Subversion Repositories c0or1k

/*
 * mmap/munmap and friends.
 *
 * Copyright (C) 2007, 2008 Bahadir Balban
 */
#include <l4/lib/math.h>
#include <vm_area.h>
#include <malloc/malloc.h>
#include INC_API(errno.h)
#include <posix/sys/types.h>
#include L4LIB_INC_ARCH(syscalls.h)
#include L4LIB_INC_ARCH(syslib.h)
#include <memory.h>
#include <task.h>
#include <mmap.h>
#include <file.h>
#include <shm.h>
#include <syscalls.h>
#include <user.h>
#include <shm.h>
 
struct vm_area *vma_new(unsigned long pfn_start, unsigned long npages,
			unsigned int flags, unsigned long file_offset)
{
	struct vm_area *vma;
 
	/* Allocate new area */
	if (!(vma = kzalloc(sizeof(struct vm_area))))
		return 0;
 
	link_init(&vma->list);
	link_init(&vma->vm_obj_list);
 
	vma->pfn_start = pfn_start;
	vma->pfn_end = pfn_start + npages;
	vma->flags = flags;
	vma->file_offset = file_offset;
 
	return vma;
}
 
/*
 * Inserts a new vma to the ordered vm area list.
 *
 * The new vma is assumed to have been correctly set up not to intersect
 * with any other existing vma.
 */
int task_insert_vma(struct vm_area *this, struct link *vma_list)
{
	struct vm_area *before, *after;
 
	/* Add if list is empty */
	if (list_empty(vma_list)) {
		list_insert_tail(&this->list, vma_list);
		return 0;
	}
 
	/* Else find the right interval */
	list_foreach_struct(before, vma_list, list) {
		after = link_to_struct(before->list.next, struct vm_area, list);
 
		/* If there's only one in list */
		if (before->list.next == vma_list) {
 
			/* Eliminate the possibility of intersection */
			BUG_ON(set_intersection(this->pfn_start, this->pfn_end,
						before->pfn_start,
						before->pfn_end));
 
			/* Add as next if greater */
			if (this->pfn_start > before->pfn_start)
				list_insert(&this->list, &before->list);
			/* Add as previous if smaller */
			else if (this->pfn_start < before->pfn_start)
				list_insert_tail(&this->list, &before->list);
			else
				BUG();
 
			return 0;
		}
 
		/* If this page is in-between two other, insert it there */
		if (before->pfn_start < this->pfn_start &&
		    after->pfn_start > this->pfn_start) {
 
			/* Eliminate possibility of intersection */
			BUG_ON(set_intersection(this->pfn_start, this->pfn_end,
						before->pfn_start,
						before->pfn_end));
			BUG_ON(set_intersection(this->pfn_start, this->pfn_end,
						after->pfn_start,
						after->pfn_end));
			list_insert(&this->list, &before->list);
 
			return 0;
		}
	}
	BUG();
}
 
/*
 * Search an empty space in the task's mmapable address region.
 *
 * This does a less than O(n) algorithm by starting the estimated region
 * and vma comparison from the beginning, once a vma is not intersected
 * that means it is an available slot. However if vma's and estimated
 * region does not go head-to-head for comparison, individual intersection
 * checks would be meaningless since any other vma could be intersecting.
 * Therefore head-to-head comparison is essential here.
 */
unsigned long find_unmapped_area(unsigned long npages, struct tcb *task)
{
	unsigned long pfn_start = __pfn(task->map_start);
	unsigned long pfn_end = pfn_start + npages;
	struct vm_area *vma;
 
	if (npages > __pfn(task->map_end - task->map_start))
		return 0;
 
	/* If no vmas, first map slot is available. */
	if (list_empty(&task->vm_area_head->list))
		return task->map_start;
 
	/* First vma to check our range against */
	vma = link_to_struct(task->vm_area_head->list.next, struct vm_area, list);
 
	/* Start searching from task's end of data to start of stack */
	while (pfn_end <= __pfn(task->map_end)) {
 
		/* If intersection, skip the vma and fast-forward to next */
		if (set_intersection(pfn_start, pfn_end,
				     vma->pfn_start, vma->pfn_end)) {
 
			/* Update interval to next available space */
			pfn_start = vma->pfn_end;
			pfn_end = pfn_start + npages;
 
			/*
			 * Decision point, no more vmas left to check.
			 * Are we out of task map area?
			 */
			if (vma->list.next == &task->vm_area_head->list) {
				if (pfn_end > __pfn(task->map_end))
					break; /* Yes, fail */
				else	/* No, success */
					return __pfn_to_addr(pfn_start);
			}
 
