Home | History | Annotate | Line # | Download | only in kern
sysv_shm.c revision 1.31
      1 /*	$NetBSD: sysv_shm.c,v 1.31 1995/06/29 11:43:17 cgd 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 /*
     54  * Provides the following externally accessible functions:
     55  *
     56  * shminit(void);		           initialization
     57  * shmexit(struct proc *)                  cleanup
     58  * shmfork(struct proc *, struct proc *, int) fork handling
     59  * shmsys(arg1, arg2, arg3, arg4);         shm{at,ctl,dt,get}(arg2, arg3, arg4)
     60  *
     61  * Structures:
     62  * shmsegs (an array of 'struct shmid_ds')
     63  * per proc array of 'struct shmmap_state'
     64  */
     65 
     66 #define	SHMSEG_FREE     	0x0200
     67 #define	SHMSEG_REMOVED  	0x0400
     68 #define	SHMSEG_ALLOCATED	0x0800
     69 #define	SHMSEG_WANTED		0x1000
     70 
     71 vm_map_t sysvshm_map;
     72 int shm_last_free, shm_nused, shm_committed;
     73 
     74 struct shm_handle {
     75 	vm_offset_t kva;
     76 };
     77 
     78 struct shmmap_state {
     79 	vm_offset_t va;
     80 	int shmid;
     81 };
     82 
     83 static void shm_deallocate_segment __P((struct shmid_ds *));
     84 static int shm_find_segment_by_key __P((key_t));
     85 struct shmid_ds *shm_find_segment_by_shmid __P((int));
     86 static int shm_delete_mapping __P((struct proc *, struct shmmap_state *));
     87 
     88 static int
     89 shm_find_segment_by_key(key)
     90 	key_t key;
     91 {
     92 	int i;
     93 
     94 	for (i = 0; i < shminfo.shmmni; i++)
     95 		if ((shmsegs[i].shm_perm.mode & SHMSEG_ALLOCATED) &&
     96 		    shmsegs[i].shm_perm.key == key)
     97 			return i;
     98 	return -1;
     99 }
    100 
    101 struct shmid_ds *
    102 shm_find_segment_by_shmid(shmid)
    103 	int shmid;
    104 {
    105 	int segnum;
    106 	struct shmid_ds *shmseg;
    107 
    108 	segnum = IPCID_TO_IX(shmid);
    109 	if (segnum < 0 || segnum >= shminfo.shmmni)
    110 		return NULL;
    111 	shmseg = &shmsegs[segnum];
    112 	if ((shmseg->shm_perm.mode & (SHMSEG_ALLOCATED | SHMSEG_REMOVED))
    113 	    != SHMSEG_ALLOCATED ||
    114 	    shmseg->shm_perm.seq != IPCID_TO_SEQ(shmid))
    115 		return NULL;
    116 	return shmseg;
    117 }
    118 
    119 static void
    120 shm_deallocate_segment(shmseg)
    121 	struct shmid_ds *shmseg;
    122 {
    123 	struct shm_handle *shm_handle;
    124 	size_t size;
    125 
    126 	shm_handle = shmseg->shm_internal;
    127 	size = (shmseg->shm_segsz + CLOFSET) & ~CLOFSET;
    128 	vm_deallocate(sysvshm_map, shm_handle->kva, size);
    129 	free((caddr_t)shm_handle, M_SHM);
    130 	shmseg->shm_internal = NULL;
    131 	shm_committed -= btoc(size);
    132 	shmseg->shm_perm.mode = SHMSEG_FREE;
    133 	shm_nused--;
    134 }
    135 
    136 static int
    137 shm_delete_mapping(p, shmmap_s)
    138 	struct proc *p;
    139 	struct shmmap_state *shmmap_s;
    140 {
    141 	struct shmid_ds *shmseg;
    142 	int segnum, result;
    143 	size_t size;
    144 
    145 	segnum = IPCID_TO_IX(shmmap_s->shmid);
    146 	shmseg = &shmsegs[segnum];
