sys/vm/vm_mmap.c

/*
 * Copyright (c) 1988 University of Utah.
 * Copyright (c) 1991 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.
 *
 * %sccs.include.redist.c%
 *
 * from: Utah $Hdr: vm_mmap.c 1.3 90/01/21$
 *
 *	@(#)vm_mmap.c	7.2 (Berkeley) 01/10/91
 */

/*
 * Mapped file (mmap) interface to VM
 */

#include "param.h"
#include "systm.h"
#include "user.h"
#include "filedesc.h"
#include "proc.h"
#include "vnode.h"
#include "specdev.h"
#include "file.h"
#include "mman.h"
#include "conf.h"

#include "../vm/vm_param.h"
#include "../vm/vm_map.h"
#include "../vm/vm_pager.h"
#include "../vm/vm_prot.h"

#ifdef DEBUG
int mmapdebug = 0;
#define MDB_FOLLOW	0x01
#define MDB_SYNC	0x02
#define MDB_MAPIT	0x04
#endif

/* ARGSUSED */
getpagesize(p, uap, retval)
	struct proc *p;
	struct args *uap;
	int *retval;
{

	*retval = NBPG * CLSIZE;
	return (0);
}

/* ARGSUSED */
sbrk(p, uap, retval)
	struct proc *p;
	struct args {
		int	incr;
	} *uap;
	int *retval;
{

	/* Not yet implemented */
	return (EOPNOTSUPP);
}

/* ARGSUSED */
sstk(p, uap, retval)
	struct proc *p;
	struct args {
		int	incr;
	} *uap;
	int *retval;
{

	/* Not yet implemented */
	return (EOPNOTSUPP);
}

smmap(p, uap, retval)
	struct proc *p;
	register struct args {
		caddr_t	addr;
		int	len;
		int	prot;
		int	flags;
		int	fd;
		off_t	pos;
	} *uap;
	int *retval;
{
	register struct filedesc *fdp = p->p_fd;
	register struct file *fp;
	struct vnode *vp;
	vm_offset_t addr;
	vm_size_t size;
	vm_prot_t prot;
	caddr_t handle;
	int mtype, error;

#ifdef DEBUG
	if (mmapdebug & MDB_FOLLOW)
		printf("mmap(%d): addr %x len %x pro %x flg %x fd %d pos %x\n",
		       p->p_pid, uap->addr, uap->len, uap->prot,
		       uap->flags, uap->fd, uap->pos);
#endif
	/*
	 * Make sure one of the sharing types is specified
	 */
	mtype = uap->flags & MAP_TYPE;
	switch (mtype) {
	case MAP_FILE:
	case MAP_ANON:
		break;
	default:
		return(EINVAL);
	}
	/*
	 * Address (if FIXED) and size must be page aligned
	 */
	size = (vm_size_t)uap->len;
	addr = (vm_offset_t)uap->addr;
	if ((size & page_mask) ||
	    (uap->flags & MAP_FIXED) && (addr & page_mask))
		return(EINVAL);
	/*
	 * Mapping file or named anonymous, get fp for validation
	 */
	if (mtype == MAP_FILE || uap->fd != -1) {
		if (((unsigned)uap->fd) >= fdp->fd_maxfiles ||
		    (fp = OFILE(fdp, uap->fd)) == NULL)
			return(EBADF);
	}
	/*
	 * If we are mapping a file we need to check various
	 * file/vnode related things.
	 */
	if (mtype == MAP_FILE) {
		/*
		 * Obtain vnode and make sure it is of appropriate type
		 */
		if (fp->f_type != DTYPE_VNODE)
			return(EINVAL);
		vp = (struct vnode *)fp->f_data;
		if (vp->v_type != VREG && vp->v_type != VCHR)
			return(EINVAL);
		/*
		 * Ensure that file protection and desired protection
		 * are compatible.  Note that we only worry about writability
		 * if mapping is shared.
		 */
		if ((uap->prot & PROT_READ) && (fp->f_flag & FREAD) == 0 ||
		    ((uap->flags & MAP_SHARED) &&
		     (uap->prot & PROT_WRITE) && (fp->f_flag & FWRITE) == 0))
			return(EACCES);
		handle = (caddr_t)vp;
	} else if (uap->fd != -1)
		handle = (caddr_t)fp;
	else
		handle = NULL;
	/*
	 * Map protections to MACH style
	 */
	prot = VM_PROT_NONE;
	if (uap->prot & PROT_READ)
		prot |= VM_PROT_READ;
	if (uap->prot & PROT_WRITE)
		prot |= VM_PROT_WRITE;
	if (uap->prot & PROT_EXEC)
		prot |= VM_PROT_EXECUTE;

