]>
Commit | Line | Data |
---|---|---|
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) 2013 Inktank Storage, Inc. | |
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 <iostream> | |
16 | #include <sstream> | |
17 | ||
18 | #include "ECBackend.h" | |
19 | #include "messages/MOSDPGPush.h" | |
20 | #include "messages/MOSDPGPushReply.h" | |
21 | #include "messages/MOSDECSubOpWrite.h" | |
22 | #include "messages/MOSDECSubOpWriteReply.h" | |
23 | #include "messages/MOSDECSubOpRead.h" | |
24 | #include "messages/MOSDECSubOpReadReply.h" | |
25 | #include "ECMsgTypes.h" | |
26 | ||
27 | #include "PrimaryLogPG.h" | |
28 | ||
29 | #define dout_context cct | |
30 | #define dout_subsys ceph_subsys_osd | |
31 | #define DOUT_PREFIX_ARGS this | |
32 | #undef dout_prefix | |
33 | #define dout_prefix _prefix(_dout, this) | |
34 | static ostream& _prefix(std::ostream *_dout, ECBackend *pgb) { | |
35 | return *_dout << pgb->get_parent()->gen_dbg_prefix(); | |
36 | } | |
37 | ||
38 | struct ECRecoveryHandle : public PGBackend::RecoveryHandle { | |
39 | list<ECBackend::RecoveryOp> ops; | |
40 | }; | |
41 | ||
42 | ostream &operator<<(ostream &lhs, const ECBackend::pipeline_state_t &rhs) { | |
43 | switch (rhs.pipeline_state) { | |
44 | case ECBackend::pipeline_state_t::CACHE_VALID: | |
45 | return lhs << "CACHE_VALID"; | |
46 | case ECBackend::pipeline_state_t::CACHE_INVALID: | |
47 | return lhs << "CACHE_INVALID"; | |
48 | default: | |
49 | assert(0 == "invalid pipeline state"); | |
50 | } | |
51 | return lhs; // unreachable | |
52 | } | |
53 | ||
54 | static ostream &operator<<(ostream &lhs, const map<pg_shard_t, bufferlist> &rhs) | |
55 | { | |
56 | lhs << "["; | |
57 | for (map<pg_shard_t, bufferlist>::const_iterator i = rhs.begin(); | |
58 | i != rhs.end(); | |
59 | ++i) { | |
60 | if (i != rhs.begin()) | |
61 | lhs << ", "; | |
62 | lhs << make_pair(i->first, i->second.length()); | |
63 | } | |
64 | return lhs << "]"; | |
65 | } | |
66 | ||
67 | static ostream &operator<<(ostream &lhs, const map<int, bufferlist> &rhs) | |
68 | { | |
69 | lhs << "["; | |
70 | for (map<int, bufferlist>::const_iterator i = rhs.begin(); | |
71 | i != rhs.end(); | |
72 | ++i) { | |
73 | if (i != rhs.begin()) | |
74 | lhs << ", "; | |
75 | lhs << make_pair(i->first, i->second.length()); | |
76 | } | |
77 | return lhs << "]"; | |
78 | } | |
79 | ||
80 | static ostream &operator<<( | |
81 | ostream &lhs, | |
82 | const boost::tuple<uint64_t, uint64_t, map<pg_shard_t, bufferlist> > &rhs) | |
83 | { | |
84 | return lhs << "(" << rhs.get<0>() << ", " | |
85 | << rhs.get<1>() << ", " << rhs.get<2>() << ")"; | |
86 | } | |
87 | ||
88 | ostream &operator<<(ostream &lhs, const ECBackend::read_request_t &rhs) | |
89 | { | |
90 | return lhs << "read_request_t(to_read=[" << rhs.to_read << "]" | |
91 | << ", need=" << rhs.need | |
92 | << ", want_attrs=" << rhs.want_attrs | |
93 | << ")"; | |
94 | } | |
95 | ||
96 | ostream &operator<<(ostream &lhs, const ECBackend::read_result_t &rhs) | |
97 | { | |
98 | lhs << "read_result_t(r=" << rhs.r | |
99 | << ", errors=" << rhs.errors; | |
100 | if (rhs.attrs) { | |
101 | lhs << ", attrs=" << rhs.attrs.get(); | |
102 | } else { | |
103 | lhs << ", noattrs"; | |
104 | } | |
105 | return lhs << ", returned=" << rhs.returned << ")"; | |
106 | } | |
107 | ||
108 | ostream &operator<<(ostream &lhs, const ECBackend::ReadOp &rhs) | |
109 | { | |
110 | lhs << "ReadOp(tid=" << rhs.tid; | |
111 | if (rhs.op && rhs.op->get_req()) { | |
112 | lhs << ", op="; | |
113 | rhs.op->get_req()->print(lhs); | |
114 | } | |
115 | return lhs << ", to_read=" << rhs.to_read | |
116 | << ", complete=" << rhs.complete | |
117 | << ", priority=" << rhs.priority | |
118 | << ", obj_to_source=" << rhs.obj_to_source | |
119 | << ", source_to_obj=" << rhs.source_to_obj | |
120 | << ", in_progress=" << rhs.in_progress << ")"; | |
121 | } | |
122 | ||
123 | void ECBackend::ReadOp::dump(Formatter *f) const | |
124 | { | |
125 | f->dump_unsigned("tid", tid); | |
126 | if (op && op->get_req()) { | |
127 | f->dump_stream("op") << *(op->get_req()); | |
128 | } | |
129 | f->dump_stream("to_read") << to_read; | |
130 | f->dump_stream("complete") << complete; | |
131 | f->dump_int("priority", priority); | |
132 | f->dump_stream("obj_to_source") << obj_to_source; | |
133 | f->dump_stream("source_to_obj") << source_to_obj; | |
134 | f->dump_stream("in_progress") << in_progress; | |
135 | } | |
136 | ||
137 | ostream &operator<<(ostream &lhs, const ECBackend::Op &rhs) | |
138 | { | |
139 | lhs << "Op(" << rhs.hoid | |
140 | << " v=" << rhs.version | |
141 | << " tt=" << rhs.trim_to | |
142 | << " tid=" << rhs.tid | |
143 | << " reqid=" << rhs.reqid; | |
144 | if (rhs.client_op && rhs.client_op->get_req()) { | |
145 | lhs << " client_op="; | |
146 | rhs.client_op->get_req()->print(lhs); | |
147 | } | |
148 | lhs << " roll_forward_to=" << rhs.roll_forward_to | |
149 | << " temp_added=" << rhs.temp_added | |
150 | << " temp_cleared=" << rhs.temp_cleared | |
151 | << " pending_read=" << rhs.pending_read | |
152 | << " remote_read=" << rhs.remote_read | |
153 | << " remote_read_result=" << rhs.remote_read_result | |
154 | << " pending_apply=" << rhs.pending_apply | |
155 | << " pending_commit=" << rhs.pending_commit | |
156 | << " plan.to_read=" << rhs.plan.to_read | |
157 | << " plan.will_write=" << rhs.plan.will_write | |
158 | << ")"; | |
159 | return lhs; | |
160 | } | |
161 | ||
162 | ostream &operator<<(ostream &lhs, const ECBackend::RecoveryOp &rhs) | |
163 | { | |
164 | return lhs << "RecoveryOp(" | |
165 | << "hoid=" << rhs.hoid | |
166 | << " v=" << rhs.v | |
167 | << " missing_on=" << rhs.missing_on | |
168 | << " missing_on_shards=" << rhs.missing_on_shards | |
169 | << " recovery_info=" << rhs.recovery_info | |
170 | << " recovery_progress=" << rhs.recovery_progress | |
171 | << " obc refcount=" << rhs.obc.use_count() | |
172 | << " state=" << ECBackend::RecoveryOp::tostr(rhs.state) | |
173 | << " waiting_on_pushes=" << rhs.waiting_on_pushes | |
174 | << " extent_requested=" << rhs.extent_requested | |
175 | << ")"; | |
176 | } | |
177 | ||
178 | void ECBackend::RecoveryOp::dump(Formatter *f) const | |
179 | { | |
180 | f->dump_stream("hoid") << hoid; | |
181 | f->dump_stream("v") << v; | |
182 | f->dump_stream("missing_on") << missing_on; | |
183 | f->dump_stream("missing_on_shards") << missing_on_shards; | |
184 | f->dump_stream("recovery_info") << recovery_info; | |
185 | f->dump_stream("recovery_progress") << recovery_progress; | |
186 | f->dump_stream("state") << tostr(state); | |
187 | f->dump_stream("waiting_on_pushes") << waiting_on_pushes; | |
188 | f->dump_stream("extent_requested") << extent_requested; | |
189 | } | |
190 | ||
191 | ECBackend::ECBackend( | |
192 | PGBackend::Listener *pg, | |
193 | coll_t coll, | |
194 | ObjectStore::CollectionHandle &ch, | |
195 | ObjectStore *store, | |
196 | CephContext *cct, | |
197 | ErasureCodeInterfaceRef ec_impl, | |
198 | uint64_t stripe_width) | |
199 | : PGBackend(cct, pg, store, coll, ch), | |
200 | ec_impl(ec_impl), | |
201 | sinfo(ec_impl->get_data_chunk_count(), stripe_width) { | |
202 | assert((ec_impl->get_data_chunk_count() * | |
203 | ec_impl->get_chunk_size(stripe_width)) == stripe_width); | |
204 | } | |
205 | ||
206 | PGBackend::RecoveryHandle *ECBackend::open_recovery_op() | |
207 | { | |
208 | return new ECRecoveryHandle; | |
209 | } | |
210 | ||
211 | void ECBackend::_failed_push(const hobject_t &hoid, | |
212 | pair<RecoveryMessages *, ECBackend::read_result_t &> &in) | |
213 | { | |
214 | ECBackend::read_result_t &res = in.second; | |
215 | dout(10) << __func__ << ": Read error " << hoid << " r=" | |
216 | << res.r << " errors=" << res.errors << dendl; | |
217 | dout(10) << __func__ << ": canceling recovery op for obj " << hoid | |
218 | << dendl; | |
219 | assert(recovery_ops.count(hoid)); | |
220 | eversion_t v = recovery_ops[hoid].v; | |
221 | recovery_ops.erase(hoid); | |
222 | ||
223 | list<pg_shard_t> fl; | |
224 | for (auto&& i : res.errors) { | |
225 | fl.push_back(i.first); | |
226 | } | |
227 | get_parent()->failed_push(fl, hoid); | |
228 | get_parent()->backfill_add_missing(hoid, v); | |
229 | get_parent()->finish_degraded_object(hoid); | |
230 | } | |
231 | ||
232 | struct OnRecoveryReadComplete : | |
233 | public GenContext<pair<RecoveryMessages*, ECBackend::read_result_t& > &> { | |
234 | ECBackend *pg; | |
235 | hobject_t hoid; | |
236 | set<int> want; | |
237 | OnRecoveryReadComplete(ECBackend *pg, const hobject_t &hoid) | |
238 | : pg(pg), hoid(hoid) {} | |
239 | void finish(pair<RecoveryMessages *, ECBackend::read_result_t &> &in) override { | |
240 | ECBackend::read_result_t &res = in.second; | |
241 | if (!(res.r == 0 && res.errors.empty())) { | |
242 | pg->_failed_push(hoid, in); | |
243 | return; | |
244 | } | |
245 | assert(res.returned.size() == 1); | |
246 | pg->handle_recovery_read_complete( | |
247 | hoid, | |
248 | res.returned.back(), | |
249 | res.attrs, | |
250 | in.first); | |
251 | } | |
252 | }; | |
253 | ||
254 | struct RecoveryMessages { | |
255 | map<hobject_t, | |
256 | ECBackend::read_request_t> reads; | |
257 | void read( | |
258 | ECBackend *ec, | |
259 | const hobject_t &hoid, uint64_t off, uint64_t len, | |
260 | const set<pg_shard_t> &need, | |
261 | bool attrs) { | |
262 | list<boost::tuple<uint64_t, uint64_t, uint32_t> > to_read; | |
263 | to_read.push_back(boost::make_tuple(off, len, 0)); | |
264 | assert(!reads.count(hoid)); | |
265 | reads.insert( | |
266 | make_pair( | |
267 | hoid, | |
268 | ECBackend::read_request_t( | |
269 | to_read, | |
270 | need, | |
271 | attrs, | |
272 | new OnRecoveryReadComplete( | |
273 | ec, | |
274 | hoid)))); | |
275 | } | |
276 | ||
277 | map<pg_shard_t, vector<PushOp> > pushes; | |
278 | map<pg_shard_t, vector<PushReplyOp> > push_replies; | |
279 | ObjectStore::Transaction t; | |
280 | RecoveryMessages() {} | |
281 | ~RecoveryMessages(){} | |
282 | }; | |
283 | ||
284 | void ECBackend::handle_recovery_push( | |
285 | const PushOp &op, | |
286 | RecoveryMessages *m) | |
287 | { | |
288 | ostringstream ss; | |
289 | if (get_parent()->check_failsafe_full(ss)) { | |
290 | dout(10) << __func__ << " Out of space (failsafe) processing push request: " << ss.str() << dendl; | |
291 | ceph_abort(); | |
292 | } | |
293 | ||
294 | bool oneshot = op.before_progress.first && op.after_progress.data_complete; | |
295 | ghobject_t tobj; | |
296 | if (oneshot) { | |
297 | tobj = ghobject_t(op.soid, ghobject_t::NO_GEN, | |
298 | get_parent()->whoami_shard().shard); | |
299 | } else { | |
300 | tobj = ghobject_t(get_parent()->get_temp_recovery_object(op.soid, | |
301 | op.version), | |
302 | ghobject_t::NO_GEN, | |
303 | get_parent()->whoami_shard().shard); | |
304 | if (op.before_progress.first) { | |
305 | dout(10) << __func__ << ": Adding oid " | |
306 | << tobj.hobj << " in the temp collection" << dendl; | |
307 | add_temp_obj(tobj.hobj); | |
308 | } | |
309 | } | |
310 | ||
311 | if (op.before_progress.first) { | |
312 | m->t.remove(coll, tobj); | |
313 | m->t.touch(coll, tobj); | |
314 | } | |
315 | ||
316 | if (!op.data_included.empty()) { | |
317 | uint64_t start = op.data_included.range_start(); | |
318 | uint64_t end = op.data_included.range_end(); | |
