Re: [PATCH v2 7/8] mm: refactor __mmap_region()

[Lorenzo Stoakes <lorenzo.stoakes@oracle.com>]

On Fri, Oct 25, 2024 at 10:35:20AM +0200, Vlastimil Babka wrote:
> On 10/23/24 22:38, Lorenzo Stoakes wrote:
> > We have seen bugs and resource leaks arise from the complexity of the
> > __mmap_region() function. This, and the generally deeply fragile error
> > handling logic and complexity which makes understanding the function
> > difficult make it highly desirable to refactor it into something readable.
> >
> > Achieve this by separating the function into smaller logical parts which
> > are easier to understand and follow, and which importantly very
> > significantly simplify the error handling.
> >
> > Note that we now call vms_abort_munmap_vmas() in more error paths than we
> > used to, however in cases where no abort need occur, vms->nr_pages will be
> > equal to zero and we simply exit this function without doing more than we
> > would have done previously.
> >
> > Importantly, the invocation of the driver mmap hook via mmap_file() now has
> > very simple and obvious handling (this was previously the most problematic
> > part of the mmap() operation).
> >
> > Signed-off-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
>
> Reviewed-by: Vlastimil Babka <vbabka@suse.cz>
>
> Some nits:
>
> > +static int __mmap_prepare(struct mmap_state *map, struct vma_merge_struct *vmg,
> > +	struct list_head *uf)
> >  {
> > -	struct mm_struct *mm = current->mm;
> > -	struct vm_area_struct *vma = NULL;
> > -	pgoff_t pglen = PHYS_PFN(len);
> > -	unsigned long charged = 0;
> > -	struct vma_munmap_struct vms;
> > -	struct ma_state mas_detach;
> > -	struct maple_tree mt_detach;
> > -	unsigned long end = addr + len;
> >  	int error;
> > -	VMA_ITERATOR(vmi, mm, addr);
> > -	VMG_STATE(vmg, mm, &vmi, addr, end, vm_flags, pgoff);
> > -
> > -	vmg.file = file;
> > -	/* Find the first overlapping VMA */
> > -	vma = vma_find(&vmi, end);
> > -	init_vma_munmap(&vms, &vmi, vma, addr, end, uf, /* unlock = */ false);
> > -	if (vma) {
> > -		mt_init_flags(&mt_detach, vmi.mas.tree->ma_flags & MT_FLAGS_LOCK_MASK);
> > -		mt_on_stack(mt_detach);
> > -		mas_init(&mas_detach, &mt_detach, /* addr = */ 0);
> > +	struct vma_iterator *vmi = map->vmi;
> > +	struct vma_munmap_struct *vms = &map->vms;
> > +
> > +	/* Find the first overlapping VMA and initialise unmap state. */
> > +	vms->vma = vma_find(vmi, map->end);
> > +	init_vma_munmap(vms, vmi, vms->vma, map->addr, map->end, uf,
> > +			/* unlock = */ false);
> > +
> > +	/* OK, we have overlapping VMAs - prepare to unmap them. */
> > +	if (vms->vma) {
> > +		mt_init_flags(&map->mt_detach,
> > +			      vmi->mas.tree->ma_flags & MT_FLAGS_LOCK_MASK);
> > +		mt_on_stack(map->mt_detach);
> > +		mas_init(&map->mas_detach, &map->mt_detach, /* addr = */ 0);
> >  		/* Prepare to unmap any existing mapping in the area */
> > -		error = vms_gather_munmap_vmas(&vms, &mas_detach);
> > -		if (error)
> > -			goto gather_failed;
> > +		error = vms_gather_munmap_vmas(vms, &map->mas_detach);
> > +		if (error) {
> > +			/* On error VMAs will already have been reattached. */
> > +			vms->nr_pages = 0;
> > +			return error;
> > +		}
> >
> > -		vmg.next = vms.next;
> > -		vmg.prev = vms.prev;
> > -		vma = NULL;
> > +		map->next = vms->next;
> > +		map->prev = vms->prev;
> >  	} else {
> > -		vmg.next = vma_iter_next_rewind(&vmi, &vmg.prev);
> > +		map->next = vma_iter_next_rewind(vmi, &map->prev);
> >  	}
> >
> > +	/* Set up vmg for merge attempt. */
> > +	vmg->file = map->file;
> > +	vmg->prev = map->prev;
> > +	vmg->next = map->next;
>
> It seems arbitrary to do this here. vmg->flags are set in __mmap_region().
> Why not all of this? We wouldn't need to pass vmg to __mmap_prepare() at
> all? Do any of the values chenge between here and returning and vmg needs to
> capture them as they are now? AFAICS not.

