tools: Create cephfs-journal-tool

This is for debugging/repairing CephFS journals.

Signed-off-by: John Spray <[email protected]>

src/ceph_mds.cc

@@ -27,8 +27,6 @@ using namespace std;
 
 #include "mon/MonMap.h"
 #include "mds/MDS.h"
-#include "mds/Dumper.h"
-#include "mds/Resetter.h"
 
 #include "msg/Messenger.h"
 
@@ -55,65 +53,14 @@ void usage()
        << "        connect to monitor at given address\n"
        << "  --debug_mds n\n"
        << "        debug MDS level (e.g. 10)\n"
-       << "  --dump-journal rank filename\n"
-       << "        dump the MDS journal (binary) for rank.\n"
-       << "  --dump-journal-entries rank filename\n"
-       << "        dump the MDS journal (JSON) for rank.\n"
        << "  --journal-check rank\n"
        << "        replay the journal for rank, then exit\n"
        << "  --hot-standby rank\n"
        << "        start up as a hot standby for rank\n"
-       << "  --reset-journal rank\n"
-       << "        discard the MDS journal for rank, and replace it with a single\n"
-       << "        event that updates/resets inotable and sessionmap on replay.\n"
        << dendl;
   generic_server_usage();
 }
 
-static int do_cmds_special_action(const std::string &action,
-				  const std::string &dump_file, int rank)
-{
-  common_init_finish(g_ceph_context, CINIT_FLAG_NO_DAEMON_ACTIONS);
-
-  if (action == "dump-journal") {
-    dout(0) << "dumping journal for mds." << rank << " to " << dump_file << dendl;
-    Dumper journal_dumper;
-    journal_dumper.init(rank);
-    journal_dumper.dump(dump_file.c_str());
-    journal_dumper.shutdown();
-  } else if (action == "dump-journal-entries") {
-    Dumper journal_dumper;
-    journal_dumper.init(rank);
-    journal_dumper.dump_entries();
-    journal_dumper.shutdown();
-  } else if (action == "undump-journal") {
-    dout(0) << "undumping journal for mds." << rank << " from " << dump_file << dendl;
-    Dumper journal_dumper;
-    journal_dumper.init(rank);
-    journal_dumper.undump(dump_file.c_str());
-    journal_dumper.shutdown();
-  } else if (action == "reset-journal") {
-    dout(0) << "resetting journal" << dendl;
-    Resetter resetter;
-    resetter.init(rank);
-    resetter.reset();
-    resetter.shutdown();
-  } else {
-    assert(0);
-  }
-  return 0;
-}
-
-static void set_special_action(std::string &dest, const std::string &act)
-{
-  if (!dest.empty()) {
-    derr << "Parse error! Can't specify more than one action. You "
-	 << "specified both " << act << " and " << dest << "\n" << dendl;
-    usage();
-    exit(1);
-  }
-  dest = act;
-}
 
 static int parse_rank(const char *opt_name, const std::string &val)
 {
@@ -148,44 +95,13 @@ int main(int argc, const char **argv)
 
   // mds specific args
   int shadow = 0;
-  int rank = -1;
   std::string dump_file;
 
   std::string val, action;
   for (std::vector<const char*>::iterator i = args.begin(); i != args.end(); ) {
     if (ceph_argparse_double_dash(args, i)) {
       break;
     }
-    else if (ceph_argparse_witharg(args, i, &val, "--dump-journal", (char*)NULL)) {
-      set_special_action(action, "dump-journal");
-      rank = parse_rank("dump-journal", val);
-      if (i == args.end()) {
-	derr << "error parsing --dump-journal: you must give a second "
-	     << "dump-journal argument: the filename to dump the journal to. "
-	     << "\n" << dendl;
-	usage();
-      }
-      dump_file = *i++;
-    }
-    else if (ceph_argparse_witharg(args, i, &val, "--undump-journal", (char*)NULL)) {
-      set_special_action(action, "undump-journal");
-      rank = parse_rank("undump-journal", val);
-      if (i == args.end()) {
-	derr << "error parsing --undump-journal: you must give a second "
-	     << "undump-journal argument: the filename to undump the journal from. "
-	     << "\n" << dendl;
-	usage();
-      }
-      dump_file = *i++;
-    }
-    else if (ceph_argparse_witharg(args, i, &val, "--dump-journal-entries", (char*)NULL)){
-      set_special_action(action, "dump-journal-entries");
-      rank = parse_rank("dump-journal-entries", val);
-    }
-    else if (ceph_argparse_witharg(args, i, &val, "--reset-journal", (char*)NULL)) {
-      set_special_action(action, "reset-journal");
-      rank = parse_rank("reset-journal", val);
-    }
     else if (ceph_argparse_witharg(args, i, &val, "--journal-check", (char*)NULL)) {
       int r = parse_rank("journal-check", val);
       if (shadow) {
@@ -220,11 +136,6 @@ int main(int argc, const char **argv)
 
   pick_addresses(g_ceph_context, CEPH_PICK_ADDRESS_PUBLIC);
 