319 | assert(op.data.length() == (end - start)); | |
320 | ||
321 | m->t.write( | |
322 | coll, | |
323 | tobj, | |
324 | start, | |
325 | op.data.length(), | |
326 | op.data); | |
327 | } else { | |
328 | assert(op.data.length() == 0); | |
329 | } | |
330 | ||
331 | if (op.before_progress.first) { | |
332 | assert(op.attrset.count(string("_"))); | |
333 | m->t.setattrs( | |
334 | coll, | |
335 | tobj, | |
336 | op.attrset); | |
337 | } | |
338 | ||
339 | if (op.after_progress.data_complete && !oneshot) { | |
340 | dout(10) << __func__ << ": Removing oid " | |
341 | << tobj.hobj << " from the temp collection" << dendl; | |
342 | clear_temp_obj(tobj.hobj); | |
343 | m->t.remove(coll, ghobject_t( | |
344 | op.soid, ghobject_t::NO_GEN, get_parent()->whoami_shard().shard)); | |
345 | m->t.collection_move_rename( | |
346 | coll, tobj, | |
347 | coll, ghobject_t( | |
348 | op.soid, ghobject_t::NO_GEN, get_parent()->whoami_shard().shard)); | |
349 | } | |
350 | if (op.after_progress.data_complete) { | |
351 | if ((get_parent()->pgb_is_primary())) { | |
352 | assert(recovery_ops.count(op.soid)); | |
353 | assert(recovery_ops[op.soid].obc); | |
354 | get_parent()->on_local_recover( | |
355 | op.soid, | |
356 | op.recovery_info, | |
357 | recovery_ops[op.soid].obc, | |
358 | false, | |
359 | &m->t); | |
360 | } else { | |
361 | get_parent()->on_local_recover( | |
362 | op.soid, | |
363 | op.recovery_info, | |
364 | ObjectContextRef(), | |
365 | false, | |
366 | &m->t); | |
367 | } | |
368 | } | |
369 | m->push_replies[get_parent()->primary_shard()].push_back(PushReplyOp()); | |
370 | m->push_replies[get_parent()->primary_shard()].back().soid = op.soid; | |
371 | } | |
372 | ||
373 | void ECBackend::handle_recovery_push_reply( | |
374 | const PushReplyOp &op, | |
375 | pg_shard_t from, | |
376 | RecoveryMessages *m) | |
377 | { | |
378 | if (!recovery_ops.count(op.soid)) | |
379 | return; | |
380 | RecoveryOp &rop = recovery_ops[op.soid]; | |
381 | assert(rop.waiting_on_pushes.count(from)); | |
382 | rop.waiting_on_pushes.erase(from); | |
383 | continue_recovery_op(rop, m); | |
384 | } | |
385 | ||
386 | void ECBackend::handle_recovery_read_complete( | |
387 | const hobject_t &hoid, | |
388 | boost::tuple<uint64_t, uint64_t, map<pg_shard_t, bufferlist> > &to_read, | |
389 | boost::optional<map<string, bufferlist> > attrs, | |
390 | RecoveryMessages *m) | |
391 | { | |
392 | dout(10) << __func__ << ": returned " << hoid << " " | |
393 | << "(" << to_read.get<0>() | |
394 | << ", " << to_read.get<1>() | |
395 | << ", " << to_read.get<2>() | |
396 | << ")" | |
397 | << dendl; | |
398 | assert(recovery_ops.count(hoid)); | |
399 | RecoveryOp &op = recovery_ops[hoid]; | |
400 | assert(op.returned_data.empty()); | |
401 | map<int, bufferlist*> target; | |
402 | for (set<shard_id_t>::iterator i = op.missing_on_shards.begin(); | |
403 | i != op.missing_on_shards.end(); | |
404 | ++i) { | |
405 | target[*i] = &(op.returned_data[*i]); | |
406 | } | |
407 | map<int, bufferlist> from; | |
408 | for(map<pg_shard_t, bufferlist>::iterator i = to_read.get<2>().begin(); | |
409 | i != to_read.get<2>().end(); | |
410 | ++i) { | |
411 | from[i->first.shard].claim(i->second); | |
412 | } | |
413 | dout(10) << __func__ << ": " << from << dendl; | |
414 | int r = ECUtil::decode(sinfo, ec_impl, from, target); | |
415 | assert(r == 0); | |
416 | if (attrs) { | |
417 | op.xattrs.swap(*attrs); | |
418 | ||
419 | if (!op.obc) { | |
420 | // attrs only reference the origin bufferlist (decode from | |
421 | // ECSubReadReply message) whose size is much greater than attrs | |
422 | // in recovery. If obc cache it (get_obc maybe cache the attr), | |
423 | // this causes the whole origin bufferlist would not be free | |
424 | // until obc is evicted from obc cache. So rebuild the | |
425 | // bufferlist before cache it. | |
426 | for (map<string, bufferlist>::iterator it = op.xattrs.begin(); | |
427 | it != op.xattrs.end(); | |
428 | ++it) { | |
429 | it->second.rebuild(); | |
430 | } | |
431 | // Need to remove ECUtil::get_hinfo_key() since it should not leak out | |
432 | // of the backend (see bug #12983) | |
433 | map<string, bufferlist> sanitized_attrs(op.xattrs); | |
434 | sanitized_attrs.erase(ECUtil::get_hinfo_key()); | |
435 | op.obc = get_parent()->get_obc(hoid, sanitized_attrs); | |
436 | assert(op.obc); | |
437 | op.recovery_info.size = op.obc->obs.oi.size; | |
438 | op.recovery_info.oi = op.obc->obs.oi; | |
439 | } | |
440 | ||
441 | ECUtil::HashInfo hinfo(ec_impl->get_chunk_count()); | |
442 | if (op.obc->obs.oi.size > 0) { | |
443 | assert(op.xattrs.count(ECUtil::get_hinfo_key())); | |
444 | bufferlist::iterator bp = op.xattrs[ECUtil::get_hinfo_key()].begin(); | |
445 | ::decode(hinfo, bp); | |
446 | } | |
447 | op.hinfo = unstable_hashinfo_registry.lookup_or_create(hoid, hinfo); | |
448 | } | |
449 | assert(op.xattrs.size()); | |
450 | assert(op.obc); | |
451 | continue_recovery_op(op, m); | |
452 | } | |
453 | ||
454 | struct SendPushReplies : public Context { | |
455 | PGBackend::Listener *l; | |
456 | epoch_t epoch; | |
457 | map<int, MOSDPGPushReply*> replies; | |
458 | SendPushReplies( | |
459 | PGBackend::Listener *l, | |
460 | epoch_t epoch, | |
461 | map<int, MOSDPGPushReply*> &in) : l(l), epoch(epoch) { | |
462 | replies.swap(in); | |
463 | } | |
464 | void finish(int) override { | |
465 | for (map<int, MOSDPGPushReply*>::iterator i = replies.begin(); | |
466 | i != replies.end(); | |
467 | ++i) { | |
468 | l->send_message_osd_cluster(i->first, i->second, epoch); | |
469 | } | |
470 | replies.clear(); | |
471 | } | |
472 | ~SendPushReplies() override { | |
473 | for (map<int, MOSDPGPushReply*>::iterator i = replies.begin(); | |
474 | i != replies.end(); | |
475 | ++i) { | |
476 | i->second->put(); | |
477 | } | |
478 | replies.clear(); | |
479 | } | |
480 | }; | |
481 | ||
482 | void ECBackend::dispatch_recovery_messages(RecoveryMessages &m, int priority) | |
483 | { | |
484 | for (map<pg_shard_t, vector<PushOp> >::iterator i = m.pushes.begin(); | |
485 | i != m.pushes.end(); | |
486 | m.pushes.erase(i++)) { | |
487 | MOSDPGPush *msg = new MOSDPGPush(); | |
488 | msg->set_priority(priority); | |
489 | msg->map_epoch = get_parent()->get_epoch(); | |
490 | msg->min_epoch = get_parent()->get_last_peering_reset_epoch(); | |
491 | msg->from = get_parent()->whoami_shard(); | |
492 | msg->pgid = spg_t(get_parent()->get_info().pgid.pgid, i->first.shard); | |
493 | msg->pushes.swap(i->second); | |
494 | msg->compute_cost(cct); | |
495 | get_parent()->send_message( | |
496 | i->first.osd, | |
497 | msg); | |
498 | } | |
499 | map<int, MOSDPGPushReply*> replies; | |
500 | for (map<pg_shard_t, vector<PushReplyOp> >::iterator i = | |
501 | m.push_replies.begin(); | |
502 | i != m.push_replies.end(); | |
503 | m.push_replies.erase(i++)) { | |
504 | MOSDPGPushReply *msg = new MOSDPGPushReply(); | |
505 | msg->set_priority(priority); | |
506 | msg->map_epoch = get_parent()->get_epoch(); | |
507 | msg->min_epoch = get_parent()->get_last_peering_reset_epoch(); | |
508 | msg->from = get_parent()->whoami_shard(); | |
509 | msg->pgid = spg_t(get_parent()->get_info().pgid.pgid, i->first.shard); | |
510 | msg->replies.swap(i->second); | |
511 | msg->compute_cost(cct); | |
512 | replies.insert(make_pair(i->first.osd, msg)); | |
513 | } | |
514 | ||
515 | if (!replies.empty()) { | |
516 | (m.t).register_on_complete( | |
517 | get_parent()->bless_context( | |
518 | new SendPushReplies( | |
519 | get_parent(), | |
520 | get_parent()->get_epoch(), | |
521 | replies))); | |
522 | get_parent()->queue_transaction(std::move(m.t)); | |
523 | } | |
524 | ||
525 | if (m.reads.empty()) | |
526 | return; | |
527 | start_read_op( | |
528 | priority, | |
529 | m.reads, | |
530 | OpRequestRef(), | |
531 | false, true); | |
532 | } | |
533 | ||
534 | void ECBackend::continue_recovery_op( | |
535 | RecoveryOp &op, | |
536 | RecoveryMessages *m) | |
537 | { | |
538 | dout(10) << __func__ << ": continuing " << op << dendl; | |
539 | while (1) { | |
540 | switch (op.state) { | |
541 | case RecoveryOp::IDLE: { | |
542 | // start read | |
543 | op.state = RecoveryOp::READING; | |
544 | assert(!op.recovery_progress.data_complete); | |
545 | set<int> want(op.missing_on_shards.begin(), op.missing_on_shards.end()); | |
546 | uint64_t from = op.recovery_progress.data_recovered_to; | |
547 | uint64_t amount = get_recovery_chunk_size(); | |
548 | ||
549 | if (op.recovery_progress.first && op.obc) { | |
550 | /* We've got the attrs and the hinfo, might as well use them */ | |
551 | op.hinfo = get_hash_info(op.hoid); | |
552 | assert(op.hinfo); | |
553 | op.xattrs = op.obc->attr_cache; | |
554 | ::encode(*(op.hinfo), op.xattrs[ECUtil::get_hinfo_key()]); | |
555 | } | |
556 | ||
557 | set<pg_shard_t> to_read; | |
558 | int r = get_min_avail_to_read_shards( | |
559 | op.hoid, want, true, false, &to_read); | |
560 | if (r != 0) { | |
561 | // we must have lost a recovery source | |
562 | assert(!op.recovery_progress.first); | |
563 | dout(10) << __func__ << ": canceling recovery op for obj " << op.hoid | |
564 | << dendl; | |
565 | get_parent()->cancel_pull(op.hoid); | |
566 | recovery_ops.erase(op.hoid); | |
567 | return; | |
568 | } | |
569 | m->read( | |
570 | this, | |
571 | op.hoid, | |
572 | op.recovery_progress.data_recovered_to, | |
573 | amount, | |
574 | to_read, | |
575 | op.recovery_progress.first && !op.obc); | |
576 | op.extent_requested = make_pair( | |
577 | from, | |
578 | amount); | |
579 | dout(10) << __func__ << ": IDLE return " << op << dendl; | |
580 | return; | |
581 | } | |
582 | case RecoveryOp::READING: { | |
583 | // read completed, start write | |
584 | assert(op.xattrs.size()); | |
585 | assert(op.returned_data.size()); | |
586 | op.state = RecoveryOp::WRITING; | |
587 | ObjectRecoveryProgress after_progress = op.recovery_progress; | |
588 | after_progress.data_recovered_to += op.extent_requested.second; | |
589 | after_progress.first = false; | |
590 | if (after_progress.data_recovered_to >= op.obc->obs.oi.size) { | |
591 | after_progress.data_recovered_to = | |
592 | sinfo.logical_to_next_stripe_offset( | |
593 | op.obc->obs.oi.size); | |
594 | after_progress.data_complete = true; | |
595 | } | |
596 | for (set<pg_shard_t>::iterator mi = op.missing_on.begin(); | |
597 | mi != op.missing_on.end(); | |
598 | ++mi) { | |
599 | assert(op.returned_data.count(mi->shard)); | |
600 | m->pushes[*mi].push_back(PushOp()); | |
601 | PushOp &pop = m->pushes[*mi].back(); | |
602 | pop.soid = op.hoid; | |
603 | pop.version = op.v; | |
604 | pop.data = op.returned_data[mi->shard]; | |
605 | dout(10) << __func__ << ": before_progress=" << op.recovery_progress | |
606 | << ", after_progress=" << after_progress | |
607 | << ", pop.data.length()=" << pop.data.length() | |
608 | << ", size=" << op.obc->obs.oi.size << dendl; | |
609 | assert( | |
610 | pop.data.length() == | |
611 | sinfo.aligned_logical_offset_to_chunk_offset( | |
612 | after_progress.data_recovered_to - | |
613 | op.recovery_progress.data_recovered_to) | |
614 | ); | |
615 | if (pop.data.length()) | |
616 | pop.data_included.insert( | |
617 | sinfo.aligned_logical_offset_to_chunk_offset( | |
618 | op.recovery_progress.data_recovered_to), | |
619 | pop.data.length() | |
