FreeBSD/Linux Kernel Cross Reference
sys/kern/sysv_sem.c
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 #include <sys/cdefs.h>
10 __FBSDID("$FreeBSD: src/sys/kern/sysv_sem.c,v 1.70.2.4 2005/05/05 18:53:13 rwatson Exp $");
11
12 #include "opt_sysvipc.h"
13
14 #include <sys/param.h>
15 #include <sys/systm.h>
16 #include <sys/sysproto.h>
17 #include <sys/eventhandler.h>
18 #include <sys/kernel.h>
19 #include <sys/proc.h>
20 #include <sys/lock.h>
21 #include <sys/module.h>
22 #include <sys/mutex.h>
23 #include <sys/sem.h>
24 #include <sys/syscall.h>
25 #include <sys/sysent.h>
26 #include <sys/sysctl.h>
27 #include <sys/malloc.h>
28 #include <sys/jail.h>
29
30 static MALLOC_DEFINE(M_SEM, "sem", "SVID compatible semaphores");
31
32 #ifdef SEM_DEBUG
33 #define DPRINTF(a) printf a
34 #else
35 #define DPRINTF(a)
36 #endif
37
38 static void seminit(void);
39 static int sysvsem_modload(struct module *, int, void *);
40 static int semunload(void);
41 static void semexit_myhook(void *arg, struct proc *p);
42 static int sysctl_sema(SYSCTL_HANDLER_ARGS);
43 static int semvalid(int semid, struct semid_kernel *semakptr);
44
45 #ifndef _SYS_SYSPROTO_H_
46 struct __semctl_args;
47 int __semctl(struct thread *td, struct __semctl_args *uap);
48 struct semget_args;
49 int semget(struct thread *td, struct semget_args *uap);
50 struct semop_args;
51 int semop(struct thread *td, struct semop_args *uap);
52 #endif
53
54 static struct sem_undo *semu_alloc(struct thread *td);
55 static int semundo_adjust(struct thread *td, struct sem_undo **supptr,
56 int semid, int semnum, int adjval);
57 static void semundo_clear(int semid, int semnum);
58
59 /* XXX casting to (sy_call_t *) is bogus, as usual. */
60 static sy_call_t *semcalls[] = {
61 (sy_call_t *)__semctl, (sy_call_t *)semget,
62 (sy_call_t *)semop
63 };
64
65 static struct mtx sem_mtx; /* semaphore global lock */
66 static int semtot = 0;
67 static struct semid_kernel *sema; /* semaphore id pool */
68 static struct mtx *sema_mtx; /* semaphore id pool mutexes*/
69 static struct sem *sem; /* semaphore pool */
70 SLIST_HEAD(, sem_undo) semu_list; /* list of active undo structures */
71 static int *semu; /* undo structure pool */
72 static eventhandler_tag semexit_tag;
73
74 #define SEMUNDO_MTX sem_mtx
75 #define SEMUNDO_LOCK() mtx_lock(&SEMUNDO_MTX);
76 #define SEMUNDO_UNLOCK() mtx_unlock(&SEMUNDO_MTX);
77 #define SEMUNDO_LOCKASSERT(how) mtx_assert(&SEMUNDO_MTX, (how));
78
79 struct sem {
80 u_short semval; /* semaphore value */
81 pid_t sempid; /* pid of last operation */
82 u_short semncnt; /* # awaiting semval > cval */
83 u_short semzcnt; /* # awaiting semval = 0 */
84 };
85
86 /*
87 * Undo structure (one per process)
88 */
89 struct sem_undo {
90 SLIST_ENTRY(sem_undo) un_next; /* ptr to next active undo structure */
91 struct proc *un_proc; /* owner of this structure */
92 short un_cnt; /* # of active entries */
93 struct undo {
94 short un_adjval; /* adjust on exit values */
95 short un_num; /* semaphore # */
96 int un_id; /* semid */
97 } un_ent[1]; /* undo entries */
98 };
99
100 /*
101 * Configuration parameters
102 */
103 #ifndef SEMMNI
104 #define SEMMNI 10 /* # of semaphore identifiers */
105 #endif
106 #ifndef SEMMNS
107 #define SEMMNS 60 /* # of semaphores in system */
108 #endif
109 #ifndef SEMUME
110 #define SEMUME 10 /* max # of undo entries per process */
111 #endif
112 #ifndef SEMMNU
113 #define SEMMNU 30 /* # of undo structures in system */
114 #endif
115
116 /* shouldn't need tuning */
117 #ifndef SEMMAP
118 #define SEMMAP 30 /* # of entries in semaphore map */
119 #endif
120 #ifndef SEMMSL
121 #define SEMMSL SEMMNS /* max # of semaphores per id */
122 #endif
123 #ifndef SEMOPM
124 #define SEMOPM 100 /* max # of operations per semop call */
125 #endif
126
127 #define SEMVMX 32767 /* semaphore maximum value */
128 #define SEMAEM 16384 /* adjust on exit max value */
129
130 /*
131 * Due to the way semaphore memory is allocated, we have to ensure that
132 * SEMUSZ is properly aligned.
