysctl_data = &newsize;
1179 
1180 	newsize = seminfo.semmni;
1181 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
1182 	if (error || newp == NULL)
1183 		return error;
1184 
1185 	return semrealloc(newsize, seminfo.semmns, seminfo.semmnu);
1186 }
1187 
1188 static int
1189 sysctl_ipc_semmns(SYSCTLFN_ARGS)
1190 {
1191 	int newsize, error;
1192 	struct sysctlnode node;
1193 	node = *rnode;
1194 	node.sysctl_data = &newsize;
1195 
1196 	newsize = seminfo.semmns;
1197 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
1198 	if (error || newp == NULL)
1199 		return error;
1200 
1201 	return semrealloc(seminfo.semmni, newsize, seminfo.semmnu);
1202 }
1203 
1204 static int
1205 sysctl_ipc_semmnu(SYSCTLFN_ARGS)
1206 {
1207 	int newsize, error;
1208 	struct sysctlnode node;
1209 	node = *rnode;
1210 	node.sysctl_data = &newsize;
1211 
1212 	newsize = seminfo.semmnu;
1213 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
1214 	if (error || newp == NULL)
1215 		return error;
1216 
1217 	return semrealloc(seminfo.semmni, seminfo.semmns, newsize);
1218 }
1219 
1220 SYSCTL_SETUP(sysctl_ipc_sem_setup, "sysctl kern.ipc subtree setup")
1221 {
1222 	const struct sysctlnode *node = NULL;
1223 
1224 	sysctl_createv(clog, 0, NULL, &node,
1225 		CTLFLAG_PERMANENT,
1226 		CTLTYPE_NODE, "ipc",
1227 		SYSCTL_DESCR("SysV IPC options"),
1228 		NULL, 0, NULL, 0,
1229 		CTL_KERN, KERN_SYSVIPC, CTL_EOL);
1230 
1231 	if (node == NULL)
1232 		return;
1233 
1234 	sysctl_createv(clog, 0, &node, NULL,
1235 		CTLFLAG_PERMANENT | CTLFLAG_READWRITE,
1236 		CTLTYPE_INT, "semmni",
1237 		SYSCTL_DESCR("Max number of number of semaphore identifiers"),
1238 		sysctl_ipc_semmni, 0, &seminfo.semmni, 0,
1239 		CTL_CREATE, CTL_EOL);
1240 	sysctl_createv(clog, 0, &node, NULL,
1241 		CTLFLAG_PERMANENT | CTLFLAG_READWRITE,
1242 		CTLTYPE_INT, "semmns",
1243 		SYSCTL_DESCR("Max number of number of semaphores in system"),
1244 		sysctl_ipc_semmns, 0, &seminfo.semmns, 0,
1245 		CTL_CREATE, CTL_EOL);
1246 	sysctl_createv(clog, 0, &node, NULL,
1247 		CTLFLAG_PERMANENT | CTLFLAG_READWRITE,
1248 		CTLTYPE_INT, "semmnu",
1249 		SYSCTL_DESCR("Max number of undo structures in system"),
1250 		sysctl_ipc_semmnu, 0, &seminfo.semmnu, 0,
1251 		CTL_CREATE, CTL_EOL);
1252 }
1253