proc->thread stage 2: MAJOR revamping of system calls, ucred, jail API,
[dragonfly.git] / sys / kern / sysv_shm.c
1 /* $FreeBSD: src/sys/kern/sysv_shm.c,v 1.45.2.6 2002/10/22 20:45:03 fjoe Exp $ */
2 /* $DragonFly: src/sys/kern/sysv_shm.c,v 1.3 2003/06/23 17:55:41 dillon Exp $ */
3 /*      $NetBSD: sysv_shm.c,v 1.23 1994/07/04 23:25:12 glass Exp $      */
4
5 /*
6  * Copyright (c) 1994 Adam Glass and Charles Hannum.  All rights reserved.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. All advertising materials mentioning features or use of this software
17  *    must display the following acknowledgement:
18  *      This product includes software developed by Adam Glass and Charles
19  *      Hannum.
20  * 4. The names of the authors may not be used to endorse or promote products
21  *    derived from this software without specific prior written permission.
22  *
23  * THIS SOFTWARE IS PROVIDED BY THE AUTHORS ``AS IS'' AND ANY EXPRESS OR
24  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
25  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
26  * IN NO EVENT SHALL THE AUTHORS BE LIABLE FOR ANY DIRECT, INDIRECT,
27  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
28  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
29  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
30  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
31  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
32  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
33  */
34
35 #include "opt_compat.h"
36 #include "opt_sysvipc.h"
37
38 #include <sys/param.h>
39 #include <sys/systm.h>
40 #include <sys/sysproto.h>
41 #include <sys/kernel.h>
42 #include <sys/sysctl.h>
43 #include <sys/shm.h>
44 #include <sys/proc.h>
45 #include <sys/malloc.h>
46 #include <sys/mman.h>
47 #include <sys/stat.h>
48 #include <sys/sysent.h>
49 #include <sys/jail.h>
50
51 #include <vm/vm.h>
52 #include <vm/vm_param.h>
53 #include <sys/lock.h>
54 #include <vm/pmap.h>
55 #include <vm/vm_object.h>
56 #include <vm/vm_map.h>
57 #include <vm/vm_page.h>
58 #include <vm/vm_pager.h>
59
60 static MALLOC_DEFINE(M_SHM, "shm", "SVID compatible shared memory segments");
61
62 struct oshmctl_args;
63 static int oshmctl __P((struct proc *p, struct oshmctl_args *uap));
64
65 static int shmget_allocate_segment __P((struct proc *p, struct shmget_args *uap, int mode));
66 static int shmget_existing __P((struct proc *p, struct shmget_args *uap, int mode, int segnum));
67
68 /* XXX casting to (sy_call_t *) is bogus, as usual. */
69 static sy_call_t *shmcalls[] = {
70         (sy_call_t *)shmat, (sy_call_t *)oshmctl,
71         (sy_call_t *)shmdt, (sy_call_t *)shmget,
72         (sy_call_t *)shmctl
73 };
74
75 #define SHMSEG_FREE             0x0200
76 #define SHMSEG_REMOVED          0x0400
77 #define SHMSEG_ALLOCATED        0x0800
78 #define SHMSEG_WANTED           0x1000
79
80 static int shm_last_free, shm_nused, shm_committed, shmalloced;
81 static struct shmid_ds  *shmsegs;
82
83 struct shm_handle {
84         /* vm_offset_t kva; */
85         vm_object_t shm_object;
86 };
87
88 struct shmmap_state {
89         vm_offset_t va;
90         int shmid;
91 };
92
93 static void shm_deallocate_segment __P((struct shmid_ds *));
94 static int shm_find_segment_by_key __P((key_t));
95 static struct shmid_ds *shm_find_segment_by_shmid __P((int));
