aio/migratepages: make aio migrate pages sane

The arbitrary restriction on page counts offered by the core
migrate_page_move_mapping() code results in rather suspicious looking
fiddling with page reference counts in the aio_migratepage() operation.
To fix this, make migrate_page_move_mapping() take an extra_count parameter
that allows aio to tell the code about its own reference count on the page
being migrated.

While cleaning up aio_migratepage(), make it validate that the old page
being passed in is actually what aio_migratepage() expects to prevent
misbehaviour in the case of races.

Signed-off-by: Benjamin LaHaise <bcrl@kvack.org>

diff --git a/fs/aio.c b/fs/aio.c
index fd1c0ba..efa708b 100644 (file)
--- a/fs/aio.c
+++ b/fs/aio.c
@@ -244,9 +244,14 @@ static void aio_free_ring(struct kioctx *ctx)
        int i;
 
        for (i = 0; i < ctx->nr_pages; i++) {
+               struct page *page;
                pr_debug("pid(%d) [%d] page->count=%d\n", current->pid, i,
                                page_count(ctx->ring_pages[i]));
-               put_page(ctx->ring_pages[i]);
+               page = ctx->ring_pages[i];
+               if (!page)
+                       continue;
+               ctx->ring_pages[i] = NULL;
+               put_page(page);
        }
 
        put_aio_ring_file(ctx);
@@ -280,18 +285,38 @@ static int aio_migratepage(struct address_space *mapping, struct page *new,
        unsigned long flags;
        int rc;
 
+       rc = 0;
+
+       /* Make sure the old page hasn't already been changed */
+       spin_lock(&mapping->private_lock);
+       ctx = mapping->private_data;
+       if (ctx) {
+               pgoff_t idx;
+               spin_lock_irqsave(&ctx->completion_lock, flags);
+               idx = old->index;
+               if (idx < (pgoff_t)ctx->nr_pages) {
+                       if (ctx->ring_pages[idx] != old)
+                               rc = -EAGAIN;
+               } else
+                       rc = -EINVAL;
+               spin_unlock_irqrestore(&ctx->completion_lock, flags);
+       } else
+               rc = -EINVAL;
+       spin_unlock(&mapping->private_lock);
+
+       if (rc != 0)
+               return rc;
+
        /* Writeback must be complete */
        BUG_ON(PageWriteback(old));
-       put_page(old);
+       get_page(new);
 
-       rc = migrate_page_move_mapping(mapping, new, old, NULL, mode);
+       rc = migrate_page_move_mapping(mapping, new, old, NULL, mode, 1);
        if (rc != MIGRATEPAGE_SUCCESS) {
-               get_page(old);
+               put_page(new);
                return rc;
        }
 
-       get_page(new);
-
        /* We can potentially race against kioctx teardown here.  Use the
         * address_space's private data lock to protect the mapping's
         * private_data.
@@ -303,13 +328,24 @@ static int aio_migratepage(struct address_space *mapping, struct page *new,
                spin_lock_irqsave(&ctx->completion_lock, flags);
                migrate_page_copy(new, old);
                idx = old->index;
-               if (idx < (pgoff_t)ctx->nr_pages)
-                       ctx->ring_pages[idx] = new;
+               if (idx < (pgoff_t)ctx->nr_pages) {
+                       /* And only do the move if things haven't changed */
+                       if (ctx->ring_pages[idx] == old)
+                               ctx->ring_pages[idx] = new;
+                       else
+                               rc = -EAGAIN;
+               } else
+                       rc = -EINVAL;
                spin_unlock_irqrestore(&ctx->completion_lock, flags);
        } else
                rc = -EBUSY;
        spin_unlock(&mapping->private_lock);
 
+       if (rc == MIGRATEPAGE_SUCCESS)
+               put_page(old);
+       else
+               put_page(new);
+
        return rc;
 }
 #endif

diff --git a/include/linux/migrate.h b/include/linux/migrate.h
index b7717d7..f015c05 100644 (file)
--- a/include/linux/migrate.h
+++ b/include/linux/migrate.h
@@ -55,7 +55,8 @@ extern int migrate_huge_page_move_mapping(struct address_space *mapping,
                                  struct page *newpage, struct page *page);
 extern int migrate_page_move_mapping(struct address_space *mapping,
                struct page *newpage, struct page *page,
-               struct buffer_head *head, enum migrate_mode mode);
+               struct buffer_head *head, enum migrate_mode mode,
+               int extra_count);
 #else
 
 static inline void putback_lru_pages(struct list_head *l) {}

diff --git a/mm/migrate.c b/mm/migrate.c
index e9b7102..9194375 100644 (file)
--- a/mm/migrate.c
+++ b/mm/migrate.c
@@ -317,14 +317,15 @@ static inline bool buffer_migrate_lock_buffers(struct buffer_head *head,
  */
 int migrate_page_move_mapping(struct address_space *mapping,
                struct page *newpage, struct page *page,
-               struct buffer_head *head, enum migrate_mode mode)
+               struct buffer_head *head, enum migrate_mode mode,
+               int extra_count)
 {
-       int expected_count = 0;
+       int expected_count = 1 + extra_count;
        void **pslot;
 
        if (!mapping) {
                /* Anonymous page without mapping */
-               if (page_count(page) != 1)
+               if (page_count(page) != expected_count)
                        return -EAGAIN;
                return MIGRATEPAGE_SUCCESS;
        }
@@ -334,7 +335,7 @@ int migrate_page_move_mapping(struct address_space *mapping,
        pslot = radix_tree_lookup_slot(&mapping->page_tree,
                                        page_index(page));
 
-       expected_count = 2 + page_has_private(page);
+       expected_count += 1 + page_has_private(page);
        if (page_count(page) != expected_count ||
                radix_tree_deref_slot_protected(pslot, &mapping->tree_lock) != page) {
                spin_unlock_irq(&mapping->tree_lock);
@@ -584,7 +585,7 @@ int migrate_page(struct address_space *mapping,
 
        BUG_ON(PageWriteback(page));    /* Writeback must be complete */
 
-       rc = migrate_page_move_mapping(mapping, newpage, page, NULL, mode);
+       rc = migrate_page_move_mapping(mapping, newpage, page, NULL, mode, 0);
 
        if (rc != MIGRATEPAGE_SUCCESS)
                return rc;
@@ -611,7 +612,7 @@ int buffer_migrate_page(struct address_space *mapping,
 
        head = page_buffers(page);
 
-       rc = migrate_page_move_mapping(mapping, newpage, page, head, mode);
+       rc = migrate_page_move_mapping(mapping, newpage, page, head, mode, 0);
 
        if (rc != MIGRATEPAGE_SUCCESS)
                return rc;