1/*-
2 * Implementation of SVID semaphores
3 *
4 * Author: Daniel Boulet
5 *
6 * This software is provided ``AS IS'' without any warranties of any kind.
7 */
8/*-
9 * Copyright (c) 2003-2005 McAfee, Inc.
10 * All rights reserved.
11 *
12 * This software was developed for the FreeBSD Project in part by McAfee
13 * Research, the Security Research Division of McAfee, Inc under DARPA/SPAWAR
14 * contract N66001-01-C-8035 ("CBOSS"), as part of the DARPA CHATS research
15 * program.
16 *
17 * Redistribution and use in source and binary forms, with or without
18 * modification, are permitted provided that the following conditions
19 * are met:
20 * 1. Redistributions of source code must retain the above copyright
21 * notice, this list of conditions and the following disclaimer.
22 * 2. Redistributions in binary form must reproduce the above copyright
23 * notice, this list of conditions and the following disclaimer in the
24 * documentation and/or other materials provided with the distribution.
25 *
26 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36 * SUCH DAMAGE.
37 */
38
39#include <sys/cdefs.h>
40__FBSDID("$FreeBSD: head/sys/kern/sysv_sem.c 145230 2005-04-18 11:51:10Z rwatson $");
41
42#include "opt_sysvipc.h"
43#include "opt_mac.h"
44
45#include <sys/param.h>
46#include <sys/systm.h>
47#include <sys/sysproto.h>
48#include <sys/eventhandler.h>
49#include <sys/kernel.h>
50#include <sys/proc.h>
51#include <sys/lock.h>
52#include <sys/module.h>
53#include <sys/mutex.h>
54#include <sys/sem.h>
55#include <sys/syscall.h>
56#include <sys/sysent.h>
57#include <sys/sysctl.h>
58#include <sys/malloc.h>
59#include <sys/jail.h>
60#include <sys/mac.h>
61
62static MALLOC_DEFINE(M_SEM, "sem", "SVID compatible semaphores");
63
64#ifdef SEM_DEBUG
65#define DPRINTF(a) printf a
66#else
67#define DPRINTF(a)
68#endif
69#ifdef MAC_DEBUG
70#define MPRINTF(a) printf a
71#else
72#define MPRINTF(a)
73#endif
74
75static void seminit(void);
76static int sysvsem_modload(struct module *, int, void *);
77static int semunload(void);
78static void semexit_myhook(void *arg, struct proc *p);
79static int sysctl_sema(SYSCTL_HANDLER_ARGS);
80static int semvalid(int semid, struct semid_kernel *semakptr);
81
82#ifndef _SYS_SYSPROTO_H_
83struct __semctl_args;
84int __semctl(struct thread *td, struct __semctl_args *uap);
85struct semget_args;
86int semget(struct thread *td, struct semget_args *uap);
87struct semop_args;
88int semop(struct thread *td, struct semop_args *uap);
89#endif
90
91static struct sem_undo *semu_alloc(struct thread *td);
92static int semundo_adjust(struct thread *td, struct sem_undo **supptr,
93 int semid, int semnum, int adjval);
94static void semundo_clear(int semid, int semnum);
95
96/* XXX casting to (sy_call_t *) is bogus, as usual. */
97static sy_call_t *semcalls[] = {
98 (sy_call_t *)__semctl, (sy_call_t *)semget,
99 (sy_call_t *)semop
100};
101
102static struct mtx sem_mtx; /* semaphore global lock */
103static int semtot = 0;
104static struct semid_kernel *sema; /* semaphore id pool */
105static struct mtx *sema_mtx; /* semaphore id pool mutexes*/
106static struct sem *sem; /* semaphore pool */
107SLIST_HEAD(, sem_undo) semu_list; /* list of active undo structures */
108static int *semu; /* undo structure pool */
109static eventhandler_tag semexit_tag;
110
111#define SEMUNDO_MTX sem_mtx
112#define SEMUNDO_LOCK() mtx_lock(&SEMUNDO_MTX);
113#define SEMUNDO_UNLOCK() mtx_unlock(&SEMUNDO_MTX);
114#define SEMUNDO_LOCKASSERT(how) mtx_assert(&SEMUNDO_MTX, (how));
115
116struct sem {
117 u_short semval; /* semaphore value */
118 pid_t sempid; /* pid of last operation */
119 u_short semncnt; /* # awaiting semval > cval */
120 u_short semzcnt; /* # awaiting semval = 0 */
121};
122
123/*
124 * Undo structure (one per process)
125 */
126struct sem_undo {
127 SLIST_ENTRY(sem_undo) un_next; /* ptr to next active undo structure */
128 struct proc *un_proc; /* owner of this structure */
129 short un_cnt; /* # of active entries */
130 struct undo {
131 short un_adjval; /* adjust on exit values */
132 short un_num; /* semaphore # */
133 int un_id; /* semid */
134 } un_ent[1]; /* undo entries */
135};
136
137/*
138 * Configuration parameters
139 */
140#ifndef SEMMNI
141#define SEMMNI 10 /* # of semaphore identifiers */
142#endif
143#ifndef SEMMNS
144#define SEMMNS 60 /* # of semaphores in system */
145#endif
146#ifndef SEMUME
147#define SEMUME 10 /* max # of undo entries per process */
148#endif
149#ifndef SEMMNU
150#define SEMMNU 30 /* # of undo structures in system */
151#endif
152
153/* shouldn't need tuning */
154#ifndef SEMMAP
155#define SEMMAP 30 /* # of entries in semaphore map */
156#endif
157#ifndef SEMMSL
158#define SEMMSL SEMMNS /* max # of semaphores per id */
159#endif
160#ifndef SEMOPM
161#define SEMOPM 100 /* max # of operations per semop call */
162#endif
163
164#define SEMVMX 32767 /* semaphore maximum value */
165#define SEMAEM 16384 /* adjust on exit max value */
166
167/*
168 * Due to the way semaphore memory is allocated, we have to ensure that
169 * SEMUSZ is properly aligned.
170 */
171
172#define SEM_ALIGN(bytes) (((bytes) + (sizeof(long) - 1)) & ~(sizeof(long) - 1))
173
174/* actual size of an undo structure */
175#define SEMUSZ SEM_ALIGN(offsetof(struct sem_undo, un_ent[SEMUME]))
176
177/*
178 * Macro to find a particular sem_undo vector
179 */
180#define SEMU(ix) \
181 ((struct sem_undo *)(((intptr_t)semu)+ix * seminfo.semusz))
182
183/*
184 * semaphore info struct
185 */
186struct seminfo seminfo = {
187 SEMMAP, /* # of entries in semaphore map */
188 SEMMNI, /* # of semaphore identifiers */
189 SEMMNS, /* # of semaphores in system */
190 SEMMNU, /* # of undo structures in system */
191 SEMMSL, /* max # of semaphores per id */
192 SEMOPM, /* max # of operations per semop call */
193 SEMUME, /* max # of undo entries per process */
194 SEMUSZ, /* size in bytes of undo structure */
195 SEMVMX, /* semaphore maximum value */
196 SEMAEM /* adjust on exit max value */
197};
198
199SYSCTL_DECL(_kern_ipc);
200SYSCTL_INT(_kern_ipc, OID_AUTO, semmap, CTLFLAG_RW, &seminfo.semmap, 0,
201 "Number of entries in the semaphore map");
202SYSCTL_INT(_kern_ipc, OID_AUTO, semmni, CTLFLAG_RDTUN, &seminfo.semmni, 0,
203 "Number of semaphore identifiers");
204SYSCTL_INT(_kern_ipc, OID_AUTO, semmns, CTLFLAG_RDTUN, &seminfo.semmns, 0,
205 "Maximum number of semaphores in the system");
206SYSCTL_INT(_kern_ipc, OID_AUTO, semmnu, CTLFLAG_RDTUN, &seminfo.semmnu, 0,
207 "Maximum number of undo structures in the system");
208SYSCTL_INT(_kern_ipc, OID_AUTO, semmsl, CTLFLAG_RW, &seminfo.semmsl, 0,
209 "Max semaphores per id");
210SYSCTL_INT(_kern_ipc, OID_AUTO, semopm, CTLFLAG_RDTUN, &seminfo.semopm, 0,
211 "Max operations per semop call");
212SYSCTL_INT(_kern_ipc, OID_AUTO, semume, CTLFLAG_RDTUN, &seminfo.semume, 0,
213 "Max undo entries per process");
214SYSCTL_INT(_kern_ipc, OID_AUTO, semusz, CTLFLAG_RDTUN, &seminfo.semusz, 0,
215 "Size in bytes of undo structure");
216SYSCTL_INT(_kern_ipc, OID_AUTO, semvmx, CTLFLAG_RW, &seminfo.semvmx, 0,
217 "Semaphore maximum value");
218SYSCTL_INT(_kern_ipc, OID_AUTO, semaem, CTLFLAG_RW, &seminfo.semaem, 0,
219 "Adjust on exit max value");
220SYSCTL_PROC(_kern_ipc, OID_AUTO, sema, CTLFLAG_RD,
221 NULL, 0, sysctl_sema, "", "");
222
223static void
224seminit(void)
225{
226 int i;
227
228 TUNABLE_INT_FETCH("kern.ipc.semmap", &seminfo.semmap);
229 TUNABLE_INT_FETCH("kern.ipc.semmni", &seminfo.semmni);
230 TUNABLE_INT_FETCH("kern.ipc.semmns", &seminfo.semmns);
231 TUNABLE_INT_FETCH("kern.ipc.semmnu", &seminfo.semmnu);
232 TUNABLE_INT_FETCH("kern.ipc.semmsl", &seminfo.semmsl);
233 TUNABLE_INT_FETCH("kern.ipc.semopm", &seminfo.semopm);
234 TUNABLE_INT_FETCH("kern.ipc.semume", &seminfo.semume);
235 TUNABLE_INT_FETCH("kern.ipc.semusz", &seminfo.semusz);
236 TUNABLE_INT_FETCH("kern.ipc.semvmx", &seminfo.semvmx);
237 TUNABLE_INT_FETCH("kern.ipc.semaem", &seminfo.semaem);
238
239 sem = malloc(sizeof(struct sem) * seminfo.semmns, M_SEM, M_WAITOK);
240 sema = malloc(sizeof(struct semid_kernel) * seminfo.semmni, M_SEM,
241 M_WAITOK);
242 sema_mtx = malloc(sizeof(struct mtx) * seminfo.semmni, M_SEM,
243 M_WAITOK | M_ZERO);
244 semu = malloc(seminfo.semmnu * seminfo.semusz, M_SEM, M_WAITOK);
245
246 for (i = 0; i < seminfo.semmni; i++) {
247 sema[i].u.sem_base = 0;
248 sema[i].u.sem_perm.mode = 0;
249 sema[i].u.sem_perm.seq = 0;
250#ifdef MAC
251 mac_init_sysv_sema(&sema[i]);
252#endif
253 }
254 for (i = 0; i < seminfo.semmni; i++)
255 mtx_init(&sema_mtx[i], "semid", NULL, MTX_DEF);
256 for (i = 0; i < seminfo.semmnu; i++) {
257 struct sem_undo *suptr = SEMU(i);
258 suptr->un_proc = NULL;
259 }
260 SLIST_INIT(&semu_list);
261 mtx_init(&sem_mtx, "sem", NULL, MTX_DEF);
262 semexit_tag = EVENTHANDLER_REGISTER(process_exit, semexit_myhook, NULL,
263 EVENTHANDLER_PRI_ANY);
264}
265
266static int
267semunload(void)
268{
269 int i;
270
271 if (semtot != 0)
272 return (EBUSY);
273
274 EVENTHANDLER_DEREGISTER(process_exit, semexit_tag);
275#ifdef MAC
276 for (i = 0; i < seminfo.semmni; i++)
277 mac_destroy_sysv_sema(&sema[i]);
278#endif
279 free(sem, M_SEM);
280 free(sema, M_SEM);
281 free(semu, M_SEM);
282 for (i = 0; i < seminfo.semmni; i++)
283 mtx_destroy(&sema_mtx[i]);
284 mtx_destroy(&sem_mtx);
285 return (0);
286}
287
288static int
289sysvsem_modload(struct module *module, int cmd, void *arg)
290{
291 int error = 0;
292
293 switch (cmd) {
294 case MOD_LOAD:
295 seminit();
296 break;
297 case MOD_UNLOAD:
298 error = semunload();
299 break;
300 case MOD_SHUTDOWN:
301 break;
302 default:
303 error = EINVAL;
304 break;
305 }
306 return (error);
307}
308
309static moduledata_t sysvsem_mod = {
310 "sysvsem",
311 &sysvsem_modload,
312 NULL
313};
314
315SYSCALL_MODULE_HELPER(semsys);
316SYSCALL_MODULE_HELPER(__semctl);
317SYSCALL_MODULE_HELPER(semget);
318SYSCALL_MODULE_HELPER(semop);
319
320DECLARE_MODULE(sysvsem, sysvsem_mod,
321 SI_SUB_SYSV_SEM, SI_ORDER_FIRST);
322MODULE_VERSION(sysvsem, 1);
323
324/*
325 * Entry point for all SEM calls
326 *
327 * MPSAFE
328 */
329int
330semsys(td, uap)
331 struct thread *td;
332 /* XXX actually varargs. */
333 struct semsys_args /* {
334 int which;
335 int a2;
336 int a3;
337 int a4;
338 int a5;
339 } */ *uap;
340{
341 int error;
342
343 if (!jail_sysvipc_allowed && jailed(td->td_ucred))
344 return (ENOSYS);
345 if (uap->which < 0 ||
346 uap->which >= sizeof(semcalls)/sizeof(semcalls[0]))
347 return (EINVAL);
348 error = (*semcalls[uap->which])(td, &uap->a2);
349 return (error);
350}
351
352/*
353 * Allocate a new sem_undo structure for a process
354 * (returns ptr to structure or NULL if no more room)
355 */
356
357static struct sem_undo *
358semu_alloc(td)
359 struct thread *td;
360{
361 int i;
362 struct sem_undo *suptr;
363 struct sem_undo **supptr;
364 int attempt;
365
366 SEMUNDO_LOCKASSERT(MA_OWNED);
367 /*
368 * Try twice to allocate something.
