Home | History | Annotate | Line # | Download | only in kern
sysv_shm.c revision 1.77
      1 /*	$NetBSD: sysv_shm.c,v 1.77 2004/04/25 16:42:41 simonb Exp $	*/
      2 
      3 /*-
      4  * Copyright (c) 1999 The NetBSD Foundation, Inc.
      5  * All rights reserved.
      6  *
      7  * This code is derived from software contributed to The NetBSD Foundation
      8  * by Jason R. Thorpe of the Numerical Aerospace Simulation Facility,
      9  * NASA Ames Research Center.
     10  *
     11  * Redistribution and use in source and binary forms, with or without
     12  * modification, are permitted provided that the following conditions
     13  * are met:
     14  * 1. Redistributions of source code must retain the above copyright
     15  *    notice, this list of conditions and the following disclaimer.
     16  * 2. Redistributions in binary form must reproduce the above copyright
     17  *    notice, this list of conditions and the following disclaimer in the
     18  *    documentation and/or other materials provided with the distribution.
     19  * 3. All advertising materials mentioning features or use of this software
     20  *    must display the following acknowledgement:
     21  *	This product includes software developed by the NetBSD
     22  *	Foundation, Inc. and its contributors.
     23  * 4. Neither the name of The NetBSD Foundation nor the names of its
     24  *    contributors may be used to endorse or promote products derived
     25  *    from this software without specific prior written permission.
     26  *
     27  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
     28  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
     29  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
     30  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
     31  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
     32  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
     33  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
     34  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
     35  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
     36  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
     37  * POSSIBILITY OF SUCH DAMAGE.
     38  */
     39 
     40 /*
     41  * Copyright (c) 1994 Adam Glass and Charles M. Hannum.  All rights reserved.
     42  *
     43  * Redistribution and use in source and binary forms, with or without
     44  * modification, are permitted provided that the following conditions
     45  * are met:
     46  * 1. Redistributions of source code must retain the above copyright
     47  *    notice, this list of conditions and the following disclaimer.
     48  * 2. Redistributions in binary form must reproduce the above copyright
     49  *    notice, this list of conditions and the following disclaimer in the
     50  *    documentation and/or other materials provided with the distribution.
     51  * 3. All advertising materials mentioning features or use of this software
     52  *    must display the following acknowledgement:
     53  *	This product includes software developed by Adam Glass and Charles M.