-  // Check for special actions
-  if (!action.empty()) {
-    return do_cmds_special_action(action, dump_file, rank);
-  }
-
   // Normal startup
   if (g_conf->name.has_default_id()) {
     derr << "must specify '-i name' with the ceph-mds instance name" << dendl;

src/mds/Makefile.am

@@ -1,7 +1,5 @@
 libmds_la_SOURCES = \
 	mds/Capability.cc \
-	mds/Dumper.cc \
-	mds/Resetter.cc \
 	mds/MDS.cc \
 	mds/flock.cc \
 	mds/locks.c \
@@ -25,7 +23,6 @@ libmds_la_SOURCES = \
 	mds/snap.cc \
 	mds/SessionMap.cc \
 	mds/MDLog.cc \
-	mds/MDSUtility.cc \
 	common/TrackedOp.cc
 libmds_la_LIBADD = $(LIBOSDC)
 noinst_LTLIBRARIES += libmds.la
@@ -39,7 +36,6 @@ noinst_HEADERS += \
 	mds/CDir.h \
 	mds/CInode.h \
 	mds/Capability.h \
-	mds/Dumper.h \
 	mds/InoTable.h \
 	mds/LocalLock.h \
 	mds/Locker.h \
@@ -53,10 +49,8 @@ noinst_HEADERS += \
 	mds/MDSTable.h \
 	mds/MDSTableServer.h \
 	mds/MDSTableClient.h \
-	mds/MDSUtility.h \
 	mds/Mutation.h \
 	mds/Migrator.h \
-	mds/Resetter.h \
 	mds/ScatterLock.h \
 	mds/Server.h \
 	mds/SessionMap.h \

src/mds/events/ENoOp.h

@@ -18,11 +18,11 @@
 #include "../LogEvent.h"
 
 class ENoOp : public LogEvent {
-  uint32_t size;
+  uint32_t pad_size;
 
 public:
-  ENoOp() : LogEvent(EVENT_NOOP), size(0) { }
-  ENoOp(uint32_t size_) : LogEvent(EVENT_NOOP), size(size){ }
+  ENoOp() : LogEvent(EVENT_NOOP), pad_size(0) { }
+  ENoOp(uint32_t size_) : LogEvent(EVENT_NOOP), pad_size(size_){ }
 
   void encode(bufferlist& bl) const;
   void decode(bufferlist::iterator& bl);

src/mds/journal.cc

@@ -857,9 +857,8 @@ void EMetaBlob::get_inodes(
 void EMetaBlob::get_dentries(std::map<dirfrag_t, std::set<std::string> > &dentries)
 {
   for (std::map<dirfrag_t, dirlump>::iterator i = lump_map.begin(); i != lump_map.end(); ++i) {
-    inodeno_t const dir_ino = i->first.ino;
     dirlump &dl = i->second;
-    dirfrag_t &df = i->first;
+    dirfrag_t const &df = i->first;
 
     // Get all bits
     dl._decode_bits();
@@ -2996,27 +2995,32 @@ void EResetJournal::replay(MDS *mds)
 
 void ENoOp::encode(bufferlist &bl) const
 {
-  ::encode(size, bl);
-  uint32_t const pad_bytes = size - sizeof(size);
-  uint32_t const pad = 0xdeadbeef;
-  for (unsigned int i = 0; i < pad_bytes / sizeof(uint32_t); ++i) {
+  ENCODE_START(2, 2, bl);
+  ::encode(pad_size, bl);
+  uint8_t const pad = 0xff;
+  for (unsigned int i = 0; i < pad_size; ++i) {
     ::encode(pad, bl);
   }
+  ENCODE_FINISH(bl);
 }
 
 
 void ENoOp::decode(bufferlist::iterator &bl)
 {
-  ::decode(size, bl);
-  if (bl.get_remaining() != (size - sizeof(size))) {
+  DECODE_START(2, bl);
+  ::decode(pad_size, bl);
+  if (bl.get_remaining() != pad_size) {
     // This is spiritually an assertion, but expressing in a way that will let
     // journal debug tools catch it and recognise a malformed entry.
     throw buffer::end_of_buffer();
+  } else {
+    bl.advance(pad_size);
   }
+  DECODE_FINISH(bl);
 }
 
 
 void ENoOp::replay(MDS *mds)
 {
-  dout(4) << "ENoOp::replay, " << size << " bytes skipped in journal" << dendl;
+  dout(4) << "ENoOp::replay, " << pad_size << " bytes skipped in journal" << dendl;
 }

src/tools/Makefile.am

@@ -71,6 +71,18 @@ rados_SOURCES += common/obj_bencher.cc # needs cleanup so it can go in libcommon
 rados_LDADD = libcls_lock_client.la $(LIBRADOS) $(CEPH_GLOBAL)
 bin_PROGRAMS += rados
 