	error = vm_mmap(p->p_map, &addr, size, prot,
			uap->flags, handle, (vm_offset_t)uap->pos);
	if (error == 0)
		*retval = (int) addr;
	return(error);
}

msync(p, uap, retval)
	struct proc *p;
	struct args {
		char	*addr;
		int	len;
	} *uap;
	int *retval;
{
	vm_offset_t addr, objoff, oaddr;
	vm_size_t size, osize;
	vm_prot_t prot, mprot;
	vm_inherit_t inherit;
	vm_object_t object;
	boolean_t shared;
	int rv;

#ifdef DEBUG
	if (mmapdebug & (MDB_FOLLOW|MDB_SYNC))
		printf("msync(%d): addr %x len %x\n",
		       p->p_pid, uap->addr, uap->len);
#endif
	if (((int)uap->addr & page_mask) || (uap->len & page_mask))
		return(EINVAL);
	addr = oaddr = (vm_offset_t)uap->addr;
	osize = (vm_size_t)uap->len;
	/*
	 * Region must be entirely contained in a single entry
	 */
	if (!vm_map_is_allocated(p->p_map, addr, addr+osize, TRUE))
		return(EINVAL);
	/*
	 * Determine the object associated with that entry
	 * (object is returned locked on KERN_SUCCESS)
	 */
	rv = vm_region(p->p_map, &addr, &size, &prot, &mprot,
		       &inherit, &shared, &object, &objoff);
	if (rv != KERN_SUCCESS)
		return(EINVAL);
#ifdef DEBUG
	if (mmapdebug & MDB_SYNC)
		printf("msync: region: object %x addr %x size %d objoff %d\n",
		       object, addr, size, objoff);
#endif
	/*
	 * Do not msync non-vnoded backed objects.
	 */
	if (object->internal || object->pager == vm_pager_null ||
	    object->pager->pg_type != PG_VNODE) {
		vm_object_unlock(object);
		return(EINVAL);
	}
	objoff += oaddr - addr;
	if (osize == 0)
		osize = size;
#ifdef DEBUG
	if (mmapdebug & MDB_SYNC)
		printf("msync: cleaning/flushing object range [%x-%x)\n",
		       objoff, objoff+osize);
#endif
	if (prot & VM_PROT_WRITE)
		vm_object_page_clean(object, objoff, objoff+osize);
	/*
	 * (XXX)
	 * Bummer, gotta flush all cached pages to ensure
	 * consistency with the file system cache.
	 */
	vm_object_page_remove(object, objoff, objoff+osize);
	vm_object_unlock(object);
	return(0);
}

munmap(p, uap, retval)
	register struct proc *p;
	register struct args {
		caddr_t	addr;
		int	len;
	} *uap;
	int *retval;
{
	vm_offset_t addr;
	vm_size_t size;

#ifdef DEBUG
	if (mmapdebug & MDB_FOLLOW)
		printf("munmap(%d): addr %x len %x\n",
		       p->p_pid, uap->addr, uap->len);
#endif

	addr = (vm_offset_t) uap->addr;
	size = (vm_size_t) uap->len;
	if ((addr & page_mask) || (size & page_mask))
		return(EINVAL);
	if (size == 0)
		return(0);
	if (!vm_map_is_allocated(p->p_map, addr, addr+size, FALSE))
		return(EINVAL);
	/* returns nothing but KERN_SUCCESS anyway */
	(void) vm_map_remove(p->p_map, addr, addr+size);
	return(0);
}

munmapfd(fd)
{
#ifdef DEBUG
	if (mmapdebug & MDB_FOLLOW)
		printf("munmapfd(%d): fd %d\n", u.u_procp->p_pid, fd);
#endif

