uvm_mmap.c

来自「基于组件方式开发操作系统的OSKIT源代码」· C语言 代码 · 共 1,306 行 · 第 1/2 页

	case KERN_PAGES_LOCKED:	/* XXXCDC: uvm doesn't return this */
		return (EBUSY);
	default:
		return (EINVAL);
	}
	/*NOTREACHED*/
}

/*
 * sys_munmap: unmap a users memory
 */

int
sys_munmap(p, v, retval)
	struct proc *p;
	void *v;
	register_t *retval;
{
	struct sys_munmap_args /* {
		syscallarg(caddr_t) addr;
		syscallarg(size_t) len;
	} */ *uap = v;
	vaddr_t addr;
	vsize_t size, pageoff;
	vm_map_t map;
	vaddr_t vm_min_address = VM_MIN_ADDRESS;
	struct vm_map_entry *dead_entries;

	/*
	 * get syscall args...
	 */

	addr = (vaddr_t) SCARG(uap, addr);
	size = (vsize_t) SCARG(uap, len);
	
	/*
	 * align the address to a page boundary, and adjust the size accordingly
	 */

	pageoff = (addr & PAGE_MASK);
	addr -= pageoff;
	size += pageoff;
	size = (vsize_t) round_page(size);

	if ((int)size < 0)
		return (EINVAL);
	if (size == 0)
		return (0);

	/*
	 * Check for illegal addresses.  Watch out for address wrap...
	 * Note that VM_*_ADDRESS are not constants due to casts (argh).
	 */
	if (VM_MAXUSER_ADDRESS > 0 && addr + size > VM_MAXUSER_ADDRESS)
		return (EINVAL);
	if (vm_min_address > 0 && addr < vm_min_address)
		return (EINVAL);
	if (addr > addr + size)
		return (EINVAL);
	map = &p->p_vmspace->vm_map;


	vm_map_lock(map);	/* lock map so we can checkprot */

	/*
	 * interesting system call semantic: make sure entire range is 
	 * allocated before allowing an unmap.
	 */

	if (!uvm_map_checkprot(map, addr, addr + size, VM_PROT_NONE)) {
		vm_map_unlock(map);
		return (EINVAL);
	}

	/*
	 * doit!
	 */
	(void) uvm_unmap_remove(map, addr, addr + size, &dead_entries);

	vm_map_unlock(map);	/* and unlock */

	if (dead_entries != NULL)
		uvm_unmap_detach(dead_entries, 0);

	return (0);
}

/*
 * sys_mprotect: the mprotect system call
 */

int
sys_mprotect(p, v, retval)
	struct proc *p;
	void *v;
	register_t *retval;
{
	struct sys_mprotect_args /* {
		syscallarg(caddr_t) addr;
		syscallarg(int) len;
		syscallarg(int) prot;
	} */ *uap = v;
	vaddr_t addr;
	vsize_t size, pageoff;
	vm_prot_t prot;
	int rv;

	/*
	 * extract syscall args from uap
	 */

	addr = (vaddr_t)SCARG(uap, addr);
	size = (vsize_t)SCARG(uap, len);
	prot = SCARG(uap, prot) & VM_PROT_ALL;

	/*
	 * align the address to a page boundary, and adjust the size accordingly
	 */
	pageoff = (addr & PAGE_MASK);
	addr -= pageoff;
	size += pageoff;
	size = (vsize_t) round_page(size);
	if ((int)size < 0)
		return (EINVAL);

	/*
	 * doit
	 */
#ifdef OSKIT
	{
	    extern vm_map_t kmem_map;	/* kernel malloc submap */
	    vm_map_t map;

	    /*
	     * We want to use mprotect in kernel malloc area.  To do so,
	     * if addr is in the submap (kmem_map), we have to pass kmem_map
	     * instead.  Otherwise, we will fail because uvm_map_protect
	     * returns KERN_INVALID_ARGUMENT. 
	     */
	    if (/*p == &proc0 */
		p->p_vmspace == &vmspace0
		&& kmem_map->header.start <= (vaddr_t)addr 
		&& (vaddr_t)addr < kmem_map->header.end) {
		map = kmem_map; 
	    } else {
		map = &p->p_vmspace->vm_map;
	    }
	    rv = uvm_map_protect(map, addr, addr+size, prot, FALSE);
	}
#else
	rv = uvm_map_protect(&p->p_vmspace->vm_map, 
			   addr, addr+size, prot, FALSE);
#endif

	if (rv == KERN_SUCCESS)
		return (0);
	if (rv == KERN_PROTECTION_FAILURE)
		return (EACCES);
	return (EINVAL);
}

