Home | History | Annotate | Line # | Download | only in kern
sysv_shm.c revision 1.86.10.1
      1 /*	$NetBSD: sysv_shm.c,v 1.86.10.1 2006/03/08 00:53:41 elad 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 1.86.10.1 2006/03/08 00:53:41 elad 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 static MALLOC_DEFINE(M_SHM, "shm", "SVID compatible shared memory segments");
     92 
     93 /*
     94  * Provides the following externally accessible functions:
     95  *
     96  * shminit(void);		                 initialization
     97  * shmexit(struct vmspace *)                     cleanup
     98  * shmfork(struct vmspace *, struct vmspace *)   fork handling
     99  *
    100  * Structures:
    101  * shmsegs (an array of 'struct shmid_ds')
    102  * per proc array of 'struct shmmap_state'
    103  */
    104 
    105 int shm_nused;
    106 struct	shmid_ds *shmsegs;
    107 
    108 struct shmmap_entry {
    109 	SLIST_ENTRY(shmmap_entry) next;
    110 	vaddr_t va;
    111 	int shmid;
    112 };
    113 
    114 static int	shm_last_free, shm_committed;
    115 
    116 static POOL_INIT(shmmap_entry_pool, sizeof(struct shmmap_entry), 0, 0, 0,
    117     "shmmp", &pool_allocator_nointr);
    118 
    119 struct shmmap_state {
    120 	unsigned int nitems;
    121 	unsigned int nrefs;
    122 	SLIST_HEAD(, shmmap_entry) entries;
    123 };
    124 
    125 static int shm_find_segment_by_key(key_t);
    126 static void shm_deallocate_segment(struct shmid_ds *);
    127 static void shm_delete_mapping(struct vmspace *, struct shmmap_state *,
    128 			       struct shmmap_entry *);
    129 static int shmget_existing(struct proc *, struct sys_shmget_args *,
    130 			   int, int, register_t *);
    131 static int shmget_allocate_segment(struct proc *, struct sys_shmget_args *,
    132 				   int, register_t *);
    133 static struct shmmap_state *shmmap_getprivate(struct proc *);
    134 static struct shmmap_entry *shm_find_mapping(struct shmmap_state *, vaddr_t);
    135 
    136 static int
    137 shm_find_segment_by_key(key_t key)
    138 {
    139 	int i;
    140 
    141 	for (i = 0; i < shminfo.shmmni; i++)
    142 		if ((shmsegs[i].shm_perm.mode & SHMSEG_ALLOCATED) &&
    143 		    shmsegs[i].shm_perm._key == key)
    144 			return i;
    145 	return -1;
    146 }
    147 
    148 static struct shmid_ds *
    149 shm_find_segment_by_shmid(int shmid)
    150 {
    151 	int segnum;
    152 	struct shmid_ds *shmseg;
    153 
    154 	segnum = IPCID_TO_IX(shmid);
    155 	if (segnum < 0 || segnum >= shminfo.shmmni)
    156 		return NULL;
    157 	shmseg = &shmsegs[segnum];
    158 	if ((shmseg->shm_perm.mode & SHMSEG_ALLOCATED) == 0)
    159 		return NULL;
    160 	if ((shmseg->shm_perm.mode & (SHMSEG_REMOVED|SHMSEG_RMLINGER)) == SHMSEG_REMOVED)
    161 		return NULL;
    162 	if (shmseg->shm_perm._seq != IPCID_TO_SEQ(shmid))
    163 		return NULL;
    164 	return shmseg;
    165 }
    166 
    167 static void
    168 shm_deallocate_segment(struct shmid_ds *shmseg)
    169 {
    170 	struct uvm_object *uobj = shmseg->_shm_internal;
    171 	size_t size = (shmseg->shm_segsz + PGOFSET) & ~PGOFSET;
    172 
    173 #ifdef SHMDEBUG
    174 	printf("shm freeing key 0x%lx seq 0x%x\n",
    175 	       shmseg->shm_perm._key, shmseg->shm_perm._seq);
    176 #endif
    177 