	/*
	 * XXX -- should vm_deallocate any regions mapped to this file
	 */
	OFILEFLAGS(u.u_procp->p_fd, fd) &= ~UF_MAPPED;
}

mprotect(p, uap, retval)
	struct proc *p;
	struct args {
		char	*addr;
		int	len;
		int	prot;
	} *uap;
	int *retval;
{
	vm_offset_t addr;
	vm_size_t size;
	register vm_prot_t prot;

#ifdef DEBUG
	if (mmapdebug & MDB_FOLLOW)
		printf("mprotect(%d): addr %x len %x prot %d\n",
		       p->p_pid, uap->addr, uap->len, uap->prot);
#endif

	addr = (vm_offset_t) uap->addr;
	size = (vm_size_t) uap->len;
	if ((addr & page_mask) || (size & page_mask))
		return(EINVAL);
	/*
	 * Map protections
	 */
	prot = VM_PROT_NONE;
	if (uap->prot & PROT_READ)
		prot |= VM_PROT_READ;
	if (uap->prot & PROT_WRITE)
		prot |= VM_PROT_WRITE;
	if (uap->prot & PROT_EXEC)
		prot |= VM_PROT_EXECUTE;

	switch (vm_map_protect(p->p_map, addr, addr+size, prot, FALSE)) {
	case KERN_SUCCESS:
		return (0);
	case KERN_PROTECTION_FAILURE:
		return (EACCES);
	}
	return (EINVAL);
}

/* ARGSUSED */
madvise(p, uap, retval)
	struct proc *p;
	struct args {
		char	*addr;
		int	len;
		int	behav;
	} *uap;
	int *retval;
{

	/* Not yet implemented */
	return (EOPNOTSUPP);
}

/* ARGSUSED */
mincore(p, uap, retval)
	struct proc *p;
	struct args {
		char	*addr;
		int	len;
		char	*vec;
	} *uap;
	int *retval;
{

	/* Not yet implemented */
	return (EOPNOTSUPP);
}

/*
 * Internal version of mmap.
 * Currently used by mmap, exec, and sys5 shared memory.
 * Handle is:
 *	MAP_FILE: a vnode pointer
 *	MAP_ANON: NULL or a file pointer
 */
vm_mmap(map, addr, size, prot, flags, handle, foff)
	register vm_map_t map;
	register vm_offset_t *addr;
	register vm_size_t size;
	vm_prot_t prot;
	register int flags;
	caddr_t handle;		/* XXX should be vp */
	vm_offset_t foff;
{
	register vm_pager_t pager;
	boolean_t fitit;
	vm_object_t object;
	struct vnode *vp;
	int type;
	int rv = KERN_SUCCESS;

	if (size == 0)
		return (0);

	if ((flags & MAP_FIXED) == 0) {
		fitit = TRUE;
		*addr = round_page(*addr);
	} else {
		fitit = FALSE;
		(void) vm_deallocate(map, *addr, size);
	}

	/*
	 * Lookup/allocate pager.  All except an unnamed anonymous lookup
	 * gain a reference to ensure continued existance of the object.
	 * (XXX the exception is to appease the pageout daemon)
	 */
	if ((flags & MAP_TYPE) == MAP_ANON)
		type = PG_DFLT;
	else {
		vp = (struct vnode *)handle;
		if (vp->v_type == VCHR) {
			type = PG_DEVICE;
			handle = (caddr_t)vp->v_rdev;
		} else
			type = PG_VNODE;
	}
	pager = vm_pager_allocate(type, handle, size, prot);
	if (pager == VM_PAGER_NULL)
		return (type == PG_DEVICE ? EINVAL : ENOMEM);
	/*
	 * Find object and release extra reference gained by lookup
	 */
	object = vm_object_lookup(pager);
	vm_object_deallocate(object);

