1 // -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
2 // vim: ts=8 sw=2 smarttab
4 * Ceph - scalable distributed file system
6 * Copyright (C) 2004-2006 Sage Weil <sage@newdream.net>
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.
15 #include "SnapServer.h"
17 #include "osd/OSDMap.h"
18 #include "osdc/Objecter.h"
19 #include "mon/MonClient.h"
21 #include "include/types.h"
22 #include "messages/MMDSTableRequest.h"
23 #include "messages/MRemoveSnaps.h"
25 #include "msg/Messenger.h"
27 #include "common/config.h"
28 #include "include/assert.h"
30 #define dout_context g_ceph_context
31 #define dout_subsys ceph_subsys_mds
33 #define dout_prefix *_dout << "mds." << rank << ".snap "
36 void SnapServer::reset_state()
38 last_snap = 1; /* snapid 1 reserved for initial root snaprealm */
40 need_to_purge.clear();
42 // find any removed snapshot in data pools
43 if (mds) { // only if I'm running in a live MDS
44 snapid_t first_free = 0;
45 mds->objecter->with_osdmap([&](const OSDMap& o) {
46 for (const auto p : mds->mdsmap->get_data_pools()) {
47 const pg_pool_t *pi = o.get_pg_pool(p);
49 // If pool isn't in OSDMap yet then can't have any snaps
50 // needing removal, skip.
53 if (!pi->removed_snaps.empty() &&
54 pi->removed_snaps.range_end() > first_free)
55 first_free = pi->removed_snaps.range_end();
58 if (first_free > last_snap)
59 last_snap = first_free;
66 void SnapServer::_prepare(bufferlist &bl, uint64_t reqid, mds_rank_t bymds)
68 bufferlist::iterator p = bl.begin();
78 ::decode(info.ino, p);
80 ::decode(info.name, p);
81 ::decode(info.stamp, p);
82 info.snapid = ++last_snap;
83 info.long_name = "create";
84 pending_update[version] = info;
85 dout(10) << "prepare v" << version << " create " << info << dendl;
87 pending_noop.insert(version);
88 dout(10) << "prepare v" << version << " noop" << dendl;
91 ::encode(last_snap, bl);
95 case TABLE_OP_DESTROY:
99 ::decode(ino, p); // not used, currently.
103 // bump last_snap... we use it as a version value on the snaprealm.
106 pending_destroy[version] = pair<snapid_t,snapid_t>(snapid, last_snap);
107 dout(10) << "prepare v" << version << " destroy " << snapid << " seq " << last_snap << dendl;
110 ::encode(last_snap, bl);
114 case TABLE_OP_UPDATE:
117 ::decode(info.ino, p);
118 ::decode(info.snapid, p);
119 ::decode(info.name, p);
120 ::decode(info.stamp, p);
121 info.long_name = "update";
124 // bump last_snap... we use it as a version value on the snaprealm.
126 pending_update[version] = info;
127 dout(10) << "prepare v" << version << " update " << info << dendl;
130 ::encode(last_snap, bl);
140 bool SnapServer::_is_prepared(version_t tid) const
143 pending_update.count(tid) ||
144 pending_destroy.count(tid);
147 bool SnapServer::_commit(version_t tid, MMDSTableRequest *req)
149 if (pending_update.count(tid)) {
150 SnapInfo &info = pending_update[tid];
152 if (info.long_name.empty())
155 opname.swap(info.long_name);
156 if (info.stamp == utime_t() && snaps.count(info.snapid))
157 info.stamp = snaps[info.snapid].stamp;
158 dout(7) << "commit " << tid << " " << opname << " " << info << dendl;
159 snaps[info.snapid] = info;
160 pending_update.erase(tid);
163 else if (pending_destroy.count(tid)) {
164 snapid_t sn = pending_destroy[tid].first;
165 snapid_t seq = pending_destroy[tid].second;
166 dout(7) << "commit " << tid << " destroy " << sn << " seq " << seq << dendl;
169 for (const auto p : mds->mdsmap->get_data_pools()) {
170 need_to_purge[p].insert(sn);
171 need_to_purge[p].insert(seq);
174 pending_destroy.erase(tid);
176 else if (pending_noop.count(tid)) {
177 dout(7) << "commit " << tid << " noop" << dendl;
178 pending_noop.erase(tid);
189 void SnapServer::_rollback(version_t tid)
191 if (pending_update.count(tid)) {
192 SnapInfo &info = pending_update[tid];
194 if (info.long_name.empty())
197 opname.swap(info.long_name);
198 dout(7) << "rollback " << tid << " " << opname << " " << info << dendl;
199 pending_update.erase(tid);
202 else if (pending_destroy.count(tid)) {
203 dout(7) << "rollback " << tid << " destroy " << pending_destroy[tid] << dendl;
