os: remove read ops from transactions

diff --git a/src/ebofs/Ebofs.cc b/src/ebofs/Ebofs.cc
index e05bbdc21cad27638aebc5e7be0a5094bf8d6b2f..5a79e4ebffc8c6a4b3aec0ff215966fa16d9f12a 100644 (file)
--- a/src/ebofs/Ebofs.cc
+++ b/src/ebofs/Ebofs.cc
@@ -2464,64 +2464,7 @@ unsigned Ebofs::_apply_transaction(Transaction& t)
   while (t.have_op()) {
     int op = t.get_op();
     switch (op) {
-    case Transaction::OP_READ:
-      {
-        pobject_t oid;
-       t.get_oid(oid);
-        __u64 offset, len;
-       t.get_length(offset);
-       t.get_length(len);
-        bufferlist *pbl;
-       t.get_pbl(pbl);
-        if (_read(oid, offset, len, *pbl) < 0) {
-          dout(7) << "apply_transaction fail on _read" << dendl;
-          r &= bit;
-        }
-      }
-      break;
-
-    case Transaction::OP_STAT:
-      {
-        pobject_t oid;
-       t.get_oid(oid);
-        struct stat *st;
-       t.get_pstat(st);
-        if (_stat(oid, st) < 0) {
-          dout(7) << "apply_transaction fail on _stat" << dendl;
-          r &= bit;
-        }
-      }
-      break;
-
-    case Transaction::OP_GETATTR:
-      {
-        pobject_t oid;
-       t.get_oid(oid);
-       const char *attrname;
-       t.get_attrname(attrname);
-        pair<void*,int*> pattrval;
-       t.get_pattrval(pattrval);
-        if ((*(pattrval.second) = _getattr(oid, attrname, pattrval.first, *(pattrval.second))) < 0) {
-          dout(7) << "apply_transaction fail on _getattr" << dendl;
-          r &= bit;
-        }        
-      }
-      break;
-
-    case Transaction::OP_GETATTRS:
-      {
-        pobject_t oid;
-       t.get_oid(oid);
-        map<string,bufferptr> *pset;
-       t.get_pattrset(pset);
-        if (_getattrs(oid, *pset) < 0) {
-          dout(7) << "apply_transaction fail on _getattrs" << dendl;
-          r &= bit;
-        }        
-      }
-      break;
-
-
+ 
     case Transaction::OP_WRITE:
       {
         pobject_t oid;

diff --git a/src/os/FileStore.cc b/src/os/FileStore.cc
index dbe84670d5bda06d0d93f7b846c8e081f527ec8a..26cf7a10ac2021fad042273d8908ca53e011bbea 100644 (file)
--- a/src/os/FileStore.cc
+++ b/src/os/FileStore.cc
@@ -426,7 +426,7 @@ unsigned FileStore::apply_transaction(Transaction &t, Context *onsafe)
     return ObjectStore::apply_transaction(t, onsafe);
 
   // create transaction
-  int len = t.get_len() * 30;  // very conservative!  FIXME FIXME FIXME
+  int len = t.get_btrfs_len();
   dout(20) << "apply_transaction allocation btrfs usertrans len " << len << dendl;
   btrfs_ioctl_usertrans *trans =
     (btrfs_ioctl_usertrans *)new char[sizeof(*trans) + len * sizeof(trans->ops[0])];
@@ -439,56 +439,7 @@ unsigned FileStore::apply_transaction(Transaction &t, Context *onsafe)
     int op = t.get_op();
 
