From: Sage Weil <sage@redhat.com>
Date: Fri, 10 Apr 2015 23:49:07 +0000 (-0700)
Subject: os/newstore: use O_DIRECT is write is page-aligned
X-Git-Tag: v9.1.0~242^2~75
X-Git-Url: http://git-server-git.apps.pok.os.sepia.ceph.com/?a=commitdiff_plain;h=c67c9a2bee177183bdff1cc9ef56d13f438a2efe;p=ceph.git

os/newstore: use O_DIRECT is write is page-aligned

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

diff --git a/src/common/config_opts.h b/src/common/config_opts.h
index abb7a7a000fd..2c3d8ee52a6e 100644
--- a/src/common/config_opts.h
+++ b/src/common/config_opts.h
@@ -803,6 +803,7 @@ OPTION(newstore_nid_prealloc, OPT_INT, 1024)
 OPTION(newstore_overlay_max_length, OPT_INT, 65536)
 OPTION(newstore_overlay_max, OPT_INT, 32)
 OPTION(newstore_open_by_handle, OPT_BOOL, true)
+OPTION(newstore_o_direct, OPT_BOOL, true)
 
 OPTION(filestore_omap_backend, OPT_STR, "leveldb")
 
diff --git a/src/os/newstore/NewStore.cc b/src/os/newstore/NewStore.cc
index 81dd130b2920..8515eb5d0111 100644
--- a/src/os/newstore/NewStore.cc
+++ b/src/os/newstore/NewStore.cc
@@ -2277,7 +2277,14 @@ int NewStore::_do_wal_transaction(wal_transaction_t& wt)
       {
 	dout(20) << __func__ << " write " << p->fid << " "
 		 << p->offset << "~" << p->length << dendl;
-	int fd = _open_fid(p->fid, O_RDWR);
+	unsigned flags = O_RDWR;
+	if (g_conf->newstore_o_direct &&
+	    (p->offset & ~CEPH_PAGE_MASK) == 0 &&
+	    (p->length & ~CEPH_PAGE_MASK) == 0) {
+	  dout(20) << __func__ << " page-aligned, using O_DIRECT" << dendl;
+	  flags |= O_DIRECT;
+	}
+	int fd = _open_fid(p->fid, flags);
 	if (fd < 0)
 	  return fd;
 	int r = ::lseek64(fd, p->offset, SEEK_SET);
@@ -3022,12 +3029,19 @@ int NewStore::_do_write(TransContext *txc,
       o->onode.size == 0 ||
       o->onode.data_map.empty()) {
     _do_overlay_clear(txc, o);
+    unsigned flags = O_RDWR;
+    if (g_conf->newstore_o_direct &&
+	(offset & ~CEPH_PAGE_MASK) == 0 &&
+	(length & ~CEPH_PAGE_MASK) == 0) {
+      dout(20) << __func__ << " page-aligned, using O_DIRECT" << dendl;
+      flags |= O_DIRECT;
+    }
     if (o->onode.data_map.empty()) {
       // create
       fragment_t &f = o->onode.data_map[0];
       f.offset = 0;
       f.length = MAX(offset + length, o->onode.size);
-      fd = _create_fid(txc, &f.fid, O_RDWR);
+      fd = _create_fid(txc, &f.fid, flags);
       if (fd < 0) {
 	r = fd;
 	goto out;
@@ -3039,7 +3053,7 @@ int NewStore::_do_write(TransContext *txc,
       // append (possibly with gap)
       assert(o->onode.data_map.size() == 1);
       fragment_t &f = o->onode.data_map.rbegin()->second;
-      fd = _open_fid(f.fid, O_RDWR);
+      fd = _open_fid(f.fid, flags);
       if (fd < 0) {
 	r = fd;
 	goto out;