sysv_shm.c revision 10358 
1/*	$Id: sysv_shm.c,v 1.6 1995/07/29 11:40:15 bde 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
64/*
65 * System initialization
66 */
67
68extern void shminit();				/* should be static*/
69SYSINIT(sysv_shm, SI_SUB_SYSV_SHM, SI_ORDER_FIRST, shminit, NULL)
70
71int	oshmctl();
72int	shmat(), shmctl(), shmdt(), shmget();
73int	(*shmcalls[])() = { shmat, oshmctl, shmdt, shmget, shmctl };
74
75#define	SHMSEG_FREE     	0x0200
76#define	SHMSEG_REMOVED  	0x0400
77#define	SHMSEG_ALLOCATED	0x0800
78#define	SHMSEG_WANTED		0x1000
79
80vm_map_t sysvshm_map;
81int shm_last_free, shm_nused, shm_committed;
82struct shmid_ds	*shmsegs;
83
84struct shm_handle {
85	vm_offset_t kva;
86};
87
88struct shmmap_state {
89	vm_offset_t va;
90	int shmid;
91};
92
93static void shm_deallocate_segment __P((struct shmid_ds *));
94static int shm_find_segment_by_key __P((key_t));
95static struct shmid_ds *shm_find_segment_by_shmid __P((int));
96static int shm_delete_mapping __P((struct proc *, struct shmmap_state *));
97
98static int
99shm_find_segment_by_key(key)
100	key_t key;
101{
102	int i;
103
104	for (i = 0; i < shminfo.shmmni; i++)
105		if ((shmsegs[i].shm_perm.mode & SHMSEG_ALLOCATED) &&
106		    shmsegs[i].shm_perm.key == key)
107			return i;
108	return -1;
109}
110
111static struct shmid_ds *
112shm_find_segment_by_shmid(shmid)
113	int shmid;
114{
115	int segnum;
116	struct shmid_ds *shmseg;
117
118	segnum = IPCID_TO_IX(shmid);
119	if (segnum < 0 || segnum >= shminfo.shmmni)
120		return NULL;
121	shmseg = &shmsegs[segnum];
122	if ((shmseg->shm_perm.mode & (SHMSEG_ALLOCATED | SHMSEG_REMOVED))
123	    != SHMSEG_ALLOCATED ||
124	    shmseg->shm_perm.seq != IPCID_TO_SEQ(shmid))
125		return NULL;
126	return shmseg;
127}
128
129static void
130shm_deallocate_segment(shmseg)
131	struct shmid_ds *shmseg;
132{
133	struct shm_handle *shm_handle;
134	size_t size;
135
136	shm_handle = shmseg->shm_internal;
137	size = (shmseg->shm_segsz + CLOFSET) & ~CLOFSET;
138	(void) vm_map_remove(sysvshm_map, shm_handle->kva, shm_handle->kva + size);
139	free((caddr_t)shm_handle, M_SHM);
140	shmseg->shm_internal = NULL;
141	shm_committed -= btoc(size);
142	shm_nused--;
143	shmseg->shm_perm.mode = SHMSEG_FREE;
144}
145
146static int
147shm_delete_mapping(p, shmmap_s)
148	struct proc *p;
149	struct shmmap_state *shmmap_s;
150{
151	struct shmid_ds *shmseg;
152	int segnum, result;
153	size_t size;
154
155	segnum = IPCID_TO_IX(shmmap_s->shmid);
156	shmseg = &shmsegs[segnum];
157	size = (shmseg->shm_segsz + CLOFSET) & ~CLOFSET;
158	result = vm_map_remove(&p->p_vmspace->vm_map, shmmap_s->va, shmmap_s->va + size);
159	if (result != KERN_SUCCESS)
160		return EINVAL;
161	shmmap_s->shmid = -1;
162	shmseg->shm_dtime = time.tv_sec;
163	if ((--shmseg->shm_nattch <= 0) &&
164	    (shmseg->shm_perm.mode & SHMSEG_REMOVED)) {
165		shm_deallocate_segment(shmseg);
166		shm_last_free = segnum;
167	}
168	return 0;
169}
170
171struct shmdt_args {
172	void *shmaddr;
173};
174int
175shmdt(p, uap, retval)
176	struct proc *p;
177	struct shmdt_args *uap;
178	int *retval;
179{
180	struct shmmap_state *shmmap_s;