369 * (we'll purge an empty structure after the first pass so
370 * two passes are always enough)
371 */
372
373 for (attempt = 0; attempt < 2; attempt++) {
374 /*
375 * Look for a free structure.
376 * Fill it in and return it if we find one.
377 */
378
379 for (i = 0; i < seminfo.semmnu; i++) {
380 suptr = SEMU(i);
381 if (suptr->un_proc == NULL) {
382 SLIST_INSERT_HEAD(&semu_list, suptr, un_next);
383 suptr->un_cnt = 0;
384 suptr->un_proc = td->td_proc;
385 return(suptr);
386 }
387 }
388
389 /*
390 * We didn't find a free one, if this is the first attempt
391 * then try to free a structure.
392 */
393
394 if (attempt == 0) {
395 /* All the structures are in use - try to free one */
396 int did_something = 0;
397
398 SLIST_FOREACH_PREVPTR(suptr, supptr, &semu_list,
399 un_next) {
400 if (suptr->un_cnt == 0) {
401 suptr->un_proc = NULL;
402 did_something = 1;
403 *supptr = SLIST_NEXT(suptr, un_next);
404 break;
405 }
406 }
407
408 /* If we didn't free anything then just give-up */
409 if (!did_something)
410 return(NULL);
411 } else {
412 /*
413 * The second pass failed even though we freed
414 * something after the first pass!
415 * This is IMPOSSIBLE!
416 */
417 panic("semu_alloc - second attempt failed");
418 }
419 }
420 return (NULL);
421}
422
423/*
424 * Adjust a particular entry for a particular proc
425 */
426
427static int
428semundo_adjust(td, supptr, semid, semnum, adjval)
429 struct thread *td;
430 struct sem_undo **supptr;
431 int semid, semnum;
432 int adjval;
433{
434 struct proc *p = td->td_proc;
435 struct sem_undo *suptr;
436 struct undo *sunptr;
437 int i;
438
439 SEMUNDO_LOCKASSERT(MA_OWNED);
440 /* Look for and remember the sem_undo if the caller doesn't provide
441 it */
442
443 suptr = *supptr;
444 if (suptr == NULL) {
445 SLIST_FOREACH(suptr, &semu_list, un_next) {
446 if (suptr->un_proc == p) {
447 *supptr = suptr;
448 break;
449 }
450 }
451 if (suptr == NULL) {
452 if (adjval == 0)
453 return(0);
454 suptr = semu_alloc(td);
455 if (suptr == NULL)
456 return(ENOSPC);
457 *supptr = suptr;
458 }
459 }
460
461 /*
462 * Look for the requested entry and adjust it (delete if adjval becomes
463 * 0).
464 */
465 sunptr = &suptr->un_ent[0];
466 for (i = 0; i < suptr->un_cnt; i++, sunptr++) {
467 if (sunptr->un_id != semid || sunptr->un_num != semnum)
468 continue;
469 if (adjval != 0) {
470 adjval += sunptr->un_adjval;
471 if (adjval > seminfo.semaem || adjval < -seminfo.semaem)
472 return (ERANGE);
473 }
474 sunptr->un_adjval = adjval;
475 if (sunptr->un_adjval == 0) {
476 suptr->un_cnt--;
477 if (i < suptr->un_cnt)
478 suptr->un_ent[i] =
479 suptr->un_ent[suptr->un_cnt];
480 }
481 return(0);
482 }
483
484 /* Didn't find the right entry - create it */
485 if (adjval == 0)
486 return(0);
487 if (adjval > seminfo.semaem || adjval < -seminfo.semaem)
488 return (ERANGE);
489 if (suptr->un_cnt != seminfo.semume) {
490 sunptr = &suptr->un_ent[suptr->un_cnt];
491 suptr->un_cnt++;
492 sunptr->un_adjval = adjval;
493 sunptr->un_id = semid; sunptr->un_num = semnum;
494 } else
495 return(EINVAL);
496 return(0);
497}
498
499static void
500semundo_clear(semid, semnum)
501 int semid, semnum;
502{
503 struct sem_undo *suptr;
504
505 SEMUNDO_LOCKASSERT(MA_OWNED);
506 SLIST_FOREACH(suptr, &semu_list, un_next) {
507 struct undo *sunptr = &suptr->un_ent[0];
508 int i = 0;
509
510 while (i < suptr->un_cnt) {
511 if (sunptr->un_id == semid) {
512 if (semnum == -1 || sunptr->un_num == semnum) {
513 suptr->un_cnt--;
514 if (i < suptr->un_cnt) {
515 suptr->un_ent[i] =
516 suptr->un_ent[suptr->un_cnt];
517 continue;
518 }
519 }
520 if (semnum != -1)
521 break;
522 }
523 i++, sunptr++;
524 }
525 }
526}
527
528static int
529semvalid(semid, semakptr)
530 int semid;
531 struct semid_kernel *semakptr;
532{
533
534 return ((semakptr->u.sem_perm.mode & SEM_ALLOC) == 0 ||
535 semakptr->u.sem_perm.seq != IPCID_TO_SEQ(semid) ? EINVAL : 0);
536}
537
538/*
539 * Note that the user-mode half of this passes a union, not a pointer
540 */
541#ifndef _SYS_SYSPROTO_H_
542struct __semctl_args {
543 int semid;
544 int semnum;
545 int cmd;
546 union semun *arg;
547};
548#endif
549
550/*
551 * MPSAFE
552 */
553int
554__semctl(td, uap)
555 struct thread *td;
556 struct __semctl_args *uap;
557{
558 int semid = uap->semid;
559 int semnum = uap->semnum;
560 int cmd = uap->cmd;
561 u_short *array;
562 union semun *arg = uap->arg;
563 union semun real_arg;
564 struct ucred *cred = td->td_ucred;
565 int i, rval, error;
566 struct semid_ds sbuf;
567 struct semid_kernel *semakptr;
568 struct mtx *sema_mtxp;
569 u_short usval, count;
570
571 DPRINTF(("call to semctl(%d, %d, %d, 0x%x)\n",
572 semid, semnum, cmd, arg));
573 if (!jail_sysvipc_allowed && jailed(td->td_ucred))
574 return (ENOSYS);
575
576 array = NULL;
577
578 switch(cmd) {
579 case SEM_STAT:
580 if (semid < 0 || semid >= seminfo.semmni)
581 return (EINVAL);
582 if ((error = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
583 return (error);
584 semakptr = &sema[semid];
585 sema_mtxp = &sema_mtx[semid];
586 mtx_lock(sema_mtxp);
587 if ((semakptr->u.sem_perm.mode & SEM_ALLOC) == 0) {
588 error = EINVAL;
589 goto done2;
590 }
591 if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
592 goto done2;
593#ifdef MAC
594 error = mac_check_sysv_semctl(cred, semakptr, cmd);
595 if (error != 0) {
596 MPRINTF(("mac_check_sysv_semctl returned %d\n",
597 error));
598 goto done2;
599 }
600#endif
601 mtx_unlock(sema_mtxp);
602 error = copyout(&semakptr->u, real_arg.buf,
603 sizeof(struct semid_ds));
604 rval = IXSEQ_TO_IPCID(semid, semakptr->u.sem_perm);
605 if (error == 0)
606 td->td_retval[0] = rval;
607 return (error);
608 }
609
610 semid = IPCID_TO_IX(semid);
611 if (semid < 0 || semid >= seminfo.semmni)
612 return (EINVAL);
613
614 semakptr = &sema[semid];
615 sema_mtxp = &sema_mtx[semid];
616#ifdef MAC
617 mtx_lock(sema_mtxp);
618 error = mac_check_sysv_semctl(cred, semakptr, cmd);
619 if (error != 0) {
620 MPRINTF(("mac_check_sysv_semctl returned %d\n", error));
621 mtx_unlock(sema_mtxp);
622 return (error);
623 }
624 mtx_unlock(sema_mtxp);
625#endif
626
627 error = 0;
628 rval = 0;
629
630 switch (cmd) {
631 case IPC_RMID:
632 mtx_lock(sema_mtxp);
633 if ((error = semvalid(uap->semid, semakptr)) != 0)
634 goto done2;
635 if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_M)))
636 goto done2;
637 semakptr->u.sem_perm.cuid = cred->cr_uid;
638 semakptr->u.sem_perm.uid = cred->cr_uid;
639 semtot -= semakptr->u.sem_nsems;
640 for (i = semakptr->u.sem_base - sem; i < semtot; i++)
641 sem[i] = sem[i + semakptr->u.sem_nsems];
642 for (i = 0; i < seminfo.semmni; i++) {
643 if ((sema[i].u.sem_perm.mode & SEM_ALLOC) &&
644 sema[i].u.sem_base > semakptr->u.sem_base)
645 sema[i].u.sem_base -= semakptr->u.sem_nsems;
646 }
647 semakptr->u.sem_perm.mode = 0;
648#ifdef MAC
649 mac_cleanup_sysv_sema(semakptr);
650#endif
651 SEMUNDO_LOCK();
652 semundo_clear(semid, -1);
653 SEMUNDO_UNLOCK();
654 wakeup(semakptr);
655 break;
656
657 case IPC_SET:
658 if ((error = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
659 goto done2;
660 if ((error = copyin(real_arg.buf, &sbuf, sizeof(sbuf))) != 0)
661 goto done2;
662 mtx_lock(sema_mtxp);
663 if ((error = semvalid(uap->semid, semakptr)) != 0)
664 goto done2;
665 if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_M)))
666 goto done2;
667 semakptr->u.sem_perm.uid = sbuf.sem_perm.uid;
668 semakptr->u.sem_perm.gid = sbuf.sem_perm.gid;
669 semakptr->u.sem_perm.mode = (semakptr->u.sem_perm.mode &
670 ~0777) | (sbuf.sem_perm.mode & 0777);
671 semakptr->u.sem_ctime = time_second;
672 break;
673
674 case IPC_STAT:
675 if ((error = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
676 goto done2;
677 mtx_lock(sema_mtxp);
678 if ((error = semvalid(uap->semid, semakptr)) != 0)
679 goto done2;
680 if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
681 goto done2;
682 sbuf = semakptr->u;
683 mtx_unlock(sema_mtxp);
684 error = copyout(&semakptr->u, real_arg.buf,
685 sizeof(struct semid_ds));
686 break;
687
688 case GETNCNT:
689 mtx_lock(sema_mtxp);
690 if ((error = semvalid(uap->semid, semakptr)) != 0)
691 goto done2;
692 if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
693 goto done2;
694 if (semnum < 0 || semnum >= semakptr->u.sem_nsems) {
695 error = EINVAL;
696 goto done2;
697 }
698 rval = semakptr->u.sem_base[semnum].semncnt;
699 break;
700
701 case GETPID:
702 mtx_lock(sema_mtxp);
703 if ((error = semvalid(uap->semid, semakptr)) != 0)
704 goto done2;
