Home | History | Annotate | Line # | Download | only in kern
sysv_shm.c revision 1.40
      1 /*	$NetBSD: sysv_shm.c,v 1.40 1997/10/09 08:35:13 drochner Exp $	*/
      2 
      3 /*
      4  * Copyright (c) 1994 Adam Glass and Charles Hannum.  All rights reserved.
      5  *
      6  * Redistribution and use in source and binary forms, with or without
      7  * modification, are permitted provided that the following conditions
      8  * are met:
      9  * 1. Redistributions of source code must retain the above copyright
     10  *    notice, this list of conditions and the following disclaimer.
     11  * 2. Redistributions in binary form must reproduce the above copyright
     12  *    notice, this list of conditions and the following disclaimer in the
     13  *    documentation and/or other materials provided with the distribution.
     14  * 3. All advertising materials mentioning features or use of this software
     15  *    must display the following acknowledgement:
     16  *	This product includes software developed by Adam Glass and Charles
     17  *	Hannum.
     18  * 4. The names of the authors may not be used to endorse or promote products
     19  *    derived from this software without specific prior written permission.
     20  *
     21  * THIS SOFTWARE IS PROVIDED BY THE AUTHORS ``AS IS'' AND ANY EXPRESS OR
     22  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
     23  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
     24  * IN NO EVENT SHALL THE AUTHORS BE LIABLE FOR ANY DIRECT, INDIRECT,
     25  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
     26  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
     27  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
     28  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
     29  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
     30  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
     31  */
     32 
     33 #include <sys/types.h>
     34 #include <sys/param.h>
     35 #include <sys/kernel.h>
     36 #include <sys/shm.h>
     37 #include <sys/proc.h>
     38 #include <sys/uio.h>
     39 #include <sys/time.h>
     40 #include <sys/malloc.h>
     41 #include <sys/mman.h>
     42 #include <sys/systm.h>
     43 #include <sys/stat.h>
     44 
     45 #include <sys/mount.h>
     46 #include <sys/syscallargs.h>
     47 
     48 #include <vm/vm.h>
     49 #include <vm/vm_map.h>
     50 #include <vm/vm_map.h>
     51 #include <vm/vm_kern.h>
     52 
     53 struct shmid_ds *shm_find_segment_by_shmid __P((int));
     54 void shmexit __P((struct proc *));
     55 
     56 /*
     57  * Provides the following externally accessible functions:
     58  *
     59  * shminit(void);		           initialization
     60  * shmexit(struct proc *)                  cleanup
     61  * shmfork(struct proc *, struct proc *)   fork handling
     62  * shmsys(arg1, arg2, arg3, arg4);         shm{at,ctl,dt,get}(arg2, arg3, arg4)
     63  *
     64  * Structures:
     65  * shmsegs (an array of 'struct shmid_ds')
     66  * per proc array of 'struct shmmap_state'
     67  */
     68 
     69 #define	SHMSEG_FREE     	0x0200