133 */
134
135 #define SEM_ALIGN(bytes) (((bytes) + (sizeof(long) - 1)) & ~(sizeof(long) - 1))
136
137 /* actual size of an undo structure */
138 #define SEMUSZ SEM_ALIGN(offsetof(struct sem_undo, un_ent[SEMUME]))
139
140 /*
141 * Macro to find a particular sem_undo vector
142 */
143 #define SEMU(ix) \
144 ((struct sem_undo *)(((intptr_t)semu)+ix * seminfo.semusz))
145
146 /*
147 * semaphore info struct
148 */
149 struct seminfo seminfo = {
150 SEMMAP, /* # of entries in semaphore map */
151 SEMMNI, /* # of semaphore identifiers */
152 SEMMNS, /* # of semaphores in system */
153 SEMMNU, /* # of undo structures in system */
154 SEMMSL, /* max # of semaphores per id */
155 SEMOPM, /* max # of operations per semop call */
156 SEMUME, /* max # of undo entries per process */
157 SEMUSZ, /* size in bytes of undo structure */
158 SEMVMX, /* semaphore maximum value */
159 SEMAEM /* adjust on exit max value */
160 };
161
162 SYSCTL_DECL(_kern_ipc);
163 SYSCTL_INT(_kern_ipc, OID_AUTO, semmap, CTLFLAG_RW, &seminfo.semmap, 0,
164 "Number of entries in the semaphore map");
165 SYSCTL_INT(_kern_ipc, OID_AUTO, semmni, CTLFLAG_RDTUN, &seminfo.semmni, 0,
166 "Number of semaphore identifiers");
167 SYSCTL_INT(_kern_ipc, OID_AUTO, semmns, CTLFLAG_RDTUN, &seminfo.semmns, 0,
168 "Maximum number of semaphores in the system");
169 SYSCTL_INT(_kern_ipc, OID_AUTO, semmnu, CTLFLAG_RDTUN, &seminfo.semmnu, 0,
170 "Maximum number of undo structures in the system");
171 SYSCTL_INT(_kern_ipc, OID_AUTO, semmsl, CTLFLAG_RW, &seminfo.semmsl, 0,
172 "Max semaphores per id");
173 SYSCTL_INT(_kern_ipc, OID_AUTO, semopm, CTLFLAG_RDTUN, &seminfo.semopm, 0,
174 "Max operations per semop call");
175 SYSCTL_INT(_kern_ipc, OID_AUTO, semume, CTLFLAG_RDTUN, &seminfo.semume, 0,
176 "Max undo entries per process");
177 SYSCTL_INT(_kern_ipc, OID_AUTO, semusz, CTLFLAG_RDTUN, &seminfo.semusz, 0,
178 "Size in bytes of undo structure");
179 SYSCTL_INT(_kern_ipc, OID_AUTO, semvmx, CTLFLAG_RW, &seminfo.semvmx, 0,
180 "Semaphore maximum value");
181 SYSCTL_INT(_kern_ipc, OID_AUTO, semaem, CTLFLAG_RW, &seminfo.semaem, 0,
182 "Adjust on exit max value");
183 SYSCTL_PROC(_kern_ipc, OID_AUTO, sema, CTLFLAG_RD,
184 NULL, 0, sysctl_sema, "", "");
185
186 static void
187 seminit(void)
188 {
189 int i;
190
191 TUNABLE_INT_FETCH("kern.ipc.semmap", &seminfo.semmap);
192 TUNABLE_INT_FETCH("kern.ipc.semmni", &seminfo.semmni);
193 TUNABLE_INT_FETCH("kern.ipc.semmns", &seminfo.semmns);
194 TUNABLE_INT_FETCH("kern.ipc.semmnu", &seminfo.semmnu);
195 TUNABLE_INT_FETCH("kern.ipc.semmsl", &seminfo.semmsl);
196 TUNABLE_INT_FETCH("kern.ipc.semopm", &seminfo.semopm);
197 TUNABLE_INT_FETCH("kern.ipc.semume", &seminfo.semume);
198 TUNABLE_INT_FETCH("kern.ipc.semusz", &seminfo.semusz);
199 TUNABLE_INT_FETCH("kern.ipc.semvmx", &seminfo.semvmx);
200 TUNABLE_INT_FETCH("kern.ipc.semaem", &seminfo.semaem);
201
202 sem = malloc(sizeof(struct sem) * seminfo.semmns, M_SEM, M_WAITOK);
203 sema = malloc(sizeof(struct semid_kernel) * seminfo.semmni, M_SEM,
204 M_WAITOK);
205 sema_mtx = malloc(sizeof(struct mtx) * seminfo.semmni, M_SEM,
206 M_WAITOK | M_ZERO);
207 semu = malloc(seminfo.semmnu * seminfo.semusz, M_SEM, M_WAITOK);
208
209 for (i = 0; i < seminfo.semmni; i++) {
210 sema[i].u.sem_base = 0;
211 sema[i].u.sem_perm.mode = 0;
212 sema[i].u.sem_perm.seq = 0;
213 }
214 for (i = 0; i < seminfo.semmni; i++)
215 mtx_init(&sema_mtx[i], "semid", NULL, MTX_DEF);
216 for (i = 0; i < seminfo.semmnu; i++) {
217 struct sem_undo *suptr = SEMU(i);
218 suptr->un_proc = NULL;
219 }
220 SLIST_INIT(&semu_list);
221 mtx_init(&sem_mtx, "sem", NULL, MTX_DEF);
222 semexit_tag = EVENTHANDLER_REGISTER(process_exit, semexit_myhook, NULL,
223 EVENTHANDLER_PRI_ANY);
224 }
225
226 static int
227 semunload(void)
228 {
229 int i;
230
231 if (semtot != 0)
232 return (EBUSY);
233
234 EVENTHANDLER_DEREGISTER(process_exit, semexit_tag);
235 free(sem, M_SEM);
236 free(sema, M_SEM);
