libceph: fix page calculation for non-page-aligned io
authorSage Weil <sage@newdream.net>
Mon, 13 Jun 2011 23:20:18 +0000 (16:20 -0700)
committerSage Weil <sage@newdream.net>
Mon, 13 Jun 2011 23:26:17 +0000 (16:26 -0700)
Set the page count correctly for non-page-aligned IO.  We were already
doing this correctly for alignment, but not the page count.  Fixes
DIRECT_IO writes from unaligned pages.

Signed-off-by: Sage Weil <sage@newdream.net>
net/ceph/osd_client.c

index 9cb627a..7330c27 100644 (file)
@@ -477,8 +477,9 @@ struct ceph_osd_request *ceph_osdc_new_request(struct ceph_osd_client *osdc,
        calc_layout(osdc, vino, layout, off, plen, req, ops);
        req->r_file_layout = *layout;  /* keep a copy */
 
-       /* in case it differs from natural alignment that calc_layout
-          filled in for us */
+       /* in case it differs from natural (file) alignment that
+          calc_layout filled in for us */
+       req->r_num_pages = calc_pages_for(page_align, *plen);
        req->r_page_alignment = page_align;
 
        ceph_osdc_build_request(req, off, plen, ops,
@@ -2027,8 +2028,9 @@ static struct ceph_msg *get_reply(struct ceph_connection *con,
                int want = calc_pages_for(req->r_page_alignment, data_len);
 
                if (unlikely(req->r_num_pages < want)) {
-                       pr_warning("tid %lld reply %d > expected %d pages\n",
-                                  tid, want, m->nr_pages);
+                       pr_warning("tid %lld reply has %d bytes %d pages, we"
+                                  " had only %d pages ready\n", tid, data_len,
+                                  want, req->r_num_pages);
                        *skip = 1;
                        ceph_msg_put(m);
                        m = NULL;