     70 #define	SHMSEG_REMOVED  	0x0400
     71 #define	SHMSEG_ALLOCATED	0x0800
     72 #define	SHMSEG_WANTED		0x1000
     73 
     74 int shm_last_free, shm_nused, shm_committed;
     75 
     76 struct shm_handle {
     77 	vm_object_t shm_object;
     78 };
     79 
     80 struct shmmap_state {
     81 	vm_offset_t va;
     82 	int shmid;
     83 };
     84 
     85 static int shm_find_segment_by_key __P((key_t));
     86 static void shm_deallocate_segment __P((struct shmid_ds *));
     87 static int shm_delete_mapping __P((struct proc *, struct shmmap_state *));
     88 static int shmget_existing __P((struct proc *, struct sys_shmget_args *,
     89 				int, int, register_t *));
     90 static int shmget_allocate_segment __P((struct proc *, struct sys_shmget_args *,
     91 					int, register_t *));
     92 
     93 static int
     94 shm_find_segment_by_key(key)
     95 	key_t key;
     96 {
     97 	int i;
     98 
     99 	for (i = 0; i < shminfo.shmmni; i++)
    100 		if ((shmsegs[i].shm_perm.mode & SHMSEG_ALLOCATED) &&
    101 		    shmsegs[i].shm_perm.key == key)
    102 			return i;
    103 	return -1;
    104 }
    105 
    106 struct shmid_ds *
    107 shm_find_segment_by_shmid(shmid)
    108 	int shmid;
    109 {
    110 	int segnum;
    111 	struct shmid_ds *shmseg;
    112 
    113 	segnum = IPCID_TO_IX(shmid);
    114 	if (segnum < 0 || segnum >= shminfo.shmmni)
    115 		return NULL;
    116 	shmseg = &shmsegs[segnum];
    117 	if ((shmseg->shm_perm.mode & (SHMSEG_ALLOCATED | SHMSEG_REMOVED))
    118 	    != SHMSEG_ALLOCATED ||
    119 	    shmseg->shm_perm.seq != IPCID_TO_SEQ(shmid))
    120 		return NULL;
    121 	return shmseg;
    122 }
    123 
    124 static void
    125 shm_deallocate_segment(shmseg)
    126 	struct shmid_ds *shmseg;
    127 {
    128 	struct shm_handle *shm_handle;
    129 	size_t size;
    130 
    131 	shm_handle = shmseg->shm_internal;
    132 	size = (shmseg->shm_segsz + CLOFSET) & ~CLOFSET;
    133 	vm_object_deallocate(shm_handle->shm_object);
    134 	free((caddr_t)shm_handle, M_SHM);
    135 	shmseg->shm_internal = NULL;
    136 	shm_committed -= btoc(size);
    137 	shmseg->shm_perm.mode = SHMSEG_FREE;
    138 	shm_nused--;
    139 }
    140 
    141 static int
    142 shm_delete_mapping(p, shmmap_s)
    143 	struct proc *p;
    144 	struct shmmap_state *shmmap_s;
    145 {
    146 	struct shmid_ds *shmseg;
    147 	int segnum, result;
    148 	size_t size;
    149 
    150 	segnum = IPCID_TO_IX(shmmap_s->shmid);
    151 	shmseg = &shmsegs[segnum];
    152 	size = (shmseg->shm_segsz + CLOFSET) & ~CLOFSET;
    153 	result = vm_map_remove(&p->p_vmspace->vm_map,
    154 			       shmmap_s->va, shmmap_s->va + size);
    155 	if (result != KERN_SUCCESS)
    156 		return EINVAL;
    157 	shmmap_s->shmid = -1;
    158 	shmseg->shm_dtime = time.tv_sec;
    159 	if ((--shmseg->shm_nattch <= 0) &&
    160 	    (shmseg->shm_perm.mode & SHMSEG_REMOVED)) {
    161 		shm_deallocate_segment(shmseg);
    162 		shm_last_free = segnum;
    163 	}
    164 	return 0;
    165 }
    166 
    167 int
    168 sys_shmdt(p, v, retval)
    169 	struct proc *p;
