sys/kern/sysv_shm.c

/*
 * Copyright (c) 1988 University of Utah.
 * Copyright (c) 1990 The Regents of the University of California.
 * All rights reserved.
 *
 * This code is derived from software contributed to Berkeley by
 * the Systems Programming Group of the University of Utah Computer
 * Science Department. Originally from University of Wisconsin.
 *
 * %sccs.include.redist.c%
 *
 * from: Utah $Hdr: uipc_shm.c 1.9 89/08/14$
 *
 *	@(#)sysv_shm.c	7.12 (Berkeley) 03/17/91
 */

/*
 * System V shared memory routines.
 * TEMPORARY, until mmap is in place;
 * needed now for HP-UX compatibility and X server (yech!).
 */

#ifdef SYSVSHM

#include "param.h"
#include "systm.h"
#include "user.h"
#include "kernel.h"
#include "proc.h"
#include "shm.h"
#include "malloc.h"
#include "mman.h"
#include "../vm/vm_param.h"
#include "../vm/vm_map.h"
#include "../vm/vm_kern.h"
#include "../vm/vm_inherit.h"
#include "../vm/vm_pager.h"

#ifdef HPUXCOMPAT
#include "hp300/hpux/hpux.h"
#endif

int	shmat(), shmctl(), shmdt(), shmget();
int	(*shmcalls[])() = { shmat, shmctl, shmdt, shmget };
int	shmtot = 0;

/*
 * Per process internal structure for managing segments.
 * Each process using shm will have an array of ``shmseg'' of these.
 */
struct	shmdesc {
	vm_offset_t	shmd_uva;
	int		shmd_id;
};

/*
 * Per segment internal structure (shm_handle).
 */
struct	shmhandle {
	vm_offset_t	shmh_kva;
	caddr_t		shmh_id;
};

vm_map_t shm_map;	/* address space for shared memory segments */

shminit()
{
	register int i;
	vm_offset_t whocares1, whocares2;

	shm_map = kmem_suballoc(kernel_map, &whocares1, &whocares2,
				shminfo.shmall * NBPG, FALSE);
	if (shminfo.shmmni > SHMMMNI)
		shminfo.shmmni = SHMMMNI;
	for (i = 0; i < shminfo.shmmni; i++) {
		shmsegs[i].shm_perm.mode = 0;
		shmsegs[i].shm_perm.seq = 0;
	}
}

/*
 * Entry point for all SHM calls
 */
shmsys(p, uap, retval)
	struct proc *p;
	struct args {
		u_int which;
	} *uap;
	int *retval;
{

	if (uap->which >= sizeof(shmcalls)/sizeof(shmcalls[0]))
		return (EINVAL);
	return ((*shmcalls[uap->which])(p, &uap[1], retval));
}

/*
 * Get a shared memory segment
 */
shmget(p, uap, retval)
	struct proc *p;
	register struct args {
		key_t key;
		int size;
		int shmflg;
	} *uap;
	int *retval;
{
	register struct shmid_ds *shp;
	register struct ucred *cred = p->p_ucred;
	register int i;
	int error, size, rval = 0;
	register struct shmhandle *shmh;

	/* look up the specified shm_id */
	if (uap->key != IPC_PRIVATE) {
		for (i = 0; i < shminfo.shmmni; i++)
			if ((shmsegs[i].shm_perm.mode & SHM_ALLOC) &&
			    shmsegs[i].shm_perm.key == uap->key) {
				rval = i;
				break;
			}
	} else
		i = shminfo.shmmni;

