From dbee69927803ccf7aa592ca8718335dde16bb2f3 Mon Sep 17 00:00:00 2001
From: Yehuda Sadeh <yehuda@hq.newdream.net>
Date: Fri, 24 Jun 2011 14:09:30 -0700
Subject: [PATCH] rgw: atomic put

---
 src/include/rados/librados.hpp |  3 +++
 src/librados.cc                |  7 +++++++
 src/rgw/rgw_access.h           |  5 +++++
 src/rgw/rgw_op.cc              | 23 +++++++++++++++++++----
 src/rgw/rgw_rados.cc           | 28 ++++++++++++++++++++++++++++
 src/rgw/rgw_rados.h            |  4 ++++
 6 files changed, 66 insertions(+), 4 deletions(-)

diff --git a/src/include/rados/librados.hpp b/src/include/rados/librados.hpp
index 77ac98e42da22..f0ee082d18e0d 100644
--- a/src/include/rados/librados.hpp
+++ b/src/include/rados/librados.hpp
@@ -122,6 +122,9 @@ namespace librados
     void rmxattr(const char *name);
     void setxattr(const char *name, const bufferlist& bl);
     void tmap_update(const bufferlist& cmdbl);
+    void clone_range(uint64_t dst_off,
+                     const std::string& src_oid, uint64_t src_off,
+                     size_t len);
 
     void exec(const char *cls, const char *method, bufferlist& bl);
 
diff --git a/src/librados.cc b/src/librados.cc
index 98aa9116be1ec..c121307e01c10 100644
--- a/src/librados.cc
+++ b/src/librados.cc
@@ -195,6 +195,13 @@ void librados::ObjectOperation::tmap_update(const bufferlist& cmdbl)
   o->tmap_update(c);
 }
 
+void librados::ObjectOperation::clone_range(uint64_t dst_off,
+                     const std::string& src_oid, uint64_t src_off,
+                     size_t len)
+{
+  ::ObjectOperation *o = (::ObjectOperation *)impl;
+  o->clone_range(src_oid, src_off, len, dst_off);
+}
 
 
 librados::WatchCtx::
diff --git a/src/rgw/rgw_access.h b/src/rgw/rgw_access.h
index 1c563e12d292b..751c324684cbb 100644
--- a/src/rgw/rgw_access.h
+++ b/src/rgw/rgw_access.h
@@ -160,6 +160,11 @@ public:
   virtual int clone_range(rgw_obj& dst_obj, off_t dst_ofs,
                           rgw_obj& src_obj, off_t src_ofs,
                           size_t size) = 0;
+
+  virtual int clone_obj(rgw_obj& dst_obj, off_t dst_ofs,
+                        rgw_obj& src_obj, off_t src_ofs,
+                        size_t size,
+                        map<string, bufferlist> attrs) { return -ENOTSUP; }
  /**
    * a simple object read without keeping state
    */
diff --git a/src/rgw/rgw_op.cc b/src/rgw/rgw_op.cc
index 5b192a3f403a7..08a2e49885a85 100644
--- a/src/rgw/rgw_op.cc
+++ b/src/rgw/rgw_op.cc
@@ -19,6 +19,7 @@ using namespace std;
 using ceph::crypto::MD5;
 
 static string mp_ns = "multipart";
+static string tmp_ns = "tmp";
 
 class MultipartMetaFilter : public RGWAccessListFilter {
 public:
@@ -560,6 +561,12 @@ void RGWPutObj::execute()
     rgw_obj obj;
     if (!multipart) {
       oid = s->object_str;
+      obj.set_ns(tmp_ns);
+
+      char buf[33];
+      gen_rand_alphanumeric(buf, sizeof(buf) - 1);
+      oid.append("_");
+      oid.append(buf);
     } else {
       oid = s->object_str;
       string upload_id = s->args.get("uploadId");
@@ -644,11 +651,19 @@ void RGWPutObj::execute()
 
     get_request_metadata(s, attrs);
 
-    ret = rgwstore->put_obj_meta(s->user.user_id, obj, NULL, attrs, false);
-    if (ret < 0)
-      goto done;
+    if (!multipart) {
+      rgw_obj dst_obj(s->bucket_str, s->object_str);
+      ret = rgwstore->clone_obj(dst_obj, 0, obj, 0, s->obj_size, attrs);
+      if (ret < 0)
+        goto done;
+      ret = rgwstore->delete_obj(s->user.user_id, obj);
+      if (ret < 0)
+        goto done;
+    } else {
+      ret = rgwstore->put_obj_meta(s->user.user_id, obj, NULL, attrs, false);
+      if (ret < 0)
+        goto done;
 
-    if (multipart) {
       bl.clear();
       map<string, bufferlist> meta_attrs;
       RGWUploadPartInfo info;
diff --git a/src/rgw/rgw_rados.cc b/src/rgw/rgw_rados.cc
index de6b80c494e65..7b5f0537306b6 100644
--- a/src/rgw/rgw_rados.cc
+++ b/src/rgw/rgw_rados.cc
@@ -865,6 +865,34 @@ int RGWRados::clone_range(rgw_obj& dst_obj, off_t dst_ofs,
   return io_ctx.clone_range(dst_oid, dst_ofs, src_oid, src_ofs, size);
 }
 
+int RGWRados::clone_obj(rgw_obj& dst_obj, off_t dst_ofs,
+                        rgw_obj& src_obj, off_t src_ofs, size_t size,
+                        map<string, bufferlist> attrs)
+{
+  std::string& bucket = dst_obj.bucket;
+  std::string& dst_oid = dst_obj.object;
+  std::string& src_oid = src_obj.object;
+  librados::IoCtx io_ctx;
+
+  int r = open_bucket_ctx(bucket, io_ctx);
+  if (r < 0)
+    return r;
+
+  io_ctx.locator_set_key(dst_obj.key);
+  ObjectOperation op;
+  op.create(false);
+  map<string, bufferlist>::iterator iter;
+  for (iter = attrs.begin(); iter != attrs.end(); ++iter) {
+    const string& name = iter->first;
+    bufferlist& bl = iter->second;
+    op.setxattr(name.c_str(), bl);
+  }
+  op.clone_range(dst_ofs, src_oid, src_ofs, size);
+
+  bufferlist outbl;
+  int ret = io_ctx.operate(dst_oid, &op, &outbl);
+  return ret;
+}
 
 int RGWRados::get_obj(void **handle, rgw_obj& obj,
             char **data, off_t ofs, off_t end)
diff --git a/src/rgw/rgw_rados.h b/src/rgw/rgw_rados.h
index 9e31160e125a7..8948a95cf7069 100644
--- a/src/rgw/rgw_rados.h
+++ b/src/rgw/rgw_rados.h
@@ -54,6 +54,10 @@ public:
   virtual bool aio_completed(void *handle);
   virtual int clone_range(rgw_obj& dst_obj, off_t dst_ofs,
                           rgw_obj& src_obj, off_t src_ofs, size_t size);
+  virtual int clone_obj(rgw_obj& dst_obj, off_t dst_ofs,
+                        rgw_obj& src_obj, off_t src_ofs, size_t size,
+                        map<string, bufferlist> attrs);
+
   /** Copy an object, with many extra options */
   virtual int copy_obj(std::string& id, rgw_obj& dest_obj,
                rgw_obj& src_obj,
-- 
2.39.5