237 free(semu, M_SEM);
238 for (i = 0; i < seminfo.semmni; i++)
239 mtx_destroy(&sema_mtx[i]);
240 mtx_destroy(&sem_mtx);
241 return (0);
242 }
243
244 static int
245 sysvsem_modload(struct module *module, int cmd, void *arg)
246 {
247 int error = 0;
248
249 switch (cmd) {
250 case MOD_LOAD:
251 seminit();
252 break;
253 case MOD_UNLOAD:
254 error = semunload();
255 break;
256 case MOD_SHUTDOWN:
257 break;
258 default:
259 error = EINVAL;
260 break;
261 }
262 return (error);
263 }
264
265 static moduledata_t sysvsem_mod = {
266 "sysvsem",
267 &sysvsem_modload,
268 NULL
269 };
270
271 SYSCALL_MODULE_HELPER(semsys);
272 SYSCALL_MODULE_HELPER(__semctl);
273 SYSCALL_MODULE_HELPER(semget);
274 SYSCALL_MODULE_HELPER(semop);
275
276 DECLARE_MODULE(sysvsem, sysvsem_mod,
277 SI_SUB_SYSV_SEM, SI_ORDER_FIRST);
278 MODULE_VERSION(sysvsem, 1);
279
280 /*
281 * Entry point for all SEM calls
282 *
283 * MPSAFE
284 */
285 int
286 semsys(td, uap)
287 struct thread *td;
288 /* XXX actually varargs. */
289 struct semsys_args /* {
290 int which;
291 int a2;
292 int a3;
293 int a4;
294 int a5;
295 } */ *uap;
296 {
297 int error;
298
299 if (!jail_sysvipc_allowed && jailed(td->td_ucred))
300 return (ENOSYS);
301 if (uap->which < 0 ||
302 uap->which >= sizeof(semcalls)/sizeof(semcalls[0]))
303 return (EINVAL);
304 error = (*semcalls[uap->which])(td, &uap->a2);
305 return (error);
306 }
307
308 /*
309 * Allocate a new sem_undo structure for a process
310 * (returns ptr to structure or NULL if no more room)
311 */
312
313 static struct sem_undo *
314 semu_alloc(td)
315 struct thread *td;
316 {
317 int i;
318 struct sem_undo *suptr;
319 struct sem_undo **supptr;
320 int attempt;
321
322 SEMUNDO_LOCKASSERT(MA_OWNED);
323 /*
324 * Try twice to allocate something.
325 * (we'll purge an empty structure after the first pass so
326 * two passes are always enough)
327 */
328
329 for (attempt = 0; attempt < 2; attempt++) {
330 /*
331 * Look for a free structure.
332 * Fill it in and return it if we find one.
333 */
334
335 for (i = 0; i < seminfo.semmnu; i++) {
336 suptr = SEMU(i);
337 if (suptr->un_proc == NULL) {
338 SLIST_INSERT_HEAD(&semu_list, suptr, un_next);
339 suptr->un_cnt = 0;
340 suptr->un_proc = td->td_proc;
341 return(suptr);
342 }
343 }
344
345 /*
346 * We didn't find a free one, if this is the first attempt
347 * then try to free a structure.
348 */
349
350 if (attempt == 0) {
351 /* All the structures are in use - try to free one */
352 int did_something = 0;
353
354 SLIST_FOREACH_PREVPTR(suptr, supptr, &semu_list,
355 un_next) {
356 if (suptr->un_cnt == 0) {
357 suptr->un_proc = NULL;
358 did_something = 1;
359 *supptr = SLIST_NEXT(suptr, un_next);
360 break;
361 }
362 }
363
364 /* If we didn't free anything then just give-up */
365 if (!did_something)
366 return(NULL);
367 } else {
368 /*
369 * The second pass failed even though we freed
370 * something after the first pass!
371 * This is IMPOSSIBLE!
372 */
373 panic("semu_alloc - second attempt failed");
374 }
375 }
376 return (NULL);
377 }
378
379 /*
380 * Adjust a particular entry for a particular proc
381 */
382
383 static int
384 semundo_adjust(td, supptr, semid, semnum, adjval)
385 struct thread *td;
386 struct sem_undo **supptr;
387 int semid, semnum;
388 int adjval;
389 {
390 struct proc *p = td->td_proc;
391 struct sem_undo *suptr;
392 struct undo *sunptr;
393 int i;
394
395 SEMUNDO_LOCKASSERT(MA_OWNED);
396 /* Look for and remember the sem_undo if the caller doesn't provide
397 it */
398
399 suptr = *supptr;
400 if (suptr == NULL) {
401 SLIST_FOREACH(suptr, &semu_list, un_next) {
402 if (suptr->un_proc == p) {
403 *supptr = suptr;
404 break;
405 }
406 }
407 if (suptr == NULL) {
408 if (adjval == 0)
409 return(0);
410 suptr = semu_alloc(td);
411 if (suptr == NULL)
412 return(ENOSPC);
413 *supptr = suptr;
414 }
415 }
416
417 /*
418 * Look for the requested entry and adjust it (delete if adjval becomes
419 * 0).
