mds: versioned dentry, dir items, inode_map

diff --git a/src/TODO b/src/TODO
index 1fe4ef8a60310f60c4285ba4d9166dca3fec3210..b15b10d00def476d26ff2fad5a5635a261dad07f 100644 (file)
--- a/src/TODO
+++ b/src/TODO
@@ -233,7 +233,19 @@ todo
 - mds metadata versioning
   - (dir) inode versions..
 
+  primary file link -> old inode
+  primary dir link -> multiversion inode
+  remote link -> multiversion inode
+
+- for simplicity, don't replicate any snapshot data.
+- or issue any leases or capabilities... the content is static!
+- 
+
+- need rrealms in fraginfo_t
+- rename() needs to create a new realm if src/dst realms differ and (rrealms, or open_children, or not subtree leaf) (similar logic to the anchor update)
+
 - will snapshots and CAS play nice?
+  - cas object refs should follow same deletion semantics as non-cas objects.
 
 - mds server ops
 

diff --git a/src/csyn.cc b/src/csyn.cc
index 0980acf51c0e2058dff5e8f06761bfc5dd32871b..12cbbd7b4ace82540f2ba9984fcef76a4432087e 100644 (file)
--- a/src/csyn.cc
+++ b/src/csyn.cc
@@ -36,8 +36,8 @@ using namespace std;
 #include <sys/stat.h>
 #include <fcntl.h>
 
-int main(int argc, const char **argv, char *envp[]) {
-
+int main(int argc, const char **argv, char *envp[]) 
+{
   //cerr << "csyn starting" << std::endl;
   vector<const char*> args;
   argv_to_vec(argc, argv, args);

diff --git a/src/mds/CDentry.cc b/src/mds/CDentry.cc
index 8fc44b462b66c43bbbee9b5673d270f67dfca881..527cb270908707b1b73e0a4916e416517bea098f 100644 (file)
--- a/src/mds/CDentry.cc
+++ b/src/mds/CDentry.cc
@@ -46,6 +46,15 @@ ostream& operator<<(ostream& out, CDentry& dn)
   
   out << "[dentry " << path;
   
+  if (true || dn.first != 0 || dn.last != CEPH_NOSNAP) {
+    out << " [" << dn.first << ",";
+    if (dn.last == CEPH_NOSNAP) 
+      out << "head";
+    else
+      out << dn.last;
+    out << ']';
+  }
+
   if (dn.is_auth()) {
     out << " auth";
     if (dn.is_replicated()) 

diff --git a/src/mds/CDentry.h b/src/mds/CDentry.h
index fab43d0509c60ef24652a14d6851ff73c6641400..a6a1ec9cc2a9200908a6a46b77ac4c01aaddb266 100644 (file)
--- a/src/mds/CDentry.h
+++ b/src/mds/CDentry.h
@@ -76,9 +76,15 @@ class CDentry : public MDSCacheObject, public LRUObject {
     return *this < *(CDentry*)r;
   }
 
- protected:
+public:
   nstring name;
+  snapid_t first, last;
+
+  dentry_key_t key() { 
+    return dentry_key_t(last, name.c_str()); 
+  }
 
+protected:
   inodeno_t remote_ino;      // if remote dentry
   unsigned char remote_d_type;
 
@@ -113,19 +119,11 @@ public:
   SimpleLock lock;
 
 
-
  public:
   // cons
-  CDentry() :
-    remote_ino(0), remote_d_type(0),
-    inode(0), dir(0),
-    version(0), projected_version(0),
-    xlist_dirty(this),
-    dir_offset(0),
-    auth_pins(0), nested_auth_pins(0), nested_anchors(0),
-    lock(this, CEPH_LOCK_DN, WAIT_LOCK_OFFSET) { }
   CDentry(const nstring& n, CInode *in) :
     name(n),
+    first(0), last(CEPH_NOSNAP),
     remote_ino(0), remote_d_type(0),
     inode(in), dir(0),
     version(0), projected_version(0),
@@ -135,6 +133,7 @@ public:
     lock(this, CEPH_LOCK_DN, WAIT_LOCK_OFFSET) { }
   CDentry(const nstring& n, inodeno_t ino, unsigned char dt, CInode *in=0) :
     name(n),
+    first(0), last(CEPH_NOSNAP),
     remote_ino(ino), remote_d_type(dt),
     inode(in), dir(0),
     version(0), projected_version(0),

