Re: [PATCH v7 1/4] vmalloc: Add __vmalloc_node_try_addr function

[Edgecombe, Rick P]

Forgot to include this:

Reviewed-by: Kees Cook <keescook@xxxxxxxxxxxx>

On Mon, 2018-10-01 at 14:38 -0700, Rick Edgecombe wrote:
> Create __vmalloc_node_try_addr function that tries to allocate at a specific
> address and supports caller specified behavior for whether any lazy purging
> happens if there is a collision.
> 
> This new function draws from the __vmalloc_node_range implementation. Attempts
> to merge the two into a single allocator resulted in logic that was difficult
> to follow, so they are left separate.
> 
> Signed-off-by: Rick Edgecombe <rick.p.edgecombe@xxxxxxxxx>
> ---
> Âinclude/linux/vmalloc.h |ÂÂÂ3 +
> Âmm/vmalloc.cÂÂÂÂÂÂÂÂÂÂÂÂ| 177
> +++++++++++++++++++++++++++++++++++++++++++++++-
> Â2 files changed, 179 insertions(+), 1 deletion(-)
> 
> diff --git a/include/linux/vmalloc.h b/include/linux/vmalloc.h
> index 398e9c9..c7712c8 100644
> --- a/include/linux/vmalloc.h
> +++ b/include/linux/vmalloc.h
> @@ -82,6 +82,9 @@ extern void *__vmalloc_node_range(unsigned long size,
> unsigned long align,
> Â			unsigned long start, unsigned long end, gfp_t
> gfp_mask,
> Â			pgprot_t prot, unsigned long vm_flags, int node,
> Â			const void *caller);
> +extern void *__vmalloc_node_try_addr(unsigned long addr, unsigned long size,
> +			gfp_t gfp_mask,	pgprot_t prot, unsigned long
> vm_flags,
> +			int node, int try_purge, const void *caller);
> Â#ifndef CONFIG_MMU
> Âextern void *__vmalloc_node_flags(unsigned long size, int node, gfp_t flags);
> Âstatic inline void *__vmalloc_node_flags_caller(unsigned long size, int node,
> diff --git a/mm/vmalloc.c b/mm/vmalloc.c
> index a728fc4..1954458 100644
> --- a/mm/vmalloc.c
> +++ b/mm/vmalloc.c
> @@ -1709,6 +1709,181 @@ static void *__vmalloc_area_node(struct vm_struct
> *area, gfp_t gfp_mask,
> Â	return NULL;
> Â}
> Â
> +static bool pvm_find_next_prev(unsigned long end,
> +			ÂÂÂÂÂÂÂstruct vmap_area **pnext,
> +			ÂÂÂÂÂÂÂstruct vmap_area **pprev);
> +
> +/* Try to allocate a region of KVA of the specified address and size. */
> +static struct vmap_area *try_alloc_vmap_area(unsigned long addr,
> +				unsigned long size, int node, gfp_t gfp_mask,
> +				int try_purge)
> +{
> +	struct vmap_area *va;
> +	struct vmap_area *cur_va = NULL;
> +	struct vmap_area *first_before = NULL;
> +	int need_purge = 0;
> +	int blocked = 0;
> +	int purged = 0;
> +	unsigned long addr_end;
> +
> +	WARN_ON(!size);
> +	WARN_ON(offset_in_page(size));
> +
> +	addr_end = addr + size;
> +	if (addr > addr_end)
> +		return ERR_PTR(-EOVERFLOW);
> +
> +	might_sleep();
> +
> +	va = kmalloc_node(sizeof(struct vmap_area),
> +			gfp_mask & GFP_RECLAIM_MASK, node);
> +	if (unlikely(!va))
> +		return ERR_PTR(-ENOMEM);
> +
> +	/*
> +	Â* Only scan the relevant parts containing pointers to other objects
> +	Â* to avoid false negatives.
> +	Â*/
> +	kmemleak_scan_area(&va->rb_node, SIZE_MAX, gfp_mask &
> GFP_RECLAIM_MASK);
> +
> +retry:
> +	spin_lock(&vmap_area_lock);
> +
> +	pvm_find_next_prev(addr, &cur_va, &first_before);
> +
> +	if (!cur_va)
> +		goto found;
> +
> +	/*
> +	Â* If there is no VA that starts before the target address, start the
> +	Â* check from the closest VA in order to cover the case where the
> +	Â* allocation overlaps at the end.
> +	Â*/
> +	if (first_before && addr < first_before->va_end)
> +		cur_va = first_before;
> +
> +	/* Linearly search through to make sure there is a hole */
> +	while (cur_va->va_start < addr_end) {
> +		if (cur_va->va_end > addr) {
> +			if (cur_va->flags & VM_LAZY_FREE) {
> +				need_purge = 1;
> +			} else {
> +				blocked = 1;
> +				break;
> +			}
> +		}
> +
> +		if (list_is_last(&cur_va->list, &vmap_area_list))