    170 	void *v;
    171 	register_t *retval;
    172 {
    173 	struct sys_shmdt_args /* {
    174 		syscallarg(void *) shmaddr;
    175 	} */ *uap = v;
    176 	struct shmmap_state *shmmap_s;
    177 	int i;
    178 
    179 	shmmap_s = (struct shmmap_state *)p->p_vmspace->vm_shm;
    180 	if (shmmap_s == NULL)
    181 		return EINVAL;
    182 
    183 	for (i = 0; i < shminfo.shmseg; i++, shmmap_s++)
    184 		if (shmmap_s->shmid != -1 &&
    185 		    shmmap_s->va == (vm_offset_t)SCARG(uap, shmaddr))
    186 			break;
    187 	if (i == shminfo.shmseg)
    188 		return EINVAL;
    189 	return shm_delete_mapping(p, shmmap_s);
    190 }
    191 
    192 int
    193 sys_shmat(p, v, retval)
    194 	struct proc *p;
    195 	void *v;
    196 	register_t *retval;
    197 {
    198 	struct sys_shmat_args /* {
    199 		syscallarg(int) shmid;
    200 		syscallarg(void *) shmaddr;
    201 		syscallarg(int) shmflg;
    202 	} */ *uap = v;
    203 	int error, i, flags;
    204 	struct ucred *cred = p->p_ucred;
    205 	struct shmid_ds *shmseg;
    206 	struct shmmap_state *shmmap_s = NULL;
    207 	struct shm_handle *shm_handle;
    208 	vm_offset_t attach_va;
    209 	vm_prot_t prot;
    210 	vm_size_t size;
    211 	int rv;
    212 
    213 	shmmap_s = (struct shmmap_state *)p->p_vmspace->vm_shm;
    214 	if (shmmap_s == NULL) {
    215 		size = shminfo.shmseg * sizeof(struct shmmap_state);
    216 		shmmap_s = malloc(size, M_SHM, M_WAITOK);
    217 		for (i = 0; i < shminfo.shmseg; i++)
    218 			shmmap_s[i].shmid = -1;
    219 		p->p_vmspace->vm_shm = (caddr_t)shmmap_s;
    220 	}
    221 	shmseg = shm_find_segment_by_shmid(SCARG(uap, shmid));
    222 	if (shmseg == NULL)
    223 		return EINVAL;
    224 	error = ipcperm(cred, &shmseg->shm_perm,
    225 		    (SCARG(uap, shmflg) & SHM_RDONLY) ? IPC_R : IPC_R|IPC_W);
    226 	if (error)
    227 		return error;
    228 	for (i = 0; i < shminfo.shmseg; i++) {
    229 		if (shmmap_s->shmid == -1)
    230 			break;
    231 		shmmap_s++;
    232 	}
    233 	if (i >= shminfo.shmseg)
    234 		return EMFILE;
    235 	size = (shmseg->shm_segsz + CLOFSET) & ~CLOFSET;
    236 	prot = VM_PROT_READ;
    237 	if ((SCARG(uap, shmflg) & SHM_RDONLY) == 0)
    238 		prot |= VM_PROT_WRITE;
    239 	flags = MAP_ANON | MAP_SHARED;
    240 	if (SCARG(uap, shmaddr)) {
    241 		flags |= MAP_FIXED;
    242 		if (SCARG(uap, shmflg) & SHM_RND)
    243 			attach_va =
    244 			    (vm_offset_t)SCARG(uap, shmaddr) & ~(SHMLBA-1);
    245 		else if (((vm_offset_t)SCARG(uap, shmaddr) & (SHMLBA-1)) == 0)
    246 			attach_va = (vm_offset_t)SCARG(uap, shmaddr);
    247 		else
    248 			return EINVAL;
    249 	} else {
    250 		/* This is just a hint to vm_mmap() about where to put it. */
    251 		attach_va =
    252 		    round_page(p->p_vmspace->vm_taddr + MAXTSIZ + MAXDSIZ);
    253 	}
    254 	shm_handle = shmseg->shm_internal;
    255 	vm_object_reference(shm_handle->shm_object);
    256 	rv = vm_map_find(&p->p_vmspace->vm_map, shm_handle->shm_object,
    257 		0, &attach_va, size, (flags & MAP_FIXED)?0:1);
    258 	if (rv != KERN_SUCCESS) {
    259 		return ENOMEM;
    260 	}
    261 	vm_map_protect(&p->p_vmspace->vm_map, attach_va, attach_va + size,
    262 		       prot, 0);
    263 	vm_map_inherit(&p->p_vmspace->vm_map,
    264 		attach_va, attach_va + size, VM_INHERIT_SHARE);
    265 
    266 	shmmap_s->va = attach_va;
    267 	shmmap_s->shmid = SCARG(uap, shmid);
    268 	shmseg->shm_lpid = p->p_pid;
    269 	shmseg->shm_atime = time.tv_sec;
    270 	shmseg->shm_nattch++;
    271 	*retval = attach_va;
    272 	return 0;
    273 }
    274 
    275 int
    276 sys_shmctl(p, v, retval)
    277 	struct proc *p;
    278 	void *v;
    279 	register_t *retval;
    280 {
    281 	struct sys_shmctl_args /* {
    282 		syscallarg(int) shmid;
    283 		syscallarg(int) cmd;
    284 		syscallarg(struct shmid_ds *) buf;
    285 	} */ *uap = v;
    286 	int error;
    287 	struct ucred *cred = p->p_ucred;
    288 	struct shmid_ds inbuf;
    289 	struct shmid_ds *shmseg;
    290 
    291 	shmseg = shm_find_segment_by_shmid(SCARG(uap, shmid));
    292 	if (shmseg == NULL)
    293 		return EINVAL;
    294 	switch (SCARG(uap, cmd)) {
    295 	case IPC_STAT:
    296 		if ((error = ipcperm(cred, &shmseg->shm_perm, IPC_R)) != 0)
    297 			return error;
    298 		error = copyout((caddr_t)shmseg, SCARG(uap, buf),
    299 				sizeof(inbuf));
    300 		if (error)
    301 			return error;
    302 		break;
    303 	case IPC_SET:
    304 		if ((error = ipcperm(cred, &shmseg->shm_perm, IPC_M)) != 0)
    305 			return error;
    306 		error = copyin(SCARG(uap, buf), (caddr_t)&inbuf,
    307 			       sizeof(inbuf));
    308 		if (error)
    309 			return error;
    310 		shmseg->shm_perm.uid = inbuf.shm_perm.uid;
    311 		shmseg->shm_perm.gid = inbuf.shm_perm.gid;
    312 		shmseg->shm_perm.mode =
    313 		    (shmseg->shm_perm.mode & ~ACCESSPERMS) |
    314 		    (inbuf.shm_perm.mode & ACCESSPERMS);
    315 		shmseg->shm_ctime = time.tv_sec;
    316 		break;
    317 	case IPC_RMID:
    318 		if ((error = ipcperm(cred, &shmseg->shm_perm, IPC_M)) != 0)
    319 			return error;
    320 		shmseg->shm_perm.key = IPC_PRIVATE;
    321 		shmseg->shm_perm.mode |= SHMSEG_REMOVED;
    322 		if (shmseg->shm_nattch <= 0) {
    323 			shm_deallocate_segment(shmseg);
    324 			shm_last_free = IPCID_TO_IX(SCARG(uap, shmid));
    325 		}
    326 		break;
    327 	case SHM_LOCK:
    328 	case SHM_UNLOCK:
    329 	default:
    330 		return EINVAL;
    331 	}
    332 	return 0;
    333 }
    334 
    335 static int
    336 shmget_existing(p, uap, mode, segnum, retval)
    337 	struct proc *p;
    338 	struct sys_shmget_args /* {
    339 		syscallarg(key_t) key;
    340 		syscallarg(int) size;
    341 		syscallarg(int) shmflg;
    342 	} */ *uap;
    343 	int mode;
    344 	int segnum;
    345 	register_t *retval;
    346 {
    347 	struct shmid_ds *shmseg;
    348 	struct ucred *cred = p->p_ucred;
    349 	int error;
    350 
    351 	shmseg = &shmsegs[segnum];
    352 	if (shmseg->shm_perm.mode & SHMSEG_REMOVED) {
    353 		/*
    354 		 * This segment is in the process of being allocated.  Wait
    355 		 * until it's done, and look the key up again (in case the
    356 		 * allocation failed or it was freed).
    357 		 */
    358 		shmseg->shm_perm.mode |= SHMSEG_WANTED;
    359 		error = tsleep((caddr_t)shmseg, PLOCK | PCATCH, "shmget", 0);
    360 		if (error)
    361 			return error;
    362 		return EAGAIN;
    363 	}
    364 	if ((error = ipcperm(cred, &shmseg->shm_perm, mode)) != 0)
    365 		return error;
    366 	if (SCARG(uap, size) && SCARG(uap, size) > shmseg->shm_segsz)
    367 		return EINVAL;
    368 	if ((SCARG(uap, shmflg) & (IPC_CREAT | IPC_EXCL)) ==
    369 	    (IPC_CREAT | IPC_EXCL))
    370 		return EEXIST;
    371 	*retval = IXSEQ_TO_IPCID(segnum, shmseg->shm_perm);
    372 	return 0;
    373 }
    374 
    375 static int
    376 shmget_allocate_segment(p, uap, mode, retval)
    377 	struct proc *p;
    378 	struct sys_shmget_args /* {
    379 		syscallarg(key_t) key;
    380 		syscallarg(int) size;
    381 		syscallarg(int) shmflg;
    382 	} */ *uap;
    383 	int mode;
    384 	register_t *retval;
    385 {
    386 	int i, segnum, shmid, size;
    387 	struct ucred *cred = p->p_ucred;
    388 	struct shmid_ds *shmseg;
    389 	struct shm_handle *shm_handle;
    390 	vm_pager_t pager;
    391 	int error = 0;
    392 
    393 	if (SCARG(uap, size) < shminfo.shmmin ||
    394 	    SCARG(uap, size) > shminfo.shmmax)
    395 		return EINVAL;
    396 	if (shm_nused >= shminfo.shmmni) /* any shmids left? */
    397 		return ENOSPC;
    398 	size = (SCARG(uap, size) + CLOFSET) & ~CLOFSET;
    399 	if (shm_committed + btoc(size) > shminfo.shmall)
    400 		return ENOMEM;
    401 	if (shm_last_free < 0) {
    402 		for (i = 0; i < shminfo.shmmni; i++)
    403 			if (shmsegs[i].shm_perm.mode & SHMSEG_FREE)
    404 				break;
    405 		if (i == shminfo.shmmni)
    406 			panic("shmseg free count inconsistent");
    407 		segnum = i;
    408 	} else  {
    409 		segnum = shm_last_free;
    410 		shm_last_free = -1;
    411 	}
    412 	shmseg = &shmsegs[segnum];
    413 	/*
    414 	 * In case we sleep in malloc(), mark the segment present but deleted
    415 	 * so that noone else tries to create the same key.
    416 	 */
    417 	shmseg->shm_perm.mode = SHMSEG_ALLOCATED | SHMSEG_REMOVED;
    418 	shmseg->shm_perm.key = SCARG(uap, key);
    419 	shmseg->shm_perm.seq = (shmseg->shm_perm.seq + 1) & 0x7fff;
    420 	shm_handle = (struct shm_handle *)
    421 	    malloc(sizeof(struct shm_handle), M_SHM, M_WAITOK);
    422 	shmid = IXSEQ_TO_IPCID(segnum, shmseg->shm_perm);
    423 
    424 	shm_handle->shm_object = vm_object_allocate(size);
    425 	if (shm_handle->shm_object == NULL) {
    426 		/* XXX cannot happen */
    427 		error = ENOMEM;
    428 		goto out;
    429 	}
    430 	/*
    431 	 * We make sure that we have allocated a pager before we need
    432 	 * to.
    433 	 */
    434 	pager = vm_pager_allocate(PG_DFLT, 0, size, VM_PROT_DEFAULT, 0);
    435 	if (pager == NULL) {
    436 		error = ENOMEM;
    437 		goto out;
    438 	}
    439 	vm_object_setpager(shm_handle->shm_object, pager, 0, 0);
    440 	shmseg->shm_internal = shm_handle;
    441 	shmseg->shm_perm.cuid = shmseg->shm_perm.uid = cred->cr_uid;
    442 	shmseg->shm_perm.cgid = shmseg->shm_perm.gid = cred->cr_gid;
    443 	shmseg->shm_perm.mode = (shmseg->shm_perm.mode & SHMSEG_WANTED) |
    444 	    (mode & ACCESSPERMS) | SHMSEG_ALLOCATED;
    445 	shmseg->shm_segsz = SCARG(uap, size);
    446 	shmseg->shm_cpid = p->p_pid;
    447 	shmseg->shm_lpid = shmseg->shm_nattch = 0;
    448 	shmseg->shm_atime = shmseg->shm_dtime = 0;
    449 	shmseg->shm_ctime = time.tv_sec;
    450 	shm_committed += btoc(size);
    451 	shm_nused++;
    452 
    453 out:
    454 	if (error) {
    455 		if (shm_handle->shm_object != NULL)
    456 			vm_object_deallocate(shm_handle->shm_object);
    457 		free(shm_handle, M_SHM);
    458 		shmseg->shm_perm.mode = (shmseg->shm_perm.mode & SHMSEG_WANTED)
    459 		    | SHMSEG_FREE;
    460 	} else
    461 		*retval = shmid;
    462 	if (shmseg->shm_perm.mode & SHMSEG_WANTED) {
    463 		/*
    464 		 * Somebody else wanted this key while we were asleep.  Wake
    465 		 * them up now.