diff --git a/src/mds/CDir.cc b/src/mds/CDir.cc
index e59f514acf289cc794642578fa6ea5e941690eb0..2fa9f78c3717642c92c551f69e197ae68a7a2d67 100644 (file)
--- a/src/mds/CDir.cc
+++ b/src/mds/CDir.cc
@@ -204,10 +204,10 @@ CDentry* CDir::add_null_dentry(const nstring& dname)
   dn->version = get_projected_version();
   
   // add to dir
-  assert(items.count(dn->name.c_str()) == 0);
+  assert(items.count(dn->key()) == 0);
   //assert(null_items.count(dn->name) == 0);
 
-  items[dn->name.c_str()] = dn;
+  items[dn->key()] = dn;
   nnull++;
 
   dout(12) << "add_null_dentry " << *dn << dendl;
@@ -236,10 +236,10 @@ CDentry* CDir::add_primary_dentry(const nstring& dname, CInode *in)
   dn->version = get_projected_version();
   
   // add to dir
-  assert(items.count(dn->name.c_str()) == 0);
+  assert(items.count(dn->key()) == 0);
   //assert(null_items.count(dn->name) == 0);
 
-  items[dn->name.c_str()] = dn;
+  items[dn->key()] = dn;
   link_inode_work( dn, in );
 
   dout(12) << "add_primary_dentry " << *dn << dendl;
@@ -267,10 +267,10 @@ CDentry* CDir::add_remote_dentry(const nstring& dname, inodeno_t ino, unsigned c
   dn->version = get_projected_version();
   
   // add to dir
-  assert(items.count(dn->name.c_str()) == 0);
+  assert(items.count(dn->key()) == 0);
   //assert(null_items.count(dn->name) == 0);
 
-  items[dn->name.c_str()] = dn;
+  items[dn->key()] = dn;
   nitems++;
 
   dout(12) << "add_remote_dentry " << *dn << dendl;
@@ -300,8 +300,8 @@ void CDir::remove_dentry(CDentry *dn)
   }
   
   // remove from list
-  assert(items.count(dn->name.c_str()) == 1);
-  items.erase(dn->name.c_str());
+  assert(items.count(dn->key()) == 1);
+  items.erase(dn->key());
 
   // adjust dirty counter?
   if (dn->state_test(CDentry::STATE_DIRTY))
@@ -501,9 +501,9 @@ void CDir::steal_dentry(CDentry *dn)
 {
   dout(15) << "steal_dentry " << *dn << dendl;
 
-  items[dn->name.c_str()] = dn;
+  items[dn->key()] = dn;
 
-  dn->dir->items.erase(dn->name.c_str());
+  dn->dir->items.erase(dn->key());
   if (dn->dir->items.empty())
     dn->dir->put(PIN_CHILD);
 
@@ -627,7 +627,7 @@ void CDir::split(int bits, list<CDir*>& subs, list<Context*>& waiters, bool repl
     CDir::map_t::iterator p = items.begin();
     
     CDentry *dn = p->second;
-    frag_t subfrag = inode->pick_dirfrag(p->first);
+    frag_t subfrag = inode->pick_dirfrag(dn->name);
     int n = (subfrag.value() & (subfrag.mask() ^ frag.mask())) >> subfrag.mask_shift();
     dout(15) << " subfrag " << subfrag << " n=" << n << " for " << p->first << dendl;
     CDir *f = subfrags[n];

diff --git a/src/mds/CDir.h b/src/mds/CDir.h
index f8961da098a7ab48e7d07a2432db8fd7542f347d..ea59183eb5f8a3a738a0f264d23683479ac5ad3a 100644 (file)
--- a/src/mds/CDir.h
+++ b/src/mds/CDir.h
@@ -45,6 +45,9 @@ class Context;
 class CDirDiscover;
 