705 if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
706 goto done2;
707 if (semnum < 0 || semnum >= semakptr->u.sem_nsems) {
708 error = EINVAL;
709 goto done2;
710 }
711 rval = semakptr->u.sem_base[semnum].sempid;
712 break;
713
714 case GETVAL:
715 mtx_lock(sema_mtxp);
716 if ((error = semvalid(uap->semid, semakptr)) != 0)
717 goto done2;
718 if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
719 goto done2;
720 if (semnum < 0 || semnum >= semakptr->u.sem_nsems) {
721 error = EINVAL;
722 goto done2;
723 }
724 rval = semakptr->u.sem_base[semnum].semval;
725 break;
726
727 case GETALL:
728 if ((error = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
729 goto done2;
730 array = malloc(sizeof(*array) * semakptr->u.sem_nsems, M_TEMP,
731 M_WAITOK);
732 mtx_lock(sema_mtxp);
733 if ((error = semvalid(uap->semid, semakptr)) != 0)
734 goto done2;
735 if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
736 goto done2;
737 for (i = 0; i < semakptr->u.sem_nsems; i++)
738 array[i] = semakptr->u.sem_base[i].semval;
739 mtx_unlock(sema_mtxp);
740 error = copyout(array, real_arg.array,
741 i * sizeof(real_arg.array[0]));
742 break;
743
744 case GETZCNT:
745 mtx_lock(sema_mtxp);
746 if ((error = semvalid(uap->semid, semakptr)) != 0)
747 goto done2;
748 if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
749 goto done2;
750 if (semnum < 0 || semnum >= semakptr->u.sem_nsems) {
751 error = EINVAL;
752 goto done2;
753 }
754 rval = semakptr->u.sem_base[semnum].semzcnt;
755 break;
756
757 case SETVAL:
758 if ((error = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
759 goto done2;
760 mtx_lock(sema_mtxp);
761 if ((error = semvalid(uap->semid, semakptr)) != 0)
762 goto done2;
763 if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_W)))
764 goto done2;
765 if (semnum < 0 || semnum >= semakptr->u.sem_nsems) {
766 error = EINVAL;
767 goto done2;
768 }
769 if (real_arg.val < 0 || real_arg.val > seminfo.semvmx) {
770 error = ERANGE;
771 goto done2;
772 }
773 semakptr->u.sem_base[semnum].semval = real_arg.val;
774 SEMUNDO_LOCK();
775 semundo_clear(semid, semnum);
776 SEMUNDO_UNLOCK();
777 wakeup(semakptr);
778 break;
779
780 case SETALL:
781 mtx_lock(sema_mtxp);
782raced:
783 if ((error = semvalid(uap->semid, semakptr)) != 0)
784 goto done2;
785 count = semakptr->u.sem_nsems;
786 mtx_unlock(sema_mtxp);
787 if ((error = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
788 goto done2;
789 array = malloc(sizeof(*array) * count, M_TEMP, M_WAITOK);
790 error = copyin(real_arg.array, array, count * sizeof(*array));
791 if (error)
792 break;
793 mtx_lock(sema_mtxp);
794 if ((error = semvalid(uap->semid, semakptr)) != 0)
795 goto done2;
796 /* we could have raced? */
797 if (count != semakptr->u.sem_nsems) {
798 free(array, M_TEMP);
799 array = NULL;
800 goto raced;
801 }
802 if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_W)))
803 goto done2;
804 for (i = 0; i < semakptr->u.sem_nsems; i++) {
805 usval = array[i];
806 if (usval > seminfo.semvmx) {
807 error = ERANGE;
808 break;
809 }
810 semakptr->u.sem_base[i].semval = usval;
811 }
812 SEMUNDO_LOCK();
813 semundo_clear(semid, -1);
814 SEMUNDO_UNLOCK();
815 wakeup(semakptr);
816 break;
817
818 default:
819 error = EINVAL;
820 break;
821 }
822
823 if (error == 0)
824 td->td_retval[0] = rval;
825done2:
826 if (mtx_owned(sema_mtxp))
827 mtx_unlock(sema_mtxp);
828 if (array != NULL)
829 free(array, M_TEMP);
830 return(error);
831}
832
833#ifndef _SYS_SYSPROTO_H_
834struct semget_args {
835 key_t key;
836 int nsems;
837 int semflg;
838};
839#endif
840
841/*
842 * MPSAFE
843 */
844int
845semget(td, uap)
846 struct thread *td;
847 struct semget_args *uap;
848{
849 int semid, error = 0;
850 int key = uap->key;
851 int nsems = uap->nsems;
852 int semflg = uap->semflg;
853 struct ucred *cred = td->td_ucred;
854
855 DPRINTF(("semget(0x%x, %d, 0%o)\n", key, nsems, semflg));
856 if (!jail_sysvipc_allowed && jailed(td->td_ucred))
857 return (ENOSYS);
858
859 mtx_lock(&Giant);
860 if (key != IPC_PRIVATE) {
861 for (semid = 0; semid < seminfo.semmni; semid++) {
862 if ((sema[semid].u.sem_perm.mode & SEM_ALLOC) &&
863 sema[semid].u.sem_perm.key == key)
864 break;
865 }
866 if (semid < seminfo.semmni) {
867 DPRINTF(("found public key\n"));
868 if ((error = ipcperm(td, &sema[semid].u.sem_perm,
869 semflg & 0700))) {
870 goto done2;
871 }
872 if (nsems > 0 && sema[semid].u.sem_nsems < nsems) {
873 DPRINTF(("too small\n"));
874 error = EINVAL;
875 goto done2;
876 }
877 if ((semflg & IPC_CREAT) && (semflg & IPC_EXCL)) {
878 DPRINTF(("not exclusive\n"));
879 error = EEXIST;
880 goto done2;
881 }
882#ifdef MAC
883 error = mac_check_sysv_semget(cred, &sema[semid]);
884 if (error != 0) {
885 MPRINTF(("mac_check_sysv_semget returned %d\n",
886 error));
887 goto done2;
888 }
889#endif
890 goto found;
891 }
892 }
893
894 DPRINTF(("need to allocate the semid_kernel\n"));
895 if (key == IPC_PRIVATE || (semflg & IPC_CREAT)) {
896 if (nsems <= 0 || nsems > seminfo.semmsl) {
897 DPRINTF(("nsems out of range (0<%d<=%d)\n", nsems,
898 seminfo.semmsl));
899 error = EINVAL;
900 goto done2;
901 }
902 if (nsems > seminfo.semmns - semtot) {
903 DPRINTF((
904 "not enough semaphores left (need %d, got %d)\n",
905 nsems, seminfo.semmns - semtot));
906 error = ENOSPC;
907 goto done2;
908 }
909 for (semid = 0; semid < seminfo.semmni; semid++) {
910 if ((sema[semid].u.sem_perm.mode & SEM_ALLOC) == 0)
911 break;
912 }
913 if (semid == seminfo.semmni) {
914 DPRINTF(("no more semid_kernel's available\n"));
915 error = ENOSPC;
916 goto done2;
917 }
918 DPRINTF(("semid %d is available\n", semid));
919 sema[semid].u.sem_perm.key = key;
920 sema[semid].u.sem_perm.cuid = cred->cr_uid;
921 sema[semid].u.sem_perm.uid = cred->cr_uid;
922 sema[semid].u.sem_perm.cgid = cred->cr_gid;
923 sema[semid].u.sem_perm.gid = cred->cr_gid;
924 sema[semid].u.sem_perm.mode = (semflg & 0777) | SEM_ALLOC;
925 sema[semid].u.sem_perm.seq =
926 (sema[semid].u.sem_perm.seq + 1) & 0x7fff;
927 sema[semid].u.sem_nsems = nsems;
928 sema[semid].u.sem_otime = 0;
929 sema[semid].u.sem_ctime = time_second;
930 sema[semid].u.sem_base = &sem[semtot];
931 semtot += nsems;
932 bzero(sema[semid].u.sem_base,
933 sizeof(sema[semid].u.sem_base[0])*nsems);
934#ifdef MAC
935 mac_create_sysv_sema(cred, &sema[semid]);
936#endif
937 DPRINTF(("sembase = 0x%x, next = 0x%x\n",
938 sema[semid].u.sem_base, &sem[semtot]));
939 } else {
940 DPRINTF(("didn't find it and wasn't asked to create it\n"));
941 error = ENOENT;
942 goto done2;
943 }
944
945found:
946 td->td_retval[0] = IXSEQ_TO_IPCID(semid, sema[semid].u.sem_perm);
947done2:
948 mtx_unlock(&Giant);
949 return (error);
950}
951
952#ifndef _SYS_SYSPROTO_H_
953struct semop_args {
954 int semid;
955 struct sembuf *sops;
956 size_t nsops;
957};
958#endif
959
960/*
961 * MPSAFE
962 */
963int
964semop(td, uap)
965 struct thread *td;
966 struct semop_args *uap;
967{
968#define SMALL_SOPS 8
969 struct sembuf small_sops[SMALL_SOPS];
970 int semid = uap->semid;
971 size_t nsops = uap->nsops;
972 struct sembuf *sops;
973 struct semid_kernel *semakptr;
974 struct sembuf *sopptr = 0;
975 struct sem *semptr = 0;
976 struct sem_undo *suptr;
977 struct mtx *sema_mtxp;
978 size_t i, j, k;
979 int error;
980 int do_wakeup, do_undos;
981
982 DPRINTF(("call to semop(%d, 0x%x, %u)\n", semid, sops, nsops));
983
984 if (!jail_sysvipc_allowed && jailed(td->td_ucred))
985 return (ENOSYS);
986
987 semid = IPCID_TO_IX(semid); /* Convert back to zero origin */
988
989 if (semid < 0 || semid >= seminfo.semmni)
990 return (EINVAL);
991
992 /* Allocate memory for sem_ops */
993 if (nsops <= SMALL_SOPS)
994 sops = small_sops;
995 else if (nsops <= seminfo.semopm)
996 sops = malloc(nsops * sizeof(*sops), M_TEMP, M_WAITOK);
997 else {
998 DPRINTF(("too many sops (max=%d, nsops=%d)\n", seminfo.semopm,
999 nsops));
1000 return (E2BIG);
1001 }
1002 if ((error = copyin(uap->sops, sops, nsops * sizeof(sops[0]))) != 0) {
1003 DPRINTF(("error = %d from copyin(%08x, %08x, %d)\n", error,
1004 uap->sops, sops, nsops * sizeof(sops[0])));
1005 if (sops != small_sops)
1006 free(sops, M_SEM);
1007 return (error);
1008 }
1009
1010 semakptr = &sema[semid];
1011 sema_mtxp = &sema_mtx[semid];
1012 mtx_lock(sema_mtxp);
1013 if ((semakptr->u.sem_perm.mode & SEM_ALLOC) == 0) {
1014 error = EINVAL;
1015 goto done2;
1016 }
1017 if (semakptr->u.sem_perm.seq != IPCID_TO_SEQ(uap->semid)) {
1018 error = EINVAL;
1019 goto done2;
1020 }
1021 /*
1022 * Initial pass thru sops to see what permissions are needed.