	/*
	 * Anonymous memory.
	 */
	if ((flags & MAP_TYPE) == MAP_ANON) {
		rv = vm_allocate_with_pager(map, addr, size, fitit,
					    pager, (vm_offset_t)foff, TRUE);
		if (rv != KERN_SUCCESS) {
			if (handle == NULL)
				vm_pager_deallocate(pager);
			else
				vm_object_deallocate(object);
			goto out;
		}
		/*
		 * Don't cache anonymous objects.
		 * Loses the reference gained by vm_pager_allocate.
		 */
		(void) pager_cache(object, FALSE);
#ifdef DEBUG
		if (mmapdebug & MDB_MAPIT)
			printf("vm_mmap(%d): ANON *addr %x size %x pager %x\n",
			       u.u_procp->p_pid, *addr, size, pager);
#endif
	}
	/*
	 * Must be type MAP_FILE.
	 * Distinguish between character special and regular files.
	 */
	else if (vp->v_type == VCHR) {
		rv = vm_allocate_with_pager(map, addr, size, fitit,
					    pager, (vm_offset_t)foff, FALSE);
		/*
		 * Uncache the object and lose the reference gained
		 * by vm_pager_allocate().  If the call to
		 * vm_allocate_with_pager() was sucessful, then we
		 * gained an additional reference ensuring the object
		 * will continue to exist.  If the call failed then
		 * the deallocate call below will terminate the
		 * object which is fine.
		 */
		(void) pager_cache(object, FALSE);
		if (rv != KERN_SUCCESS)
			goto out;
	}
	/*
	 * A regular file
	 */
	else {
#ifdef DEBUG
		if (object == VM_OBJECT_NULL)
			printf("vm_mmap: no object: vp %x, pager %x\n",
			       vp, pager);
#endif
		/*
		 * Map it directly.
		 * Allows modifications to go out to the vnode.
		 */
		if (flags & MAP_SHARED) {
			rv = vm_allocate_with_pager(map, addr, size,
						    fitit, pager,
						    (vm_offset_t)foff, FALSE);
			if (rv != KERN_SUCCESS) {
				vm_object_deallocate(object);
				goto out;
			}
			/*
			 * Don't cache the object.  This is the easiest way
			 * of ensuring that data gets back to the filesystem
			 * because vnode_pager_deallocate() will fsync the
			 * vnode.  pager_cache() will lose the extra ref.
			 */
			if (prot & VM_PROT_WRITE)
				pager_cache(object, FALSE);
			else
				vm_object_deallocate(object);
		}
		/*
		 * Copy-on-write of file.  Two flavors.
		 * MAP_COPY is true COW, you essentially get a snapshot of
		 * the region at the time of mapping.  MAP_PRIVATE means only
		 * that your changes are not reflected back to the object.
		 * Changes made by others will be seen.
		 */
		else {
			vm_map_t tmap;
			vm_offset_t off;

			/* locate and allocate the target address space */
			rv = vm_map_find(map, VM_OBJECT_NULL, (vm_offset_t)0,
					 addr, size, fitit);
			if (rv != KERN_SUCCESS) {
				vm_object_deallocate(object);
				goto out;
			}
			tmap = vm_map_create(pmap_create(size), VM_MIN_ADDRESS,
					     VM_MIN_ADDRESS+size, TRUE);
			off = VM_MIN_ADDRESS;
			rv = vm_allocate_with_pager(tmap, &off, size,
						    TRUE, pager,
						    (vm_offset_t)foff, FALSE);
			if (rv != KERN_SUCCESS) {
				vm_object_deallocate(object);
				vm_map_deallocate(tmap);
				goto out;
			}
			/*
			 * (XXX)
			 * MAP_PRIVATE implies that we see changes made by
			 * others.  To ensure that we need to guarentee that
			 * no copy object is created (otherwise original
			 * pages would be pushed to the copy object and we
			 * would never see changes made by others).  We
			 * totally sleeze it right now by marking the object
			 * internal temporarily.
			 */
			if ((flags & MAP_COPY) == 0)
				object->internal = TRUE;
			rv = vm_map_copy(map, tmap, *addr, size, off,
					 FALSE, FALSE);
			object->internal = FALSE;
			/*
			 * (XXX)
			 * My oh my, this only gets worse...
			 * Force creation of a shadow object so that
			 * vm_map_fork will do the right thing.
			 */
			if ((flags & MAP_COPY) == 0) {
				vm_map_t tmap;
				vm_map_entry_t tentry;
				vm_object_t tobject;
				vm_offset_t toffset;
				vm_prot_t tprot;
				boolean_t twired, tsu;