	/* create a new shared segment if necessary */
	if (i == shminfo.shmmni) {
		if ((uap->shmflg & IPC_CREAT) == 0)
			return (ENOENT);
		if (uap->size < shminfo.shmmin || uap->size > shminfo.shmmax)
			return (EINVAL);
		for (i = 0; i < shminfo.shmmni; i++)
			if ((shmsegs[i].shm_perm.mode & SHM_ALLOC) == 0) {
				rval = i;
				break;
			}
		if (i == shminfo.shmmni)
			return (ENOSPC);
		size = clrnd(btoc(uap->size));
		if (shmtot + size > shminfo.shmall)
			return (ENOMEM);
		shp = &shmsegs[rval];
		/*
		 * We need to do a couple of things to ensure consistency
		 * in case we sleep in malloc().  We mark segment as
		 * allocated so that other shmgets() will not allocate it.
		 * We mark it as "destroyed" to insure that shmvalid() is
		 * false making most operations fail (XXX).  We set the key,
		 * so that other shmget()s will fail.
		 */
		shp->shm_perm.mode = SHM_ALLOC | SHM_DEST;
		shp->shm_perm.key = uap->key;
		shmh = (struct shmhandle *)
			malloc(sizeof(struct shmhandle), M_SHM, M_WAITOK);
		shmh->shmh_kva = 0;
		shmh->shmh_id = (caddr_t)(0xc0000000|rval);	/* XXX */
		error = vm_mmap(shm_map, &shmh->shmh_kva, ctob(size),
				VM_PROT_ALL, MAP_ANON, shmh->shmh_id, 0);
		if (error) {
			free((caddr_t)shmh, M_SHM);
			shp->shm_perm.mode = 0;
			return(ENOMEM);
		}
		shp->shm_handle = (void *) shmh;
		shmtot += size;
		shp->shm_perm.cuid = shp->shm_perm.uid = cred->cr_uid;
		shp->shm_perm.cgid = shp->shm_perm.gid = cred->cr_gid;
		shp->shm_perm.mode = SHM_ALLOC | (uap->shmflg&0777);
		shp->shm_segsz = uap->size;
		shp->shm_cpid = p->p_pid;
		shp->shm_lpid = shp->shm_nattch = 0;
		shp->shm_atime = shp->shm_dtime = 0;
		shp->shm_ctime = time.tv_sec;
	} else {
		shp = &shmsegs[rval];
		/* XXX: probably not the right thing to do */
		if (shp->shm_perm.mode & SHM_DEST)
			return (EBUSY);
		if (error = ipcaccess(&shp->shm_perm, uap->shmflg&0777, cred))
			return (error);
		if (uap->size && uap->size > shp->shm_segsz)
			return (EINVAL);
		if ((uap->shmflg&IPC_CREAT) && (uap->shmflg&IPC_EXCL))
			return (EEXIST);
	}
	*retval = shp->shm_perm.seq * SHMMMNI + rval;
	return (0);
}

/*
 * Shared memory control
 */
/* ARGSUSED */
shmctl(p, uap, retval)
	struct proc *p;
	register struct args {
		int shmid;
		int cmd;
		caddr_t buf;
	} *uap;
	int *retval;
{
	register struct shmid_ds *shp;
	register struct ucred *cred = p->p_ucred;
	struct shmid_ds sbuf;
	int error;

	if (error = shmvalid(uap->shmid))
		return (error);
	shp = &shmsegs[uap->shmid % SHMMMNI];
	switch (uap->cmd) {
	case IPC_STAT:
		if (error = ipcaccess(&shp->shm_perm, IPC_R, cred))
			return (error);
		return (copyout((caddr_t)shp, uap->buf, sizeof(*shp)));

	case IPC_SET:
		if (cred->cr_uid && cred->cr_uid != shp->shm_perm.uid &&
		    cred->cr_uid != shp->shm_perm.cuid)
			return (EPERM);
		if (error = copyin(uap->buf, (caddr_t)&sbuf, sizeof sbuf))
			return (error);
		shp->shm_perm.uid = sbuf.shm_perm.uid;
		shp->shm_perm.gid = sbuf.shm_perm.gid;
		shp->shm_perm.mode = (shp->shm_perm.mode & ~0777)
			| (sbuf.shm_perm.mode & 0777);
		shp->shm_ctime = time.tv_sec;
		break;

	case IPC_RMID:
		if (cred->cr_uid && cred->cr_uid != shp->shm_perm.uid &&
		    cred->cr_uid != shp->shm_perm.cuid)
			return (EPERM);
		/* set ctime? */
		shp->shm_perm.key = IPC_PRIVATE;
		shp->shm_perm.mode |= SHM_DEST;
		if (shp->shm_nattch <= 0)
			shmfree(shp);
		break;