1023 * Also perform any checks that don't need repeating on each
1024 * attempt to satisfy the request vector.
1025 */
1026 j = 0; /* permission needed */
1027 do_undos = 0;
1028 for (i = 0; i < nsops; i++) {
1029 sopptr = &sops[i];
1030 if (sopptr->sem_num >= semakptr->u.sem_nsems) {
1031 error = EFBIG;
1032 goto done2;
1033 }
1034 if (sopptr->sem_flg & SEM_UNDO && sopptr->sem_op != 0)
1035 do_undos = 1;
1036 j |= (sopptr->sem_op == 0) ? SEM_R : SEM_A;
1037 }
1038
1039 if ((error = ipcperm(td, &semakptr->u.sem_perm, j))) {
1040 DPRINTF(("error = %d from ipaccess\n", error));
1041 goto done2;
1042 }
1043#ifdef MAC
1044 error = mac_check_sysv_semop(td->td_ucred, semakptr, j);
1045 if (error != 0) {
1046 MPRINTF(("mac_check_sysv_semop returned %d\n", error));
1047 goto done2;
1048 }
1049#endif
1050
1051 /*
1052 * Loop trying to satisfy the vector of requests.
1053 * If we reach a point where we must wait, any requests already
1054 * performed are rolled back and we go to sleep until some other
1055 * process wakes us up. At this point, we start all over again.
1056 *
1057 * This ensures that from the perspective of other tasks, a set
1058 * of requests is atomic (never partially satisfied).
1059 */
1060 for (;;) {
1061 do_wakeup = 0;
1062 error = 0; /* error return if necessary */
1063
1064 for (i = 0; i < nsops; i++) {
1065 sopptr = &sops[i];
1066 semptr = &semakptr->u.sem_base[sopptr->sem_num];
1067
1068 DPRINTF((
1069 "semop: semakptr=%x, sem_base=%x, "
1070 "semptr=%x, sem[%d]=%d : op=%d, flag=%s\n",
1071 semakptr, semakptr->u.sem_base, semptr,
1072 sopptr->sem_num, semptr->semval, sopptr->sem_op,
1073 (sopptr->sem_flg & IPC_NOWAIT) ?
1074 "nowait" : "wait"));
1075
1076 if (sopptr->sem_op < 0) {
1077 if (semptr->semval + sopptr->sem_op < 0) {
1078 DPRINTF(("semop: can't do it now\n"));
1079 break;
1080 } else {
1081 semptr->semval += sopptr->sem_op;
1082 if (semptr->semval == 0 &&
1083 semptr->semzcnt > 0)
1084 do_wakeup = 1;
1085 }
1086 } else if (sopptr->sem_op == 0) {
1087 if (semptr->semval != 0) {
1088 DPRINTF(("semop: not zero now\n"));
1089 break;
1090 }
1091 } else if (semptr->semval + sopptr->sem_op >
1092 seminfo.semvmx) {
1093 error = ERANGE;
1094 break;
1095 } else {
1096 if (semptr->semncnt > 0)
1097 do_wakeup = 1;
1098 semptr->semval += sopptr->sem_op;
1099 }
1100 }
1101
1102 /*
1103 * Did we get through the entire vector?
1104 */
1105 if (i >= nsops)
1106 goto done;
1107
1108 /*
1109 * No ... rollback anything that we've already done
1110 */
1111 DPRINTF(("semop: rollback 0 through %d\n", i-1));
1112 for (j = 0; j < i; j++)
1113 semakptr->u.sem_base[sops[j].sem_num].semval -=
1114 sops[j].sem_op;
1115
1116 /* If we detected an error, return it */
1117 if (error != 0)
1118 goto done2;
1119
1120 /*
1121 * If the request that we couldn't satisfy has the
1122 * NOWAIT flag set then return with EAGAIN.
1123 */
1124 if (sopptr->sem_flg & IPC_NOWAIT) {
1125 error = EAGAIN;
1126 goto done2;
1127 }
1128
1129 if (sopptr->sem_op == 0)
1130 semptr->semzcnt++;
1131 else
1132 semptr->semncnt++;
1133
1134 DPRINTF(("semop: good night!\n"));
1135 error = msleep(semakptr, sema_mtxp, (PZERO - 4) | PCATCH,
1136 "semwait", 0);
1137 DPRINTF(("semop: good morning (error=%d)!\n", error));
1138 /* return code is checked below, after sem[nz]cnt-- */
1139
1140 /*
1141 * Make sure that the semaphore still exists
1142 */
1143 if ((semakptr->u.sem_perm.mode & SEM_ALLOC) == 0 ||
1144 semakptr->u.sem_perm.seq != IPCID_TO_SEQ(uap->semid)) {
1145 error = EIDRM;
1146 goto done2;
1147 }
1148
1149 /*
1150 * The semaphore is still alive. Readjust the count of
1151 * waiting processes.
1152 */
1153 if (sopptr->sem_op == 0)
1154 semptr->semzcnt--;
1155 else
1156 semptr->semncnt--;
1157
1158 /*
1159 * Is it really morning, or was our sleep interrupted?
1160 * (Delayed check of msleep() return code because we
1161 * need to decrement sem[nz]cnt either way.)
1162 */
1163 if (error != 0) {
1164 error = EINTR;
1165 goto done2;
1166 }
1167 DPRINTF(("semop: good morning!\n"));
1168 }
1169
1170done:
1171 /*
1172 * Process any SEM_UNDO requests.
1173 */
1174 if (do_undos) {
1175 SEMUNDO_LOCK();
1176 suptr = NULL;
1177 for (i = 0; i < nsops; i++) {
1178 /*
1179 * We only need to deal with SEM_UNDO's for non-zero
1180 * op's.
1181 */
1182 int adjval;
1183
1184 if ((sops[i].sem_flg & SEM_UNDO) == 0)
1185 continue;
1186 adjval = sops[i].sem_op;
1187 if (adjval == 0)
1188 continue;
1189 error = semundo_adjust(td, &suptr, semid,
1190 sops[i].sem_num, -adjval);
1191 if (error == 0)
1192 continue;
1193
1194 /*
1195 * Oh-Oh! We ran out of either sem_undo's or undo's.
1196 * Rollback the adjustments to this point and then
1197 * rollback the semaphore ups and down so we can return
1198 * with an error with all structures restored. We
1199 * rollback the undo's in the exact reverse order that
1200 * we applied them. This guarantees that we won't run
1201 * out of space as we roll things back out.
1202 */
1203 for (j = 0; j < i; j++) {
1204 k = i - j - 1;
1205 if ((sops[k].sem_flg & SEM_UNDO) == 0)
1206 continue;
1207 adjval = sops[k].sem_op;
1208 if (adjval == 0)
1209 continue;
1210 if (semundo_adjust(td, &suptr, semid,
1211 sops[k].sem_num, adjval) != 0)
1212 panic("semop - can't undo undos");
1213 }
1214
1215 for (j = 0; j < nsops; j++)
1216 semakptr->u.sem_base[sops[j].sem_num].semval -=
1217 sops[j].sem_op;
1218
1219 DPRINTF(("error = %d from semundo_adjust\n", error));
1220 SEMUNDO_UNLOCK();
1221 goto done2;
1222 } /* loop through the sops */
1223 SEMUNDO_UNLOCK();
1224 } /* if (do_undos) */
1225
1226 /* We're definitely done - set the sempid's and time */
1227 for (i = 0; i < nsops; i++) {
1228 sopptr = &sops[i];
1229 semptr = &semakptr->u.sem_base[sopptr->sem_num];
1230 semptr->sempid = td->td_proc->p_pid;
1231 }
1232 semakptr->u.sem_otime = time_second;
1233
1234 /*
1235 * Do a wakeup if any semaphore was up'd whilst something was
1236 * sleeping on it.
1237 */
1238 if (do_wakeup) {
1239 DPRINTF(("semop: doing wakeup\n"));
1240 wakeup(semakptr);
1241 DPRINTF(("semop: back from wakeup\n"));
1242 }
1243 DPRINTF(("semop: done\n"));
1244 td->td_retval[0] = 0;
1245done2:
1246 mtx_unlock(sema_mtxp);
1247 if (sops != small_sops)
1248 free(sops, M_SEM);
1249 return (error);
1250}
1251
1252/*
1253 * Go through the undo structures for this process and apply the adjustments to
1254 * semaphores.
1255 */
1256static void
1257semexit_myhook(arg, p)
1258 void *arg;
1259 struct proc *p;
1260{
1261 struct sem_undo *suptr;
1262 struct sem_undo **supptr;
1263
1264 /*
1265 * Go through the chain of undo vectors looking for one
1266 * associated with this process.
1267 */
1268 SEMUNDO_LOCK();
1269 SLIST_FOREACH_PREVPTR(suptr, supptr, &semu_list, un_next) {
1270 if (suptr->un_proc == p)
1271 break;
1272 }
1273 SEMUNDO_UNLOCK();
1274
1275 if (suptr == NULL)
1276 return;
1277
1278 DPRINTF(("proc @%08x has undo structure with %d entries\n", p,
1279 suptr->un_cnt));
1280
1281 /*
1282 * If there are any active undo elements then process them.
1283 */
1284 if (suptr->un_cnt > 0) {
1285 int ix;
1286
1287 for (ix = 0; ix < suptr->un_cnt; ix++) {
1288 int semid = suptr->un_ent[ix].un_id;
1289 int semnum = suptr->un_ent[ix].un_num;
1290 int adjval = suptr->un_ent[ix].un_adjval;
1291 struct semid_kernel *semakptr;
1292 struct mtx *sema_mtxp;
1293
1294 semakptr = &sema[semid];
1295 sema_mtxp = &sema_mtx[semid];
1296 mtx_lock(sema_mtxp);
1297 SEMUNDO_LOCK();
1298 if ((semakptr->u.sem_perm.mode & SEM_ALLOC) == 0)
1299 panic("semexit - semid not allocated");
1300 if (semnum >= semakptr->u.sem_nsems)
1301 panic("semexit - semnum out of range");
1302
1303 DPRINTF((
1304 "semexit: %08x id=%d num=%d(adj=%d) ; sem=%d\n",
1305 suptr->un_proc, suptr->un_ent[ix].un_id,
1306 suptr->un_ent[ix].un_num,
1307 suptr->un_ent[ix].un_adjval,
1308 semakptr->u.sem_base[semnum].semval));
1309
1310 if (adjval < 0) {
1311 if (semakptr->u.sem_base[semnum].semval <
1312 -adjval)
1313 semakptr->u.sem_base[semnum].semval = 0;
1314 else
1315 semakptr->u.sem_base[semnum].semval +=
1316 adjval;
1317 } else
1318 semakptr->u.sem_base[semnum].semval += adjval;
1319
1320 wakeup(semakptr);
1321 DPRINTF(("semexit: back from wakeup\n"));
1322 mtx_unlock(sema_mtxp);
1323 SEMUNDO_UNLOCK();
1324 }
1325 }
1326
1327 /*
1328 * Deallocate the undo vector.
1329 */
1330 DPRINTF(("removing vector\n"));
1331 suptr->un_proc = NULL;
1332 *supptr = SLIST_NEXT(suptr, un_next);
1333}
1334
1335static int
1336sysctl_sema(SYSCTL_HANDLER_ARGS)
1337{
1338
1339 return (SYSCTL_OUT(req, sema,
1340 sizeof(struct semid_kernel) * seminfo.semmni));
1341}
2 * Implementation of SVID semaphores
3 *
4 * Author: Daniel Boulet
5 *
6 * This software is provided ``AS IS'' without any warranties of any kind.