96 static int shm_delete_mapping __P((struct proc *, struct shmmap_state *));
97 static void shmrealloc __P((void));
98 static void shminit __P((void *));
99
100 /*
101  * Tuneable values
102  */
103 #ifndef SHMMAXPGS
104 #define SHMMAXPGS       8192    /* note: sysv shared memory is swap backed */
105 #endif
106 #ifndef SHMMAX
107 #define SHMMAX  (SHMMAXPGS*PAGE_SIZE)
108 #endif
109 #ifndef SHMMIN
110 #define SHMMIN  1
111 #endif
112 #ifndef SHMMNI
113 #define SHMMNI  192
114 #endif
115 #ifndef SHMSEG
116 #define SHMSEG  128
117 #endif
118 #ifndef SHMALL
119 #define SHMALL  (SHMMAXPGS)
120 #endif
121
122 struct  shminfo shminfo = {
123         SHMMAX,
124         SHMMIN,
125         SHMMNI,
126         SHMSEG,
127         SHMALL
128 };
129
130 static int shm_use_phys;
131
132 TUNABLE_INT("kern.ipc.shmmin", &shminfo.shmmin);
133 TUNABLE_INT("kern.ipc.shmmni", &shminfo.shmmni);
134 TUNABLE_INT("kern.ipc.shmseg", &shminfo.shmseg);
135 TUNABLE_INT("kern.ipc.shmmaxpgs", &shminfo.shmall);
136 TUNABLE_INT("kern.ipc.shm_use_phys", &shm_use_phys);
137
138 SYSCTL_DECL(_kern_ipc);
139 SYSCTL_INT(_kern_ipc, OID_AUTO, shmmax, CTLFLAG_RW, &shminfo.shmmax, 0, "");
140 SYSCTL_INT(_kern_ipc, OID_AUTO, shmmin, CTLFLAG_RW, &shminfo.shmmin, 0, "");
141 SYSCTL_INT(_kern_ipc, OID_AUTO, shmmni, CTLFLAG_RD, &shminfo.shmmni, 0, "");
142 SYSCTL_INT(_kern_ipc, OID_AUTO, shmseg, CTLFLAG_RW, &shminfo.shmseg, 0, "");
143 SYSCTL_INT(_kern_ipc, OID_AUTO, shmall, CTLFLAG_RW, &shminfo.shmall, 0, "");
144 SYSCTL_INT(_kern_ipc, OID_AUTO, shm_use_phys, CTLFLAG_RW, &shm_use_phys, 0, "");
145
146 static int
147 shm_find_segment_by_key(key)
148         key_t key;
149 {
150         int i;
151
152         for (i = 0; i < shmalloced; i++)
153                 if ((shmsegs[i].shm_perm.mode & SHMSEG_ALLOCATED) &&
154                     shmsegs[i].shm_perm.key == key)
155                         return i;
156         return -1;
157 }
158
159 static struct shmid_ds *
160 shm_find_segment_by_shmid(shmid)
161         int shmid;
162 {
163         int segnum;
164         struct shmid_ds *shmseg;
165
166         segnum = IPCID_TO_IX(shmid);
167         if (segnum < 0 || segnum >= shmalloced)
168                 return NULL;
169         shmseg = &shmsegs[segnum];
170         if ((shmseg->shm_perm.mode & (SHMSEG_ALLOCATED | SHMSEG_REMOVED))
171             != SHMSEG_ALLOCATED ||
172             shmseg->shm_perm.seq != IPCID_TO_SEQ(shmid))
173                 return NULL;
174         return shmseg;
175 }
176
177 static void
178 shm_deallocate_segment(shmseg)
179         struct shmid_ds *shmseg;
180 {
181         struct shm_handle *shm_handle;
182         size_t size;
183
184         shm_handle = shmseg->shm_internal;
185         vm_object_deallocate(shm_handle->shm_object);
186         free((caddr_t)shm_handle, M_SHM);
187         shmseg->shm_internal = NULL;
188         size = round_page(shmseg->shm_segsz);
189         shm_committed -= btoc(size);
190         shm_nused--;
191         shmseg->shm_perm.mode = SHMSEG_FREE;
192 }
193
194 static int
195 shm_delete_mapping(p, shmmap_s)
196         struct proc *p;
197         struct shmmap_state *shmmap_s;
198 {
199         struct shmid_ds *shmseg;
200         int segnum, result;
201         size_t size;
202
203         segnum = IPCID_TO_IX(shmmap_s->shmid);
204         shmseg = &shmsegs[segnum];
205         size = round_page(shmseg->shm_segsz);