620 | ); | |
621 | if (op.recovery_progress.first) { | |
622 | pop.attrset = op.xattrs; | |
623 | } | |
624 | pop.recovery_info = op.recovery_info; | |
625 | pop.before_progress = op.recovery_progress; | |
626 | pop.after_progress = after_progress; | |
627 | if (*mi != get_parent()->primary_shard()) | |
628 | get_parent()->begin_peer_recover( | |
629 | *mi, | |
630 | op.hoid); | |
631 | } | |
632 | op.returned_data.clear(); | |
633 | op.waiting_on_pushes = op.missing_on; | |
634 | op.recovery_progress = after_progress; | |
635 | dout(10) << __func__ << ": READING return " << op << dendl; | |
636 | return; | |
637 | } | |
638 | case RecoveryOp::WRITING: { | |
639 | if (op.waiting_on_pushes.empty()) { | |
640 | if (op.recovery_progress.data_complete) { | |
641 | op.state = RecoveryOp::COMPLETE; | |
642 | for (set<pg_shard_t>::iterator i = op.missing_on.begin(); | |
643 | i != op.missing_on.end(); | |
644 | ++i) { | |
645 | if (*i != get_parent()->primary_shard()) { | |
646 | dout(10) << __func__ << ": on_peer_recover on " << *i | |
647 | << ", obj " << op.hoid << dendl; | |
648 | get_parent()->on_peer_recover( | |
649 | *i, | |
650 | op.hoid, | |
651 | op.recovery_info); | |
652 | } | |
653 | } | |
654 | object_stat_sum_t stat; | |
655 | stat.num_bytes_recovered = op.recovery_info.size; | |
656 | stat.num_keys_recovered = 0; // ??? op ... omap_entries.size(); ? | |
657 | stat.num_objects_recovered = 1; | |
658 | get_parent()->on_global_recover(op.hoid, stat, false); | |
659 | dout(10) << __func__ << ": WRITING return " << op << dendl; | |
660 | recovery_ops.erase(op.hoid); | |
661 | return; | |
662 | } else { | |
663 | op.state = RecoveryOp::IDLE; | |
664 | dout(10) << __func__ << ": WRITING continue " << op << dendl; | |
665 | continue; | |
666 | } | |
667 | } | |
668 | return; | |
669 | } | |
670 | // should never be called once complete | |
671 | case RecoveryOp::COMPLETE: | |
672 | default: { | |
673 | ceph_abort(); | |
674 | }; | |
675 | } | |
676 | } | |
677 | } | |
678 | ||
679 | void ECBackend::run_recovery_op( | |
680 | RecoveryHandle *_h, | |
681 | int priority) | |
682 | { | |
683 | ECRecoveryHandle *h = static_cast<ECRecoveryHandle*>(_h); | |
684 | RecoveryMessages m; | |
685 | for (list<RecoveryOp>::iterator i = h->ops.begin(); | |
686 | i != h->ops.end(); | |
687 | ++i) { | |
688 | dout(10) << __func__ << ": starting " << *i << dendl; | |
689 | assert(!recovery_ops.count(i->hoid)); | |
690 | RecoveryOp &op = recovery_ops.insert(make_pair(i->hoid, *i)).first->second; | |
691 | continue_recovery_op(op, &m); | |
692 | } | |
693 | ||
694 | dispatch_recovery_messages(m, priority); | |
695 | send_recovery_deletes(priority, h->deletes); | |
696 | delete _h; | |
697 | } | |
698 | ||
699 | int ECBackend::recover_object( | |
700 | const hobject_t &hoid, | |
701 | eversion_t v, | |
702 | ObjectContextRef head, | |
703 | ObjectContextRef obc, | |
704 | RecoveryHandle *_h) | |
705 | { | |
706 | ECRecoveryHandle *h = static_cast<ECRecoveryHandle*>(_h); | |
707 | h->ops.push_back(RecoveryOp()); | |
708 | h->ops.back().v = v; | |
709 | h->ops.back().hoid = hoid; | |
710 | h->ops.back().obc = obc; | |
711 | h->ops.back().recovery_info.soid = hoid; | |
712 | h->ops.back().recovery_info.version = v; | |
713 | if (obc) { | |
714 | h->ops.back().recovery_info.size = obc->obs.oi.size; | |
715 | h->ops.back().recovery_info.oi = obc->obs.oi; | |
716 | } | |
717 | if (hoid.is_snap()) { | |
718 | if (obc) { | |
719 | assert(obc->ssc); | |
720 | h->ops.back().recovery_info.ss = obc->ssc->snapset; | |
721 | } else if (head) { | |
722 | assert(head->ssc); | |
723 | h->ops.back().recovery_info.ss = head->ssc->snapset; | |
724 | } else { | |
725 | assert(0 == "neither obc nor head set for a snap object"); | |
726 | } | |
727 | } | |
728 | h->ops.back().recovery_progress.omap_complete = true; | |
729 | for (set<pg_shard_t>::const_iterator i = | |
730 | get_parent()->get_actingbackfill_shards().begin(); | |
731 | i != get_parent()->get_actingbackfill_shards().end(); | |
732 | ++i) { | |
733 | dout(10) << "checking " << *i << dendl; | |
734 | if (get_parent()->get_shard_missing(*i).is_missing(hoid)) { | |
735 | h->ops.back().missing_on.insert(*i); | |
736 | h->ops.back().missing_on_shards.insert(i->shard); | |
737 | } | |
738 | } | |
739 | dout(10) << __func__ << ": built op " << h->ops.back() << dendl; | |
740 | return 0; | |
741 | } | |
742 | ||
743 | bool ECBackend::can_handle_while_inactive( | |
744 | OpRequestRef _op) | |
745 | { | |
746 | return false; | |
747 | } | |
748 | ||
749 | bool ECBackend::_handle_message( | |
750 | OpRequestRef _op) | |
751 | { | |
752 | dout(10) << __func__ << ": " << *_op->get_req() << dendl; | |
753 | int priority = _op->get_req()->get_priority(); | |
754 | switch (_op->get_req()->get_type()) { | |
755 | case MSG_OSD_EC_WRITE: { | |
756 | // NOTE: this is non-const because handle_sub_write modifies the embedded | |
757 | // ObjectStore::Transaction in place (and then std::move's it). It does | |
758 | // not conflict with ECSubWrite's operator<<. | |
759 | MOSDECSubOpWrite *op = static_cast<MOSDECSubOpWrite*>( | |
760 | _op->get_nonconst_req()); | |
761 | handle_sub_write(op->op.from, _op, op->op, _op->pg_trace); | |
762 | return true; | |
763 | } | |
764 | case MSG_OSD_EC_WRITE_REPLY: { | |
765 | const MOSDECSubOpWriteReply *op = static_cast<const MOSDECSubOpWriteReply*>( | |
766 | _op->get_req()); | |
767 | handle_sub_write_reply(op->op.from, op->op, _op->pg_trace); | |
768 | return true; | |
769 | } | |
770 | case MSG_OSD_EC_READ: { | |
771 | const MOSDECSubOpRead *op = static_cast<const MOSDECSubOpRead*>(_op->get_req()); | |
772 | MOSDECSubOpReadReply *reply = new MOSDECSubOpReadReply; | |
773 | reply->pgid = get_parent()->primary_spg_t(); | |
774 | reply->map_epoch = get_parent()->get_epoch(); | |
775 | reply->min_epoch = get_parent()->get_interval_start_epoch(); | |
776 | handle_sub_read(op->op.from, op->op, &(reply->op), _op->pg_trace); | |
777 | reply->trace = _op->pg_trace; | |
778 | get_parent()->send_message_osd_cluster( | |
779 | op->op.from.osd, reply, get_parent()->get_epoch()); | |
780 | return true; | |
781 | } | |
782 | case MSG_OSD_EC_READ_REPLY: { | |
783 | // NOTE: this is non-const because handle_sub_read_reply steals resulting | |
784 | // buffers. It does not conflict with ECSubReadReply operator<<. | |
785 | MOSDECSubOpReadReply *op = static_cast<MOSDECSubOpReadReply*>( | |
786 | _op->get_nonconst_req()); | |
787 | RecoveryMessages rm; | |
788 | handle_sub_read_reply(op->op.from, op->op, &rm, _op->pg_trace); | |
789 | dispatch_recovery_messages(rm, priority); | |
790 | return true; | |
791 | } | |
792 | case MSG_OSD_PG_PUSH: { | |
793 | const MOSDPGPush *op = static_cast<const MOSDPGPush *>(_op->get_req()); | |
794 | RecoveryMessages rm; | |
795 | for (vector<PushOp>::const_iterator i = op->pushes.begin(); | |
796 | i != op->pushes.end(); | |
797 | ++i) { | |
798 | handle_recovery_push(*i, &rm); | |
799 | } | |
800 | dispatch_recovery_messages(rm, priority); | |
801 | return true; | |
802 | } | |
803 | case MSG_OSD_PG_PUSH_REPLY: { | |
804 | const MOSDPGPushReply *op = static_cast<const MOSDPGPushReply *>( | |
805 | _op->get_req()); | |
806 | RecoveryMessages rm; | |
807 | for (vector<PushReplyOp>::const_iterator i = op->replies.begin(); | |
808 | i != op->replies.end(); | |
809 | ++i) { | |
810 | handle_recovery_push_reply(*i, op->from, &rm); | |
811 | } | |
812 | dispatch_recovery_messages(rm, priority); | |
813 | return true; | |
814 | } | |
815 | default: | |
816 | return false; | |
817 | } | |
818 | return false; | |
819 | } | |
820 | ||
821 | struct SubWriteCommitted : public Context { | |
822 | ECBackend *pg; | |
823 | OpRequestRef msg; | |
824 | ceph_tid_t tid; | |
825 | eversion_t version; | |
826 | eversion_t last_complete; | |
827 | const ZTracer::Trace trace; | |
828 | SubWriteCommitted( | |
829 | ECBackend *pg, | |
830 | OpRequestRef msg, | |
831 | ceph_tid_t tid, | |
832 | eversion_t version, | |
833 | eversion_t last_complete, | |
834 | const ZTracer::Trace &trace) | |
835 | : pg(pg), msg(msg), tid(tid), | |
836 | version(version), last_complete(last_complete), trace(trace) {} | |
837 | void finish(int) override { | |
838 | if (msg) | |
839 | msg->mark_event("sub_op_committed"); | |
840 | pg->sub_write_committed(tid, version, last_complete, trace); | |
841 | } | |
842 | }; | |
843 | void ECBackend::sub_write_committed( | |
844 | ceph_tid_t tid, eversion_t version, eversion_t last_complete, | |
845 | const ZTracer::Trace &trace) { | |
846 | if (get_parent()->pgb_is_primary()) { | |
847 | ECSubWriteReply reply; | |
848 | reply.tid = tid; | |
849 | reply.last_complete = last_complete; | |
850 | reply.committed = true; | |
851 | reply.from = get_parent()->whoami_shard(); | |
852 | handle_sub_write_reply( | |
853 | get_parent()->whoami_shard(), | |
854 | reply, trace); | |
855 | } else { | |
856 | get_parent()->update_last_complete_ondisk(last_complete); | |
857 | MOSDECSubOpWriteReply *r = new MOSDECSubOpWriteReply; | |
858 | r->pgid = get_parent()->primary_spg_t(); | |
859 | r->map_epoch = get_parent()->get_epoch(); | |
860 | r->min_epoch = get_parent()->get_interval_start_epoch(); | |
861 | r->op.tid = tid; | |
862 | r->op.last_complete = last_complete; | |
863 | r->op.committed = true; | |
864 | r->op.from = get_parent()->whoami_shard(); | |
865 | r->set_priority(CEPH_MSG_PRIO_HIGH); | |
866 | r->trace = trace; | |
867 | r->trace.event("sending sub op commit"); | |
868 | get_parent()->send_message_osd_cluster( | |
869 | get_parent()->primary_shard().osd, r, get_parent()->get_epoch()); | |
870 | } | |
871 | } | |
872 | ||
873 | struct SubWriteApplied : public Context { | |
874 | ECBackend *pg; | |
875 | OpRequestRef msg; | |
876 | ceph_tid_t tid; | |
877 | eversion_t version; | |
878 | const ZTracer::Trace trace; | |
879 | SubWriteApplied( | |
880 | ECBackend *pg, | |
881 | OpRequestRef msg, | |
882 | ceph_tid_t tid, | |
883 | eversion_t version, | |
884 | const ZTracer::Trace &trace) | |
885 | : pg(pg), msg(msg), tid(tid), version(version), trace(trace) {} | |
886 | void finish(int) override { | |
887 | if (msg) | |
888 | msg->mark_event("sub_op_applied"); | |
889 | pg->sub_write_applied(tid, version, trace); | |
890 | } | |
891 | }; | |
892 | void ECBackend::sub_write_applied( | |
893 | ceph_tid_t tid, eversion_t version, | |
894 | const ZTracer::Trace &trace) { | |
895 | parent->op_applied(version); | |
896 | if (get_parent()->pgb_is_primary()) { | |
897 | ECSubWriteReply reply; | |
898 | reply.from = get_parent()->whoami_shard(); | |
899 | reply.tid = tid; | |
900 | reply.applied = true; | |
901 | handle_sub_write_reply( | |
902 | get_parent()->whoami_shard(), | |
903 | reply, trace); | |
904 | } else { | |
905 | MOSDECSubOpWriteReply *r = new MOSDECSubOpWriteReply; | |
906 | r->pgid = get_parent()->primary_spg_t(); | |
907 | r->map_epoch = get_parent()->get_epoch(); | |
908 | r->min_epoch = get_parent()->get_interval_start_epoch(); | |
909 | r->op.from = get_parent()->whoami_shard(); | |
910 | r->op.tid = tid; | |
911 | r->op.applied = true; | |
912 | r->set_priority(CEPH_MSG_PRIO_HIGH); | |
913 | r->trace = trace; | |
914 | r->trace.event("sending sub op apply"); | |
915 | get_parent()->send_message_osd_cluster( | |