181	int i;
182
183	shmmap_s = (struct shmmap_state *)p->p_vmspace->vm_shm;
184	for (i = 0; i < shminfo.shmseg; i++, shmmap_s++)
185		if (shmmap_s->shmid != -1 &&
186		    shmmap_s->va == (vm_offset_t)uap->shmaddr)
187			break;
188	if (i == shminfo.shmseg)
189		return EINVAL;
190	return shm_delete_mapping(p, shmmap_s);
191}
192
193struct shmat_args {
194	int shmid;
195	void *shmaddr;
196	int shmflg;
197};
198int
199shmat(p, uap, retval)
200	struct proc *p;
201	struct shmat_args *uap;
202	int *retval;
203{
204	int error, i, flags;
205	struct ucred *cred = p->p_ucred;
206	struct shmid_ds *shmseg;
207	struct shmmap_state *shmmap_s = NULL;
208	vm_offset_t attach_va;
209	vm_prot_t prot;
210	vm_size_t size;
211
212	shmmap_s = (struct shmmap_state *)p->p_vmspace->vm_shm;
213	if (shmmap_s == NULL) {
214		size = shminfo.shmseg * sizeof(struct shmmap_state);
215		shmmap_s = malloc(size, M_SHM, M_WAITOK);
216		for (i = 0; i < shminfo.shmseg; i++)
217			shmmap_s[i].shmid = -1;
218		p->p_vmspace->vm_shm = (caddr_t)shmmap_s;
219	}
220	shmseg = shm_find_segment_by_shmid(uap->shmid);
221	if (shmseg == NULL)
222		return EINVAL;
223	error = ipcperm(cred, &shmseg->shm_perm,
224	    (uap->shmflg & SHM_RDONLY) ? IPC_R : IPC_R|IPC_W);
225	if (error)
226		return error;
227	for (i = 0; i < shminfo.shmseg; i++) {
228		if (shmmap_s->shmid == -1)
229			break;
230		shmmap_s++;
231	}
232	if (i >= shminfo.shmseg)
233		return EMFILE;
234	size = (shmseg->shm_segsz + CLOFSET) & ~CLOFSET;
235	prot = VM_PROT_READ;
236	if ((uap->shmflg & SHM_RDONLY) == 0)
237		prot |= VM_PROT_WRITE;
238	flags = MAP_ANON | MAP_SHARED;
239	if (uap->shmaddr) {
240		flags |= MAP_FIXED;
241		if (uap->shmflg & SHM_RND)
242			attach_va = (vm_offset_t)uap->shmaddr & ~(SHMLBA-1);
243		else if (((vm_offset_t)uap->shmaddr & (SHMLBA-1)) == 0)
244			attach_va = (vm_offset_t)uap->shmaddr;
245		else
246			return EINVAL;
247	} else {
248		/* This is just a hint to vm_mmap() about where to put it. */
249		attach_va = round_page(p->p_vmspace->vm_daddr + MAXDSIZ);
250	}
251	error = vm_mmap(&p->p_vmspace->vm_map, &attach_va, size, prot,
252	    VM_PROT_DEFAULT, flags, (caddr_t) uap->shmid, 0);
253	if (error)
254		return error;
255	shmmap_s->va = attach_va;
256	shmmap_s->shmid = uap->shmid;
257	shmseg->shm_lpid = p->p_pid;
258	shmseg->shm_atime = time.tv_sec;
259	shmseg->shm_nattch++;
260	*retval = attach_va;
261	return 0;
262}
263
264struct oshmid_ds {
265	struct	ipc_perm shm_perm;	/* operation perms */
266	int	shm_segsz;		/* size of segment (bytes) */
267	ushort	shm_cpid;		/* pid, creator */
268	ushort	shm_lpid;		/* pid, last operation */
269	short	shm_nattch;		/* no. of current attaches */
270	time_t	shm_atime;		/* last attach time */
271	time_t	shm_dtime;		/* last detach time */
272	time_t	shm_ctime;		/* last change time */
273	void	*shm_handle;		/* internal handle for shm segment */
274};
275
276struct oshmctl_args {
277	int shmid;
278	int cmd;
279	struct oshmid_ds *ubuf;
280};
281
282int
283oshmctl(p, uap, retval)
284	struct proc *p;
285	struct oshmctl_args *uap;
286	int *retval;
287{
288#ifdef COMPAT_43
289	int error;
