mm: consistent truncate and invalidate loops
Hugh Dickins [Tue, 26 Jul 2011 00:12:25 +0000 (17:12 -0700)]
Make the pagevec_lookup loops in truncate_inode_pages_range(),
invalidate_mapping_pages() and invalidate_inode_pages2_range() more
consistent with each other.

They were relying upon page->index of an unlocked page, but apologizing
for it: accept it, embrace it, add comments and WARN_ONs, and simplify the
index handling.

invalidate_inode_pages2_range() had special handling for a wrapped
page->index + 1 = 0 case; but MAX_LFS_FILESIZE doesn't let us anywhere
near there, and a corrupt page->index in the radix_tree could cause more
trouble than that would catch.  Remove that wrapped handling.

invalidate_inode_pages2_range() uses min() to limit the pagevec_lookup
when near the end of the range: copy that into the other two, although
it's less useful than you might think (it limits the use of the buffer,
rather than the indices looked up).

Signed-off-by: Hugh Dickins <hughd@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/filemap.c
mm/truncate.c

index 2780be4..10a1711 100644 (file)
@@ -128,6 +128,7 @@ void __delete_from_page_cache(struct page *page)
 
        radix_tree_delete(&mapping->page_tree, page->index);
        page->mapping = NULL;
+       /* Leave page->index set: truncation lookup relies upon it */
        mapping->nrpages--;
        __dec_zone_page_state(page, NR_FILE_PAGES);
        if (PageSwapBacked(page))
@@ -483,6 +484,7 @@ int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
                        spin_unlock_irq(&mapping->tree_lock);
                } else {
                        page->mapping = NULL;
+                       /* Leave page->index set: truncation relies upon it */
                        spin_unlock_irq(&mapping->tree_lock);
                        mem_cgroup_uncharge_cache_page(page);
                        page_cache_release(page);
index c924764..dc45901 100644 (file)
@@ -199,9 +199,6 @@ int invalidate_inode_page(struct page *page)
  * The first pass will remove most pages, so the search cost of the second pass
  * is low.
  *
- * When looking at page->index outside the page lock we need to be careful to
- * copy it into a local to avoid races (it could change at any time).
- *
  * We pass down the cache-hot hint to the page freeing code.  Even if the
  * mapping is large, it is probably the case that the final pages are the most
  * recently touched, and freeing happens in ascending file offset order.
@@ -210,10 +207,10 @@ void truncate_inode_pages_range(struct address_space *mapping,
                                loff_t lstart, loff_t lend)
 {
        const pgoff_t start = (lstart + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
-       pgoff_t end;
        const unsigned partial = lstart & (PAGE_CACHE_SIZE - 1);
        struct pagevec pvec;
-       pgoff_t next;
+       pgoff_t index;
+       pgoff_t end;
        int i;
 
        cleancache_flush_inode(mapping);
@@ -224,24 +221,21 @@ void truncate_inode_pages_range(struct address_space *mapping,
        end = (lend >> PAGE_CACHE_SHIFT);
 
        pagevec_init(&pvec, 0);
-       next = start;
-       while (next <= end &&
-              pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
+       index = start;
+       while (index <= end && pagevec_lookup(&pvec, mapping, index,
+                       min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1)) {
                mem_cgroup_uncharge_start();
                for (i = 0; i < pagevec_count(&pvec); i++) {
                        struct page *page = pvec.pages[i];
-                       pgoff_t page_index = page->index;
 
-                       if (page_index > end) {
-                               next = page_index;
+                       /* We rely upon deletion not changing page->index */
+                       index = page->index;
+                       if (index > end)
                                break;
-                       }
 
-                       if (page_index > next)
-                               next = page_index;
-                       next++;
                        if (!trylock_page(page))
                                continue;
+                       WARN_ON(page->index != index);
                        if (PageWriteback(page)) {
                                unlock_page(page);
                                continue;
@@ -252,6 +246,7 @@ void truncate_inode_pages_range(struct address_space *mapping,
                pagevec_release(&pvec);
                mem_cgroup_uncharge_end();
                cond_resched();
+               index++;
        }
 
        if (partial) {
@@ -264,13 +259,14 @@ void truncate_inode_pages_range(struct address_space *mapping,
                }
        }
 
-       next = start;
+       index = start;
        for ( ; ; ) {
                cond_resched();
-               if (!pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
-                       if (next == start)
+               if (!pagevec_lookup(&pvec, mapping, index,
+                       min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1)) {
+                       if (index == start)
                                break;
-                       next = start;
+                       index = start;
                        continue;
                }
                if (pvec.pages[0]->index > end) {
@@ -281,18 +277,20 @@ void truncate_inode_pages_range(struct address_space *mapping,
                for (i = 0; i < pagevec_count(&pvec); i++) {
                        struct page *page = pvec.pages[i];
 
