mds/rename: Handle referent inode rollback

If the destination dentry exists and is referent
remote, then the rollback should take care of
restoring the referent inode and the add back
the referent inode number to the referent_inode
list of the primary/real inode. This patch takes
care of the same.

Fixes: https://tracker.ceph.com/issues/54205
Signed-off-by: Kotresh HR <khiremat@redhat.com>

diff --git a/src/mds/Server.cc b/src/mds/Server.cc
index d12b696dad11da8d52bc87681fad9d358fa9cd34..4f2fc1c029684b02a648d0788a2ebf94026cf2fa 100644 (file)
--- a/src/mds/Server.cc
+++ b/src/mds/Server.cc
@@ -10913,23 +10913,29 @@ void Server::handle_peer_rename_prep(const MDRequestRef& mdr)
   rollback.orig_src.dirfrag_old_mtime = srcdn->get_dir()->get_projected_fnode()->fragstat.mtime;
   rollback.orig_src.dirfrag_old_rctime = srcdn->get_dir()->get_projected_fnode()->rstat.rctime;
   rollback.orig_src.dname = srcdn->get_name();
+  rollback.orig_src.referent_ino = 0;
   if (srcdnl->is_primary())
     rollback.orig_src.ino = srcdnl->get_inode()->ino();
   else {
     ceph_assert(srcdnl->is_remote() || srcdnl->is_referent_remote());
     rollback.orig_src.remote_ino = srcdnl->get_remote_ino();
     rollback.orig_src.remote_d_type = srcdnl->get_remote_d_type();
+    if (srcdnl->is_referent_remote())
+      rollback.orig_src.referent_ino = srcdnl->get_referent_ino();
   }
   
   rollback.orig_dest.dirfrag = destdn->get_dir()->dirfrag();
   rollback.orig_dest.dirfrag_old_mtime = destdn->get_dir()->get_projected_fnode()->fragstat.mtime;
   rollback.orig_dest.dirfrag_old_rctime = destdn->get_dir()->get_projected_fnode()->rstat.rctime;
   rollback.orig_dest.dname = destdn->get_name();
+  rollback.orig_dest.referent_ino = 0;
   if (destdnl->is_primary())
     rollback.orig_dest.ino = destdnl->get_inode()->ino();
   else if (destdnl->is_remote() || destdnl->is_referent_remote()) {
     rollback.orig_dest.remote_ino = destdnl->get_remote_ino();
     rollback.orig_dest.remote_d_type = destdnl->get_remote_d_type();
+    if (destdnl->is_referent_remote())
+      rollback.orig_dest.referent_ino = destdnl->get_referent_ino();
   }
   
   if (straydn) {
@@ -11296,12 +11302,16 @@ void Server::do_rename_rollback(bufferlist &rbl, mds_rank_t leader, const MDRequ
     dout(10) << " destdir not found" << dendl;
 
   CInode *in = NULL;
+  CInode *referent_in = nullptr;
   if (rollback.orig_src.ino) {
     in = mdcache->get_inode(rollback.orig_src.ino);
     if (in && in->is_dir())
       ceph_assert(srcdn && destdn);
-  } else
+  } else {
     in = mdcache->get_inode(rollback.orig_src.remote_ino);
+    if (rollback.orig_src.referent_ino)
+      referent_in = mdcache->get_inode(rollback.orig_src.referent_ino);
+  }
 
   CDir *straydir = NULL;
   CDentry *straydn = NULL;
@@ -11320,12 +11330,16 @@ void Server::do_rename_rollback(bufferlist &rbl, mds_rank_t leader, const MDRequ
   }
 
   CInode *target = NULL;
+  CInode *target_referent_in = nullptr;
   if (rollback.orig_dest.ino) {
     target = mdcache->get_inode(rollback.orig_dest.ino);
     if (target)
       ceph_assert(destdn && straydn);
-  } else if (rollback.orig_dest.remote_ino)
+  } else if (rollback.orig_dest.remote_ino) {
     target = mdcache->get_inode(rollback.orig_dest.remote_ino);
+    if (rollback.orig_dest.referent_ino)
+      target_referent_in = mdcache->get_inode(rollback.orig_dest.referent_ino);
+  }
 