290	struct ucred *cred = p->p_ucred;
291	struct shmid_ds *shmseg;
292	struct oshmid_ds outbuf;
293
294	shmseg = shm_find_segment_by_shmid(uap->shmid);
295	if (shmseg == NULL)
296		return EINVAL;
297	switch (uap->cmd) {
298	case IPC_STAT:
299		error = ipcperm(cred, &shmseg->shm_perm, IPC_R);
300		if (error)
301			return error;
302		outbuf.shm_perm = shmseg->shm_perm;
303		outbuf.shm_segsz = shmseg->shm_segsz;
304		outbuf.shm_cpid = shmseg->shm_cpid;
305		outbuf.shm_lpid = shmseg->shm_lpid;
306		outbuf.shm_nattch = shmseg->shm_nattch;
307		outbuf.shm_atime = shmseg->shm_atime;
308		outbuf.shm_dtime = shmseg->shm_dtime;
309		outbuf.shm_ctime = shmseg->shm_ctime;
310		outbuf.shm_handle = shmseg->shm_internal;
311		error = copyout((caddr_t)&outbuf, uap->ubuf, sizeof(outbuf));
312		if (error)
313			return error;
314		break;
315	default:
316		return shmctl(p, uap, retval);
317	}
318	return 0;
319#else
320	return EINVAL;
321#endif
322}
323
324struct shmctl_args {
325	int shmid;
326	int cmd;
327	struct shmid_ds *ubuf;
328};
329int
330shmctl(p, uap, retval)
331	struct proc *p;
332	struct shmctl_args *uap;
333	int *retval;
334{
335	int error;
336	struct ucred *cred = p->p_ucred;
337	struct shmid_ds inbuf;
338	struct shmid_ds *shmseg;
339
340	shmseg = shm_find_segment_by_shmid(uap->shmid);
341	if (shmseg == NULL)
342		return EINVAL;
343	switch (uap->cmd) {
344	case IPC_STAT:
345		error = ipcperm(cred, &shmseg->shm_perm, IPC_R);
346		if (error)
347			return error;
348		error = copyout((caddr_t)shmseg, uap->ubuf, sizeof(inbuf));
349		if (error)
350			return error;
351		break;
352	case IPC_SET:
353		error = ipcperm(cred, &shmseg->shm_perm, IPC_M);
354		if (error)
355			return error;
356		error = copyin(uap->ubuf, (caddr_t)&inbuf, sizeof(inbuf));
357		if (error)
358			return error;
359		shmseg->shm_perm.uid = inbuf.shm_perm.uid;
360		shmseg->shm_perm.gid = inbuf.shm_perm.gid;
361		shmseg->shm_perm.mode =
362		    (shmseg->shm_perm.mode & ~ACCESSPERMS) |
363		    (inbuf.shm_perm.mode & ACCESSPERMS);
364		shmseg->shm_ctime = time.tv_sec;
365		break;
366	case IPC_RMID:
367		error = ipcperm(cred, &shmseg->shm_perm, IPC_M);
368		if (error)
369			return error;
370		shmseg->shm_perm.key = IPC_PRIVATE;
371		shmseg->shm_perm.mode |= SHMSEG_REMOVED;
372		if (shmseg->shm_nattch <= 0) {
373			shm_deallocate_segment(shmseg);
374			shm_last_free = IPCID_TO_IX(uap->shmid);
375		}
376		break;
377#if 0
378	case SHM_LOCK:
379	case SHM_UNLOCK:
380#endif
381	default:
382		return EINVAL;
383	}
384	return 0;
385}
386
387struct shmget_args {
388	key_t key;
389	size_t size;
390	int shmflg;
391};
392static int
393shmget_existing(p, uap, mode, segnum, retval)
394	struct proc *p;
395	struct shmget_args *uap;
396	int mode;
397	int segnum;
398	int *retval;
399{
400	struct shmid_ds *shmseg;
401	struct ucred *cred = p->p_ucred;
402	int error;
403
404	shmseg = &shmsegs[segnum];
405	if (shmseg->shm_perm.mode & SHMSEG_REMOVED) {
406		/*
407		 * This segment is in the process of being allocated.  Wait
408		 * until it's done, and look the key up again (in case the
409		 * allocation failed or it was freed).