 
+
+
+
 ostream& operator<<(ostream& out, class CDir& dir);
 
 
@@ -193,7 +196,8 @@ public:
 
 public:
   //typedef hash_map<string, CDentry*> map_t;   // there is a bug somewhere, valgrind me.
-  typedef map<const char *, CDentry*, ltstr> map_t;
+  //typedef map<const char *, CDentry*, ltstr> map_t;
+  typedef map<dentry_key_t, CDentry*, ltdentrykey> map_t;
 protected:
 
   // contents
@@ -282,12 +286,14 @@ protected:
 
   // -- dentries and inodes --
  public:
-  CDentry* lookup(const string& s) {
-    nstring ns(s);
-    return lookup(ns);
+  CDentry* lookup(const string& n, snapid_t snap=CEPH_NOSNAP) {
+    return lookup(n.c_str(), snap);
+  }
+  CDentry* lookup(const nstring& ns, snapid_t snap=CEPH_NOSNAP) {
+    return lookup(ns.c_str(), snap);
   }
-  CDentry* lookup(const nstring& ns) {
-    map_t::iterator iter = items.find(ns.c_str());
+  CDentry* lookup(const char *n, snapid_t snap=CEPH_NOSNAP) {
+    map_t::iterator iter = items.find(dentry_key_t(snap, n));
     if (iter == items.end()) 
       return 0;
     else

diff --git a/src/mds/CInode.h b/src/mds/CInode.h
index 498ff9fc8ba517a40b456338c4717a87042ffc3d..62b6d1adf91ea7832dc0b9db15e59f6c5cd90257 100644 (file)
--- a/src/mds/CInode.h
+++ b/src/mds/CInode.h
@@ -137,6 +137,8 @@ class CInode : public MDSCacheObject {
   SnapRealm        *snaprealm;
 
   SnapRealm        *containing_realm;
+  snapid_t                   snapid;      // 0 = multiversion OR head
+  map<snapid_t, old_inode_t> old_inodes;  // key = last, value.first = first
 
   off_t last_journaled;       // log offset for the last time i was journaled
   off_t last_open_journaled;  // log offset for the last journaled EOpen
@@ -260,7 +262,8 @@ private:
   // ---------------------------
   CInode(MDCache *c, bool auth=true) : 
     mdcache(c),
-    snaprealm(0),
+    snaprealm(0), containing_realm(0),
+    snapid(0),    
     last_journaled(0), last_open_journaled(0), 
     //hack_accessed(true),
     stickydir_ref(0),
@@ -313,6 +316,7 @@ private:
 
 
   inodeno_t ino() const { return inode.ino; }
+  vinodeno_t vino() const { return vinodeno_t(inode.ino, snapid); }
   inode_t& get_inode() { return inode; }
   CDentry* get_parent_dn() { return parent; }
   CDentry* get_projected_parent_dn() { return projected_parent ? projected_parent:parent; }

diff --git a/src/mds/MDCache.cc b/src/mds/MDCache.cc
index b099c932489a9e642168252153f5cd973b635673..8d57fbba0277ed17ddd4d0d98fca32573daf1494 100644 (file)
--- a/src/mds/MDCache.cc
+++ b/src/mds/MDCache.cc
@@ -164,8 +164,8 @@ CInode *MDCache::create_inode()
 void MDCache::add_inode(CInode *in) 
 {
   // add to lru, inode map
-  assert(inode_map.count(in->ino()) == 0);  // should be no dup inos!
-  inode_map[ in->ino() ] = in;
+  assert(inode_map.count(in->vino()) == 0);  // should be no dup inos!
+  inode_map[ in->vino() ] = in;
 
   if (in->ino() < MDS_INO_BASE) {
     base_inodes.insert(in);
@@ -188,7 +188,7 @@ void MDCache::remove_inode(CInode *o)
   }
 
