x86/mm: Check if PUD is large when validating a kernel address
Mel Gorman [Mon, 11 Feb 2013 14:52:36 +0000 (14:52 +0000)]
commit 0ee364eb316348ddf3e0dfcd986f5f13f528f821 upstream.

A user reported the following oops when a backup process reads
/proc/kcore:

 BUG: unable to handle kernel paging request at ffffbb00ff33b000
 IP: [<ffffffff8103157e>] kern_addr_valid+0xbe/0x110
 [...]

 Call Trace:
  [<ffffffff811b8aaa>] read_kcore+0x17a/0x370
  [<ffffffff811ad847>] proc_reg_read+0x77/0xc0
  [<ffffffff81151687>] vfs_read+0xc7/0x130
  [<ffffffff811517f3>] sys_read+0x53/0xa0
  [<ffffffff81449692>] system_call_fastpath+0x16/0x1b

Investigation determined that the bug triggered when reading
system RAM at the 4G mark. On this system, that was the first
address using 1G pages for the virt->phys direct mapping so the
PUD is pointing to a physical address, not a PMD page.

The problem is that the page table walker in kern_addr_valid() is
not checking pud_large() and treats the physical address as if
it was a PMD.  If it happens to look like pmd_none then it'll
silently fail, probably returning zeros instead of real data. If
the data happens to look like a present PMD though, it will be
walked resulting in the oops above.

This patch adds the necessary pud_large() check.

Unfortunately the problem was not readily reproducible and now
they are running the backup program without accessing
/proc/kcore so the patch has not been validated but I think it
makes sense.

Signed-off-by: Mel Gorman <mgorman@suse.de>
Reviewed-by: Rik van Riel <riel@redhat.coM>
Reviewed-by: Michal Hocko <mhocko@suse.cz>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: linux-mm@kvack.org
Link: http://lkml.kernel.org/r/20130211145236.GX21389@suse.de
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

arch/x86/include/asm/pgtable.h
arch/x86/mm/init_64.c

index c3520d7..3f3dd52 100644 (file)
@@ -142,6 +142,11 @@ static inline unsigned long pmd_pfn(pmd_t pmd)
        return (pmd_val(pmd) & PTE_PFN_MASK) >> PAGE_SHIFT;
 }
 
+static inline unsigned long pud_pfn(pud_t pud)
+{
+       return (pud_val(pud) & PTE_PFN_MASK) >> PAGE_SHIFT;
+}
+
 #define pte_page(pte)  pfn_to_page(pte_pfn(pte))
 
 static inline int pmd_large(pmd_t pte)
index fc18be0..faf7a68 100644 (file)
@@ -821,6 +821,9 @@ int kern_addr_valid(unsigned long addr)
        if (pud_none(*pud))
                return 0;
 
+       if (pud_large(*pud))
+               return pfn_valid(pud_pfn(*pud));
+
        pmd = pmd_offset(pud, addr);
        if (pmd_none(*pmd))
                return 0;