mds: don't skip inodes in journal that may be trimmed during replay

During replay we trim non-auth inodes on EExport or EImportFinish abort.
Subtree trimming may be delayed, too.

Skip parents if the diri is in the same blob, or if it is journaled in the
current segment *and* it is in a subtree that is unambiguously auth.  We can't
easily be more precise than that because the actual event we care about on
replay is EExport, but the migrator doesn't twiddle auth bits to false until
later.

Also, reset last_journaled on import.

This fixes replay bugs like

2011-04-13 18:15:18.064029 7f65588ef710 mds1.journal EImportStart.replay 10000000015 bounds []
2011-04-13 18:15:18.064034 7f65588ef710 mds1.journal EMetaBlob.replay 2 dirlumps by unknown0
2011-04-13 18:15:18.064040 7f65588ef710 mds1.journal EMetaBlob.replay dir 10000000010
2011-04-13 18:15:18.064046 7f65588ef710 mds1.journal EMetaBlob.replay missing dir ino  10000000010
mds/journal.cc: In function 'void EMetaBlob::replay(MDS*, LogSegment*)', in thread '0x7f65588ef710'
mds/journal.cc: 407: FAILED assert(0)
 ceph version 0.25-683-g653580a (commit:653580ae84c471c34872f14a0308c78af71f7243)
 1: (ceph::__ceph_assert_fail(char const*, char const*, int, char const*)+0x53) [0xa53d26]
 2: (EMetaBlob::replay(MDS*, LogSegment*)+0x7eb) [0x7a737d]

Fixes: #994
Signed-off-by: Sage Weil <sage@newdream.net>

diff --git a/src/mds/Migrator.cc b/src/mds/Migrator.cc
index c5798cfb41864349b984a54ff864c868574e3f3d..cf714d549a144ec404892f5047041715fee595fe 100644 (file)
--- a/src/mds/Migrator.cc
+++ b/src/mds/Migrator.cc
@@ -2142,7 +2142,7 @@ void Migrator::import_finish(CDir *dir)
 
 
 void Migrator::decode_import_inode(CDentry *dn, bufferlist::iterator& blp, int oldauth,
-                                  LogSegment *ls,
+                                  LogSegment *ls, uint64_t log_offset,
                                   map<CInode*, map<client_t,Capability::Export> >& cap_imports,
                                   list<ScatterLock*>& updated_scatterlocks)
 {  
@@ -2165,6 +2165,9 @@ void Migrator::decode_import_inode(CDentry *dn, bufferlist::iterator& blp, int o
   // state after link  -- or not!  -sage
   in->decode_import(blp, ls);  // cap imports are noted for later action
 
+  // note that we are journaled at this log offset
+  in->last_journaled = log_offset;
+
   // caps
   decode_import_inode_caps(in, blp, cap_imports);
 
@@ -2338,7 +2341,7 @@ int Migrator::decode_import_dir(bufferlist::iterator& blp,
     }
     else if (icode == 'I') {
       // inode
-      decode_import_inode(dn, blp, oldauth, ls, cap_imports, updated_scatterlocks);
+      decode_import_inode(dn, blp, oldauth, ls, le->get_start_off(), cap_imports, updated_scatterlocks);
     }
     
     // add dentry to journal entry

diff --git a/src/mds/Migrator.h b/src/mds/Migrator.h
index 086e09ea756702e197660bc81b1786a5094d6539..e7816339c6947dbb51d6638c78f52bcd4f24ce59 100644 (file)
--- a/src/mds/Migrator.h
+++ b/src/mds/Migrator.h
@@ -235,7 +235,7 @@ public:
 
 public:
   void decode_import_inode(CDentry *dn, bufferlist::iterator& blp, int oldauth, 
-                          LogSegment *ls,
+                          LogSegment *ls, uint64_t log_offset,
                           map<CInode*, map<client_t,Capability::Export> >& cap_imports,
                           list<ScatterLock*>& updated_scatterlocks);
   void decode_import_inode_caps(CInode *in,

diff --git a/src/mds/Server.cc b/src/mds/Server.cc
index f3b339ed41cc16f84b47763f5b2796d88e4e89e0..4c5840c27cc919464c46926a5adf1da94b377971 100644 (file)
--- a/src/mds/Server.cc
+++ b/src/mds/Server.cc
@@ -5016,7 +5016,7 @@ version_t Server::_rename_prepare_import(MDRequest *mdr, CDentry *srcdn, bufferl
   list<ScatterLock*> updated_scatterlocks;  // we clear_updated explicitly below
   mdcache->migrator->decode_import_inode(srcdn, blp, 
                                         srcdn->authority().first,
-                                        mdr->ls,
+                                        mdr->ls, 0,
                                         mdr->more()->cap_imports, updated_scatterlocks);
   srcdnl->get_inode()->filelock.clear_dirty();  
   srcdnl->get_inode()->nestlock.clear_dirty();  

diff --git a/src/mds/events/EMetaBlob.h b/src/mds/events/EMetaBlob.h
index 35e762eb14fa0dc48759b60a4e2297049b7c97b6..deb4ed3b9ed2da362e97bf9b1af88cf4fa933f90 100644 (file)
--- a/src/mds/events/EMetaBlob.h
+++ b/src/mds/events/EMetaBlob.h
@@ -633,47 +633,7 @@ private:
   static const int TO_AUTH_SUBTREE_ROOT = 0;  // default.
   static const int TO_ROOT = 1;
   