206         result = vm_map_remove(&p->p_vmspace->vm_map, shmmap_s->va, shmmap_s->va + size);
207         if (result != KERN_SUCCESS)
208                 return EINVAL;
209         shmmap_s->shmid = -1;
210         shmseg->shm_dtime = time_second;
211         if ((--shmseg->shm_nattch <= 0) &&
212             (shmseg->shm_perm.mode & SHMSEG_REMOVED)) {
213                 shm_deallocate_segment(shmseg);
214                 shm_last_free = segnum;
215         }
216         return 0;
217 }
218
219 #ifndef _SYS_SYSPROTO_H_
220 struct shmdt_args {
221         void *shmaddr;
222 };
223 #endif
224
225 int
226 shmdt(struct shmdt_args *uap)
227 {
228         struct proc *p = curproc;
229         struct shmmap_state *shmmap_s;
230         int i;
231
232         if (!jail_sysvipc_allowed && p->p_ucred->cr_prison != NULL)
233                 return (ENOSYS);
234
235         shmmap_s = (struct shmmap_state *)p->p_vmspace->vm_shm;
236         if (shmmap_s == NULL)
237             return EINVAL;
238         for (i = 0; i < shminfo.shmseg; i++, shmmap_s++)
239                 if (shmmap_s->shmid != -1 &&
240                     shmmap_s->va == (vm_offset_t)uap->shmaddr)
241                         break;
242         if (i == shminfo.shmseg)
243                 return EINVAL;
244         return shm_delete_mapping(p, shmmap_s);
245 }
246
247 #ifndef _SYS_SYSPROTO_H_
248 struct shmat_args {
249         int shmid;
250         void *shmaddr;
251         int shmflg;
252 };
253 #endif
254
255 int
256 shmat(struct shmat_args *uap)
257 {
258         struct proc *p = curproc;
259         int error, i, flags;
260         struct shmid_ds *shmseg;
261         struct shmmap_state *shmmap_s = NULL;
262         struct shm_handle *shm_handle;
263         vm_offset_t attach_va;
264         vm_prot_t prot;
265         vm_size_t size;
266         int rv;
267
268         if (!jail_sysvipc_allowed && p->p_ucred->cr_prison != NULL)
269                 return (ENOSYS);
270
271         shmmap_s = (struct shmmap_state *)p->p_vmspace->vm_shm;
272         if (shmmap_s == NULL) {
273                 size = shminfo.shmseg * sizeof(struct shmmap_state);
274                 shmmap_s = malloc(size, M_SHM, M_WAITOK);
275                 for (i = 0; i < shminfo.shmseg; i++)
276                         shmmap_s[i].shmid = -1;
277                 p->p_vmspace->vm_shm = (caddr_t)shmmap_s;
278         }
279         shmseg = shm_find_segment_by_shmid(uap->shmid);
280         if (shmseg == NULL)
281                 return EINVAL;
282         error = ipcperm(p, &shmseg->shm_perm,
283             (uap->shmflg & SHM_RDONLY) ? IPC_R : IPC_R|IPC_W);
284         if (error)
285                 return error;
286         for (i = 0; i < shminfo.shmseg; i++) {
287                 if (shmmap_s->shmid == -1)
288                         break;
289                 shmmap_s++;
290         }
291         if (i >= shminfo.shmseg)
292                 return EMFILE;
293         size = round_page(shmseg->shm_segsz);
294 #ifdef VM_PROT_READ_IS_EXEC
295         prot = VM_PROT_READ | VM_PROT_EXECUTE;
296 #else
297         prot = VM_PROT_READ;
298 #endif
299         if ((uap->shmflg & SHM_RDONLY) == 0)
300                 prot |= VM_PROT_WRITE;
301         flags = MAP_ANON | MAP_SHARED;
302         if (uap->shmaddr) {
303                 flags |= MAP_FIXED;
304                 if (uap->shmflg & SHM_RND)