Yeah I sort of felt we are 'preparing' things.

This _was_ necessary before when we were ickily leaning on the vmg to store
state and not mutate it which is not ideal.

So yeah I think probably it's better to just put _all_ the merge stuff in
__mmap_region() rather than just some.

This aligns with the need to strip out the 'reset' logic from
vma_merge_new_range().

>
> > +
> >  	/* Check against address space limit. */
> > -	if (!may_expand_vm(mm, vm_flags, pglen - vms.nr_pages)) {
> > -		error = -ENOMEM;
> > -		goto abort_munmap;
> > -	}
> > +	if (!may_expand_vm(map->mm, map->flags, map->pglen - vms->nr_pages))
> > +		return -ENOMEM;
> >
> > -	/*
> > -	 * Private writable mapping: check memory availability
> > -	 */
> > -	if (accountable_mapping(file, vm_flags)) {
> > -		charged = pglen;
> > -		charged -= vms.nr_accounted;
> > -		if (charged) {
> > -			error = security_vm_enough_memory_mm(mm, charged);
> > +	/* Private writable mapping: check memory availability. */
> > +	if (accountable_mapping(map->file, map->flags)) {
> > +		map->charged = map->pglen;
> > +		map->charged -= vms->nr_accounted;
> > +		if (map->charged) {
> > +			error = security_vm_enough_memory_mm(map->mm, map->charged);
> >  			if (error)
> > -				goto abort_munmap;
> > +				return error;
> >  		}
> >
> > -		vms.nr_accounted = 0;
> > -		vm_flags |= VM_ACCOUNT;
> > -		vmg.flags = vm_flags;
> > +		vms->nr_accounted = 0;
> > +		map->flags |= VM_ACCOUNT;
> >  	}
> >
> >  	/*
> > -	 * clear PTEs while the vma is still in the tree so that rmap
> > +	 * Clear PTEs while the vma is still in the tree so that rmap
> >  	 * cannot race with the freeing later in the truncate scenario.
> >  	 * This is also needed for mmap_file(), which is why vm_ops
> >  	 * close function is called.
> >  	 */
> > -	vms_clean_up_area(&vms, &mas_detach);
> > -	vma = vma_merge_new_range(&vmg);
> > -	if (vma)
> > -		goto expanded;
> > +	vms_clean_up_area(vms, &map->mas_detach);
> > +
> > +	return 0;
> > +}
> > +
>
> <snip>
>
> > +static int __mmap_new_vma(struct mmap_state *map, struct vma_merge_struct *vmg,
> > +			  struct vm_area_struct **vmap)
> > +{
> > +	struct vma_iterator *vmi = map->vmi;
> > +	int error = 0;
> > +	bool merged = false;
> > +	struct vm_area_struct *vma;
> > +
> >  	/*
> >  	 * Determine the object being mapped and call the appropriate
> >  	 * specific mapper. the address has already been validated, but
> >  	 * not unmapped, but the maps are removed from the list.
> >  	 */
> > -	vma = vm_area_alloc(mm);
> > -	if (!vma) {
> > -		error = -ENOMEM;
> > -		goto unacct_error;
> > -	}
> > +	vma = vm_area_alloc(map->mm);
> > +	if (!vma)
> > +		return -ENOMEM;
> >
> > -	vma_iter_config(&vmi, addr, end);
> > -	vma_set_range(vma, addr, end, pgoff);
> > -	vm_flags_init(vma, vm_flags);
> > -	vma->vm_page_prot = vm_get_page_prot(vm_flags);
> > +	vma_iter_config(vmi, map->addr, map->end);
> > +	vma_set_range(vma, map->addr, map->end, map->pgoff);
> > +	vm_flags_init(vma, map->flags);
> > +	vma->vm_page_prot = vm_get_page_prot(map->flags);
> >
> > -	if (vma_iter_prealloc(&vmi, vma)) {
> > +	if (vma_iter_prealloc(vmi, vma)) {
> >  		error = -ENOMEM;
