From 0d3338c2d2271bbf6f4c7774ad175f7ee42d07f6 Mon Sep 17 00:00:00 2001
From: Matt Benjamin <mbenjamin@redhat.com>
Date: Thu, 8 Aug 2019 16:21:24 -0400
Subject: [PATCH] rgw_file: dont deadlock in advance_mtime()

Fixes: https://tracker.ceph.com/issues/41173

Signed-off-by: Matt Benjamin <mbenjamin@redhat.com>
---
 src/rgw/rgw_file.cc | 17 ++++++++++++-----
 src/rgw/rgw_file.h  |  8 ++++----
 2 files changed, 16 insertions(+), 9 deletions(-)

diff --git a/src/rgw/rgw_file.cc b/src/rgw/rgw_file.cc
index b1ae5773cb3..da6c487a2b4 100644
--- a/src/rgw/rgw_file.cc
+++ b/src/rgw/rgw_file.cc
@@ -589,7 +589,7 @@ namespace rgw {
       /* save attrs */
       rgw_fh->encode_attrs(ux_key, ux_attrs);
       if (st)
-        rgw_fh->stat(st);
+        rgw_fh->stat(st, RGWFileHandle::FLAG_LOCKED);
       get<0>(mkr) = rgw_fh;
     } else {
       get<1>(mkr) = -EIO;
@@ -720,7 +720,7 @@ namespace rgw {
 	  parent->set_ctime(real_clock::to_timespec(t));
 	}
         if (st)
-          (void) rgw_fh->stat(st);
+	  (void) rgw_fh->stat(st, RGWFileHandle::FLAG_LOCKED);
 
         rgw_fh->set_etag(*(req.get_attr(RGW_ATTR_ETAG)));
         rgw_fh->set_acls(*(req.get_attr(RGW_ATTR_ACL))); 
@@ -1264,8 +1264,11 @@ namespace rgw {
     }
 
     bool initial_off;
+    char* mk{nullptr};
+
     if (likely(!! get<const char*>(&offset))) {
-      initial_off = ! get<const char*>(offset);
+      mk = const_cast<char*>(get<const char*>(offset));
+      initial_off = !mk;
     } else {
       initial_off = (*get<uint64_t*>(offset) == 0);
     }
@@ -1478,10 +1481,14 @@ namespace rgw {
     }
   }
 
-  void RGWFileHandle::advance_mtime() {
+  void RGWFileHandle::advance_mtime(uint32_t flags) {
     /* intended for use on directories, fast-forward mtime so as to
      * ensure a new, higher value for the change attribute */
-    lock_guard guard(mtx);
+    unique_lock uniq(mtx, std::defer_lock);
+    if (likely(! (flags & RGWFileHandle::FLAG_LOCKED))) {
+      uniq.lock();
+    }
+
     /* advance mtime only if stored mtime is older than the
      * configured namespace expiration */
     auto now = real_clock::now();
diff --git a/src/rgw/rgw_file.h b/src/rgw/rgw_file.h
index fa8dfa00d58..1600df5cf1e 100644
--- a/src/rgw/rgw_file.h
+++ b/src/rgw/rgw_file.h
@@ -238,7 +238,7 @@ namespace rgw {
     };
 
     void clear_state();
-    void advance_mtime();
+    void advance_mtime(uint32_t flags = FLAG_NONE);
 
     boost::variant<file, directory> variant_type;
 
@@ -436,7 +436,7 @@ namespace rgw {
 	state.ctime = st->st_ctim;
     }
 
-    int stat(struct stat* st) {
+    int stat(struct stat* st, uint32_t flags = FLAG_NONE) {
       /* partial Unix attrs */
       memset(st, 0, sizeof(struct stat));
       st->st_dev = state.dev;
@@ -450,7 +450,7 @@ namespace rgw {
       switch (fh.fh_type) {
       case RGW_FS_TYPE_DIRECTORY:
 	/* virtual directories are always invalid */
-	advance_mtime();
+	advance_mtime(flags);
 	st->st_nlink = state.nlink;
 	break;
       case RGW_FS_TYPE_FILE:
@@ -1077,7 +1077,7 @@ namespace rgw {
       fh_key fhk = parent->make_fhk(obj_name);
 
       lsubdout(get_context(), rgw, 10)
-	<< __func__ << " lookup called on "
+	<< __func__ << " called on "
 	<< parent->object_name() << " for " << key_name
 	<< " (" << obj_name << ")"
 	<< " -> " << fhk