420 */
421 sunptr = &suptr->un_ent[0];
422 for (i = 0; i < suptr->un_cnt; i++, sunptr++) {
423 if (sunptr->un_id != semid || sunptr->un_num != semnum)
424 continue;
425 if (adjval != 0) {
426 adjval += sunptr->un_adjval;
427 if (adjval > seminfo.semaem || adjval < -seminfo.semaem)
428 return (ERANGE);
429 }
430 sunptr->un_adjval = adjval;
431 if (sunptr->un_adjval == 0) {
432 suptr->un_cnt--;
433 if (i < suptr->un_cnt)
434 suptr->un_ent[i] =
435 suptr->un_ent[suptr->un_cnt];
436 }
437 return(0);
438 }
439
440 /* Didn't find the right entry - create it */
441 if (adjval == 0)
442 return(0);
443 if (adjval > seminfo.semaem || adjval < -seminfo.semaem)
444 return (ERANGE);
445 if (suptr->un_cnt != seminfo.semume) {
446 sunptr = &suptr->un_ent[suptr->un_cnt];
447 suptr->un_cnt++;
448 sunptr->un_adjval = adjval;
449 sunptr->un_id = semid; sunptr->un_num = semnum;
450 } else
451 return(EINVAL);
452 return(0);
453 }
454
455 static void
456 semundo_clear(semid, semnum)
457 int semid, semnum;
458 {
459 struct sem_undo *suptr;
460
461 SEMUNDO_LOCKASSERT(MA_OWNED);
462 SLIST_FOREACH(suptr, &semu_list, un_next) {
463 struct undo *sunptr = &suptr->un_ent[0];
464 int i = 0;
465
466 while (i < suptr->un_cnt) {
467 if (sunptr->un_id == semid) {
468 if (semnum == -1 || sunptr->un_num == semnum) {
469 suptr->un_cnt--;
470 if (i < suptr->un_cnt) {
471 suptr->un_ent[i] =
472 suptr->un_ent[suptr->un_cnt];
473 continue;
474 }
475 }
476 if (semnum != -1)
477 break;
478 }
479 i++, sunptr++;
480 }
481 }
482 }
483
484 static int
485 semvalid(semid, semakptr)
486 int semid;
487 struct semid_kernel *semakptr;
488 {
489
490 return ((semakptr->u.sem_perm.mode & SEM_ALLOC) == 0 ||
491 semakptr->u.sem_perm.seq != IPCID_TO_SEQ(semid) ? EINVAL : 0);
492 }
493
494 /*
495 * Note that the user-mode half of this passes a union, not a pointer
496 */
497 #ifndef _SYS_SYSPROTO_H_
498 struct __semctl_args {
499 int semid;
500 int semnum;
501 int cmd;
502 union semun *arg;
503 };
504 #endif
505
506 /*
507 * MPSAFE
508 */
509 int
510 __semctl(td, uap)
511 struct thread *td;
512 struct __semctl_args *uap;
513 {
514 int semid = uap->semid;
515 int semnum = uap->semnum;
516 int cmd = uap->cmd;
517 u_short *array;
518 union semun *arg = uap->arg;
519 union semun real_arg;
520 struct ucred *cred = td->td_ucred;
521 int i, rval, error;
522 struct semid_ds sbuf;
523 struct semid_kernel *semakptr;
524 struct mtx *sema_mtxp;
525 u_short usval, count;
526
527 DPRINTF(("call to semctl(%d, %d, %d, 0x%x)\n",
528 semid, semnum, cmd, arg));
529 if (!jail_sysvipc_allowed && jailed(td->td_ucred))
530 return (ENOSYS);
531
532 array = NULL;
533
534 switch(cmd) {
535 case SEM_STAT:
536 if (semid < 0 || semid >= seminfo.semmni)
537 return (EINVAL);
538 if ((error = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
539 return (error);
540 semakptr = &sema[semid];
541 sema_mtxp = &sema_mtx[semid];
542 mtx_lock(sema_mtxp);
543 if ((semakptr->u.sem_perm.mode & SEM_ALLOC) == 0) {
544 error = EINVAL;
545 goto done2;
546 }
547 if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
548 goto done2;
549 mtx_unlock(sema_mtxp);
550 error = copyout(&semakptr->u, real_arg.buf,
551 sizeof(struct semid_ds));
552 rval = IXSEQ_TO_IPCID(semid, semakptr->u.sem_perm);
553 if (error == 0)
554 td->td_retval[0] = rval;
555 return (error);
556 }
557
558 semid = IPCID_TO_IX(semid);
559 if (semid < 0 || semid >= seminfo.semmni)
560 return (EINVAL);
561
562 semakptr = &sema[semid];
563 sema_mtxp = &sema_mtx[semid];
564
565 error = 0;
566 rval = 0;
567
568 switch (cmd) {
569 case IPC_RMID:
570 mtx_lock(sema_mtxp);
571 if ((error = semvalid(uap->semid, semakptr)) != 0)
572 goto done2;
573 if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_M)))
574 goto done2;
575 semakptr->u.sem_perm.cuid = cred->cr_uid;
576 semakptr->u.sem_perm.uid = cred->cr_uid;
577 semtot -= semakptr->u.sem_nsems;
578 for (i = semakptr->u.sem_base - sem; i < semtot; i++)
579 sem[i] = sem[i + semakptr->u.sem_nsems];
580 for (i = 0; i < seminfo.semmni; i++) {
581 if ((sema[i].u.sem_perm.mode & SEM_ALLOC) &&
582 sema[i].u.sem_base > semakptr->u.sem_base)
583 sema[i].u.sem_base -= semakptr->u.sem_nsems;
584 }
585 semakptr->u.sem_perm.mode = 0;
586 SEMUNDO_LOCK();
587 semundo_clear(semid, -1);
588 SEMUNDO_UNLOCK();
589 wakeup(semakptr);
590 break;
591
592 case IPC_SET:
593 if ((error = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
594 goto done2;
595 if ((error = copyin(real_arg.buf, &sbuf, sizeof(sbuf))) != 0)
596 goto done2;
597 mtx_lock(sema_mtxp);
598 if ((error = semvalid(uap->semid, semakptr)) != 0)
599 goto done2;
600 if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_M)))
601 goto done2;