     54  *	Hannum.
     55  * 4. The names of the authors may not be used to endorse or promote products
     56  *    derived from this software without specific prior written permission.
     57  *
     58  * THIS SOFTWARE IS PROVIDED BY THE AUTHORS ``AS IS'' AND ANY EXPRESS OR
     59  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
     60  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
     61  * IN NO EVENT SHALL THE AUTHORS BE LIABLE FOR ANY DIRECT, INDIRECT,
     62  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
     63  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
     64  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
     65  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
     66  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
     67  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
     68  */
     69 
     70 #include <sys/cdefs.h>
     71 __KERNEL_RCSID(0, "$NetBSD: sysv_shm.c,v 1.77 2004/04/25 16:42:41 simonb Exp $");
     72 
     73 #define SYSVSHM
     74 
     75 #include <sys/param.h>
     76 #include <sys/kernel.h>
     77 #include <sys/shm.h>
     78 #include <sys/malloc.h>
     79 #include <sys/mman.h>
     80 #include <sys/stat.h>
     81 #include <sys/sysctl.h>
     82 #include <sys/mount.h>		/* XXX for <sys/syscallargs.h> */
     83 #include <sys/sa.h>
     84 #include <sys/syscallargs.h>
     85 #include <sys/queue.h>
     86 #include <sys/pool.h>
     87 
     88 #include <uvm/uvm_extern.h>
     89 #include <uvm/uvm_object.h>
     90 
     91 struct shmid_ds *shm_find_segment_by_shmid(int, int);
     92 
     93 static MALLOC_DEFINE(M_SHM, "shm", "SVID compatible shared memory segments");
     94 
     95 /*
     96  * Provides the following externally accessible functions:
     97  *
     98  * shminit(void);		                 initialization
     99  * shmexit(struct vmspace *)                     cleanup
    100  * shmfork(struct vmspace *, struct vmspace *)   fork handling
    101  *
    102  * Structures:
    103  * shmsegs (an array of 'struct shmid_ds')
    104  * per proc array of 'struct shmmap_state'
    105  */
    106 
    107 #define	SHMSEG_FREE     	0x0200
    108 #define	SHMSEG_REMOVED  	0x0400
    109 #define	SHMSEG_ALLOCATED	0x0800
    110 #define	SHMSEG_WANTED		0x1000
    111 
    112 static int	shm_last_free, shm_nused, shm_committed;
    113 struct	shmid_ds *shmsegs;
    114 
    115 struct shm_handle {
    116 	struct uvm_object *shm_object;
    117 };
    118 
    119 struct shmmap_entry {
    120 	SLIST_ENTRY(shmmap_entry) next;
    121 	vaddr_t va;
    122 	int shmid;
    123 };
    124 
    125 static POOL_INIT(shmmap_entry_pool, sizeof(struct shmmap_entry), 0, 0, 0,
    126     "shmmp", 0);
    127 
    128 struct shmmap_state {
    129 	unsigned int nitems;
    130 	unsigned int nrefs;
    131 	SLIST_HEAD(, shmmap_entry) entries;
    132 };
    133 
    134 static int shm_find_segment_by_key(key_t);
    135 static void shm_deallocate_segment(struct shmid_ds *);
    136 static void shm_delete_mapping(struct vmspace *, struct shmmap_state *,
    137 			       struct shmmap_entry *);
    138 static int shmget_existing(struct proc *, struct sys_shmget_args *,
    139 			   int, int, register_t *);
    140 static int shmget_allocate_segment(struct proc *, struct sys_shmget_args *,
    141 				   int, register_t *);
    142 static struct shmmap_state *shmmap_getprivate(struct proc *);
    143 static struct shmmap_entry *shm_find_mapping(struct shmmap_state *, vaddr_t);
    144 
    145 static int
    146 shm_find_segment_by_key(key)
    147 	key_t key;
    148 {
    149 	int i;
    150 
    151 	for (i = 0; i < shminfo.shmmni; i++)
    152 		if ((shmsegs[i].shm_perm.mode & SHMSEG_ALLOCATED) &&
    153 		    shmsegs[i].shm_perm._key == key)
    154 			return i;
    155 	return -1;
    156 }
    157 
    158 struct shmid_ds *
    159 shm_find_segment_by_shmid(shmid, findremoved)
    160 	int shmid;
    161 	int findremoved;
    162 {
    163 	int segnum;
    164 	struct shmid_ds *shmseg;
    165 
    166 	segnum = IPCID_TO_IX(shmid);
    167 	if (segnum < 0 || segnum >= shminfo.shmmni)
    168 		return NULL;
    169 	shmseg = &shmsegs[segnum];
    170 	if ((shmseg->shm_perm.mode & SHMSEG_ALLOCATED) == 0)
    171 		return NULL;
    172 	if (!findremoved && ((shmseg->shm_perm.mode & SHMSEG_REMOVED) != 0))
    173 		return NULL;
    174 	if (shmseg->shm_perm._seq != IPCID_TO_SEQ(shmid))
    175 		return NULL;
    176 	return shmseg;
    177 }
    178 
    179 static void
    180 shm_deallocate_segment(shmseg)
    181 	struct shmid_ds *shmseg;
    182 {
    183 	struct shm_handle *shm_handle = shmseg->_shm_internal;
    184 	struct uvm_object *uobj = shm_handle->shm_object;
    185 	size_t size = (shmseg->shm_segsz + PGOFSET) & ~PGOFSET;
    186 
    187 	(*uobj->pgops->pgo_detach)(uobj);
    188 	free((caddr_t)shm_handle, M_SHM);
    189 	shmseg->_shm_internal = NULL;
    190 	shm_committed -= btoc(size);
    191 	shmseg->shm_perm.mode = SHMSEG_FREE;
    192 	shm_nused--;
    193 }
    194 
    195 static void
    196 shm_delete_mapping(vm, shmmap_s, shmmap_se)
    197 	struct vmspace *vm;
    198 	struct shmmap_state *shmmap_s;
    199 	struct shmmap_entry *shmmap_se;
    200 {
    201 	struct shmid_ds *shmseg;
    202 	int segnum;
    203 	size_t size;
    204 
    205 	segnum = IPCID_TO_IX(shmmap_se->shmid);
    206 #ifdef DEBUG
    207 	if (segnum < 0 || segnum >= shminfo.shmmni)
    208 		panic("shm_delete_mapping: vmspace %p state %p entry %p - "
    209 		    "entry segment ID bad (%d)",
    210 		    vm, shmmap_s, shmmap_se, segnum);
    211 #endif
    212 	shmseg = &shmsegs[segnum];
    213 	size = (shmseg->shm_segsz + PGOFSET) & ~PGOFSET;
    214 	uvm_deallocate(&vm->vm_map, shmmap_se->va, size);
    215 	SLIST_REMOVE(&shmmap_s->entries, shmmap_se, shmmap_entry, next);
    216 	shmmap_s->nitems--;
    217 	pool_put(&shmmap_entry_pool, shmmap_se);
    218 	shmseg->shm_dtime = time.tv_sec;
    219 	if ((--shmseg->shm_nattch <= 0) &&
    220 	    (shmseg->shm_perm.mode & SHMSEG_REMOVED)) {
    221 		shm_deallocate_segment(shmseg);
    222 		shm_last_free = segnum;
    223 	}
    224 }
    225 
    226 /*
    227  * Get a non-shared shm map for that vmspace.
    228  * 3 cases:
    229  *   - no shm map present: create a fresh one
    230  *   - a shm map with refcount=1, just used by ourselves: fine
    231  *   - a shared shm map: copy to a fresh one and adjust refcounts
    232  */
    233 static struct shmmap_state *
    234 shmmap_getprivate(struct proc *p)
    235 {
    236 	struct shmmap_state *oshmmap_s, *shmmap_s;
    237 	struct shmmap_entry *oshmmap_se, *shmmap_se;
    238 
    239 	oshmmap_s = (struct shmmap_state *)p->p_vmspace->vm_shm;
    240 	if (oshmmap_s && oshmmap_s->nrefs == 1)
    241 		return (oshmmap_s);
    242 
    243 	shmmap_s = malloc(sizeof(struct shmmap_state), M_SHM, M_WAITOK);
    244 	memset(shmmap_s, 0, sizeof(struct shmmap_state));
    245 	shmmap_s->nrefs = 1;
    246 	SLIST_INIT(&shmmap_s->entries);
    247 	p->p_vmspace->vm_shm = (caddr_t)shmmap_s;
    248 
    249 	if (!oshmmap_s)
    250 		return (shmmap_s);
    251 
    252 #ifdef SHMDEBUG
    253 	printf("shmmap_getprivate: vm %p split (%d entries), was used by %d\n",
    254 	       p->p_vmspace, oshmmap_s->nitems, oshmmap_s->nrefs);
    255 #endif
    256 	SLIST_FOREACH(oshmmap_se, &oshmmap_s->entries, next) {
    257 		shmmap_se = pool_get(&shmmap_entry_pool, PR_WAITOK);
    258 		shmmap_se->va = oshmmap_se->va;
    259 		shmmap_se->shmid = oshmmap_se->shmid;
    260 		SLIST_INSERT_HEAD(&shmmap_s->entries, shmmap_se, next);
    261 	}
    262 	shmmap_s->nitems = oshmmap_s->nitems;
    263 	oshmmap_s->nrefs--;
    264 	return (shmmap_s);
    265 }
    266 
    267 static struct shmmap_entry *
    268 shm_find_mapping(map, va)
    269 	struct shmmap_state *map;
    270 	vaddr_t va;
    271 {
    272 	struct shmmap_entry *shmmap_se;
    273 
    274 	SLIST_FOREACH(shmmap_se, &map->entries, next) {
    275 		if (shmmap_se->va == va)
    276 			return shmmap_se;
    277 	}
    278 	return 0;
    279 }
    280 
    281 int
    282 sys_shmdt(l, v, retval)
    283 	struct lwp *l;
    284 	void *v;
    285 	register_t *retval;
    286 {
    287 	struct sys_shmdt_args /* {
    288 		syscallarg(const void *) shmaddr;
    289 	} */ *uap = v;
    290 	struct proc *p = l->l_proc;
    291 	struct shmmap_state *shmmap_s, *shmmap_s1;
    292 	struct shmmap_entry *shmmap_se;
    293 
    294 	shmmap_s = (struct shmmap_state *)p->p_vmspace->vm_shm;
    295 	if (shmmap_s == NULL)
    296 		return EINVAL;
    297 
    298 	shmmap_se = shm_find_mapping(shmmap_s, (vaddr_t)SCARG(uap, shmaddr));
    299 	if (!shmmap_se)
    300 		return EINVAL;
    301 
    302 	shmmap_s1 = shmmap_getprivate(p);
    303 	if (shmmap_s1 != shmmap_s) {
    304 		/* map has been copied, lookup entry in new map */
    305 		shmmap_se = shm_find_mapping(shmmap_s1,
    306 					     (vaddr_t)SCARG(uap, shmaddr));
    307 		KASSERT(shmmap_se != NULL);
    308 	}
    309 #ifdef SHMDEBUG
    310 	printf("shmdt: vm %p: remove %d @%lx\n",
    311 	       p->p_vmspace, shmmap_se->shmid, shmmap_se->va);
    312 #endif
    313 	shm_delete_mapping(p->p_vmspace, shmmap_s1, shmmap_se);
    314 	return 0;
    315 }
    316 
    317 int
    318 sys_shmat(l, v, retval)
    319 	struct lwp *l;
    320 	void *v;
    321 	register_t *retval;
