[PATCH 1/1] crash_dump: fix non-pae kdump kernel memory accesses

[Jiri Slaby <jirislaby@gmail.com>]

Non-PAE 32-bit dump kernels may wrap an address around 4G and
poke unwanted space. ptes there are 32-bit long, and since
pfn << PAGE_SIZE may exceed this limit, high pfn bits are cropped
and wrong address mapped by kmap_atomic_pfn in copy_oldmem_page.

Don't allow this behavior in non-PAE kdump kernels by checking
pfns passed into copy_oldmem_page. In the case of failure,
userspace process gets EFAULT.

Signed-off-by: Jiri Slaby <jirislaby@gmail.com>
Cc: Vivek Goyal <vgoyal@redhat.com>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Cc: Simon Horman <horms@verge.net.au>
Cc: Paul Mundt <lethal@linux-sh.org>
Cc: Ingo Molnar <mingo@elte.hu>
---
 arch/x86/kernel/crash_dump_32.c |   16 ++++++++++++++++
 1 files changed, 16 insertions(+), 0 deletions(-)

diff --git a/arch/x86/kernel/crash_dump_32.c b/arch/x86/kernel/crash_dump_32.c
index f7cdb3b..b78dab8 100644
--- a/arch/x86/kernel/crash_dump_32.c
+++ b/arch/x86/kernel/crash_dump_32.c
@@ -16,6 +16,19 @@ static void *kdump_buf_page;
 /* Stores the physical address of elf header of crash image. */
 unsigned long long elfcorehdr_addr = ELFCORE_ADDR_MAX;
 
+#ifndef CONFIG_X86_PAE
+/* non-PAE kdump kernel executed from a PAE one will crop high pte bits and
+   poke unwanted space counting again from address 0, we don't want that */
+static inline bool is_crashed_pfn_valid(unsigned long pfn)
+{
+	/* on non-PAE pte must fit into unsigned long
+	   in fact the test is (pfn & 0x000fffff) */
+	return pte_pfn(pfn_pte(pfn, __pgprot(0))) == pfn;
+}
+#else
+static inline bool is_crashed_pfn_valid(unsigned long pfn) { return true; }
+#endif
+
 /**
  * copy_oldmem_page - copy one page from "oldmem"
  * @pfn: page frame number to be copied
@@ -41,6 +54,9 @@ ssize_t copy_oldmem_page(unsigned long pfn, char *buf,
 	if (!csize)
 		return 0;
 
+	if (!is_crashed_pfn_valid(pfn))
+		return -EFAULT;
+
 	vaddr = kmap_atomic_pfn(pfn, KM_PTE0);
 
 	if (!userbuf) {
-- 
1.6.4.2