602 semakptr->u.sem_perm.uid = sbuf.sem_perm.uid;
603 semakptr->u.sem_perm.gid = sbuf.sem_perm.gid;
604 semakptr->u.sem_perm.mode = (semakptr->u.sem_perm.mode &
605 ~0777) | (sbuf.sem_perm.mode & 0777);
606 semakptr->u.sem_ctime = time_second;
607 break;
608
609 case IPC_STAT:
610 if ((error = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
611 goto done2;
612 mtx_lock(sema_mtxp);
613 if ((error = semvalid(uap->semid, semakptr)) != 0)
614 goto done2;
615 if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
616 goto done2;
617 sbuf = semakptr->u;
618 mtx_unlock(sema_mtxp);
619 error = copyout(&semakptr->u, real_arg.buf,
620 sizeof(struct semid_ds));
621 break;
622
623 case GETNCNT:
624 mtx_lock(sema_mtxp);
625 if ((error = semvalid(uap->semid, semakptr)) != 0)
626 goto done2;
627 if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
628 goto done2;
629 if (semnum < 0 || semnum >= semakptr->u.sem_nsems) {
630 error = EINVAL;
631 goto done2;
632 }
633 rval = semakptr->u.sem_base[semnum].semncnt;
634 break;
635
636 case GETPID:
637 mtx_lock(sema_mtxp);
638 if ((error = semvalid(uap->semid, semakptr)) != 0)
639 goto done2;
640 if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
641 goto done2;
642 if (semnum < 0 || semnum >= semakptr->u.sem_nsems) {
643 error = EINVAL;
644 goto done2;
645 }
646 rval = semakptr->u.sem_base[semnum].sempid;
647 break;
648
649 case GETVAL:
650 mtx_lock(sema_mtxp);
651 if ((error = semvalid(uap->semid, semakptr)) != 0)
652 goto done2;
653 if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
654 goto done2;
655 if (semnum < 0 || semnum >= semakptr->u.sem_nsems) {
656 error = EINVAL;
657 goto done2;
658 }
659 rval = semakptr->u.sem_base[semnum].semval;
660 break;
661
662 case GETALL:
663 if ((error = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
664 goto done2;
665 array = malloc(sizeof(*array) * semakptr->u.sem_nsems, M_TEMP,
666 M_WAITOK);
667 mtx_lock(sema_mtxp);
668 if ((error = semvalid(uap->semid, semakptr)) != 0)
669 goto done2;
670 if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
671 goto done2;
672 for (i = 0; i < semakptr->u.sem_nsems; i++)
673 array[i] = semakptr->u.sem_base[i].semval;
674 mtx_unlock(sema_mtxp);
675 error = copyout(array, real_arg.array,
676 i * sizeof(real_arg.array[0]));
677 break;
678
679 case GETZCNT:
680 mtx_lock(sema_mtxp);
681 if ((error = semvalid(uap->semid, semakptr)) != 0)
682 goto done2;
683 if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_R)))
684 goto done2;
685 if (semnum < 0 || semnum >= semakptr->u.sem_nsems) {
686 error = EINVAL;
687 goto done2;
688 }
689 rval = semakptr->u.sem_base[semnum].semzcnt;
690 break;
691
692 case SETVAL:
693 if ((error = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
694 goto done2;
695 mtx_lock(sema_mtxp);
696 if ((error = semvalid(uap->semid, semakptr)) != 0)
697 goto done2;
698 if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_W)))
699 goto done2;
700 if (semnum < 0 || semnum >= semakptr->u.sem_nsems) {
701 error = EINVAL;
702 goto done2;
703 }
704 if (real_arg.val < 0 || real_arg.val > seminfo.semvmx) {
705 error = ERANGE;
706 goto done2;
707 }
708 semakptr->u.sem_base[semnum].semval = real_arg.val;
709 SEMUNDO_LOCK();
710 semundo_clear(semid, semnum);
711 SEMUNDO_UNLOCK();
712 wakeup(semakptr);
713 break;
714
715 case SETALL:
716 mtx_lock(sema_mtxp);
717 raced:
718 if ((error = semvalid(uap->semid, semakptr)) != 0)
719 goto done2;
720 count = semakptr->u.sem_nsems;
721 mtx_unlock(sema_mtxp);
722 if ((error = copyin(arg, &real_arg, sizeof(real_arg))) != 0)
723 goto done2;
724 array = malloc(sizeof(*array) * count, M_TEMP, M_WAITOK);
725 error = copyin(real_arg.array, array, count * sizeof(*array));
726 if (error)
727 break;
728 mtx_lock(sema_mtxp);
729 if ((error = semvalid(uap->semid, semakptr)) != 0)
730 goto done2;
731 /* we could have raced? */
732 if (count != semakptr->u.sem_nsems) {
733 free(array, M_TEMP);
734 array = NULL;
735 goto raced;
736 }
737 if ((error = ipcperm(td, &semakptr->u.sem_perm, IPC_W)))
738 goto done2;
739 for (i = 0; i < semakptr->u.sem_nsems; i++) {
740 usval = array[i];
741 if (usval > seminfo.semvmx) {
742 error = ERANGE;
743 break;
744 }
745 semakptr->u.sem_base[i].semval = usval;
746 }
747 SEMUNDO_LOCK();
748 semundo_clear(semid, -1);
749 SEMUNDO_UNLOCK();
750 wakeup(semakptr);
751 break;
752
753 default:
754 error = EINVAL;
755 break;
756 }
757
758 if (error == 0)
759 td->td_retval[0] = rval;
760 done2:
761 if (mtx_owned(sema_mtxp))
762 mtx_unlock(sema_mtxp);
763 if (array != NULL)
764 free(array, M_TEMP);
765 return(error);
766 }
767
768 #ifndef _SYS_SYSPROTO_H_
769 struct semget_args {
770 key_t key;
771 int nsems;
772 int semflg;
773 };
774 #endif