			/* Otherwise get next vma entry */
			vma = link_to_struct(vma->list.next,
					 struct vm_area, list);
			continue;
		}
		BUG_ON(pfn_start + npages > __pfn(task->map_end));
		return __pfn_to_addr(pfn_start);
	}
 
	return 0;
}
 
/* Validate an address that is a possible candidate for an mmap() region */
int mmap_address_validate(struct tcb *task, unsigned long map_address,
			  unsigned int vm_flags)
{
	if (map_address == 0)
		return 0;
 
	/* Private mappings can only go in task address space */
	if (vm_flags & VMA_PRIVATE) {
		if ((map_address >= task->start &&
	    	     map_address < task->end) ||
		    (map_address >= __pfn_to_addr(cont_mem_regions.utcb->start) &&
		     map_address < __pfn_to_addr(cont_mem_regions.utcb->end))) {
			return 1;
		} else
			return 0;
	/*
	 * Shared mappings can go in task, utcb, and shared
	 * memory address space,
	 */
	} else if (vm_flags & VMA_SHARED) {
		if ((map_address >= task->start &&
	    	     map_address < task->end) ||
		    (map_address >= __pfn_to_addr(cont_mem_regions.shmem->start) &&
	    	     map_address < __pfn_to_addr(cont_mem_regions.shmem->end)))
			return 1;
		else
			return 0;
	} else
		BUG();
}
 
/*
 * Returns a suitably mmap'able address. It allocates
 * differently for shared and private areas.
 */
unsigned long mmap_new_address(struct tcb *task, unsigned int flags,
			       unsigned int npages)
{
	if (flags & VMA_SHARED)
		return (unsigned long)shm_new_address(npages);
	else
		return find_unmapped_area(npages, task);
}
 
/*
 * Side note:
 * Why in do_mmap() shm files have devzero mapped behind separately but
 * anonymous files map devzero directly? Because private anonymous files get
 * shadow objects in front when written to. Shm files are not private, so they
 * stay where they are and just grow. Other processes can reach and map them.
 */
 
/*
 * Maps the given file with given flags at the given page offset to the given
 * task's address space at the specified virtual memory address and length.
 *
 * The actual paging in/out of the file from/into memory pages is handled by
 * the file's pager upon page faults.
 */
void *do_mmap(struct vm_file *mapfile, unsigned long file_offset,
	      struct tcb *task, unsigned long map_address,
	      unsigned int flags, unsigned int npages)
{
	struct vm_obj_link *vmo_link, *vmo_link2;
	unsigned long file_npages;
	struct vm_area *new;
	int err;
 
	/* Set up devzero if none given */
	if (!mapfile) {
	       if (flags & VMA_ANONYMOUS) {
			BUG_ON(!(mapfile = get_devzero()));
			file_offset = 0;
	       } else
		       return PTR_ERR(-EINVAL);
	}
 
	/* Get total file pages, check if mapping is within file size */
	file_npages = __pfn(page_align_up(mapfile->length));
	if (npages > file_npages - file_offset) {
		printf("%s: Trying to map %d pages from page %lu, "
		       "but file length is %lu\n", __FUNCTION__,
		       npages, file_offset, file_npages);
		return PTR_ERR(-EINVAL);
	}
 
	/* Check invalid page size */
	if (npages == 0) {
		printf("Trying to map %d pages.\n", npages);
		return PTR_ERR(-EINVAL);
	}
	if (npages > __pfn(TASK_SIZE)) {
		printf("Trying to map too many pages: %d\n", npages);
		return PTR_ERR(-ENOMEM);
	}
 
	/* Check invalid map address */
	if (!mmap_address_validate(task, map_address, flags)) {
		if (flags & VMA_FIXED)
			return PTR_ERR(-EINVAL);
		else if (!(map_address = mmap_new_address(task, flags, npages)))
			return PTR_ERR(-ENOMEM);
	}
 
	/* Unmap any existing vmas that overlap with the new mapping */
	if ((err = do_munmap(task, map_address, npages)) < 0)
		return PTR_ERR(err);
 
	/* For valid regions that aren't allocated by us, create the vma. */
	if (!(new = vma_new(__pfn(map_address), npages, flags,
			    __pfn(file_offset))))
		return PTR_ERR(-ENOMEM);
 
	/* Attach the file as the first vm object of this vma */
	if (!(vmo_link = vm_objlink_create())) {
		kfree(new);
		return PTR_ERR(-ENOMEM);
	}
 
	/* Attach link to object */
	vm_link_object(vmo_link, &mapfile->vm_obj);
 
	/* Add link to vma list */
	list_insert_tail(&vmo_link->list, &new->vm_obj_list);
 