#ifdef HPUXCOMPAT
	case SHM_LOCK:
	case SHM_UNLOCK:
		/* don't really do anything, but make them think we did */
		if ((p->p_flag & SHPUX) == 0)
			return (EINVAL);
		if (cred->cr_uid && cred->cr_uid != shp->shm_perm.uid &&
		    cred->cr_uid != shp->shm_perm.cuid)
			return (EPERM);
		break;
#endif

	default:
		return (EINVAL);
	}
	return (0);
}

/*
 * Attach to shared memory segment.
 */
shmat(p, uap, retval)
	struct proc *p;
	register struct args {
		int	shmid;
		caddr_t	shmaddr;
		int	shmflg;
	} *uap;
	int *retval;
{
	register struct shmid_ds *shp;
	register int size;
	caddr_t uva;
	int error;
	int flags;
	vm_prot_t prot;
	struct shmdesc *shmd;

	/*
	 * Allocate descriptors now (before validity check)
	 * in case malloc() blocks.
	 */
	shmd = (struct shmdesc *)p->p_vmspace->vm_shm;
	size = shminfo.shmseg * sizeof(struct shmdesc);
	if (shmd == NULL) {
		shmd = (struct shmdesc *)malloc(size, M_SHM, M_WAITOK);
		bzero((caddr_t)shmd, size);
		p->p_vmspace->vm_shm = (caddr_t)shmd;
	}
	if (error = shmvalid(uap->shmid))
		return (error);
	shp = &shmsegs[uap->shmid % SHMMMNI];
	if (shp->shm_handle == NULL)
		panic("shmat NULL handle");
	if (error = ipcaccess(&shp->shm_perm,
	    (uap->shmflg&SHM_RDONLY) ? IPC_R : IPC_R|IPC_W, p->p_ucred))
		return (error);
	uva = uap->shmaddr;
	if (uva && ((int)uva & (SHMLBA-1))) {
		if (uap->shmflg & SHM_RND)
			uva = (caddr_t) ((int)uva & ~(SHMLBA-1));
		else
			return (EINVAL);
	}
	/*
	 * Make sure user doesn't use more than their fair share
	 */
	for (size = 0; size < shminfo.shmseg; size++) {
		if (shmd->shmd_uva == 0)
			break;
		shmd++;
	}
	if (size >= shminfo.shmseg)
		return (EMFILE);
	size = ctob(clrnd(btoc(shp->shm_segsz)));
	prot = VM_PROT_READ;
	if ((uap->shmflg & SHM_RDONLY) == 0)
		prot |= VM_PROT_WRITE;
	flags = MAP_ANON|MAP_SHARED;
	if (uva)
		flags |= MAP_FIXED;
	else
		uva = (caddr_t)0x1000000;	/* XXX */
	error = vm_mmap(p->p_vmspace->vm_map, &uva, (vm_size_t)size, prot,
	    flags, ((struct shmhandle *)shp->shm_handle)->shmh_id, 0);
	if (error)
		return(error);
	shmd->shmd_uva = (vm_offset_t)uva;
	shmd->shmd_id = uap->shmid;
	/*
	 * Fill in the remaining fields
	 */
	shp->shm_lpid = p->p_pid;
	shp->shm_atime = time.tv_sec;
	shp->shm_nattch++;
	*retval = (int) uva;
	return (0);
}

/*
 * Detach from shared memory segment.
 */
/* ARGSUSED */
shmdt(p, uap, retval)
	struct proc *p;
	struct args {
		caddr_t	shmaddr;
	} *uap;
	int *retval;
{
	register struct shmdesc *shmd;
	register int i;

	shmd = (struct shmdesc *)p->p_vmspace->vm_shm;
	for (i = 0; i < shminfo.shmseg; i++, shmd++)
		if (shmd->shmd_uva &&
		    shmd->shmd_uva == (vm_offset_t)uap->shmaddr)
			break;
	if (i == shminfo.shmseg)
		return(EINVAL);
	shmufree(p, shmd);
	shmsegs[shmd->shmd_id % SHMMMNI].shm_lpid = p->p_pid;
}

shmfork(p1, p2, isvfork)
	struct proc *p1, *p2;
	int isvfork;
{
	register struct shmdesc *shmd;
	register int size;