    178 	(*uobj->pgops->pgo_detach)(uobj);
    179 	shmseg->_shm_internal = NULL;
    180 	shm_committed -= btoc(size);
    181 	shmseg->shm_perm.mode = SHMSEG_FREE;
    182 	shm_nused--;
    183 }
    184 
    185 static void
    186 shm_delete_mapping(struct vmspace *vm, struct shmmap_state *shmmap_s,
    187     struct shmmap_entry *shmmap_se)
    188 {
    189 	struct shmid_ds *shmseg;
    190 	int segnum;
    191 	size_t size;
    192 
    193 	segnum = IPCID_TO_IX(shmmap_se->shmid);
    194 #ifdef DEBUG
    195 	if (segnum < 0 || segnum >= shminfo.shmmni)
    196 		panic("shm_delete_mapping: vmspace %p state %p entry %p - "
    197 		    "entry segment ID bad (%d)",
    198 		    vm, shmmap_s, shmmap_se, segnum);
    199 #endif
    200 	shmseg = &shmsegs[segnum];
    201 	size = (shmseg->shm_segsz + PGOFSET) & ~PGOFSET;
    202 	uvm_deallocate(&vm->vm_map, shmmap_se->va, size);
    203 	SLIST_REMOVE(&shmmap_s->entries, shmmap_se, shmmap_entry, next);
    204 	shmmap_s->nitems--;
    205 	pool_put(&shmmap_entry_pool, shmmap_se);
    206 	shmseg->shm_dtime = time.tv_sec;
    207 	if ((--shmseg->shm_nattch <= 0) &&
    208 	    (shmseg->shm_perm.mode & SHMSEG_REMOVED)) {
    209 		shm_deallocate_segment(shmseg);
    210 		shm_last_free = segnum;
    211 	}
    212 }
    213 
    214 /*
    215  * Get a non-shared shm map for that vmspace.
    216  * 3 cases:
    217  *   - no shm map present: create a fresh one
    218  *   - a shm map with refcount=1, just used by ourselves: fine
    219  *   - a shared shm map: copy to a fresh one and adjust refcounts
    220  */
    221 static struct shmmap_state *
    222 shmmap_getprivate(struct proc *p)
    223 {
    224 	struct shmmap_state *oshmmap_s, *shmmap_s;
    225 	struct shmmap_entry *oshmmap_se, *shmmap_se;
    226 
    227 	oshmmap_s = (struct shmmap_state *)p->p_vmspace->vm_shm;
    228 	if (oshmmap_s && oshmmap_s->nrefs == 1)
    229 		return (oshmmap_s);
    230 
    231 	shmmap_s = malloc(sizeof(struct shmmap_state), M_SHM, M_WAITOK);
    232 	memset(shmmap_s, 0, sizeof(struct shmmap_state));
    233 	shmmap_s->nrefs = 1;
    234 	SLIST_INIT(&shmmap_s->entries);
    235 	p->p_vmspace->vm_shm = (caddr_t)shmmap_s;
    236 
    237 	if (!oshmmap_s)
    238 		return (shmmap_s);
    239 
    240 #ifdef SHMDEBUG
    241 	printf("shmmap_getprivate: vm %p split (%d entries), was used by %d\n",
    242 	       p->p_vmspace, oshmmap_s->nitems, oshmmap_s->nrefs);
    243 #endif
    244 	SLIST_FOREACH(oshmmap_se, &oshmmap_s->entries, next) {
    245 		shmmap_se = pool_get(&shmmap_entry_pool, PR_WAITOK);
    246 		shmmap_se->va = oshmmap_se->va;
    247 		shmmap_se->shmid = oshmmap_se->shmid;
    248 		SLIST_INSERT_HEAD(&shmmap_s->entries, shmmap_se, next);
    249 	}
    250 	shmmap_s->nitems = oshmmap_s->nitems;
    251 	oshmmap_s->nrefs--;
    252 	return (shmmap_s);
    253 }
    254 
    255 static struct shmmap_entry *
    256 shm_find_mapping(struct shmmap_state *map, vaddr_t va)
    257 {
    258 	struct shmmap_entry *shmmap_se;
    259 
    260 	SLIST_FOREACH(shmmap_se, &map->entries, next) {
    261 		if (shmmap_se->va == va)
    262 			return shmmap_se;
    263 	}
    264 	return 0;
    265 }
    266 
    267 int
    268 sys_shmdt(struct lwp *l, void *v, register_t *retval)
    269 {
    270 	struct sys_shmdt_args /* {