/*
 * sys_minherit: the minherit system call
 */

int
sys_minherit(p, v, retval)
	struct proc *p;
	void *v;
	register_t *retval;
{
	struct sys_minherit_args /* {
		syscallarg(caddr_t) addr;
		syscallarg(int) len;
		syscallarg(int) inherit;
	} */ *uap = v;
	vaddr_t addr;
	vsize_t size, pageoff;
	vm_inherit_t inherit;
	
	addr = (vaddr_t)SCARG(uap, addr);
	size = (vsize_t)SCARG(uap, len);
	inherit = SCARG(uap, inherit);
	/*
	 * align the address to a page boundary, and adjust the size accordingly
	 */

	pageoff = (addr & PAGE_MASK);
	addr -= pageoff;
	size += pageoff;
	size = (vsize_t) round_page(size);

	if ((int)size < 0)
		return (EINVAL);
	
	switch (uvm_map_inherit(&p->p_vmspace->vm_map, addr, addr+size,
			 inherit)) {
	case KERN_SUCCESS:
		return (0);
	case KERN_PROTECTION_FAILURE:
		return (EACCES);
	}
	return (EINVAL);
}

/*
 * sys_madvise: give advice about memory usage.
 */

/* ARGSUSED */
int
sys_madvise(p, v, retval)
	struct proc *p;
	void *v;
	register_t *retval;
{
	struct sys_madvise_args /* {
		syscallarg(caddr_t) addr;
		syscallarg(size_t) len;
		syscallarg(int) behav;
	} */ *uap = v;
	vaddr_t addr;
	vsize_t size, pageoff;
	int advice, rv;;
	
	addr = (vaddr_t)SCARG(uap, addr);
	size = (vsize_t)SCARG(uap, len);
	advice = SCARG(uap, behav);

	/*
	 * align the address to a page boundary, and adjust the size accordingly
	 */
	pageoff = (addr & PAGE_MASK);
	addr -= pageoff;
	size += pageoff;
	size = (vsize_t) round_page(size);

	if ((ssize_t)size <= 0)
		return (EINVAL);

	switch (advice) {
	case MADV_NORMAL:
	case MADV_RANDOM:
	case MADV_SEQUENTIAL:
		rv = uvm_map_advice(&p->p_vmspace->vm_map, addr, addr + size,
		    advice);
		break;

	case MADV_WILLNEED:
		/*
		 * Activate all these pages, pre-faulting them in if
		 * necessary.
		 */
		/*
		 * XXX IMPLEMENT ME.
		 * Should invent a "weak" mode for uvm_fault()
		 * which would only do the PGO_LOCKED pgo_get().
		 */
		return (0);

	case MADV_DONTNEED:
		/*
		 * Deactivate all these pages.  We don't need them
		 * any more.  We don't, however, toss the data in
		 * the pages.
		 */
		rv = uvm_map_clean(&p->p_vmspace->vm_map, addr, addr + size,
		    PGO_DEACTIVATE);
		break;

	case MADV_FREE:
		/*
		 * These pages contain no valid data, and may be
		 * garbage-collected.  Toss all resources, including
		 * any swap space in use.
		 */
		rv = uvm_map_clean(&p->p_vmspace->vm_map, addr, addr + size,
		    PGO_FREE);
		break;

	case MADV_SPACEAVAIL:
		/*
		 * XXXMRG What is this?  I think it's:
		 *
		 *	Ensure that we have allocated backing-store
		 *	for these pages.
		 *
		 * This is going to require changes to the page daemon,
		 * as it will free swap space allocated to pages in core.
		 * There's also what to do for device/file/anonymous memory.
		 */
		return (EINVAL);

	default:
		return (EINVAL);
	}

	switch (rv) {
	case KERN_SUCCESS:
		return (0);
	case KERN_NO_SPACE:
		return (EAGAIN);
	case KERN_INVALID_ADDRESS:
		return (ENOMEM);
	case KERN_FAILURE:
		return (EIO);
	}

	return (EINVAL);
}

/*
 * sys_mlock: memory lock
 */

int
sys_mlock(p, v, retval)
	struct proc *p;
	void *v;
	register_t *retval;
{
	struct sys_mlock_args /* {
		syscallarg(const void *) addr;
		syscallarg(size_t) len;
	} */ *uap = v;
	vaddr_t addr;
	vsize_t size, pageoff;
	int error;