410		 */
411		shmseg->shm_perm.mode |= SHMSEG_WANTED;
412		error = tsleep((caddr_t)shmseg, PLOCK | PCATCH, "shmget", 0);
413		if (error)
414			return error;
415		return EAGAIN;
416	}
417	error = ipcperm(cred, &shmseg->shm_perm, mode);
418	if (error)
419		return error;
420	if (uap->size && uap->size > shmseg->shm_segsz)
421		return EINVAL;
422	if (uap->shmflg & (IPC_CREAT | IPC_EXCL) == (IPC_CREAT | IPC_EXCL))
423		return EEXIST;
424	*retval = IXSEQ_TO_IPCID(segnum, shmseg->shm_perm);
425	return 0;
426}
427
428static int
429shmget_allocate_segment(p, uap, mode, retval)
430	struct proc *p;
431	struct shmget_args *uap;
432	int mode;
433	int *retval;
434{
435	int i, segnum, result, shmid, size;
436	struct ucred *cred = p->p_ucred;
437	struct shmid_ds *shmseg;
438	struct shm_handle *shm_handle;
439
440	if (uap->size < shminfo.shmmin || uap->size > shminfo.shmmax)
441		return EINVAL;
442	if (shm_nused >= shminfo.shmmni) /* any shmids left? */
443		return ENOSPC;
444	size = (uap->size + CLOFSET) & ~CLOFSET;
445	if (shm_committed + btoc(size) > shminfo.shmall)
446		return ENOMEM;
447	if (shm_last_free < 0) {
448		for (i = 0; i < shminfo.shmmni; i++)
449			if (shmsegs[i].shm_perm.mode & SHMSEG_FREE)
450				break;
451		if (i == shminfo.shmmni)
452			panic("shmseg free count inconsistent");
453		segnum = i;
454	} else  {
455		segnum = shm_last_free;
456		shm_last_free = -1;
457	}
458	shmseg = &shmsegs[segnum];
459	/*
460	 * In case we sleep in malloc(), mark the segment present but deleted
461	 * so that noone else tries to create the same key.
462	 */
463	shmseg->shm_perm.mode = SHMSEG_ALLOCATED | SHMSEG_REMOVED;
464	shmseg->shm_perm.key = uap->key;
465	shmseg->shm_perm.seq = (shmseg->shm_perm.seq + 1) & 0x7fff;
466	shm_handle = (struct shm_handle *)
467	    malloc(sizeof(struct shm_handle), M_SHM, M_WAITOK);
468	shmid = IXSEQ_TO_IPCID(segnum, shmseg->shm_perm);
469	result = vm_mmap(sysvshm_map, &shm_handle->kva, size, VM_PROT_ALL,
470	    VM_PROT_DEFAULT, MAP_ANON, (caddr_t) shmid, 0);
471	if (result != KERN_SUCCESS) {
472		shmseg->shm_perm.mode = SHMSEG_FREE;
473		shm_last_free = segnum;
474		free((caddr_t)shm_handle, M_SHM);
475		/* Just in case. */
476		wakeup((caddr_t)shmseg);
477		return ENOMEM;
478	}
479	shmseg->shm_internal = shm_handle;
480	shmseg->shm_perm.cuid = shmseg->shm_perm.uid = cred->cr_uid;
481	shmseg->shm_perm.cgid = shmseg->shm_perm.gid = cred->cr_gid;
482	shmseg->shm_perm.mode = (shmseg->shm_perm.mode & SHMSEG_WANTED) |
483	    (mode & ACCESSPERMS) | SHMSEG_ALLOCATED;
484	shmseg->shm_segsz = uap->size;
485	shmseg->shm_cpid = p->p_pid;
486	shmseg->shm_lpid = shmseg->shm_nattch = 0;
487	shmseg->shm_atime = shmseg->shm_dtime = 0;
488	shmseg->shm_ctime = time.tv_sec;
489	shm_committed += btoc(size);
490	shm_nused++;
491	if (shmseg->shm_perm.mode & SHMSEG_WANTED) {
492		/*
493		 * Somebody else wanted this key while we were asleep.  Wake
494		 * them up now.