     switch (op) {
-    case Transaction::OP_READ:
-      {
-       coll_t cid;
-       pobject_t oid;
-       __u64 offset, len;
-       t.get_cid(cid);
-       t.get_oid(oid);
-       t.get_length(offset);
-       t.get_length(len);
-       bufferlist *pbl;
-       t.get_pbl(pbl);
-       read(cid, oid, offset, len, *pbl);
-      }
-      break;
-    case Transaction::OP_STAT:
-      {
-       coll_t cid;
-       t.get_cid(cid);
-       pobject_t oid;
-       t.get_oid(oid);
-       struct stat *st;
-       t.get_pstat(st);
-       stat(cid, oid, st);
-      }
-      break;
-    case Transaction::OP_GETATTR:
-      {
-       coll_t cid;
-       t.get_cid(cid);
-       pobject_t oid;
-       t.get_oid(oid);
-       const char *attrname;
-       t.get_attrname(attrname);
-       pair<void*,int*> pattrval;
-       t.get_pattrval(pattrval);
-       *pattrval.second = getattr(cid, oid, attrname, pattrval.first, *pattrval.second);
-      }
-      break;
-    case Transaction::OP_GETATTRS:
-      {
-       coll_t cid;
-       t.get_cid(cid);
-       pobject_t oid;
-       t.get_oid(oid);
-       map<string,bufferptr> *pset;
-       t.get_pattrset(pset);
-       getattrs(cid, oid, *pset);
-      }
-      break;
-      
+
     case Transaction::OP_WRITE:
     case Transaction::OP_ZERO:   // write actual zeros.
       {
@@ -616,24 +567,32 @@ unsigned FileStore::apply_transaction(Transaction &t, Context *onsafe)
        trans->len++;
       }
       break;
+
+
     case Transaction::OP_SETATTRS:
+    case Transaction::OP_COLL_SETATTRS:
       {
-       coll_t cid;
-       t.get_cid(cid);
-       pobject_t oid;
-       t.get_oid(oid);
-       map<string,bufferptr> *pattrset;
-       t.get_pattrset(pattrset);
-       //setattrs(cid, oid, *pattrset, 0);
-       
        // make note of old attrs
        map<string,bufferptr> oldattrs;
-       getattrs(cid, oid, oldattrs);
-       
-       dout(10) << "setattrs " << cid << " " << oid << dendl;
        char *fn = new char[80];
        str.push_back(fn);
-       get_coname(cid, oid, fn);
+
+       if (op == Transaction::OP_SETATTRS) {
+         coll_t cid;
+         t.get_cid(cid);
+         pobject_t oid;
+         t.get_oid(oid);
+         getattrs(cid, oid, oldattrs);
+         get_coname(cid, oid, fn);
+       } else {
+         coll_t cid;
+         t.get_cid(cid);
+         collection_getattrs(cid, oldattrs);
+         get_cdir(cid, fn);
+       }
+       map<string,bufferptr> *pattrset;
+       t.get_pattrset(pattrset);
+       
        for (map<string,bufferptr>::iterator p = pattrset->begin();
             p != pattrset->end();
             p++) {
@@ -713,10 +672,10 @@ unsigned FileStore::apply_transaction(Transaction &t, Context *onsafe)
        trans->ops[trans->len].args[0] = -2;
        trans->ops[trans->len].args[1] = -1;
        trans->len++;
-       trans->ops[trans->len].op = BTRFS_IOC_USERTRANS_CLONE;
+       trans->ops[trans->len].op = BTRFS_IOC_USERTRANS_CLOSE;
        trans->ops[trans->len].args[0] = -1;
        trans->len++;
-       trans->ops[trans->len].op = BTRFS_IOC_USERTRANS_CLONE;
+       trans->ops[trans->len].op = BTRFS_IOC_USERTRANS_CLOSE;
        trans->ops[trans->len].args[0] = -2;
        trans->len++;
       }
@@ -817,9 +776,9 @@ unsigned FileStore::apply_transaction(Transaction &t, Context *onsafe)
        trans->ops[trans->len].args[0] = (__u64)fn;
        trans->ops[trans->len].args[1] = (__u64)attrname;
        trans->len++;
-     }
+      }
       break;
-      
+
       
     default:
       cerr << "bad op " << op << std::endl;
@@ -828,6 +787,7 @@ unsigned FileStore::apply_transaction(Transaction &t, Context *onsafe)
   }  
 
   dout(20) << "apply_transaction final btrfs usertrans len is " << trans->len << dendl;
+  assert((int)trans->len <= (int)len);
 
   // apply
   int r = 0;

diff --git a/src/os/JournalingObjectStore.h b/src/os/JournalingObjectStore.h
index ff31a3735135aab9d6f83fc51005af5a8c499f93..fcb6c794b946044488d4d878074927ef296832f4 100644 (file)
--- a/src/os/JournalingObjectStore.h
+++ b/src/os/JournalingObjectStore.h
@@ -200,7 +200,7 @@ protected:
     } else
       queue_commit_waiter(onsafe);
   }
-
+  
   void journal_collection_setattrs(coll_t cid, map<string,bufferptr>& aset, Context *onsafe) {
     if (journal && journal->is_writeable()) {
       Transaction t;
@@ -211,7 +211,7 @@ protected:
     } else
       queue_commit_waiter(onsafe);
   }
-
+  
   void journal_sync(Context *onsafe) {
     if (journal) {  
       // journal empty transaction

diff --git a/src/os/ObjectStore.h b/src/os/ObjectStore.h
index b3b2caeb7fb534b0a5db167572564926bcd57fa3..8c9932c3279a0a34bf188234c49ac3cc76e8fd9a 100644 (file)
--- a/src/os/ObjectStore.h
+++ b/src/os/ObjectStore.h
@@ -43,6 +43,9 @@ using std::list;
  */
 