	/*
	 * extract syscall args from uap
	 */
	addr = (vaddr_t)SCARG(uap, addr);
	size = (vsize_t)SCARG(uap, len);

	/*
	 * align the address to a page boundary and adjust the size accordingly
	 */
	pageoff = (addr & PAGE_MASK);
	addr -= pageoff;
	size += pageoff;
	size = (vsize_t) round_page(size);
	
	/* disallow wrap-around. */
	if (addr + (int)size < addr)
		return (EINVAL);

	if (atop(size) + uvmexp.wired > uvmexp.wiredmax)
		return (EAGAIN);

#ifdef pmap_wired_count
	if (size + ptoa(pmap_wired_count(vm_map_pmap(&p->p_vmspace->vm_map))) >
			p->p_rlimit[RLIMIT_MEMLOCK].rlim_cur)
		return (EAGAIN);
#else
	if ((error = suser(p->p_ucred, &p->p_acflag)) != 0)
		return (error);
#endif

	error = uvm_map_pageable(&p->p_vmspace->vm_map, addr, addr+size, FALSE,
	    0);
	return (error == KERN_SUCCESS ? 0 : ENOMEM);
}

/*
 * sys_munlock: unlock wired pages
 */

int
sys_munlock(p, v, retval)
	struct proc *p;
	void *v;
	register_t *retval;
{
	struct sys_munlock_args /* {
		syscallarg(const void *) addr;
		syscallarg(size_t) len;
	} */ *uap = v;
	vaddr_t addr;
	vsize_t size, pageoff;
	int error;

	/*
	 * extract syscall args from uap
	 */

	addr = (vaddr_t)SCARG(uap, addr);
	size = (vsize_t)SCARG(uap, len);

	/*
	 * align the address to a page boundary, and adjust the size accordingly
	 */
	pageoff = (addr & PAGE_MASK);
	addr -= pageoff;
	size += pageoff;
	size = (vsize_t) round_page(size);

	/* disallow wrap-around. */
	if (addr + (int)size < addr)
		return (EINVAL);

#ifndef pmap_wired_count
	if ((error = suser(p->p_ucred, &p->p_acflag)) != 0)
		return (error);
#endif

	error = uvm_map_pageable(&p->p_vmspace->vm_map, addr, addr+size, TRUE,
	    0);
	return (error == KERN_SUCCESS ? 0 : ENOMEM);
}

/*
 * sys_mlockall: lock all pages mapped into an address space.
 */

int
sys_mlockall(p, v, retval)
	struct proc *p;
	void *v;
	register_t *retval;
{
	struct sys_mlockall_args /* {
		syscallarg(int) flags;
	} */ *uap = v;
	int error, flags;

	flags = SCARG(uap, flags);

	if (flags == 0 ||
	    (flags & ~(MCL_CURRENT|MCL_FUTURE)) != 0)
		return (EINVAL);

#ifndef pmap_wired_count
	if ((error = suser(p->p_ucred, &p->p_acflag)) != 0)
		return (error);
#endif

	error = uvm_map_pageable_all(&p->p_vmspace->vm_map, flags,
	    p->p_rlimit[RLIMIT_MEMLOCK].rlim_cur);
	switch (error) {
	case KERN_SUCCESS:
		error = 0;
		break;

	case KERN_NO_SPACE:	/* XXX overloaded */
		error = ENOMEM;
		break;

	default:
		/*
		 * "Some or all of the memory could not be locked when
		 * the call was made."
		 */
		error = EAGAIN;
	}

	return (error);
}

/*
 * sys_munlockall: unlock all pages mapped into an address space.
 */

int
sys_munlockall(p, v, retval)
	struct proc *p;
	void *v;
	register_t *retval;
{

	(void) uvm_map_pageable_all(&p->p_vmspace->vm_map, 0, 0);
	return (0);
}

/*
 * uvm_mmap: internal version of mmap
 *
 * - used by sys_mmap, exec, and sysv shm
 * - handle is a vnode pointer or NULL for MAP_ANON (XXX: not true,
 *	sysv shm uses "named anonymous memory")
 * - caller must page-align the file offset
 */

int
uvm_mmap(map, addr, size, prot, maxprot, flags, handle, foff, locklimit)
	vm_map_t map;
	vaddr_t *addr;
	vsize_t size;
	vm_prot_t prot, maxprot;
	int flags;
	caddr_t handle;		/* XXX: VNODE? */
	voff_t foff;
	vsize_t locklimit;
{
	struct uvm_object *uobj;
#ifndef OSKIT
	struct vnode *vp;
#endif
	int retval;
	int advice = UVM_ADV_NORMAL;
	uvm_flag_t uvmflag = 0;