-                       if (page->index > end)
+                       /* We rely upon deletion not changing page->index */
+                       index = page->index;
+                       if (index > end)
                                break;
+
                        lock_page(page);
+                       WARN_ON(page->index != index);
                        wait_on_page_writeback(page);
                        truncate_inode_page(mapping, page);
-                       if (page->index > next)
-                               next = page->index;
-                       next++;
                        unlock_page(page);
                }
                pagevec_release(&pvec);
                mem_cgroup_uncharge_end();
+               index++;
        }
        cleancache_flush_inode(mapping);
 }
@@ -333,35 +331,26 @@ unsigned long invalidate_mapping_pages(struct address_space *mapping,
                pgoff_t start, pgoff_t end)
 {
        struct pagevec pvec;
-       pgoff_t next = start;
+       pgoff_t index = start;
        unsigned long ret;
        unsigned long count = 0;
        int i;
 
        pagevec_init(&pvec, 0);
-       while (next <= end &&
-                       pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
+       while (index <= end && pagevec_lookup(&pvec, mapping, index,
+                       min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1)) {
                mem_cgroup_uncharge_start();
                for (i = 0; i < pagevec_count(&pvec); i++) {
                        struct page *page = pvec.pages[i];
-                       pgoff_t index;
-                       int lock_failed;
-
-                       lock_failed = !trylock_page(page);
 
-                       /*
-                        * We really shouldn't be looking at the ->index of an
-                        * unlocked page.  But we're not allowed to lock these
-                        * pages.  So we rely upon nobody altering the ->index
-                        * of this (pinned-by-us) page.
-                        */
+                       /* We rely upon deletion not changing page->index */
                        index = page->index;
-                       if (index > next)
-                               next = index;
-                       next++;
-                       if (lock_failed)
-                               continue;
+                       if (index > end)
+                               break;
 
+                       if (!trylock_page(page))
+                               continue;
+                       WARN_ON(page->index != index);
                        ret = invalidate_inode_page(page);
                        unlock_page(page);
                        /*
@@ -371,12 +360,11 @@ unsigned long invalidate_mapping_pages(struct address_space *mapping,
                        if (!ret)
                                deactivate_page(page);
                        count += ret;
-                       if (next > end)
-                               break;
                }
                pagevec_release(&pvec);
                mem_cgroup_uncharge_end();
                cond_resched();
+               index++;
        }
        return count;
 }
@@ -442,37 +430,32 @@ int invalidate_inode_pages2_range(struct address_space *mapping,
                                  pgoff_t start, pgoff_t end)
 {
        struct pagevec pvec;
-       pgoff_t next;
+       pgoff_t index;
        int i;
        int ret = 0;
        int ret2 = 0;
        int did_range_unmap = 0;
-       int wrapped = 0;
 
        cleancache_flush_inode(mapping);
        pagevec_init(&pvec, 0);
-       next = start;
-       while (next <= end && !wrapped &&
-               pagevec_lookup(&pvec, mapping, next,
-                       min(end - next, (pgoff_t)PAGEVEC_SIZE - 1) + 1)) {
+       index = start;
+       while (index <= end && pagevec_lookup(&pvec, mapping, index,
+                       min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1)) {
                mem_cgroup_uncharge_start();
                for (i = 0; i < pagevec_count(&pvec); i++) {
                        struct page *page = pvec.pages[i];
-                       pgoff_t page_index;
+
+                       /* We rely upon deletion not changing page->index */
+                       index = page->index;
+                       if (index > end)
+                               break;
 
                        lock_page(page);
+                       WARN_ON(page->index != index);
                        if (page->mapping != mapping) {
                                unlock_page(page);
                                continue;
                        }
-                       page_index = page->index;
-                       next = page_index + 1;
-                       if (next == 0)
-                               wrapped = 1;
-                       if (page_index > end) {
-                               unlock_page(page);
-                               break;
-                       }
                        wait_on_page_writeback(page);
                        if (page_mapped(page)) {
                                if (!did_range_unmap) {
@@ -480,9 +463,9 @@ int invalidate_inode_pages2_range(struct address_space *mapping,
                                         * Zap the rest of the file in one hit.
                                         */
                                        unmap_mapping_range(mapping,
-                                          (loff_t)page_index<<PAGE_CACHE_SHIFT,
-                                          (loff_t)(end - page_index + 1)
-                                                       << PAGE_CACHE_SHIFT,
+                                          (loff_t)index << PAGE_CACHE_SHIFT,
+                                          (loff_t)(1 + end - index)
+                                                        << PAGE_CACHE_SHIFT,
                                            0);
                                        did_range_unmap = 1;
                                } else {
@@ -490,8 +473,8 @@ int invalidate_inode_pages2_range(struct address_space *mapping,
                                         * Just zap this page
                                         */
                                        unmap_mapping_range(mapping,
-                                         (loff_t)page_index<<PAGE_CACHE_SHIFT,
-                                         PAGE_CACHE_SIZE, 0);
+                                          (loff_t)index << PAGE_CACHE_SHIFT,
+                                          PAGE_CACHE_SIZE, 0);
                                }
                        }
                        BUG_ON(page_mapped(page));
@@ -507,6 +490,7 @@ int invalidate_inode_pages2_range(struct address_space *mapping,
                pagevec_release(&pvec);
                mem_cgroup_uncharge_end();
                cond_resched();
+               index++;
        }
        cleancache_flush_inode(mapping);
        return ret;