 
+static inline void encode(const map<string,bufferptr> *attrset, bufferlist &bl) {
+  ::encode(*attrset, bl);
+}
 
 class ObjectStore {
 public:
@@ -73,16 +76,11 @@ public:
    */
   class Transaction {
   public:
-    static const int OP_READ =          1;  // oid, offset, len, pbl
-    static const int OP_STAT =          2;  // oid, pstat
-    static const int OP_GETATTR =       3;  // oid, attrname, pattrval
-    static const int OP_GETATTRS =      4;  // oid, pattrset
-
     static const int OP_WRITE =        10;  // oid, offset, len, bl
     static const int OP_ZERO =         11;  // oid, offset, len
     static const int OP_TRUNCATE =     12;  // oid, len
     static const int OP_REMOVE =       13;  // oid
-    static const int OP_SETATTR =      14;  // oid, attrname, attrval
+    static const int OP_SETATTR =      14;  // oid, attrname, bl
     static const int OP_SETATTRS =     15;  // oid, attrset
     static const int OP_RMATTR =       16;  // oid, attrname
     static const int OP_CLONE =        17;  // oid, newoid
@@ -93,27 +91,29 @@ public:
     static const int OP_RMCOLL =       21;  // cid
     static const int OP_COLL_ADD =     22;  // cid, oid
     static const int OP_COLL_REMOVE =  23;  // cid, oid
-    static const int OP_COLL_SETATTR = 24;  // cid, attrname, attrval
+    static const int OP_COLL_SETATTR = 24;  // cid, attrname, bl
     static const int OP_COLL_RMATTR =  25;  // cid, attrname
     static const int OP_COLL_SETATTRS = 26;  // cid, attrset
 
   private:
+    int len;
+    int blen;  // for btrfs transactions
     list<int8_t> ops;
     list<bufferlist> bls;
     list<pobject_t> oids;
     list<coll_t> cids;
     list<int64_t> lengths;
+
+    // for these guys, just use a pointer.
+    // but, decode to a full value, and create pointers to that.
     list<const char*> attrnames;
     list<string> attrnames2;
-
-    // for reads only (not encoded)
-    list<bufferlist*> pbls;
-    list<struct stat*> psts;
-    list< pair<void*,int*> > pattrvals;
-    list< map<string,bufferptr>* > pattrsets;
+    list<map<string,bufferptr> *> attrsets;
+    list<map<string,bufferptr> > attrsets2;
 
   public:
-    int get_len() { return ops.size(); } // FIXME maintain a counter?
+    int get_len() { return len ? len : ops.size(); }
+    int get_btrfs_len() { return blen; }
 
     bool have_op() {
       return !ops.empty();
@@ -144,54 +144,9 @@ public:
       p = attrnames.front();
       attrnames.pop_front();
     }
-    void get_pbl(bufferlist* &pbl) {
-      pbl = pbls.front();
-      pbls.pop_front();
-    }
-    void get_pstat(struct stat* &pst) {
-      pst = psts.front();
-      psts.pop_front();
-    }
-    void get_pattrval(pair<void*,int*>& p) {
-      p = pattrvals.front();
-      pattrvals.pop_front();
-    }
     void get_pattrset(map<string,bufferptr>* &ps) {
-      ps = pattrsets.front();
-      pattrsets.pop_front();
-    }
-      
-
-    void read(coll_t cid, pobject_t oid, __u64 off, size_t len, bufferlist *pbl) {
-      int op = OP_READ;
-      ops.push_back(op);
-      cids.push_back(cid);
-      oids.push_back(oid);
-      lengths.push_back(off);
-      lengths.push_back(len);
-      pbls.push_back(pbl);
-    }
-    void stat(coll_t cid, pobject_t oid, struct stat *st) {
-      int op = OP_STAT;
-      ops.push_back(op);
-      cids.push_back(cid);
-      oids.push_back(oid);
-      psts.push_back(st);
-    }
-    void getattr(coll_t cid, pobject_t oid, const char* name, void* val, int *plen) {
-      int op = OP_GETATTR;
-      ops.push_back(op);
-      cids.push_back(cid);
-      oids.push_back(oid);
-      attrnames.push_back(name);
-      pattrvals.push_back(pair<void*,int*>(val,plen));
-    }
-    void getattrs(coll_t cid, pobject_t oid, map<string,bufferptr>& aset) {
-      int op = OP_GETATTRS;
-      ops.push_back(op);
-      cids.push_back(cid);
-      oids.push_back(oid);
-      pattrsets.push_back(&aset);
+      ps = attrsets.front();
+      attrsets.pop_front();
     }
 