				tmap = map;
				vm_map_lookup(&tmap, *addr, VM_PROT_WRITE,
					      &tentry, &tobject, &toffset,
					      &tprot, &twired, &tsu);
				vm_map_lookup_done(tmap, tentry);
			}
			/*
			 * (XXX)
			 * Map copy code cannot detect sharing unless a
			 * sharing map is involved.  So we cheat and write
			 * protect everything ourselves.  Note we cannot
			 * use vm_object_pmap_copy() because that relies
			 * on the page copy_on_write bit which isn't
			 * always accurate with shared objects.
			 */
			vm_object_pmap_force_copy(object, (vm_offset_t)foff,
					    (vm_offset_t)foff+size);
			vm_object_deallocate(object);
			vm_map_deallocate(tmap);
			if (rv != KERN_SUCCESS)
				goto out;
		}
#ifdef DEBUG
		if (mmapdebug & MDB_MAPIT)
			printf("vm_mmap(%d): FILE *addr %x size %x pager %x\n",
			       u.u_procp->p_pid, *addr, size, pager);
#endif
	}
	/*
	 * Correct protection (default is VM_PROT_ALL).
	 * Note that we set the maximum protection.  This may not be
	 * entirely correct.  Maybe the maximum protection should be based
	 * on the object permissions where it makes sense (e.g. a vnode).
	 *
	 * Changed my mind: leave max prot at VM_PROT_ALL.
	 */
	if (prot != VM_PROT_ALL) {
		rv = vm_map_protect(map, *addr, *addr+size, prot, FALSE);
		if (rv != KERN_SUCCESS) {
			(void) vm_deallocate(map, *addr, size);
			goto out;
		}
	}
	/*
	 * Shared memory is also shared with children.
	 */
	if (flags & MAP_SHARED) {
		rv = vm_inherit(map, *addr, size, VM_INHERIT_SHARE);
		if (rv != KERN_SUCCESS) {
			(void) vm_deallocate(map, *addr, size);
			goto out;
		}
	}
out:
#ifdef DEBUG
	if (mmapdebug & MDB_MAPIT)
		printf("vm_mmap: rv %d\n", rv);
#endif
	switch (rv) {
	case KERN_SUCCESS:
		return (0);
	case KERN_INVALID_ADDRESS:
	case KERN_NO_SPACE:
		return (ENOMEM);
	case KERN_PROTECTION_FAILURE:
		return (EACCES);
	default:
		return (EINVAL);
	}
}

/*
 * Internal bastardized version of MACHs vm_region system call.
 * Given address and size it returns map attributes as well
 * as the (locked) object mapped at that location.
 */
vm_region(map, addr, size, prot, max_prot, inheritance, shared, object, objoff)
	vm_map_t	map;
	vm_offset_t	*addr;		/* IN/OUT */
	vm_size_t	*size;		/* OUT */
	vm_prot_t	*prot;		/* OUT */
	vm_prot_t	*max_prot;	/* OUT */
	vm_inherit_t	*inheritance;	/* OUT */
	boolean_t	*shared;	/* OUT */
	vm_object_t	*object;	/* OUT */
	vm_offset_t	*objoff;	/* OUT */
{
	vm_map_entry_t	tmp_entry;
	register
	vm_map_entry_t	entry;
	register
	vm_offset_t	tmp_offset;
	vm_offset_t	start;

	if (map == VM_MAP_NULL)
		return(KERN_INVALID_ARGUMENT);

	start = *addr;

	vm_map_lock_read(map);
	if (!vm_map_lookup_entry(map, start, &tmp_entry)) {
		if ((entry = tmp_entry->next) == &map->header) {
			vm_map_unlock_read(map);
		   	return(KERN_NO_SPACE);
		}
		start = entry->start;
		*addr = start;
	} else
		entry = tmp_entry;