    147 	size = (shmseg->shm_segsz + CLOFSET) & ~CLOFSET;
    148 	result = vm_deallocate(&p->p_vmspace->vm_map, shmmap_s->va, size);
    149 	if (result != KERN_SUCCESS)
    150 		return EINVAL;
    151 	shmmap_s->shmid = -1;
    152 	shmseg->shm_dtime = time.tv_sec;
    153 	if ((--shmseg->shm_nattch <= 0) &&
    154 	    (shmseg->shm_perm.mode & SHMSEG_REMOVED)) {
    155 		shm_deallocate_segment(shmseg);
    156 		shm_last_free = segnum;
    157 	}
    158 	return 0;
    159 }
    160 
    161 int
    162 shmdt(p, uap, retval)
    163 	struct proc *p;
    164 	struct shmdt_args /* {
    165 		syscallarg(void *) shmaddr;
    166 	} */ *uap;
    167 	register_t *retval;
    168 {
    169 	struct shmmap_state *shmmap_s;
    170 	int i;
    171 
    172 	shmmap_s = (struct shmmap_state *)p->p_vmspace->vm_shm;
    173 	for (i = 0; i < shminfo.shmseg; i++, shmmap_s++)
    174 		if (shmmap_s->shmid != -1 &&
    175 		    shmmap_s->va == (vm_offset_t)SCARG(uap, shmaddr))
    176 			break;
    177 	if (i == shminfo.shmseg)
    178 		return EINVAL;
    179 	return shm_delete_mapping(p, shmmap_s);
    180 }
    181 
    182 int
    183 shmat(p, uap, retval)
    184 	struct proc *p;
    185 	struct shmat_args /* {
    186 		syscallarg(int) shmid;
    187 		syscallarg(void *) shmaddr;
    188 		syscallarg(int) shmflag;
    189 	} */ *uap;
    190 	register_t *retval;
    191 {
    192 	int error, i, flags;
    193 	struct ucred *cred = p->p_ucred;
    194 	struct shmid_ds *shmseg;
    195 	struct shmmap_state *shmmap_s = NULL;
    196 	vm_offset_t attach_va;
    197 	vm_prot_t prot;
    198 	vm_size_t size;
    199 
    200 	shmmap_s = (struct shmmap_state *)p->p_vmspace->vm_shm;
    201 	if (shmmap_s == NULL) {
    202 		size = shminfo.shmseg * sizeof(struct shmmap_state);
    203 		shmmap_s = malloc(size, M_SHM, M_WAITOK);
    204 		for (i = 0; i < shminfo.shmseg; i++)
    205 			shmmap_s[i].shmid = -1;
    206 		p->p_vmspace->vm_shm = (caddr_t)shmmap_s;
    207 	}
    208 	shmseg = shm_find_segment_by_shmid(SCARG(uap, shmid));
    209 	if (shmseg == NULL)
    210 		return EINVAL;
    211 	if (error = ipcperm(cred, &shmseg->shm_perm,
    212 	    (SCARG(uap, shmflg) & SHM_RDONLY) ? IPC_R : IPC_R|IPC_W))
    213 		return error;
    214 	for (i = 0; i < shminfo.shmseg; i++) {
    215 		if (shmmap_s->shmid == -1)
    216 			break;
    217 		shmmap_s++;
    218 	}
    219 	if (i >= shminfo.shmseg)
    220 		return EMFILE;
    221 	size = (shmseg->shm_segsz + CLOFSET) & ~CLOFSET;
    222 	prot = VM_PROT_READ;
    223 	if ((SCARG(uap, shmflg) & SHM_RDONLY) == 0)
    224 		prot |= VM_PROT_WRITE;
    225 	flags = MAP_ANON | MAP_SHARED;
    226 	if (SCARG(uap, shmaddr)) {
    227 		flags |= MAP_FIXED;
    228 		if (SCARG(uap, shmflg) & SHM_RND)
    229 			attach_va =
    230 			    (vm_offset_t)SCARG(uap, shmaddr) & ~(SHMLBA-1);
    231 		else if (((vm_offset_t)SCARG(uap, shmaddr) & (SHMLBA-1)) == 0)
    232 			attach_va = (vm_offset_t)SCARG(uap, shmaddr);
    233 		else
    234 			return EINVAL;
    235 	} else {
    236 		/* This is just a hint to vm_mmap() about where to put it. */
    237 		attach_va =