    271 		syscallarg(const void *) shmaddr;
    272 	} */ *uap = v;
    273 	struct proc *p = l->l_proc;
    274 	struct shmmap_state *shmmap_s, *shmmap_s1;
    275 	struct shmmap_entry *shmmap_se;
    276 
    277 	shmmap_s = (struct shmmap_state *)p->p_vmspace->vm_shm;
    278 	if (shmmap_s == NULL)
    279 		return EINVAL;
    280 
    281 	shmmap_se = shm_find_mapping(shmmap_s, (vaddr_t)SCARG(uap, shmaddr));
    282 	if (!shmmap_se)
    283 		return EINVAL;
    284 
    285 	shmmap_s1 = shmmap_getprivate(p);
    286 	if (shmmap_s1 != shmmap_s) {
    287 		/* map has been copied, lookup entry in new map */
    288 		shmmap_se = shm_find_mapping(shmmap_s1,
    289 					     (vaddr_t)SCARG(uap, shmaddr));
    290 		KASSERT(shmmap_se != NULL);
    291 	}
    292 #ifdef SHMDEBUG
    293 	printf("shmdt: vm %p: remove %d @%lx\n",
    294 	       p->p_vmspace, shmmap_se->shmid, shmmap_se->va);
    295 #endif
    296 	shm_delete_mapping(p->p_vmspace, shmmap_s1, shmmap_se);
    297 	return 0;
    298 }
    299 
    300 int
    301 sys_shmat(struct lwp *l, void *v, register_t *retval)
    302 {
    303 	struct sys_shmat_args /* {
    304 		syscallarg(int) shmid;
    305 		syscallarg(const void *) shmaddr;
    306 		syscallarg(int) shmflg;
    307 	} */ *uap = v;
    308 	int error, flags;
    309 	struct proc *p = l->l_proc;
    310 	kauth_cred_t cred = p->p_cred;
    311 	struct shmid_ds *shmseg;
    312 	struct shmmap_state *shmmap_s;
    313 	struct uvm_object *uobj;
    314 	vaddr_t attach_va;
    315 	vm_prot_t prot;
    316 	vsize_t size;
    317 	struct shmmap_entry *shmmap_se;
    318 
    319 	shmseg = shm_find_segment_by_shmid(SCARG(uap, shmid));
    320 	if (shmseg == NULL)
    321 		return EINVAL;
    322 	error = ipcperm(cred, &shmseg->shm_perm,
    323 		    (SCARG(uap, shmflg) & SHM_RDONLY) ? IPC_R : IPC_R|IPC_W);
    324 	if (error)
    325 		return error;
    326 
    327 	shmmap_s = (struct shmmap_state *)p->p_vmspace->vm_shm;
    328 	if (shmmap_s && shmmap_s->nitems >= shminfo.shmseg)
    329 		return EMFILE;
    330 
    331 	size = (shmseg->shm_segsz + PGOFSET) & ~PGOFSET;
    332 	prot = VM_PROT_READ;
    333 	if ((SCARG(uap, shmflg) & SHM_RDONLY) == 0)
    334 		prot |= VM_PROT_WRITE;
    335 	flags = MAP_ANON | MAP_SHARED;
    336 	if (SCARG(uap, shmaddr)) {
    337 		flags |= MAP_FIXED;
    338 		if (SCARG(uap, shmflg) & SHM_RND)
    339 			attach_va =
    340 			    (vaddr_t)SCARG(uap, shmaddr) & ~(SHMLBA-1);
    341 		else if (((vaddr_t)SCARG(uap, shmaddr) & (SHMLBA-1)) == 0)
    342 			attach_va = (vaddr_t)SCARG(uap, shmaddr);
    343 		else
    344 			return EINVAL;
    345 	} else {
    346 		/* This is just a hint to uvm_mmap() about where to put it. */
    347 		attach_va = p->p_emul->e_vm_default_addr(p,
    348 		    (vaddr_t)p->p_vmspace->vm_daddr, size);
    349 	}
    350 	uobj = shmseg->_shm_internal;
    351 	(*uobj->pgops->pgo_reference)(uobj);
    352 	error = uvm_map(&p->p_vmspace->vm_map, &attach_va, size,
    353 	    uobj, 0, 0,
    354 	    UVM_MAPFLAG(prot, prot, UVM_INH_SHARE, UVM_ADV_RANDOM, 0));
    355 	if (error) {
    356 		(*uobj->pgops->pgo_detach)(uobj);
    357 		return error;
    358 	}
    359 	shmmap_se = pool_get(&shmmap_entry_pool, PR_WAITOK);
    360 	shmmap_se->va = attach_va;
    361 	shmmap_se->shmid = SCARG(uap, shmid);
