From: Simon Gao <simon29rock@gmail.com>
Date: Wed, 3 Jun 2020 03:04:25 +0000 (+0800)
Subject: mds: rdlock file/nest lock when accumulating stats of subtree dirfrags
X-Git-Tag: v17.0.0~609^2~11
X-Git-Url: http://git.apps.os.sepia.ceph.com/?a=commitdiff_plain;h=c8c3ba439abab26736f0caa9919362c0b447dbac;p=ceph.git

mds: rdlock file/nest lock when accumulating stats of subtree dirfrags

Signed-off-by: Simon Gao <simon29rock@gmail.com>
Signed-off-by: "Yan, Zheng" <zyan@redhat.com>
---

diff --git a/src/include/ceph_fs.h b/src/include/ceph_fs.h
index fe98eef892428..f0cac9e1a0db2 100644
--- a/src/include/ceph_fs.h
+++ b/src/include/ceph_fs.h
@@ -415,6 +415,7 @@ enum {
 	CEPH_MDS_OP_ENQUEUE_SCRUB  = 0x01503,
 	CEPH_MDS_OP_REPAIR_FRAGSTATS = 0x01504,
 	CEPH_MDS_OP_REPAIR_INODESTATS = 0x01505,
+	CEPH_MDS_OP_RDLOCK_FRAGSSTATS = 0x01507
 };
 
 extern const char *ceph_mds_op_name(int op);
diff --git a/src/mds/CInode.cc b/src/mds/CInode.cc
index 3e771e30429a0..601a144e8c563 100644
--- a/src/mds/CInode.cc
+++ b/src/mds/CInode.cc
@@ -4811,7 +4811,12 @@ next:
     }
 
     bool check_dirfrag_rstats() {
-      return immediate(DIRFRAGS, 0);
+      if (in->has_subtree_root_dirfrag()) {
+	in->mdcache->rdlock_dirfrags_stats(in, get_internal_callback(DIRFRAGS));
+	return false;
+      } else {
+	return immediate(DIRFRAGS, 0);
+      }
     }
 
     bool _dirfrags(int rval) {
diff --git a/src/mds/MDCache.cc b/src/mds/MDCache.cc
index 234c630bef3cc..ade41bcb8f8a1 100644
--- a/src/mds/MDCache.cc
+++ b/src/mds/MDCache.cc
@@ -9650,6 +9650,9 @@ void MDCache::dispatch_request(MDRequestRef& mdr)
     case CEPH_MDS_OP_REPAIR_INODESTATS:
       repair_inode_stats_work(mdr);
       break;
+    case CEPH_MDS_OP_RDLOCK_FRAGSSTATS:
+      rdlock_dirfrags_stats_work(mdr);
+      break;
     default:
       ceph_abort();
     }
@@ -13056,6 +13059,40 @@ do_rdlocks:
   mds->server->respond_to_request(mdr, 0);
 }
 
+void MDCache::rdlock_dirfrags_stats(CInode *diri, MDSInternalContext* fin)
+{
+  MDRequestRef mdr = request_start_internal(CEPH_MDS_OP_RDLOCK_FRAGSSTATS);
+  mdr->pin(diri);
+  mdr->internal_op_private = diri;
+  mdr->internal_op_finish = fin;
+  return rdlock_dirfrags_stats_work(mdr);
+}
+
+void MDCache::rdlock_dirfrags_stats_work(MDRequestRef& mdr)
+{
+  CInode *diri = static_cast<CInode*>(mdr->internal_op_private);
+  dout(10) << __func__ << " " << *diri << dendl;
+  if (!diri->is_auth()) {
+    mds->server->respond_to_request(mdr, -ESTALE);
+    return;
+  }
+  if (!diri->is_dir()) {
+    mds->server->respond_to_request(mdr, -ENOTDIR);
+    return;
+  }
+
+  MutationImpl::LockOpVec lov;
+  lov.add_rdlock(&diri->dirfragtreelock);
+  lov.add_rdlock(&diri->nestlock);
+  lov.add_rdlock(&diri->filelock);
+  if (!mds->locker->acquire_locks(mdr, lov))
+    return;
+  dout(10) << __func__ << " start dirfrags : " << *diri << dendl;
+
+  mds->server->respond_to_request(mdr, 0);
+  return;
+}
+
 void MDCache::flush_dentry(std::string_view path, Context *fin)
 {
   if (is_readonly()) {
diff --git a/src/mds/MDCache.h b/src/mds/MDCache.h
index 8c0957646ad4e..2db916d70a73c 100644
--- a/src/mds/MDCache.h
+++ b/src/mds/MDCache.h
@@ -940,6 +940,7 @@ class MDCache {
 		     Formatter *f, Context *fin);
   void repair_inode_stats(CInode *diri);
   void repair_dirfrag_stats(CDir *dir);
+  void rdlock_dirfrags_stats(CInode *diri, MDSInternalContext *fin);
 
   // my leader
   MDSRank *mds;
@@ -1125,6 +1126,7 @@ class MDCache {
   void enqueue_scrub_work(MDRequestRef& mdr);
   void repair_inode_stats_work(MDRequestRef& mdr);
   void repair_dirfrag_stats_work(MDRequestRef& mdr);
+  void rdlock_dirfrags_stats_work(MDRequestRef& mdr);
 
   ceph::unordered_map<inodeno_t,CInode*> inode_map;  // map of head inodes by ino
   map<vinodeno_t, CInode*> snap_inode_map;  // map of snap inodes by ino
diff --git a/src/mds/MDSContinuation.h b/src/mds/MDSContinuation.h
index 97bae912a2a4b..e01522522556d 100644
--- a/src/mds/MDSContinuation.h
+++ b/src/mds/MDSContinuation.h
@@ -21,7 +21,7 @@
 class MDSContinuation : public Continuation {
 protected:
   Server *server;
-  MDSContext *get_internal_callback(int stage) {
+  MDSInternalContext *get_internal_callback(int stage) {
     return new MDSInternalContextWrapper(server->mds, get_callback(stage));
   }
   MDSIOContextBase *get_io_callback(int stage) {