7 */
8/*-
9 * Copyright (c) 2003-2005 McAfee, Inc.
10 * All rights reserved.
11 *
12 * This software was developed for the FreeBSD Project in part by McAfee
13 * Research, the Security Research Division of McAfee, Inc under DARPA/SPAWAR
14 * contract N66001-01-C-8035 ("CBOSS"), as part of the DARPA CHATS research
15 * program.
16 *
17 * Redistribution and use in source and binary forms, with or without
18 * modification, are permitted provided that the following conditions
19 * are met:
20 * 1. Redistributions of source code must retain the above copyright
21 * notice, this list of conditions and the following disclaimer.
22 * 2. Redistributions in binary form must reproduce the above copyright
23 * notice, this list of conditions and the following disclaimer in the
24 * documentation and/or other materials provided with the distribution.
25 *
26 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36 * SUCH DAMAGE.
37 */
38
39#include <sys/cdefs.h>
40__FBSDID("$FreeBSD: head/sys/kern/sysv_sem.c 145230 2005-04-18 11:51:10Z rwatson $");
41
42#include "opt_sysvipc.h"
43#include "opt_mac.h"
44
45#include <sys/param.h>
46#include <sys/systm.h>
47#include <sys/sysproto.h>
48#include <sys/eventhandler.h>
49#include <sys/kernel.h>
50#include <sys/proc.h>
51#include <sys/lock.h>
52#include <sys/module.h>
53#include <sys/mutex.h>
54#include <sys/sem.h>
55#include <sys/syscall.h>
56#include <sys/sysent.h>
57#include <sys/sysctl.h>
58#include <sys/malloc.h>
59#include <sys/jail.h>
60#include <sys/mac.h>
61
62static MALLOC_DEFINE(M_SEM, "sem", "SVID compatible semaphores");
63
64#ifdef SEM_DEBUG
65#define DPRINTF(a) printf a
66#else
67#define DPRINTF(a)
68#endif
69#ifdef MAC_DEBUG
70#define MPRINTF(a) printf a
71#else
72#define MPRINTF(a)
73#endif
74
75static void seminit(void);
76static int sysvsem_modload(struct module *, int, void *);
77static int semunload(void);
78static void semexit_myhook(void *arg, struct proc *p);
79static int sysctl_sema(SYSCTL_HANDLER_ARGS);
80static int semvalid(int semid, struct semid_kernel *semakptr);
81
82#ifndef _SYS_SYSPROTO_H_
83struct __semctl_args;
84int __semctl(struct thread *td, struct __semctl_args *uap);
85struct semget_args;
86int semget(struct thread *td, struct semget_args *uap);
87struct semop_args;
88int semop(struct thread *td, struct semop_args *uap);
89#endif
90
91static struct sem_undo *semu_alloc(struct thread *td);
92static int semundo_adjust(struct thread *td, struct sem_undo **supptr,
93 int semid, int semnum, int adjval);
94static void semundo_clear(int semid, int semnum);
95
96/* XXX casting to (sy_call_t *) is bogus, as usual. */
97static sy_call_t *semcalls[] = {
98 (sy_call_t *)__semctl, (sy_call_t *)semget,
99 (sy_call_t *)semop
100};
101
102static struct mtx sem_mtx; /* semaphore global lock */
103static int semtot = 0;
104static struct semid_kernel *sema; /* semaphore id pool */
105static struct mtx *sema_mtx; /* semaphore id pool mutexes*/
106static struct sem *sem; /* semaphore pool */
107SLIST_HEAD(, sem_undo) semu_list; /* list of active undo structures */
108static int *semu; /* undo structure pool */
109static eventhandler_tag semexit_tag;
110
111#define SEMUNDO_MTX sem_mtx
112#define SEMUNDO_LOCK() mtx_lock(&SEMUNDO_MTX);
113#define SEMUNDO_UNLOCK() mtx_unlock(&SEMUNDO_MTX);
114#define SEMUNDO_LOCKASSERT(how) mtx_assert(&SEMUNDO_MTX, (how));
115
116struct sem {
117 u_short semval; /* semaphore value */
118 pid_t sempid; /* pid of last operation */
119 u_short semncnt; /* # awaiting semval > cval */
120 u_short semzcnt; /* # awaiting semval = 0 */
121};
122
123/*
124 * Undo structure (one per process)
125 */
126struct sem_undo {
127 SLIST_ENTRY(sem_undo) un_next; /* ptr to next active undo structure */
128 struct proc *un_proc; /* owner of this structure */
129 short un_cnt; /* # of active entries */
130 struct undo {
131 short un_adjval; /* adjust on exit values */
132 short un_num; /* semaphore # */
133 int un_id; /* semid */
134 } un_ent[1]; /* undo entries */
135};
136
137/*
138 * Configuration parameters
139 */
140#ifndef SEMMNI
141#define SEMMNI 10 /* # of semaphore identifiers */
142#endif
143#ifndef SEMMNS
144#define SEMMNS 60 /* # of semaphores in system */
145#endif
146#ifndef SEMUME
147#define SEMUME 10 /* max # of undo entries per process */
148#endif
149#ifndef SEMMNU
150#define SEMMNU 30 /* # of undo structures in system */
151#endif
152
153/* shouldn't need tuning */
154#ifndef SEMMAP
155#define SEMMAP 30 /* # of entries in semaphore map */
156#endif
157#ifndef SEMMSL
158#define SEMMSL SEMMNS /* max # of semaphores per id */
159#endif
160#ifndef SEMOPM
161#define SEMOPM 100 /* max # of operations per semop call */
162#endif
163
164#define SEMVMX 32767 /* semaphore maximum value */
165#define SEMAEM 16384 /* adjust on exit max value */
166
167/*
168 * Due to the way semaphore memory is allocated, we have to ensure that
169 * SEMUSZ is properly aligned.
170 */
171
172#define SEM_ALIGN(bytes) (((bytes) + (sizeof(long) - 1)) & ~(sizeof(long) - 1))
173
174/* actual size of an undo structure */
175#define SEMUSZ SEM_ALIGN(offsetof(struct sem_undo, un_ent[SEMUME]))
176
177/*
178 * Macro to find a particular sem_undo vector
179 */
180#define SEMU(ix) \
181 ((struct sem_undo *)(((intptr_t)semu)+ix * seminfo.semusz))
182
183/*
184 * semaphore info struct
185 */
186struct seminfo seminfo = {
187 SEMMAP, /* # of entries in semaphore map */
188 SEMMNI, /* # of semaphore identifiers */
189 SEMMNS, /* # of semaphores in system */
190 SEMMNU, /* # of undo structures in system */
191 SEMMSL, /* max # of semaphores per id */
192 SEMOPM, /* max # of operations per semop call */
193 SEMUME, /* max # of undo entries per process */
194 SEMUSZ, /* size in bytes of undo structure */
195 SEMVMX, /* semaphore maximum value */
196 SEMAEM /* adjust on exit max value */
197};
198
199SYSCTL_DECL(_kern_ipc);
200SYSCTL_INT(_kern_ipc, OID_AUTO, semmap, CTLFLAG_RW, &seminfo.semmap, 0,
201 "Number of entries in the semaphore map");
202SYSCTL_INT(_kern_ipc, OID_AUTO, semmni, CTLFLAG_RDTUN, &seminfo.semmni, 0,
203 "Number of semaphore identifiers");
204SYSCTL_INT(_kern_ipc, OID_AUTO, semmns, CTLFLAG_RDTUN, &seminfo.semmns, 0,
205 "Maximum number of semaphores in the system");
206SYSCTL_INT(_kern_ipc, OID_AUTO, semmnu, CTLFLAG_RDTUN, &seminfo.semmnu, 0,
207 "Maximum number of undo structures in the system");
208SYSCTL_INT(_kern_ipc, OID_AUTO, semmsl, CTLFLAG_RW, &seminfo.semmsl, 0,
209 "Max semaphores per id");
210SYSCTL_INT(_kern_ipc, OID_AUTO, semopm, CTLFLAG_RDTUN, &seminfo.semopm, 0,
211 "Max operations per semop call");
212SYSCTL_INT(_kern_ipc, OID_AUTO, semume, CTLFLAG_RDTUN, &seminfo.semume, 0,
213 "Max undo entries per process");
214SYSCTL_INT(_kern_ipc, OID_AUTO, semusz, CTLFLAG_RDTUN, &seminfo.semusz, 0,
215 "Size in bytes of undo structure");
216SYSCTL_INT(_kern_ipc, OID_AUTO, semvmx, CTLFLAG_RW, &seminfo.semvmx, 0,
217 "Semaphore maximum value");
218SYSCTL_INT(_kern_ipc, OID_AUTO, semaem, CTLFLAG_RW, &seminfo.semaem, 0,
219 "Adjust on exit max value");
220SYSCTL_PROC(_kern_ipc, OID_AUTO, sema, CTLFLAG_RD,
221 NULL, 0, sysctl_sema, "", "");
222
223static void
224seminit(void)
225{
226 int i;
227
228 TUNABLE_INT_FETCH("kern.ipc.semmap", &seminfo.semmap);
229 TUNABLE_INT_FETCH("kern.ipc.semmni", &seminfo.semmni);
230 TUNABLE_INT_FETCH("kern.ipc.semmns", &seminfo.semmns);
231 TUNABLE_INT_FETCH("kern.ipc.semmnu", &seminfo.semmnu);
232 TUNABLE_INT_FETCH("kern.ipc.semmsl", &seminfo.semmsl);
233 TUNABLE_INT_FETCH("kern.ipc.semopm", &seminfo.semopm);
234 TUNABLE_INT_FETCH("kern.ipc.semume", &seminfo.semume);
235 TUNABLE_INT_FETCH("kern.ipc.semusz", &seminfo.semusz);
236 TUNABLE_INT_FETCH("kern.ipc.semvmx", &seminfo.semvmx);
237 TUNABLE_INT_FETCH("kern.ipc.semaem", &seminfo.semaem);
238
239 sem = malloc(sizeof(struct sem) * seminfo.semmns, M_SEM, M_WAITOK);
240 sema = malloc(sizeof(struct semid_kernel) * seminfo.semmni, M_SEM,
241 M_WAITOK);
242 sema_mtx = malloc(sizeof(struct mtx) * seminfo.semmni, M_SEM,
243 M_WAITOK | M_ZERO);
244 semu = malloc(seminfo.semmnu * seminfo.semusz, M_SEM, M_WAITOK);
245
246 for (i = 0; i < seminfo.semmni; i++) {
247 sema[i].u.sem_base = 0;
248 sema[i].u.sem_perm.mode = 0;
249 sema[i].u.sem_perm.seq = 0;
250#ifdef MAC
251 mac_init_sysv_sema(&sema[i]);
252#endif
253 }
254 for (i = 0; i < seminfo.semmni; i++)
255 mtx_init(&sema_mtx[i], "semid", NULL, MTX_DEF);
256 for (i = 0; i < seminfo.semmnu; i++) {
257 struct sem_undo *suptr = SEMU(i);
258 suptr->un_proc = NULL;
259 }
260 SLIST_INIT(&semu_list);
261 mtx_init(&sem_mtx, "sem", NULL, MTX_DEF);
262 semexit_tag = EVENTHANDLER_REGISTER(process_exit, semexit_myhook, NULL,
263 EVENTHANDLER_PRI_ANY);
264}
265
266static int
267semunload(void)
268{
269 int i;
270
271 if (semtot != 0)
272 return (EBUSY);
273
274 EVENTHANDLER_DEREGISTER(process_exit, semexit_tag);
275#ifdef MAC
276 for (i = 0; i < seminfo.semmni; i++)
277 mac_destroy_sysv_sema(&sema[i]);
278#endif
279 free(sem, M_SEM);
280 free(sema, M_SEM);
281 free(semu, M_SEM);
282 for (i = 0; i < seminfo.semmni; i++)
283 mtx_destroy(&sema_mtx[i]);
284 mtx_destroy(&sem_mtx);
285 return (0);
286}
287
288static int
289sysvsem_modload(struct module *module, int cmd, void *arg)
290{
291 int error = 0;
292
293 switch (cmd) {
294 case MOD_LOAD:
295 seminit();
296 break;
297 case MOD_UNLOAD:
298 error = semunload();
299 break;
300 case MOD_SHUTDOWN:
301 break;
302 default:
303 error = EINVAL;
304 break;
305 }
306 return (error);
307}
308
309static moduledata_t sysvsem_mod = {
310 "sysvsem",
311 &sysvsem_modload,
312 NULL
313};
314
315SYSCALL_MODULE_HELPER(semsys);
316SYSCALL_MODULE_HELPER(__semctl);
317SYSCALL_MODULE_HELPER(semget);
318SYSCALL_MODULE_HELPER(semop);
319
320DECLARE_MODULE(sysvsem, sysvsem_mod,
321 SI_SUB_SYSV_SEM, SI_ORDER_FIRST);
322MODULE_VERSION(sysvsem, 1);
323
324/*
325 * Entry point for all SEM calls
326 *
327 * MPSAFE
328 */
329int
330semsys(td, uap)
331 struct thread *td;
332 /* XXX actually varargs. */
333 struct semsys_args /* {
334 int which;
335 int a2;
336 int a3;
337 int a4;
338 int a5;
339 } */ *uap;
340{
341 int error;
342
343 if (!jail_sysvipc_allowed && jailed(td->td_ucred))
344 return (ENOSYS);
345 if (uap->which < 0 ||
346 uap->which >= sizeof(semcalls)/sizeof(semcalls[0]))
347 return (EINVAL);
348 error = (*semcalls[uap->which])(td, &uap->a2);
349 return (error);
350}
351
352/*
353 * Allocate a new sem_undo structure for a process
354 * (returns ptr to structure or NULL if no more room)
355 */
356
357static struct sem_undo *
358semu_alloc(td)
359 struct thread *td;
360{
361 int i;
362 struct sem_undo *suptr;
363 struct sem_undo **supptr;
364 int attempt;
365
366 SEMUNDO_LOCKASSERT(MA_OWNED);
367 /*
368 * Try twice to allocate something.
