cephfs: InodeStat versioning

* Use the new feature bit CEPHFS_FEATURE_REPLY_ENCODING for encoding.
* encode/decode in the new format.
* The old format is maintained for backward compatibility.

Fixes: http://tracker.ceph.com/issues/24444
Signed-off-by: Jos Collin <jcollin@redhat.com>

diff --git a/src/client/Client.cc b/src/client/Client.cc
index 111d6c05b8b45766a4b168bce697591da7772a7d..14880f8cc21303dea74424dc28c34d6c72fe7b34 100644 (file)
--- a/src/client/Client.cc
+++ b/src/client/Client.cc
@@ -1068,7 +1068,13 @@ void Client::insert_readdir_results(MetaRequest *request, MetaSession *session,
 
   MClientReply *reply = request->reply;
   ConnectionRef con = request->reply->get_connection();
-  uint64_t features = con->get_features();
+  uint64_t features;
+  if(session->mds_features.test(CEPHFS_FEATURE_REPLY_ENCODING)) {
+    features = (uint64_t)-1;
+  }
+  else {
+    features = con->get_features();
+  }
 
   dir_result_t *dirp = request->dirp;
   assert(dirp);
@@ -1266,7 +1272,13 @@ Inode* Client::insert_trace(MetaRequest *request, MetaSession *session)
   }
 
   ConnectionRef con = request->reply->get_connection();
-  uint64_t features = con->get_features();
+  uint64_t features;
+  if (session->mds_features.test(CEPHFS_FEATURE_REPLY_ENCODING)) {
+    features = (uint64_t)-1;
+  }
+  else {
+    features = con->get_features();
+  }
   ldout(cct, 10) << " features 0x" << hex << features << dec << dendl;
 
   // snap trace
@@ -2072,6 +2084,7 @@ void Client::handle_client_session(MClientSession *m)
        _closed_mds_session(session);
        break;
       }
+      session->mds_features = std::move(m->supported_features);
 
       renew_caps(session);
       session->state = MetaSession::STATE_OPEN;

diff --git a/src/client/MetaSession.h b/src/client/MetaSession.h
index 08159ac08b5715635a428ea9d464002b617d3919..0879abdbe9f1ad61c3cb5355ae6bd6789dfaf673 100644 (file)
--- a/src/client/MetaSession.h
+++ b/src/client/MetaSession.h
@@ -24,6 +24,7 @@ struct MetaSession {
   utime_t cap_ttl, last_cap_renew_request;
   uint64_t cap_renew_seq;
   entity_addrvec_t addrs;
+  feature_bitset_t mds_features;
 
   enum {
     STATE_NEW, // Unused

diff --git a/src/mds/CInode.cc b/src/mds/CInode.cc
index e17241b23e2fd662f9995b72d7d51b325b641ff1..ad87cec5a369a605950899690c1fb964f58ddbe2 100644 (file)
--- a/src/mds/CInode.cc
+++ b/src/mds/CInode.cc
@@ -47,6 +47,7 @@
 
 #include "mds/MDSContinuation.h"
 #include "mds/InoTable.h"
+#include "cephfs_features.h"
 
 #define dout_context g_ceph_context
 #define dout_subsys ceph_subsys_mds
@@ -3317,7 +3318,6 @@ int CInode::encode_inodestat(bufferlist& bl, Session *session,
 {
   client_t client = session->get_client();
   assert(snapid);
-  assert(session->connection);
   
   bool valid = true;
 
@@ -3577,62 +3577,103 @@ int CInode::encode_inodestat(bufferlist& bl, Session *session,
   /*
    * note: encoding matches MClientReply::InodeStat
    */
-  encode(oi->ino, bl);
-  encode(snapid, bl);
-  encode(oi->rdev, bl);
-  encode(version, bl);
-
-  encode(xattr_version, bl);
-
-  encode(ecap, bl);
-  {
-    ceph_file_layout legacy_layout;
-    layout.to_legacy(&legacy_layout);
-    encode(legacy_layout, bl);
-  }
-  encode(any_i->ctime, bl);
-  encode(file_i->mtime, bl);
-  encode(file_i->atime, bl);
-  encode(file_i->time_warp_seq, bl);
-  encode(file_i->size, bl);
-  encode(max_size, bl);
-  encode(file_i->truncate_size, bl);
-  encode(file_i->truncate_seq, bl);
-
-  encode(auth_i->mode, bl);
-  encode((uint32_t)auth_i->uid, bl);
-  encode((uint32_t)auth_i->gid, bl);
-
-  encode(link_i->nlink, bl);
-
-  encode(file_i->dirstat.nfiles, bl);
-  encode(file_i->dirstat.nsubdirs, bl);
-  encode(file_i->rstat.rbytes, bl);
-  encode(file_i->rstat.rfiles, bl);
-  encode(file_i->rstat.rsubdirs, bl);
-  encode(file_i->rstat.rctime, bl);
-
-  dirfragtree.encode(bl);
-
-  encode(symlink, bl);
-  if (session->connection->has_feature(CEPH_FEATURE_DIRLAYOUTHASH)) {
+  if (session->info.has_feature(CEPHFS_FEATURE_REPLY_ENCODING)) {
+    ENCODE_START(1, 1, bl);
+    encode(oi->ino, bl);
+    encode(snapid, bl);
+    encode(oi->rdev, bl);
+    encode(version, bl);
+    encode(xattr_version, bl);
+    encode(ecap, bl);
+    {
+      ceph_file_layout legacy_layout;
+      layout.to_legacy(&legacy_layout);
+      encode(legacy_layout, bl);
+    }
+    encode(any_i->ctime, bl);
+    encode(file_i->mtime, bl);
+    encode(file_i->atime, bl);
+    encode(file_i->time_warp_seq, bl);
+    encode(file_i->size, bl);
+    encode(max_size, bl);
+    encode(file_i->truncate_size, bl);
+    encode(file_i->truncate_seq, bl);
+    encode(auth_i->mode, bl);
+    encode((uint32_t)auth_i->uid, bl);
+    encode((uint32_t)auth_i->gid, bl);
+    encode(link_i->nlink, bl);
+    encode(file_i->dirstat.nfiles, bl);
+    encode(file_i->dirstat.nsubdirs, bl);
+    encode(file_i->rstat.rbytes, bl);
+    encode(file_i->rstat.rfiles, bl);
+    encode(file_i->rstat.rsubdirs, bl);
+    encode(file_i->rstat.rctime, bl);
+    dirfragtree.encode(bl);
+    encode(symlink, bl);
     encode(file_i->dir_layout, bl);
-  }
-  encode(xbl, bl);
-  if (session->connection->has_feature(CEPH_FEATURE_MDS_INLINE_DATA)) {
+    encode(xbl, bl);
     encode(inline_version, bl);
     encode(inline_data, bl);
-  }
-  if (session->connection->has_feature(CEPH_FEATURE_MDS_QUOTA)) {
     mempool_inode *policy_i = ppolicy ? pi : oi;
     encode(policy_i->quota, bl);
-  }
-  if (session->connection->has_feature(CEPH_FEATURE_FS_FILE_LAYOUT_V2)) {
     encode(layout.pool_ns, bl);
-  }
-  if (session->connection->has_feature(CEPH_FEATURE_FS_BTIME)) {
     encode(any_i->btime, bl);
     encode(any_i->change_attr, bl);
+    ENCODE_FINISH(bl);
+  }
+  else {
+    assert(session->connection);
+
+    encode(oi->ino, bl);
+    encode(snapid, bl);
+    encode(oi->rdev, bl);
+    encode(version, bl);
+    encode(xattr_version, bl);
+    encode(ecap, bl);
+    {
+      ceph_file_layout legacy_layout;
+      layout.to_legacy(&legacy_layout);
+      encode(legacy_layout, bl);
+    }
+    encode(any_i->ctime, bl);
+    encode(file_i->mtime, bl);
+    encode(file_i->atime, bl);
+    encode(file_i->time_warp_seq, bl);
+    encode(file_i->size, bl);
+    encode(max_size, bl);
+    encode(file_i->truncate_size, bl);
+    encode(file_i->truncate_seq, bl);
+    encode(auth_i->mode, bl);
+    encode((uint32_t)auth_i->uid, bl);
+    encode((uint32_t)auth_i->gid, bl);
+    encode(link_i->nlink, bl);
+    encode(file_i->dirstat.nfiles, bl);
+    encode(file_i->dirstat.nsubdirs, bl);
+    encode(file_i->rstat.rbytes, bl);
+    encode(file_i->rstat.rfiles, bl);
+    encode(file_i->rstat.rsubdirs, bl);
+    encode(file_i->rstat.rctime, bl);
+    dirfragtree.encode(bl);
+    encode(symlink, bl);
+    if (session->connection->has_feature(CEPH_FEATURE_DIRLAYOUTHASH)) {
+      encode(file_i->dir_layout, bl);
+    }
+    encode(xbl, bl);
+    if (session->connection->has_feature(CEPH_FEATURE_MDS_INLINE_DATA)) {
+      encode(inline_version, bl);
+      encode(inline_data, bl);
+    }
+    if (session->connection->has_feature(CEPH_FEATURE_MDS_QUOTA)) {
+      mempool_inode *policy_i = ppolicy ? pi : oi;
+      encode(policy_i->quota, bl);
+    }
+    if (session->connection->has_feature(CEPH_FEATURE_FS_FILE_LAYOUT_V2)) {
+      encode(layout.pool_ns, bl);
+    }
+    if (session->connection->has_feature(CEPH_FEATURE_FS_BTIME)) {
+      encode(any_i->btime, bl);
+      encode(any_i->change_attr, bl);
+    }
   }
 
   return valid;

diff --git a/src/messages/MClientReply.h b/src/messages/MClientReply.h
index 558e43e54482c539593ece15aee51b9ad426bb16..8706aec17203918d2a7f7b0cdaa34a5ee1b04702 100644 (file)
--- a/src/messages/MClientReply.h
+++ b/src/messages/MClientReply.h
@@ -130,73 +130,116 @@ struct InodeStat {
 
  public:
   InodeStat() {}
-  InodeStat(bufferlist::const_iterator& p, uint64_t features) {
+  InodeStat(bufferlist::const_iterator& p, const uint64_t features) {
     decode(p, features);
   }
 
-  void decode(bufferlist::const_iterator &p, uint64_t features) {
+  void decode(bufferlist::const_iterator &p, const uint64_t features) {
     using ceph::decode;
-    decode(vino.ino, p);
-    decode(vino.snapid, p);
-    decode(rdev, p);
-    decode(version, p);
-    decode(xattr_version, p);
-    decode(cap, p);
-    {
-      ceph_file_layout legacy_layout;
-      decode(legacy_layout, p);
-      layout.from_legacy(legacy_layout);
-    }
-    decode(ctime, p);
-    decode(mtime, p);
-    decode(atime, p);
-    decode(time_warp_seq, p);
-    decode(size, p);
-    decode(max_size, p);
-    decode(truncate_size, p);
-    decode(truncate_seq, p);
-    decode(mode, p);
-    decode(uid, p);
-    decode(gid, p);
-    decode(nlink, p);
-    decode(dirstat.nfiles, p);
-    decode(dirstat.nsubdirs, p);
-    decode(rstat.rbytes, p);
-    decode(rstat.rfiles, p);
-    decode(rstat.rsubdirs, p);
-    decode(rstat.rctime, p);
-
-    decode(dirfragtree, p);
-
-    decode(symlink, p);
-    
-    if (features & CEPH_FEATURE_DIRLAYOUTHASH)
+    if (features == (uint64_t)-1) {
+      DECODE_START(1, p);
+      decode(vino.ino, p);
+      decode(vino.snapid, p);
+      decode(rdev, p);
+      decode(version, p);
+      decode(xattr_version, p);
+      decode(cap, p);
+      {
+        ceph_file_layout legacy_layout;
+        decode(legacy_layout, p);
+        layout.from_legacy(legacy_layout);
+      }
+      decode(ctime, p);
+      decode(mtime, p);
+      decode(atime, p);
+      decode(time_warp_seq, p);
+      decode(size, p);
+      decode(max_size, p);
+      decode(truncate_size, p);
+      decode(truncate_seq, p);
+      decode(mode, p);
+      decode(uid, p);
+      decode(gid, p);
+      decode(nlink, p);
+      decode(dirstat.nfiles, p);
+      decode(dirstat.nsubdirs, p);
+      decode(rstat.rbytes, p);
+      decode(rstat.rfiles, p);
+      decode(rstat.rsubdirs, p);
+      decode(rstat.rctime, p);
+      decode(dirfragtree, p);
+      decode(symlink, p);
       decode(dir_layout, p);
-    else
-      memset(&dir_layout, 0, sizeof(dir_layout));
-
-    decode(xattrbl, p);
-
-    if (features & CEPH_FEATURE_MDS_INLINE_DATA) {
+      decode(xattrbl, p);
       decode(inline_version, p);
       decode(inline_data, p);
-    } else {
-      inline_version = CEPH_INLINE_NONE;
-    }
-
-    if (features & CEPH_FEATURE_MDS_QUOTA)
       decode(quota, p);
-    else
-      quota = quota_info_t{};
-
-    if ((features & CEPH_FEATURE_FS_FILE_LAYOUT_V2))
       decode(layout.pool_ns, p);
-    if ((features & CEPH_FEATURE_FS_BTIME)) {
       decode(btime, p);
       decode(change_attr, p);
-    } else {
-      btime = utime_t();
-      change_attr = 0;
+      DECODE_FINISH(p);
+    }
+    else {
+      decode(vino.ino, p);
+      decode(vino.snapid, p);
+      decode(rdev, p);
+      decode(version, p);
+      decode(xattr_version, p);
+      decode(cap, p);
+      {
+        ceph_file_layout legacy_layout;
+        decode(legacy_layout, p);
+        layout.from_legacy(legacy_layout);
+      }
+      decode(ctime, p);
+      decode(mtime, p);
+      decode(atime, p);
+      decode(time_warp_seq, p);
+      decode(size, p);
+      decode(max_size, p);
+      decode(truncate_size, p);
+      decode(truncate_seq, p);
+      decode(mode, p);
+      decode(uid, p);
+      decode(gid, p);
+      decode(nlink, p);
+      decode(dirstat.nfiles, p);
+      decode(dirstat.nsubdirs, p);
+      decode(rstat.rbytes, p);
+      decode(rstat.rfiles, p);
+      decode(rstat.rsubdirs, p);
+      decode(rstat.rctime, p);
+      decode(dirfragtree, p);
+      decode(symlink, p);
+      if (features & CEPH_FEATURE_DIRLAYOUTHASH)
+        decode(dir_layout, p);
+      else
+        memset(&dir_layout, 0, sizeof(dir_layout));
+
+      decode(xattrbl, p);
+
+      if (features & CEPH_FEATURE_MDS_INLINE_DATA) {
+        decode(inline_version, p);
+        decode(inline_data, p);
+      } else {
+        inline_version = CEPH_INLINE_NONE;
+      }
+
+      if (features & CEPH_FEATURE_MDS_QUOTA)
+        decode(quota, p);
+      else
+        quota = quota_info_t{};
+
+      if ((features & CEPH_FEATURE_FS_FILE_LAYOUT_V2))
+        decode(layout.pool_ns, p);
+
+      if ((features & CEPH_FEATURE_FS_BTIME)) {
+        decode(btime, p);
+        decode(change_attr, p);
+      } else {
+        btime = utime_t();
+        change_attr = 0;
+      }
     }
   }