os/bluestore: new write path

- simplified wal_op_t.  we still have overlays in there, although that
  might need to get removed soon too.
- init_csum cleanup
- totally new write path

Signed-off-by: Sage Weil <sage@redhat.com>

src/os/bluestore/BlueStore.h

@@ -938,6 +938,41 @@ private:
   // --------------------------------------------------------
   // write ops
 
+  struct WriteContext {
+    unsigned fadvise_flags;                    ///< write flags
+    bool buffered;                             ///< buffered write
+
+    //map<uint64_t,bluestore_lextent_t> lex_new; ///< new lextents
+    vector<bluestore_lextent_t> lex_old;       ///< must deref blobs
+    vector<bluestore_blob_t*> blob_new;        ///< new blobs
+    vector<bufferlist> bl_new;                 ///< new data, for above blobs
+
+    WriteContext() : fadvise_flags(0), buffered(false) {}
+  };
+
+  void _do_write_small(
+    TransContext *txc,
+    CollectionRef &c,
+    OnodeRef o,
+    uint64_t offset, uint64_t length,
+    bufferlist::iterator& blp,
+    WriteContext *wctx);
+  void _do_write_big(
+    TransContext *txc,
+    CollectionRef &c,
+    OnodeRef o,
+    uint64_t offset, uint64_t length,
+    bufferlist::iterator& blp,
+    WriteContext *wctx);
+  int _do_alloc_write(
+    TransContext *txc,
+    WriteContext *wctx);
+  void _wctx_finish(
+    TransContext *txc,
+    CollectionRef& c,
+    OnodeRef o,
+    WriteContext *wctx);
+
   int _do_transaction(Transaction *t,
 		      TransContext *txc,
 		      ThreadPool::TPHandle *handle);
@@ -971,18 +1006,6 @@ private:
 		       OnodeRef o, bufferlist *bl,
 		       uint64_t offset, uint64_t *length,
 		       uint64_t chunk_size);
-  int _do_allocate(TransContext *txc,
-		   CollectionRef& c,
-		   OnodeRef o,
-		   uint64_t offset, uint64_t length,
-		   uint32_t fadvise_flags,
-		   bool allow_overlay,
-		   uint64_t *alloc_offset,
-		   uint64_t *alloc_length,
-		   uint64_t *cow_head_extent,
-		   uint64_t *cow_tail_extent,
-		   uint64_t *rmw_cow_head,
-		   uint64_t *rmw_cow_tail);
   int _do_write(TransContext *txc,
 		CollectionRef &c,
 		OnodeRef o,
@@ -992,15 +1015,6 @@ private:
   int _touch(TransContext *txc,
 	     CollectionRef& c,
 	     OnodeRef& o);
-  int _do_write_zero(TransContext *txc,
-		     CollectionRef &c,
-		     OnodeRef o,
-		     uint64_t offset, uint64_t length);
-  void _do_zero_tail_extent(
-    TransContext *txc,
-    CollectionRef& c,
-    OnodeRef& o,
-    uint64_t offset);
   int _do_zero(TransContext *txc,
 	       CollectionRef& c,
 	       OnodeRef& o,

