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>

1 files changed, 44 insertions, 8 deletions

diff --git a/fs/aio.c b/fs/aio.c
index fd1c0baf15bb..efa708b29054 100644
--- 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