sysv_shm.c revision 8876 
1/*	$Id: sysv_shm.c,v 1.4 1995/02/20 22:23:13 davidg Exp $ */
2/*	$NetBSD: sysv_shm.c,v 1.23 1994/07/04 23:25:12 glass Exp $	*/
3
4/*
5 * Copyright (c) 1994 Adam Glass and Charles Hannum.  All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 * 1. Redistributions of source code must retain the above copyright
11 *    notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 *    notice, this list of conditions and the following disclaimer in the
14 *    documentation and/or other materials provided with the distribution.
15 * 3. All advertising materials mentioning features or use of this software
16 *    must display the following acknowledgement:
17 *	This product includes software developed by Adam Glass and Charles
18 *	Hannum.
19 * 4. The names of the authors may not be used to endorse or promote products
20 *    derived from this software without specific prior written permission.
21 *
22 * THIS SOFTWARE IS PROVIDED BY THE AUTHORS ``AS IS'' AND ANY EXPRESS OR
23 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
24 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
25 * IN NO EVENT SHALL THE AUTHORS BE LIABLE FOR ANY DIRECT, INDIRECT,
26 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
27 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
31 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
32 */
33
34#include <sys/types.h>
35#include <sys/param.h>
36#include <sys/kernel.h>
37#include <sys/shm.h>
38#include <sys/proc.h>
39#include <sys/uio.h>
40#include <sys/time.h>
41#include <sys/malloc.h>
42#include <sys/mman.h>
43#include <sys/systm.h>
44#include <sys/stat.h>
45
46#include <vm/vm.h>
47#include <vm/vm_map.h>
48#include <vm/vm_map.h>
49#include <vm/vm_kern.h>
50
51/*
52 * Provides the following externally accessible functions:
53 *
54 * shminit(void);		           initialization
55 * shmexit(struct proc *)                  cleanup
56 * shmfork(struct proc *, struct proc *, int) fork handling
57 * shmsys(arg1, arg2, arg3, arg4);         shm{at,ctl,dt,get}(arg2, arg3, arg4)
58 *
59 * Structures:
60 * shmsegs (an array of 'struct shmid_ds')
61 * per proc array of 'struct shmmap_state'
62 */
63
64int	oshmctl();
65int	shmat(), shmctl(), shmdt(), shmget();
66int	(*shmcalls[])() = { shmat, oshmctl, shmdt, shmget, shmctl };
67
68#define	SHMSEG_FREE     	0x0200
69#define	SHMSEG_REMOVED  	0x0400
70#define	SHMSEG_ALLOCATED	0x0800
71#define	SHMSEG_WANTED		0x1000
72
73vm_map_t sysvshm_map;
74int shm_last_free, shm_nused, shm_committed;
75
76struct shm_handle {
77	vm_offset_t kva;
78};
79
80struct shmmap_state {
81	vm_offset_t va;
82	int shmid;
83};
84
85static void shm_deallocate_segment __P((struct shmid_ds *));
86static int shm_find_segment_by_key __P((key_t));
87static struct shmid_ds *shm_find_segment_by_shmid __P((int));
88static int shm_delete_mapping __P((struct proc *, struct shmmap_state *));
89
90static int
91shm_find_segment_by_key(key)
92	key_t key;
93{
94	int i;
95
96	for (i = 0; i < shminfo.shmmni; i++)
97		if ((shmsegs[i].shm_perm.mode & SHMSEG_ALLOCATED) &&
98		    shmsegs[i].shm_perm.key == key)
99			return i;
100	return -1;
101}
102
103static struct shmid_ds *