916 | get_parent()->primary_shard().osd, r, get_parent()->get_epoch()); | |
917 | } | |
918 | } | |
919 | ||
920 | void ECBackend::handle_sub_write( | |
921 | pg_shard_t from, | |
922 | OpRequestRef msg, | |
923 | ECSubWrite &op, | |
924 | const ZTracer::Trace &trace, | |
925 | Context *on_local_applied_sync) | |
926 | { | |
927 | if (msg) | |
928 | msg->mark_started(); | |
929 | trace.event("handle_sub_write"); | |
930 | assert(!get_parent()->get_log().get_missing().is_missing(op.soid)); | |
931 | if (!get_parent()->pgb_is_primary()) | |
932 | get_parent()->update_stats(op.stats); | |
933 | ObjectStore::Transaction localt; | |
934 | if (!op.temp_added.empty()) { | |
935 | add_temp_objs(op.temp_added); | |
936 | } | |
937 | if (op.backfill) { | |
938 | for (set<hobject_t>::iterator i = op.temp_removed.begin(); | |
939 | i != op.temp_removed.end(); | |
940 | ++i) { | |
941 | dout(10) << __func__ << ": removing object " << *i | |
942 | << " since we won't get the transaction" << dendl; | |
943 | localt.remove( | |
944 | coll, | |
945 | ghobject_t( | |
946 | *i, | |
947 | ghobject_t::NO_GEN, | |
948 | get_parent()->whoami_shard().shard)); | |
949 | } | |
950 | } | |
951 | clear_temp_objs(op.temp_removed); | |
952 | get_parent()->log_operation( | |
953 | op.log_entries, | |
954 | op.updated_hit_set_history, | |
955 | op.trim_to, | |
956 | op.roll_forward_to, | |
957 | !op.backfill, | |
958 | localt); | |
959 | ||
960 | PrimaryLogPG *_rPG = dynamic_cast<PrimaryLogPG *>(get_parent()); | |
961 | if (_rPG && !_rPG->is_undersized() && | |
962 | (unsigned)get_parent()->whoami_shard().shard >= ec_impl->get_data_chunk_count()) | |
963 | op.t.set_fadvise_flag(CEPH_OSD_OP_FLAG_FADVISE_DONTNEED); | |
964 | ||
965 | if (on_local_applied_sync) { | |
966 | dout(10) << "Queueing onreadable_sync: " << on_local_applied_sync << dendl; | |
967 | localt.register_on_applied_sync(on_local_applied_sync); | |
968 | } | |
969 | localt.register_on_commit( | |
970 | get_parent()->bless_context( | |
971 | new SubWriteCommitted( | |
972 | this, msg, op.tid, | |
973 | op.at_version, | |
974 | get_parent()->get_info().last_complete, trace))); | |
975 | localt.register_on_applied( | |
976 | get_parent()->bless_context( | |
977 | new SubWriteApplied(this, msg, op.tid, op.at_version, trace))); | |
978 | vector<ObjectStore::Transaction> tls; | |
979 | tls.reserve(2); | |
980 | tls.push_back(std::move(op.t)); | |
981 | tls.push_back(std::move(localt)); | |
982 | get_parent()->queue_transactions(tls, msg); | |
983 | } | |
984 | ||
985 | void ECBackend::handle_sub_read( | |
986 | pg_shard_t from, | |
987 | const ECSubRead &op, | |
988 | ECSubReadReply *reply, | |
989 | const ZTracer::Trace &trace) | |
990 | { | |
991 | trace.event("handle sub read"); | |
992 | shard_id_t shard = get_parent()->whoami_shard().shard; | |
993 | for(auto i = op.to_read.begin(); | |
994 | i != op.to_read.end(); | |
995 | ++i) { | |
996 | int r = 0; | |
997 | ECUtil::HashInfoRef hinfo; | |
998 | if (!get_parent()->get_pool().allows_ecoverwrites()) { | |
999 | hinfo = get_hash_info(i->first); | |
1000 | if (!hinfo) { | |
1001 | r = -EIO; | |
1002 | get_parent()->clog_error() << "Corruption detected: object " << i->first | |
1003 | << " is missing hash_info"; | |
1004 | dout(5) << __func__ << ": No hinfo for " << i->first << dendl; | |
1005 | goto error; | |
1006 | } | |
1007 | } | |
1008 | for (auto j = i->second.begin(); j != i->second.end(); ++j) { | |
1009 | bufferlist bl; | |
1010 | r = store->read( | |
1011 | ch, | |
1012 | ghobject_t(i->first, ghobject_t::NO_GEN, shard), | |
1013 | j->get<0>(), | |
1014 | j->get<1>(), | |
1015 | bl, j->get<2>()); | |
1016 | if (r < 0) { | |
1017 | get_parent()->clog_error() << "Error " << r | |
1018 | << " reading object " | |
1019 | << i->first; | |
1020 | dout(5) << __func__ << ": Error " << r | |
1021 | << " reading " << i->first << dendl; | |
1022 | goto error; | |
1023 | } else { | |
1024 | dout(20) << __func__ << " read request=" << j->get<1>() << " r=" << r << " len=" << bl.length() << dendl; | |
1025 | reply->buffers_read[i->first].push_back( | |
1026 | make_pair( | |
1027 | j->get<0>(), | |
1028 | bl) | |
1029 | ); | |
1030 | } | |
1031 | ||
1032 | if (!get_parent()->get_pool().allows_ecoverwrites()) { | |
1033 | // This shows that we still need deep scrub because large enough files | |
1034 | // are read in sections, so the digest check here won't be done here. | |
1035 | // Do NOT check osd_read_eio_on_bad_digest here. We need to report | |
1036 | // the state of our chunk in case other chunks could substitute. | |
1037 | assert(hinfo->has_chunk_hash()); | |
1038 | if ((bl.length() == hinfo->get_total_chunk_size()) && | |
1039 | (j->get<0>() == 0)) { | |
1040 | dout(20) << __func__ << ": Checking hash of " << i->first << dendl; | |
1041 | bufferhash h(-1); | |
1042 | h << bl; | |
1043 | if (h.digest() != hinfo->get_chunk_hash(shard)) { | |
1044 | get_parent()->clog_error() << "Bad hash for " << i->first << " digest 0x" | |
1045 | << hex << h.digest() << " expected 0x" << hinfo->get_chunk_hash(shard) << dec; | |
1046 | dout(5) << __func__ << ": Bad hash for " << i->first << " digest 0x" | |
1047 | << hex << h.digest() << " expected 0x" << hinfo->get_chunk_hash(shard) << dec << dendl; | |
1048 | r = -EIO; | |
1049 | goto error; | |
1050 | } | |
1051 | } | |
1052 | } | |
1053 | } | |
1054 | continue; | |
1055 | error: | |
1056 | // Do NOT check osd_read_eio_on_bad_digest here. We need to report | |
1057 | // the state of our chunk in case other chunks could substitute. | |
1058 | reply->buffers_read.erase(i->first); | |
1059 | reply->errors[i->first] = r; | |
1060 | } | |
1061 | for (set<hobject_t>::iterator i = op.attrs_to_read.begin(); | |
1062 | i != op.attrs_to_read.end(); | |
1063 | ++i) { | |
1064 | dout(10) << __func__ << ": fulfilling attr request on " | |
1065 | << *i << dendl; | |
1066 | if (reply->errors.count(*i)) | |
1067 | continue; | |
1068 | int r = store->getattrs( | |
1069 | ch, | |
1070 | ghobject_t( | |
1071 | *i, ghobject_t::NO_GEN, get_parent()->whoami_shard().shard), | |
1072 | reply->attrs_read[*i]); | |
1073 | if (r < 0) { | |
1074 | reply->buffers_read.erase(*i); | |
1075 | reply->errors[*i] = r; | |
1076 | } | |
1077 | } | |
1078 | reply->from = get_parent()->whoami_shard(); | |
1079 | reply->tid = op.tid; | |
1080 | } | |
1081 | ||
1082 | void ECBackend::handle_sub_write_reply( | |
1083 | pg_shard_t from, | |
1084 | const ECSubWriteReply &op, | |
1085 | const ZTracer::Trace &trace) | |
1086 | { | |
1087 | map<ceph_tid_t, Op>::iterator i = tid_to_op_map.find(op.tid); | |
1088 | assert(i != tid_to_op_map.end()); | |
1089 | if (op.committed) { | |
1090 | trace.event("sub write committed"); | |
1091 | assert(i->second.pending_commit.count(from)); | |
1092 | i->second.pending_commit.erase(from); | |
1093 | if (from != get_parent()->whoami_shard()) { | |
1094 | get_parent()->update_peer_last_complete_ondisk(from, op.last_complete); | |
1095 | } | |
1096 | } | |
1097 | if (op.applied) { | |
1098 | trace.event("sub write applied"); | |
1099 | assert(i->second.pending_apply.count(from)); | |
1100 | i->second.pending_apply.erase(from); | |
1101 | } | |
1102 | ||
1103 | if (i->second.pending_apply.empty() && i->second.on_all_applied) { | |
1104 | dout(10) << __func__ << " Calling on_all_applied on " << i->second << dendl; | |
1105 | i->second.on_all_applied->complete(0); | |
1106 | i->second.on_all_applied = 0; | |
1107 | i->second.trace.event("ec write all applied"); | |
1108 | } | |
1109 | if (i->second.pending_commit.empty() && i->second.on_all_commit) { | |
1110 | dout(10) << __func__ << " Calling on_all_commit on " << i->second << dendl; | |
1111 | i->second.on_all_commit->complete(0); | |
1112 | i->second.on_all_commit = 0; | |
1113 | i->second.trace.event("ec write all committed"); | |
1114 | } | |
1115 | check_ops(); | |
1116 | } | |
1117 | ||
1118 | void ECBackend::handle_sub_read_reply( | |
1119 | pg_shard_t from, | |
1120 | ECSubReadReply &op, | |
1121 | RecoveryMessages *m, | |
1122 | const ZTracer::Trace &trace) | |
1123 | { | |
1124 | trace.event("ec sub read reply"); | |
1125 | dout(10) << __func__ << ": reply " << op << dendl; | |
1126 | map<ceph_tid_t, ReadOp>::iterator iter = tid_to_read_map.find(op.tid); | |
1127 | if (iter == tid_to_read_map.end()) { | |
1128 | //canceled | |
1129 | dout(20) << __func__ << ": dropped " << op << dendl; | |
1130 | return; | |
1131 | } | |
1132 | ReadOp &rop = iter->second; | |
1133 | for (auto i = op.buffers_read.begin(); | |
1134 | i != op.buffers_read.end(); | |
1135 | ++i) { | |
1136 | assert(!op.errors.count(i->first)); // If attribute error we better not have sent a buffer | |
1137 | if (!rop.to_read.count(i->first)) { | |
1138 | // We canceled this read! @see filter_read_op | |
1139 | dout(20) << __func__ << " to_read skipping" << dendl; | |
1140 | continue; | |
1141 | } | |
1142 | list<boost::tuple<uint64_t, uint64_t, uint32_t> >::const_iterator req_iter = | |
1143 | rop.to_read.find(i->first)->second.to_read.begin(); | |
1144 | list< | |
1145 | boost::tuple< | |
1146 | uint64_t, uint64_t, map<pg_shard_t, bufferlist> > >::iterator riter = | |
1147 | rop.complete[i->first].returned.begin(); | |
1148 | for (list<pair<uint64_t, bufferlist> >::iterator j = i->second.begin(); | |
1149 | j != i->second.end(); | |
1150 | ++j, ++req_iter, ++riter) { | |
1151 | assert(req_iter != rop.to_read.find(i->first)->second.to_read.end()); | |
1152 | assert(riter != rop.complete[i->first].returned.end()); | |
1153 | pair<uint64_t, uint64_t> adjusted = | |
1154 | sinfo.aligned_offset_len_to_chunk( | |
1155 | make_pair(req_iter->get<0>(), req_iter->get<1>())); | |
1156 | assert(adjusted.first == j->first); | |
1157 | riter->get<2>()[from].claim(j->second); | |
1158 | } | |
1159 | } | |
1160 | for (auto i = op.attrs_read.begin(); | |
1161 | i != op.attrs_read.end(); | |
1162 | ++i) { | |
1163 | assert(!op.errors.count(i->first)); // if read error better not have sent an attribute | |
1164 | if (!rop.to_read.count(i->first)) { | |
1165 | // We canceled this read! @see filter_read_op | |
1166 | dout(20) << __func__ << " to_read skipping" << dendl; | |
1167 | continue; | |
1168 | } | |
1169 | rop.complete[i->first].attrs = map<string, bufferlist>(); | |
1170 | (*(rop.complete[i->first].attrs)).swap(i->second); | |
1171 | } | |
1172 | for (auto i = op.errors.begin(); | |
1173 | i != op.errors.end(); | |
1174 | ++i) { | |
1175 | rop.complete[i->first].errors.insert( | |
1176 | make_pair( | |
1177 | from, | |
1178 | i->second)); | |
1179 | dout(20) << __func__ << " shard=" << from << " error=" << i->second << dendl; | |
1180 | } | |
1181 | ||
1182 | map<pg_shard_t, set<ceph_tid_t> >::iterator siter = | |
1183 | shard_to_read_map.find(from); | |
1184 | assert(siter != shard_to_read_map.end()); | |
1185 | assert(siter->second.count(op.tid)); | |
1186 | siter->second.erase(op.tid); | |
1187 | ||
1188 | assert(rop.in_progress.count(from)); | |
1189 | rop.in_progress.erase(from); | |
1190 | unsigned is_complete = 0; | |
1191 | // For redundant reads check for completion as each shard comes in, | |
1192 | // or in a non-recovery read check for completion once all the shards read. | |
1193 | if (rop.do_redundant_reads || rop.in_progress.empty()) { | |
