osd: Add automatic repair for DBObjectMap bug

Add repair command to ceph-osdomap-tool too

Under some situations the previous rm_keys() code would
generated a corrupt complete table.  There is no way to
figure out what the table should look like now.  By removing
the entries we fix the corruption and aren't much worse off
because the corruption caused some deleted keys to re-appear.

This doesn't breaking the parent/child relationship during
repair because some of the keys may still be contained
in the parent.

Signed-off-by: David Zafman <dzafman@redhat.com>

src/os/ObjectMap.h

@@ -138,7 +138,7 @@ class ObjectMap {
     const SequencerPosition *spos=0   ///< [in] Sequencer
     ) { return 0; }
 
-  virtual int check(std::ostream &out) { return 0; }
+  virtual int check(std::ostream &out, bool repair = false) { return 0; }
 
   typedef KeyValueDB::GenericIteratorImpl ObjectMapIteratorImpl;
   typedef ceph::shared_ptr<ObjectMapIteratorImpl> ObjectMapIterator;

src/os/filestore/DBObjectMap.cc

@@ -55,7 +55,7 @@ static void append_escaped(const string &in, string *out)
   }
 }
 
-int DBObjectMap::check(std::ostream &out)
+int DBObjectMap::check(std::ostream &out, bool repair)
 {
   int errors = 0;
   map<uint64_t, uint64_t> parent_to_num_children;
@@ -78,18 +78,25 @@ int DBObjectMap::check(std::ostream &out)
            complete_iter->next()) {
          if (prev && prev >= complete_iter->key()) {
              out << "Bad complete for " << header.oid << std::endl;
-             errors++;
              complete_error = true;
              break;
          }
          prev = string(complete_iter->value().c_str(), complete_iter->value().length() - 1);
       }
       if (complete_error) {
-        out << "Complete mapping:" << std::endl;
+        out << "Complete mapping for " << header.seq << " :" << std::endl;
         for (complete_iter->seek_to_first(); complete_iter->valid();
              complete_iter->next()) {
           out << complete_iter->key() << " -> " << string(complete_iter->value().c_str(), complete_iter->value().length() - 1) << std::endl;
         }
+        if (repair) {
+          KeyValueDB::Transaction t = db->get_transaction();
+          t->rmkeys_by_prefix(USER_PREFIX + header_key(header.seq) + COMPLETE_PREFIX);
+          db->submit_transaction(t);
+          out << "Cleared complete mapping to repair" << std::endl;
+        } else {
+          errors++;  // Only count when not repaired
+        }
       }
 
       if (header.parent == 0)
@@ -1051,6 +1058,12 @@ int DBObjectMap::init(bool do_upgrade)
     state.v = 2;
     state.seq = 1;
   }
+  ostringstream ss;
+  int errors = check(ss, true);
+  if (errors) {
+    dout(5) << ss.str() << dendl;
+    return -EINVAL;
+  }
   dout(20) << "(init)dbobjectmap: seq is " << state.seq << dendl;
   return 0;
 }

src/os/filestore/DBObjectMap.h

@@ -213,7 +213,7 @@ class DBObjectMap : public ObjectMap {
   int upgrade_to_v2();
 
   /// Consistency check, debug, there must be no parallel writes
-  int check(std::ostream &out) override;
+  int check(std::ostream &out, bool repair = false) override;
 
   /// Ensure that all previous operations are durable
   int sync(const ghobject_t *oid=0, const SequencerPosition *spos=0) override;

src/tools/ceph_osdomap_tool.cc

@@ -37,7 +37,7 @@ int main(int argc, char **argv) {
     ("debug", "Additional debug output from DBObjectMap")
     ("oid", po::value<string>(&oid), "Restrict to this object id when dumping objects")
     ("command", po::value<string>(&cmd),
-     "command arg is one of [dump-raw-keys, dump-raw-key-vals, dump-objects, dump-objects-with-keys, check, dump-headers], mandatory")
+     "command arg is one of [dump-raw-keys, dump-raw-key-vals, dump-objects, dump-objects-with-keys, check, dump-headers, repair], mandatory")
     ;
   po::positional_options_description p;
   p.add("command", 1);
@@ -109,6 +109,9 @@ int main(int argc, char **argv) {
     std::cerr << "Output: " << out.str() << std::endl;
     goto done;
   }
+  // We don't call omap.init() here because it will repair
+  // the DBObjectMap which we might want to examine for diagnostic
+  // reasons.  Instead use --command repair.
   r = 0;
 
 
@@ -157,14 +160,15 @@ int main(int argc, char **argv) {
 	j->value().hexdump(std::cout);
       }
     }
-  } else if (cmd == "check") {
-    r = omap.check(std::cout);
+  } else if (cmd == "check" || cmd == "repair") {
+    bool repair = (cmd == "repair");
+    r = omap.check(std::cout, repair);
     if (r > 0) {
       std::cerr << "check got " << r << " error(s)" << std::endl;
       r = 1;
       goto done;
     }
-    std::cout << "check succeeded" << std::endl;
+    std::cout << (repair ? "repair" : "check") << " succeeded" << std::endl;
   } else if (cmd == "dump-headers") {
     vector<DBObjectMap::_Header> headers;
     r = omap.list_object_headers(&headers);