> +			break;
> +
> +		cur_va = list_next_entry(cur_va, list);
> +	}
> +
> +	/*
> +	Â* If a non-lazy free va blocks the allocation, or
> +	Â* we are not supposed to purge, but we need to, the
> +	Â* allocation fails.
> +	Â*/
> +	if (blocked || (need_purge && !try_purge))
> +		goto fail;
> +
> +	if (try_purge && need_purge) {
> +		/* if purged once before, give up */
> +		if (purged)
> +			goto fail;
> +
> +		/*
> +		Â* If the va blocking the allocation is set to
> +		Â* be purged then purge all vmap_areas that are
> +		Â* set to purged since this will flush the TLBs
> +		Â* anyway.
> +		Â*/
> +		spin_unlock(&vmap_area_lock);
> +		purge_vmap_area_lazy();
> +		need_purge = 0;
> +		purged = 1;
> +		goto retry;
> +	}
> +
> +found:
> +	va->va_start = addr;
> +	va->va_end = addr_end;
> +	va->flags = 0;
> +	__insert_vmap_area(va);
> +	spin_unlock(&vmap_area_lock);
> +
> +	return va;
> +fail:
> +	spin_unlock(&vmap_area_lock);
> +	kfree(va);
> +	if (need_purge && !blocked)
> +		return ERR_PTR(-EUCLEAN);
> +	return ERR_PTR(-EBUSY);
> +}
> +
> +/**
> + *	__vmalloc_try_addrÂÂ-ÂÂtry to alloc at a specific address
> + *	@addr:		address to try
> + *	@size:		size to try
> + *	@gfp_mask:	flags for the page level allocator
> + *	@prot:		protection mask for the allocated pages
> + *	@vm_flags:	additional vm area flags (e.g. %VM_NO_GUARD)
> + *	@node:		node to use for allocation or NUMA_NO_NODE
> + *	@try_purge:	try to purge if needed to fulfill and allocation
> + *	@caller:	caller's return address
> + *
> + *	Try to allocate at the specific address. If it succeeds the address
> is
> + *	returned. If it fails an EBUSY ERR_PTR is returned. If try_purge is
> + *	zero, it will return an EUCLEAN ERR_PTR if it could have allocated
> if it
> + *	was allowed to purge. It may trigger TLB flushes if a purge is
> needed,
> + *	and try_purge is set.
> + */
> +void *__vmalloc_node_try_addr(unsigned long addr, unsigned long size,
> +			gfp_t gfp_mask,	pgprot_t prot, unsigned long
> vm_flags,
> +			int node, int try_purge, const void *caller)
> +{
> +	struct vmap_area *va;
> +	struct vm_struct *area;
> +	void *alloc_addr;
> +	unsigned long real_size = size;
> +
> +	size = PAGE_ALIGN(size);
> +	if (!size || (size >> PAGE_SHIFT) > totalram_pages)
> +		return NULL;
> +
> +	WARN_ON(in_interrupt());
> +
> +	if (!(vm_flags & VM_NO_GUARD))
> +		size += PAGE_SIZE;
> +
> +	va = try_alloc_vmap_area(addr, size, node, gfp_mask, try_purge);
> +	if (IS_ERR(va))
> +		goto fail;
> +
> +	area = kzalloc_node(sizeof(*area), gfp_mask & GFP_RECLAIM_MASK,
> node);
> +	if (unlikely(!area)) {
> +		warn_alloc(gfp_mask, NULL, "kmalloc: allocation failure");
> +		return ERR_PTR(-ENOMEM);
> +	}
> +
> +	setup_vmalloc_vm(area, va, vm_flags, caller);
> +
> +	alloc_addr = __vmalloc_area_node(area, gfp_mask, prot, node);
> +	if (!alloc_addr) {
> +		warn_alloc(gfp_mask, NULL,
> +			"vmalloc: allocation failure: %lu bytes", real_size);
> +		return ERR_PTR(-ENOMEM);
> +	}
> +
> +	clear_vm_uninitialized_flag(area);
> +
> +	kmemleak_vmalloc(area, real_size, gfp_mask);
> +
> +	return alloc_addr;
> +fail:
> +	return va;
> +}
> +
> Â/**
> Â *	__vmalloc_node_rangeÂÂ-ÂÂallocate virtually contiguous memory
> Â *	@size:		allocation size
> @@ -2355,7 +2530,6 @@ void free_vm_area(struct vm_struct *area)
> Â}
> ÂEXPORT_SYMBOL_GPL(free_vm_area);
> Â
> -#ifdef CONFIG_SMP
> Âstatic struct vmap_area *node_to_va(struct rb_node *n)
> Â{
> Â	return rb_entry_safe(n, struct vmap_area, rb_node);
> @@ -2403,6 +2577,7 @@ static bool pvm_find_next_prev(unsigned long end,
> Â	return true;
> Â}
> Â
> +#ifdef CONFIG_SMP
> Â/**
> Â * pvm_determine_end - find the highest aligned address between two
> vmap_areas
> Â * @pnext: in/out arg for the next vmap_area