    362 	shmmap_s = shmmap_getprivate(p);
    363 #ifdef SHMDEBUG
    364 	printf("shmat: vm %p: add %d @%lx\n", p->p_vmspace, shmmap_se->shmid, attach_va);
    365 #endif
    366 	SLIST_INSERT_HEAD(&shmmap_s->entries, shmmap_se, next);
    367 	shmmap_s->nitems++;
    368 	shmseg->shm_lpid = p->p_pid;
    369 	shmseg->shm_atime = time.tv_sec;
    370 	shmseg->shm_nattch++;
    371 
    372 	retval[0] = attach_va;
    373 	return 0;
    374 }
    375 
    376 int
    377 sys___shmctl13(struct lwp *l, void *v, register_t *retval)
    378 {
    379 	struct sys___shmctl13_args /* {
    380 		syscallarg(int) shmid;
    381 		syscallarg(int) cmd;
    382 		syscallarg(struct shmid_ds *) buf;
    383 	} */ *uap = v;
    384 	struct proc *p = l->l_proc;
    385 	struct shmid_ds shmbuf;
    386 	int cmd, error;
    387 
    388 	cmd = SCARG(uap, cmd);
    389 
    390 	if (cmd == IPC_SET) {
    391 		error = copyin(SCARG(uap, buf), &shmbuf, sizeof(shmbuf));
    392 		if (error)
    393 			return (error);
    394 	}
    395 
    396 	error = shmctl1(p, SCARG(uap, shmid), cmd,
    397 	    (cmd == IPC_SET || cmd == IPC_STAT) ? &shmbuf : NULL);
    398 
    399 	if (error == 0 && cmd == IPC_STAT)
    400 		error = copyout(&shmbuf, SCARG(uap, buf), sizeof(shmbuf));
    401 
    402 	return (error);
    403 }
    404 
    405 int
    406 shmctl1(struct proc *p, int shmid, int cmd, struct shmid_ds *shmbuf)
    407 {
    408 	kauth_cred_t cred = p->p_cred;
    409 	struct shmid_ds *shmseg;
    410 	int error = 0;
    411 
    412 	shmseg = shm_find_segment_by_shmid(shmid);
    413 	if (shmseg == NULL)
    414 		return EINVAL;
    415 	switch (cmd) {
    416 	case IPC_STAT:
    417 		if ((error = ipcperm(cred, &shmseg->shm_perm, IPC_R)) != 0)
    418 			return error;
    419 		memcpy(shmbuf, shmseg, sizeof(struct shmid_ds));
    420 		break;
    421 	case IPC_SET:
    422 		if ((error = ipcperm(cred, &shmseg->shm_perm, IPC_M)) != 0)
    423 			return error;
    424 		shmseg->shm_perm.uid = shmbuf->shm_perm.uid;
    425 		shmseg->shm_perm.gid = shmbuf->shm_perm.gid;
    426 		shmseg->shm_perm.mode =
    427 		    (shmseg->shm_perm.mode & ~ACCESSPERMS) |
    428 		    (shmbuf->shm_perm.mode & ACCESSPERMS);
    429 		shmseg->shm_ctime = time.tv_sec;
    430 		break;
    431 	case IPC_RMID:
    432 		if ((error = ipcperm(cred, &shmseg->shm_perm, IPC_M)) != 0)
    433 			return error;
    434 		shmseg->shm_perm._key = IPC_PRIVATE;
    435 		shmseg->shm_perm.mode |= SHMSEG_REMOVED;
    436 		if (shmseg->shm_nattch <= 0) {
    437 			shm_deallocate_segment(shmseg);
    438 			shm_last_free = IPCID_TO_IX(shmid);
    439 		}
    440 		break;
    441 	case SHM_LOCK:
    442 	case SHM_UNLOCK:
    443 	default:
    444 		return EINVAL;
    445 	}
    446 	return 0;
    447 }
    448 
    449 static int
    450 shmget_existing(struct proc *p, struct sys_shmget_args *uap, int mode,
    451     int segnum, register_t *retval)
    452 {
    453 	struct shmid_ds *shmseg;
    454 	kauth_cred_t cred = p->p_cred;
    455 	int error;
    456 
    457 	shmseg = &shmsegs[segnum];
    458 	if (shmseg->shm_perm.mode & SHMSEG_REMOVED) {
    459 		/*
    460 		 * This segment is in the process of being allocated.  Wait
    461 		 * until it's done, and look the key up again (in case the
    462 		 * allocation failed or it was freed).
