ceph: queue_cap_snap should always queue dirty context
authorSage Weil <sage@newdream.net>
Tue, 13 Apr 2010 18:41:22 +0000 (11:41 -0700)
committerSage Weil <sage@newdream.net>
Tue, 13 Apr 2010 19:28:31 +0000 (12:28 -0700)
This simplifies the calling convention, and fixes a bug where we queue a
capsnap with a context other than i_head_snapc (the one that matches the
dirty pages).  The result was a BUG at fs/ceph/caps.c:2178 on writeback
completion when a capsnap matching the writeback snapc could not be found.

Signed-off-by: Sage Weil <sage@newdream.net>
fs/ceph/snap.c
fs/ceph/super.h

index 7e3e5f9..d197431 100644 (file)
@@ -430,8 +430,7 @@ static int dup_array(u64 **dst, __le64 *src, int num)
  * Caller must hold snap_rwsem for read (i.e., the realm topology won't
  * change).
  */
-void ceph_queue_cap_snap(struct ceph_inode_info *ci,
-                        struct ceph_snap_context *snapc)
+void ceph_queue_cap_snap(struct ceph_inode_info *ci)
 {
        struct inode *inode = &ci->vfs_inode;
        struct ceph_cap_snap *capsnap;
@@ -450,10 +449,11 @@ void ceph_queue_cap_snap(struct ceph_inode_info *ci,
                   as no new writes are allowed to start when pending, so any
                   writes in progress now were started before the previous
                   cap_snap.  lucky us. */
-               dout("queue_cap_snap %p snapc %p seq %llu used %d"
-                    " already pending\n", inode, snapc, snapc->seq, used);
+               dout("queue_cap_snap %p already pending\n", inode);
                kfree(capsnap);
        } else if (ci->i_wrbuffer_ref_head || (used & CEPH_CAP_FILE_WR)) {
+               struct ceph_snap_context *snapc = ci->i_head_snapc;
+
                igrab(inode);
 
                atomic_set(&capsnap->nref, 1);
@@ -462,7 +462,6 @@ void ceph_queue_cap_snap(struct ceph_inode_info *ci,
                INIT_LIST_HEAD(&capsnap->flushing_item);
 
                capsnap->follows = snapc->seq - 1;
-               capsnap->context = ceph_get_snap_context(snapc);
                capsnap->issued = __ceph_caps_issued(ci, NULL);
                capsnap->dirty = __ceph_caps_dirty(ci);
 
@@ -479,7 +478,7 @@ void ceph_queue_cap_snap(struct ceph_inode_info *ci,
                   snapshot. */
                capsnap->dirty_pages = ci->i_wrbuffer_ref_head;
                ci->i_wrbuffer_ref_head = 0;
-               ceph_put_snap_context(ci->i_head_snapc);
+               capsnap->context = snapc;
                ci->i_head_snapc = NULL;
                list_add_tail(&capsnap->ci_item, &ci->i_cap_snaps);
 
@@ -603,7 +602,7 @@ more:
                                if (lastinode)
                                        iput(lastinode);
                                lastinode = inode;
-                               ceph_queue_cap_snap(ci, realm->cached_context);
+                               ceph_queue_cap_snap(ci);
                                spin_lock(&realm->inodes_with_caps_lock);
                        }
                        spin_unlock(&realm->inodes_with_caps_lock);
@@ -825,8 +824,7 @@ void ceph_handle_snap(struct ceph_mds_client *mdsc,
                        spin_unlock(&realm->inodes_with_caps_lock);
                        spin_unlock(&inode->i_lock);
 
-                       ceph_queue_cap_snap(ci,
-                                           ci->i_snap_realm->cached_context);
+                       ceph_queue_cap_snap(ci);
 
                        iput(inode);
                        continue;
index 65d1203..4c07aca 100644 (file)
@@ -714,8 +714,7 @@ extern int ceph_update_snap_trace(struct ceph_mds_client *m,
 extern void ceph_handle_snap(struct ceph_mds_client *mdsc,
                             struct ceph_mds_session *session,
                             struct ceph_msg *msg);
-extern void ceph_queue_cap_snap(struct ceph_inode_info *ci,
-                               struct ceph_snap_context *snapc);
+extern void ceph_queue_cap_snap(struct ceph_inode_info *ci);
 extern int __ceph_finish_cap_snap(struct ceph_inode_info *ci,
                                  struct ceph_cap_snap *capsnap);
 extern void ceph_cleanup_empty_realms(struct ceph_mds_client *mdsc);