[PATCH] unpaged: COW on VM_UNPAGED
authorHugh Dickins <hugh@veritas.com>
Tue, 22 Nov 2005 05:32:17 +0000 (21:32 -0800)
committerLinus Torvalds <torvalds@g5.osdl.org>
Tue, 22 Nov 2005 17:13:42 +0000 (09:13 -0800)
Remove the BUG_ON(vma->vm_flags & VM_UNPAGED) from do_wp_page, and let it do
Copy-On-Write without touching the VM_UNPAGED's page counts - but this is
incomplete, because the anonymous page it inserts will itself need to be
handled, here and in other functions - next patch.

We still don't copy the page if the pfn is invalid, because the
copy_user_highpage interface does not allow it.  But that's not been a problem
in the past: can be added in later if the need arises.

Signed-off-by: Hugh Dickins <hugh@veritas.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
mm/memory.c

index ece04963158e4c61cb89c0bd38736d2dc59f76d1..107b619cfb16767624522a722d545426c8530eeb 100644 (file)
@@ -1277,22 +1277,28 @@ static int do_wp_page(struct mm_struct *mm, struct vm_area_struct *vma,
                unsigned long address, pte_t *page_table, pmd_t *pmd,
                spinlock_t *ptl, pte_t orig_pte)
 {
-       struct page *old_page, *new_page;
+       struct page *old_page, *src_page, *new_page;
        unsigned long pfn = pte_pfn(orig_pte);
        pte_t entry;
        int ret = VM_FAULT_MINOR;
 
-       BUG_ON(vma->vm_flags & VM_UNPAGED);
-
        if (unlikely(!pfn_valid(pfn))) {
                /*
                 * Page table corrupted: show pte and kill process.
+                * Or it's an attempt to COW an out-of-map VM_UNPAGED
+                * entry, which copy_user_highpage does not support.
                 */
                print_bad_pte(vma, orig_pte, address);
                ret = VM_FAULT_OOM;
                goto unlock;
        }
        old_page = pfn_to_page(pfn);
+       src_page = old_page;
+
+       if (unlikely(vma->vm_flags & VM_UNPAGED)) {
+               old_page = NULL;
+               goto gotten;
+       }
 
        if (PageAnon(old_page) && !TestSetPageLocked(old_page)) {
                int reuse = can_share_swap_page(old_page);
@@ -1313,11 +1319,12 @@ static int do_wp_page(struct mm_struct *mm, struct vm_area_struct *vma,
         * Ok, we need to copy. Oh, well..
         */
        page_cache_get(old_page);
+gotten:
        pte_unmap_unlock(page_table, ptl);
 
        if (unlikely(anon_vma_prepare(vma)))
                goto oom;
-       if (old_page == ZERO_PAGE(address)) {
+       if (src_page == ZERO_PAGE(address)) {
                new_page = alloc_zeroed_user_highpage(vma, address);
                if (!new_page)
                        goto oom;
@@ -1325,7 +1332,7 @@ static int do_wp_page(struct mm_struct *mm, struct vm_area_struct *vma,
                new_page = alloc_page_vma(GFP_HIGHUSER, vma, address);
                if (!new_page)
                        goto oom;
-               copy_user_highpage(new_page, old_page, address);
+               copy_user_highpage(new_page, src_page, address);
        }
 
        /*
@@ -1333,11 +1340,14 @@ static int do_wp_page(struct mm_struct *mm, struct vm_area_struct *vma,
         */
        page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
        if (likely(pte_same(*page_table, orig_pte))) {
-               page_remove_rmap(old_page);
-               if (!PageAnon(old_page)) {
+               if (old_page) {
+                       page_remove_rmap(old_page);
+                       if (!PageAnon(old_page)) {
+                               dec_mm_counter(mm, file_rss);
+                               inc_mm_counter(mm, anon_rss);
+                       }
+               } else
                        inc_mm_counter(mm, anon_rss);
-                       dec_mm_counter(mm, file_rss);
-               }
                flush_cache_page(vma, address, pfn);
                entry = mk_pte(new_page, vma->vm_page_prot);
                entry = maybe_mkwrite(pte_mkdirty(entry), vma);
@@ -1351,13 +1361,16 @@ static int do_wp_page(struct mm_struct *mm, struct vm_area_struct *vma,
                new_page = old_page;
                ret |= VM_FAULT_WRITE;
        }
-       page_cache_release(new_page);
-       page_cache_release(old_page);
+       if (new_page)
+               page_cache_release(new_page);
+       if (old_page)
+               page_cache_release(old_page);
 unlock:
        pte_unmap_unlock(page_table, ptl);
        return ret;
 oom:
-       page_cache_release(old_page);
+       if (old_page)
+               page_cache_release(old_page);
        return VM_FAULT_OOM;
 }