    463 		 */
    464 		shmseg->shm_perm.mode |= SHMSEG_WANTED;
    465 		error = tsleep((caddr_t)shmseg, PLOCK | PCATCH, "shmget", 0);
    466 		if (error)
    467 			return error;
    468 		return EAGAIN;
    469 	}
    470 	if ((error = ipcperm(cred, &shmseg->shm_perm, mode)) != 0)
    471 		return error;
    472 	if (SCARG(uap, size) && SCARG(uap, size) > shmseg->shm_segsz)
    473 		return EINVAL;
    474 	if ((SCARG(uap, shmflg) & (IPC_CREAT | IPC_EXCL)) ==
    475 	    (IPC_CREAT | IPC_EXCL))
    476 		return EEXIST;
    477 	*retval = IXSEQ_TO_IPCID(segnum, shmseg->shm_perm);
    478 	return 0;
    479 }
    480 
    481 static int
    482 shmget_allocate_segment(struct proc *p, struct sys_shmget_args *uap, int mode,
    483     register_t *retval)
    484 {
    485 	int i, segnum, shmid, size;
    486 	kauth_cred_t cred = p->p_cred;
    487 	struct shmid_ds *shmseg;
    488 	int error = 0;
    489 
    490 	if (SCARG(uap, size) < shminfo.shmmin ||
    491 	    SCARG(uap, size) > shminfo.shmmax)
    492 		return EINVAL;
    493 	if (shm_nused >= shminfo.shmmni) /* any shmids left? */
    494 		return ENOSPC;
    495 	size = (SCARG(uap, size) + PGOFSET) & ~PGOFSET;
    496 	if (shm_committed + btoc(size) > shminfo.shmall)
    497 		return ENOMEM;
    498 	if (shm_last_free < 0) {
    499 		for (i = 0; i < shminfo.shmmni; i++)
    500 			if (shmsegs[i].shm_perm.mode & SHMSEG_FREE)
    501 				break;
    502 		if (i == shminfo.shmmni)
    503 			panic("shmseg free count inconsistent");
    504 		segnum = i;
    505 	} else  {
    506 		segnum = shm_last_free;
    507 		shm_last_free = -1;
    508 	}
    509 	shmseg = &shmsegs[segnum];
    510 	/*
    511 	 * In case we sleep in malloc(), mark the segment present but deleted
    512 	 * so that noone else tries to create the same key.
    513 	 */
    514 	shmseg->shm_perm.mode = SHMSEG_ALLOCATED | SHMSEG_REMOVED;
    515 	shmseg->shm_perm._key = SCARG(uap, key);
    516 	shmseg->shm_perm._seq = (shmseg->shm_perm._seq + 1) & 0x7fff;
    517 	shmid = IXSEQ_TO_IPCID(segnum, shmseg->shm_perm);
    518 
    519 	shmseg->_shm_internal = uao_create(size, 0);
    520 
    521 	shmseg->shm_perm.cuid = shmseg->shm_perm.uid = kauth_cred_geteuid(cred);
    522 	shmseg->shm_perm.cgid = shmseg->shm_perm.gid = kauth_cred_getegid(cred);
    523 	shmseg->shm_perm.mode = (shmseg->shm_perm.mode & SHMSEG_WANTED) |
    524 	    (mode & (ACCESSPERMS|SHMSEG_RMLINGER)) | SHMSEG_ALLOCATED;
    525 	shmseg->shm_segsz = SCARG(uap, size);
    526 	shmseg->shm_cpid = p->p_pid;
    527 	shmseg->shm_lpid = shmseg->shm_nattch = 0;
    528 	shmseg->shm_atime = shmseg->shm_dtime = 0;
    529 	shmseg->shm_ctime = time.tv_sec;
    530 	shm_committed += btoc(size);
    531 	shm_nused++;
    532 
    533 	*retval = shmid;
    534 	if (shmseg->shm_perm.mode & SHMSEG_WANTED) {
    535 		/*
    536 		 * Somebody else wanted this key while we were asleep.  Wake
    537 		 * them up now.