-  void add_dir_context(CDir *dir, int mode = TO_AUTH_SUBTREE_ROOT) {
-    // already have this dir?  (we must always add in order)
-    if (lump_map.count(dir->dirfrag())) 
-      return;
-
-    if (mode == TO_AUTH_SUBTREE_ROOT) {
-      //return;  // hack: for comparison purposes.. what if NO context?
-
-      // subtree root?
-      if (dir->is_subtree_root() && dir->is_auth())
-       return;
-
-      // was the inode journaled since the last subtree_map?
-      if (//false &&  // for benchmarking
-         last_subtree_map &&
-         dir->inode->last_journaled >= last_subtree_map) {
-       /*
-       cout << " inode " << dir->inode->inode.ino 
-            << " last journaled at " << dir->inode->last_journaled
-            << " and last_subtree_map is " << last_subtree_map 
-            << std::endl;
-       */
-       return;
-      }
-    }
-    
-    // stop at root/stray
-    CInode *diri = dir->get_inode();
-    if (!diri->get_projected_parent_dn())
-      return;
-
-    // journaled?
-
-    // add parent dn
-    CDentry *parent = diri->get_projected_parent_dn();
-    add_dir_context(parent->get_dir(), mode);
-    add_dentry(parent, false);
-  }
-
-
-
+  void add_dir_context(CDir *dir, int mode = TO_AUTH_SUBTREE_ROOT);
  
   void print(ostream& out) const {
     out << "[metablob";

diff --git a/src/mds/journal.cc b/src/mds/journal.cc
index 6a739e01bd11d81e6e0b007677cb091be979dad2..7e77443f9b5a6699bc4bec8180498257ddb54011 100644 (file)
--- a/src/mds/journal.cc
+++ b/src/mds/journal.cc
@@ -317,6 +317,83 @@ EMetaBlob::EMetaBlob(MDLog *mdlog) : root(NULL),
                                     my_offset(mdlog ? mdlog->get_write_pos() : 0) //, _segment(0)
 { }
 
+void EMetaBlob::add_dir_context(CDir *dir, int mode)
+{
+  MDS *mds = dir->cache->mds;
+
+  list<CDentry*> parents;
+
+  // it may be okay not to include the maybe items, if
+  //  - we journaled the maybe child inode in this segment
+  //  - that subtree turns out to be unambiguously auth
+  list<CDentry*> maybe;
+  bool maybenot = false;
+
+  while (true) {
+    // already have this dir?  (we must always add in order)
+    if (lump_map.count(dir->dirfrag())) {
+      dout(20) << "EMetaBlob::add_dir_context(" << dir << ") have lump " << dir->dirfrag() << dendl;
+      break;
+    }
+      
+    // stop at root/stray
+    CInode *diri = dir->get_inode();
+    CDentry *parent = diri->get_projected_parent_dn();
+
+    if (!parent)
+      break;
+
+    if (mode == TO_AUTH_SUBTREE_ROOT) {
+      // subtree root?
+      if (dir->is_subtree_root()) {
+       if (dir->is_auth() && !dir->is_ambiguous_auth()) {
+         // it's an auth subtree, we don't need maybe (if any), and we're done.
+         dout(20) << "EMetaBlob::add_dir_context(" << dir << ") reached unambig auth subtree, don't need " << maybe
+                  << " at " << *dir << dendl;
+         maybe.clear();
+         break;
+       } else {
+         dout(20) << "EMetaBlob::add_dir_context(" << dir << ") reached ambig or !auth subtree, need " << maybe
+                  << " at " << *dir << dendl;
+         // we need the maybe list after all!
+         parents.splice(parents.begin(), maybe);
+         maybenot = false;
+       }
+      }
+      
+      // was the inode journaled in this blob?
+      if (my_offset && diri->last_journaled == my_offset) {
+       dout(20) << "EMetaBlob::add_dir_context(" << dir << ") already have diri this blob " << *diri << dendl;
+       break;
+      }
+
+      // have we journaled this inode since the last subtree map?
+      if (!maybenot && last_subtree_map && diri->last_journaled >= last_subtree_map) {
+       dout(20) << "EMetaBlob::add_dir_context(" << dir << ") already have diri in this segment (" 
+                << diri->last_journaled << " >= " << last_subtree_map << "), setting maybenot flag "
+                << *diri << dendl;
+       maybenot = true;
+      }
+    }
+
+    if (maybenot) {
+      dout(25) << "EMetaBlob::add_dir_context(" << dir << ")      maybe " << *parent << dendl;
+      maybe.push_front(parent);
+    } else {
+      dout(25) << "EMetaBlob::add_dir_context(" << dir << ") definitely " << *parent << dendl;
+      parents.push_front(parent);
+    }
+    
+    dir = parent->get_dir();
+  }
+  
+  parents.splice(parents.begin(), maybe);
+
+  dout(20) << "EMetaBlob::add_dir_context final: " << parents << dendl;
+  for (list<CDentry*>::iterator p = parents.begin(); p != parents.end(); p++)
+    add_dentry(*p, false);
+}
+
 void EMetaBlob::update_segment(LogSegment *ls)
 {
   // atids?