1194 | for (map<hobject_t, read_result_t>::const_iterator iter = | |
1195 | rop.complete.begin(); | |
1196 | iter != rop.complete.end(); | |
1197 | ++iter) { | |
1198 | set<int> have; | |
1199 | for (map<pg_shard_t, bufferlist>::const_iterator j = | |
1200 | iter->second.returned.front().get<2>().begin(); | |
1201 | j != iter->second.returned.front().get<2>().end(); | |
1202 | ++j) { | |
1203 | have.insert(j->first.shard); | |
1204 | dout(20) << __func__ << " have shard=" << j->first.shard << dendl; | |
1205 | } | |
1206 | set<int> want_to_read, dummy_minimum; | |
1207 | get_want_to_read_shards(&want_to_read); | |
1208 | int err; | |
1209 | if ((err = ec_impl->minimum_to_decode(want_to_read, have, &dummy_minimum)) < 0) { | |
1210 | dout(20) << __func__ << " minimum_to_decode failed" << dendl; | |
1211 | if (rop.in_progress.empty()) { | |
1212 | // If we don't have enough copies and we haven't sent reads for all shards | |
1213 | // we can send the rest of the reads, if any. | |
1214 | if (!rop.do_redundant_reads) { | |
1215 | int r = send_all_remaining_reads(iter->first, rop); | |
1216 | if (r == 0) { | |
1217 | // We added to in_progress and not incrementing is_complete | |
1218 | continue; | |
1219 | } | |
1220 | // Couldn't read any additional shards so handle as completed with errors | |
1221 | } | |
1222 | // We don't want to confuse clients / RBD with objectstore error | |
1223 | // values in particular ENOENT. We may have different error returns | |
1224 | // from different shards, so we'll return minimum_to_decode() error | |
1225 | // (usually EIO) to reader. It is likely an error here is due to a | |
1226 | // damaged pg. | |
1227 | rop.complete[iter->first].r = err; | |
1228 | ++is_complete; | |
1229 | } | |
1230 | } else { | |
1231 | assert(rop.complete[iter->first].r == 0); | |
1232 | if (!rop.complete[iter->first].errors.empty()) { | |
1233 | if (cct->_conf->osd_read_ec_check_for_errors) { | |
1234 | dout(10) << __func__ << ": Not ignoring errors, use one shard err=" << err << dendl; | |
1235 | err = rop.complete[iter->first].errors.begin()->second; | |
1236 | rop.complete[iter->first].r = err; | |
1237 | } else { | |
1238 | get_parent()->clog_warn() << "Error(s) ignored for " | |
1239 | << iter->first << " enough copies available"; | |
1240 | dout(10) << __func__ << " Error(s) ignored for " << iter->first | |
1241 | << " enough copies available" << dendl; | |
1242 | rop.complete[iter->first].errors.clear(); | |
1243 | } | |
1244 | } | |
1245 | ++is_complete; | |
1246 | } | |
1247 | } | |
1248 | } | |
1249 | if (rop.in_progress.empty() || is_complete == rop.complete.size()) { | |
1250 | dout(20) << __func__ << " Complete: " << rop << dendl; | |
1251 | rop.trace.event("ec read complete"); | |
1252 | complete_read_op(rop, m); | |
1253 | } else { | |
1254 | dout(10) << __func__ << " readop not complete: " << rop << dendl; | |
1255 | } | |
1256 | } | |
1257 | ||
1258 | void ECBackend::complete_read_op(ReadOp &rop, RecoveryMessages *m) | |
1259 | { | |
1260 | map<hobject_t, read_request_t>::iterator reqiter = | |
1261 | rop.to_read.begin(); | |
1262 | map<hobject_t, read_result_t>::iterator resiter = | |
1263 | rop.complete.begin(); | |
1264 | assert(rop.to_read.size() == rop.complete.size()); | |
1265 | for (; reqiter != rop.to_read.end(); ++reqiter, ++resiter) { | |
1266 | if (reqiter->second.cb) { | |
1267 | pair<RecoveryMessages *, read_result_t &> arg( | |
1268 | m, resiter->second); | |
1269 | reqiter->second.cb->complete(arg); | |
1270 | reqiter->second.cb = NULL; | |
1271 | } | |
1272 | } | |
1273 | tid_to_read_map.erase(rop.tid); | |
1274 | } | |
1275 | ||
1276 | struct FinishReadOp : public GenContext<ThreadPool::TPHandle&> { | |
1277 | ECBackend *ec; | |
1278 | ceph_tid_t tid; | |
1279 | FinishReadOp(ECBackend *ec, ceph_tid_t tid) : ec(ec), tid(tid) {} | |
1280 | void finish(ThreadPool::TPHandle &handle) override { | |
1281 | auto ropiter = ec->tid_to_read_map.find(tid); | |
1282 | assert(ropiter != ec->tid_to_read_map.end()); | |
1283 | int priority = ropiter->second.priority; | |
1284 | RecoveryMessages rm; | |
1285 | ec->complete_read_op(ropiter->second, &rm); | |
1286 | ec->dispatch_recovery_messages(rm, priority); | |
1287 | } | |
1288 | }; | |
1289 | ||
1290 | void ECBackend::filter_read_op( | |
1291 | const OSDMapRef& osdmap, | |
1292 | ReadOp &op) | |
1293 | { | |
1294 | set<hobject_t> to_cancel; | |
1295 | for (map<pg_shard_t, set<hobject_t> >::iterator i = op.source_to_obj.begin(); | |
1296 | i != op.source_to_obj.end(); | |
1297 | ++i) { | |
1298 | if (osdmap->is_down(i->first.osd)) { | |
1299 | to_cancel.insert(i->second.begin(), i->second.end()); | |
1300 | op.in_progress.erase(i->first); | |
1301 | continue; | |
1302 | } | |
1303 | } | |
1304 | ||
1305 | if (to_cancel.empty()) | |
1306 | return; | |
1307 | ||
1308 | for (map<pg_shard_t, set<hobject_t> >::iterator i = op.source_to_obj.begin(); | |
1309 | i != op.source_to_obj.end(); | |
1310 | ) { | |
1311 | for (set<hobject_t>::iterator j = i->second.begin(); | |
1312 | j != i->second.end(); | |
1313 | ) { | |
1314 | if (to_cancel.count(*j)) | |
1315 | i->second.erase(j++); | |
1316 | else | |
1317 | ++j; | |
1318 | } | |
1319 | if (i->second.empty()) { | |
1320 | op.source_to_obj.erase(i++); | |
1321 | } else { | |
1322 | assert(!osdmap->is_down(i->first.osd)); | |
1323 | ++i; | |
1324 | } | |
1325 | } | |
1326 | ||
1327 | for (set<hobject_t>::iterator i = to_cancel.begin(); | |
1328 | i != to_cancel.end(); | |
1329 | ++i) { | |
1330 | get_parent()->cancel_pull(*i); | |
1331 | ||
1332 | assert(op.to_read.count(*i)); | |
1333 | read_request_t &req = op.to_read.find(*i)->second; | |
1334 | dout(10) << __func__ << ": canceling " << req | |
1335 | << " for obj " << *i << dendl; | |
1336 | assert(req.cb); | |
1337 | delete req.cb; | |
1338 | req.cb = NULL; | |
1339 | ||
1340 | op.to_read.erase(*i); | |
1341 | op.complete.erase(*i); | |
1342 | recovery_ops.erase(*i); | |
1343 | } | |
1344 | ||
1345 | if (op.in_progress.empty()) { | |
1346 | get_parent()->schedule_recovery_work( | |
1347 | get_parent()->bless_gencontext( | |
1348 | new FinishReadOp(this, op.tid))); | |
1349 | } | |
1350 | } | |
1351 | ||
1352 | void ECBackend::check_recovery_sources(const OSDMapRef& osdmap) | |
1353 | { | |
1354 | set<ceph_tid_t> tids_to_filter; | |
1355 | for (map<pg_shard_t, set<ceph_tid_t> >::iterator | |
1356 | i = shard_to_read_map.begin(); | |
1357 | i != shard_to_read_map.end(); | |
1358 | ) { | |
1359 | if (osdmap->is_down(i->first.osd)) { | |
1360 | tids_to_filter.insert(i->second.begin(), i->second.end()); | |
1361 | shard_to_read_map.erase(i++); | |
1362 | } else { | |
1363 | ++i; | |
1364 | } | |
1365 | } | |
1366 | for (set<ceph_tid_t>::iterator i = tids_to_filter.begin(); | |
1367 | i != tids_to_filter.end(); | |
1368 | ++i) { | |
1369 | map<ceph_tid_t, ReadOp>::iterator j = tid_to_read_map.find(*i); | |
1370 | assert(j != tid_to_read_map.end()); | |
1371 | filter_read_op(osdmap, j->second); | |
1372 | } | |
1373 | } | |
1374 | ||
1375 | void ECBackend::on_change() | |
1376 | { | |
1377 | dout(10) << __func__ << dendl; | |
1378 | ||
1379 | completed_to = eversion_t(); | |
1380 | committed_to = eversion_t(); | |
1381 | pipeline_state.clear(); | |
1382 | waiting_reads.clear(); | |
1383 | waiting_state.clear(); | |
1384 | waiting_commit.clear(); | |
1385 | for (auto &&op: tid_to_op_map) { | |
1386 | cache.release_write_pin(op.second.pin); | |
1387 | } | |
1388 | tid_to_op_map.clear(); | |
1389 | ||
1390 | for (map<ceph_tid_t, ReadOp>::iterator i = tid_to_read_map.begin(); | |
1391 | i != tid_to_read_map.end(); | |
1392 | ++i) { | |
1393 | dout(10) << __func__ << ": cancelling " << i->second << dendl; | |
1394 | for (map<hobject_t, read_request_t>::iterator j = | |
1395 | i->second.to_read.begin(); | |
1396 | j != i->second.to_read.end(); | |
1397 | ++j) { | |
1398 | delete j->second.cb; | |
1399 | j->second.cb = 0; | |
1400 | } | |
1401 | } | |
1402 | tid_to_read_map.clear(); | |
1403 | in_progress_client_reads.clear(); | |
1404 | shard_to_read_map.clear(); | |
1405 | clear_recovery_state(); | |
1406 | } | |
1407 | ||
1408 | void ECBackend::clear_recovery_state() | |
1409 | { | |
1410 | recovery_ops.clear(); | |
1411 | } | |
1412 | ||
1413 | void ECBackend::on_flushed() | |
1414 | { | |
1415 | } | |
1416 | ||
1417 | void ECBackend::dump_recovery_info(Formatter *f) const | |
1418 | { | |
1419 | f->open_array_section("recovery_ops"); | |
1420 | for (map<hobject_t, RecoveryOp>::const_iterator i = recovery_ops.begin(); | |
1421 | i != recovery_ops.end(); | |
1422 | ++i) { | |
1423 | f->open_object_section("op"); | |
1424 | i->second.dump(f); | |
1425 | f->close_section(); | |
1426 | } | |
1427 | f->close_section(); | |
1428 | f->open_array_section("read_ops"); | |
1429 | for (map<ceph_tid_t, ReadOp>::const_iterator i = tid_to_read_map.begin(); | |
1430 | i != tid_to_read_map.end(); | |
1431 | ++i) { | |
1432 | f->open_object_section("read_op"); | |
1433 | i->second.dump(f); | |
1434 | f->close_section(); | |
1435 | } | |
1436 | f->close_section(); | |
1437 | } | |
1438 | ||
1439 | void ECBackend::submit_transaction( | |
1440 | const hobject_t &hoid, | |
1441 | const object_stat_sum_t &delta_stats, | |
1442 | const eversion_t &at_version, | |
1443 | PGTransactionUPtr &&t, | |
1444 | const eversion_t &trim_to, | |
1445 | const eversion_t &roll_forward_to, | |
1446 | const vector<pg_log_entry_t> &log_entries, | |
1447 | boost::optional<pg_hit_set_history_t> &hset_history, | |
1448 | Context *on_local_applied_sync, | |
1449 | Context *on_all_applied, | |
1450 | Context *on_all_commit, | |
1451 | ceph_tid_t tid, | |
1452 | osd_reqid_t reqid, | |
1453 | OpRequestRef client_op | |
1454 | ) | |
1455 | { | |
1456 | assert(!tid_to_op_map.count(tid)); | |
1457 | Op *op = &(tid_to_op_map[tid]); | |
1458 | op->hoid = hoid; | |
1459 | op->delta_stats = delta_stats; | |
1460 | op->version = at_version; | |
1461 | op->trim_to = trim_to; | |
1462 | op->roll_forward_to = MAX(roll_forward_to, committed_to); | |
1463 | op->log_entries = log_entries; | |
1464 | std::swap(op->updated_hit_set_history, hset_history); | |
1465 | op->on_local_applied_sync = on_local_applied_sync; | |
1466 | op->on_all_applied = on_all_applied; | |
1467 | op->on_all_commit = on_all_commit; | |
1468 | op->tid = tid; | |
1469 | op->reqid = reqid; | |
1470 | op->client_op = client_op; | |
1471 | if (client_op) | |
1472 | op->trace = client_op->pg_trace; | |
1473 | ||
1474 | dout(10) << __func__ << ": op " << *op << " starting" << dendl; | |
1475 | start_rmw(op, std::move(t)); | |
1476 | dout(10) << "onreadable_sync: " << op->on_local_applied_sync << dendl; | |
1477 | } | |
1478 | ||
1479 | void ECBackend::call_write_ordered(std::function<void(void)> &&cb) { | |
1480 | if (!waiting_state.empty()) { | |
1481 | waiting_state.back().on_write.emplace_back(std::move(cb)); | |
1482 | } else if (!waiting_reads.empty()) { | |
1483 | waiting_reads.back().on_write.emplace_back(std::move(cb)); | |
1484 | } else { | |
1485 | // Nothing earlier in the pipeline, just call it | |
1486 | cb(); | |
1487 | } | |
1488 | } | |
1489 | ||
1490 | void ECBackend::get_all_avail_shards( | |
1491 | const hobject_t &hoid, | |
1492 | set<int> &have, | |