	/*
	 * Copy parents descriptive information
	 */
	size = shminfo.shmseg * sizeof(struct shmdesc);
	shmd = (struct shmdesc *)malloc(size, M_SHM, M_WAITOK);
	bcopy((caddr_t)p1->p_vmspace->vm_shm, (caddr_t)shmd, size);
	p2->p_vmspace->vm_shm = (caddr_t)shmd;
	/*
	 * Increment reference counts
	 */
	for (size = 0; size < shminfo.shmseg; size++, shmd++)
		if (shmd->shmd_uva)
			shmsegs[shmd->shmd_id % SHMMMNI].shm_nattch++;
}

shmexit(p)
	struct proc *p;
{
	register struct shmdesc *shmd;
	register int i;

	shmd = (struct shmdesc *)p->p_vmspace->vm_shm;
	for (i = 0; i < shminfo.shmseg; i++, shmd++)
		if (shmd->shmd_uva)
			shmufree(p, shmd);
	free((caddr_t)p->p_vmspace->vm_shm, M_SHM);
	p->p_vmspace->vm_shm = NULL;
}

shmvalid(id)
	register int id;
{
	register struct shmid_ds *shp;

	if (id < 0 || (id % SHMMMNI) >= shminfo.shmmni)
		return(EINVAL);
	shp = &shmsegs[id % SHMMMNI];
	if (shp->shm_perm.seq == (id / SHMMMNI) &&
	    (shp->shm_perm.mode & (SHM_ALLOC|SHM_DEST)) == SHM_ALLOC)
		return(0);
	return(EINVAL);
}

/*
 * Free user resources associated with a shared memory segment
 */
shmufree(p, shmd)
	struct proc *p;
	struct shmdesc *shmd;
{
	register struct shmid_ds *shp;

	shp = &shmsegs[shmd->shmd_id % SHMMMNI];
	(void) vm_deallocate(p->p_vmspace->vm_map, shmd->shmd_uva,
			     ctob(clrnd(btoc(shp->shm_segsz))));
	shmd->shmd_id = 0;
	shmd->shmd_uva = 0;
	shp->shm_dtime = time.tv_sec;
	if (--shp->shm_nattch <= 0 && (shp->shm_perm.mode & SHM_DEST))
		shmfree(shp);
}

/*
 * Deallocate resources associated with a shared memory segment
 */
shmfree(shp)
	register struct shmid_ds *shp;
{
	caddr_t kva;

	if (shp->shm_handle == NULL)
		panic("shmfree");
	/*
	 * Lose our lingering object reference by deallocating space
	 * in kernel.  Pager will also be deallocated as a side-effect.
	 */
	vm_deallocate(shm_map,
		      ((struct shmhandle *)shp->shm_handle)->shmh_kva,
		      clrnd(btoc(shp->shm_segsz)));
	free((caddr_t)shp->shm_handle, M_SHM);
	shp->shm_handle = NULL;
	shmtot -= clrnd(btoc(shp->shm_segsz));
	shp->shm_perm.mode = 0;
	/*
	 * Increment the sequence number to ensure that outstanding
	 * shmids for this segment will be invalid in the event that
	 * the segment is reallocated.  Note that shmids must be
	 * positive as decreed by SVID.
	 */
	shp->shm_perm.seq++;
	if ((int)(shp->shm_perm.seq * SHMMMNI) < 0)
		shp->shm_perm.seq = 0;
}

/*
 * XXX This routine would be common to all sysV style IPC
 *     (if the others were implemented).
 */
ipcaccess(ipc, mode, cred)
	register struct ipc_perm *ipc;
	int mode;
	register struct ucred *cred;
{
	register int m;

	if (cred->cr_uid == 0)
		return(0);
	/*
	 * Access check is based on only one of owner, group, public.
	 * If not owner, then check group.
	 * If not a member of the group, then check public access.
	 */
	mode &= 0700;
	m = ipc->mode;
	if (cred->cr_uid != ipc->uid && cred->cr_uid != ipc->cuid) {
		m <<= 3;
		if (!groupmember(ipc->gid, cred) &&
		    !groupmember(ipc->cgid, cred))
			m <<= 3;
	}
	if ((mode&m) == mode)
		return (0);
	return (EACCES);
}
#endif /* SYSVSHM */