   // can't use is_auth() in the resolve stage
   mds_rank_t whoami = mds->get_nodeid();
@@ -11348,9 +11362,15 @@ void Server::do_rename_rollback(bufferlist &rbl, mds_rank_t leader, const MDRequ
     if (rollback.orig_src.ino) {
       ceph_assert(in);
       srcdn->push_projected_linkage(in);
-    } else
-      srcdn->push_projected_linkage(rollback.orig_src.remote_ino,
-                                   rollback.orig_src.remote_d_type);
+    } else {
+      if (rollback.orig_src.referent_ino) {
+       ceph_assert(referent_in);
+        srcdn->push_projected_linkage(referent_in, rollback.orig_src.remote_ino, rollback.orig_src.referent_ino);
+      } else {
+        srcdn->push_projected_linkage(rollback.orig_src.remote_ino,
+                                      rollback.orig_src.remote_d_type);
+      }
+    }
   }
 
   map<client_t,ref_t<MClientSnap>> splits[2];
@@ -11413,8 +11433,13 @@ void Server::do_rename_rollback(bufferlist &rbl, mds_rank_t leader, const MDRequ
     if (rollback.orig_dest.ino && target) {
       destdn->push_projected_linkage(target);
     } else if (rollback.orig_dest.remote_ino) {
-      destdn->push_projected_linkage(rollback.orig_dest.remote_ino,
-                                    rollback.orig_dest.remote_d_type);
+      if (rollback.orig_dest.referent_ino) {
+       ceph_assert(target_referent_in);
+        destdn->push_projected_linkage(target_referent_in, rollback.orig_dest.remote_ino, rollback.orig_dest.referent_ino);
+      } else {
+        destdn->push_projected_linkage(rollback.orig_dest.remote_ino,
+                                       rollback.orig_dest.remote_d_type);
+      }
     } else {
       // the dentry will be trimmed soon, it's ok to have wrong linkage
       if (rollback.orig_dest.ino)
@@ -11452,8 +11477,15 @@ void Server::do_rename_rollback(bufferlist &rbl, mds_rank_t leader, const MDRequ
       else
        ceph_assert(rollback.orig_dest.remote_ino &&
               rollback.orig_dest.remote_ino == rollback.orig_src.ino);
-    } else
+    } else {
       ti->nlink++;
+      //add referent inode back to the list
+      if (rollback.orig_dest.referent_ino) {
+        ti->add_referent_ino(rollback.orig_dest.referent_ino);
+        dout(10) << __func__ << " referent_inodes " << std::hex << ti->get_referent_inodes()
+                 << " referent ino added " << rollback.orig_dest.referent_ino << dendl;
+      }
+    }
 
     if (!projected)
       target->reset_inode(ti);

diff --git a/src/mds/events/EPeerUpdate.h b/src/mds/events/EPeerUpdate.h
index 087caa77996abee4022116c1770b54672bce4dee..772bd81dc3007b68645832a27d711cdc90827e31 100644 (file)
--- a/src/mds/events/EPeerUpdate.h
+++ b/src/mds/events/EPeerUpdate.h
@@ -74,6 +74,7 @@ struct rename_rollback {
     std::string dname;
     char remote_d_type;
     utime_t old_ctime;
+    inodeno_t referent_ino;
 
     drec() : remote_d_type((char)S_IFREG) {}
 

diff --git a/src/mds/journal.cc b/src/mds/journal.cc
index b1888bda3a0c3e1876778d563d39917e06e7251e..e786a02a3ae77b9160f68b11fe647e72c81902a6 100644 (file)
--- a/src/mds/journal.cc
+++ b/src/mds/journal.cc
@@ -2655,7 +2655,7 @@ void rmdir_rollback::generate_test_instances(std::list<rmdir_rollback*>& ls)
 
 void rename_rollback::drec::encode(bufferlist &bl) const
 {
-  ENCODE_START(2, 2, bl);
+  ENCODE_START(3, 2, bl);
   encode(dirfrag, bl);
   encode(dirfrag_old_mtime, bl);
   encode(dirfrag_old_rctime, bl);
@@ -2664,12 +2664,13 @@ void rename_rollback::drec::encode(bufferlist &bl) const
   encode(dname, bl);
   encode(remote_d_type, bl);
   encode(old_ctime, bl);
+  encode(referent_ino, bl);
   ENCODE_FINISH(bl);
 }
 
 void rename_rollback::drec::decode(bufferlist::const_iterator &bl)
 {
-  DECODE_START_LEGACY_COMPAT_LEN(2, 2, 2, bl);
+  DECODE_START_LEGACY_COMPAT_LEN(3, 2, 2, bl);
   decode(dirfrag, bl);
   decode(dirfrag_old_mtime, bl);
   decode(dirfrag_old_rctime, bl);
@@ -2678,6 +2679,8 @@ void rename_rollback::drec::decode(bufferlist::const_iterator &bl)
   decode(dname, bl);
   decode(remote_d_type, bl);
   decode(old_ctime, bl);
+  if (struct_v >= 3)
+    decode(referent_ino, bl);
   DECODE_FINISH(bl);
 }
 
@@ -2689,6 +2692,7 @@ void rename_rollback::drec::dump(Formatter *f) const
   f->dump_int("ino", ino);
   f->dump_int("remote ino", remote_ino);
   f->dump_string("dname", dname);
+  f->dump_int("referent_ino", referent_ino);
   uint32_t type = DTTOIF(remote_d_type) & S_IFMT; // convert to type entries
   string type_string;
   switch(type) {