1493 | map<shard_id_t, pg_shard_t> &shards, | |
1494 | bool for_recovery) | |
1495 | { | |
1496 | for (set<pg_shard_t>::const_iterator i = | |
1497 | get_parent()->get_acting_shards().begin(); | |
1498 | i != get_parent()->get_acting_shards().end(); | |
1499 | ++i) { | |
1500 | dout(10) << __func__ << ": checking acting " << *i << dendl; | |
1501 | const pg_missing_t &missing = get_parent()->get_shard_missing(*i); | |
1502 | if (!missing.is_missing(hoid)) { | |
1503 | assert(!have.count(i->shard)); | |
1504 | have.insert(i->shard); | |
1505 | assert(!shards.count(i->shard)); | |
1506 | shards.insert(make_pair(i->shard, *i)); | |
1507 | } | |
1508 | } | |
1509 | ||
1510 | if (for_recovery) { | |
1511 | for (set<pg_shard_t>::const_iterator i = | |
1512 | get_parent()->get_backfill_shards().begin(); | |
1513 | i != get_parent()->get_backfill_shards().end(); | |
1514 | ++i) { | |
1515 | if (have.count(i->shard)) { | |
1516 | assert(shards.count(i->shard)); | |
1517 | continue; | |
1518 | } | |
1519 | dout(10) << __func__ << ": checking backfill " << *i << dendl; | |
1520 | assert(!shards.count(i->shard)); | |
1521 | const pg_info_t &info = get_parent()->get_shard_info(*i); | |
1522 | const pg_missing_t &missing = get_parent()->get_shard_missing(*i); | |
1523 | if (hoid < info.last_backfill && | |
1524 | !missing.is_missing(hoid)) { | |
1525 | have.insert(i->shard); | |
1526 | shards.insert(make_pair(i->shard, *i)); | |
1527 | } | |
1528 | } | |
1529 | ||
1530 | map<hobject_t, set<pg_shard_t>>::const_iterator miter = | |
1531 | get_parent()->get_missing_loc_shards().find(hoid); | |
1532 | if (miter != get_parent()->get_missing_loc_shards().end()) { | |
1533 | for (set<pg_shard_t>::iterator i = miter->second.begin(); | |
1534 | i != miter->second.end(); | |
1535 | ++i) { | |
1536 | dout(10) << __func__ << ": checking missing_loc " << *i << dendl; | |
1537 | auto m = get_parent()->maybe_get_shard_missing(*i); | |
1538 | if (m) { | |
1539 | assert(!(*m).is_missing(hoid)); | |
1540 | } | |
1541 | have.insert(i->shard); | |
1542 | shards.insert(make_pair(i->shard, *i)); | |
1543 | } | |
1544 | } | |
1545 | } | |
1546 | } | |
1547 | ||
1548 | int ECBackend::get_min_avail_to_read_shards( | |
1549 | const hobject_t &hoid, | |
1550 | const set<int> &want, | |
1551 | bool for_recovery, | |
1552 | bool do_redundant_reads, | |
1553 | set<pg_shard_t> *to_read) | |
1554 | { | |
1555 | // Make sure we don't do redundant reads for recovery | |
1556 | assert(!for_recovery || !do_redundant_reads); | |
1557 | ||
1558 | set<int> have; | |
1559 | map<shard_id_t, pg_shard_t> shards; | |
1560 | ||
1561 | get_all_avail_shards(hoid, have, shards, for_recovery); | |
1562 | ||
1563 | set<int> need; | |
1564 | int r = ec_impl->minimum_to_decode(want, have, &need); | |
1565 | if (r < 0) | |
1566 | return r; | |
1567 | ||
1568 | if (do_redundant_reads) { | |
1569 | need.swap(have); | |
1570 | } | |
1571 | ||
1572 | if (!to_read) | |
1573 | return 0; | |
1574 | ||
1575 | for (set<int>::iterator i = need.begin(); | |
1576 | i != need.end(); | |
1577 | ++i) { | |
1578 | assert(shards.count(shard_id_t(*i))); | |
1579 | to_read->insert(shards[shard_id_t(*i)]); | |
1580 | } | |
1581 | return 0; | |
1582 | } | |
1583 | ||
1584 | int ECBackend::get_remaining_shards( | |
1585 | const hobject_t &hoid, | |
1586 | const set<int> &avail, | |
1587 | set<pg_shard_t> *to_read, | |
1588 | bool for_recovery) | |
1589 | { | |
1590 | assert(to_read); | |
1591 | ||
1592 | set<int> have; | |
1593 | map<shard_id_t, pg_shard_t> shards; | |
1594 | ||
1595 | get_all_avail_shards(hoid, have, shards, for_recovery); | |
1596 | ||
1597 | for (set<int>::iterator i = have.begin(); | |
1598 | i != have.end(); | |
1599 | ++i) { | |
1600 | assert(shards.count(shard_id_t(*i))); | |
1601 | if (avail.find(*i) == avail.end()) | |
1602 | to_read->insert(shards[shard_id_t(*i)]); | |
1603 | } | |
1604 | return 0; | |
1605 | } | |
1606 | ||
1607 | void ECBackend::start_read_op( | |
1608 | int priority, | |
1609 | map<hobject_t, read_request_t> &to_read, | |
1610 | OpRequestRef _op, | |
1611 | bool do_redundant_reads, | |
1612 | bool for_recovery) | |
1613 | { | |
1614 | ceph_tid_t tid = get_parent()->get_tid(); | |
1615 | assert(!tid_to_read_map.count(tid)); | |
1616 | auto &op = tid_to_read_map.emplace( | |
1617 | tid, | |
1618 | ReadOp( | |
1619 | priority, | |
1620 | tid, | |
1621 | do_redundant_reads, | |
1622 | for_recovery, | |
1623 | _op, | |
1624 | std::move(to_read))).first->second; | |
1625 | dout(10) << __func__ << ": starting " << op << dendl; | |
1626 | if (_op) { | |
1627 | op.trace = _op->pg_trace; | |
1628 | op.trace.event("start ec read"); | |
1629 | } | |
1630 | do_read_op(op); | |
1631 | } | |
1632 | ||
1633 | void ECBackend::do_read_op(ReadOp &op) | |
1634 | { | |
1635 | int priority = op.priority; | |
1636 | ceph_tid_t tid = op.tid; | |
1637 | ||
1638 | dout(10) << __func__ << ": starting read " << op << dendl; | |
1639 | ||
1640 | map<pg_shard_t, ECSubRead> messages; | |
1641 | for (map<hobject_t, read_request_t>::iterator i = op.to_read.begin(); | |
1642 | i != op.to_read.end(); | |
1643 | ++i) { | |
1644 | bool need_attrs = i->second.want_attrs; | |
1645 | for (set<pg_shard_t>::const_iterator j = i->second.need.begin(); | |
1646 | j != i->second.need.end(); | |
1647 | ++j) { | |
1648 | if (need_attrs) { | |
1649 | messages[*j].attrs_to_read.insert(i->first); | |
1650 | need_attrs = false; | |
1651 | } | |
1652 | op.obj_to_source[i->first].insert(*j); | |
1653 | op.source_to_obj[*j].insert(i->first); | |
1654 | } | |
1655 | for (list<boost::tuple<uint64_t, uint64_t, uint32_t> >::const_iterator j = | |
1656 | i->second.to_read.begin(); | |
1657 | j != i->second.to_read.end(); | |
1658 | ++j) { | |
1659 | pair<uint64_t, uint64_t> chunk_off_len = | |
1660 | sinfo.aligned_offset_len_to_chunk(make_pair(j->get<0>(), j->get<1>())); | |
1661 | for (set<pg_shard_t>::const_iterator k = i->second.need.begin(); | |
1662 | k != i->second.need.end(); | |
1663 | ++k) { | |
1664 | messages[*k].to_read[i->first].push_back( | |
1665 | boost::make_tuple( | |
1666 | chunk_off_len.first, | |
1667 | chunk_off_len.second, | |
1668 | j->get<2>())); | |
1669 | } | |
1670 | assert(!need_attrs); | |
1671 | } | |
1672 | } | |
1673 | ||
1674 | for (map<pg_shard_t, ECSubRead>::iterator i = messages.begin(); | |
1675 | i != messages.end(); | |
1676 | ++i) { | |
1677 | op.in_progress.insert(i->first); | |
1678 | shard_to_read_map[i->first].insert(op.tid); | |
1679 | i->second.tid = tid; | |
1680 | MOSDECSubOpRead *msg = new MOSDECSubOpRead; | |
1681 | msg->set_priority(priority); | |
1682 | msg->pgid = spg_t( | |
1683 | get_parent()->whoami_spg_t().pgid, | |
1684 | i->first.shard); | |
1685 | msg->map_epoch = get_parent()->get_epoch(); | |
1686 | msg->min_epoch = get_parent()->get_interval_start_epoch(); | |
1687 | msg->op = i->second; | |
1688 | msg->op.from = get_parent()->whoami_shard(); | |
1689 | msg->op.tid = tid; | |
1690 | if (op.trace) { | |
1691 | // initialize a child span for this shard | |
1692 | msg->trace.init("ec sub read", nullptr, &op.trace); | |
1693 | msg->trace.keyval("shard", i->first.shard.id); | |
1694 | } | |
1695 | get_parent()->send_message_osd_cluster( | |
1696 | i->first.osd, | |
1697 | msg, | |
1698 | get_parent()->get_epoch()); | |
1699 | } | |
1700 | dout(10) << __func__ << ": started " << op << dendl; | |
1701 | } | |
1702 | ||
1703 | ECUtil::HashInfoRef ECBackend::get_hash_info( | |
1704 | const hobject_t &hoid, bool checks, const map<string,bufferptr> *attrs) | |
1705 | { | |
1706 | dout(10) << __func__ << ": Getting attr on " << hoid << dendl; | |
1707 | ECUtil::HashInfoRef ref = unstable_hashinfo_registry.lookup(hoid); | |
1708 | if (!ref) { | |
1709 | dout(10) << __func__ << ": not in cache " << hoid << dendl; | |
1710 | struct stat st; | |
1711 | int r = store->stat( | |
1712 | ch, | |
1713 | ghobject_t(hoid, ghobject_t::NO_GEN, get_parent()->whoami_shard().shard), | |
1714 | &st); | |
1715 | ECUtil::HashInfo hinfo(ec_impl->get_chunk_count()); | |
1716 | // XXX: What does it mean if there is no object on disk? | |
1717 | if (r >= 0) { | |
1718 | dout(10) << __func__ << ": found on disk, size " << st.st_size << dendl; | |
1719 | bufferlist bl; | |
1720 | if (attrs) { | |
1721 | map<string, bufferptr>::const_iterator k = attrs->find(ECUtil::get_hinfo_key()); | |
1722 | if (k == attrs->end()) { | |
1723 | dout(5) << __func__ << " " << hoid << " missing hinfo attr" << dendl; | |
1724 | } else { | |
1725 | bl.push_back(k->second); | |
1726 | } | |
1727 | } else { | |
1728 | r = store->getattr( | |
1729 | ch, | |
1730 | ghobject_t(hoid, ghobject_t::NO_GEN, get_parent()->whoami_shard().shard), | |
1731 | ECUtil::get_hinfo_key(), | |
1732 | bl); | |
1733 | if (r < 0) { | |
1734 | dout(5) << __func__ << ": getattr failed: " << cpp_strerror(r) << dendl; | |
1735 | bl.clear(); // just in case | |
1736 | } | |
1737 | } | |
1738 | if (bl.length() > 0) { | |
1739 | bufferlist::iterator bp = bl.begin(); | |
1740 | ::decode(hinfo, bp); | |
1741 | if (checks && hinfo.get_total_chunk_size() != (uint64_t)st.st_size) { | |
1742 | dout(0) << __func__ << ": Mismatch of total_chunk_size " | |
1743 | << hinfo.get_total_chunk_size() << dendl; | |
1744 | return ECUtil::HashInfoRef(); | |
1745 | } | |
1746 | } else if (st.st_size > 0) { // If empty object and no hinfo, create it | |
1747 | return ECUtil::HashInfoRef(); | |
1748 | } | |
1749 | } | |
1750 | ref = unstable_hashinfo_registry.lookup_or_create(hoid, hinfo); | |
1751 | } | |
1752 | return ref; | |
1753 | } | |
1754 | ||
1755 | void ECBackend::start_rmw(Op *op, PGTransactionUPtr &&t) | |
1756 | { | |
1757 | assert(op); | |
1758 | ||
1759 | op->plan = ECTransaction::get_write_plan( | |
1760 | sinfo, | |
1761 | std::move(t), | |
1762 | [&](const hobject_t &i) { | |
1763 | ECUtil::HashInfoRef ref = get_hash_info(i, false); | |
1764 | if (!ref) { | |
1765 | derr << __func__ << ": get_hash_info(" << i << ")" | |
1766 | << " returned a null pointer and there is no " | |
1767 | << " way to recover from such an error in this " | |
1768 | << " context" << dendl; | |
1769 | ceph_abort(); | |
1770 | } | |
1771 | return ref; | |
1772 | }, | |
1773 | get_parent()->get_dpp()); | |
1774 | ||
1775 | dout(10) << __func__ << ": " << *op << dendl; | |
1776 | ||
1777 | waiting_state.push_back(*op); | |
1778 | check_ops(); | |
1779 | } | |
1780 | ||
1781 | bool ECBackend::try_state_to_reads() | |
1782 | { | |
1783 | if (waiting_state.empty()) | |
1784 | return false; | |
1785 | ||
1786 | Op *op = &(waiting_state.front()); | |
1787 | if (op->requires_rmw() && pipeline_state.cache_invalid()) { | |
1788 | assert(get_parent()->get_pool().allows_ecoverwrites()); | |
1789 | dout(20) << __func__ << ": blocking " << *op | |
1790 | << " because it requires an rmw and the cache is invalid " | |
1791 | << pipeline_state | |
1792 | << dendl; | |
1793 | return false; | |
1794 | } | |
1795 | ||
1796 | if (op->invalidates_cache()) { | |
1797 | dout(20) << __func__ << ": invalidating cache after this op" | |
1798 | << dendl; | |
1799 | pipeline_state.invalidate(); | |
1800 | op->using_cache = false; | |
1801 | } else { | |
1802 | op->using_cache = pipeline_state.caching_enabled(); | |
1803 | } | |
1804 | ||