775
776 /*
777 * MPSAFE
778 */
779 int
780 semget(td, uap)
781 struct thread *td;
782 struct semget_args *uap;
783 {
784 int semid, error = 0;
785 int key = uap->key;
786 int nsems = uap->nsems;
787 int semflg = uap->semflg;
788 struct ucred *cred = td->td_ucred;
789
790 DPRINTF(("semget(0x%x, %d, 0%o)\n", key, nsems, semflg));
791 if (!jail_sysvipc_allowed && jailed(td->td_ucred))
792 return (ENOSYS);
793
794 mtx_lock(&Giant);
795 if (key != IPC_PRIVATE) {
796 for (semid = 0; semid < seminfo.semmni; semid++) {
797 if ((sema[semid].u.sem_perm.mode & SEM_ALLOC) &&
798 sema[semid].u.sem_perm.key == key)
799 break;
800 }
801 if (semid < seminfo.semmni) {
802 DPRINTF(("found public key\n"));
803 if ((error = ipcperm(td, &sema[semid].u.sem_perm,
804 semflg & 0700))) {
805 goto done2;
806 }
807 if (nsems > 0 && sema[semid].u.sem_nsems < nsems) {
808 DPRINTF(("too small\n"));
809 error = EINVAL;
810 goto done2;
811 }
812 if ((semflg & IPC_CREAT) && (semflg & IPC_EXCL)) {
813 DPRINTF(("not exclusive\n"));
814 error = EEXIST;
815 goto done2;
816 }
817 goto found;
818 }
819 }
820
821 DPRINTF(("need to allocate the semid_kernel\n"));
822 if (key == IPC_PRIVATE || (semflg & IPC_CREAT)) {
823 if (nsems <= 0 || nsems > seminfo.semmsl) {
824 DPRINTF(("nsems out of range (0<%d<=%d)\n", nsems,
825 seminfo.semmsl));
826 error = EINVAL;
827 goto done2;
828 }
829 if (nsems > seminfo.semmns - semtot) {
830 DPRINTF((
831 "not enough semaphores left (need %d, got %d)\n",
832 nsems, seminfo.semmns - semtot));
833 error = ENOSPC;
834 goto done2;
835 }
836 for (semid = 0; semid < seminfo.semmni; semid++) {
837 if ((sema[semid].u.sem_perm.mode & SEM_ALLOC) == 0)
838 break;
839 }
840 if (semid == seminfo.semmni) {
841 DPRINTF(("no more semid_kernel's available\n"));
842 error = ENOSPC;
843 goto done2;
844 }
845 DPRINTF(("semid %d is available\n", semid));
846 sema[semid].u.sem_perm.key = key;
847 sema[semid].u.sem_perm.cuid = cred->cr_uid;
848 sema[semid].u.sem_perm.uid = cred->cr_uid;
849 sema[semid].u.sem_perm.cgid = cred->cr_gid;
850 sema[semid].u.sem_perm.gid = cred->cr_gid;
851 sema[semid].u.sem_perm.mode = (semflg & 0777) | SEM_ALLOC;
852 sema[semid].u.sem_perm.seq =
853 (sema[semid].u.sem_perm.seq + 1) & 0x7fff;
854 sema[semid].u.sem_nsems = nsems;
855 sema[semid].u.sem_otime = 0;
856 sema[semid].u.sem_ctime = time_second;
857 sema[semid].u.sem_base = &sem[semtot];
858 semtot += nsems;
859 bzero(sema[semid].u.sem_base,
860 sizeof(sema[semid].u.sem_base[0])*nsems);
861 DPRINTF(("sembase = 0x%x, next = 0x%x\n",
862 sema[semid].u.sem_base, &sem[semtot]));
863 } else {
864 DPRINTF(("didn't find it and wasn't asked to create it\n"));
865 error = ENOENT;
866 goto done2;
867 }
868
869 found:
870 td->td_retval[0] = IXSEQ_TO_IPCID(semid, sema[semid].u.sem_perm);
871 done2:
872 mtx_unlock(&Giant);
873 return (error);
874 }
875
876 #ifndef _SYS_SYSPROTO_H_
877 struct semop_args {
878 int semid;
879 struct sembuf *sops;
880 size_t nsops;
881 };
882 #endif
883
884 /*
885 * MPSAFE
886 */
887 int
888 semop(td, uap)
889 struct thread *td;
890 struct semop_args *uap;
891 {
892 #define SMALL_SOPS 8
893 struct sembuf small_sops[SMALL_SOPS];
894 int semid = uap->semid;
895 size_t nsops = uap->nsops;
896 struct sembuf *sops;
897 struct semid_kernel *semakptr;
898 struct sembuf *sopptr = 0;
899 struct sem *semptr = 0;
900 struct sem_undo *suptr;
901 struct mtx *sema_mtxp;
902 size_t i, j, k;
903 int error;
904 int do_wakeup, do_undos;
905
906 DPRINTF(("call to semop(%d, 0x%x, %u)\n", semid, sops, nsops));
907
908 if (!jail_sysvipc_allowed && jailed(td->td_ucred))
909 return (ENOSYS);
910
911 semid = IPCID_TO_IX(semid); /* Convert back to zero origin */
912
913 if (semid < 0 || semid >= seminfo.semmni)
914 return (EINVAL);
915
916 /* Allocate memory for sem_ops */
917 if (nsops <= SMALL_SOPS)
918 sops = small_sops;
919 else if (nsops <= seminfo.semopm)
920 sops = malloc(nsops * sizeof(*sops), M_TEMP, M_WAITOK);
921 else {
922 DPRINTF(("too many sops (max=%d, nsops=%d)\n", seminfo.semopm,
923 nsops));
924 return (E2BIG);
925 }
926 if ((error = copyin(uap->sops, sops, nsops * sizeof(sops[0]))) != 0) {
927 DPRINTF(("error = %d from copyin(%08x, %08x, %d)\n", error,
928 uap->sops, sops, nsops * sizeof(sops[0])));
929 if (sops != small_sops)
930 free(sops, M_SEM);
931 return (error);
932 }
933
934 semakptr = &sema[semid];
935 sema_mtxp = &sema_mtx[semid];
936 mtx_lock(sema_mtxp);
937 if ((semakptr->u.sem_perm.mode & SEM_ALLOC) == 0) {
938 error = EINVAL;
939 goto done2;
940 }
941 if (semakptr->u.sem_perm.seq != IPCID_TO_SEQ(uap->semid)) {
942 error = EINVAL;
943 goto done2;
944 }
945 /*
946 * Initial pass thru sops to see what permissions are needed.