     void write(coll_t cid, pobject_t oid, __u64 off, size_t len, const bufferlist& bl) {
@@ -202,6 +157,8 @@ public:
       lengths.push_back(off);
       lengths.push_back(len);
       bls.push_back(bl);
+      len++;
+      blen += 3 + bl.buffers().size();
     }
     void zero(coll_t cid, pobject_t oid, __u64 off, size_t len) {
       int op = OP_ZERO;
@@ -210,6 +167,8 @@ public:
       oids.push_back(oid);
       lengths.push_back(off);
       lengths.push_back(len);
+      len++;
+      blen += 3 + 1;
     }
     void trim_from_cache(coll_t cid, pobject_t oid, __u64 off, size_t len) {
       int op = OP_TRIMCACHE;
@@ -218,6 +177,7 @@ public:
       oids.push_back(oid);
       lengths.push_back(off);
       lengths.push_back(len);
+      len++;
     }
     void truncate(coll_t cid, pobject_t oid, __u64 off) {
       int op = OP_TRUNCATE;
@@ -225,12 +185,16 @@ public:
       cids.push_back(cid);
       oids.push_back(oid);
       lengths.push_back(off);
+      len++;
+      blen++;
     }
     void remove(coll_t cid, pobject_t oid) {
       int op = OP_REMOVE;
       ops.push_back(op);
       cids.push_back(cid);
       oids.push_back(oid);
+      len++;
+      blen++;
     }
     void setattr(coll_t cid, pobject_t oid, const char* name, const void* val, int len) {
       int op = OP_SETATTR;
@@ -238,17 +202,20 @@ public:
       cids.push_back(cid);
       oids.push_back(oid);
       attrnames.push_back(name);
-      //attrvals.push_back(pair<const void*,int>(val,len));
       bufferlist bl;
       bl.append((char*)val,len);
       bls.push_back(bl);
+      len++;
+      blen++;
     }
     void setattrs(coll_t cid, pobject_t oid, map<string,bufferptr>& attrset) {
       int op = OP_SETATTRS;
       ops.push_back(op);
       cids.push_back(cid);
       oids.push_back(oid);
-      pattrsets.push_back(&attrset);
+      attrsets.push_back(&attrset);
+      len++;
+      blen += 5 + attrset.size();     // HACK allowance for removing old attrs
     }
     void rmattr(coll_t cid, pobject_t oid, const char* name) {
       int op = OP_RMATTR;
@@ -256,6 +223,8 @@ public:
       cids.push_back(cid);
       oids.push_back(oid);
       attrnames.push_back(name);
+      len++;
+      blen++;
     }
     void clone(coll_t cid, pobject_t oid, pobject_t noid) {
       int op = OP_CLONE;
@@ -263,16 +232,22 @@ public:
       cids.push_back(cid);
       oids.push_back(oid);
       oids.push_back(noid);
+      len++;
+      blen += 5;
     }
     void create_collection(coll_t cid) {
       int op = OP_MKCOLL;
       ops.push_back(op);
       cids.push_back(cid);
+      len++;
+      blen++;
     }
     void remove_collection(coll_t cid) {
       int op = OP_RMCOLL;
       ops.push_back(op);
       cids.push_back(cid);
+      len++;
+      blen++;
     }
     void collection_add(coll_t cid, coll_t ocid, pobject_t oid) {
       int op = OP_COLL_ADD;
@@ -280,13 +255,17 @@ public:
       cids.push_back(cid);
       cids.push_back(ocid);
       oids.push_back(oid);
+      len++;
+      blen++;
     }
     void collection_remove(coll_t cid, pobject_t oid) {
       int op = OP_COLL_REMOVE;
       ops.push_back(op);
       cids.push_back(cid);
       oids.push_back(oid);
-    }
+      len++;
+       blen++;
+   }
     void collection_setattr(coll_t cid, const char* name, const void* val, int len) {
       int op = OP_COLL_SETATTR;
       ops.push_back(op);
@@ -295,24 +274,31 @@ public:
       bufferlist bl;
       bl.append((char*)val, len);
       bls.push_back(bl);
+      len++;
+      blen++;
     }
     void collection_rmattr(coll_t cid, const char* name) {
       int op = OP_COLL_RMATTR;
       ops.push_back(op);
       cids.push_back(cid);
       attrnames.push_back(name);
+      len++;
+      blen++;
     }
     void collection_setattrs(coll_t cid, map<string,bufferptr>& aset) {
       int op = OP_COLL_SETATTRS;
       ops.push_back(op);
       cids.push_back(cid);
-      pattrsets.push_back(&aset);
+      attrsets.push_back(&aset);
+      len++;
+      blen += 5 + aset.size();
     }
 