305                         attach_va = (vm_offset_t)uap->shmaddr & ~(SHMLBA-1);
306                 else if (((vm_offset_t)uap->shmaddr & (SHMLBA-1)) == 0)
307                         attach_va = (vm_offset_t)uap->shmaddr;
308                 else
309                         return EINVAL;
310         } else {
311                 /* This is just a hint to vm_map_find() about where to put it. */
312                 attach_va = round_page((vm_offset_t)p->p_vmspace->vm_taddr + maxtsiz + maxdsiz);
313         }
314
315         shm_handle = shmseg->shm_internal;
316         vm_object_reference(shm_handle->shm_object);
317         rv = vm_map_find(&p->p_vmspace->vm_map, shm_handle->shm_object,
318                 0, &attach_va, size, (flags & MAP_FIXED)?0:1, prot, prot, 0);
319         if (rv != KERN_SUCCESS) {
320                 return ENOMEM;
321         }
322         vm_map_inherit(&p->p_vmspace->vm_map,
323                 attach_va, attach_va + size, VM_INHERIT_SHARE);
324
325         shmmap_s->va = attach_va;
326         shmmap_s->shmid = uap->shmid;
327         shmseg->shm_lpid = p->p_pid;
328         shmseg->shm_atime = time_second;
329         shmseg->shm_nattch++;
330         p->p_retval[0] = attach_va;
331         return 0;
332 }
333
334 struct oshmid_ds {
335         struct  ipc_perm shm_perm;      /* operation perms */
336         int     shm_segsz;              /* size of segment (bytes) */
337         ushort  shm_cpid;               /* pid, creator */
338         ushort  shm_lpid;               /* pid, last operation */
339         short   shm_nattch;             /* no. of current attaches */
340         time_t  shm_atime;              /* last attach time */
341         time_t  shm_dtime;              /* last detach time */
342         time_t  shm_ctime;              /* last change time */
343         void    *shm_handle;            /* internal handle for shm segment */
344 };
345
346 struct oshmctl_args {
347         int shmid;
348         int cmd;
349         struct oshmid_ds *ubuf;
350 };
351
352 static int
353 oshmctl(p, uap)
354         struct proc *p;
355         struct oshmctl_args *uap;
356 {
357 #ifdef COMPAT_43
358         int error;
359         struct shmid_ds *shmseg;
360         struct oshmid_ds outbuf;
361
362         if (!jail_sysvipc_allowed && p->p_ucred->cr_prison != NULL)
363                 return (ENOSYS);
364
365         shmseg = shm_find_segment_by_shmid(uap->shmid);
366         if (shmseg == NULL)
367                 return EINVAL;
368         switch (uap->cmd) {
369         case IPC_STAT:
370                 error = ipcperm(p, &shmseg->shm_perm, IPC_R);
371                 if (error)
372                         return error;
373                 outbuf.shm_perm = shmseg->shm_perm;
374                 outbuf.shm_segsz = shmseg->shm_segsz;
375                 outbuf.shm_cpid = shmseg->shm_cpid;
376                 outbuf.shm_lpid = shmseg->shm_lpid;
377                 outbuf.shm_nattch = shmseg->shm_nattch;
378                 outbuf.shm_atime = shmseg->shm_atime;
379                 outbuf.shm_dtime = shmseg->shm_dtime;
380                 outbuf.shm_ctime = shmseg->shm_ctime;
381                 outbuf.shm_handle = shmseg->shm_internal;
382                 error = copyout((caddr_t)&outbuf, uap->ubuf, sizeof(outbuf));
383                 if (error)