	*prot = entry->protection;
	*max_prot = entry->max_protection;
	*inheritance = entry->inheritance;

	tmp_offset = entry->offset + (start - entry->start);
	*size = (entry->end - start);

	if (entry->is_a_map) {
		register vm_map_t share_map;
		vm_size_t share_size;

		share_map = entry->object.share_map;

		vm_map_lock_read(share_map);
		(void) vm_map_lookup_entry(share_map, tmp_offset, &tmp_entry);

		if ((share_size = (tmp_entry->end - tmp_offset)) < *size)
			*size = share_size;

		vm_object_lock(tmp_entry->object);
		*object = tmp_entry->object.vm_object;
		*objoff = tmp_entry->offset + (tmp_offset - tmp_entry->start);

		*shared = (share_map->ref_count != 1);
		vm_map_unlock_read(share_map);
	} else {
		vm_object_lock(entry->object);
		*object = entry->object.vm_object;
		*objoff = tmp_offset;

		*shared = FALSE;
	}

	vm_map_unlock_read(map);

	return(KERN_SUCCESS);
}

/*
 * Yet another bastard routine.
 */
vm_allocate_with_pager(map, addr, size, fitit, pager, poffset, internal)
	register vm_map_t	map;
	register vm_offset_t	*addr;
	register vm_size_t	size;
	boolean_t		fitit;
	vm_pager_t		pager;
	vm_offset_t		poffset;
	boolean_t		internal;
{
	register vm_object_t	object;
	register int		result;

	if (map == VM_MAP_NULL)
		return(KERN_INVALID_ARGUMENT);

	*addr = trunc_page(*addr);
	size = round_page(size);

	/*
	 *	Lookup the pager/paging-space in the object cache.
	 *	If it's not there, then create a new object and cache
	 *	it.
	 */
	object = vm_object_lookup(pager);
	vm_stat.lookups++;
	if (object == VM_OBJECT_NULL) {
		object = vm_object_allocate(size);
		vm_object_enter(object, pager);
	} else
		vm_stat.hits++;
	object->internal = internal;

	result = vm_map_find(map, object, poffset, addr, size, fitit);
	if (result != KERN_SUCCESS)
		vm_object_deallocate(object);
	else if (pager != vm_pager_null)
		vm_object_setpager(object, pager, (vm_offset_t) 0, TRUE);
	return(result);
}

/*
 * XXX: this routine belongs in vm_map.c.
 *
 * Returns TRUE if the range [start - end) is allocated in either
 * a single entry (single_entry == TRUE) or multiple contiguous
 * entries (single_entry == FALSE).
 *
 * start and end should be page aligned.
 */
boolean_t
vm_map_is_allocated(map, start, end, single_entry)
	vm_map_t map;
	vm_offset_t start, end;
	boolean_t single_entry;
{
	vm_map_entry_t mapent;
	register vm_offset_t nend;

	vm_map_lock_read(map);

	/*
	 * Start address not in any entry
	 */
	if (!vm_map_lookup_entry(map, start, &mapent)) {
		vm_map_unlock_read(map);
		return (FALSE);
	}
	/*
	 * Find the maximum stretch of contiguously allocated space
	 */
	nend = mapent->end;
	if (!single_entry) {
		mapent = mapent->next;
		while (mapent != &map->header && mapent->start == nend) {
			nend = mapent->end;
			mapent = mapent->next;
		}
	}

	vm_map_unlock_read(map);
	return (end <= nend);
}

#include "../vm/vm_page.h"

/*
 * Doesn't trust the COW bit in the page structure.
 * vm_fault can improperly set it.
 */
void
vm_object_pmap_force_copy(object, start, end)
	register vm_object_t	object;
	register vm_offset_t	start;
	register vm_offset_t	end;
{
	register vm_page_t	p;

	if (object == VM_OBJECT_NULL)
		return;

	vm_object_lock(object);
	p = (vm_page_t) queue_first(&object->memq);
	while (!queue_end(&object->memq, (queue_entry_t) p)) {
		if (start <= p->offset && p->offset < end) {
			pmap_copy_on_write(VM_PAGE_TO_PHYS(p));
			p->copy_on_write = TRUE;
		}
		p = (vm_page_t) queue_next(&p->listq);
	}
	vm_object_unlock(object);
}