mds/journal: Journal referent remote dentry

Add machinery to journal the referent remote dentry.
The call to metablob's add_remote_dentry is just
adding remote dentry and not referent remote dentry
yet. This will be fixed as part of operations like
link, unlink, rename which makes use of referent
inodes.

Fixes: https://tracker.ceph.com/issues/54205
Signed-off-by: Kotresh HR <khiremat@redhat.com>

diff --git a/src/mds/MDCache.cc b/src/mds/MDCache.cc
index 51926b46ccd10e2db5b09e59e2f15de1bc39fc92..65f4b83f8889a8b04c5496d13c5deb9b533f1f09 100644 (file)
--- a/src/mds/MDCache.cc
+++ b/src/mds/MDCache.cc
@@ -577,8 +577,10 @@ void MDCache::_create_system_file(CDir *dir, std::string_view name, CInode *in,
   } else {
     predirty_journal_parents(mut, &le->metablob, in, dir, PREDIRTY_DIR, 1);
     journal_dirty_inode(mut.get(), &le->metablob, in);
+    //TODO: A referent inode for system file ??
     dn->push_projected_linkage(in->ino(), in->d_type());
-    le->metablob.add_remote_dentry(dn, true, in->ino(), in->d_type());
+    dout(10) << __func__ << " add remote dentry " << *dn << dendl;
+    le->metablob.add_remote_dentry(dn, true, in->ino(), in->d_type(), 0, nullptr);
     le->metablob.add_root(true, in);
   }
   if (mdir)

diff --git a/src/mds/Server.cc b/src/mds/Server.cc
index 23772c8f8b13f2d822023d1b298f42041c3cde36..a77ca844afb0dee612082a159d9007dcde4dc35f 100644 (file)
--- a/src/mds/Server.cc
+++ b/src/mds/Server.cc
@@ -7749,7 +7749,8 @@ void Server::_link_local(const MDRequestRef& mdr, CDentry *dn, CInode *targeti,
   le->metablob.add_client_req(mdr->reqid, mdr->client_request->get_oldest_client_tid());
   mdcache->predirty_journal_parents(mdr, &le->metablob, targeti, dn->get_dir(), PREDIRTY_DIR, 1);      // new dn
   mdcache->predirty_journal_parents(mdr, &le->metablob, targeti, 0, PREDIRTY_PRIMARY);           // targeti
-  le->metablob.add_remote_dentry(dn, true, targeti->ino(), targeti->d_type());  // new remote
+  // TODO: Pass referent inode upon creation. It's adding just remote dentry now
+  le->metablob.add_remote_dentry(dn, true, targeti->ino(), targeti->d_type(), 0, nullptr);  // new remote
   mdcache->journal_dirty_inode(mdr.get(), &le->metablob, targeti);
 
   // do this after predirty_*, to avoid funky extra dnl arg
@@ -7867,7 +7868,8 @@ void Server::_link_remote(const MDRequestRef& mdr, bool inc, CDentry *dn, CInode
   if (inc) {
     dn->pre_dirty();
     mdcache->predirty_journal_parents(mdr, &le->metablob, targeti, dn->get_dir(), PREDIRTY_DIR, 1);
-    le->metablob.add_remote_dentry(dn, true, targeti->ino(), targeti->d_type()); // new remote
+    // TODO: Pass referent inode upon creation. It's adding just remote dentry now
+    le->metablob.add_remote_dentry(dn, true, targeti->ino(), targeti->d_type(), 0, nullptr); // new remote
     dn->push_projected_linkage(targeti->ino(), targeti->d_type());
   } else {
     dn->pre_dirty();
@@ -9879,7 +9881,8 @@ void Server::_rename_prepare(const MDRequestRef& mdr,
       destdn->first = mdcache->get_global_snaprealm()->get_newest_seq() + 1;
 
     if (destdn->is_auth())
-      metablob->add_remote_dentry(destdn, true, srcdnl->get_remote_ino(), srcdnl->get_remote_d_type());
+      // TODO: Pass referent inode upon creation. It's adding just remote dentry now
+      metablob->add_remote_dentry(destdn, true, srcdnl->get_remote_ino(), srcdnl->get_remote_d_type(), 0, nullptr);
 
     if (srci->is_auth() ) { // it's remote
       if (mdr->peer_request) {

diff --git a/src/mds/events/EMetaBlob.h b/src/mds/events/EMetaBlob.h
index 02607ce8ba137a6202344a4c31b3be82ed3c5367..e9470d74e3bd05d684e70a7b991e91b4d5a80b2c 100644 (file)
--- a/src/mds/events/EMetaBlob.h
+++ b/src/mds/events/EMetaBlob.h
@@ -147,18 +147,24 @@ public:
     version_t dnv = 0;
     inodeno_t ino = 0;
     unsigned char d_type = '\0';
+    inodeno_t referent_ino = 0;
+    CInode::inode_const_ptr referent_inode;      // if it's not XXX should not be part of mempool; wait for std::pmr to simplify
     bool dirty = false;
 