    538 		 */
    539 		shmseg->shm_perm.mode &= ~SHMSEG_WANTED;
    540 		wakeup((caddr_t)shmseg);
    541 	}
    542 	return error;
    543 }
    544 
    545 int
    546 sys_shmget(struct lwp *l, void *v, register_t *retval)
    547 {
    548 	struct sys_shmget_args /* {
    549 		syscallarg(key_t) key;
    550 		syscallarg(int) size;
    551 		syscallarg(int) shmflg;
    552 	} */ *uap = v;
    553 	struct proc *p = l->l_proc;
    554 	int segnum, mode, error;
    555 
    556 	mode = SCARG(uap, shmflg) & ACCESSPERMS;
    557 	if (SCARG(uap, shmflg) & _SHM_RMLINGER)
    558 		mode |= SHMSEG_RMLINGER;
    559 
    560 #ifdef SHMDEBUG
    561 	printf("shmget: key 0x%lx size 0x%x shmflg 0x%x mode 0x%x\n",
    562         	SCARG(uap, key), SCARG(uap, size), SCARG(uap, shmflg), mode);
    563 #endif
    564 
    565 	if (SCARG(uap, key) != IPC_PRIVATE) {
    566 again:
    567 		segnum = shm_find_segment_by_key(SCARG(uap, key));
    568 		if (segnum >= 0) {
    569 			error = shmget_existing(p, uap, mode, segnum, retval);
    570 			if (error == EAGAIN)
    571 				goto again;
    572 			return error;
    573 		}
    574 		if ((SCARG(uap, shmflg) & IPC_CREAT) == 0)
    575 			return ENOENT;
    576 	}
    577 	return shmget_allocate_segment(p, uap, mode, retval);
    578 }
    579 
    580 void
    581 shmfork(struct vmspace *vm1, struct vmspace *vm2)
    582 {
    583 	struct shmmap_state *shmmap_s;
    584 	struct shmmap_entry *shmmap_se;
    585 
    586 	vm2->vm_shm = vm1->vm_shm;
    587 
    588 	if (vm1->vm_shm == NULL)
    589 		return;
    590 
    591 #ifdef SHMDEBUG
    592 	printf("shmfork %p->%p\n", vm1, vm2);
    593 #endif
    594 
    595 	shmmap_s = (struct shmmap_state *)vm1->vm_shm;
    596 
    597 	SLIST_FOREACH(shmmap_se, &shmmap_s->entries, next)
    598 		shmsegs[IPCID_TO_IX(shmmap_se->shmid)].shm_nattch++;
    599 	shmmap_s->nrefs++;
    600 }
    601 
    602 void
    603 shmexit(struct vmspace *vm)
    604 {
    605 	struct shmmap_state *shmmap_s;
    606 	struct shmmap_entry *shmmap_se;
    607 
    608 	shmmap_s = (struct shmmap_state *)vm->vm_shm;
    609 	if (shmmap_s == NULL)
    610 		return;
    611 
    612 	vm->vm_shm = NULL;
    613 
    614 	if (--shmmap_s->nrefs > 0) {
    615 #ifdef SHMDEBUG
    616 		printf("shmexit: vm %p drop ref (%d entries), now used by %d\n",
    617 		       vm, shmmap_s->nitems, shmmap_s->nrefs);
    618 #endif
    619 		SLIST_FOREACH(shmmap_se, &shmmap_s->entries, next)
    620 			shmsegs[IPCID_TO_IX(shmmap_se->shmid)].shm_nattch--;
    621 		return;
    622 	}
    623 
    624 #ifdef SHMDEBUG
    625 	printf("shmexit: vm %p cleanup (%d entries)\n", vm, shmmap_s->nitems);
    626 #endif
    627 	while (!SLIST_EMPTY(&shmmap_s->entries)) {
    628 		shmmap_se = SLIST_FIRST(&shmmap_s->entries);
    629 		shm_delete_mapping(vm, shmmap_s, shmmap_se);
    630 	}
    631 	KASSERT(shmmap_s->nitems == 0);
    632 	free(shmmap_s, M_SHM);
    633 }
    634 
    635 void
    636 shminit(void)
    637 {
    638 	int i, sz;
    639 	vaddr_t v;
    640 
    641 	/* Allocate pageable memory for our structures */
    642 	sz = shminfo.shmmni * sizeof(struct shmid_ds);
    643 	v = uvm_km_alloc(kernel_map, round_page(sz), 0, UVM_KMF_WIRED);
    644 	if (v == 0)
    645 		panic("sysv_shm: cannot allocate memory");
    646 	shmsegs = (void *)v;
    647 
    648 	shminfo.shmmax *= PAGE_SIZE;
    649 
    650 	for (i = 0; i < shminfo.shmmni; i++) {
    651 		shmsegs[i].shm_perm.mode = SHMSEG_FREE;
    652 		shmsegs[i].shm_perm._seq = 0;
    653 	}
    654 	shm_last_free = 0;
    655 	shm_nused = 0;
    656 	shm_committed = 0;
    657 }
    658