   // remove from inode map
-  inode_map.erase(o->ino());    
+  inode_map.erase(o->vino());    
 
   if (o->ino() < MDS_INO_BASE) {
     assert(base_inodes.count(o));
@@ -1939,7 +1939,7 @@ void MDCache::rejoin_walk(CDir *dir, MMDSCacheRejoin *rejoin)
       dout(15) << " add_weak_primary_dentry " << *dn << dendl;
       assert(dn->is_primary());
       assert(dn->inode->is_dir());
-      rejoin->add_weak_primary_dentry(dir->dirfrag(), p->first, dn->get_inode()->ino());
+      rejoin->add_weak_primary_dentry(dir->dirfrag(), dn->name.c_str(), dn->get_inode()->ino());
       dn->get_inode()->get_nested_dirfrags(nested);
     }
   } else {
@@ -1952,7 +1952,7 @@ void MDCache::rejoin_walk(CDir *dir, MMDSCacheRejoin *rejoin)
         ++p) {
       CDentry *dn = p->second;
       dout(15) << " add_strong_dentry " << *dn << dendl;
-      rejoin->add_strong_dentry(dir->dirfrag(), p->first, 
+      rejoin->add_strong_dentry(dir->dirfrag(), dn->name, 
                                dn->is_primary() ? dn->get_inode()->ino():inodeno_t(0),
                                dn->is_remote() ? dn->get_remote_ino():inodeno_t(0),
                                dn->is_remote() ? dn->get_remote_d_type():0, 
@@ -2269,7 +2269,7 @@ void MDCache::rejoin_scour_survivor_replicas(int from, MMDSCacheRejoin *ack)
 
   // FIXME: what about root and stray inodes.
   
-  for (hash_map<inodeno_t,CInode*>::iterator p = inode_map.begin();
+  for (hash_map<vinodeno_t,CInode*>::iterator p = inode_map.begin();
        p != inode_map.end();
        ++p) {
     CInode *in = p->second;
@@ -2938,7 +2938,7 @@ void MDCache::identify_files_to_recover()
   */
 
   dout(10) << "identify_files_to_recover" << dendl;
-  for (hash_map<inodeno_t,CInode*>::iterator p = inode_map.begin();
+  for (hash_map<vinodeno_t,CInode*>::iterator p = inode_map.begin();
        p != inode_map.end();
        ++p) {
     CInode *in = p->second;
@@ -3475,9 +3475,9 @@ void MDCache::trim_non_auth()
 
   if (lru.lru_get_size() == 0) {
     // root, stray, etc.?
-    hash_map<inodeno_t,CInode*>::iterator p = inode_map.begin();
+    hash_map<vinodeno_t,CInode*>::iterator p = inode_map.begin();
     while (p != inode_map.end()) {
-      hash_map<inodeno_t,CInode*>::iterator next = p;
+      hash_map<vinodeno_t,CInode*>::iterator next = p;
       ++next;
       CInode *in = p->second;
       if (!in->is_auth()) {
@@ -4314,7 +4314,7 @@ int MDCache::path_traverse(MDRequest *mdr, Message *req,     // who
     
     // MISS.  dentry doesn't exist.
     dout(12) << "traverse: miss on dentry " << path[depth] << " in " << *curdir << dendl;
-    
+
     if (curdir->is_auth()) {
       // dentry is mine.
       if (curdir->is_complete()) {
@@ -6771,7 +6771,7 @@ void MDCache::show_cache()
 {
   dout(7) << "show_cache" << dendl;
   