384                         return error;
385                 break;
386         default:
387                 /* XXX casting to (sy_call_t *) is bogus, as usual. */
388                 return ((sy_call_t *)shmctl)(uap);
389         }
390         return 0;
391 #else
392         return EINVAL;
393 #endif
394 }
395
396 #ifndef _SYS_SYSPROTO_H_
397 struct shmctl_args {
398         int shmid;
399         int cmd;
400         struct shmid_ds *buf;
401 };
402 #endif
403
404 int
405 shmctl(struct shmctl_args *uap)
406 {
407         struct proc *p = curproc;
408         int error;
409         struct shmid_ds inbuf;
410         struct shmid_ds *shmseg;
411
412         if (!jail_sysvipc_allowed && p->p_ucred->cr_prison != NULL)
413                 return (ENOSYS);
414
415         shmseg = shm_find_segment_by_shmid(uap->shmid);
416         if (shmseg == NULL)
417                 return EINVAL;
418         switch (uap->cmd) {
419         case IPC_STAT:
420                 error = ipcperm(p, &shmseg->shm_perm, IPC_R);
421                 if (error)
422                         return error;
423                 error = copyout((caddr_t)shmseg, uap->buf, sizeof(inbuf));
424                 if (error)
425                         return error;
426                 break;
427         case IPC_SET:
428                 error = ipcperm(p, &shmseg->shm_perm, IPC_M);
429                 if (error)
430                         return error;
431                 error = copyin(uap->buf, (caddr_t)&inbuf, sizeof(inbuf));
432                 if (error)
433                         return error;
434                 shmseg->shm_perm.uid = inbuf.shm_perm.uid;
435                 shmseg->shm_perm.gid = inbuf.shm_perm.gid;
436                 shmseg->shm_perm.mode =
437                     (shmseg->shm_perm.mode & ~ACCESSPERMS) |
438                     (inbuf.shm_perm.mode & ACCESSPERMS);
439                 shmseg->shm_ctime = time_second;
440                 break;
441         case IPC_RMID:
442                 error = ipcperm(p, &shmseg->shm_perm, IPC_M);
443                 if (error)
444                         return error;
445                 shmseg->shm_perm.key = IPC_PRIVATE;
446                 shmseg->shm_perm.mode |= SHMSEG_REMOVED;
447                 if (shmseg->shm_nattch <= 0) {
448                         shm_deallocate_segment(shmseg);
449                         shm_last_free = IPCID_TO_IX(uap->shmid);
450                 }
451                 break;
452 #if 0
453         case SHM_LOCK:
454         case SHM_UNLOCK:
455 #endif
456         default:
457                 return EINVAL;
458         }
459         return 0;
460 }
461
462 #ifndef _SYS_SYSPROTO_H_
463 struct shmget_args {
464         key_t key;
465         size_t size;
466         int shmflg;
467 };
468 #endif
469
470 static int
471 shmget_existing(p, uap, mode, segnum)
472         struct proc *p;
473         struct shmget_args *uap;
474         int mode;
475         int segnum;
476 {
477         struct shmid_ds *shmseg;
478         int error;
479
480         shmseg = &shmsegs[segnum];
481         if (shmseg->shm_perm.mode & SHMSEG_REMOVED) {
482                 /*
483                  * This segment is in the process of being allocated.  Wait
484                  * until it's done, and look the key up again (in case the
485                  * allocation failed or it was freed).