1805 | waiting_state.pop_front(); | |
1806 | waiting_reads.push_back(*op); | |
1807 | ||
1808 | if (op->using_cache) { | |
1809 | cache.open_write_pin(op->pin); | |
1810 | ||
1811 | extent_set empty; | |
1812 | for (auto &&hpair: op->plan.will_write) { | |
1813 | auto to_read_plan_iter = op->plan.to_read.find(hpair.first); | |
1814 | const extent_set &to_read_plan = | |
1815 | to_read_plan_iter == op->plan.to_read.end() ? | |
1816 | empty : | |
1817 | to_read_plan_iter->second; | |
1818 | ||
1819 | extent_set remote_read = cache.reserve_extents_for_rmw( | |
1820 | hpair.first, | |
1821 | op->pin, | |
1822 | hpair.second, | |
1823 | to_read_plan); | |
1824 | ||
1825 | extent_set pending_read = to_read_plan; | |
1826 | pending_read.subtract(remote_read); | |
1827 | ||
1828 | if (!remote_read.empty()) { | |
1829 | op->remote_read[hpair.first] = std::move(remote_read); | |
1830 | } | |
1831 | if (!pending_read.empty()) { | |
1832 | op->pending_read[hpair.first] = std::move(pending_read); | |
1833 | } | |
1834 | } | |
1835 | } else { | |
1836 | op->remote_read = op->plan.to_read; | |
1837 | } | |
1838 | ||
1839 | dout(10) << __func__ << ": " << *op << dendl; | |
1840 | ||
1841 | if (!op->remote_read.empty()) { | |
1842 | assert(get_parent()->get_pool().allows_ecoverwrites()); | |
1843 | objects_read_async_no_cache( | |
1844 | op->remote_read, | |
1845 | [this, op](map<hobject_t,pair<int, extent_map> > &&results) { | |
1846 | for (auto &&i: results) { | |
1847 | op->remote_read_result.emplace(i.first, i.second.second); | |
1848 | } | |
1849 | check_ops(); | |
1850 | }); | |
1851 | } | |
1852 | ||
1853 | return true; | |
1854 | } | |
1855 | ||
1856 | bool ECBackend::try_reads_to_commit() | |
1857 | { | |
1858 | if (waiting_reads.empty()) | |
1859 | return false; | |
1860 | Op *op = &(waiting_reads.front()); | |
1861 | if (op->read_in_progress()) | |
1862 | return false; | |
1863 | waiting_reads.pop_front(); | |
1864 | waiting_commit.push_back(*op); | |
1865 | ||
1866 | dout(10) << __func__ << ": starting commit on " << *op << dendl; | |
1867 | dout(20) << __func__ << ": " << cache << dendl; | |
1868 | ||
1869 | get_parent()->apply_stats( | |
1870 | op->hoid, | |
1871 | op->delta_stats); | |
1872 | ||
1873 | if (op->using_cache) { | |
1874 | for (auto &&hpair: op->pending_read) { | |
1875 | op->remote_read_result[hpair.first].insert( | |
1876 | cache.get_remaining_extents_for_rmw( | |
1877 | hpair.first, | |
1878 | op->pin, | |
1879 | hpair.second)); | |
1880 | } | |
1881 | op->pending_read.clear(); | |
1882 | } else { | |
1883 | assert(op->pending_read.empty()); | |
1884 | } | |
1885 | ||
1886 | map<shard_id_t, ObjectStore::Transaction> trans; | |
1887 | for (set<pg_shard_t>::const_iterator i = | |
1888 | get_parent()->get_actingbackfill_shards().begin(); | |
1889 | i != get_parent()->get_actingbackfill_shards().end(); | |
1890 | ++i) { | |
1891 | trans[i->shard]; | |
1892 | } | |
1893 | ||
1894 | op->trace.event("start ec write"); | |
1895 | ||
1896 | map<hobject_t,extent_map> written; | |
1897 | if (op->plan.t) { | |
1898 | ECTransaction::generate_transactions( | |
1899 | op->plan, | |
1900 | ec_impl, | |
1901 | get_parent()->get_info().pgid.pgid, | |
1902 | (get_osdmap()->require_osd_release < CEPH_RELEASE_KRAKEN), | |
1903 | sinfo, | |
1904 | op->remote_read_result, | |
1905 | op->log_entries, | |
1906 | &written, | |
1907 | &trans, | |
1908 | &(op->temp_added), | |
1909 | &(op->temp_cleared), | |
1910 | get_parent()->get_dpp()); | |
1911 | } | |
1912 | ||
1913 | dout(20) << __func__ << ": " << cache << dendl; | |
1914 | dout(20) << __func__ << ": written: " << written << dendl; | |
1915 | dout(20) << __func__ << ": op: " << *op << dendl; | |
1916 | ||
1917 | if (!get_parent()->get_pool().allows_ecoverwrites()) { | |
1918 | for (auto &&i: op->log_entries) { | |
1919 | if (i.requires_kraken()) { | |
1920 | derr << __func__ << ": log entry " << i << " requires kraken" | |
1921 | << " but overwrites are not enabled!" << dendl; | |
1922 | ceph_abort(); | |
1923 | } | |
1924 | } | |
1925 | } | |
1926 | ||
1927 | map<hobject_t,extent_set> written_set; | |
1928 | for (auto &&i: written) { | |
1929 | written_set[i.first] = i.second.get_interval_set(); | |
1930 | } | |
1931 | dout(20) << __func__ << ": written_set: " << written_set << dendl; | |
1932 | assert(written_set == op->plan.will_write); | |
1933 | ||
1934 | if (op->using_cache) { | |
1935 | for (auto &&hpair: written) { | |
1936 | dout(20) << __func__ << ": " << hpair << dendl; | |
1937 | cache.present_rmw_update(hpair.first, op->pin, hpair.second); | |
1938 | } | |
1939 | } | |
1940 | op->remote_read.clear(); | |
1941 | op->remote_read_result.clear(); | |
1942 | ||
1943 | dout(10) << "onreadable_sync: " << op->on_local_applied_sync << dendl; | |
1944 | ObjectStore::Transaction empty; | |
1945 | bool should_write_local = false; | |
1946 | ECSubWrite local_write_op; | |
1947 | for (set<pg_shard_t>::const_iterator i = | |
1948 | get_parent()->get_actingbackfill_shards().begin(); | |
1949 | i != get_parent()->get_actingbackfill_shards().end(); | |
1950 | ++i) { | |
1951 | op->pending_apply.insert(*i); | |
1952 | op->pending_commit.insert(*i); | |
1953 | map<shard_id_t, ObjectStore::Transaction>::iterator iter = | |
1954 | trans.find(i->shard); | |
1955 | assert(iter != trans.end()); | |
1956 | bool should_send = get_parent()->should_send_op(*i, op->hoid); | |
1957 | const pg_stat_t &stats = | |
1958 | should_send ? | |
1959 | get_info().stats : | |
1960 | parent->get_shard_info().find(*i)->second.stats; | |
1961 | ||
1962 | ECSubWrite sop( | |
1963 | get_parent()->whoami_shard(), | |
1964 | op->tid, | |
1965 | op->reqid, | |
1966 | op->hoid, | |
1967 | stats, | |
1968 | should_send ? iter->second : empty, | |
1969 | op->version, | |
1970 | op->trim_to, | |
1971 | op->roll_forward_to, | |
1972 | op->log_entries, | |
1973 | op->updated_hit_set_history, | |
1974 | op->temp_added, | |
1975 | op->temp_cleared, | |
1976 | !should_send); | |
1977 | ||
1978 | ZTracer::Trace trace; | |
1979 | if (op->trace) { | |
1980 | // initialize a child span for this shard | |
1981 | trace.init("ec sub write", nullptr, &op->trace); | |
1982 | trace.keyval("shard", i->shard.id); | |
1983 | } | |
1984 | ||
1985 | if (*i == get_parent()->whoami_shard()) { | |
1986 | should_write_local = true; | |
1987 | local_write_op.claim(sop); | |
1988 | } else { | |
1989 | MOSDECSubOpWrite *r = new MOSDECSubOpWrite(sop); | |
1990 | r->pgid = spg_t(get_parent()->primary_spg_t().pgid, i->shard); | |
1991 | r->map_epoch = get_parent()->get_epoch(); | |
1992 | r->min_epoch = get_parent()->get_interval_start_epoch(); | |
1993 | r->trace = trace; | |
1994 | get_parent()->send_message_osd_cluster( | |
1995 | i->osd, r, get_parent()->get_epoch()); | |
1996 | } | |
1997 | } | |
1998 | if (should_write_local) { | |
1999 | handle_sub_write( | |
2000 | get_parent()->whoami_shard(), | |
2001 | op->client_op, | |
2002 | local_write_op, | |
2003 | op->trace, | |
2004 | op->on_local_applied_sync); | |
2005 | op->on_local_applied_sync = 0; | |
2006 | } | |
2007 | ||
2008 | for (auto i = op->on_write.begin(); | |
2009 | i != op->on_write.end(); | |
2010 | op->on_write.erase(i++)) { | |
2011 | (*i)(); | |
2012 | } | |
2013 | ||
2014 | return true; | |
2015 | } | |
2016 | ||
2017 | bool ECBackend::try_finish_rmw() | |
2018 | { | |
2019 | if (waiting_commit.empty()) | |
2020 | return false; | |
2021 | Op *op = &(waiting_commit.front()); | |
2022 | if (op->write_in_progress()) | |
2023 | return false; | |
2024 | waiting_commit.pop_front(); | |
2025 | ||
2026 | dout(10) << __func__ << ": " << *op << dendl; | |
2027 | dout(20) << __func__ << ": " << cache << dendl; | |
2028 | ||
2029 | if (op->roll_forward_to > completed_to) | |
2030 | completed_to = op->roll_forward_to; | |
2031 | if (op->version > committed_to) | |
2032 | committed_to = op->version; | |
2033 | ||
2034 | if (get_osdmap()->require_osd_release >= CEPH_RELEASE_KRAKEN) { | |
2035 | if (op->version > get_parent()->get_log().get_can_rollback_to() && | |
2036 | waiting_reads.empty() && | |
2037 | waiting_commit.empty()) { | |
2038 | // submit a dummy transaction to kick the rollforward | |
2039 | auto tid = get_parent()->get_tid(); | |
2040 | Op *nop = &(tid_to_op_map[tid]); | |
2041 | nop->hoid = op->hoid; | |
2042 | nop->trim_to = op->trim_to; | |
2043 | nop->roll_forward_to = op->version; | |
2044 | nop->tid = tid; | |
2045 | nop->reqid = op->reqid; | |
2046 | waiting_reads.push_back(*nop); | |
2047 | } | |
2048 | } | |
2049 | ||
2050 | if (op->using_cache) { | |
2051 | cache.release_write_pin(op->pin); | |
2052 | } | |
2053 | tid_to_op_map.erase(op->tid); | |
2054 | ||
2055 | if (waiting_reads.empty() && | |
2056 | waiting_commit.empty()) { | |
2057 | pipeline_state.clear(); | |
2058 | dout(20) << __func__ << ": clearing pipeline_state " | |
2059 | << pipeline_state | |
2060 | << dendl; | |
2061 | } | |
2062 | return true; | |
2063 | } | |
2064 | ||
2065 | void ECBackend::check_ops() | |
2066 | { | |
2067 | while (try_state_to_reads() || | |
2068 | try_reads_to_commit() || | |
2069 | try_finish_rmw()); | |
2070 | } | |
2071 | ||
2072 | int ECBackend::objects_read_sync( | |
2073 | const hobject_t &hoid, | |
2074 | uint64_t off, | |
2075 | uint64_t len, | |
2076 | uint32_t op_flags, | |
2077 | bufferlist *bl) | |
2078 | { | |
2079 | return -EOPNOTSUPP; | |
2080 | } | |
2081 | ||
2082 | void ECBackend::objects_read_async( | |
2083 | const hobject_t &hoid, | |
2084 | const list<pair<boost::tuple<uint64_t, uint64_t, uint32_t>, | |
2085 | pair<bufferlist*, Context*> > > &to_read, | |
2086 | Context *on_complete, | |
2087 | bool fast_read) | |
2088 | { | |
2089 | map<hobject_t,std::list<boost::tuple<uint64_t, uint64_t, uint32_t> > > | |
2090 | reads; | |
2091 | ||
2092 | uint32_t flags = 0; | |
2093 | extent_set es; | |
2094 | for (list<pair<boost::tuple<uint64_t, uint64_t, uint32_t>, | |
2095 | pair<bufferlist*, Context*> > >::const_iterator i = | |
2096 | to_read.begin(); | |
2097 | i != to_read.end(); | |
2098 | ++i) { | |
2099 | pair<uint64_t, uint64_t> tmp = | |
2100 | sinfo.offset_len_to_stripe_bounds( | |
2101 | make_pair(i->first.get<0>(), i->first.get<1>())); | |
2102 | ||
2103 | extent_set esnew; | |
2104 | esnew.insert(tmp.first, tmp.second); | |
2105 | es.union_of(esnew); | |
2106 | flags |= i->first.get<2>(); | |
2107 | } | |
2108 | ||
2109 | if (!es.empty()) { | |
2110 | auto &offsets = reads[hoid]; | |
2111 | for (auto j = es.begin(); | |
2112 | j != es.end(); | |
2113 | ++j) { | |
2114 | offsets.push_back( | |
2115 | boost::make_tuple( | |
2116 | j.get_start(), | |
2117 | j.get_len(), | |
2118 | flags)); | |
2119 | } | |
2120 | } | |
2121 | ||
2122 | struct cb { | |
2123 | ECBackend *ec; | |
2124 | hobject_t hoid; | |
2125 | list<pair<boost::tuple<uint64_t, uint64_t, uint32_t>, | |
2126 | pair<bufferlist*, Context*> > > to_read; | |
2127 | unique_ptr<Context> on_complete; | |
2128 | cb(const cb&) = delete; | |
2129 | cb(cb &&) = default; | |
2130 | cb(ECBackend *ec, | |
2131 | const hobject_t &hoid, | |
2132 | const list<pair<boost::tuple<uint64_t, uint64_t, uint32_t>, | |
2133 | pair<bufferlist*, Context*> > > &to_read, | |
2134 | Context *on_complete) | |
2135 | : ec(ec), | |
2136 | hoid(hoid), | |
2137 | to_read(to_read), | |