104shm_find_segment_by_shmid(shmid)
105	int shmid;
106{
107	int segnum;
108	struct shmid_ds *shmseg;
109
110	segnum = IPCID_TO_IX(shmid);
111	if (segnum < 0 || segnum >= shminfo.shmmni)
112		return NULL;
113	shmseg = &shmsegs[segnum];
114	if ((shmseg->shm_perm.mode & (SHMSEG_ALLOCATED | SHMSEG_REMOVED))
115	    != SHMSEG_ALLOCATED ||
116	    shmseg->shm_perm.seq != IPCID_TO_SEQ(shmid))
117		return NULL;
118	return shmseg;
119}
120
121static void
122shm_deallocate_segment(shmseg)
123	struct shmid_ds *shmseg;
124{
125	struct shm_handle *shm_handle;
126	size_t size;
127
128	shm_handle = shmseg->shm_internal;
129	size = (shmseg->shm_segsz + CLOFSET) & ~CLOFSET;
130	(void) vm_map_remove(sysvshm_map, shm_handle->kva, shm_handle->kva + size);
131	free((caddr_t)shm_handle, M_SHM);
132	shmseg->shm_internal = NULL;
133	shm_committed -= btoc(size);
134	shm_nused--;
135	shmseg->shm_perm.mode = SHMSEG_FREE;
136}
137
138static int
139shm_delete_mapping(p, shmmap_s)
140	struct proc *p;
141	struct shmmap_state *shmmap_s;
142{
143	struct shmid_ds *shmseg;
144	int segnum, result;
145	size_t size;
146
147	segnum = IPCID_TO_IX(shmmap_s->shmid);
148	shmseg = &shmsegs[segnum];
149	size = (shmseg->shm_segsz + CLOFSET) & ~CLOFSET;
150	result = vm_map_remove(&p->p_vmspace->vm_map, shmmap_s->va, shmmap_s->va + size);
151	if (result != KERN_SUCCESS)
152		return EINVAL;
153	shmmap_s->shmid = -1;
154	shmseg->shm_dtime = time.tv_sec;
155	if ((--shmseg->shm_nattch <= 0) &&
156	    (shmseg->shm_perm.mode & SHMSEG_REMOVED)) {
157		shm_deallocate_segment(shmseg);
158		shm_last_free = segnum;
159	}
160	return 0;
161}
162
163struct shmdt_args {
164	void *shmaddr;
165};
166int
167shmdt(p, uap, retval)
168	struct proc *p;
169	struct shmdt_args *uap;
170	int *retval;
171{
172	struct shmmap_state *shmmap_s;
173	int i;
174
175	shmmap_s = (struct shmmap_state *)p->p_vmspace->vm_shm;
176	for (i = 0; i < shminfo.shmseg; i++, shmmap_s++)
177		if (shmmap_s->shmid != -1 &&
178		    shmmap_s->va == (vm_offset_t)uap->shmaddr)
179			break;
180	if (i == shminfo.shmseg)
181		return EINVAL;
182	return shm_delete_mapping(p, shmmap_s);
183}
184
185struct shmat_args {
186	int shmid;
187	void *shmaddr;
188	int shmflg;
189};
190int
191shmat(p, uap, retval)
192	struct proc *p;
193	struct shmat_args *uap;
194	int *retval;
195{
196	int error, i, flags;
197	struct ucred *cred = p->p_ucred;
198	struct shmid_ds *shmseg;
199	struct shmmap_state *shmmap_s = NULL;
200	vm_offset_t attach_va;
201	vm_prot_t prot;
202	vm_size_t size;
203
204	shmmap_s = (struct shmmap_state *)p->p_vmspace->vm_shm;
205	if (shmmap_s == NULL) {
206		size = shminfo.shmseg * sizeof(struct shmmap_state);
207		shmmap_s = malloc(size, M_SHM, M_WAITOK);
208		for (i = 0; i < shminfo.shmseg; i++)
209			shmmap_s[i].shmid = -1;
210		p->p_vmspace->vm_shm = (caddr_t)shmmap_s;
211	}
212	shmseg = shm_find_segment_by_shmid(uap->shmid);
213	if (shmseg == NULL)
214		return EINVAL;
215	error = ipcperm(cred, &shmseg->shm_perm,
216	    (uap->shmflg & SHM_RDONLY) ? IPC_R : IPC_R|IPC_W);
217	if (error)
218		return error;
219	for (i = 0; i < shminfo.shmseg; i++) {
220		if (shmmap_s->shmid == -1)
221			break;
222		shmmap_s++;
223	}
224	if (i >= shminfo.shmseg)