-  for (hash_map<inodeno_t,CInode*>::iterator it = inode_map.begin();
+  for (hash_map<vinodeno_t,CInode*>::iterator it = inode_map.begin();
        it != inode_map.end();
        it++) {
     // unlinked?
@@ -6810,7 +6810,7 @@ void MDCache::dump_cache()
   ofstream myfile;
   myfile.open(fn);
   
-  for (hash_map<inodeno_t,CInode*>::iterator it = inode_map.begin();
+  for (hash_map<vinodeno_t,CInode*>::iterator it = inode_map.begin();
        it != inode_map.end();
        it++) {
     list<CDir*> dfs;

diff --git a/src/mds/MDCache.h b/src/mds/MDCache.h
index f0e74c8d24febbb48b7c988689f164ba075f9f3b..22844a7aade9f85e3e1977532d343f5a0e0d60c6 100644 (file)
--- a/src/mds/MDCache.h
+++ b/src/mds/MDCache.h
@@ -343,9 +343,9 @@ class MDCache {
   // -- my cache --
   LRU lru;   // dentry lru for expiring items from cache
  protected:
-  hash_map<inodeno_t,CInode*>   inode_map;  // map of inodes by ino
-  CInode *root;                             // root inode
-  CInode *stray;                            // my stray dir
+  hash_map<vinodeno_t,CInode*> inode_map;  // map of inodes by ino
+  CInode *root;                            // root inode
+  CInode *stray;                           // my stray dir
 
   set<CInode*> base_inodes;  // inodes < MDS_INO_BASE (root, stray, etc.)
 
@@ -653,15 +653,17 @@ public:
   bool did_shutdown_log_cap;
 
   // inode_map
-  bool have_inode( inodeno_t ino ) { return inode_map.count(ino) ? true:false; }
-  CInode* get_inode( inodeno_t ino ) {
-    if (have_inode(ino))
-      return inode_map[ino];
+  bool have_inode( inodeno_t ino ) { return have_inode(vinodeno_t(ino, 0)); }
+  bool have_inode( vinodeno_t vino ) { return inode_map.count(vino) ? true:false; }
+  CInode* get_inode( inodeno_t ino, snapid_t s=0 ) {
+    vinodeno_t vino(ino,s);
+    if (have_inode(vino))
+      return inode_map[vino];
     return NULL;
   }
   CDir* get_dirfrag(dirfrag_t df) {
     if (!have_inode(df.ino)) return NULL;
-    return inode_map[df.ino]->get_dirfrag(df.frag);
+    return get_inode(df.ino)->get_dirfrag(df.frag);
   }
   /*
   void get_dirfrags_under(dirfrag_t df, list<CDir*>& ls) {
@@ -853,7 +855,7 @@ private:
   CInode *hack_pick_random_inode() {
     assert(!inode_map.empty());
     int n = rand() % inode_map.size();
-    hash_map<inodeno_t,CInode*>::iterator p = inode_map.begin();
+    hash_map<vinodeno_t,CInode*>::iterator p = inode_map.begin();
     while (n--) p++;
     return p->second;
   }

diff --git a/src/mds/Server.cc b/src/mds/Server.cc
index 81af42a867cbbfe2ead3950fd64eeb70942aa1cb..00ac84f7d0cdab20ccdf29fb29ec979d52a50f3c 100644 (file)
--- a/src/mds/Server.cc
+++ b/src/mds/Server.cc
@@ -1917,7 +1917,7 @@ void Server::handle_client_readdir(MDRequest *mdr)
     dout(12) << "including inode " << *in << dendl;
     
     // dentry
-    ::encode(it->first, dnbl);
+    ::encode(dn->name, dnbl);
     mds->locker->issue_client_lease(dn, client, dnbl, mdr->now, mdr->session);
 
     // inode

diff --git a/src/mds/mdstypes.h b/src/mds/mdstypes.h
index d4194f4f9e31988051ee1352611b4ef579255bf2..5089a479f4edc23a68cfb3546491ca9569783c22 100644 (file)
--- a/src/mds/mdstypes.h
+++ b/src/mds/mdstypes.h
@@ -150,6 +150,40 @@ inline ostream& operator<<(ostream &out, const frag_info_t &f) {
             << ")";    
 }
 
