splice: partial write handling fix

currently if partial write has happened while ->commit_write() then page wasn't marked as accessed and rebalanced.

index 2fca6eb..bb1bf62 100644
@@ -682,18 +682,25 @@ find_page:
 	}
 
 	ret = mapping->a_ops->commit_write(file, page, offset, offset+this_len);
-	if (!ret) {
+	if (ret) {
+		if(ret == AOP_TRUNCATED_PAGE) {
+			page_cache_release(page);
+			goto find_page;
+		}
+		if (ret < 0)
+			goto out;
 		/*
-		 * Return the number of bytes written and mark page as
-		 * accessed, we are now done!
+		 * Partial write has happened, so 'ret' already initialized by
+		 * number of bytes written, Where is nothing we have to do here.
 		 */
+	} else
 		ret = this_len;
-		mark_page_accessed(page);
-		balance_dirty_pages_ratelimited(mapping);
-	} else if (ret == AOP_TRUNCATED_PAGE) {
-		page_cache_release(page);
-		goto find_page;
-	}
+	/*
+	 * Return the number of bytes written and mark page as
+	 * accessed, we are now done!
+	 */
+	mark_page_accessed(page);
+	balance_dirty_pages_ratelimited(mapping);
 out:
 	page_cache_release(page);
 	unlock_page(page);

The ->commit_write() return values aren't very well designed imho. Is your fix correct getting the pipe_to_file() return value correct now?

Jens Axboe

I think yes, after my patch this path look like follows:

        ret = mapping->a_ops->commit_write(file, page, offset, offset+this_len);
        if (ret) {
                if(ret == AOP_TRUNCATED_PAGE) {
                        page_cache_release(page);
                        goto find_page;
                }
                if (ret < 0)
                        goto out;
                /*
                 * Partial write has happened, so 'ret' already initialized by
                 * number of bytes written, Where is nothing we have to do here.
                 */
        } else
                ret = this_len;
        /*
         * Return the number of bytes written and mark page as
         * accessed, we are now done!
         */
        mark_page_accessed(page);
        balance_dirty_pages_ratelimited(mapping);
out:
        page_cache_release(page);
        unlock_page(page);
out_ret:
        return ret;
}

But off corse this patch does't fix issues spotted by Nick.