mds: don't reset readdir offset if client supports hash order dentry

Now the ordering of dentries is stable across directory fragmentation.
There is no need to reset readdir offset if directory get fragmented
in the middle of readdir.

Signed-off-by: Yan, Zheng <zyan@redhat.com>
(cherry picked from commit 98a01af2cd8e01f14ab7be32c31a43571ef8bb87)

Signed-off-by: Greg Farnum <gfarnum@redhat.com

diff --git a/src/client/Client.cc b/src/client/Client.cc
index ad8aca627643653c3859947119db8fba3142bf68..9755446c400fdd33bf7b2f4fe998b51a376fb237 100644 (file)
--- a/src/client/Client.cc
+++ b/src/client/Client.cc
@@ -1125,10 +1125,11 @@ void Client::insert_readdir_results(MetaRequest *request, MetaSession *session,
     if (fg != dst.frag) {
       ldout(cct, 10) << "insert_trace got new frag " << fg << " -> " << dst.frag << dendl;
       fg = dst.frag;
-      readdir_offset = 2;
-      readdir_start.clear();
-      if (!hash_order)
+      if (!hash_order) {
+       readdir_offset = 2;
+       readdir_start.clear();
        dirp->offset = dir_result_t::make_fpos(fg, readdir_offset, false);
+      }
     }
 
     ldout(cct, 10) << __func__ << " " << numdn << " readdir items, end=" << end
@@ -1184,13 +1185,12 @@ void Client::insert_readdir_results(MetaRequest *request, MetaSession *session,
       ldout(cct, 15) << __func__ << "  " << hex << dn->offset << dec << ": '" << dname << "' -> " << in->ino << dendl;
     }
 
-    if (end) {
-      dirp->last_name.clear();
-      dirp->next_offset = 2;
-    } else {
+    if (numdn > 0)
       dirp->last_name = dname;
+    if (end)
+      dirp->next_offset = 2;
+    else
       dirp->next_offset = readdir_offset;
-    }
 
     if (dir->is_empty())
       close_dir(dir);
@@ -6918,10 +6918,12 @@ void Client::_readdir_next_frag(dir_result_t *dirp)
   ldout(cct, 10) << "_readdir_next_frag advance from " << dirp->buffer_frag << " to " << fg << dendl;
 
   if (dirp->hash_order()) {
+    // keep last_name
     int64_t new_offset = dir_result_t::make_fpos(fg.value(), 2, true);
     if (dirp->offset < new_offset) // don't decrease offset
       dirp->offset = new_offset;
   } else {
+    dirp->last_name.clear();
     dirp->offset = dir_result_t::make_fpos(fg, 2, false);
     _readdir_rechoose_frag(dirp);
   }
@@ -7208,7 +7210,7 @@ int Client::readdir_r_cb(dir_result_t *d, add_dirent_cb_t cb, void *p)
        return r;
     }
 
-    if (!dirp->last_name.empty()) {
+    if (dirp->next_offset > 2) {
       ldout(cct, 10) << " fetching next chunk of this frag" << dendl;
       _readdir_drop_dirp_buffer(dirp);
       continue;  // more!

diff --git a/src/mds/Server.cc b/src/mds/Server.cc
index 514f4d958e9ec816b5d5643ac22605dc47259d80..313744718b0b18015868fa9c283ea476c5a256c8 100644 (file)
--- a/src/mds/Server.cc
+++ b/src/mds/Server.cc
@@ -3276,7 +3276,8 @@ void Server::handle_client_readdir(MDRequestRef& mdr)
   frag_t fg = (__u32)req->head.args.readdir.frag;
   unsigned req_flags = (__u32)req->head.args.readdir.flags;
   string offset_str = req->get_path2();
-  dout(10) << " frag " << fg << " offset '" << offset_str << "'" << dendl;
+  dout(10) << " frag " << fg << " offset '" << offset_str << "'"
+          << " flags " << req_flags << dendl;
 
   __u32 offset_hash = 0;
   if (!offset_str.empty())
@@ -3284,10 +3285,20 @@ void Server::handle_client_readdir(MDRequestRef& mdr)
 
   // does the frag exist?
   if (diri->dirfragtree[fg.value()] != fg) {
-    frag_t newfg = diri->dirfragtree[fg.value()];
+    frag_t newfg;
+    if (req_flags & CEPH_READDIR_REPLY_BITFLAGS) {
+      if (fg.contains((unsigned)offset_hash)) {
+       newfg = diri->dirfragtree[offset_hash];
+      } else {
+       // client actually wants next frag
+       newfg = diri->dirfragtree[fg.value()];
+      }
+    } else {
+      offset_str.clear();
+      newfg = diri->dirfragtree[fg.value()];
+    }
     dout(10) << " adjust frag " << fg << " -> " << newfg << " " << diri->dirfragtree << dendl;
     fg = newfg;
-    offset_str.clear();
   }
   
   CDir *dir = try_open_auth_dirfrag(diri, fg, mdr);