	/*
	 * If the file is a shm file, also map devzero behind it. i.e.
	 * vma -> vm_link -> vm_link
	 * 	     |          |
	 * 	     v          v
	 * 	  shm_file	devzero
	 *
	 * So that faults go through shm file and then devzero, as in
	 * the shadow object copy_on_write setup in fault.c
	 */
	if (mapfile->type == VM_FILE_SHM) {
		struct vm_file *dzero = get_devzero();
 
		/* Attach the file as the first vm object of this vma */
		if (!(vmo_link2 = vm_objlink_create())) {
			kfree(new);
			kfree(vmo_link);
			return PTR_ERR(-ENOMEM);
		}
		vm_link_object(vmo_link2, &dzero->vm_obj);
		list_insert_tail(&vmo_link2->list, &new->vm_obj_list);
	}
 
	/* Finished initialising the vma, add it to task */
	dprintf("%s: Mapping 0x%lx - 0x%lx\n", __FUNCTION__,
		map_address, map_address + __pfn_to_addr(npages));
	task_insert_vma(new, &task->vm_area_head->list);
 
	/*
	 * If area is going to be used going downwards, (i.e. as a stack)
	 * we return the *end* of the area as the start address.
	 */
	if (flags & VMA_GROWSDOWN)
		map_address += __pfn_to_addr(npages);
 
	return (void *)map_address;
}
 
/* mmap system call implementation */
void *__sys_mmap(struct tcb *task, void *start, size_t length, int prot,
	      	 int flags, int fd, unsigned long file_offset)
{
	unsigned int vmflags = 0;
	struct vm_file *file = 0;
 
	/* Check file validity */
	if (!(flags & MAP_ANONYMOUS))
		if (fd < 0 || fd > TASK_FILES_MAX ||
		    !task->files->fd[fd].vmfile)
			return PTR_ERR(-EBADF);
 
	/* Check file offset is page aligned */
	if (!is_page_aligned(file_offset))
		return PTR_ERR(-EINVAL);
 
	/* TODO:
	 * Check that @start does not already have a mapping.
	 * Check that pfn + npages range is within the file range.
	 * Check that posix flags passed match those defined in vm_area.h
	 */
	if (flags & MAP_ANONYMOUS) {
		file = 0;
		vmflags |= VMA_ANONYMOUS;
	} else {
		file = task->files->fd[fd].vmfile;
	}
 
	if (flags & MAP_FIXED)
		vmflags |= VMA_FIXED;
 
	if (flags & MAP_PRIVATE)
		/* This means COW, if writeable. */
		vmflags |= VMA_PRIVATE;
	else	/* This also means COW, if writeable and anonymous */
		vmflags |= VMA_SHARED;
 
	if (flags & MAP_GROWSDOWN)
		vmflags |= VMA_GROWSDOWN;
 
	if (prot & PROT_READ)
		vmflags |= VM_READ;
	if (prot & PROT_WRITE)
		vmflags |= VM_WRITE;
	if (prot & PROT_EXEC)
		vmflags |= VM_EXEC;
 
	/*
	 * Currently MAP_SHARED && MAP_ANONYMOUS mappings use the
	 * shm interface to create virtual shared memory files and
	 * do_mmap is internally called through this interface.
	 *
	 * FIXME: A common method of creating virtual shm files
	 * should be used by both sys_mmap and sys_shmget. With the
	 * current method, a task that guesses the shmid of an
	 * anonymous shared mmap can attach to it via shmat.
	 */
	if ((flags & MAP_ANONYMOUS) &&
	    (flags & MAP_SHARED)) {
		/* Create a new shared memory virtual file */
		l4id_t shmid =	sys_shmget(IPC_PRIVATE,
					   page_align_up(length),
					   0);
 
		/* Find and mmap the file via do_shmat() */
		return sys_shmat(task, shmid, 0, 0);
	}
 
	return do_mmap(file, file_offset, task, (unsigned long)start,
		       vmflags, __pfn(page_align_up(length)));
}
 
void *sys_mmap(struct tcb *task, struct sys_mmap_args *args)
{
 
	struct sys_mmap_args *mapped_args;
	void *ret;
 
	if (!(mapped_args = pager_get_user_page(task, args,
						sizeof(*args),
						VM_READ | VM_WRITE)))
		return PTR_ERR(-EINVAL);
 
	ret = __sys_mmap(task, mapped_args->start, mapped_args->length,
			 mapped_args->prot, mapped_args->flags, mapped_args->fd,
			 mapped_args->offset);
 
	return ret;
}
 
/* Sets the end of data segment for sender */
int sys_brk(struct tcb *sender, void *ds_end)
{
	return 0;
}