    466 		 */
    467 		shmseg->shm_perm.mode &= ~SHMSEG_WANTED;
    468 		wakeup((caddr_t)shmseg);
    469 	}
    470 	return error;
    471 }
    472 
    473 int
    474 sys_shmget(p, v, retval)
    475 	struct proc *p;
    476 	void *v;
    477 	register_t *retval;
    478 {
    479 	struct sys_shmget_args /* {
    480 		syscallarg(key_t) key;
    481 		syscallarg(int) size;
    482 		syscallarg(int) shmflg;
    483 	} */ *uap = v;
    484 	int segnum, mode, error;
    485 
    486 	mode = SCARG(uap, shmflg) & ACCESSPERMS;
    487 	if (SCARG(uap, key) != IPC_PRIVATE) {
    488 	again:
    489 		segnum = shm_find_segment_by_key(SCARG(uap, key));
    490 		if (segnum >= 0) {
    491 			error = shmget_existing(p, uap, mode, segnum, retval);
    492 			if (error == EAGAIN)
    493 				goto again;
    494 			return error;
    495 		}
    496 		if ((SCARG(uap, shmflg) & IPC_CREAT) == 0)
    497 			return ENOENT;
    498 	}
    499 	return shmget_allocate_segment(p, uap, mode, retval);
    500 }
    501 
    502 void
    503 shmfork(p1, p2)
    504 	struct proc *p1, *p2;
    505 {
    506 	struct shmmap_state *shmmap_s;
    507 	size_t size;
    508 	int i;
    509 
    510 	if (p1->p_vmspace->vm_shm == NULL) {
    511 		p2->p_vmspace->vm_shm = NULL;
    512 		return;
    513 	}
    514 
    515 	size = shminfo.shmseg * sizeof(struct shmmap_state);
    516 	shmmap_s = malloc(size, M_SHM, M_WAITOK);
    517 	bcopy((caddr_t)p1->p_vmspace->vm_shm, (caddr_t)shmmap_s, size);
    518 	p2->p_vmspace->vm_shm = (caddr_t)shmmap_s;
    519 	for (i = 0; i < shminfo.shmseg; i++, shmmap_s++)
    520 		if (shmmap_s->shmid != -1)
    521 			shmsegs[IPCID_TO_IX(shmmap_s->shmid)].shm_nattch++;
    522 }
    523 
    524 void
    525 shmexit(p)
    526 	struct proc *p;
    527 {
    528 	struct shmmap_state *shmmap_s;
    529 	int i;
    530 
    531 	shmmap_s = (struct shmmap_state *)p->p_vmspace->vm_shm;
    532 	if (shmmap_s == NULL)
    533 		return;
    534 	for (i = 0; i < shminfo.shmseg; i++, shmmap_s++)
    535 		if (shmmap_s->shmid != -1)
    536 			shm_delete_mapping(p, shmmap_s);
    537 	free((caddr_t)p->p_vmspace->vm_shm, M_SHM);
    538 	p->p_vmspace->vm_shm = NULL;
    539 }
    540 
    541 void
    542 shminit()
    543 {
    544 	int i;
    545 
    546 	shminfo.shmmax *= NBPG;
    547 
    548 	for (i = 0; i < shminfo.shmmni; i++) {
    549 		shmsegs[i].shm_perm.mode = SHMSEG_FREE;
    550 		shmsegs[i].shm_perm.seq = 0;
    551 	}
    552 	shm_last_free = 0;
    553 	shm_nused = 0;
    554 	shm_committed = 0;
    555 }
    556