     remotebit(std::string_view d, std::string_view an, snapid_t df, snapid_t dl, version_t v, inodeno_t i, unsigned char dt, bool dr) : 
       dn(d), alternate_name(an), dnfirst(df), dnlast(dl), dnv(v), ino(i), d_type(dt), dirty(dr) { }
+    remotebit(std::string_view d, std::string_view an, snapid_t df, snapid_t dl, version_t v, inodeno_t i, unsigned char dt, inodeno_t ref_ino, const CInode::inode_const_ptr& ref_inode, bool dr) :
+      dn(d), alternate_name(an), dnfirst(df), dnlast(dl), dnv(v), ino(i), d_type(dt), referent_ino(ref_ino), referent_inode(ref_inode), dirty(dr) {}
     explicit remotebit(bufferlist::const_iterator &p) { decode(p); }
     remotebit() = default;
 
-    void encode(bufferlist& bl) const;
+    void encode(bufferlist& bl, uint64_t features) const;
     void decode(bufferlist::const_iterator &bl);
     void print(std::ostream& out) const {
       out << " remotebit dn " << dn << " [" << dnfirst << "," << dnlast << "] dnv " << dnv
          << " ino " << ino
+         << " referent ino " << referent_ino
+         << " referent inode " << referent_inode
          << " dirty=" << dirty;
       if (!alternate_name.empty()) {
         out << " altn " << binstrprint(alternate_name, 8);
@@ -168,7 +174,7 @@ public:
     void dump(Formatter *f) const;
     static void generate_test_instances(std::list<remotebit*>& ls);
   };
-  WRITE_CLASS_ENCODER(remotebit)
+  WRITE_CLASS_ENCODER_FEATURES(remotebit)
 
   /*
    * nullbit - a null dentry
@@ -288,7 +294,7 @@ public:
       using ceph::encode;
       if (!dn_decoded) return;
       encode(dfull, dnbl, features);
-      encode(dremote, dnbl);
+      encode(dremote, dnbl, features);
       encode(dnull, dnbl);
     }
     void _decode_bits() const { 
@@ -425,21 +431,37 @@ private:
   }
 
   void add_remote_dentry(CDentry *dn, bool dirty) {
-    add_remote_dentry(add_dir(dn->get_dir(), false), dn, dirty, 0, 0);
+    add_remote_dentry(add_dir(dn->get_dir(), false), dn, dirty, 0, 0, 0, nullptr);
   }
-  void add_remote_dentry(CDentry *dn, bool dirty, inodeno_t rino, int rdt) {
-    add_remote_dentry(add_dir(dn->get_dir(), false), dn, dirty, rino, rdt);
+  void add_remote_dentry(CDentry *dn, bool dirty, inodeno_t rino, int rdt, inodeno_t referent_ino, CInode *ref_in) {
+    add_remote_dentry(add_dir(dn->get_dir(), false), dn, dirty, rino, rdt, referent_ino, ref_in);
   }
   void add_remote_dentry(dirlump& lump, CDentry *dn, bool dirty, 
-                        inodeno_t rino=0, unsigned char rdt=0) {
+                        inodeno_t rino=0, unsigned char rdt=0, inodeno_t referent_ino=0, CInode *ref_in=nullptr) {
     dn->check_corruption(false);
+    /* In multi-version inode, i.e., a file has hardlinks and the primary link is being deleted,
+     * the primary inode is added as remote in the journal. In this case, it will not have a
+     * referent inode. So referent_ino=0 and ref_in=nullptr.
+     */
     if (!rino) {
       rino = dn->get_projected_linkage()->get_remote_ino();
       rdt = dn->get_projected_linkage()->get_remote_d_type();
+      referent_ino = dn->get_projected_linkage()->get_referent_ino();
+      ref_in = dn->get_projected_linkage()->get_referent_inode();
     }
+
     lump.nremote++;
-    lump.add_dremote(dn->get_name(), dn->get_alternate_name(), dn->first, dn->last,
-                    dn->get_projected_version(), rino, rdt, dirty);
+    if (ref_in) {
+      ceph_assert(referent_ino > 0);
+      const auto& ref_pi = ref_in->get_projected_inode();
+      ceph_assert(ref_pi->version > 0);
+      lump.add_dremote(dn->get_name(), dn->get_alternate_name(), dn->first, dn->last,
+                       dn->get_projected_version(), rino, rdt, referent_ino, ref_pi, dirty);
+    } else {
+      ceph_assert(referent_ino == 0);
+      lump.add_dremote(dn->get_name(), dn->get_alternate_name(), dn->first, dn->last,
+                       dn->get_projected_version(), rino, rdt, dirty);
+    }
   }
 
   // return remote pointer to to-be-journaled inode
@@ -502,8 +524,8 @@ private:
     add_dentry(lump, dn, dn->is_dirty(), dirty_parent, dirty_pool);
   }
   void add_dentry(dirlump& lump, CDentry *dn, bool dirty, bool dirty_parent, bool dirty_pool) {
-    // primary or remote
-    if (dn->get_projected_linkage()->is_remote()) {
+    // primary or remote or referent_remote
+    if (dn->get_projected_linkage()->is_remote() || dn->get_projected_linkage()->is_referent_remote()) {
       add_remote_dentry(dn, dirty);
     } else if (dn->get_projected_linkage()->is_null()) {
       add_null_dentry(dn, dirty);
@@ -607,7 +629,7 @@ private:
 };
 WRITE_CLASS_ENCODER_FEATURES(EMetaBlob)
 WRITE_CLASS_ENCODER_FEATURES(EMetaBlob::fullbit)
-WRITE_CLASS_ENCODER(EMetaBlob::remotebit)
+WRITE_CLASS_ENCODER_FEATURES(EMetaBlob::remotebit)
 WRITE_CLASS_ENCODER(EMetaBlob::nullbit)
 WRITE_CLASS_ENCODER_FEATURES(EMetaBlob::dirlump)
 

diff --git a/src/mds/journal.cc b/src/mds/journal.cc
index 4bbf02a7413cf3c7ffd114859880abcfc99c2265..53f3bb7e588685480f0b8407cddf60d290dc4ebe 100644 (file)
--- a/src/mds/journal.cc
+++ b/src/mds/journal.cc
@@ -683,10 +683,9 @@ void EMetaBlob::fullbit::update_inode(MDSRank *mds, CInode *in)
 }
 
 // EMetaBlob::remotebit
-
-void EMetaBlob::remotebit::encode(bufferlist& bl) const
+void EMetaBlob::remotebit::encode(bufferlist& bl, uint64_t features) const
 {
-  ENCODE_START(3, 2, bl);
+  ENCODE_START(4, 2, bl);
   encode(dn, bl);
   encode(std::tuple{
     dnfirst,
@@ -697,12 +696,15 @@ void EMetaBlob::remotebit::encode(bufferlist& bl) const
       dirty,
   }, bl, 0);
   encode(alternate_name, bl);
+  encode(referent_ino, bl);
+  if (referent_ino)
+    encode(*referent_inode, bl, features);
   ENCODE_FINISH(bl);
 }
 
 void EMetaBlob::remotebit::decode(bufferlist::const_iterator &bl)
 {
-  DECODE_START_LEGACY_COMPAT_LEN(3, 2, 2, bl);
+  DECODE_START_LEGACY_COMPAT_LEN(4, 2, 2, bl);
   decode(dn, bl);
   decode(dnfirst, bl);
   decode(dnlast, bl);
@@ -712,6 +714,16 @@ void EMetaBlob::remotebit::decode(bufferlist::const_iterator &bl)
   decode(dirty, bl);
   if (struct_v >= 3)
     decode(alternate_name, bl);
+  if (struct_v >= 4) {
+    decode(referent_ino, bl);
+    if (referent_ino) {
+      auto _inode = CInode::allocate_inode();
+      decode(*_inode, bl);
+      referent_inode = std::move(_inode);
+    } else {
+      referent_inode = NULL;
+    }
+  }
   DECODE_FINISH(bl);
 }
 
@@ -745,14 +757,16 @@ void EMetaBlob::remotebit::dump(Formatter *f) const
   f->dump_string("d_type", type_string);
   f->dump_string("dirty", dirty ? "true" : "false");
   f->dump_string("alternate_name", alternate_name);
+  f->dump_int("referentino", referent_ino);
 }
 
 void EMetaBlob::remotebit::
 generate_test_instances(std::list<EMetaBlob::remotebit*>& ls)
 {
-  remotebit *remote = new remotebit("/test/dn", "", 0, 10, 15, 1, IFTODT(S_IFREG), false);
+  auto _inode = CInode::allocate_inode();
+  remotebit *remote = new remotebit("/test/dn", "", 0, 10, 15, 1, IFTODT(S_IFREG), 2, _inode, false);
   ls.push_back(remote);
-  remote = new remotebit("/test/dn2", "foo", 0, 10, 15, 1, IFTODT(S_IFREG), false);
+  remote = new remotebit("/test/dn2", "foo", 0, 10, 15, 1, IFTODT(S_IFREG), 2, _inode, false);
   ls.push_back(remote);
 }
 

diff --git a/src/tools/ceph-dencoder/mds_types.h b/src/tools/ceph-dencoder/mds_types.h
index 1272017c438ebfea680bcd2aabc7714acb21485d..8cc39f34df994964a36f6a5cb092bc48e959d47f 100644 (file)
--- a/src/tools/ceph-dencoder/mds_types.h
+++ b/src/tools/ceph-dencoder/mds_types.h
@@ -96,7 +96,7 @@ TYPE_FEATUREFUL_NOCOPY(EImportStart)
 
 #include "mds/events/EMetaBlob.h"
 TYPE_FEATUREFUL_NOCOPY(EMetaBlob::fullbit)
-TYPE(EMetaBlob::remotebit)
+TYPE_FEATUREFUL_NOCOPY(EMetaBlob::remotebit)
 TYPE(EMetaBlob::nullbit)
 TYPE_FEATUREFUL_NOCOPY(EMetaBlob::dirlump)
 TYPE_FEATUREFUL_NOCOPY(EMetaBlob)