2138 | on_complete(on_complete) {} | |
2139 | void operator()(map<hobject_t,pair<int, extent_map> > &&results) { | |
2140 | auto dpp = ec->get_parent()->get_dpp(); | |
2141 | ldpp_dout(dpp, 20) << "objects_read_async_cb: got: " << results | |
2142 | << dendl; | |
2143 | ldpp_dout(dpp, 20) << "objects_read_async_cb: cache: " << ec->cache | |
2144 | << dendl; | |
2145 | ||
2146 | auto &got = results[hoid]; | |
2147 | ||
2148 | int r = 0; | |
2149 | for (auto &&read: to_read) { | |
2150 | if (got.first < 0) { | |
2151 | if (read.second.second) { | |
2152 | read.second.second->complete(got.first); | |
2153 | } | |
2154 | if (r == 0) | |
2155 | r = got.first; | |
2156 | } else { | |
2157 | assert(read.second.first); | |
2158 | uint64_t offset = read.first.get<0>(); | |
2159 | uint64_t length = read.first.get<1>(); | |
2160 | auto range = got.second.get_containing_range(offset, length); | |
2161 | assert(range.first != range.second); | |
2162 | assert(range.first.get_off() <= offset); | |
2163 | assert( | |
2164 | (offset + length) <= | |
2165 | (range.first.get_off() + range.first.get_len())); | |
2166 | read.second.first->substr_of( | |
2167 | range.first.get_val(), | |
2168 | offset - range.first.get_off(), | |
2169 | length); | |
2170 | if (read.second.second) { | |
2171 | read.second.second->complete(length); | |
2172 | read.second.second = nullptr; | |
2173 | } | |
2174 | } | |
2175 | } | |
2176 | to_read.clear(); | |
2177 | if (on_complete) { | |
2178 | on_complete.release()->complete(r); | |
2179 | } | |
2180 | } | |
2181 | ~cb() { | |
2182 | for (auto &&i: to_read) { | |
2183 | delete i.second.second; | |
2184 | } | |
2185 | to_read.clear(); | |
2186 | } | |
2187 | }; | |
2188 | objects_read_and_reconstruct( | |
2189 | reads, | |
2190 | fast_read, | |
2191 | make_gen_lambda_context< | |
2192 | map<hobject_t,pair<int, extent_map> > &&, cb>( | |
2193 | cb(this, | |
2194 | hoid, | |
2195 | to_read, | |
2196 | on_complete))); | |
2197 | } | |
2198 | ||
2199 | struct CallClientContexts : | |
2200 | public GenContext<pair<RecoveryMessages*, ECBackend::read_result_t& > &> { | |
2201 | hobject_t hoid; | |
2202 | ECBackend *ec; | |
2203 | ECBackend::ClientAsyncReadStatus *status; | |
2204 | list<boost::tuple<uint64_t, uint64_t, uint32_t> > to_read; | |
2205 | CallClientContexts( | |
2206 | hobject_t hoid, | |
2207 | ECBackend *ec, | |
2208 | ECBackend::ClientAsyncReadStatus *status, | |
2209 | const list<boost::tuple<uint64_t, uint64_t, uint32_t> > &to_read) | |
2210 | : hoid(hoid), ec(ec), status(status), to_read(to_read) {} | |
2211 | void finish(pair<RecoveryMessages *, ECBackend::read_result_t &> &in) override { | |
2212 | ECBackend::read_result_t &res = in.second; | |
2213 | extent_map result; | |
2214 | if (res.r != 0) | |
2215 | goto out; | |
2216 | assert(res.returned.size() == to_read.size()); | |
2217 | assert(res.r == 0); | |
2218 | assert(res.errors.empty()); | |
2219 | for (auto &&read: to_read) { | |
2220 | pair<uint64_t, uint64_t> adjusted = | |
2221 | ec->sinfo.offset_len_to_stripe_bounds( | |
2222 | make_pair(read.get<0>(), read.get<1>())); | |
2223 | assert(res.returned.front().get<0>() == adjusted.first && | |
2224 | res.returned.front().get<1>() == adjusted.second); | |
2225 | map<int, bufferlist> to_decode; | |
2226 | bufferlist bl; | |
2227 | for (map<pg_shard_t, bufferlist>::iterator j = | |
2228 | res.returned.front().get<2>().begin(); | |
2229 | j != res.returned.front().get<2>().end(); | |
2230 | ++j) { | |
2231 | to_decode[j->first.shard].claim(j->second); | |
2232 | } | |
2233 | int r = ECUtil::decode( | |
2234 | ec->sinfo, | |
2235 | ec->ec_impl, | |
2236 | to_decode, | |
2237 | &bl); | |
2238 | if (r < 0) { | |
2239 | res.r = r; | |
2240 | goto out; | |
2241 | } | |
2242 | bufferlist trimmed; | |
2243 | trimmed.substr_of( | |
2244 | bl, | |
2245 | read.get<0>() - adjusted.first, | |
2246 | MIN(read.get<1>(), | |
2247 | bl.length() - (read.get<0>() - adjusted.first))); | |
2248 | result.insert( | |
2249 | read.get<0>(), trimmed.length(), std::move(trimmed)); | |
2250 | res.returned.pop_front(); | |
2251 | } | |
2252 | out: | |
2253 | status->complete_object(hoid, res.r, std::move(result)); | |
2254 | ec->kick_reads(); | |
2255 | } | |
2256 | }; | |
2257 | ||
2258 | void ECBackend::objects_read_and_reconstruct( | |
2259 | const map<hobject_t, | |
2260 | std::list<boost::tuple<uint64_t, uint64_t, uint32_t> > | |
2261 | > &reads, | |
2262 | bool fast_read, | |
2263 | GenContextURef<map<hobject_t,pair<int, extent_map> > &&> &&func) | |
2264 | { | |
2265 | in_progress_client_reads.emplace_back( | |
2266 | reads.size(), std::move(func)); | |
2267 | if (!reads.size()) { | |
2268 | kick_reads(); | |
2269 | return; | |
2270 | } | |
2271 | ||
2272 | set<int> want_to_read; | |
2273 | get_want_to_read_shards(&want_to_read); | |
2274 | ||
2275 | map<hobject_t, read_request_t> for_read_op; | |
2276 | for (auto &&to_read: reads) { | |
2277 | set<pg_shard_t> shards; | |
2278 | int r = get_min_avail_to_read_shards( | |
2279 | to_read.first, | |
2280 | want_to_read, | |
2281 | false, | |
2282 | fast_read, | |
2283 | &shards); | |
2284 | assert(r == 0); | |
2285 | ||
2286 | CallClientContexts *c = new CallClientContexts( | |
2287 | to_read.first, | |
2288 | this, | |
2289 | &(in_progress_client_reads.back()), | |
2290 | to_read.second); | |
2291 | for_read_op.insert( | |
2292 | make_pair( | |
2293 | to_read.first, | |
2294 | read_request_t( | |
2295 | to_read.second, | |
2296 | shards, | |
2297 | false, | |
2298 | c))); | |
2299 | } | |
2300 | ||
2301 | start_read_op( | |
2302 | CEPH_MSG_PRIO_DEFAULT, | |
2303 | for_read_op, | |
2304 | OpRequestRef(), | |
2305 | fast_read, false); | |
2306 | return; | |
2307 | } | |
2308 | ||
2309 | ||
2310 | int ECBackend::send_all_remaining_reads( | |
2311 | const hobject_t &hoid, | |
2312 | ReadOp &rop) | |
2313 | { | |
2314 | set<int> already_read; | |
2315 | const set<pg_shard_t>& ots = rop.obj_to_source[hoid]; | |
2316 | for (set<pg_shard_t>::iterator i = ots.begin(); i != ots.end(); ++i) | |
2317 | already_read.insert(i->shard); | |
2318 | dout(10) << __func__ << " have/error shards=" << already_read << dendl; | |
2319 | set<pg_shard_t> shards; | |
2320 | int r = get_remaining_shards(hoid, already_read, &shards, rop.for_recovery); | |
2321 | if (r) | |
2322 | return r; | |
2323 | if (shards.empty()) | |
2324 | return -EIO; | |
2325 | ||
2326 | dout(10) << __func__ << " Read remaining shards " << shards << dendl; | |
2327 | ||
2328 | // TODOSAM: this doesn't seem right | |
2329 | list<boost::tuple<uint64_t, uint64_t, uint32_t> > offsets = | |
2330 | rop.to_read.find(hoid)->second.to_read; | |
2331 | GenContext<pair<RecoveryMessages *, read_result_t& > &> *c = | |
2332 | rop.to_read.find(hoid)->second.cb; | |
2333 | ||
2334 | map<hobject_t, read_request_t> for_read_op; | |
2335 | for_read_op.insert( | |
2336 | make_pair( | |
2337 | hoid, | |
2338 | read_request_t( | |
2339 | offsets, | |
2340 | shards, | |
2341 | false, | |
2342 | c))); | |
2343 | ||
2344 | rop.to_read.swap(for_read_op); | |
2345 | do_read_op(rop); | |
2346 | return 0; | |
2347 | } | |
2348 | ||
2349 | int ECBackend::objects_get_attrs( | |
2350 | const hobject_t &hoid, | |
2351 | map<string, bufferlist> *out) | |
2352 | { | |
2353 | int r = store->getattrs( | |
2354 | ch, | |
2355 | ghobject_t(hoid, ghobject_t::NO_GEN, get_parent()->whoami_shard().shard), | |
2356 | *out); | |
2357 | if (r < 0) | |
2358 | return r; | |
2359 | ||
2360 | for (map<string, bufferlist>::iterator i = out->begin(); | |
2361 | i != out->end(); | |
2362 | ) { | |
2363 | if (ECUtil::is_hinfo_key_string(i->first)) | |
2364 | out->erase(i++); | |
2365 | else | |
2366 | ++i; | |
2367 | } | |
2368 | return r; | |
2369 | } | |
2370 | ||
2371 | void ECBackend::rollback_append( | |
2372 | const hobject_t &hoid, | |
2373 | uint64_t old_size, | |
2374 | ObjectStore::Transaction *t) | |
2375 | { | |
2376 | assert(old_size % sinfo.get_stripe_width() == 0); | |
2377 | t->truncate( | |
2378 | coll, | |
2379 | ghobject_t(hoid, ghobject_t::NO_GEN, get_parent()->whoami_shard().shard), | |
2380 | sinfo.aligned_logical_offset_to_chunk_offset( | |
2381 | old_size)); | |
2382 | } | |
2383 | ||
2384 | void ECBackend::be_deep_scrub( | |
2385 | const hobject_t &poid, | |
2386 | uint32_t seed, | |
2387 | ScrubMap::object &o, | |
2388 | ThreadPool::TPHandle &handle) { | |
2389 | bufferhash h(-1); // we always used -1 | |
2390 | int r; | |
2391 | uint64_t stride = cct->_conf->osd_deep_scrub_stride; | |
2392 | if (stride % sinfo.get_chunk_size()) | |
2393 | stride += sinfo.get_chunk_size() - (stride % sinfo.get_chunk_size()); | |
2394 | uint64_t pos = 0; | |
2395 | ||
2396 | uint32_t fadvise_flags = CEPH_OSD_OP_FLAG_FADVISE_SEQUENTIAL | CEPH_OSD_OP_FLAG_FADVISE_DONTNEED; | |
2397 | ||
2398 | while (true) { | |
2399 | bufferlist bl; | |
2400 | handle.reset_tp_timeout(); | |
2401 | r = store->read( | |
2402 | ch, | |
2403 | ghobject_t( | |
2404 | poid, ghobject_t::NO_GEN, get_parent()->whoami_shard().shard), | |
2405 | pos, | |
2406 | stride, bl, | |
2407 | fadvise_flags); | |
2408 | if (r < 0) | |
2409 | break; | |
2410 | if (bl.length() % sinfo.get_chunk_size()) { | |
2411 | r = -EIO; | |
2412 | break; | |
2413 | } | |
2414 | pos += r; | |
2415 | h << bl; | |
2416 | if ((unsigned)r < stride) | |
2417 | break; | |
2418 | } | |
2419 | ||
2420 | if (r == -EIO) { | |
2421 | dout(0) << "_scan_list " << poid << " got " | |
2422 | << r << " on read, read_error" << dendl; | |
2423 | o.read_error = true; | |
2424 | return; | |
2425 | } | |
2426 | ||
2427 | ECUtil::HashInfoRef hinfo = get_hash_info(poid, false, &o.attrs); | |
2428 | if (!hinfo) { | |
2429 | dout(0) << "_scan_list " << poid << " could not retrieve hash info" << dendl; | |
2430 | o.read_error = true; | |
2431 | o.digest_present = false; | |
2432 | return; | |
2433 | } else { | |
2434 | if (!get_parent()->get_pool().allows_ecoverwrites()) { | |
2435 | assert(hinfo->has_chunk_hash()); | |
2436 | if (hinfo->get_total_chunk_size() != pos) { | |
2437 | dout(0) << "_scan_list " << poid << " got incorrect size on read" << dendl; | |
2438 | o.ec_size_mismatch = true; | |
2439 | return; | |
2440 | } | |
2441 | ||
2442 | if (hinfo->get_chunk_hash(get_parent()->whoami_shard().shard) != h.digest()) { | |
2443 | dout(0) << "_scan_list " << poid << " got incorrect hash on read" << dendl; | |
2444 | o.ec_hash_mismatch = true; | |
2445 | return; | |
2446 | } | |
2447 | ||
2448 | /* We checked above that we match our own stored hash. We cannot | |
2449 | * send a hash of the actual object, so instead we simply send | |
2450 | * our locally stored hash of shard 0 on the assumption that if | |
2451 | * we match our chunk hash and our recollection of the hash for | |
2452 | * chunk 0 matches that of our peers, there is likely no corruption. | |
2453 | */ | |
2454 | o.digest = hinfo->get_chunk_hash(0); | |
2455 | o.digest_present = true; | |
2456 | } else { | |
2457 | /* Hack! We must be using partial overwrites, and partial overwrites | |
2458 | * don't support deep-scrub yet | |
2459 | */ | |
2460 | o.digest = 0; | |
2461 | o.digest_present = true; | |
2462 | } | |
2463 | } | |
2464 | ||
2465 | o.omap_digest = seed; | |
2466 | o.omap_digest_present = true; | |
2467 | } |