    238 		    round_page(p->p_vmspace->vm_taddr + MAXTSIZ + MAXDSIZ);
    239 	}
    240 	error = vm_mmap(&p->p_vmspace->vm_map, &attach_va, size, prot,
    241 	    VM_PROT_DEFAULT, flags, (caddr_t)(long)SCARG(uap, shmid), 0);
    242 	if (error)
    243 		return error;
    244 	shmmap_s->va = attach_va;
    245 	shmmap_s->shmid = SCARG(uap, shmid);
    246 	shmseg->shm_lpid = p->p_pid;
    247 	shmseg->shm_atime = time.tv_sec;
    248 	shmseg->shm_nattch++;
    249 	*retval = attach_va;
    250 	return 0;
    251 }
    252 
    253 int
    254 shmctl(p, uap, retval)
    255 	struct proc *p;
    256 	struct shmctl_args /* {
    257 		syscallarg(int) shmid;
    258 		syscallarg(int) cmd;
    259 		syscallarg(struct shmid_ds *) buf;
    260 	} */ *uap;
    261 	register_t *retval;
    262 {
    263 	int error, segnum;
    264 	struct ucred *cred = p->p_ucred;
    265 	struct shmid_ds inbuf;
    266 	struct shmid_ds *shmseg;
    267 
    268 	shmseg = shm_find_segment_by_shmid(SCARG(uap, shmid));
    269 	if (shmseg == NULL)
    270 		return EINVAL;
    271 	switch (SCARG(uap, cmd)) {
    272 	case IPC_STAT:
    273 		if (error = ipcperm(cred, &shmseg->shm_perm, IPC_R))
    274 			return error;
    275 		if (error = copyout((caddr_t)shmseg, SCARG(uap, buf),
    276 		    sizeof(inbuf)))
    277 			return error;
    278 		break;
    279 	case IPC_SET:
    280 		if (error = ipcperm(cred, &shmseg->shm_perm, IPC_M))
    281 			return error;
    282 		if (error = copyin(SCARG(uap, buf), (caddr_t)&inbuf,
    283 		    sizeof(inbuf)))
    284 			return error;
    285 		shmseg->shm_perm.uid = inbuf.shm_perm.uid;
    286 		shmseg->shm_perm.gid = inbuf.shm_perm.gid;
    287 		shmseg->shm_perm.mode =
    288 		    (shmseg->shm_perm.mode & ~ACCESSPERMS) |
    289 		    (inbuf.shm_perm.mode & ACCESSPERMS);
    290 		shmseg->shm_ctime = time.tv_sec;
    291 		break;
    292 	case IPC_RMID:
    293 		if (error = ipcperm(cred, &shmseg->shm_perm, IPC_M))
    294 			return error;
    295 		shmseg->shm_perm.key = IPC_PRIVATE;
    296 		shmseg->shm_perm.mode |= SHMSEG_REMOVED;
    297 		if (shmseg->shm_nattch <= 0) {
    298 			shm_deallocate_segment(shmseg);
    299 			shm_last_free = IPCID_TO_IX(SCARG(uap, shmid));
    300 		}
    301 		break;
    302 	case SHM_LOCK:
    303 	case SHM_UNLOCK:
    304 	default:
    305 		return EINVAL;
    306 	}
    307 	return 0;
    308 }
    309 
    310 static int
    311 shmget_existing(p, uap, mode, segnum, retval)
    312 	struct proc *p;
    313 	struct shmget_args /* {
    314 		syscallarg(key_t) key;
    315 		syscallarg(int) size;
    316 		syscallarg(int) shmflag;
    317 	} */ *uap;
    318 	int mode;
    319 	int segnum;
    320 	register_t *retval;
    321 {
    322 	struct shmid_ds *shmseg;
    323 	struct ucred *cred = p->p_ucred;
    324 	int error;
    325 
    326 	shmseg = &shmsegs[segnum];
    327 	if (shmseg->shm_perm.mode & SHMSEG_REMOVED) {
    328 		/*
    329 		 * This segment is in the process of being allocated.  Wait
    330 		 * until it's done, and look the key up again (in case the
    331 		 * allocation failed or it was freed).