495		 */
496		shmseg->shm_perm.mode &= ~SHMSEG_WANTED;
497		wakeup((caddr_t)shmseg);
498	}
499	*retval = shmid;
500	return 0;
501}
502
503int
504shmget(p, uap, retval)
505	struct proc *p;
506	struct shmget_args *uap;
507	int *retval;
508{
509	int segnum, mode, error;
510
511	mode = uap->shmflg & ACCESSPERMS;
512	if (uap->key != IPC_PRIVATE) {
513	again:
514		segnum = shm_find_segment_by_key(uap->key);
515		if (segnum >= 0) {
516			error = shmget_existing(p, uap, mode, segnum, retval);
517			if (error == EAGAIN)
518				goto again;
519			return error;
520		}
521		if ((uap->shmflg & IPC_CREAT) == 0)
522			return ENOENT;
523	}
524	return shmget_allocate_segment(p, uap, mode, retval);
525}
526
527struct shmsys_args {
528	u_int	which;
529};
530int
531shmsys(p, uap, retval)
532	struct proc *p;
533	struct shmsys_args *uap;
534	int *retval;
535{
536
537	if (uap->which >= sizeof(shmcalls)/sizeof(shmcalls[0]))
538		return EINVAL;
539	return ((*shmcalls[uap->which])(p, &uap[1], retval));
540}
541
542void
543shmfork(p1, p2, isvfork)
544	struct proc *p1, *p2;
545	int isvfork;
546{
547	struct shmmap_state *shmmap_s;
548	size_t size;
549	int i;
550
551	size = shminfo.shmseg * sizeof(struct shmmap_state);
552	shmmap_s = malloc(size, M_SHM, M_WAITOK);
553	bcopy((caddr_t)p1->p_vmspace->vm_shm, (caddr_t)shmmap_s, size);
554	p2->p_vmspace->vm_shm = (caddr_t)shmmap_s;
555	for (i = 0; i < shminfo.shmseg; i++, shmmap_s++)
556		if (shmmap_s->shmid != -1)
557			shmsegs[IPCID_TO_IX(shmmap_s->shmid)].shm_nattch++;
558}
559
560void
561shmexit(p)
562	struct proc *p;
563{
564	struct shmmap_state *shmmap_s;
565	int i;
566
567	shmmap_s = (struct shmmap_state *)p->p_vmspace->vm_shm;
568	for (i = 0; i < shminfo.shmseg; i++, shmmap_s++)
569		if (shmmap_s->shmid != -1)
570			shm_delete_mapping(p, shmmap_s);
571	free((caddr_t)p->p_vmspace->vm_shm, M_SHM);
572	p->p_vmspace->vm_shm = NULL;
573}
574
575void
576shminit()
577{
578	int i;
579	vm_offset_t garbage1, garbage2;
580
581	/* actually this *should* be pageable.  SHM_{LOCK,UNLOCK} */
582	sysvshm_map = kmem_suballoc(kernel_map, &garbage1, &garbage2,
583				    shminfo.shmall * NBPG, TRUE);
584	for (i = 0; i < shminfo.shmmni; i++) {
585		shmsegs[i].shm_perm.mode = SHMSEG_FREE;
586		shmsegs[i].shm_perm.seq = 0;
587	}
588	shm_last_free = 0;
589	shm_nused = 0;
590	shm_committed = 0;
591}
592