From: Milind Changire <mchangir@redhat.com>
Date: Mon, 25 Sep 2023 12:49:50 +0000 (+0530)
Subject: mds: do not simplify fragset
X-Git-Tag: v18.2.4~303^2
X-Git-Url: http://git.apps.os.sepia.ceph.com/?a=commitdiff_plain;h=b388fbd46f47f41012e0bbb72494d25ae691ec1e;p=ceph.git

mds: do not simplify fragset

Problem:
Frags in simplified fragset aren't found as is on the replica during scrub.

fragset::simplify() computes the bit representation of the least common
ancestor of the frags.  When this representation is forwarded to the replicas,
the ScrubStack::handle_scrub() method in the OP_QUEUEDIR case simply searches
this simplified representation in the list of (unsimplified) frags ...
which it fails to find.
Hence we get to see the "no frag 10*" type of logs.

Solution:
Do not simplify fragset when forwarding fragset to replica for scrub.

Fixes: https://tracker.ceph.com/issues/62658
Signed-off-by: Milind Changire <mchangir@redhat.com>
(cherry picked from commit f3c0424bb583423db80009c29f73e484aad840b4)
---

diff --git a/src/mds/ScrubStack.cc b/src/mds/ScrubStack.cc
index 6d799343f1496..742c464f4d37a 100644
--- a/src/mds/ScrubStack.cc
+++ b/src/mds/ScrubStack.cc
@@ -320,7 +320,7 @@ void ScrubStack::scrub_dir_inode(CInode *in, bool *added_children, bool *done)
 
   frag_vec_t frags;
   in->dirfragtree.get_leaves(frags);
-  dout(20) << __func__ << "recursive mode, frags " << frags << dendl;
+  dout(20) << __func__ << " recursive mode, frags " << frags << dendl;
   for (auto &fg : frags) {
     if (queued.contains(fg))
       continue;
@@ -366,7 +366,6 @@ void ScrubStack::scrub_dir_inode(CInode *in, bool *added_children, bool *done)
     scrub_r.tag = header->get_tag();
 
     for (auto& p : scrub_remote) {
-      p.second.simplify();
       dout(20) << __func__ << " forward " << p.second  << " to mds." << p.first << dendl;
       auto r = make_message<MMDSScrub>(MMDSScrub::OP_QUEUEDIR, in->ino(),
 				       std::move(p.second), header->get_tag(),