    332 		 */
    333 		shmseg->shm_perm.mode |= SHMSEG_WANTED;
    334 		if (error =
    335 		    tsleep((caddr_t)shmseg, PLOCK | PCATCH, "shmget", 0))
    336 			return error;
    337 		return EAGAIN;
    338 	}
    339 	if (error = ipcperm(cred, &shmseg->shm_perm, mode))
    340 		return error;
    341 	if (SCARG(uap, size) && SCARG(uap, size) > shmseg->shm_segsz)
    342 		return EINVAL;
    343 	if (SCARG(uap, shmflg) & (IPC_CREAT | IPC_EXCL) ==
    344 	    (IPC_CREAT | IPC_EXCL))
    345 		return EEXIST;
    346 	*retval = IXSEQ_TO_IPCID(segnum, shmseg->shm_perm);
    347 	return 0;
    348 }
    349 
    350 static int
    351 shmget_allocate_segment(p, uap, mode, retval)
    352 	struct proc *p;
    353 	struct shmget_args /* {
    354 		syscallarg(key_t) key;
    355 		syscallarg(int) size;
    356 		syscallarg(int) shmflag;
    357 	} */ *uap;
    358 	int mode;
    359 	register_t *retval;
    360 {
    361 	int i, segnum, result, shmid, size;
    362 	struct ucred *cred = p->p_ucred;
    363 	struct shmid_ds *shmseg;
    364 	struct shm_handle *shm_handle;
    365 
    366 	if (SCARG(uap, size) < shminfo.shmmin ||
    367 	    SCARG(uap, size) > shminfo.shmmax)
    368 		return EINVAL;
    369 	if (shm_nused >= shminfo.shmmni) /* any shmids left? */
    370 		return ENOSPC;
    371 	size = (SCARG(uap, size) + CLOFSET) & ~CLOFSET;
    372 	if (shm_committed + btoc(size) > shminfo.shmall)
    373 		return ENOMEM;
    374 	if (shm_last_free < 0) {
    375 		for (i = 0; i < shminfo.shmmni; i++)
    376 			if (shmsegs[i].shm_perm.mode & SHMSEG_FREE)
    377 				break;
    378 		if (i == shminfo.shmmni)
    379 			panic("shmseg free count inconsistent");
    380 		segnum = i;
    381 	} else  {
    382 		segnum = shm_last_free;
    383 		shm_last_free = -1;
    384 	}
    385 	shmseg = &shmsegs[segnum];
    386 	/*
    387 	 * In case we sleep in malloc(), mark the segment present but deleted
    388 	 * so that noone else tries to create the same key.
    389 	 */
    390 	shmseg->shm_perm.mode = SHMSEG_ALLOCATED | SHMSEG_REMOVED;
    391 	shmseg->shm_perm.key = SCARG(uap, key);
    392 	shmseg->shm_perm.seq = (shmseg->shm_perm.seq + 1) & 0x7fff;
    393 	shm_handle = (struct shm_handle *)
    394 	    malloc(sizeof(struct shm_handle), M_SHM, M_WAITOK);
    395 	shmid = IXSEQ_TO_IPCID(segnum, shmseg->shm_perm);
    396 	result = vm_mmap(sysvshm_map, &shm_handle->kva, size, VM_PROT_ALL,
    397 	    VM_PROT_DEFAULT, MAP_ANON, (caddr_t)(long)shmid, 0);
    398 	if (result != KERN_SUCCESS) {
    399 		shmseg->shm_perm.mode = SHMSEG_FREE;
    400 		shm_last_free = segnum;
    401 		free((caddr_t)shm_handle, M_SHM);
    402 		/* Just in case. */
    403 		wakeup((caddr_t)shmseg);
    404 		return ENOMEM;
    405 	}
    406 	shmseg->shm_internal = shm_handle;
    407 	shmseg->shm_perm.cuid = shmseg->shm_perm.uid = cred->cr_uid;
    408 	shmseg->shm_perm.cgid = shmseg->shm_perm.gid = cred->cr_gid;
    409 	shmseg->shm_perm.mode = (shmseg->shm_perm.mode & SHMSEG_WANTED) |
    410 	    (mode & ACCESSPERMS) | SHMSEG_ALLOCATED;
    411 	shmseg->shm_segsz = SCARG(uap, size);
    412 	shmseg->shm_cpid = p->p_pid;
    413 	shmseg->shm_lpid = shmseg->shm_nattch = 0;
    414 	shmseg->shm_atime = shmseg->shm_dtime = 0;
    415 	shmseg->shm_ctime = time.tv_sec;
    416 	shm_committed += btoc(size);
    417 	shm_nused++;
    418 	if (shmseg->shm_perm.mode & SHMSEG_WANTED) {
    419 		/*
    420 		 * Somebody else wanted this key while we were asleep.  Wake
    421 		 * them up now.