    322 {
    323 	struct sys_shmat_args /* {
    324 		syscallarg(int) shmid;
    325 		syscallarg(const void *) shmaddr;
    326 		syscallarg(int) shmflg;
    327 	} */ *uap = v;
    328 	struct proc *p = l->l_proc;
    329 	vaddr_t attach_va;
    330 	int error;
    331 
    332 	error = shmat1(p, SCARG(uap, shmid), SCARG(uap, shmaddr),
    333 	    SCARG(uap, shmflg), &attach_va, 0);
    334 	if (error != 0)
    335 		return error;
    336 	retval[0] = attach_va;
    337 	return 0;
    338 }
    339 
    340 int
    341 shmat1(p, shmid, shmaddr, shmflg, attachp, findremoved)
    342 	struct proc *p;
    343 	int shmid;
    344 	const void *shmaddr;
    345 	int shmflg;
    346 	vaddr_t *attachp;
    347 	int findremoved;
    348 {
    349 	int error, flags;
    350 	struct ucred *cred = p->p_ucred;
    351 	struct shmid_ds *shmseg;
    352 	struct shmmap_state *shmmap_s;
    353 	struct uvm_object *uobj;
    354 	vaddr_t attach_va;
    355 	vm_prot_t prot;
    356 	vsize_t size;
    357 	struct shmmap_entry *shmmap_se;
    358 
    359 	shmseg = shm_find_segment_by_shmid(shmid, findremoved);
    360 	if (shmseg == NULL)
    361 		return EINVAL;
    362 	error = ipcperm(cred, &shmseg->shm_perm,
    363 		    (shmflg & SHM_RDONLY) ? IPC_R : IPC_R|IPC_W);
    364 	if (error)
    365 		return error;
    366 
    367 	shmmap_s = (struct shmmap_state *)p->p_vmspace->vm_shm;
    368 	if (shmmap_s && shmmap_s->nitems >= shminfo.shmseg)
    369 		return EMFILE;
    370 
    371 	size = (shmseg->shm_segsz + PGOFSET) & ~PGOFSET;
    372 	prot = VM_PROT_READ;
    373 	if ((shmflg & SHM_RDONLY) == 0)
    374 		prot |= VM_PROT_WRITE;
    375 	flags = MAP_ANON | MAP_SHARED;
    376 	if (shmaddr) {
    377 		flags |= MAP_FIXED;
    378 		if (shmflg & SHM_RND)
    379 			attach_va =
    380 			    (vaddr_t)shmaddr & ~(SHMLBA-1);
    381 		else if (((vaddr_t)shmaddr & (SHMLBA-1)) == 0)
    382 			attach_va = (vaddr_t)shmaddr;
    383 		else
    384 			return EINVAL;
    385 	} else {
    386 		/* This is just a hint to uvm_mmap() about where to put it. */
    387 		attach_va = VM_DEFAULT_ADDRESS(p->p_vmspace->vm_daddr, size);
    388 	}
    389 	uobj = ((struct shm_handle *)shmseg->_shm_internal)->shm_object;
    390 	(*uobj->pgops->pgo_reference)(uobj);
    391 	error = uvm_map(&p->p_vmspace->vm_map, &attach_va, size,
    392 	    uobj, 0, 0,
    393 	    UVM_MAPFLAG(prot, prot, UVM_INH_SHARE, UVM_ADV_RANDOM, 0));
    394 	if (error) {
    395 		(*uobj->pgops->pgo_detach)(uobj);
    396 		return error;
    397 	}
    398 	shmmap_se = pool_get(&shmmap_entry_pool, PR_WAITOK);
    399 	shmmap_se->va = attach_va;
    400 	shmmap_se->shmid = shmid;
    401 	shmmap_s = shmmap_getprivate(p);
    402 #ifdef SHMDEBUG
    403 	printf("shmat: vm %p: add %d @%lx\n", p->p_vmspace, shmid, attach_va);
    404 #endif
    405 	SLIST_INSERT_HEAD(&shmmap_s->entries, shmmap_se, next);
    406 	shmmap_s->nitems++;
    407 	shmseg->shm_lpid = p->p_pid;
    408 	shmseg->shm_atime = time.tv_sec;
    409 	shmseg->shm_nattch++;
    410 	*attachp = attach_va;
    411 	return 0;
    412 }
    413 
    414 int
    415 sys___shmctl13(l, v, retval)
    416 	struct lwp *l;
    417 	void *v;
    418 	register_t *retval;
    419 {
    420 	struct sys___shmctl13_args /* {
    421 		syscallarg(int) shmid;
    422 		syscallarg(int) cmd;
    423 		syscallarg(struct shmid_ds *) buf;
    424 	} */ *uap = v;
    425 	struct proc *p = l->l_proc;
    426 	struct shmid_ds shmbuf;
    427 	int cmd, error;
    428 
    429 	cmd = SCARG(uap, cmd);
    430 
    431 	if (cmd == IPC_SET) {
    432 		error = copyin(SCARG(uap, buf), &shmbuf, sizeof(shmbuf));
    433 		if (error)
    434 			return (error);
    435 	}
    436 
    437 	error = shmctl1(p, SCARG(uap, shmid), cmd,
    438 	    (cmd == IPC_SET || cmd == IPC_STAT) ? &shmbuf : NULL);
    439 
    440 	if (error == 0 && cmd == IPC_STAT)
    441 		error = copyout(&shmbuf, SCARG(uap, buf), sizeof(shmbuf));
    442 
    443 	return (error);
    444 }
    445 
    446 int
    447 shmctl1(p, shmid, cmd, shmbuf)
    448 	struct proc *p;
    449 	int shmid;
    450 	int cmd;
    451 	struct shmid_ds *shmbuf;
    452 {
    453 	struct ucred *cred = p->p_ucred;
    454 	struct shmid_ds *shmseg;
    455 	int error = 0;
    456 
    457 	shmseg = shm_find_segment_by_shmid(shmid, 0);
    458 	if (shmseg == NULL)
    459 		return EINVAL;
    460 	switch (cmd) {
    461 	case IPC_STAT:
    462 		if ((error = ipcperm(cred, &shmseg->shm_perm, IPC_R)) != 0)
    463 			return error;
    464 		memcpy(shmbuf, shmseg, sizeof(struct shmid_ds));
    465 		break;
    466 	case IPC_SET:
    467 		if ((error = ipcperm(cred, &shmseg->shm_perm, IPC_M)) != 0)
    468 			return error;
    469 		shmseg->shm_perm.uid = shmbuf->shm_perm.uid;
    470 		shmseg->shm_perm.gid = shmbuf->shm_perm.gid;
    471 		shmseg->shm_perm.mode =
    472 		    (shmseg->shm_perm.mode & ~ACCESSPERMS) |
    473 		    (shmbuf->shm_perm.mode & ACCESSPERMS);
    474 		shmseg->shm_ctime = time.tv_sec;
    475 		break;
    476 	case IPC_RMID:
    477 		if ((error = ipcperm(cred, &shmseg->shm_perm, IPC_M)) != 0)
    478 			return error;
    479 		shmseg->shm_perm._key = IPC_PRIVATE;
    480 		shmseg->shm_perm.mode |= SHMSEG_REMOVED;
    481 		if (shmseg->shm_nattch <= 0) {
    482 			shm_deallocate_segment(shmseg);
    483 			shm_last_free = IPCID_TO_IX(shmid);
    484 		}
    485 		break;
    486 	case SHM_LOCK:
    487 	case SHM_UNLOCK:
    488 	default:
    489 		return EINVAL;
    490 	}
    491 	return 0;
    492 }
    493 
    494 static int
    495 shmget_existing(p, uap, mode, segnum, retval)
    496 	struct proc *p;
    497 	struct sys_shmget_args /* {
    498 		syscallarg(key_t) key;
    499 		syscallarg(size_t) size;
    500 		syscallarg(int) shmflg;
    501 	} */ *uap;
    502 	int mode;
    503 	int segnum;
    504 	register_t *retval;
    505 {
    506 	struct shmid_ds *shmseg;
    507 	struct ucred *cred = p->p_ucred;
    508 	int error;
    509 
    510 	shmseg = &shmsegs[segnum];
    511 	if (shmseg->shm_perm.mode & SHMSEG_REMOVED) {
    512 		/*
    513 		 * This segment is in the process of being allocated.  Wait
    514 		 * until it's done, and look the key up again (in case the
    515 		 * allocation failed or it was freed).