225		return EMFILE;
226	size = (shmseg->shm_segsz + CLOFSET) & ~CLOFSET;
227	prot = VM_PROT_READ;
228	if ((uap->shmflg & SHM_RDONLY) == 0)
229		prot |= VM_PROT_WRITE;
230	flags = MAP_ANON | MAP_SHARED;
231	if (uap->shmaddr) {
232		flags |= MAP_FIXED;
233		if (uap->shmflg & SHM_RND)
234			attach_va = (vm_offset_t)uap->shmaddr & ~(SHMLBA-1);
235		else if (((vm_offset_t)uap->shmaddr & (SHMLBA-1)) == 0)
236			attach_va = (vm_offset_t)uap->shmaddr;
237		else
238			return EINVAL;
239	} else {
240		/* This is just a hint to vm_mmap() about where to put it. */
241		attach_va = round_page(p->p_vmspace->vm_daddr + MAXDSIZ);
242	}
243	error = vm_mmap(&p->p_vmspace->vm_map, &attach_va, size, prot,
244	    VM_PROT_DEFAULT, flags, (caddr_t) uap->shmid, 0);
245	if (error)
246		return error;
247	shmmap_s->va = attach_va;
248	shmmap_s->shmid = uap->shmid;
249	shmseg->shm_lpid = p->p_pid;
250	shmseg->shm_atime = time.tv_sec;
251	shmseg->shm_nattch++;
252	*retval = attach_va;
253	return 0;
254}
255
256struct oshmid_ds {
257	struct	ipc_perm shm_perm;	/* operation perms */
258	int	shm_segsz;		/* size of segment (bytes) */
259	ushort	shm_cpid;		/* pid, creator */
260	ushort	shm_lpid;		/* pid, last operation */
261	short	shm_nattch;		/* no. of current attaches */
262	time_t	shm_atime;		/* last attach time */
263	time_t	shm_dtime;		/* last detach time */
264	time_t	shm_ctime;		/* last change time */
265	void	*shm_handle;		/* internal handle for shm segment */
266};
267
268struct oshmctl_args {
269	int shmid;
270	int cmd;
271	struct oshmid_ds *ubuf;
272};
273
274int
275oshmctl(p, uap, retval)
276	struct proc *p;
277	struct oshmctl_args *uap;
278	int *retval;
279{
280#ifdef COMPAT_43
281	int error;
282	struct ucred *cred = p->p_ucred;
283	struct shmid_ds *shmseg;
284	struct oshmid_ds outbuf;
285
286	shmseg = shm_find_segment_by_shmid(uap->shmid);
287	if (shmseg == NULL)
288		return EINVAL;
289	switch (uap->cmd) {
290	case IPC_STAT:
291		error = ipcperm(cred, &shmseg->shm_perm, IPC_R);
292		if (error)
293			return error;
294		outbuf.shm_perm = shmseg->shm_perm;
295		outbuf.shm_segsz = shmseg->shm_segsz;
296		outbuf.shm_cpid = shmseg->shm_cpid;
297		outbuf.shm_lpid = shmseg->shm_lpid;
298		outbuf.shm_nattch = shmseg->shm_nattch;
299		outbuf.shm_atime = shmseg->shm_atime;
300		outbuf.shm_dtime = shmseg->shm_dtime;
301		outbuf.shm_ctime = shmseg->shm_ctime;
302		outbuf.shm_handle = shmseg->shm_internal;
303		error = copyout((caddr_t)&outbuf, uap->ubuf, sizeof(outbuf));
304		if (error)
305			return error;
306		break;
307	default:
308		return shmctl(p, uap, retval);
309	}
310	return 0;
311#else
312	return EINVAL;
313#endif
314}
315
316struct shmctl_args {
317	int shmid;
318	int cmd;
319	struct shmid_ds *ubuf;
320};
321int
322shmctl(p, uap, retval)
323	struct proc *p;
324	struct shmctl_args *uap;
325	int *retval;
326{
327	int error;
328	struct ucred *cred = p->p_ucred;
329	struct shmid_ds inbuf;
330	struct shmid_ds *shmseg;
331
332	shmseg = shm_find_segment_by_shmid(uap->shmid);
333	if (shmseg == NULL)
334		return EINVAL;
335	switch (uap->cmd) {
336	case IPC_STAT:
337		error = ipcperm(cred, &shmseg->shm_perm, IPC_R);
338		if (error)
339			return error;
