os/bluestore: add FLAG_NOCACHE flag; do not cache unbuffered writes

Add a Buffer flag to mark that a buffer should not be cached once it is
stable.

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

diff --git a/src/os/bluestore/BlueStore.cc b/src/os/bluestore/BlueStore.cc
index 96867224cbc40402e14fffb5f6399fa9340db325..7e9ef9a3dc1ff63f64c8cce48616d1aa15b8e399 100644 (file)
--- a/src/os/bluestore/BlueStore.cc
+++ b/src/os/bluestore/BlueStore.cc
@@ -443,9 +443,12 @@ static void get_wal_key(uint64_t seq, string *out)
 
 ostream& operator<<(ostream& out, const BlueStore::Buffer& b)
 {
-  return out << "buffer(space " << b.space << " 0x" << std::hex
-            << b.offset << "~" << b.length << std::dec
-            << " " << BlueStore::Buffer::get_state_name(b.state) << ")";
+  out << "buffer(space " << b.space << " 0x" << std::hex
+      << b.offset << "~" << b.length << std::dec
+      << " " << BlueStore::Buffer::get_state_name(b.state);
+  if (b.flags)
+    out << " " << BlueStore::Buffer::get_flag_name(b.flags);
+  return out << ")";
 }
 
 void BlueStore::BufferCache::trim(uint64_t keep)
@@ -577,6 +580,27 @@ void BlueStore::BufferSpace::read(
   }
 }
 
+void BlueStore::BufferSpace::finish_write(uint64_t seq)
+{
+  auto i = writing.begin();
+  while (i != writing.end()) {
+    Buffer *b = &*i;
+    dout(20) << __func__ << " " << *b << dendl;
+    assert(b->is_writing());
+    if (b->seq <= seq) {
+      if (b->flags & Buffer::FLAG_NOCACHE) {
+       ++i;
+       _rm_buffer(b);
+      } else {
+       b->state = Buffer::STATE_CLEAN;
+       writing.erase(i++);
+      }
+    } else {
+      ++i;
+    }
+  }
+}
+
 
 // Bnode
 
@@ -5849,7 +5873,7 @@ int BlueStore::_do_write(
     wctx.comp_blob_size = comp_min_blob_size;
 
   // write in buffer cache
-  o->bc.write(txc->seq, offset, bl);
+  o->bc.write(txc->seq, offset, bl, wctx.buffered ? 0 : Buffer::FLAG_NOCACHE);
 
   bufferlist::iterator p = bl.begin();
   if (offset / min_alloc_size == (end - 1) / min_alloc_size &&

diff --git a/src/os/bluestore/BlueStore.h b/src/os/bluestore/BlueStore.h
index 4dd0adc6a2f4e7c1e81d9b9b9a1070385e3bf59a..ec9597947487fecb7c3f9faf9eeb87fe123c2a82 100644 (file)
--- a/src/os/bluestore/BlueStore.h
+++ b/src/os/bluestore/BlueStore.h
@@ -117,9 +117,20 @@ public:
       default: return "???";
       }
     }
+    enum {
+      FLAG_NOCACHE = 1,  ///< trim when done WRITING (do not become CLEAN)
+      // NOTE: fix operator<< when you define a second flag
+    };
+    static const char *get_flag_name(int s) {
+      switch (s) {
+      case FLAG_NOCACHE: return "nocache";
+      default: return "???";
+      }
+    }
 
     BufferSpace *space;
-    unsigned state;            ///< STATE_*
+    uint32_t state;            ///< STATE_*
+    uint32_t flags;            ///< FLAG_*
     uint64_t seq;
     uint64_t offset, length;
     bufferlist data;
@@ -127,10 +138,13 @@ public:
     boost::intrusive::list_member_hook<> lru_item;
     boost::intrusive::list_member_hook<> state_item;
 
-    Buffer(BufferSpace *space, unsigned s, uint64_t q, uint64_t o, uint64_t l)
-      : space(space), state(s), seq(q), offset(o), length(l) {}
-    Buffer(BufferSpace *space, unsigned s, uint64_t q, uint64_t o, bufferlist& b)
-      : space(space), state(s), seq(q), offset(o), length(b.length()), data(b) {}
+    Buffer(BufferSpace *space, unsigned s, uint64_t q, uint64_t o, uint64_t l,
+          unsigned f = 0)
+      : space(space), state(s), flags(f), seq(q), offset(o), length(l) {}
+    Buffer(BufferSpace *space, unsigned s, uint64_t q, uint64_t o, bufferlist& b,
+          unsigned f = 0)
+      : space(space), state(s), flags(f), seq(q), offset(o),
+       length(b.length()), data(b) {}
 
     bool is_clean() const {
       return state == STATE_CLEAN;
@@ -203,6 +217,9 @@ public:
        writing.push_back(*b);
       }
     }
+    void _rm_buffer(Buffer *b) {
+      _rm_buffer(buffer_map.find(b->offset));
+    }
     void _rm_buffer(map<uint64_t,std::unique_ptr<Buffer>>::iterator p) {
       cache->size -= p->second->length;
       cache->lru.erase(cache->lru.iterator_to(*p->second));
@@ -236,22 +253,12 @@ public:
 
     void discard(uint64_t offset, uint64_t length);
 
-    void write(uint64_t seq, uint64_t offset, bufferlist& bl) {
+    void write(uint64_t seq, uint64_t offset, bufferlist& bl, unsigned flags) {
       discard(offset, bl.length());
-      _add_buffer(new Buffer(this, Buffer::STATE_WRITING, seq, offset, bl));
-    }
-    void finish_write(uint64_t seq) {
-      auto i = writing.begin();
-      while (i != writing.end()) {
-       assert(i->is_writing());
-       if (i->seq <= seq) {
-         i->state = Buffer::STATE_CLEAN;
-         writing.erase(i++);
-       } else {
-         ++i;
-       }
-      }
+      _add_buffer(new Buffer(this, Buffer::STATE_WRITING, seq, offset, bl,
+                            flags));
     }
+    void finish_write(uint64_t seq);
     void did_read(uint64_t offset, bufferlist& bl) {
       discard(offset, bl.length());
       _add_buffer(new Buffer(this, Buffer::STATE_CLEAN, 0, offset, bl));