mm: remove VM_LOCK_RMAP code
Rik van Riel [Fri, 5 Mar 2010 21:42:10 +0000 (13:42 -0800)]
When a VMA is in an inconsistent state during setup or teardown, the worst
that can happen is that the rmap code will not be able to find the page.

The mapping is in the process of being torn down (PTEs just got
invalidated by munmap), or set up (no PTEs have been instantiated yet).

It is also impossible for the rmap code to follow a pointer to an already
freed VMA, because the rmap code holds the anon_vma->lock, which the VMA
teardown code needs to take before the VMA is removed from the anon_vma
chain.

Hence, we should not need the VM_LOCK_RMAP locking at all.

Signed-off-by: Rik van Riel <riel@redhat.com>
Cc: Nick Piggin <npiggin@suse.de>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Larry Woodman <lwoodman@redhat.com>
Cc: Lee Schermerhorn <Lee.Schermerhorn@hp.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

include/linux/mm.h
mm/mmap.c
mm/rmap.c

index 8e2841a..3899395 100644 (file)
@@ -97,11 +97,7 @@ extern unsigned int kobjsize(const void *objp);
 #define VM_NORESERVE   0x00200000      /* should the VM suppress accounting */
 #define VM_HUGETLB     0x00400000      /* Huge TLB Page VM */
 #define VM_NONLINEAR   0x00800000      /* Is non-linear (remap_file_pages) */
-#ifdef CONFIG_MMU
-#define VM_LOCK_RMAP   0x01000000      /* Do not follow this rmap (mmu mmap) */
-#else
 #define VM_MAPPED_COPY 0x01000000      /* T if mapped copy of data (nommu mmap) */
-#endif
 #define VM_INSERTPAGE  0x02000000      /* The vma has had "vm_insert_page()" done on it */
 #define VM_ALWAYSDUMP  0x04000000      /* Always include in core dumps */
 
index 6a0c15d..f1b4448 100644 (file)
--- a/mm/mmap.c
+++ b/mm/mmap.c
@@ -554,9 +554,7 @@ again:                      remove_next = 1 + (end > next->vm_end);
                 */
                if (importer && !importer->anon_vma) {
                        /* Block reverse map lookups until things are set up. */
-                       importer->vm_flags |= VM_LOCK_RMAP;
                        if (anon_vma_clone(importer, vma)) {
-                               importer->vm_flags &= ~VM_LOCK_RMAP;
                                return -ENOMEM;
                        }
                        importer->anon_vma = anon_vma;
@@ -618,11 +616,6 @@ again:                     remove_next = 1 + (end > next->vm_end);
                __vma_unlink(mm, next, vma);
                if (file)
                        __remove_shared_vm_struct(next, file, mapping);
-               /*
-                * This VMA is now dead, no need for rmap to follow it.
-                * Call anon_vma_merge below, outside of i_mmap_lock.
-                */
-               next->vm_flags |= VM_LOCK_RMAP;
        } else if (insert) {
                /*
                 * split_vma has split insert from vma, and needs
@@ -635,20 +628,12 @@ again:                    remove_next = 1 + (end > next->vm_end);
        if (mapping)
                spin_unlock(&mapping->i_mmap_lock);
 
-       /*
-        * The current VMA has been set up. It is now safe for the
-        * rmap code to get from the pages to the ptes.
-        */
-       if (anon_vma && importer)
-               importer->vm_flags &= ~VM_LOCK_RMAP;
-
        if (remove_next) {
                if (file) {
                        fput(file);
                        if (next->vm_flags & VM_EXECUTABLE)
                                removed_exe_file_vma(mm);
                }
-               /* Protected by mmap_sem and VM_LOCK_RMAP. */
                if (next->anon_vma)
                        anon_vma_merge(vma, next);
                mm->map_count--;
index 28bcdc4..4d2fb93 100644 (file)
--- a/mm/rmap.c
+++ b/mm/rmap.c
@@ -329,18 +329,6 @@ vma_address(struct page *page, struct vm_area_struct *vma)
                /* page should be within @vma mapping range */
                return -EFAULT;
        }
-       if (unlikely(vma->vm_flags & VM_LOCK_RMAP)) {
-               /*
-                * This VMA is being unlinked or is not yet linked into the
-                * VMA tree.  Do not try to follow this rmap.  This race
-                * condition can result in page_referenced() ignoring a
-                * reference or in try_to_unmap() failing to unmap a page.
-                * The VMA cannot be freed under us because we hold the
-                * anon_vma->lock, which the munmap code takes while
-                * unlinking the anon_vmas from the VMA.
-                */
-               return -EFAULT;
-       }
        return address;
 }