mm: hugetlb: fix hugepage memory leak in mincore()
authorNaoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Tue, 15 Dec 2009 01:59:58 +0000 (17:59 -0800)
committerLinus Torvalds <torvalds@linux-foundation.org>
Tue, 15 Dec 2009 16:53:24 +0000 (08:53 -0800)
Most callers of pmd_none_or_clear_bad() check whether the target page is
in a hugepage or not, but mincore() and walk_page_range() do not check it.
 So if we use mincore() on a hugepage on x86 machine, the hugepage memory
is leaked as shown below.  This patch fixes it by extending mincore()
system call to support hugepages.

Details
=======
My test program (leak_mincore) works as follows:
 - creat() and mmap() a file on hugetlbfs (file size is 200MB == 100 hugepages,)
 - read()/write() something on it,
 - call mincore() for first ten pages and printf() the values of *vec
 - munmap() and unlink() the file on hugetlbfs

Without my patch
----------------
$ cat /proc/meminfo| grep "HugePage"
HugePages_Total:    1000
HugePages_Free:     1000
HugePages_Rsvd:        0
HugePages_Surp:        0
$ ./leak_mincore
vec[0] 0
vec[1] 0
vec[2] 0
vec[3] 0
vec[4] 0
vec[5] 0
vec[6] 0
vec[7] 0
vec[8] 0
vec[9] 0
$ cat /proc/meminfo |grep "HugePage"
HugePages_Total:    1000
HugePages_Free:      999
HugePages_Rsvd:        0
HugePages_Surp:        0
$ ls /hugetlbfs/
$

Return values in *vec from mincore() are set to 0, while the hugepage
should be in memory, and 1 hugepage is still accounted as used while
there is no file on hugetlbfs.

With my patch
-------------
$ cat /proc/meminfo| grep "HugePage"
HugePages_Total:    1000
HugePages_Free:     1000
HugePages_Rsvd:        0
HugePages_Surp:        0
$ ./leak_mincore
vec[0] 1
vec[1] 1
vec[2] 1
vec[3] 1
vec[4] 1
vec[5] 1
vec[6] 1
vec[7] 1
vec[8] 1
vec[9] 1
$ cat /proc/meminfo |grep "HugePage"
HugePages_Total:    1000
HugePages_Free:     1000
HugePages_Rsvd:        0
HugePages_Surp:        0
$ ls /hugetlbfs/
$

Return value in *vec set to 1 and no memory leaks.

[akpm@linux-foundation.org: cleanup]
[akpm@linux-foundation.org: build fix]
Signed-off-by: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Wu Fengguang <fengguang.wu@intel.com>
Cc: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Cc: Mel Gorman <mel@csn.ul.ie>
Cc: Lee Schermerhorn <lee.schermerhorn@hp.com>
Cc: Andy Whitcroft <apw@canonical.com>
Cc: David Rientjes <rientjes@google.com>
Cc: <stable@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
mm/mincore.c

index 8cb508f..7a3436e 100644 (file)
@@ -14,6 +14,7 @@
 #include <linux/syscalls.h>
 #include <linux/swap.h>
 #include <linux/swapops.h>
+#include <linux/hugetlb.h>
 
 #include <asm/uaccess.h>
 #include <asm/pgtable.h>
@@ -72,6 +73,42 @@ static long do_mincore(unsigned long addr, unsigned char *vec, unsigned long pag
        if (!vma || addr < vma->vm_start)
                return -ENOMEM;
 
+#ifdef CONFIG_HUGETLB_PAGE
+       if (is_vm_hugetlb_page(vma)) {
+               struct hstate *h;
+               unsigned long nr_huge;
+               unsigned char present;
+
+               i = 0;
+               nr = min(pages, (vma->vm_end - addr) >> PAGE_SHIFT);
+               h = hstate_vma(vma);
+               nr_huge = ((addr + pages * PAGE_SIZE - 1) >> huge_page_shift(h))
+                         - (addr >> huge_page_shift(h)) + 1;
+               nr_huge = min(nr_huge,
+                             (vma->vm_end - addr) >> huge_page_shift(h));
+               while (1) {
+                       /* hugepage always in RAM for now,
+                        * but generally it needs to be check */
+                       ptep = huge_pte_offset(current->mm,
+                                              addr & huge_page_mask(h));
+                       present = !!(ptep &&
+                                    !huge_pte_none(huge_ptep_get(ptep)));
+                       while (1) {
+                               vec[i++] = present;
+                               addr += PAGE_SIZE;
+                               /* reach buffer limit */
+                               if (i == nr)
+                                       return nr;
+                               /* check hugepage border */
+                               if (!((addr & ~huge_page_mask(h))
+                                     >> PAGE_SHIFT))
+                                       break;
+                       }
+               }
+               return nr;
+       }
+#endif
+
        /*
         * Calculate how many pages there are left in the last level of the
         * PTE array for our address.