486                  */
487                 shmseg->shm_perm.mode |= SHMSEG_WANTED;
488                 error = tsleep((caddr_t)shmseg, PLOCK | PCATCH, "shmget", 0);
489                 if (error)
490                         return error;
491                 return EAGAIN;
492         }
493         if ((uap->shmflg & (IPC_CREAT | IPC_EXCL)) == (IPC_CREAT | IPC_EXCL))
494                 return EEXIST;
495         error = ipcperm(p, &shmseg->shm_perm, mode);
496         if (error)
497                 return error;
498         if (uap->size && uap->size > shmseg->shm_segsz)
499                 return EINVAL;
500         p->p_retval[0] = IXSEQ_TO_IPCID(segnum, shmseg->shm_perm);
501         return 0;
502 }
503
504 static int
505 shmget_allocate_segment(p, uap, mode)
506         struct proc *p;
507         struct shmget_args *uap;
508         int mode;
509 {
510         int i, segnum, shmid, size;
511         struct ucred *cred = p->p_ucred;
512         struct shmid_ds *shmseg;
513         struct shm_handle *shm_handle;
514
515         if (uap->size < shminfo.shmmin || uap->size > shminfo.shmmax)
516                 return EINVAL;
517         if (shm_nused >= shminfo.shmmni) /* any shmids left? */
518                 return ENOSPC;
519         size = round_page(uap->size);
520         if (shm_committed + btoc(size) > shminfo.shmall)
521                 return ENOMEM;
522         if (shm_last_free < 0) {
523                 shmrealloc();   /* maybe expand the shmsegs[] array */
524                 for (i = 0; i < shmalloced; i++)
525                         if (shmsegs[i].shm_perm.mode & SHMSEG_FREE)
526                                 break;
527                 if (i == shmalloced)
528                         return ENOSPC;
529                 segnum = i;
530         } else  {
531                 segnum = shm_last_free;
532                 shm_last_free = -1;
533         }
534         shmseg = &shmsegs[segnum];
535         /*
536          * In case we sleep in malloc(), mark the segment present but deleted
537          * so that noone else tries to create the same key.
538          */
539         shmseg->shm_perm.mode = SHMSEG_ALLOCATED | SHMSEG_REMOVED;
540         shmseg->shm_perm.key = uap->key;
541         shmseg->shm_perm.seq = (shmseg->shm_perm.seq + 1) & 0x7fff;
542         shm_handle = (struct shm_handle *)
543             malloc(sizeof(struct shm_handle), M_SHM, M_WAITOK);
544         shmid = IXSEQ_TO_IPCID(segnum, shmseg->shm_perm);
545         
546         /*
547          * We make sure that we have allocated a pager before we need
548          * to.
549          */
550         if (shm_use_phys) {
551                 shm_handle->shm_object =
552                     vm_pager_allocate(OBJT_PHYS, 0, size, VM_PROT_DEFAULT, 0);
553         } else {
554                 shm_handle->shm_object =
555                     vm_pager_allocate(OBJT_SWAP, 0, size, VM_PROT_DEFAULT, 0);
556         }
557         vm_object_clear_flag(shm_handle->shm_object, OBJ_ONEMAPPING);
558         vm_object_set_flag(shm_handle->shm_object, OBJ_NOSPLIT);
559
560         shmseg->shm_internal = shm_handle;
561         shmseg->shm_perm.cuid = shmseg->shm_perm.uid = cred->cr_uid;
562         shmseg->shm_perm.cgid = shmseg->shm_perm.gid = cred->cr_gid;
563         shmseg->shm_perm.mode = (shmseg->shm_perm.mode & SHMSEG_WANTED) |
564             (mode & ACCESSPERMS) | SHMSEG_ALLOCATED;
565         shmseg->shm_segsz = uap->size;
566         shmseg->shm_cpid = p->p_pid;
567         shmseg->shm_lpid = shmseg->shm_nattch = 0;
568         shmseg->shm_atime = shmseg->shm_dtime = 0;
569         shmseg->shm_ctime = time_second;
570         shm_committed += btoc(size);
571         shm_nused++;
572         if (shmseg->shm_perm.mode & SHMSEG_WANTED) {
573                 /*
574                  * Somebody else wanted this key while we were asleep.  Wake
575                  * them up now.