+cephfs_journal_tool_SOURCES = \
+	tools/cephfs/cephfs-journal-tool.cc \
+	tools/cephfs/JournalTool.cc \
+	tools/cephfs/JournalFilter.cc \
+	tools/cephfs/JournalScanner.cc \
+	tools/cephfs/EventOutput.cc \
+	tools/cephfs/Dumper.cc \
+	tools/cephfs/Resetter.cc \
+	tools/cephfs/MDSUtility.cc
+cephfs_journal_tool_LDADD = $(LIBMDS) $(LIBRADOS) $(CEPH_GLOBAL)
+bin_PROGRAMS += cephfs-journal-tool
+
 if WITH_REST_BENCH
 rest_bench_SOURCES = tools/rest_bench.cc
 rest_bench_SOURCES += common/obj_bencher.cc # needs cleanup so it can go in libcommon.la
@@ -99,6 +111,13 @@ ceph_mon_store_converter_LDADD = $(LIBMON) $(LIBOS) $(CEPH_GLOBAL)
 bin_PROGRAMS += ceph_mon_store_converter
 
 noinst_HEADERS += \
+	tools/cephfs/JournalTool.h \
+	tools/cephfs/JournalScanner.h \
+	tools/cephfs/JournalFilter.h \
+	tools/cephfs/EventOutput.h \
+	tools/cephfs/Resetter.h \
+	tools/cephfs/Dumper.h \
+	tools/cephfs/MDSUtility.h \
 	tools/rados/rados_sync.h \
 	tools/common.h
 

src/mds/Dumper.cc → src/tools/cephfs/Dumper.cc

@@ -20,11 +20,12 @@
 #include "common/entity_name.h"
 #include "common/errno.h"
 #include "common/safe_io.h"
-#include "mds/Dumper.h"
 #include "mds/mdstypes.h"
 #include "mds/LogEvent.h"
 #include "osdc/Journaler.h"
 
+#include "Dumper.h"
+
 #define dout_subsys ceph_subsys_mds
 
 
@@ -219,60 +220,3 @@ void Dumper::undump(const char *dump_file)
   cout << "done." << std::endl;
 }
 
-
-/**
- * Write JSON-formatted log entries to standard out.
- */
-void Dumper::dump_entries()
-{
-  Mutex localLock("dump_entries");
-  JSONFormatter jf(true);
-
-  if (recover_journal()) {
-    return;
-  }
-
-  jf.open_array_section("log");
-  lock.Lock();
-  // Until the journal is empty, pop an event or wait for one to
-  // be available.
-  dout(10) << "Journaler read/write/size: "
-      << journaler->get_read_pos() << "/" << journaler->get_write_pos()
-      << "/" << journaler->get_write_pos() - journaler->get_read_pos() << dendl;
-  while (journaler->get_read_pos() != journaler->get_write_pos()) {
-    bufferlist entry_bl;
-    bool got_data = journaler->try_read_entry(entry_bl);
-    dout(10) << "try_read_entry: " << got_data << dendl;
-    if (got_data) {
-      LogEvent *le = LogEvent::decode(entry_bl);
-      if (!le) {
-	dout(0) << "Error decoding LogEvent" << dendl;
-	break;
-      } else {
-	jf.open_object_section("log_event");
-	jf.dump_unsigned("type", le->get_type());
-	jf.dump_unsigned("start_off", le->get_start_off());
-	jf.dump_unsigned("stamp_sec", le->get_stamp().tv.tv_sec);
-	jf.dump_unsigned("stamp_nsec", le->get_stamp().tv.tv_nsec);
-	le->dump(&jf);
-	jf.close_section();
-	delete le;
-      }
-    } else {
-      bool done = false;
-      Cond cond;
-
-      journaler->wait_for_readable(new C_SafeCond(&localLock, &cond, &done));
-      lock.Unlock();
-      localLock.Lock();
-      while (!done)
-        cond.Wait(localLock);
-      localLock.Unlock();
-      lock.Lock();
-    }
-  }
-  lock.Unlock();
-  jf.close_section();
-  jf.flush(cout);
-  return;
-}

src/mds/Dumper.h → src/tools/cephfs/Dumper.h

@@ -15,7 +15,7 @@
 #define JOURNAL_DUMPER_H_
 
 
-#include "mds/MDSUtility.h"
+#include "MDSUtility.h"
 #include "osdc/Journaler.h"
 
 /**
@@ -41,7 +41,6 @@ class Dumper : public MDSUtility {
   int recover_journal();
   void dump(const char *dumpfile);
   void undump(const char *dumpfile);
-  void dump_entries();
 };
 
 #endif /* JOURNAL_DUMPER_H_ */