> >  		goto free_vma;
> >  	}
> >
> > -	if (file) {
> > -		vma->vm_file = get_file(file);
> > -		error = mmap_file(file, vma);
> > -		if (error)
> > -			goto unmap_and_free_file_vma;
> > -
> > -		/* Drivers cannot alter the address of the VMA. */
> > -		WARN_ON_ONCE(addr != vma->vm_start);
> > -		/*
> > -		 * Drivers should not permit writability when previously it was
> > -		 * disallowed.
> > -		 */
> > -		VM_WARN_ON_ONCE(vm_flags != vma->vm_flags &&
> > -				!(vm_flags & VM_MAYWRITE) &&
> > -				(vma->vm_flags & VM_MAYWRITE));
> > -
> > -		vma_iter_config(&vmi, addr, end);
> > -		/*
> > -		 * If vm_flags changed after mmap_file(), we should try merge
> > -		 * vma again as we may succeed this time.
> > -		 */
> > -		if (unlikely(vm_flags != vma->vm_flags && vmg.prev)) {
> > -			struct vm_area_struct *merge;
> > -
> > -			vmg.flags = vma->vm_flags;
> > -			/* If this fails, state is reset ready for a reattempt. */
> > -			merge = vma_merge_new_range(&vmg);
> > -
> > -			if (merge) {
> > -				/*
> > -				 * ->mmap() can change vma->vm_file and fput
> > -				 * the original file. So fput the vma->vm_file
> > -				 * here or we would add an extra fput for file
> > -				 * and cause general protection fault
> > -				 * ultimately.
> > -				 */
> > -				fput(vma->vm_file);
> > -				vm_area_free(vma);
> > -				vma = merge;
> > -				/* Update vm_flags to pick up the change. */
> > -				vm_flags = vma->vm_flags;
> > -				goto file_expanded;
> > -			}
> > -			vma_iter_config(&vmi, addr, end);
> > -		}
> > -
> > -		vm_flags = vma->vm_flags;
> > -	} else if (vm_flags & VM_SHARED) {
> > +	if (map->file)
> > +		error = __mmap_new_file_vma(map, vmg, &vma, &merged);
> > +	else if (map->flags & VM_SHARED)
> >  		error = shmem_zero_setup(vma);
> > -		if (error)
> > -			goto free_iter_vma;
> > -	} else {
> > +	else
> >  		vma_set_anonymous(vma);
> > -	}
> > +
> > +	if (error)
> > +		goto free_iter_vma;
> > +
> > +	if (merged)
> > +		goto file_expanded;
> >
> >  #ifdef CONFIG_SPARC64
> >  	/* TODO: Fix SPARC ADI! */
> > -	WARN_ON_ONCE(!arch_validate_flags(vm_flags));
> > +	WARN_ON_ONCE(!arch_validate_flags(map->flags));
> >  #endif
> >
> >  	/* Lock the VMA since it is modified after insertion into VMA tree */
> >  	vma_start_write(vma);
> > -	vma_iter_store(&vmi, vma);
> > -	mm->map_count++;
> > +	vma_iter_store(vmi, vma);
> > +	map->mm->map_count++;
> >  	vma_link_file(vma);
> >
> >  	/*
> >  	 * vma_merge_new_range() calls khugepaged_enter_vma() too, the below
> >  	 * call covers the non-merge case.
> >  	 */
> > -	khugepaged_enter_vma(vma, vma->vm_flags);
> > +	khugepaged_enter_vma(vma, map->flags);
> >
> >  file_expanded:
> > -	file = vma->vm_file;
> >  	ksm_add_vma(vma);
>
> I'm wondering if this could go to the "non file expanded" region above. If
> we expanded a vma, it was either in ksm and stays in ksm, or it was not
> eligible and that couldn't have changed by expanding?

We could change this, but the next commit removes the need for this :)

>
> > -expanded:
> > +	*vmap = vma;
> > +	return 0;
> > +
> > +free_iter_vma:
> > +	vma_iter_free(vmi);
> > +free_vma:
> > +	vm_area_free(vma);
> > +	return error;
> > +}
> > +