369 * (we'll purge an empty structure after the first pass so
370 * two passes are always enough)
371 */
372
373 for (attempt = 0; attempt < 2; attempt++) {
374 /*
375 * Look for a free structure.
376 * Fill it in and return it if we find one.
377 */
378
379 for (i = 0; i < seminfo.semmnu; i++) {
380 suptr = SEMU(i);
381 if (suptr->un_proc == NULL) {
382 SLIST_INSERT_HEAD(&semu_list, suptr, un_next);
383 suptr->un_cnt = 0;
384 suptr->un_proc = td->td_proc;
385 return(suptr);
386 }
387 }
388
389 /*
390 * We didn't find a free one, if this is the first attempt
391 * then try to free a structure.
392 */
393
394 if (attempt == 0) {
395 /* All the structures are in use - try to free one */
396 int did_something = 0;
397
398 SLIST_FOREACH_PREVPTR(suptr, supptr, &semu_list,
399 un_next) {
400 if (suptr->un_cnt == 0) {
401 suptr->un_proc = NULL;
402 did_something = 1;
403 *supptr = SLIST_NEXT(suptr, un_next);
404 break;
405 }
406 }
407
408 /* If we didn't free anything then just give-up */
409 if (!did_something)
410 return(NULL);
411 } else {
412 /*
413 * The second pass failed even though we freed
414 * something after the first pass!
415 * This is IMPOSSIBLE!
416 */
417 panic("semu_alloc - second attempt failed");
418 }
419 }
420 return (NULL);
421}
422
423/*
424 * Adjust a particular entry for a particular proc
425 */
426
427static int
428semundo_adjust(td, supptr, semid, semnum, adjval)
429 struct thread *td;
430 struct sem_undo **supptr;
431 int semid, semnum;
432 int adjval;
433{
434 struct proc *p = td->td_proc;
435 struct sem_undo *suptr;
436 struct undo *sunptr;
437 int i;
438
439 SEMUNDO_LOCKASSERT(MA_OWNED);
440 /* Look for and remember the sem_undo if the caller doesn't provide
441 it */
442
443 suptr = *supptr;
444 if (suptr == NULL) {
445 SLIST_FOREACH(suptr, &semu_list, un_next) {
446 if (suptr->un_proc == p) {
447 *supptr = suptr;
448 break;
449 }
450 }
451 if (suptr == NULL) {
452 if (adjval == 0)
453 return(0);
454 suptr = semu_alloc(td);
455 if (suptr == NULL)
456 return(ENOSPC);
457 *supptr = suptr;
458 }
459 }
460
461 /*
462 * Look for the requested entry and adjust it (delete if adjval becomes
463 * 0).
464 */
465 sunptr = &suptr->un_ent[0];
466 for (i = 0; i < suptr->un_cnt; i++, sunptr++) {
467 if (sunptr->un_id != semid || sunptr->un_num != semnum)
468 continue;
469 if (adjval != 0) {
470 adjval += sunptr->un_adjval;
471 if (adjval > seminfo.semaem || adjval < -seminfo.semaem)
472 return (ERANGE);
473 }
474 sunptr->un_adjval = adjval;
475 if (sunptr->un_adjval == 0) {
476 suptr->un_cnt--;
477 if (i < suptr->un_cnt)
478 suptr->un_ent[i] =
479 suptr->un_ent[suptr->un_cnt];
480 }
481 return(0);
482 }
483
484 /* Didn't find the right entry - create it */
485 if (adjval == 0)
486 return(0);
487 if (adjval > seminfo.semaem || adjval < -seminfo.semaem)
488 return (ERANGE);
489 if (suptr->un_cnt != seminfo.semume) {
490 sunptr = &suptr->un_ent[suptr->un_cnt];
491 suptr->un_cnt++;
492 sunptr->un_adjval = adjval;
493 sunptr->un_id = semid; sunptr->un_num = semnum;
494 } else
495 return(EINVAL);
496 return(0);
497}
498
499static void
500semundo_clear(semid, semnum)
501 int semid, semnum;
502{
503 struct sem_undo *suptr;
504
505 SEMUNDO_LOCKASSERT(MA_OWNED);
506 SLIST_FOREACH(suptr, &semu_list, un_next) {
507 struct undo *sunptr = &suptr->un_ent[0];
508 int i = 0;
509
510 while (i < suptr->un_cnt) {
511 if (sunptr->un_id == semid) {
512 if (semnum == -1 || sunptr->un_num == semnum) {
513 suptr->un_cnt--;
514 if (i < suptr->un_cnt) {
515 suptr->un_ent[i] =
516 suptr->un_ent[suptr->un_cnt];
517 continue;
518 }
519 }
520 if (semnum != -1)
521 break;
522 }
523 i++, sunptr++;
524 }
525 }
526}
527
528static int
529semvalid(semid, semakptr)
530 int semid;
531 struct semid_kernel *semakptr;
532{
533
534 return ((semakptr->u.sem_perm.mode & SEM_ALLOC) == 0 ||
535 semakptr->u.sem_perm.seq != IPCID_TO_SEQ(semid) ? EINVAL : 0);
536}
537
538/*
539 * Note that the user-mode half of this passes a union, not a pointer
540 */
541#ifndef _SYS_SYSPROTO_H_
542struct __semctl_args {
543 int semid;
544 int semnum;
545 int cmd;
546 union semun *arg;
547};
548#endif
549
550/*
551 * MPSAFE
552 */
553int
554__semctl(td, uap)
555 struct thread *td;
556 struct __semctl_args *uap;
557{
558 int semid = uap->semid;
559 int semnum = uap->semnum;
560 int cmd = uap->cmd;
561 u_short *array;
562 union semun *arg = uap->arg;
563 union semun real_arg;
564 struct ucred *cred = td->td_ucred;
565 int i, rval, error;
566 struct semid_ds sbuf;
567 struct semid_kernel *semakptr;
568 struct mtx *sema_mtxp;
569 u_short usval, count;
570
571 DPRINTF(("call to semctl(%d, %d, %d, 0x%x)\n",
572 semid, semnum, cmd, arg));
573 if (!jail_sysvipc_allowed && jailed(td->td_ucred))
574 return (ENOSYS);
575
576 array = NULL;
577
578 switch(cmd) {
579 case SEM_STAT:
580 if (semid < 0 || semid >= seminfo.semmni)
581 return (EINVAL);
582 if ((error = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
583 return (error);
584 semakptr = &sema[semid];
585 sema_mtxp = &sema_mtx[semid];
586 mtx_lock(sema_mtxp);
587 if ((semakptr->u.sem_perm.mode & SEM_ALLOC) == 0) {
588 error = EINVAL;
589 goto done2;
590 }
591 if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
592 goto done2;
593#ifdef MAC
594 error = mac_check_sysv_semctl(cred, semakptr, cmd);
595 if (error != 0) {
596 MPRINTF(("mac_check_sysv_semctl returned %d\n",
597 error));
598 goto done2;
599 }
600#endif
601 mtx_unlock(sema_mtxp);
602 error = copyout(&semakptr->u, real_arg.buf,
603 sizeof(struct semid_ds));
604 rval = IXSEQ_TO_IPCID(semid, semakptr->u.sem_perm);
605 if (error == 0)
606 td->td_retval[0] = rval;
607 return (error);
608 }
609
610 semid = IPCID_TO_IX(semid);
611 if (semid < 0 || semid >= seminfo.semmni)
612 return (EINVAL);
613
614 semakptr = &sema[semid];
615 sema_mtxp = &sema_mtx[semid];
616#ifdef MAC
617 mtx_lock(sema_mtxp);
618 error = mac_check_sysv_semctl(cred, semakptr, cmd);
619 if (error != 0) {
620 MPRINTF(("mac_check_sysv_semctl returned %d\n", error));
621 mtx_unlock(sema_mtxp);
622 return (error);
623 }
624 mtx_unlock(sema_mtxp);
625#endif
626
627 error = 0;
628 rval = 0;
629
630 switch (cmd) {
631 case IPC_RMID:
632 mtx_lock(sema_mtxp);
633 if ((error = semvalid(uap->semid, semakptr)) != 0)
634 goto done2;
635 if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_M)))
636 goto done2;
637 semakptr->u.sem_perm.cuid = cred->cr_uid;
638 semakptr->u.sem_perm.uid = cred->cr_uid;
639 semtot -= semakptr->u.sem_nsems;
640 for (i = semakptr->u.sem_base - sem; i < semtot; i++)
641 sem[i] = sem[i + semakptr->u.sem_nsems];
642 for (i = 0; i < seminfo.semmni; i++) {
643 if ((sema[i].u.sem_perm.mode & SEM_ALLOC) &&
644 sema[i].u.sem_base > semakptr->u.sem_base)
645 sema[i].u.sem_base -= semakptr->u.sem_nsems;
646 }
647 semakptr->u.sem_perm.mode = 0;
648#ifdef MAC
649 mac_cleanup_sysv_sema(semakptr);
650#endif
651 SEMUNDO_LOCK();
652 semundo_clear(semid, -1);
653 SEMUNDO_UNLOCK();
654 wakeup(semakptr);
655 break;
656
657 case IPC_SET:
658 if ((error = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
659 goto done2;
660 if ((error = copyin(real_arg.buf, &sbuf, sizeof(sbuf))) != 0)
661 goto done2;
662 mtx_lock(sema_mtxp);
663 if ((error = semvalid(uap->semid, semakptr)) != 0)
664 goto done2;
665 if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_M)))
666 goto done2;
667 semakptr->u.sem_perm.uid = sbuf.sem_perm.uid;
668 semakptr->u.sem_perm.gid = sbuf.sem_perm.gid;
669 semakptr->u.sem_perm.mode = (semakptr->u.sem_perm.mode &
670 ~0777) | (sbuf.sem_perm.mode & 0777);
671 semakptr->u.sem_ctime = time_second;
672 break;
673
674 case IPC_STAT:
675 if ((error = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
676 goto done2;
677 mtx_lock(sema_mtxp);
678 if ((error = semvalid(uap->semid, semakptr)) != 0)
679 goto done2;
680 if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
681 goto done2;
682 sbuf = semakptr->u;
683 mtx_unlock(sema_mtxp);
684 error = copyout(&semakptr->u, real_arg.buf,
685 sizeof(struct semid_ds));
686 break;
687
688 case GETNCNT:
689 mtx_lock(sema_mtxp);
690 if ((error = semvalid(uap->semid, semakptr)) != 0)
691 goto done2;
692 if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
693 goto done2;
694 if (semnum < 0 || semnum >= semakptr->u.sem_nsems) {
695 error = EINVAL;
696 goto done2;
697 }
698 rval = semakptr->u.sem_base[semnum].semncnt;
699 break;
700
701 case GETPID:
702 mtx_lock(sema_mtxp);
703 if ((error = semvalid(uap->semid, semakptr)) != 0)
704 goto done2;