    422 		 */
    423 		shmseg->shm_perm.mode &= ~SHMSEG_WANTED;
    424 		wakeup((caddr_t)shmseg);
    425 	}
    426 	*retval = shmid;
    427 	return 0;
    428 }
    429 
    430 int
    431 shmget(p, uap, retval)
    432 	struct proc *p;
    433 	struct shmget_args /* {
    434 		syscallarg(key_t) key;
    435 		syscallarg(int) size;
    436 		syscallarg(int) shmflag;
    437 	} */ *uap;
    438 	register_t *retval;
    439 {
    440 	int segnum, mode, error;
    441 	struct shmid_ds *shmseg;
    442 
    443 	mode = SCARG(uap, shmflg) & ACCESSPERMS;
    444 	if (SCARG(uap, key) != IPC_PRIVATE) {
    445 	again:
    446 		segnum = shm_find_segment_by_key(SCARG(uap, key));
    447 		if (segnum >= 0) {
    448 			error = shmget_existing(p, uap, mode, segnum, retval);
    449 			if (error == EAGAIN)
    450 				goto again;
    451 			return error;
    452 		}
    453 		if ((SCARG(uap, shmflg) & IPC_CREAT) == 0)
    454 			return ENOENT;
    455 	}
    456 	return shmget_allocate_segment(p, uap, mode, retval);
    457 }
    458 
    459 void
    460 shmfork(p1, p2, isvfork)
    461 	struct proc *p1, *p2;
    462 	int isvfork;
    463 {
    464 	struct shmmap_state *shmmap_s;
    465 	size_t size;
    466 	int i;
    467 
    468 	size = shminfo.shmseg * sizeof(struct shmmap_state);
    469 	shmmap_s = malloc(size, M_SHM, M_WAITOK);
    470 	bcopy((caddr_t)p1->p_vmspace->vm_shm, (caddr_t)shmmap_s, size);
    471 	p2->p_vmspace->vm_shm = (caddr_t)shmmap_s;
    472 	for (i = 0; i < shminfo.shmseg; i++, shmmap_s++)
    473 		if (shmmap_s->shmid != -1)
    474 			shmsegs[IPCID_TO_IX(shmmap_s->shmid)].shm_nattch++;
    475 }
    476 
    477 void
    478 shmexit(p)
    479 	struct proc *p;
    480 {
    481 	struct shmmap_state *shmmap_s;
    482 	struct shmid_ds *shmseg;
    483 	int i;
    484 
    485 	shmmap_s = (struct shmmap_state *)p->p_vmspace->vm_shm;
    486 	for (i = 0; i < shminfo.shmseg; i++, shmmap_s++)
    487 		if (shmmap_s->shmid != -1)
    488 			shm_delete_mapping(p, shmmap_s);
    489 	free((caddr_t)p->p_vmspace->vm_shm, M_SHM);
    490 	p->p_vmspace->vm_shm = NULL;
    491 }
    492 
    493 void
    494 shminit()
    495 {
    496 	int i;
    497 	vm_offset_t garbage1, garbage2;
    498 
    499 	shminfo.shmmax *= NBPG;
    500 
    501 	/* actually this *should* be pageable.  SHM_{LOCK,UNLOCK} */
    502 	sysvshm_map = kmem_suballoc(kernel_map, &garbage1, &garbage2,
    503 				    shminfo.shmall * NBPG, TRUE);
    504 	for (i = 0; i < shminfo.shmmni; i++) {
    505 		shmsegs[i].shm_perm.mode = SHMSEG_FREE;
    506 		shmsegs[i].shm_perm.seq = 0;
    507 	}
    508 	shm_last_free = 0;
    509 	shm_nused = 0;
    510 	shm_committed = 0;
    511 }
    512