+
     // etc.
-    Transaction() {}
-    Transaction(bufferlist::iterator &p) { decode(p); }
-    Transaction(bufferlist &bl) { 
+    Transaction() : len(0) {}
+    Transaction(bufferlist::iterator &p) : len(0) { decode(p); }
+    Transaction(bufferlist &bl) : len(0) { 
       bufferlist::iterator p = bl.begin();
       decode(p); 
     }
@@ -324,6 +310,7 @@ public:
       ::encode(cids, bl);
       ::encode(lengths, bl);
       ::encode(attrnames, bl);
+      ::encode(attrsets, bl);
     }
     void decode(bufferlist::iterator &bl) {
       ::decode(ops, bl);
@@ -336,6 +323,11 @@ public:
           p != attrnames2.end();
           ++p)
        attrnames.push_back((*p).c_str());
+      ::decode(attrsets2, bl);
+      for (list<map<string,bufferptr> >::iterator p = attrsets2.begin();
+          p != attrsets2.end();
+          ++p)
+       attrsets.push_back(&(*p));
     }
   };
 
@@ -351,56 +343,6 @@ public:
     while (t.have_op()) {
       int op = t.get_op();
       switch (op) {
-      case Transaction::OP_READ:
-        {
-         coll_t cid;
-          pobject_t oid;
-          __u64 offset, len;
-         t.get_cid(cid);
-         t.get_oid(oid);
-         t.get_length(offset);
-         t.get_length(len);
-          bufferlist *pbl;
-         t.get_pbl(pbl);
-          read(cid, oid, offset, len, *pbl);
-        }
-        break;
-      case Transaction::OP_STAT:
-        {
-         coll_t cid;
-         t.get_cid(cid);
-          pobject_t oid;
-         t.get_oid(oid);
-          struct stat *st;
-         t.get_pstat(st);
-          stat(cid, oid, st);
-        }
-        break;
-      case Transaction::OP_GETATTR:
-        {
-                 coll_t cid;
-         t.get_cid(cid);
-         pobject_t oid;
-         t.get_oid(oid);
-          const char *attrname;
-         t.get_attrname(attrname);
-          pair<void*,int*> pattrval;
-         t.get_pattrval(pattrval);
-          *pattrval.second = getattr(cid, oid, attrname, pattrval.first, *pattrval.second);
-        }
-        break;
-      case Transaction::OP_GETATTRS:
-        {
-         coll_t cid;
-         t.get_cid(cid);
-          pobject_t oid;
-         t.get_oid(oid);
-          map<string,bufferptr> *pset;
-         t.get_pattrset(pset);
-          getattrs(cid, oid, *pset);
-        }
-        break;
-
       case Transaction::OP_WRITE:
         {
          coll_t cid;

diff --git a/src/osd/ReplicatedPG.cc b/src/osd/ReplicatedPG.cc
index bf61abaf0ab716aacb0f3207dbc66d8bbbbdef3c..31a24e0396122034f69dc9a48e0e3e67205feac4 100644 (file)
--- a/src/osd/ReplicatedPG.cc
+++ b/src/osd/ReplicatedPG.cc
@@ -1407,15 +1407,12 @@ void ReplicatedPG::push(pobject_t poid, int peer)
   // read data+attrs
   bufferlist bl;
   eversion_t v;
-  int vlen = sizeof(v);
+  size_t vlen = sizeof(v);
   map<string,bufferptr> attrset;
   
-  ObjectStore::Transaction t;
-  t.read(info.pgid, poid, 0, 0, &bl);
-  t.getattr(info.pgid, poid, "version", &v, &vlen);
-  t.getattrs(info.pgid, poid, attrset);
-  unsigned tr = osd->store->apply_transaction(t);
-  assert(tr == 0);  // !!!
+  osd->store->read(info.pgid, poid, 0, 0, bl);
+  osd->store->getattr(info.pgid, poid, "version", &v, vlen);
+  osd->store->getattrs(info.pgid, poid, attrset);
 
   // ok
   dout(7) << "push " << poid << " v " << v