705 if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
706 goto done2;
707 if (semnum < 0 || semnum >= semakptr->u.sem_nsems) {
708 error = EINVAL;
709 goto done2;
710 }
711 rval = semakptr->u.sem_base[semnum].sempid;
712 break;
713
714 case GETVAL:
715 mtx_lock(sema_mtxp);
716 if ((error = semvalid(uap->semid, semakptr)) != 0)
717 goto done2;
718 if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
719 goto done2;
720 if (semnum < 0 || semnum >= semakptr->u.sem_nsems) {
721 error = EINVAL;
722 goto done2;
723 }
724 rval = semakptr->u.sem_base[semnum].semval;
725 break;
726
727 case GETALL:
728 if ((error = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
729 goto done2;
730 array = malloc(sizeof(*array) * semakptr->u.sem_nsems, M_TEMP,
731 M_WAITOK);
732 mtx_lock(sema_mtxp);
733 if ((error = semvalid(uap->semid, semakptr)) != 0)
734 goto done2;
735 if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
736 goto done2;
737 for (i = 0; i < semakptr->u.sem_nsems; i++)
738 array[i] = semakptr->u.sem_base[i].semval;
739 mtx_unlock(sema_mtxp);
740 error = copyout(array, real_arg.array,
741 i * sizeof(real_arg.array[0]));
742 break;
743
744 case GETZCNT:
745 mtx_lock(sema_mtxp);
746 if ((error = semvalid(uap->semid, semakptr)) != 0)
747 goto done2;
748 if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
749 goto done2;
750 if (semnum < 0 || semnum >= semakptr->u.sem_nsems) {
751 error = EINVAL;
752 goto done2;
753 }
754 rval = semakptr->u.sem_base[semnum].semzcnt;
755 break;
756
757 case SETVAL:
758 if ((error = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
759 goto done2;
760 mtx_lock(sema_mtxp);
761 if ((error = semvalid(uap->semid, semakptr)) != 0)
762 goto done2;
763 if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_W)))
764 goto done2;
765 if (semnum < 0 || semnum >= semakptr->u.sem_nsems) {
766 error = EINVAL;
767 goto done2;
768 }
769 if (real_arg.val < 0 || real_arg.val > seminfo.semvmx) {
770 error = ERANGE;
771 goto done2;
772 }
773 semakptr->u.sem_base[semnum].semval = real_arg.val;
774 SEMUNDO_LOCK();
775 semundo_clear(semid, semnum);
776 SEMUNDO_UNLOCK();
777 wakeup(semakptr);
778 break;
779
780 case SETALL:
781 mtx_lock(sema_mtxp);
782raced:
783 if ((error = semvalid(uap->semid, semakptr)) != 0)
784 goto done2;
785 count = semakptr->u.sem_nsems;
786 mtx_unlock(sema_mtxp);
787 if ((error = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
788 goto done2;
789 array = malloc(sizeof(*array) * count, M_TEMP, M_WAITOK);
790 error = copyin(real_arg.array, array, count * sizeof(*array));
791 if (error)
792 break;
793 mtx_lock(sema_mtxp);
794 if ((error = semvalid(uap->semid, semakptr)) != 0)
795 goto done2;
796 /* we could have raced? */
797 if (count != semakptr->u.sem_nsems) {
798 free(array, M_TEMP);
799 array = NULL;
800 goto raced;
801 }
802 if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_W)))
803 goto done2;
804 for (i = 0; i < semakptr->u.sem_nsems; i++) {
805 usval = array[i];
806 if (usval > seminfo.semvmx) {
807 error = ERANGE;
808 break;
809 }
810 semakptr->u.sem_base[i].semval = usval;
811 }
812 SEMUNDO_LOCK();
813 semundo_clear(semid, -1);
814 SEMUNDO_UNLOCK();
815 wakeup(semakptr);
816 break;
817
818 default:
819 error = EINVAL;
820 break;
821 }
822
823 if (error == 0)
824 td->td_retval[0] = rval;
825done2:
826 if (mtx_owned(sema_mtxp))
827 mtx_unlock(sema_mtxp);
828 if (array != NULL)
829 free(array, M_TEMP);
830 return(error);
831}
832
833#ifndef _SYS_SYSPROTO_H_
834struct semget_args {
835 key_t key;
836 int nsems;
837 int semflg;
838};
839#endif
840
841/*
842 * MPSAFE
843 */
844int
845semget(td, uap)
846 struct thread *td;
847 struct semget_args *uap;
848{
849 int semid, error = 0;
850 int key = uap->key;
851 int nsems = uap->nsems;
852 int semflg = uap->semflg;
853 struct ucred *cred = td->td_ucred;
854
855 DPRINTF(("semget(0x%x, %d, 0%o)\n", key, nsems, semflg));
856 if (!jail_sysvipc_allowed && jailed(td->td_ucred))
857 return (ENOSYS);
858
859 mtx_lock(&Giant);
860 if (key != IPC_PRIVATE) {
861 for (semid = 0; semid < seminfo.semmni; semid++) {
862 if ((sema[semid].u.sem_perm.mode & SEM_ALLOC) &&
863 sema[semid].u.sem_perm.key == key)
864 break;
865 }
866 if (semid < seminfo.semmni) {
867 DPRINTF(("found public key\n"));
868 if ((error = ipcperm(td, &sema[semid].u.sem_perm,
869 semflg & 0700))) {
870 goto done2;
871 }
872 if (nsems > 0 && sema[semid].u.sem_nsems < nsems) {
873 DPRINTF(("too small\n"));
874 error = EINVAL;
875 goto done2;
876 }
877 if ((semflg & IPC_CREAT) && (semflg & IPC_EXCL)) {
878 DPRINTF(("not exclusive\n"));
879 error = EEXIST;
880 goto done2;
881 }
882#ifdef MAC
883 error = mac_check_sysv_semget(cred, &sema[semid]);
884 if (error != 0) {
885 MPRINTF(("mac_check_sysv_semget returned %d\n",
886 error));
887 goto done2;
888 }
889#endif
890 goto found;
891 }
892 }
893
894 DPRINTF(("need to allocate the semid_kernel\n"));
895 if (key == IPC_PRIVATE || (semflg & IPC_CREAT)) {
896 if (nsems <= 0 || nsems > seminfo.semmsl) {
897 DPRINTF(("nsems out of range (0<%d<=%d)\n", nsems,
898 seminfo.semmsl));
899 error = EINVAL;
900 goto done2;
901 }
902 if (nsems > seminfo.semmns - semtot) {
903 DPRINTF((
904 "not enough semaphores left (need %d, got %d)\n",
905 nsems, seminfo.semmns - semtot));
906 error = ENOSPC;
907 goto done2;
908 }
909 for (semid = 0; semid < seminfo.semmni; semid++) {
910 if ((sema[semid].u.sem_perm.mode & SEM_ALLOC) == 0)
911 break;
912 }
913 if (semid == seminfo.semmni) {
914 DPRINTF(("no more semid_kernel's available\n"));
915 error = ENOSPC;
916 goto done2;
917 }
918 DPRINTF(("semid %d is available\n", semid));
919 sema[semid].u.sem_perm.key = key;
920 sema[semid].u.sem_perm.cuid = cred->cr_uid;
921 sema[semid].u.sem_perm.uid = cred->cr_uid;
922 sema[semid].u.sem_perm.cgid = cred->cr_gid;
923 sema[semid].u.sem_perm.gid = cred->cr_gid;
924 sema[semid].u.sem_perm.mode = (semflg & 0777) | SEM_ALLOC;
925 sema[semid].u.sem_perm.seq =
926 (sema[semid].u.sem_perm.seq + 1) & 0x7fff;
927 sema[semid].u.sem_nsems = nsems;
928 sema[semid].u.sem_otime = 0;
929 sema[semid].u.sem_ctime = time_second;
930 sema[semid].u.sem_base = &sem[semtot];
931 semtot += nsems;
932 bzero(sema[semid].u.sem_base,
933 sizeof(sema[semid].u.sem_base[0])*nsems);
934#ifdef MAC
935 mac_create_sysv_sema(cred, &sema[semid]);
936#endif
937 DPRINTF(("sembase = 0x%x, next = 0x%x\n",
938 sema[semid].u.sem_base, &sem[semtot]));
939 } else {
940 DPRINTF(("didn't find it and wasn't asked to create it\n"));
941 error = ENOENT;
942 goto done2;
943 }
944
945found:
946 td->td_retval[0] = IXSEQ_TO_IPCID(semid, sema[semid].u.sem_perm);
947done2:
948 mtx_unlock(&Giant);
949 return (error);
950}
951
952#ifndef _SYS_SYSPROTO_H_
953struct semop_args {
954 int semid;
955 struct sembuf *sops;
956 size_t nsops;
957};
958#endif
959
960/*
961 * MPSAFE
962 */
963int
964semop(td, uap)
965 struct thread *td;
966 struct semop_args *uap;
967{
968#define SMALL_SOPS 8
969 struct sembuf small_sops[SMALL_SOPS];
970 int semid = uap->semid;
971 size_t nsops = uap->nsops;
972 struct sembuf *sops;
973 struct semid_kernel *semakptr;
974 struct sembuf *sopptr = 0;
975 struct sem *semptr = 0;
976 struct sem_undo *suptr;
977 struct mtx *sema_mtxp;
978 size_t i, j, k;
979 int error;
980 int do_wakeup, do_undos;
981
982 DPRINTF(("call to semop(%d, 0x%x, %u)\n", semid, sops, nsops));
983
984 if (!jail_sysvipc_allowed && jailed(td->td_ucred))
985 return (ENOSYS);
986
987 semid = IPCID_TO_IX(semid); /* Convert back to zero origin */
988
989 if (semid < 0 || semid >= seminfo.semmni)
990 return (EINVAL);
991
992 /* Allocate memory for sem_ops */
993 if (nsops <= SMALL_SOPS)
994 sops = small_sops;
995 else if (nsops <= seminfo.semopm)
996 sops = malloc(nsops * sizeof(*sops), M_TEMP, M_WAITOK);
997 else {
998 DPRINTF(("too many sops (max=%d, nsops=%d)\n", seminfo.semopm,
999 nsops));
1000 return (E2BIG);
1001 }
1002 if ((error = copyin(uap->sops, sops, nsops * sizeof(sops[0]))) != 0) {
1003 DPRINTF(("error = %d from copyin(%08x, %08x, %d)\n", error,
1004 uap->sops, sops, nsops * sizeof(sops[0])));
1005 if (sops != small_sops)
1006 free(sops, M_SEM);
1007 return (error);
1008 }
1009
1010 semakptr = &sema[semid];
1011 sema_mtxp = &sema_mtx[semid];
1012 mtx_lock(sema_mtxp);
1013 if ((semakptr->u.sem_perm.mode & SEM_ALLOC) == 0) {
1014 error = EINVAL;
1015 goto done2;
1016 }
1017 if (semakptr->u.sem_perm.seq != IPCID_TO_SEQ(uap->semid)) {
1018 error = EINVAL;
1019 goto done2;
1020 }
1021 /*
1022 * Initial pass thru sops to see what permissions are needed.