947 * Also perform any checks that don't need repeating on each
948 * attempt to satisfy the request vector.
949 */
950 j = 0; /* permission needed */
951 do_undos = 0;
952 for (i = 0; i < nsops; i++) {
953 sopptr = &sops[i];
954 if (sopptr->sem_num >= semakptr->u.sem_nsems) {
955 error = EFBIG;
956 goto done2;
957 }
958 if (sopptr->sem_flg & SEM_UNDO && sopptr->sem_op != 0)
959 do_undos = 1;
960 j |= (sopptr->sem_op == 0) ? SEM_R : SEM_A;
961 }
962
963 if ((error = ipcperm(td, &semakptr->u.sem_perm, j))) {
964 DPRINTF(("error = %d from ipaccess\n", error));
965 goto done2;
966 }
967
968 /*
969 * Loop trying to satisfy the vector of requests.
970 * If we reach a point where we must wait, any requests already
971 * performed are rolled back and we go to sleep until some other
972 * process wakes us up. At this point, we start all over again.
973 *
974 * This ensures that from the perspective of other tasks, a set
975 * of requests is atomic (never partially satisfied).
976 */
977 for (;;) {
978 do_wakeup = 0;
979 error = 0; /* error return if necessary */
980
981 for (i = 0; i < nsops; i++) {
982 sopptr = &sops[i];
983 semptr = &semakptr->u.sem_base[sopptr->sem_num];
984
985 DPRINTF((
986 "semop: semakptr=%x, sem_base=%x, "
987 "semptr=%x, sem[%d]=%d : op=%d, flag=%s\n",
988 semakptr, semakptr->u.sem_base, semptr,
989 sopptr->sem_num, semptr->semval, sopptr->sem_op,
990 (sopptr->sem_flg & IPC_NOWAIT) ?
991 "nowait" : "wait"));
992
993 if (sopptr->sem_op < 0) {
994 if (semptr->semval + sopptr->sem_op < 0) {
995 DPRINTF(("semop: can't do it now\n"));
996 break;
997 } else {
998 semptr->semval += sopptr->sem_op;
999 if (semptr->semval == 0 &&
1000 semptr->semzcnt > 0)
1001 do_wakeup = 1;
1002 }
1003 } else if (sopptr->sem_op == 0) {
1004 if (semptr->semval != 0) {
1005 DPRINTF(("semop: not zero now\n"));
1006 break;
1007 }
1008 } else if (semptr->semval + sopptr->sem_op >
1009 seminfo.semvmx) {
1010 error = ERANGE;
1011 break;
1012 } else {
1013 if (semptr->semncnt > 0)
1014 do_wakeup = 1;
1015 semptr->semval += sopptr->sem_op;
1016 }
1017 }
1018
1019 /*
1020 * Did we get through the entire vector?
1021 */
1022 if (i >= nsops)
1023 goto done;
1024
1025 /*
1026 * No ... rollback anything that we've already done
1027 */
1028 DPRINTF(("semop: rollback 0 through %d\n", i-1));
1029 for (j = 0; j < i; j++)
1030 semakptr->u.sem_base[sops[j].sem_num].semval -=
1031 sops[j].sem_op;
1032
1033 /* If we detected an error, return it */
1034 if (error != 0)
1035 goto done2;
1036
1037 /*
1038 * If the request that we couldn't satisfy has the
1039 * NOWAIT flag set then return with EAGAIN.
1040 */
1041 if (sopptr->sem_flg & IPC_NOWAIT) {
1042 error = EAGAIN;
1043 goto done2;
1044 }
1045
1046 if (sopptr->sem_op == 0)
1047 semptr->semzcnt++;
1048 else
1049 semptr->semncnt++;
1050
1051 DPRINTF(("semop: good night!\n"));
1052 error = msleep(semakptr, sema_mtxp, (PZERO - 4) | PCATCH,
1053 "semwait", 0);
1054 DPRINTF(("semop: good morning (error=%d)!\n", error));
1055 /* return code is checked below, after sem[nz]cnt-- */
1056
1057 /*
1058 * Make sure that the semaphore still exists
1059 */
1060 if ((semakptr->u.sem_perm.mode & SEM_ALLOC) == 0 ||
1061 semakptr->u.sem_perm.seq != IPCID_TO_SEQ(uap->semid)) {
1062 error = EIDRM;
1063 goto done2;
1064 }
1065
1066 /*
1067 * The semaphore is still alive. Readjust the count of
1068 * waiting processes.
1069 */
1070 if (sopptr->sem_op == 0)
1071 semptr->semzcnt--;
1072 else
1073 semptr->semncnt--;
1074
1075 /*
1076 * Is it really morning, or was our sleep interrupted?
1077 * (Delayed check of msleep() return code because we
1078 * need to decrement sem[nz]cnt either way.)
1079 */
1080 if (error != 0) {
1081 error = EINTR;
1082 goto done2;
1083 }
1084 DPRINTF(("semop: good morning!\n"));
1085 }
1086
1087 done:
1088 /*
1089 * Process any SEM_UNDO requests.
1090 */
1091 if (do_undos) {
1092 SEMUNDO_LOCK();
1093 suptr = NULL;
1094 for (i = 0; i < nsops; i++) {
1095 /*
1096 * We only need to deal with SEM_UNDO's for non-zero
1097 * op's.