src/os/bluestore/bluestore_types.cc

@@ -748,15 +748,10 @@ void bluestore_wal_op_t::encode(bufferlist& bl) const
 {
   ENCODE_START(1, 1, bl);
   ::encode(op, bl);
-  ::encode(extent, bl);
-  ::encode(src_extent, bl);
-  ::encode(src_rmw_head, bl);
-  ::encode(src_rmw_tail, bl);
+  ::encode(extents, bl);
+  ::encode(data, bl);
   ::encode(nid, bl);
   ::encode(overlays, bl);
-  if (!overlays.size()) {
-    ::encode(data, bl);
-  }
   ::encode(removed_overlays, bl);
   ENCODE_FINISH(bl);
 }
@@ -765,15 +760,10 @@ void bluestore_wal_op_t::decode(bufferlist::iterator& p)
 {
   DECODE_START(1, p);
   ::decode(op, p);
-  ::decode(extent, p);
-  ::decode(src_extent, p);
-  ::decode(src_rmw_head, p);
-  ::decode(src_rmw_tail, p);
+  ::decode(extents, p);
+  ::decode(data, p);
   ::decode(nid, p);
   ::decode(overlays, p);
-  if (!overlays.size()) {
-    ::decode(data, p);
-  }
   ::decode(removed_overlays, p);
   DECODE_FINISH(p);
 }
@@ -781,21 +771,21 @@ void bluestore_wal_op_t::decode(bufferlist::iterator& p)
 void bluestore_wal_op_t::dump(Formatter *f) const
 {
   f->dump_unsigned("op", (int)op);
-  f->dump_object("extent", extent);
-  f->dump_object("src_extent", src_extent);
-  f->dump_unsigned("src_rmw_head", src_rmw_head);
-  f->dump_unsigned("src_rmw_tail", src_rmw_tail);
+  f->dump_unsigned("data_len", data.length());
+  f->open_array_section("extents");
+  for (auto& e : extents) {
+    f->dump_object("extent", e);
+  }
+  f->close_section();
   f->dump_unsigned("nid", nid);
   f->open_array_section("overlays");
-  for (vector<bluestore_overlay_t>::const_iterator p = overlays.begin();
-       p != overlays.end(); ++p) {
-    f->dump_object("overlay", *p);
+  for (auto& o : overlays) {
+    f->dump_object("overlay", o);
   }
   f->close_section();
   f->open_array_section("removed_overlays");
-  for (vector<uint64_t>::const_iterator p = removed_overlays.begin();
-       p != removed_overlays.end(); ++p) {
-    f->dump_unsigned("key", *p);
+  for (auto key : removed_overlays) {
+    f->dump_unsigned("key", key);
   }
   f->close_section();
 }
@@ -805,19 +795,9 @@ void bluestore_wal_op_t::generate_test_instances(list<bluestore_wal_op_t*>& o)
   o.push_back(new bluestore_wal_op_t);
   o.push_back(new bluestore_wal_op_t);
   o.back()->op = OP_WRITE;
-  o.back()->extent.offset = 1;
-  o.back()->extent.length = 2;
-  o.back()->src_extent.offset = 10000;
-  o.back()->src_extent.length = 2;
-  o.back()->src_rmw_head = 22;
-  o.back()->src_rmw_tail = 88;
+  o.back()->extents.push_back(bluestore_pextent_t(1, 2));
+  o.back()->extents.push_back(bluestore_pextent_t(100, 5));
   o.back()->data.append("my data");
-  o.back()->nid = 3;
-  o.back()->overlays.push_back(bluestore_overlay_t());
-  o.back()->overlays.push_back(bluestore_overlay_t());
-  o.back()->overlays.back().key = 4;
-  o.back()->overlays.back().value_offset = 5;
-  o.back()->overlays.back().length = 6;
 }
 
 void bluestore_wal_transaction_t::encode(bufferlist& bl) const
@@ -865,8 +845,6 @@ void bluestore_wal_transaction_t::generate_test_instances(list<bluestore_wal_tra
   o.back()->ops.push_back(bluestore_wal_op_t());
   o.back()->ops.push_back(bluestore_wal_op_t());
   o.back()->ops.back().op = bluestore_wal_op_t::OP_WRITE;
-  o.back()->ops.back().extent.offset = 2;
-  o.back()->ops.back().extent.length = 3;
+  o.back()->ops.back().extents.push_back(bluestore_pextent_t(1,7));
   o.back()->ops.back().data.append("foodata");
-  o.back()->ops.back().nid = 4;
 }

src/os/bluestore/bluestore_types.h

@@ -415,11 +415,10 @@ struct bluestore_blob_t {
     return &csum_data[cs * i];
   }
 
-  void init_csum(unsigned type, unsigned order) {
+  void init_csum(unsigned type, unsigned order, unsigned len) {
     csum_type = type;
     csum_block_order = order;
-    csum_data.resize(get_csum_value_size() * get_ondisk_length() /
-		     get_csum_block_size());
+    csum_data.resize(get_csum_value_size() * len / get_csum_block_size());
   }
 };
 WRITE_CLASS_ENCODER(bluestore_blob_t)
@@ -606,20 +605,16 @@ WRITE_CLASS_ENCODER(bluestore_onode_t)
 struct bluestore_wal_op_t {
   typedef enum {
     OP_WRITE = 1,
-    OP_COPY = 2,
-    OP_ZERO = 4,
   } type_t;
   __u8 op = 0;
-  bluestore_pextent_t extent;
-  bluestore_pextent_t src_extent;
-  uint64_t src_rmw_head, src_rmw_tail;
+
+  vector<bluestore_pextent_t> extents;
   bufferlist data;
+
   uint64_t nid;
   vector<bluestore_overlay_t> overlays;
   vector<uint64_t> removed_overlays;
 
-  bluestore_wal_op_t() : src_rmw_head(0), src_rmw_tail(0), nid(0) {}
-
   void encode(bufferlist& bl) const;
   void decode(bufferlist::iterator& p);
   void dump(Formatter *f) const;
@@ -638,21 +633,6 @@ struct bluestore_wal_transaction_t {
 
   bluestore_wal_transaction_t() : seq(0), _bytes(-1) {}
 
-#if 0
-  no users for htis
-  uint64_t get_bytes() {
-    if (_bytes < 0) {
-      _bytes = 0;
-      for (list<bluestore_wal_op_t>::iterator p = ops.begin();
-	   p != ops.end();
-	   ++p) {
-	_bytes += p->extent.length;
-      }
-    }
-    return _bytes;
-  }
-#endif
-
   void encode(bufferlist& bl) const;
   void decode(bufferlist::iterator& p);
   void dump(Formatter *f) const;