Re: [patch] exofs: confusion between kmap() and kmap_atomic() api

[Andrew Morton <akpm@linux-foundation.org>]

On Sun, 09 May 2010 13:16:38 +0300
Boaz Harrosh <bharrosh@panasas.com> wrote:

> On 05/07/2010 12:05 PM, Dan Carpenter wrote:
> > For kmap_atomic() we call kunmap_atomic() on the returned pointer.
> > That's different from kmap() and kunmap() and so it's easy to get them
> > backwards.
> > 
> > Signed-off-by: Dan Carpenter <error27@gmail.com>
> > 
> 
> Thank you Dan, I'll push it ASAP. 

> Looks like a bad bug. So this is actually a leak, right? kunmap_atomic
> would detect the bad pointer and do nothing?

void kunmap_atomic(void *kvaddr, enum km_type type)
{
	unsigned long vaddr = (unsigned long) kvaddr & PAGE_MASK;
	enum fixed_addresses idx = type + KM_TYPE_NR*smp_processor_id();

	/*
	 * Force other mappings to Oops if they'll try to access this pte
	 * without first remap it.  Keeping stale mappings around is a bad idea
	 * also, in case the page changes cacheability attributes or becomes
	 * a protected page in a hypervisor.
	 */
	if (vaddr == __fix_to_virt(FIX_KMAP_BEGIN+idx))
		kpte_clear_flush(kmap_pte-idx, vaddr);
	else {
#ifdef CONFIG_DEBUG_HIGHMEM
		BUG_ON(vaddr < PAGE_OFFSET);
		BUG_ON(vaddr >= (unsigned long)high_memory);
#endif
	}

	pagefault_enable();
}

if CONFIG_DEBUG_HIGHMEM=y, kunmap_atomic() will go BUG.

if CONFIG_DEBUG_HIGHMEM=n, kunmap_atomic() will do nothing, leaving the
pte pointing at the old page.  Next time someone tries to use that
kmap_atomic() slot,

void *kmap_atomic_prot(struct page *page, enum km_type type, pgprot_t prot)
{
	enum fixed_addresses idx;
	unsigned long vaddr;

	/* even !CONFIG_PREEMPT needs this, for in_atomic in do_page_fault */
	pagefault_disable();

	if (!PageHighMem(page))
		return page_address(page);

	debug_kmap_atomic(type);

	idx = type + KM_TYPE_NR*smp_processor_id();
	vaddr = __fix_to_virt(FIX_KMAP_BEGIN + idx);
	BUG_ON(!pte_none(*(kmap_pte-idx)));
	set_pte(kmap_pte-idx, mk_pte(page, prot));

	return (void *)vaddr;
}

kmap_atomic_prot() will go BUG because the pte wasn't cleared.


I can only assume that this code has never been run on i386. I'd suggest
adding a "Cc: <stable@kernel.org>" to the changelog if you have
expectations that anyone will try to run it on i386.