    516 		 */
    517 		shmseg->shm_perm.mode |= SHMSEG_WANTED;
    518 		error = tsleep((caddr_t)shmseg, PLOCK | PCATCH, "shmget", 0);
    519 		if (error)
    520 			return error;
    521 		return EAGAIN;
    522 	}
    523 	if ((error = ipcperm(cred, &shmseg->shm_perm, mode)) != 0)
    524 		return error;
    525 	if (SCARG(uap, size) && SCARG(uap, size) > shmseg->shm_segsz)
    526 		return EINVAL;
    527 	if ((SCARG(uap, shmflg) & (IPC_CREAT | IPC_EXCL)) ==
    528 	    (IPC_CREAT | IPC_EXCL))
    529 		return EEXIST;
    530 	*retval = IXSEQ_TO_IPCID(segnum, shmseg->shm_perm);
    531 	return 0;
    532 }
    533 
    534 static int
    535 shmget_allocate_segment(p, uap, mode, retval)
    536 	struct proc *p;
    537 	struct sys_shmget_args /* {
    538 		syscallarg(key_t) key;
    539 		syscallarg(size_t) size;
    540 		syscallarg(int) shmflg;
    541 	} */ *uap;
    542 	int mode;
    543 	register_t *retval;
    544 {
    545 	int i, segnum, shmid, size;
    546 	struct ucred *cred = p->p_ucred;
    547 	struct shmid_ds *shmseg;
    548 	struct shm_handle *shm_handle;
    549 	int error = 0;
    550 
    551 	if (SCARG(uap, size) < shminfo.shmmin ||
    552 	    SCARG(uap, size) > shminfo.shmmax)
    553 		return EINVAL;
    554 	if (shm_nused >= shminfo.shmmni) /* any shmids left? */
    555 		return ENOSPC;
    556 	size = (SCARG(uap, size) + PGOFSET) & ~PGOFSET;
    557 	if (shm_committed + btoc(size) > shminfo.shmall)
    558 		return ENOMEM;
    559 	if (shm_last_free < 0) {
    560 		for (i = 0; i < shminfo.shmmni; i++)
    561 			if (shmsegs[i].shm_perm.mode & SHMSEG_FREE)
    562 				break;
    563 		if (i == shminfo.shmmni)
    564 			panic("shmseg free count inconsistent");
    565 		segnum = i;
    566 	} else  {
    567 		segnum = shm_last_free;
    568 		shm_last_free = -1;
    569 	}
    570 	shmseg = &shmsegs[segnum];
    571 	/*
    572 	 * In case we sleep in malloc(), mark the segment present but deleted
    573 	 * so that noone else tries to create the same key.
    574 	 */
    575 	shmseg->shm_perm.mode = SHMSEG_ALLOCATED | SHMSEG_REMOVED;
    576 	shmseg->shm_perm._key = SCARG(uap, key);
    577 	shmseg->shm_perm._seq = (shmseg->shm_perm._seq + 1) & 0x7fff;
    578 	shm_handle = (struct shm_handle *)
    579 	    malloc(sizeof(struct shm_handle), M_SHM, M_WAITOK);
    580 	shmid = IXSEQ_TO_IPCID(segnum, shmseg->shm_perm);
    581 
    582 	shm_handle->shm_object = uao_create(size, 0);
    583 
    584 	shmseg->_shm_internal = shm_handle;
    585 	shmseg->shm_perm.cuid = shmseg->shm_perm.uid = cred->cr_uid;
    586 	shmseg->shm_perm.cgid = shmseg->shm_perm.gid = cred->cr_gid;
    587 	shmseg->shm_perm.mode = (shmseg->shm_perm.mode & SHMSEG_WANTED) |
    588 	    (mode & ACCESSPERMS) | SHMSEG_ALLOCATED;
    589 	shmseg->shm_segsz = SCARG(uap, size);
    590 	shmseg->shm_cpid = p->p_pid;
    591 	shmseg->shm_lpid = shmseg->shm_nattch = 0;
    592 	shmseg->shm_atime = shmseg->shm_dtime = 0;
    593 	shmseg->shm_ctime = time.tv_sec;
    594 	shm_committed += btoc(size);
    595 	shm_nused++;
    596 
    597 	*retval = shmid;
    598 	if (shmseg->shm_perm.mode & SHMSEG_WANTED) {
    599 		/*
    600 		 * Somebody else wanted this key while we were asleep.  Wake
    601 		 * them up now.