1098 */
1099 int adjval;
1100
1101 if ((sops[i].sem_flg & SEM_UNDO) == 0)
1102 continue;
1103 adjval = sops[i].sem_op;
1104 if (adjval == 0)
1105 continue;
1106 error = semundo_adjust(td, &suptr, semid,
1107 sops[i].sem_num, -adjval);
1108 if (error == 0)
1109 continue;
1110
1111 /*
1112 * Oh-Oh! We ran out of either sem_undo's or undo's.
1113 * Rollback the adjustments to this point and then
1114 * rollback the semaphore ups and down so we can return
1115 * with an error with all structures restored. We
1116 * rollback the undo's in the exact reverse order that
1117 * we applied them. This guarantees that we won't run
1118 * out of space as we roll things back out.
1119 */
1120 for (j = 0; j < i; j++) {
1121 k = i - j - 1;
1122 if ((sops[k].sem_flg & SEM_UNDO) == 0)
1123 continue;
1124 adjval = sops[k].sem_op;
1125 if (adjval == 0)
1126 continue;
1127 if (semundo_adjust(td, &suptr, semid,
1128 sops[k].sem_num, adjval) != 0)
1129 panic("semop - can't undo undos");
1130 }
1131
1132 for (j = 0; j < nsops; j++)
1133 semakptr->u.sem_base[sops[j].sem_num].semval -=
1134 sops[j].sem_op;
1135
1136 DPRINTF(("error = %d from semundo_adjust\n", error));
1137 SEMUNDO_UNLOCK();
1138 goto done2;
1139 } /* loop through the sops */
1140 SEMUNDO_UNLOCK();
1141 } /* if (do_undos) */
1142
1143 /* We're definitely done - set the sempid's and time */
1144 for (i = 0; i < nsops; i++) {
1145 sopptr = &sops[i];
1146 semptr = &semakptr->u.sem_base[sopptr->sem_num];
1147 semptr->sempid = td->td_proc->p_pid;
1148 }
1149 semakptr->u.sem_otime = time_second;
1150
1151 /*
1152 * Do a wakeup if any semaphore was up'd whilst something was
1153 * sleeping on it.
1154 */
1155 if (do_wakeup) {
1156 DPRINTF(("semop: doing wakeup\n"));
1157 wakeup(semakptr);
1158 DPRINTF(("semop: back from wakeup\n"));
1159 }
1160 DPRINTF(("semop: done\n"));
1161 td->td_retval[0] = 0;
1162 done2:
1163 mtx_unlock(sema_mtxp);
1164 if (sops != small_sops)
1165 free(sops, M_SEM);
1166 return (error);
1167 }
1168
1169 /*
1170 * Go through the undo structures for this process and apply the adjustments to
1171 * semaphores.
1172 */
1173 static void
1174 semexit_myhook(arg, p)
1175 void *arg;
1176 struct proc *p;
1177 {
1178 struct sem_undo *suptr;
1179 struct sem_undo **supptr;
1180
1181 /*
1182 * Go through the chain of undo vectors looking for one
1183 * associated with this process.
1184 */
1185 SEMUNDO_LOCK();
1186 SLIST_FOREACH_PREVPTR(suptr, supptr, &semu_list, un_next) {
1187 if (suptr->un_proc == p)
1188 break;
1189 }
1190 SEMUNDO_UNLOCK();
1191
1192 if (suptr == NULL)
1193 return;
1194
1195 DPRINTF(("proc @%08x has undo structure with %d entries\n", p,
1196 suptr->un_cnt));
1197
1198 /*
1199 * If there are any active undo elements then process them.
1200 */
1201 if (suptr->un_cnt > 0) {
1202 int ix;
1203
1204 for (ix = 0; ix < suptr->un_cnt; ix++) {
1205 int semid = suptr->un_ent[ix].un_id;
1206 int semnum = suptr->un_ent[ix].un_num;
1207 int adjval = suptr->un_ent[ix].un_adjval;
1208 struct semid_kernel *semakptr;
1209 struct mtx *sema_mtxp;
1210
1211 semakptr = &sema[semid];
1212 sema_mtxp = &sema_mtx[semid];
1213 mtx_lock(sema_mtxp);
1214 SEMUNDO_LOCK();
1215 if ((semakptr->u.sem_perm.mode & SEM_ALLOC) == 0)
1216 panic("semexit - semid not allocated");
1217 if (semnum >= semakptr->u.sem_nsems)
1218 panic("semexit - semnum out of range");
1219
1220 DPRINTF((
1221 "semexit: %08x id=%d num=%d(adj=%d) ; sem=%d\n",
1222 suptr->un_proc, suptr->un_ent[ix].un_id,
1223 suptr->un_ent[ix].un_num,
1224 suptr->un_ent[ix].un_adjval,
1225 semakptr->u.sem_base[semnum].semval));
1226
1227 if (adjval < 0) {
1228 if (semakptr->u.sem_base[semnum].semval <
1229 -adjval)
1230 semakptr->u.sem_base[semnum].semval = 0;
1231 else
1232 semakptr->u.sem_base[semnum].semval +=
1233 adjval;
1234 } else
1235 semakptr->u.sem_base[semnum].semval += adjval;
1236
1237 wakeup(semakptr);
1238 DPRINTF(("semexit: back from wakeup\n"));
1239 mtx_unlock(sema_mtxp);
1240 SEMUNDO_UNLOCK();
1241 }
1242 }
1243
1244 /*
1245 * Deallocate the undo vector.
1246 */
1247 DPRINTF(("removing vector\n"));
1248 suptr->un_proc = NULL;
1249 *supptr = SLIST_NEXT(suptr, un_next);
1250 }
1251
1252 static int
1253 sysctl_sema(SYSCTL_HANDLER_ARGS)
1254 {
1255
1256 return (SYSCTL_OUT(req, sema,
1257 sizeof(struct semid_kernel) * seminfo.semmni));
1258 }
Cache object: 922b2b67cacd6ecf2b52f6004aa4f72f
|