ceph/src/mds/MDSTable.cc

   1 // -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
   2 // vim: ts=8 sw=2 smarttab
   3 /*
   4  * Ceph - scalable distributed file system
   5  *
   6  * Copyright (C) 2004-2006 Sage Weil <sage@newdream.net>
   7  *
   8  * This is free software; you can redistribute it and/or
   9  * modify it under the terms of the GNU Lesser General Public
  10  * License version 2.1, as published by the Free Software
  11  * Foundation.  See file COPYING.
  12  *
  13  */
  14
  15 #include "MDSTable.h"
  16
  17 #include "MDSRank.h"
  18 #include "MDLog.h"
  19
  20 #include "osdc/Filer.h"
  21
  22 #include "include/types.h"
  23
  24 #include "common/config.h"
  25 #include "common/errno.h"
  26 #include "common/Finisher.h"
  27
  28 #include "include/ceph_assert.h"
  29
  30
  31 #define dout_context g_ceph_context
  32 #define dout_subsys ceph_subsys_mds
  33 #undef dout_prefix
  34 #define dout_prefix *_dout << "mds." << rank << "." << table_name << ": "
  35
  36
  37 class MDSTableIOContext : public MDSIOContextBase
  38 {
  39   protected:
  40     MDSTable *ida;
  41     MDSRank *get_mds() override {return ida->mds;}
  42   public:
  43     explicit MDSTableIOContext(MDSTable *ida_) : ida(ida_) {
  44       ceph_assert(ida != NULL);
  45     }
  46 };
  47
  48
  49 class C_IO_MT_Save : public MDSTableIOContext {
  50   version_t version;
  51 public:
  52   C_IO_MT_Save(MDSTable *i, version_t v) : MDSTableIOContext(i), version(v) {}
  53   void finish(int r) override {
  54     ida->save_2(r, version);
  55   }
  56   void print(ostream& out) const override {
  57     out << "table_save(" << ida->table_name << ")";
  58   }
  59 };
  60
  61 void MDSTable::save(MDSContext *onfinish, version_t v)
  62 {
  63   if (v > 0 && v <= committing_version) {
  64     dout(10) << "save v " << version << " - already saving "
  65              << committing_version << " >= needed " << v << dendl;
  66     if (onfinish)
  67       waitfor_save[v].push_back(onfinish);
  68     return;
  69   }
  70
  71   dout(10) << "save v " << version << dendl;
  72   ceph_assert(is_active());
  73
  74   bufferlist bl;
  75   encode(version, bl);
  76   encode_state(bl);
  77
  78   committing_version = version;
  79
  80   if (onfinish)
  81     waitfor_save[version].push_back(onfinish);
  82
  83   // write (async)
  84   SnapContext snapc;
  85   object_t oid = get_object_name();
  86   object_locator_t oloc(mds->get_metadata_pool());
  87   mds->objecter->write_full(oid, oloc,
  88                             snapc,
  89                             bl, ceph::real_clock::now(), 0,
  90                             new C_OnFinisher(new C_IO_MT_Save(this, version),
  91                                              mds->finisher));
  92 }
  93
  94 void MDSTable::save_2(int r, version_t v)
  95 {
  96   if (r < 0) {
  97     dout(1) << "save error " << r << " v " << v << dendl;
  98     mds->clog->error() << "failed to store table " << table_name << " object,"
  99                        << " errno " << r;
 100     mds->handle_write_error(r);
 101     return;
 102   }
 103
 104   dout(10) << "save_2 v " << v << dendl;
 105   committed_version = v;
 106
 107   MDSContext::vec ls;
 108   while (!waitfor_save.empty()) {
 109     auto it = waitfor_save.begin();
 110     if (it->first > v) break;
 111     auto& v = it->second;
 112     ls.insert(ls.end(), v.begin(), v.end());
 113     waitfor_save.erase(it);
 114   }
 115   finish_contexts(g_ceph_context, ls, 0);
 116 }
 117
 118
 119 void MDSTable::reset()
 120 {
 121   reset_state();
 122   projected_version = version;
 123   state = STATE_ACTIVE;
 124 }
 125
 126
 127
 128 // -----------------------
 129
 130 class C_IO_MT_Load : public MDSTableIOContext {
 131 public:
 132   Context *onfinish;
 133   bufferlist bl;
 134   C_IO_MT_Load(MDSTable *i, Context *o) : MDSTableIOContext(i), onfinish(o) {}
 135   void finish(int r) override {
 136     ida->load_2(r, bl, onfinish);
 137   }
 138   void print(ostream& out) const override {
 139     out << "table_load(" << ida->table_name << ")";
 140   }
 141 };
 142
 143 object_t MDSTable::get_object_name() const
 144 {
 145   char n[50];
 146   if (per_mds)
 147     snprintf(n, sizeof(n), "mds%d_%s", int(rank), table_name.c_str());
 148   else
 149     snprintf(n, sizeof(n), "mds_%s", table_name.c_str());
 150   return object_t(n);
 151 }
 152
 153 void MDSTable::load(MDSContext *onfinish)
 154 {
 155   dout(10) << "load" << dendl;
 156
 157   ceph_assert(is_undef());
 158   state = STATE_OPENING;
 159
 160   C_IO_MT_Load *c = new C_IO_MT_Load(this, onfinish);
 161   object_t oid = get_object_name();
 162   object_locator_t oloc(mds->get_metadata_pool());
 163   mds->objecter->read_full(oid, oloc, CEPH_NOSNAP, &c->bl, 0,
 164                            new C_OnFinisher(c, mds->finisher));
 165 }
 166
 167 void MDSTable::load_2(int r, bufferlist& bl, Context *onfinish)
 168 {
 169   ceph_assert(is_opening());
 170   state = STATE_ACTIVE;
 171   if (r == -CEPHFS_EBLOCKLISTED) {
 172     mds->respawn();
 173     return;
 174   }
 175   if (r < 0) {
 176     derr << "load_2 could not read table: " << r << dendl;
 177     mds->clog->error() << "error reading table object '" << get_object_name()
 178                        << "' " << r << " (" << cpp_strerror(r) << ")";
 179     mds->damaged();
 180     ceph_assert(r >= 0);  // Should be unreachable because damaged() calls respawn()
 181   }
 182
 183   dout(10) << "load_2 got " << bl.length() << " bytes" << dendl;
 184   auto p = bl.cbegin();
 185
 186   try {
 187     decode(version, p);
 188     projected_version = committed_version = version;
 189     dout(10) << "load_2 loaded v" << version << dendl;
 190     decode_state(p);
 191   } catch (buffer::error &e) {
 192     mds->clog->error() << "error decoding table object '" << get_object_name()
 193                        << "': " << e.what();
 194     mds->damaged();
 195     ceph_assert(r >= 0);  // Should be unreachable because damaged() calls respawn()
 196   }
 197
 198   if (onfinish) {
 199     onfinish->complete(0);
 200   }
 201 }