    602 		 */
    603 		shmseg->shm_perm.mode &= ~SHMSEG_WANTED;
    604 		wakeup((caddr_t)shmseg);
    605 	}
    606 	return error;
    607 }
    608 
    609 int
    610 sys_shmget(l, v, retval)
    611 	struct lwp *l;
    612 	void *v;
    613 	register_t *retval;
    614 {
    615 	struct sys_shmget_args /* {
    616 		syscallarg(key_t) key;
    617 		syscallarg(int) size;
    618 		syscallarg(int) shmflg;
    619 	} */ *uap = v;
    620 	struct proc *p = l->l_proc;
    621 	int segnum, mode, error;
    622 
    623 	mode = SCARG(uap, shmflg) & ACCESSPERMS;
    624 	if (SCARG(uap, key) != IPC_PRIVATE) {
    625 	again:
    626 		segnum = shm_find_segment_by_key(SCARG(uap, key));
    627 		if (segnum >= 0) {
    628 			error = shmget_existing(p, uap, mode, segnum, retval);
    629 			if (error == EAGAIN)
    630 				goto again;
    631 			return error;
    632 		}
    633 		if ((SCARG(uap, shmflg) & IPC_CREAT) == 0)
    634 			return ENOENT;
    635 	}
    636 	return shmget_allocate_segment(p, uap, mode, retval);
    637 }
    638 
    639 void
    640 shmfork(vm1, vm2)
    641 	struct vmspace *vm1, *vm2;
    642 {
    643 	struct shmmap_state *shmmap_s;
    644 	struct shmmap_entry *shmmap_se;
    645 
    646 	vm2->vm_shm = vm1->vm_shm;
    647 
    648 	if (vm1->vm_shm == NULL)
    649 		return;
    650 
    651 #ifdef SHMDEBUG
    652 	printf("shmfork %p->%p\n", vm1, vm2);
    653 #endif
    654 
    655 	shmmap_s = (struct shmmap_state *)vm1->vm_shm;
    656 
    657 	SLIST_FOREACH(shmmap_se, &shmmap_s->entries, next)
    658 		shmsegs[IPCID_TO_IX(shmmap_se->shmid)].shm_nattch++;
    659 	shmmap_s->nrefs++;
    660 }
    661 
    662 void
    663 shmexit(vm)
    664 	struct vmspace *vm;
    665 {
    666 	struct shmmap_state *shmmap_s;
    667 	struct shmmap_entry *shmmap_se;
    668 
    669 	shmmap_s = (struct shmmap_state *)vm->vm_shm;
    670 	if (shmmap_s == NULL)
    671 		return;
    672 
    673 	vm->vm_shm = NULL;
    674 
    675 	if (--shmmap_s->nrefs > 0) {
    676 #ifdef SHMDEBUG
    677 		printf("shmexit: vm %p drop ref (%d entries), now used by %d\n",
    678 		       vm, shmmap_s->nitems, shmmap_s->nrefs);
    679 #endif
    680 		SLIST_FOREACH(shmmap_se, &shmmap_s->entries, next)
    681 			shmsegs[IPCID_TO_IX(shmmap_se->shmid)].shm_nattch--;
    682 		return;
    683 	}
    684 
    685 #ifdef SHMDEBUG
    686 	printf("shmexit: vm %p cleanup (%d entries)\n", vm, shmmap_s->nitems);
    687 #endif
    688 	while (!SLIST_EMPTY(&shmmap_s->entries)) {
    689 		shmmap_se = SLIST_FIRST(&shmmap_s->entries);
    690 		shm_delete_mapping(vm, shmmap_s, shmmap_se);
    691 	}
    692 	KASSERT(shmmap_s->nitems == 0);
    693 	free(shmmap_s, M_SHM);
    694 }
    695 
    696 void
    697 shminit()
    698 {
    699 	int i, sz;
    700 	vaddr_t v;
    701 
    702 	/* Allocate pageable memory for our structures */
    703 	sz = shminfo.shmmni * sizeof(struct shmid_ds);
    704 	if ((v = uvm_km_alloc(kernel_map, round_page(sz))) == 0)
    705 		panic("sysv_shm: cannot allocate memory");
    706 	shmsegs = (void *)v;
    707 
    708 	shminfo.shmmax *= PAGE_SIZE;
    709 
    710 	for (i = 0; i < shminfo.shmmni; i++) {
    711 		shmsegs[i].shm_perm.mode = SHMSEG_FREE;
    712 		shmsegs[i].shm_perm._seq = 0;
    713 	}
    714 	shm_last_free = 0;
    715 	shm_nused = 0;
    716 	shm_committed = 0;
    717 }
    718