576                  */
577                 shmseg->shm_perm.mode &= ~SHMSEG_WANTED;
578                 wakeup((caddr_t)shmseg);
579         }
580         p->p_retval[0] = shmid;
581         return 0;
582 }
583
584 int
585 shmget(struct shmget_args *uap)
586 {
587         struct proc *p = curproc;
588         int segnum, mode, error;
589
590         if (!jail_sysvipc_allowed && p->p_ucred->cr_prison != NULL)
591                 return (ENOSYS);
592
593         mode = uap->shmflg & ACCESSPERMS;
594         if (uap->key != IPC_PRIVATE) {
595         again:
596                 segnum = shm_find_segment_by_key(uap->key);
597                 if (segnum >= 0) {
598                         error = shmget_existing(p, uap, mode, segnum);
599                         if (error == EAGAIN)
600                                 goto again;
601                         return error;
602                 }
603                 if ((uap->shmflg & IPC_CREAT) == 0)
604                         return ENOENT;
605         }
606         return shmget_allocate_segment(p, uap, mode);
607 }
608
609 /*
610  *  shmsys_args(u_int which, int a2, ...) (VARARGS)
611  */
612 int
613 shmsys(struct shmsys_args *uap)
614 {
615         struct proc *p = curproc;
616
617         if (!jail_sysvipc_allowed && p->p_ucred->cr_prison != NULL)
618                 return (ENOSYS);
619
620         if (uap->which >= sizeof(shmcalls)/sizeof(shmcalls[0]))
621                 return EINVAL;
622         return ((*shmcalls[uap->which])(&uap->a2));
623 }
624
625 void
626 shmfork(p1, p2)
627         struct proc *p1, *p2;
628 {
629         struct shmmap_state *shmmap_s;
630         size_t size;
631         int i;
632
633         size = shminfo.shmseg * sizeof(struct shmmap_state);
634         shmmap_s = malloc(size, M_SHM, M_WAITOK);
635         bcopy((caddr_t)p1->p_vmspace->vm_shm, (caddr_t)shmmap_s, size);
636         p2->p_vmspace->vm_shm = (caddr_t)shmmap_s;
637         for (i = 0; i < shminfo.shmseg; i++, shmmap_s++)
638                 if (shmmap_s->shmid != -1)
639                         shmsegs[IPCID_TO_IX(shmmap_s->shmid)].shm_nattch++;
640 }
641
642 void
643 shmexit(p)
644         struct proc *p;
645 {
646         struct shmmap_state *shmmap_s;
647         int i;
648
649         shmmap_s = (struct shmmap_state *)p->p_vmspace->vm_shm;
650         for (i = 0; i < shminfo.shmseg; i++, shmmap_s++)
651                 if (shmmap_s->shmid != -1)
652                         shm_delete_mapping(p, shmmap_s);
653         free((caddr_t)p->p_vmspace->vm_shm, M_SHM);
654         p->p_vmspace->vm_shm = NULL;
655 }
656
657 static void
658 shmrealloc(void)
659 {
660         int i;
661         struct shmid_ds *newsegs;
662
663         if (shmalloced >= shminfo.shmmni)
664                 return;
665
666         newsegs = malloc(shminfo.shmmni * sizeof(*newsegs), M_SHM, M_WAITOK);
667         if (newsegs == NULL)
668                 return;
669         for (i = 0; i < shmalloced; i++)
670                 bcopy(&shmsegs[i], &newsegs[i], sizeof(newsegs[0]));
671         for (; i < shminfo.shmmni; i++) {
672                 shmsegs[i].shm_perm.mode = SHMSEG_FREE;
673                 shmsegs[i].shm_perm.seq = 0;
674         }
675         free(shmsegs, M_SHM);
676         shmsegs = newsegs;
677         shmalloced = shminfo.shmmni;
678 }
679
680 static void
681 shminit(dummy)
682         void *dummy;
683 {
684         int i;
685
686         shminfo.shmmax = shminfo.shmall * PAGE_SIZE;
687         shmalloced = shminfo.shmmni;
688         shmsegs = malloc(shmalloced * sizeof(shmsegs[0]), M_SHM, M_WAITOK);
689         if (shmsegs == NULL)
690                 panic("cannot allocate initial memory for sysvshm");
691         for (i = 0; i < shmalloced; i++) {
692                 shmsegs[i].shm_perm.mode = SHMSEG_FREE;
693                 shmsegs[i].shm_perm.seq = 0;
694         }
695         shm_last_free = 0;
696         shm_nused = 0;
697         shm_committed = 0;
698 }
699 SYSINIT(sysv_shm, SI_SUB_SYSV_SHM, SI_ORDER_FIRST, shminit, NULL);