340		error = copyout((caddr_t)shmseg, uap->ubuf, sizeof(inbuf));
341		if (error)
342			return error;
343		break;
344	case IPC_SET:
345		error = ipcperm(cred, &shmseg->shm_perm, IPC_M);
346		if (error)
347			return error;
348		error = copyin(uap->ubuf, (caddr_t)&inbuf, sizeof(inbuf));
349		if (error)
350			return error;
351		shmseg->shm_perm.uid = inbuf.shm_perm.uid;
352		shmseg->shm_perm.gid = inbuf.shm_perm.gid;
353		shmseg->shm_perm.mode =
354		    (shmseg->shm_perm.mode & ~ACCESSPERMS) |
355		    (inbuf.shm_perm.mode & ACCESSPERMS);
356		shmseg->shm_ctime = time.tv_sec;
357		break;
358	case IPC_RMID:
359		error = ipcperm(cred, &shmseg->shm_perm, IPC_M);
360		if (error)
361			return error;
362		shmseg->shm_perm.key = IPC_PRIVATE;
363		shmseg->shm_perm.mode |= SHMSEG_REMOVED;
364		if (shmseg->shm_nattch <= 0) {
365			shm_deallocate_segment(shmseg);
366			shm_last_free = IPCID_TO_IX(uap->shmid);
367		}
368		break;
369#if 0
370	case SHM_LOCK:
371	case SHM_UNLOCK:
372#endif
373	default:
374		return EINVAL;
375	}
376	return 0;
377}
378
379struct shmget_args {
380	key_t key;
381	size_t size;
382	int shmflg;
383};
384static int
385shmget_existing(p, uap, mode, segnum, retval)
386	struct proc *p;
387	struct shmget_args *uap;
388	int mode;
389	int segnum;
390	int *retval;
391{
392	struct shmid_ds *shmseg;
393	struct ucred *cred = p->p_ucred;
394	int error;
395
396	shmseg = &shmsegs[segnum];
397	if (shmseg->shm_perm.mode & SHMSEG_REMOVED) {
398		/*
399		 * This segment is in the process of being allocated.  Wait
400		 * until it's done, and look the key up again (in case the
401		 * allocation failed or it was freed).
402		 */
403		shmseg->shm_perm.mode |= SHMSEG_WANTED;
404		error = tsleep((caddr_t)shmseg, PLOCK | PCATCH, "shmget", 0);
405		if (error)
406			return error;
407		return EAGAIN;
408	}
409	error = ipcperm(cred, &shmseg->shm_perm, mode);
410	if (error)
411		return error;
412	if (uap->size && uap->size > shmseg->shm_segsz)
413		return EINVAL;
414	if (uap->shmflg & (IPC_CREAT | IPC_EXCL) == (IPC_CREAT | IPC_EXCL))
415		return EEXIST;
416	*retval = IXSEQ_TO_IPCID(segnum, shmseg->shm_perm);
417	return 0;
418}
419
420static int
421shmget_allocate_segment(p, uap, mode, retval)
422	struct proc *p;
423	struct shmget_args *uap;
424	int mode;
425	int *retval;
426{
427	int i, segnum, result, shmid, size;
428	struct ucred *cred = p->p_ucred;
429	struct shmid_ds *shmseg;
430	struct shm_handle *shm_handle;
431
432	if (uap->size < shminfo.shmmin || uap->size > shminfo.shmmax)
433		return EINVAL;
434	if (shm_nused >= shminfo.shmmni) /* any shmids left? */
435		return ENOSPC;
436	size = (uap->size + CLOFSET) & ~CLOFSET;
437	if (shm_committed + btoc(size) > shminfo.shmall)
438		return ENOMEM;
439	if (shm_last_free < 0) {
440		for (i = 0; i < shminfo.shmmni; i++)
441			if (shmsegs[i].shm_perm.mode & SHMSEG_FREE)
442				break;
443		if (i == shminfo.shmmni)
444			panic("shmseg free count inconsistent");
445		segnum = i;
446	} else  {
447		segnum = shm_last_free;
448		shm_last_free = -1;
449	}
450	shmseg = &shmsegs[segnum];
451	/*
452	 * In case we sleep in malloc(), mark the segment present but deleted
453	 * so that noone else tries to create the same key.