1023 * Also perform any checks that don't need repeating on each
1024 * attempt to satisfy the request vector.
1025 */
1026 j = 0; /* permission needed */
1027 do_undos = 0;
1028 for (i = 0; i < nsops; i++) {
1029 sopptr = &sops[i];
1030 if (sopptr->sem_num >= semakptr->u.sem_nsems) {
1031 error = EFBIG;
1032 goto done2;
1033 }
1034 if (sopptr->sem_flg & SEM_UNDO && sopptr->sem_op != 0)
1035 do_undos = 1;
1036 j |= (sopptr->sem_op == 0) ? SEM_R : SEM_A;
1037 }
1038
1039 if ((error = ipcperm(td, &semakptr->u.sem_perm, j))) {
1040 DPRINTF(("error = %d from ipaccess\n", error));
1041 goto done2;
1042 }
1043#ifdef MAC
1044 error = mac_check_sysv_semop(td->td_ucred, semakptr, j);
1045 if (error != 0) {
1046 MPRINTF(("mac_check_sysv_semop returned %d\n", error));
1047 goto done2;
1048 }
1049#endif
1050
1051 /*
1052 * Loop trying to satisfy the vector of requests.
1053 * If we reach a point where we must wait, any requests already
1054 * performed are rolled back and we go to sleep until some other
1055 * process wakes us up. At this point, we start all over again.
1056 *
1057 * This ensures that from the perspective of other tasks, a set
1058 * of requests is atomic (never partially satisfied).
1059 */
1060 for (;;) {
1061 do_wakeup = 0;
1062 error = 0; /* error return if necessary */
1063
1064 for (i = 0; i < nsops; i++) {
1065 sopptr = &sops[i];
1066 semptr = &semakptr->u.sem_base[sopptr->sem_num];
1067
1068 DPRINTF((
1069 "semop: semakptr=%x, sem_base=%x, "
1070 "semptr=%x, sem[%d]=%d : op=%d, flag=%s\n",
1071 semakptr, semakptr->u.sem_base, semptr,
1072 sopptr->sem_num, semptr->semval, sopptr->sem_op,
1073 (sopptr->sem_flg & IPC_NOWAIT) ?
1074 "nowait" : "wait"));
1075
1076 if (sopptr->sem_op < 0) {
1077 if (semptr->semval + sopptr->sem_op < 0) {
1078 DPRINTF(("semop: can't do it now\n"));
1079 break;
1080 } else {
1081 semptr->semval += sopptr->sem_op;
1082 if (semptr->semval == 0 &&
1083 semptr->semzcnt > 0)
1084 do_wakeup = 1;
1085 }
1086 } else if (sopptr->sem_op == 0) {
1087 if (semptr->semval != 0) {
1088 DPRINTF(("semop: not zero now\n"));
1089 break;
1090 }
1091 } else if (semptr->semval + sopptr->sem_op >
1092 seminfo.semvmx) {
1093 error = ERANGE;
1094 break;
1095 } else {
1096 if (semptr->semncnt > 0)
1097 do_wakeup = 1;
1098 semptr->semval += sopptr->sem_op;
1099 }
1100 }
1101
1102 /*
1103 * Did we get through the entire vector?
1104 */
1105 if (i >= nsops)
1106 goto done;
1107
1108 /*
1109 * No ... rollback anything that we've already done
1110 */
1111 DPRINTF(("semop: rollback 0 through %d\n", i-1));
1112 for (j = 0; j < i; j++)
1113 semakptr->u.sem_base[sops[j].sem_num].semval -=
1114 sops[j].sem_op;
1115
1116 /* If we detected an error, return it */
1117 if (error != 0)
1118 goto done2;
1119
1120 /*
1121 * If the request that we couldn't satisfy has the
1122 * NOWAIT flag set then return with EAGAIN.
1123 */
1124 if (sopptr->sem_flg & IPC_NOWAIT) {
1125 error = EAGAIN;
1126 goto done2;
1127 }
1128
1129 if (sopptr->sem_op == 0)
1130 semptr->semzcnt++;
1131 else
1132 semptr->semncnt++;
1133
1134 DPRINTF(("semop: good night!\n"));
1135 error = msleep(semakptr, sema_mtxp, (PZERO - 4) | PCATCH,
1136 "semwait", 0);
1137 DPRINTF(("semop: good morning (error=%d)!\n", error));
1138 /* return code is checked below, after sem[nz]cnt-- */
1139
1140 /*
1141 * Make sure that the semaphore still exists
1142 */
1143 if ((semakptr->u.sem_perm.mode & SEM_ALLOC) == 0 ||
1144 semakptr->u.sem_perm.seq != IPCID_TO_SEQ(uap->semid)) {
1145 error = EIDRM;
1146 goto done2;
1147 }
1148
1149 /*
1150 * The semaphore is still alive. Readjust the count of
1151 * waiting processes.
1152 */
1153 if (sopptr->sem_op == 0)
1154 semptr->semzcnt--;
1155 else
1156 semptr->semncnt--;
1157
1158 /*
1159 * Is it really morning, or was our sleep interrupted?
1160 * (Delayed check of msleep() return code because we
1161 * need to decrement sem[nz]cnt either way.)
1162 */
1163 if (error != 0) {
1164 error = EINTR;
1165 goto done2;
1166 }
1167 DPRINTF(("semop: good morning!\n"));
1168 }
1169
1170done:
1171 /*
1172 * Process any SEM_UNDO requests.
1173 */
1174 if (do_undos) {
1175 SEMUNDO_LOCK();
1176 suptr = NULL;
1177 for (i = 0; i < nsops; i++) {
1178 /*
1179 * We only need to deal with SEM_UNDO's for non-zero
1180 * op's.
1181 */
1182 int adjval;
1183
1184 if ((sops[i].sem_flg & SEM_UNDO) == 0)
1185 continue;
1186 adjval = sops[i].sem_op;
1187 if (adjval == 0)
1188 continue;
1189 error = semundo_adjust(td, &suptr, semid,
1190 sops[i].sem_num, -adjval);
1191 if (error == 0)
1192 continue;
1193
1194 /*
1195 * Oh-Oh! We ran out of either sem_undo's or undo's.
1196 * Rollback the adjustments to this point and then
1197 * rollback the semaphore ups and down so we can return
1198 * with an error with all structures restored. We
1199 * rollback the undo's in the exact reverse order that
1200 * we applied them. This guarantees that we won't run
1201 * out of space as we roll things back out.
1202 */
1203 for (j = 0; j < i; j++) {
1204 k = i - j - 1;
1205 if ((sops[k].sem_flg & SEM_UNDO) == 0)
1206 continue;
1207 adjval = sops[k].sem_op;
1208 if (adjval == 0)
1209 continue;
1210 if (semundo_adjust(td, &suptr, semid,
1211 sops[k].sem_num, adjval) != 0)
1212 panic("semop - can't undo undos");
1213 }
1214
1215 for (j = 0; j < nsops; j++)
1216 semakptr->u.sem_base[sops[j].sem_num].semval -=
1217 sops[j].sem_op;
1218
1219 DPRINTF(("error = %d from semundo_adjust\n", error));
1220 SEMUNDO_UNLOCK();
1221 goto done2;
1222 } /* loop through the sops */
1223 SEMUNDO_UNLOCK();
1224 } /* if (do_undos) */
1225
1226 /* We're definitely done - set the sempid's and time */
1227 for (i = 0; i < nsops; i++) {
1228 sopptr = &sops[i];
1229 semptr = &semakptr->u.sem_base[sopptr->sem_num];
1230 semptr->sempid = td->td_proc->p_pid;
1231 }
1232 semakptr->u.sem_otime = time_second;
1233
1234 /*
1235 * Do a wakeup if any semaphore was up'd whilst something was
1236 * sleeping on it.
1237 */
1238 if (do_wakeup) {
1239 DPRINTF(("semop: doing wakeup\n"));
1240 wakeup(semakptr);
1241 DPRINTF(("semop: back from wakeup\n"));
1242 }
1243 DPRINTF(("semop: done\n"));
1244 td->td_retval[0] = 0;
1245done2:
1246 mtx_unlock(sema_mtxp);
1247 if (sops != small_sops)
1248 free(sops, M_SEM);
1249 return (error);
1250}
1251
1252/*
1253 * Go through the undo structures for this process and apply the adjustments to
1254 * semaphores.
1255 */
1256static void
1257semexit_myhook(arg, p)
1258 void *arg;
1259 struct proc *p;
1260{
1261 struct sem_undo *suptr;
1262 struct sem_undo **supptr;
1263
1264 /*
1265 * Go through the chain of undo vectors looking for one
1266 * associated with this process.
1267 */
1268 SEMUNDO_LOCK();
1269 SLIST_FOREACH_PREVPTR(suptr, supptr, &semu_list, un_next) {
1270 if (suptr->un_proc == p)
1271 break;
1272 }
1273 SEMUNDO_UNLOCK();
1274
1275 if (suptr == NULL)
1276 return;
1277
1278 DPRINTF(("proc @%08x has undo structure with %d entries\n", p,
1279 suptr->un_cnt));
1280
1281 /*
1282 * If there are any active undo elements then process them.
1283 */
1284 if (suptr->un_cnt > 0) {
1285 int ix;
1286
1287 for (ix = 0; ix < suptr->un_cnt; ix++) {
1288 int semid = suptr->un_ent[ix].un_id;
1289 int semnum = suptr->un_ent[ix].un_num;
1290 int adjval = suptr->un_ent[ix].un_adjval;
1291 struct semid_kernel *semakptr;
1292 struct mtx *sema_mtxp;
1293
1294 semakptr = &sema[semid];
1295 sema_mtxp = &sema_mtx[semid];
1296 mtx_lock(sema_mtxp);
1297 SEMUNDO_LOCK();
1298 if ((semakptr->u.sem_perm.mode & SEM_ALLOC) == 0)
1299 panic("semexit - semid not allocated");
1300 if (semnum >= semakptr->u.sem_nsems)
1301 panic("semexit - semnum out of range");
1302
1303 DPRINTF((
1304 "semexit: %08x id=%d num=%d(adj=%d) ; sem=%d\n",
1305 suptr->un_proc, suptr->un_ent[ix].un_id,
1306 suptr->un_ent[ix].un_num,
1307 suptr->un_ent[ix].un_adjval,
1308 semakptr->u.sem_base[semnum].semval));
1309
1310 if (adjval < 0) {
1311 if (semakptr->u.sem_base[semnum].semval <
1312 -adjval)
1313 semakptr->u.sem_base[semnum].semval = 0;
1314 else
1315 semakptr->u.sem_base[semnum].semval +=
1316 adjval;
1317 } else
1318 semakptr->u.sem_base[semnum].semval += adjval;
1319
1320 wakeup(semakptr);
1321 DPRINTF(("semexit: back from wakeup\n"));
1322 mtx_unlock(sema_mtxp);
1323 SEMUNDO_UNLOCK();
1324 }
1325 }
1326
1327 /*
1328 * Deallocate the undo vector.
1329 */
1330 DPRINTF(("removing vector\n"));
1331 suptr->un_proc = NULL;
1332 *supptr = SLIST_NEXT(suptr, un_next);
1333}
1334
1335static int
1336sysctl_sema(SYSCTL_HANDLER_ARGS)
1337{
1338
1339 return (SYSCTL_OUT(req, sema,
1340 sizeof(struct semid_kernel) * seminfo.semmni));
1341}