	/*
	 * check params
	 */

	if (size == 0)
		return(0);
	if (foff & PAGE_MASK)
		return(EINVAL);
	if ((prot & maxprot) != prot)
		return(EINVAL);

	/*
	 * for non-fixed mappings, round off the suggested address.
	 * for fixed mappings, check alignment and zap old mappings.
	 */

	if ((flags & MAP_FIXED) == 0) {
		*addr = round_page(*addr);	/* round */
	} else {
		
		if (*addr & PAGE_MASK)
			return(EINVAL);
		uvmflag |= UVM_FLAG_FIXED;
		(void) uvm_unmap(map, *addr, *addr + size);	/* zap! */
	}

	/*
	 * handle anon vs. non-anon mappings.   for non-anon mappings attach
	 * to underlying vm object.
	 */

	if (flags & MAP_ANON) {
		foff = UVM_UNKNOWN_OFFSET;
		uobj = NULL;
		if ((flags & MAP_SHARED) == 0)
			/* XXX: defer amap create */
			uvmflag |= UVM_FLAG_COPYONW;
		else
			/* shared: create amap now */
			uvmflag |= UVM_FLAG_OVERLAY;

	} else {
#ifdef OSKIT
	    	oskit_iunknown_t *iunknown = (oskit_iunknown_t*)handle;
		uobj = oskit_blkio_attach(iunknown);
		if (uobj == NULL)
			return(ENOMEM);
#else
		vp = (struct vnode *) handle;	/* get vnode */
		if (vp->v_type != VCHR) {
			uobj = uvn_attach((void *) vp, (flags & MAP_SHARED) ?
			   maxprot : (maxprot & ~VM_PROT_WRITE));

			/* XXX for now, attach doesn't gain a ref */
			VREF(vp);
		} else {
			uobj = udv_attach((void *) &vp->v_rdev,
			    (flags & MAP_SHARED) ?
			    maxprot : (maxprot & ~VM_PROT_WRITE), foff, size);
			advice = UVM_ADV_RANDOM;
		}
		
		if (uobj == NULL)
			return((vp->v_type == VREG) ? ENOMEM : EINVAL);
#endif

		if ((flags & MAP_SHARED) == 0)
			uvmflag |= UVM_FLAG_COPYONW;
	}

	/*
	 * set up mapping flags
	 */

	uvmflag = UVM_MAPFLAG(prot, maxprot, 
			(flags & MAP_SHARED) ? UVM_INH_SHARE : UVM_INH_COPY,
			advice, uvmflag);

	/*
	 * do it!
	 */

	retval = uvm_map(map, addr, size, uobj, foff, 0, uvmflag);

	if (retval == KERN_SUCCESS) {
		/*
		 * POSIX 1003.1b -- if our address space was configured
		 * to lock all future mappings, wire the one we just made.
		 */
		if (prot == VM_PROT_NONE) {
			/*
			 * No more work to do in this case.
			 */
			return (0);
		}
		
		vm_map_lock(map);

		if (map->flags & VM_MAP_WIREFUTURE) {
			if ((atop(size) + uvmexp.wired) > uvmexp.wiredmax
#ifdef pmap_wired_count
			    || (locklimit != 0 && (size +
			         ptoa(pmap_wired_count(vm_map_pmap(map)))) >
			        locklimit)
#endif
			) {
				retval = KERN_RESOURCE_SHORTAGE;
				vm_map_unlock(map);
				/* unmap the region! */
				(void) uvm_unmap(map, *addr, *addr + size);
				goto bad;
			}
			/*
			 * uvm_map_pageable() always returns the map
			 * unlocked.
			 */
			retval = uvm_map_pageable(map, *addr, *addr + size,
			    FALSE, UVM_LK_ENTER);
			if (retval != KERN_SUCCESS) {
				/* unmap the region! */
				(void) uvm_unmap(map, *addr, *addr + size);
				goto bad;
			}
			return (0);
		}

		vm_map_unlock(map);

		return (0);
	}

	/*
	 * errors: first detach from the uobj, if any.
	 */
	
	if (uobj)
		uobj->pgops->pgo_detach(uobj);

 bad:
	switch (retval) {
	case KERN_INVALID_ADDRESS:
	case KERN_NO_SPACE:
		return(ENOMEM);
	case KERN_RESOURCE_SHORTAGE:
		return (EAGAIN);
	case KERN_PROTECTION_FAILURE:
		return(EACCES);
	}
	return(EINVAL);
}