204 pending_destroy.erase(tid);
207 else if (pending_noop.count(tid)) {
208 dout(7) << "rollback " << tid << " noop" << dendl;
209 pending_noop.erase(tid);
220 void SnapServer::_server_update(bufferlist& bl)
222 bufferlist::iterator p = bl.begin();
223 map<int, vector<snapid_t> > purge;
226 dout(7) << "_server_update purged " << purge << dendl;
227 for (map<int, vector<snapid_t> >::iterator p = purge.begin();
230 for (vector<snapid_t>::iterator q = p->second.begin();
231 q != p->second.end();
233 need_to_purge[p->first].erase(*q);
234 if (need_to_purge[p->first].empty())
235 need_to_purge.erase(p->first);
241 void SnapServer::handle_query(MMDSTableRequest *req)
248 void SnapServer::check_osd_map(bool force)
250 if (!force && version == last_checked_osdmap) {
251 dout(10) << "check_osd_map - version unchanged" << dendl;
254 dout(10) << "check_osd_map need_to_purge=" << need_to_purge << dendl;
256 map<int, vector<snapid_t> > all_purge;
257 map<int, vector<snapid_t> > all_purged;
259 mds->objecter->with_osdmap(
260 [this, &all_purged, &all_purge](const OSDMap& osdmap) {
261 for (const auto& p : need_to_purge) {
263 const pg_pool_t *pi = osdmap.get_pg_pool(id);
265 // The pool is gone. So are the snapshots.
266 all_purged[id] = std::vector<snapid_t>(p.second.begin(),
271 for (const auto& q : p.second) {
272 if (pi->is_removed_snap(q)) {
273 dout(10) << " osdmap marks " << q << " as removed" << dendl;
274 all_purged[id].push_back(q);
276 all_purge[id].push_back(q);
282 if (!all_purged.empty()) {
283 // prepare to remove from need_to_purge list
285 ::encode(all_purged, bl);
286 do_server_update(bl);
289 if (!all_purge.empty()) {
290 dout(10) << "requesting removal of " << all_purge << dendl;
291 MRemoveSnaps *m = new MRemoveSnaps(all_purge);
292 mon_client->send_mon_message(m);
295 last_checked_osdmap = version;
299 void SnapServer::dump(Formatter *f) const
301 f->open_object_section("snapserver");
303 f->dump_int("last_snap", last_snap.val);
305 f->open_array_section("pending_noop");
306 for(set<version_t>::const_iterator i = pending_noop.begin(); i != pending_noop.end(); ++i) {
307 f->dump_unsigned("version", *i);
311 f->open_array_section("snaps");
312 for (map<snapid_t, SnapInfo>::const_iterator i = snaps.begin(); i != snaps.end(); ++i) {
313 f->open_object_section("snap");
319 f->open_object_section("need_to_purge");
320 for (map<int, set<snapid_t> >::const_iterator i = need_to_purge.begin(); i != need_to_purge.end(); ++i) {
321 stringstream pool_id;
323 f->open_array_section(pool_id.str().c_str());
324 for (set<snapid_t>::const_iterator s = i->second.begin(); s != i->second.end(); ++s) {
325 f->dump_unsigned("snapid", s->val);
331 f->open_array_section("pending_update");
332 for(map<version_t, SnapInfo>::const_iterator i = pending_update.begin(); i != pending_update.end(); ++i) {
333 f->open_object_section("snap");
334 f->dump_unsigned("version", i->first);
335 f->open_object_section("snapinfo");
342 f->open_array_section("pending_destroy");
343 for(map<version_t, pair<snapid_t, snapid_t> >::const_iterator i = pending_destroy.begin(); i != pending_destroy.end(); ++i) {
344 f->open_object_section("snap");
345 f->dump_unsigned("version", i->first);
346 f->dump_unsigned("removed_snap", i->second.first);
347 f->dump_unsigned("seq", i->second.second);
355 void SnapServer::generate_test_instances(list<SnapServer*>& ls)
357 list<SnapInfo*> snapinfo_instances;
358 SnapInfo::generate_test_instances(snapinfo_instances);
359 SnapInfo populated_snapinfo = *(snapinfo_instances.back());
360 for (list<SnapInfo*>::iterator i = snapinfo_instances.begin(); i != snapinfo_instances.end(); ++i) {
364 SnapServer *blank = new SnapServer();
366 SnapServer *populated = new SnapServer();
367 populated->last_snap = 123;
368 populated->snaps[456] = populated_snapinfo;
369 populated->need_to_purge[2].insert(012);
370 populated->pending_update[234] = populated_snapinfo;
371 populated->pending_destroy[345].first = 567;
372 populated->pending_destroy[345].second = 768;
373 populated->pending_noop.insert(890);
375 ls.push_back(populated);