454	 */
455	shmseg->shm_perm.mode = SHMSEG_ALLOCATED | SHMSEG_REMOVED;
456	shmseg->shm_perm.key = uap->key;
457	shmseg->shm_perm.seq = (shmseg->shm_perm.seq + 1) & 0x7fff;
458	shm_handle = (struct shm_handle *)
459	    malloc(sizeof(struct shm_handle), M_SHM, M_WAITOK);
460	shmid = IXSEQ_TO_IPCID(segnum, shmseg->shm_perm);
461	result = vm_mmap(sysvshm_map, &shm_handle->kva, size, VM_PROT_ALL,
462	    VM_PROT_DEFAULT, MAP_ANON, (caddr_t) shmid, 0);
463	if (result != KERN_SUCCESS) {
464		shmseg->shm_perm.mode = SHMSEG_FREE;
465		shm_last_free = segnum;
466		free((caddr_t)shm_handle, M_SHM);
467		/* Just in case. */
468		wakeup((caddr_t)shmseg);
469		return ENOMEM;
470	}
471	shmseg->shm_internal = shm_handle;
472	shmseg->shm_perm.cuid = shmseg->shm_perm.uid = cred->cr_uid;
473	shmseg->shm_perm.cgid = shmseg->shm_perm.gid = cred->cr_gid;
474	shmseg->shm_perm.mode = (shmseg->shm_perm.mode & SHMSEG_WANTED) |
475	    (mode & ACCESSPERMS) | SHMSEG_ALLOCATED;
476	shmseg->shm_segsz = uap->size;
477	shmseg->shm_cpid = p->p_pid;
478	shmseg->shm_lpid = shmseg->shm_nattch = 0;
479	shmseg->shm_atime = shmseg->shm_dtime = 0;
480	shmseg->shm_ctime = time.tv_sec;
481	shm_committed += btoc(size);
482	shm_nused++;
483	if (shmseg->shm_perm.mode & SHMSEG_WANTED) {
484		/*
485		 * Somebody else wanted this key while we were asleep.  Wake
486		 * them up now.
487		 */
488		shmseg->shm_perm.mode &= ~SHMSEG_WANTED;
489		wakeup((caddr_t)shmseg);
490	}
491	*retval = shmid;
492	return 0;
493}
494
495int
496shmget(p, uap, retval)
497	struct proc *p;
498	struct shmget_args *uap;
499	int *retval;
500{
501	int segnum, mode, error;
502
503	mode = uap->shmflg & ACCESSPERMS;
504	if (uap->key != IPC_PRIVATE) {
505	again:
506		segnum = shm_find_segment_by_key(uap->key);
507		if (segnum >= 0) {
508			error = shmget_existing(p, uap, mode, segnum, retval);
509			if (error == EAGAIN)
510				goto again;
511			return error;
512		}
513		if ((uap->shmflg & IPC_CREAT) == 0)
514			return ENOENT;
515	}
516	return shmget_allocate_segment(p, uap, mode, retval);
517}
518
519struct shmsys_args {
520	u_int	which;
521};
522int
523shmsys(p, uap, retval)
524	struct proc *p;
525	struct shmsys_args *uap;
526	int *retval;
527{
528
529	if (uap->which >= sizeof(shmcalls)/sizeof(shmcalls[0]))
530		return EINVAL;
531	return ((*shmcalls[uap->which])(p, &uap[1], retval));
532}
533
534void
535shmfork(p1, p2, isvfork)
536	struct proc *p1, *p2;
537	int isvfork;
538{
539	struct shmmap_state *shmmap_s;
540	size_t size;
541	int i;
542
543	size = shminfo.shmseg * sizeof(struct shmmap_state);
544	shmmap_s = malloc(size, M_SHM, M_WAITOK);
545	bcopy((caddr_t)p1->p_vmspace->vm_shm, (caddr_t)shmmap_s, size);
546	p2->p_vmspace->vm_shm = (caddr_t)shmmap_s;
547	for (i = 0; i < shminfo.shmseg; i++, shmmap_s++)
548		if (shmmap_s->shmid != -1)
549			shmsegs[IPCID_TO_IX(shmmap_s->shmid)].shm_nattch++;
550}
551
552void
553shmexit(p)
554	struct proc *p;
555{
556	struct shmmap_state *shmmap_s;
557	int i;
558
559	shmmap_s = (struct shmmap_state *)p->p_vmspace->vm_shm;
560	for (i = 0; i < shminfo.shmseg; i++, shmmap_s++)
561		if (shmmap_s->shmid != -1)
562			shm_delete_mapping(p, shmmap_s);
563	free((caddr_t)p->p_vmspace->vm_shm, M_SHM);
564	p->p_vmspace->vm_shm = NULL;
565}
566
567void
568shminit()
569{
570	int i;
571	vm_offset_t garbage1, garbage2;
572
573	/* actually this *should* be pageable.  SHM_{LOCK,UNLOCK} */
574	sysvshm_map = kmem_suballoc(kernel_map, &garbage1, &garbage2,
575				    shminfo.shmall * NBPG, TRUE);
576	for (i = 0; i < shminfo.shmmni; i++) {
577		shmsegs[i].shm_perm.mode = SHMSEG_FREE;
578		shmsegs[i].shm_perm.seq = 0;
579	}
580	shm_last_free = 0;
581	shm_nused = 0;
582	shm_committed = 0;
583}
584