huge pagecache: extend mremap pmd rmap lockout to files
authorHugh Dickins <hughd@google.com>
Fri, 20 May 2016 00:12:57 +0000 (17:12 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Fri, 20 May 2016 02:12:14 +0000 (19:12 -0700)
Whatever huge pagecache implementation we go with, file rmap locking
must be added to anon rmap locking, when mremap's move_page_tables()
finds a pmd_trans_huge pmd entry: a simple change, let's do it now.

Factor out take_rmap_locks() and drop_rmap_locks() to handle the locking
for make move_ptes() and move_page_tables(), and delete the
VM_BUG_ON_VMA which rejected vm_file and required anon_vma.

Signed-off-by: Hugh Dickins <hughd@google.com>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Andres Lagar-Cavilla <andreslc@google.com>
Cc: Yang Shi <yang.shi@linaro.org>
Cc: Ning Qu <quning@gmail.com>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Andres Lagar-Cavilla <andreslc@google.com>
Cc: Konstantin Khlebnikov <koct9i@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
mm/mremap.c

index 7d98fe1adc1280a07e5e0767ceddb7ed345e166d..9dc499977924d6ec3b86dbd59533c2c7bd11b8c1 100644 (file)
@@ -70,6 +70,22 @@ static pmd_t *alloc_new_pmd(struct mm_struct *mm, struct vm_area_struct *vma,
        return pmd;
 }
 
+static void take_rmap_locks(struct vm_area_struct *vma)
+{
+       if (vma->vm_file)
+               i_mmap_lock_write(vma->vm_file->f_mapping);
+       if (vma->anon_vma)
+               anon_vma_lock_write(vma->anon_vma);
+}
+
+static void drop_rmap_locks(struct vm_area_struct *vma)
+{
+       if (vma->anon_vma)
+               anon_vma_unlock_write(vma->anon_vma);
+       if (vma->vm_file)
+               i_mmap_unlock_write(vma->vm_file->f_mapping);
+}
+
 static pte_t move_soft_dirty_pte(pte_t pte)
 {
        /*
@@ -90,8 +106,6 @@ static void move_ptes(struct vm_area_struct *vma, pmd_t *old_pmd,
                struct vm_area_struct *new_vma, pmd_t *new_pmd,
                unsigned long new_addr, bool need_rmap_locks)
 {
-       struct address_space *mapping = NULL;
-       struct anon_vma *anon_vma = NULL;
        struct mm_struct *mm = vma->vm_mm;
        pte_t *old_pte, *new_pte, pte;
        spinlock_t *old_ptl, *new_ptl;
@@ -114,16 +128,8 @@ static void move_ptes(struct vm_area_struct *vma, pmd_t *old_pmd,
         *   serialize access to individual ptes, but only rmap traversal
         *   order guarantees that we won't miss both the old and new ptes).
         */
-       if (need_rmap_locks) {
-               if (vma->vm_file) {
-                       mapping = vma->vm_file->f_mapping;
-                       i_mmap_lock_write(mapping);
-               }
-               if (vma->anon_vma) {
-                       anon_vma = vma->anon_vma;
-                       anon_vma_lock_write(anon_vma);
-               }
-       }
+       if (need_rmap_locks)
+               take_rmap_locks(vma);
 
        /*
         * We don't have to worry about the ordering of src and dst
@@ -151,10 +157,8 @@ static void move_ptes(struct vm_area_struct *vma, pmd_t *old_pmd,
                spin_unlock(new_ptl);
        pte_unmap(new_pte - 1);
        pte_unmap_unlock(old_pte - 1, old_ptl);
-       if (anon_vma)
-               anon_vma_unlock_write(anon_vma);
-       if (mapping)
-               i_mmap_unlock_write(mapping);
+       if (need_rmap_locks)
+               drop_rmap_locks(vma);
 }
 
 #define LATENCY_LIMIT  (64 * PAGE_SIZE)
@@ -193,15 +197,13 @@ unsigned long move_page_tables(struct vm_area_struct *vma,
                if (pmd_trans_huge(*old_pmd)) {
                        if (extent == HPAGE_PMD_SIZE) {
                                bool moved;
-                               VM_BUG_ON_VMA(vma->vm_file || !vma->anon_vma,
-                                             vma);
                                /* See comment in move_ptes() */
                                if (need_rmap_locks)
-                                       anon_vma_lock_write(vma->anon_vma);
+                                       take_rmap_locks(vma);
                                moved = move_huge_pmd(vma, old_addr, new_addr,
                                                    old_end, old_pmd, new_pmd);
                                if (need_rmap_locks)
-                                       anon_vma_unlock_write(vma->anon_vma);
+                                       drop_rmap_locks(vma);
                                if (moved) {
                                        need_flush = true;
                                        continue;