+struct vinodeno_t {
+  inodeno_t ino;
+  snapid_t snapid;
+
+  vinodeno_t(inodeno_t i, snapid_t s) : ino(i), snapid(s) {}
+};
+
+inline bool operator==(const vinodeno_t &l, const vinodeno_t &r) {
+  return l.ino == r.ino && l.snapid == r.snapid;
+}
+
+namespace __gnu_cxx {
+  template<> struct hash<vinodeno_t> {
+    size_t operator()(const vinodeno_t &vino) const { 
+      hash<inodeno_t> H;
+      hash<uint64_t> I;
+      return H(vino.ino) ^ I(vino.snapid);
+    }
+  };
+}
+
+
+
+
+inline ostream& operator<<(ostream &out, const vinodeno_t &vino) {
+  out << vino.ino;
+  if (vino.snapid == CEPH_NOSNAP)
+    out << ".head";
+  else if (vino.snapid)
+    out << '.' << vino.snapid;
+  return out;
+}
+
+
 struct inode_t {
   // base (immutable)
   inodeno_t ino;
@@ -241,6 +275,26 @@ struct inode_t {
 };
 WRITE_CLASS_ENCODER(inode_t)
 
+
+struct old_inode_t {
+  snapid_t first;
+  inodeno_t inode;
+  map<string,bufferptr> xattrs;
+
+  void encode(bufferlist& bl) const {
+    ::encode(first, bl);
+    ::encode(inode, bl);
+    ::encode(xattrs, bl);
+  }
+  void decode(bufferlist::iterator& bl) {
+    ::decode(first, bl);
+    ::decode(inode, bl);
+    ::decode(xattrs, bl);
+  }
+};
+WRITE_CLASS_ENCODER(old_inode_t)
+
+
 /*
  * like an inode, but for a dir frag 
  */
@@ -263,8 +317,28 @@ WRITE_CLASS_ENCODER(fnode_t)
 
 
 
+
+// =======
+// dentries
+
+typedef pair<snapid_t, const char *> dentry_key_t;
+
+struct ltdentrykey
+{
+  bool operator()(const dentry_key_t& k1,
+                 const dentry_key_t& k2) const
+  {
+    return 
+      k1.first < k2.first ||
+      (k1.first == k2.first && strcmp(k1.second, k2.second) < 0);
+  }
+};
+
+
+
+
 // =========
-// reqeusts
+// requests
 
 struct metareqid_t {
   entity_name_t name;

diff --git a/src/mds/snap.cc b/src/mds/snap.cc
index 1e6b9be647cd024a1f5568aac857ed142ed5b1e1..3bf27915f6757c1a297550ed7e93be670130aaeb 100644 (file)
--- a/src/mds/snap.cc
+++ b/src/mds/snap.cc
@@ -102,9 +102,9 @@ vector<snapid_t> *SnapRealm::get_snap_vector()
   int i = 0;
   for (set<snapid_t>::reverse_iterator p = s.rbegin(); p != s.rend(); p++)
     cached_snaps[i++] = *p;
-
-  dout(10) << "get_snap_vector " << cached_snaps << dendl;
-  //" (highwater " << cached_snaps_stamp << ")" << dendl;
+  
+  dout(10) << "get_snap_vector " << cached_snaps
+          << " (highwater " << snap_highwater << ")" << dendl;
   return &cached_snaps;
 }
 
@@ -126,7 +126,7 @@ void SnapRealm::split_at(SnapRealm *child)
           << " on " << *child->inode << dendl;
 
   // split open_children
-  dout(10) << " my children are " << open_children << dendl;
+  dout(10) << " open_children are " << open_children << dendl;
   for (set<SnapRealm*>::iterator p = open_children.begin();
        p != open_children.end(); ) {
     SnapRealm *realm = *p;