memcgroup: reinstate swapoff mod

This patch reinstates the "swapoff: scan ptes preemptibly" mod we started
with: in due course it should be rendered down into the earlier patches,
leaving us with a more straightforward mem_cgroup_charge mod to unuse_pte,
allocating with GFP_KERNEL while holding no spinlock and no atomic kmap.

Signed-off-by: Hugh Dickins <hugh@veritas.com>
Cc: Pavel Emelianov <xemul@openvz.org>
Acked-by: Balbir Singh <balbir@linux.vnet.ibm.com>
Cc: Paul Menage <menage@google.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Cc: Kirill Korotaev <dev@sw.ru>
Cc: Herbert Poetzl <herbert@13thfloor.at>
Cc: David Rientjes <rientjes@google.com>
Cc: Vaidyanathan Srinivasan <svaidy@linux.vnet.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

diff --git a/mm/swapfile.c b/mm/swapfile.c
index 35e00c3d0286d94ea2e748796168e442d048346c..02ccab5ad9d9f2fdebfb525ed6a1b653266dc642 100644 (file)
--- a/mm/swapfile.c
+++ b/mm/swapfile.c
@@ -507,11 +507,23 @@ unsigned int count_swap_pages(int type, int free)
  * just let do_wp_page work it out if a write is requested later - to
  * force COW, vm_page_prot omits write permission from any private vma.
  */
-static int unuse_pte(struct vm_area_struct *vma, pte_t *pte,
+static int unuse_pte(struct vm_area_struct *vma, pmd_t *pmd,
                unsigned long addr, swp_entry_t entry, struct page *page)
 {
+       spinlock_t *ptl;
+       pte_t *pte;
+       int ret = 1;
+
        if (mem_cgroup_charge(page, vma->vm_mm, GFP_KERNEL))
-               return -ENOMEM;
+               ret = -ENOMEM;
+
+       pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
+       if (unlikely(!pte_same(*pte, swp_entry_to_pte(entry)))) {
+               if (ret > 0)
+                       mem_cgroup_uncharge_page(page);
+               ret = 0;
+               goto out;
+       }
 
        inc_mm_counter(vma->vm_mm, anon_rss);
        get_page(page);
@@ -524,7 +536,9 @@ static int unuse_pte(struct vm_area_struct *vma, pte_t *pte,
         * immediately swapped out again after swapon.
         */
        activate_page(page);
-       return 1;
+out:
+       pte_unmap_unlock(pte, ptl);
+       return ret;
 }
 
 static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
@@ -533,21 +547,33 @@ static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
 {
        pte_t swp_pte = swp_entry_to_pte(entry);
        pte_t *pte;
-       spinlock_t *ptl;
        int ret = 0;
 
-       pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
+       /*
+        * We don't actually need pte lock while scanning for swp_pte: since
+        * we hold page lock and mmap_sem, swp_pte cannot be inserted into the
+        * page table while we're scanning; though it could get zapped, and on
+        * some architectures (e.g. x86_32 with PAE) we might catch a glimpse
+        * of unmatched parts which look like swp_pte, so unuse_pte must
+        * recheck under pte lock.  Scanning without pte lock lets it be
+        * preemptible whenever CONFIG_PREEMPT but not CONFIG_HIGHPTE.
+        */
+       pte = pte_offset_map(pmd, addr);
        do {
                /*
                 * swapoff spends a _lot_ of time in this loop!
                 * Test inline before going to call unuse_pte.
                 */
                if (unlikely(pte_same(*pte, swp_pte))) {
-                       ret = unuse_pte(vma, pte++, addr, entry, page);
-                       break;
+                       pte_unmap(pte);
+                       ret = unuse_pte(vma, pmd, addr, entry, page);
+                       if (ret)
+                               goto out;
+                       pte = pte_offset_map(pmd, addr);
                }
        } while (pte++, addr += PAGE_SIZE, addr != end);
-       pte_unmap_unlock(pte - 1, ptl);
+       pte_unmap(pte - 1);
+out:
        return ret;
 }