rbd: fix parent request size assumption
authorAlex Elder <elder@inktank.com>
Fri, 10 May 2013 21:29:22 +0000 (16:29 -0500)
committerAlex Elder <elder@inktank.com>
Mon, 13 May 2013 19:09:01 +0000 (14:09 -0500)
The code that reads object data from the parent for a copyup on
write request currently assumes that the size of that request is the
size of a "full" object from the original target image.

That is not necessarily the case.  The parent overlap could reduce
the request size below that.  To fix that assumption we need to
record the number of pages in the copyup_pages array, for both an
image request and an object request.  Rename a local variable in
rbd_img_obj_parent_read_full_callback() to reflect we're recording
the length of the parent read request, not the size of the target
object.

This resolves:
    http://tracker.ceph.com/issues/5038

Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
drivers/block/rbd.c

index 51c45e7..597b9bb 100644 (file)
@@ -224,6 +224,7 @@ struct rbd_obj_request {
                };
        };
        struct page             **copyup_pages;
+       u32                     copyup_page_count;
 
        struct ceph_osd_request *osd_req;
 
@@ -256,6 +257,7 @@ struct rbd_img_request {
                struct rbd_obj_request  *obj_request;   /* obj req initiator */
        };
        struct page             **copyup_pages;
+       u32                     copyup_page_count;
        spinlock_t              completion_lock;/* protects next_completion */
        u32                     next_completion;
        rbd_img_callback_t      callback;
@@ -2119,7 +2121,7 @@ rbd_img_obj_copyup_callback(struct rbd_obj_request *obj_request)
 {
        struct rbd_img_request *img_request;
        struct rbd_device *rbd_dev;
-       u64 length;
+       struct page **pages;
        u32 page_count;
 
        rbd_assert(obj_request->type == OBJ_REQUEST_BIO);
@@ -2129,12 +2131,14 @@ rbd_img_obj_copyup_callback(struct rbd_obj_request *obj_request)
 
        rbd_dev = img_request->rbd_dev;
        rbd_assert(rbd_dev);
-       length = (u64)1 << rbd_dev->header.obj_order;
-       page_count = (u32)calc_pages_for(0, length);
 
-       rbd_assert(obj_request->copyup_pages);
-       ceph_release_page_vector(obj_request->copyup_pages, page_count);
+       pages = obj_request->copyup_pages;
+       rbd_assert(pages != NULL);
        obj_request->copyup_pages = NULL;
+       page_count = obj_request->copyup_page_count;
+       rbd_assert(page_count);
+       obj_request->copyup_page_count = 0;
+       ceph_release_page_vector(pages, page_count);
 
        /*
         * We want the transfer count to reflect the size of the
@@ -2158,9 +2162,9 @@ rbd_img_obj_parent_read_full_callback(struct rbd_img_request *img_request)
        struct ceph_osd_client *osdc;
        struct rbd_device *rbd_dev;
        struct page **pages;
+       u32 page_count;
        int result;
-       u64 obj_size;
-       u64 xferred;
+       u64 parent_length;
 
        rbd_assert(img_request_child_test(img_request));
 
@@ -2169,19 +2173,21 @@ rbd_img_obj_parent_read_full_callback(struct rbd_img_request *img_request)
        pages = img_request->copyup_pages;
        rbd_assert(pages != NULL);
        img_request->copyup_pages = NULL;
+       page_count = img_request->copyup_page_count;
+       rbd_assert(page_count);
+       img_request->copyup_page_count = 0;
 
        orig_request = img_request->obj_request;
        rbd_assert(orig_request != NULL);
        rbd_assert(orig_request->type == OBJ_REQUEST_BIO);
        result = img_request->result;
-       obj_size = img_request->length;
-       xferred = img_request->xferred;
+       parent_length = img_request->length;
+       rbd_assert(parent_length == img_request->xferred);
        rbd_img_request_put(img_request);
 
        rbd_assert(orig_request->img_request);
        rbd_dev = orig_request->img_request->rbd_dev;
        rbd_assert(rbd_dev);
-       rbd_assert(obj_size == (u64)1 << rbd_dev->header.obj_order);
 
        if (result)
                goto out_err;
@@ -2195,11 +2201,12 @@ rbd_img_obj_parent_read_full_callback(struct rbd_img_request *img_request)
                goto out_err;
        orig_request->osd_req = osd_req;
        orig_request->copyup_pages = pages;
+       orig_request->copyup_page_count = page_count;
 
        /* Initialize the copyup op */
 
        osd_req_op_cls_init(osd_req, 0, CEPH_OSD_OP_CALL, "rbd", "copyup");
-       osd_req_op_cls_request_data_pages(osd_req, 0, pages, obj_size, 0,
+       osd_req_op_cls_request_data_pages(osd_req, 0, pages, parent_length, 0,
                                                false, false);
 
        /* Then the original write request op */
@@ -2312,6 +2319,7 @@ static int rbd_img_obj_parent_read_full(struct rbd_obj_request *obj_request)
        if (result)
                goto out_err;
        parent_request->copyup_pages = pages;
+       parent_request->copyup_page_count = page_count;
 
        parent_request->callback = rbd_img_obj_parent_read_full_callback;
        result = rbd_img_request_submit(parent_request);
@@ -2319,6 +2327,7 @@ static int rbd_img_obj_parent_read_full(struct rbd_obj_request *obj_request)
                return 0;
 
        parent_request->copyup_pages = NULL;
+       parent_request->copyup_page_count = 0;
        parent_request->obj_request = NULL;
        rbd_obj_request_put(obj_request);
 out_err: