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.
19 #include "osd/OSDMap.h"
20 #include "osd/error_code.h"
23 #include "mon/MonClient.h"
24 #include "mon/error_code.h"
26 #include "msg/Messenger.h"
27 #include "msg/Message.h"
29 #include "messages/MPing.h"
30 #include "messages/MOSDOp.h"
31 #include "messages/MOSDOpReply.h"
32 #include "messages/MOSDBackoff.h"
33 #include "messages/MOSDMap.h"
35 #include "messages/MPoolOp.h"
36 #include "messages/MPoolOpReply.h"
38 #include "messages/MGetPoolStats.h"
39 #include "messages/MGetPoolStatsReply.h"
40 #include "messages/MStatfs.h"
41 #include "messages/MStatfsReply.h"
43 #include "messages/MMonCommand.h"
45 #include "messages/MCommand.h"
46 #include "messages/MCommandReply.h"
48 #include "messages/MWatchNotify.h"
51 #include "common/Cond.h"
52 #include "common/config.h"
53 #include "common/perf_counters.h"
54 #include "common/scrub_types.h"
55 #include "include/str_list.h"
56 #include "common/errno.h"
57 #include "common/EventTrace.h"
58 #include "common/async/waiter.h"
59 #include "error_code.h"
66 using std::ostringstream
;
70 using std::stringstream
;
75 using ceph::Formatter
;
77 using std::defer_lock
;
79 using ceph::real_time
;
80 using ceph::real_clock
;
82 using ceph::mono_clock
;
83 using ceph::mono_time
;
87 using ceph::shunique_lock
;
88 using ceph::acquire_shared
;
89 using ceph::acquire_unique
;
91 namespace bc
= boost::container
;
92 namespace bs
= boost::system
;
93 namespace ca
= ceph::async
;
94 namespace cb
= ceph::buffer
;
96 #define dout_subsys ceph_subsys_objecter
98 #define dout_prefix *_dout << messenger->get_myname() << ".objecter "
102 l_osdc_first
= 123200,
106 l_osdc_op_send_bytes
,
121 l_osdc_osdop_writefull
,
122 l_osdc_osdop_writesame
,
125 l_osdc_osdop_truncate
,
128 l_osdc_osdop_sparse_read
,
129 l_osdc_osdop_clonerange
,
130 l_osdc_osdop_getxattr
,
131 l_osdc_osdop_setxattr
,
132 l_osdc_osdop_cmpxattr
,
133 l_osdc_osdop_rmxattr
,
134 l_osdc_osdop_resetxattrs
,
138 l_osdc_osdop_src_cmpxattr
,
140 l_osdc_osdop_pgls_filter
,
143 l_osdc_linger_active
,
145 l_osdc_linger_resend
,
148 l_osdc_poolop_active
,
150 l_osdc_poolop_resend
,
152 l_osdc_poolstat_active
,
153 l_osdc_poolstat_send
,
154 l_osdc_poolstat_resend
,
156 l_osdc_statfs_active
,
158 l_osdc_statfs_resend
,
160 l_osdc_command_active
,
162 l_osdc_command_resend
,
169 l_osdc_osd_session_open
,
170 l_osdc_osd_session_close
,
173 l_osdc_osdop_omap_wr
,
174 l_osdc_osdop_omap_rd
,
175 l_osdc_osdop_omap_del
,
181 inline bs::error_code
osdcode(int r
) {
182 return (r
< 0) ? bs::error_code(-r
, osd_category()) : bs::error_code();
186 // config obs ----------------------------
188 class Objecter::RequestStateHook
: public AdminSocketHook
{
189 Objecter
*m_objecter
;
191 explicit RequestStateHook(Objecter
*objecter
);
192 int call(std::string_view command
, const cmdmap_t
& cmdmap
,
195 cb::list
& out
) override
;
198 std::unique_lock
<std::mutex
> Objecter::OSDSession::get_lock(object_t
& oid
)
200 if (oid
.name
.empty())
203 static constexpr uint32_t HASH_PRIME
= 1021;
204 uint32_t h
= ceph_str_hash_linux(oid
.name
.c_str(), oid
.name
.size())
207 return {completion_locks
[h
% num_locks
], std::defer_lock
};
210 const char** Objecter::get_tracked_conf_keys() const
212 static const char *config_keys
[] = {
214 "rados_mon_op_timeout",
215 "rados_osd_op_timeout",
222 void Objecter::handle_conf_change(const ConfigProxy
& conf
,
223 const std::set
<std::string
> &changed
)
225 if (changed
.count("crush_location")) {
226 update_crush_location();
228 if (changed
.count("rados_mon_op_timeout")) {
229 mon_timeout
= conf
.get_val
<std::chrono::seconds
>("rados_mon_op_timeout");
231 if (changed
.count("rados_osd_op_timeout")) {
232 osd_timeout
= conf
.get_val
<std::chrono::seconds
>("rados_osd_op_timeout");
236 void Objecter::update_crush_location()
238 unique_lock
wl(rwlock
);
239 crush_location
= cct
->crush_location
.get_location();
242 // messages ------------------------------
245 * initialize only internal data structures, don't initiate cluster interaction
247 void Objecter::init()
249 ceph_assert(!initialized
);
252 PerfCountersBuilder
pcb(cct
, "objecter", l_osdc_first
, l_osdc_last
);
254 pcb
.add_u64(l_osdc_op_active
, "op_active", "Operations active", "actv",
255 PerfCountersBuilder::PRIO_CRITICAL
);
256 pcb
.add_u64(l_osdc_op_laggy
, "op_laggy", "Laggy operations");
257 pcb
.add_u64_counter(l_osdc_op_send
, "op_send", "Sent operations");
258 pcb
.add_u64_counter(l_osdc_op_send_bytes
, "op_send_bytes", "Sent data", NULL
, 0, unit_t(UNIT_BYTES
));
259 pcb
.add_u64_counter(l_osdc_op_resend
, "op_resend", "Resent operations");
260 pcb
.add_u64_counter(l_osdc_op_reply
, "op_reply", "Operation reply");
261 pcb
.add_u64_avg(l_osdc_oplen_avg
, "oplen_avg", "Average length of operation vector");
263 pcb
.add_u64_counter(l_osdc_op
, "op", "Operations");
264 pcb
.add_u64_counter(l_osdc_op_r
, "op_r", "Read operations", "rd",
265 PerfCountersBuilder::PRIO_CRITICAL
);
266 pcb
.add_u64_counter(l_osdc_op_w
, "op_w", "Write operations", "wr",
267 PerfCountersBuilder::PRIO_CRITICAL
);
268 pcb
.add_u64_counter(l_osdc_op_rmw
, "op_rmw", "Read-modify-write operations",
269 "rdwr", PerfCountersBuilder::PRIO_INTERESTING
);
270 pcb
.add_u64_counter(l_osdc_op_pg
, "op_pg", "PG operation");
272 pcb
.add_u64_counter(l_osdc_osdop_stat
, "osdop_stat", "Stat operations");
273 pcb
.add_u64_counter(l_osdc_osdop_create
, "osdop_create",
274 "Create object operations");
275 pcb
.add_u64_counter(l_osdc_osdop_read
, "osdop_read", "Read operations");
276 pcb
.add_u64_counter(l_osdc_osdop_write
, "osdop_write", "Write operations");
277 pcb
.add_u64_counter(l_osdc_osdop_writefull
, "osdop_writefull",
278 "Write full object operations");
279 pcb
.add_u64_counter(l_osdc_osdop_writesame
, "osdop_writesame",
280 "Write same operations");
281 pcb
.add_u64_counter(l_osdc_osdop_append
, "osdop_append",
283 pcb
.add_u64_counter(l_osdc_osdop_zero
, "osdop_zero",
284 "Set object to zero operations");
285 pcb
.add_u64_counter(l_osdc_osdop_truncate
, "osdop_truncate",
286 "Truncate object operations");
287 pcb
.add_u64_counter(l_osdc_osdop_delete
, "osdop_delete",
288 "Delete object operations");
289 pcb
.add_u64_counter(l_osdc_osdop_mapext
, "osdop_mapext",
290 "Map extent operations");
291 pcb
.add_u64_counter(l_osdc_osdop_sparse_read
, "osdop_sparse_read",
292 "Sparse read operations");
293 pcb
.add_u64_counter(l_osdc_osdop_clonerange
, "osdop_clonerange",
294 "Clone range operations");
295 pcb
.add_u64_counter(l_osdc_osdop_getxattr
, "osdop_getxattr",
296 "Get xattr operations");
297 pcb
.add_u64_counter(l_osdc_osdop_setxattr
, "osdop_setxattr",
298 "Set xattr operations");
299 pcb
.add_u64_counter(l_osdc_osdop_cmpxattr
, "osdop_cmpxattr",
300 "Xattr comparison operations");
301 pcb
.add_u64_counter(l_osdc_osdop_rmxattr
, "osdop_rmxattr",
302 "Remove xattr operations");
303 pcb
.add_u64_counter(l_osdc_osdop_resetxattrs
, "osdop_resetxattrs",
304 "Reset xattr operations");
305 pcb
.add_u64_counter(l_osdc_osdop_call
, "osdop_call",
306 "Call (execute) operations");
307 pcb
.add_u64_counter(l_osdc_osdop_watch
, "osdop_watch",
308 "Watch by object operations");
309 pcb
.add_u64_counter(l_osdc_osdop_notify
, "osdop_notify",
310 "Notify about object operations");
311 pcb
.add_u64_counter(l_osdc_osdop_src_cmpxattr
, "osdop_src_cmpxattr",
312 "Extended attribute comparison in multi operations");
313 pcb
.add_u64_counter(l_osdc_osdop_pgls
, "osdop_pgls");
314 pcb
.add_u64_counter(l_osdc_osdop_pgls_filter
, "osdop_pgls_filter");
315 pcb
.add_u64_counter(l_osdc_osdop_other
, "osdop_other", "Other operations");
317 pcb
.add_u64(l_osdc_linger_active
, "linger_active",
318 "Active lingering operations");
319 pcb
.add_u64_counter(l_osdc_linger_send
, "linger_send",
320 "Sent lingering operations");
321 pcb
.add_u64_counter(l_osdc_linger_resend
, "linger_resend",
322 "Resent lingering operations");
323 pcb
.add_u64_counter(l_osdc_linger_ping
, "linger_ping",
324 "Sent pings to lingering operations");
326 pcb
.add_u64(l_osdc_poolop_active
, "poolop_active",
327 "Active pool operations");
328 pcb
.add_u64_counter(l_osdc_poolop_send
, "poolop_send",
329 "Sent pool operations");
330 pcb
.add_u64_counter(l_osdc_poolop_resend
, "poolop_resend",
331 "Resent pool operations");
333 pcb
.add_u64(l_osdc_poolstat_active
, "poolstat_active",
334 "Active get pool stat operations");
335 pcb
.add_u64_counter(l_osdc_poolstat_send
, "poolstat_send",
336 "Pool stat operations sent");
337 pcb
.add_u64_counter(l_osdc_poolstat_resend
, "poolstat_resend",
338 "Resent pool stats");
340 pcb
.add_u64(l_osdc_statfs_active
, "statfs_active", "Statfs operations");
341 pcb
.add_u64_counter(l_osdc_statfs_send
, "statfs_send", "Sent FS stats");
342 pcb
.add_u64_counter(l_osdc_statfs_resend
, "statfs_resend",
345 pcb
.add_u64(l_osdc_command_active
, "command_active", "Active commands");
346 pcb
.add_u64_counter(l_osdc_command_send
, "command_send",
348 pcb
.add_u64_counter(l_osdc_command_resend
, "command_resend",
351 pcb
.add_u64(l_osdc_map_epoch
, "map_epoch", "OSD map epoch");
352 pcb
.add_u64_counter(l_osdc_map_full
, "map_full",
353 "Full OSD maps received");
354 pcb
.add_u64_counter(l_osdc_map_inc
, "map_inc",
355 "Incremental OSD maps received");
357 pcb
.add_u64(l_osdc_osd_sessions
, "osd_sessions",
358 "Open sessions"); // open sessions
359 pcb
.add_u64_counter(l_osdc_osd_session_open
, "osd_session_open",
361 pcb
.add_u64_counter(l_osdc_osd_session_close
, "osd_session_close",
363 pcb
.add_u64(l_osdc_osd_laggy
, "osd_laggy", "Laggy OSD sessions");
365 pcb
.add_u64_counter(l_osdc_osdop_omap_wr
, "omap_wr",
366 "OSD OMAP write operations");
367 pcb
.add_u64_counter(l_osdc_osdop_omap_rd
, "omap_rd",
368 "OSD OMAP read operations");
369 pcb
.add_u64_counter(l_osdc_osdop_omap_del
, "omap_del",
370 "OSD OMAP delete operations");
372 logger
= pcb
.create_perf_counters();
373 cct
->get_perfcounters_collection()->add(logger
);
376 m_request_state_hook
= new RequestStateHook(this);
377 auto admin_socket
= cct
->get_admin_socket();
378 int ret
= admin_socket
->register_command("objecter_requests",
379 m_request_state_hook
,
380 "show in-progress osd requests");
382 /* Don't warn on EEXIST, happens if multiple ceph clients
383 * are instantiated from one process */
384 if (ret
< 0 && ret
!= -EEXIST
) {
385 lderr(cct
) << "error registering admin socket command: "
386 << cpp_strerror(ret
) << dendl
;
389 update_crush_location();
391 cct
->_conf
.add_observer(this);
397 * ok, cluster interaction can happen
399 void Objecter::start(const OSDMap
* o
)
401 shared_lock
rl(rwlock
);
405 osdmap
->deepish_copy_from(*o
);
406 prune_pg_mapping(osdmap
->get_pools());
407 } else if (osdmap
->get_epoch() == 0) {
408 _maybe_request_map();
412 void Objecter::shutdown()
414 ceph_assert(initialized
);
416 unique_lock
wl(rwlock
);
421 cct
->_conf
.remove_observer(this);
424 while (!osd_sessions
.empty()) {
425 auto p
= osd_sessions
.begin();
426 close_session(p
->second
);
429 while(!check_latest_map_lingers
.empty()) {
430 auto i
= check_latest_map_lingers
.begin();
432 check_latest_map_lingers
.erase(i
->first
);
435 while(!check_latest_map_ops
.empty()) {
436 auto i
= check_latest_map_ops
.begin();
438 check_latest_map_ops
.erase(i
->first
);
441 while(!check_latest_map_commands
.empty()) {
442 auto i
= check_latest_map_commands
.begin();
444 check_latest_map_commands
.erase(i
->first
);
447 while(!poolstat_ops
.empty()) {
448 auto i
= poolstat_ops
.begin();
450 poolstat_ops
.erase(i
->first
);
453 while(!statfs_ops
.empty()) {
454 auto i
= statfs_ops
.begin();
456 statfs_ops
.erase(i
->first
);
459 while(!pool_ops
.empty()) {
460 auto i
= pool_ops
.begin();
462 pool_ops
.erase(i
->first
);
465 ldout(cct
, 20) << __func__
<< " clearing up homeless session..." << dendl
;
466 while(!homeless_session
->linger_ops
.empty()) {
467 auto i
= homeless_session
->linger_ops
.begin();
468 ldout(cct
, 10) << " linger_op " << i
->first
<< dendl
;
469 LingerOp
*lop
= i
->second
;
471 std::unique_lock
swl(homeless_session
->lock
);
472 _session_linger_op_remove(homeless_session
, lop
);
474 linger_ops
.erase(lop
->linger_id
);
475 linger_ops_set
.erase(lop
);
479 while(!homeless_session
->ops
.empty()) {
480 auto i
= homeless_session
->ops
.begin();
481 ldout(cct
, 10) << " op " << i
->first
<< dendl
;
484 std::unique_lock
swl(homeless_session
->lock
);
485 _session_op_remove(homeless_session
, op
);
490 while(!homeless_session
->command_ops
.empty()) {
491 auto i
= homeless_session
->command_ops
.begin();
492 ldout(cct
, 10) << " command_op " << i
->first
<< dendl
;
493 auto cop
= i
->second
;
495 std::unique_lock
swl(homeless_session
->lock
);
496 _session_command_op_remove(homeless_session
, cop
);
502 if (timer
.cancel_event(tick_event
)) {
503 ldout(cct
, 10) << " successfully canceled tick" << dendl
;
509 cct
->get_perfcounters_collection()->remove(logger
);
514 // Let go of Objecter write lock so timer thread can shutdown
517 // Outside of lock to avoid cycle WRT calls to RequestStateHook
518 // This is safe because we guarantee no concurrent calls to
519 // shutdown() with the ::initialized check at start.
520 if (m_request_state_hook
) {
521 auto admin_socket
= cct
->get_admin_socket();
522 admin_socket
->unregister_commands(m_request_state_hook
);
523 delete m_request_state_hook
;
524 m_request_state_hook
= NULL
;
528 void Objecter::_send_linger(LingerOp
*info
,
529 ceph::shunique_lock
<ceph::shared_mutex
>& sul
)
531 ceph_assert(sul
.owns_lock() && sul
.mutex() == &rwlock
);
533 fu2::unique_function
<Op::OpSig
> oncommit
;
535 std::shared_lock
watchl(info
->watch_lock
);
536 cb::list
*poutbl
= nullptr;
537 if (info
->registered
&& info
->is_watch
) {
538 ldout(cct
, 15) << "send_linger " << info
->linger_id
<< " reconnect"
540 opv
.push_back(OSDOp());
541 opv
.back().op
.op
= CEPH_OSD_OP_WATCH
;
542 opv
.back().op
.watch
.cookie
= info
->get_cookie();
543 opv
.back().op
.watch
.op
= CEPH_OSD_WATCH_OP_RECONNECT
;
544 opv
.back().op
.watch
.gen
= ++info
->register_gen
;
545 oncommit
= CB_Linger_Reconnect(this, info
);
547 ldout(cct
, 15) << "send_linger " << info
->linger_id
<< " register"
550 // TODO Augment ca::Completion with an equivalent of
551 // target so we can handle these cases better.
552 auto c
= std::make_unique
<CB_Linger_Commit
>(this, info
);
553 if (!info
->is_watch
) {
557 oncommit
= [c
= std::move(c
)](bs::error_code ec
) mutable {
562 auto o
= new Op(info
->target
.base_oid
, info
->target
.base_oloc
,
563 std::move(opv
), info
->target
.flags
| CEPH_OSD_FLAG_READ
,
564 std::move(oncommit
), info
->pobjver
);
566 o
->snapid
= info
->snap
;
567 o
->snapc
= info
->snapc
;
568 o
->mtime
= info
->mtime
;
570 o
->target
= info
->target
;
573 // do not resend this; we will send a new op to reregister
574 o
->should_resend
= false;
575 o
->ctx_budgeted
= true;
577 if (info
->register_tid
) {
578 // repeat send. cancel old registration op, if any.
579 std::unique_lock
sl(info
->session
->lock
);
580 if (info
->session
->ops
.count(info
->register_tid
)) {
581 auto o
= info
->session
->ops
[info
->register_tid
];
582 _op_cancel_map_check(o
);
583 _cancel_linger_op(o
);
588 _op_submit_with_budget(o
, sul
, &info
->register_tid
, &info
->ctx_budget
);
590 logger
->inc(l_osdc_linger_send
);
593 void Objecter::_linger_commit(LingerOp
*info
, bs::error_code ec
,
596 std::unique_lock
wl(info
->watch_lock
);
597 ldout(cct
, 10) << "_linger_commit " << info
->linger_id
<< dendl
;
598 if (info
->on_reg_commit
) {
599 info
->on_reg_commit
->defer(std::move(info
->on_reg_commit
),
601 info
->on_reg_commit
.reset();
603 if (ec
&& info
->on_notify_finish
) {
604 info
->on_notify_finish
->defer(std::move(info
->on_notify_finish
),
606 info
->on_notify_finish
.reset();
609 // only tell the user the first time we do this
610 info
->registered
= true;
611 info
->pobjver
= NULL
;
613 if (!info
->is_watch
) {
614 // make note of the notify_id
615 auto p
= outbl
.cbegin();
617 decode(info
->notify_id
, p
);
618 ldout(cct
, 10) << "_linger_commit notify_id=" << info
->notify_id
621 catch (cb::error
& e
) {
626 class CB_DoWatchError
{
628 boost::intrusive_ptr
<Objecter::LingerOp
> info
;
631 CB_DoWatchError(Objecter
*o
, Objecter::LingerOp
*i
,
633 : objecter(o
), info(i
), ec(ec
) {
634 info
->_queued_async();
637 std::unique_lock
wl(objecter
->rwlock
);
638 bool canceled
= info
->canceled
;
642 info
->handle(ec
, 0, info
->get_cookie(), 0, {});
645 info
->finished_async();
649 bs::error_code
Objecter::_normalize_watch_error(bs::error_code ec
)
651 // translate ENOENT -> ENOTCONN so that a delete->disconnection
652 // notification and a failure to reconnect because we raced with
653 // the delete appear the same to the user.
654 if (ec
== bs::errc::no_such_file_or_directory
)
655 ec
= bs::error_code(ENOTCONN
, osd_category());
659 void Objecter::_linger_reconnect(LingerOp
*info
, bs::error_code ec
)
661 ldout(cct
, 10) << __func__
<< " " << info
->linger_id
<< " = " << ec
662 << " (last_error " << info
->last_error
<< ")" << dendl
;
663 std::unique_lock
wl(info
->watch_lock
);
665 if (!info
->last_error
) {
666 ec
= _normalize_watch_error(ec
);
668 boost::asio::defer(finish_strand
, CB_DoWatchError(this, info
, ec
));
673 info
->last_error
= ec
;
676 void Objecter::_send_linger_ping(LingerOp
*info
)
678 // rwlock is locked unique
679 // info->session->lock is locked
681 if (cct
->_conf
->objecter_inject_no_watch_ping
) {
682 ldout(cct
, 10) << __func__
<< " " << info
->linger_id
<< " SKIPPING"
686 if (osdmap
->test_flag(CEPH_OSDMAP_PAUSERD
)) {
687 ldout(cct
, 10) << __func__
<< " PAUSERD" << dendl
;
691 ceph::coarse_mono_time now
= ceph::coarse_mono_clock::now();
692 ldout(cct
, 10) << __func__
<< " " << info
->linger_id
<< " now " << now
696 opv
[0].op
.op
= CEPH_OSD_OP_WATCH
;
697 opv
[0].op
.watch
.cookie
= info
->get_cookie();
698 opv
[0].op
.watch
.op
= CEPH_OSD_WATCH_OP_PING
;
699 opv
[0].op
.watch
.gen
= info
->register_gen
;
701 Op
*o
= new Op(info
->target
.base_oid
, info
->target
.base_oloc
,
702 std::move(opv
), info
->target
.flags
| CEPH_OSD_FLAG_READ
,
703 CB_Linger_Ping(this, info
, now
),
705 o
->target
= info
->target
;
706 o
->should_resend
= false;
709 _session_op_assign(info
->session
, o
);
711 info
->ping_tid
= o
->tid
;
713 logger
->inc(l_osdc_linger_ping
);
716 void Objecter::_linger_ping(LingerOp
*info
, bs::error_code ec
, ceph::coarse_mono_time sent
,
717 uint32_t register_gen
)
719 std::unique_lock
l(info
->watch_lock
);
720 ldout(cct
, 10) << __func__
<< " " << info
->linger_id
721 << " sent " << sent
<< " gen " << register_gen
<< " = " << ec
722 << " (last_error " << info
->last_error
723 << " register_gen " << info
->register_gen
<< ")" << dendl
;
724 if (info
->register_gen
== register_gen
) {
726 info
->watch_valid_thru
= sent
;
727 } else if (ec
&& !info
->last_error
) {
728 ec
= _normalize_watch_error(ec
);
729 info
->last_error
= ec
;
731 boost::asio::defer(finish_strand
, CB_DoWatchError(this, info
, ec
));
735 ldout(cct
, 20) << " ignoring old gen" << dendl
;
739 tl::expected
<ceph::timespan
,
740 bs::error_code
> Objecter::linger_check(LingerOp
*info
)
742 std::shared_lock
l(info
->watch_lock
);
744 ceph::coarse_mono_time stamp
= info
->watch_valid_thru
;
745 if (!info
->watch_pending_async
.empty())
746 stamp
= std::min(info
->watch_valid_thru
, info
->watch_pending_async
.front());
747 auto age
= ceph::coarse_mono_clock::now() - stamp
;
749 ldout(cct
, 10) << __func__
<< " " << info
->linger_id
750 << " err " << info
->last_error
751 << " age " << age
<< dendl
;
752 if (info
->last_error
)
753 return tl::unexpected(info
->last_error
);
754 // return a safe upper bound (we are truncating to ms)
758 void Objecter::linger_cancel(LingerOp
*info
)
760 unique_lock
wl(rwlock
);
761 _linger_cancel(info
);
765 void Objecter::_linger_cancel(LingerOp
*info
)
767 // rwlock is locked unique
768 ldout(cct
, 20) << __func__
<< " linger_id=" << info
->linger_id
<< dendl
;
769 if (!info
->canceled
) {
770 OSDSession
*s
= info
->session
;
771 std::unique_lock
sl(s
->lock
);
772 _session_linger_op_remove(s
, info
);
775 linger_ops
.erase(info
->linger_id
);
776 linger_ops_set
.erase(info
);
777 ceph_assert(linger_ops
.size() == linger_ops_set
.size());
779 info
->canceled
= true;
782 logger
->dec(l_osdc_linger_active
);
788 Objecter::LingerOp
*Objecter::linger_register(const object_t
& oid
,
789 const object_locator_t
& oloc
,
792 unique_lock
l(rwlock
);
794 auto info
= new LingerOp(this, ++max_linger_id
);
795 info
->target
.base_oid
= oid
;
796 info
->target
.base_oloc
= oloc
;
797 if (info
->target
.base_oloc
.key
== oid
)
798 info
->target
.base_oloc
.key
.clear();
799 info
->target
.flags
= flags
;
800 info
->watch_valid_thru
= ceph::coarse_mono_clock::now();
801 ldout(cct
, 10) << __func__
<< " info " << info
802 << " linger_id " << info
->linger_id
803 << " cookie " << info
->get_cookie()
805 linger_ops
[info
->linger_id
] = info
;
806 linger_ops_set
.insert(info
);
807 ceph_assert(linger_ops
.size() == linger_ops_set
.size());
809 info
->get(); // for the caller
813 ceph_tid_t
Objecter::linger_watch(LingerOp
*info
,
815 const SnapContext
& snapc
,
818 decltype(info
->on_reg_commit
)&& oncommit
,
821 info
->is_watch
= true;
824 info
->target
.flags
|= CEPH_OSD_FLAG_WRITE
;
827 info
->pobjver
= objver
;
828 info
->on_reg_commit
= std::move(oncommit
);
830 info
->ctx_budget
= take_linger_budget(info
);
832 shunique_lock
sul(rwlock
, ceph::acquire_unique
);
833 _linger_submit(info
, sul
);
834 logger
->inc(l_osdc_linger_active
);
837 return info
->linger_id
;
840 ceph_tid_t
Objecter::linger_notify(LingerOp
*info
,
842 snapid_t snap
, cb::list
& inbl
,
843 decltype(LingerOp::on_reg_commit
)&& onfinish
,
847 info
->target
.flags
|= CEPH_OSD_FLAG_READ
;
850 info
->pobjver
= objver
;
851 info
->on_reg_commit
= std::move(onfinish
);
852 info
->ctx_budget
= take_linger_budget(info
);
854 shunique_lock
sul(rwlock
, ceph::acquire_unique
);
855 _linger_submit(info
, sul
);
856 logger
->inc(l_osdc_linger_active
);
859 return info
->linger_id
;
862 void Objecter::_linger_submit(LingerOp
*info
,
863 ceph::shunique_lock
<ceph::shared_mutex
>& sul
)
865 ceph_assert(sul
.owns_lock() && sul
.mutex() == &rwlock
);
866 ceph_assert(info
->linger_id
);
867 ceph_assert(info
->ctx_budget
!= -1); // caller needs to have taken budget already!
869 // Populate Op::target
870 OSDSession
*s
= NULL
;
871 _calc_target(&info
->target
, nullptr);
873 // Create LingerOp<->OSDSession relation
874 int r
= _get_session(info
->target
.osd
, &s
, sul
);
876 unique_lock
sl(s
->lock
);
877 _session_linger_op_assign(s
, info
);
881 _send_linger(info
, sul
);
884 struct CB_DoWatchNotify
{
886 boost::intrusive_ptr
<Objecter::LingerOp
> info
;
887 boost::intrusive_ptr
<MWatchNotify
> msg
;
888 CB_DoWatchNotify(Objecter
*o
, Objecter::LingerOp
*i
, MWatchNotify
*m
)
889 : objecter(o
), info(i
), msg(m
) {
890 info
->_queued_async();
893 objecter
->_do_watch_notify(std::move(info
), std::move(msg
));
897 void Objecter::handle_watch_notify(MWatchNotify
*m
)
899 shared_lock
l(rwlock
);
904 LingerOp
*info
= reinterpret_cast<LingerOp
*>(m
->cookie
);
905 if (linger_ops_set
.count(info
) == 0) {
906 ldout(cct
, 7) << __func__
<< " cookie " << m
->cookie
<< " dne" << dendl
;
909 std::unique_lock
wl(info
->watch_lock
);
910 if (m
->opcode
== CEPH_WATCH_EVENT_DISCONNECT
) {
911 if (!info
->last_error
) {
912 info
->last_error
= bs::error_code(ENOTCONN
, osd_category());
914 boost::asio::defer(finish_strand
, CB_DoWatchError(this, info
,
918 } else if (!info
->is_watch
) {
919 // we have CEPH_WATCH_EVENT_NOTIFY_COMPLETE; we can do this inline
920 // since we know the only user (librados) is safe to call in
921 // fast-dispatch context
922 if (info
->notify_id
&&
923 info
->notify_id
!= m
->notify_id
) {
924 ldout(cct
, 10) << __func__
<< " reply notify " << m
->notify_id
925 << " != " << info
->notify_id
<< ", ignoring" << dendl
;
926 } else if (info
->on_notify_finish
) {
927 info
->on_notify_finish
->defer(
928 std::move(info
->on_notify_finish
),
929 osdcode(m
->return_code
), std::move(m
->get_data()));
931 // if we race with reconnect we might get a second notify; only
932 // notify the caller once!
933 info
->on_notify_finish
= nullptr;
936 boost::asio::defer(finish_strand
, CB_DoWatchNotify(this, info
, m
));
940 void Objecter::_do_watch_notify(boost::intrusive_ptr
<LingerOp
> info
,
941 boost::intrusive_ptr
<MWatchNotify
> m
)
943 ldout(cct
, 10) << __func__
<< " " << *m
<< dendl
;
945 shared_lock
l(rwlock
);
946 ceph_assert(initialized
);
948 if (info
->canceled
) {
953 // notify completion?
954 ceph_assert(info
->is_watch
);
955 ceph_assert(info
->handle
);
956 ceph_assert(m
->opcode
!= CEPH_WATCH_EVENT_DISCONNECT
);
961 case CEPH_WATCH_EVENT_NOTIFY
:
962 info
->handle({}, m
->notify_id
, m
->cookie
, m
->notifier_gid
, std::move(m
->bl
));
967 info
->finished_async();
970 bool Objecter::ms_dispatch(Message
*m
)
972 ldout(cct
, 10) << __func__
<< " " << cct
<< " " << *m
<< dendl
;
973 switch (m
->get_type()) {
974 // these we exlusively handle
975 case CEPH_MSG_OSD_OPREPLY
:
976 handle_osd_op_reply(static_cast<MOSDOpReply
*>(m
));
979 case CEPH_MSG_OSD_BACKOFF
:
980 handle_osd_backoff(static_cast<MOSDBackoff
*>(m
));
983 case CEPH_MSG_WATCH_NOTIFY
:
984 handle_watch_notify(static_cast<MWatchNotify
*>(m
));
988 case MSG_COMMAND_REPLY
:
989 if (m
->get_source().type() == CEPH_ENTITY_TYPE_OSD
) {
990 handle_command_reply(static_cast<MCommandReply
*>(m
));
996 case MSG_GETPOOLSTATSREPLY
:
997 handle_get_pool_stats_reply(static_cast<MGetPoolStatsReply
*>(m
));
1000 case CEPH_MSG_POOLOP_REPLY
:
1001 handle_pool_op_reply(static_cast<MPoolOpReply
*>(m
));
1004 case CEPH_MSG_STATFS_REPLY
:
1005 handle_fs_stats_reply(static_cast<MStatfsReply
*>(m
));
1008 // these we give others a chance to inspect
1011 case CEPH_MSG_OSD_MAP
:
1012 handle_osd_map(static_cast<MOSDMap
*>(m
));
1018 void Objecter::_scan_requests(
1022 map
<int64_t, bool> *pool_full_map
,
1023 map
<ceph_tid_t
, Op
*>& need_resend
,
1024 list
<LingerOp
*>& need_resend_linger
,
1025 map
<ceph_tid_t
, CommandOp
*>& need_resend_command
,
1026 ceph::shunique_lock
<ceph::shared_mutex
>& sul
)
1028 ceph_assert(sul
.owns_lock() && sul
.mutex() == &rwlock
);
1030 list
<LingerOp
*> unregister_lingers
;
1032 std::unique_lock
sl(s
->lock
);
1034 // check for changed linger mappings (_before_ regular ops)
1035 auto lp
= s
->linger_ops
.begin();
1036 while (lp
!= s
->linger_ops
.end()) {
1037 auto op
= lp
->second
;
1038 ceph_assert(op
->session
== s
);
1039 // check_linger_pool_dne() may touch linger_ops; prevent iterator
1042 ldout(cct
, 10) << " checking linger op " << op
->linger_id
<< dendl
;
1043 bool unregister
, force_resend_writes
= cluster_full
;
1044 int r
= _recalc_linger_op_target(op
, sul
);
1046 force_resend_writes
= force_resend_writes
||
1047 (*pool_full_map
)[op
->target
.base_oloc
.pool
];
1049 case RECALC_OP_TARGET_NO_ACTION
:
1050 if (!skipped_map
&& !force_resend_writes
)
1053 case RECALC_OP_TARGET_NEED_RESEND
:
1054 need_resend_linger
.push_back(op
);
1055 _linger_cancel_map_check(op
);
1057 case RECALC_OP_TARGET_POOL_DNE
:
1058 _check_linger_pool_dne(op
, &unregister
);
1060 ldout(cct
, 10) << " need to unregister linger op "
1061 << op
->linger_id
<< dendl
;
1063 unregister_lingers
.push_back(op
);
1069 // check for changed request mappings
1070 auto p
= s
->ops
.begin();
1071 while (p
!= s
->ops
.end()) {
1073 ++p
; // check_op_pool_dne() may touch ops; prevent iterator invalidation
1074 ldout(cct
, 10) << " checking op " << op
->tid
<< dendl
;
1075 _prune_snapc(osdmap
->get_new_removed_snaps(), op
);
1076 bool force_resend_writes
= cluster_full
;
1078 force_resend_writes
= force_resend_writes
||
1079 (*pool_full_map
)[op
->target
.base_oloc
.pool
];
1080 int r
= _calc_target(&op
->target
,
1081 op
->session
? op
->session
->con
.get() : nullptr);
1083 case RECALC_OP_TARGET_NO_ACTION
:
1084 if (!skipped_map
&& !(force_resend_writes
&& op
->target
.respects_full()))
1087 case RECALC_OP_TARGET_NEED_RESEND
:
1088 _session_op_remove(op
->session
, op
);
1089 need_resend
[op
->tid
] = op
;
1090 _op_cancel_map_check(op
);
1092 case RECALC_OP_TARGET_POOL_DNE
:
1093 _check_op_pool_dne(op
, &sl
);
1099 auto cp
= s
->command_ops
.begin();
1100 while (cp
!= s
->command_ops
.end()) {
1101 auto c
= cp
->second
;
1103 ldout(cct
, 10) << " checking command " << c
->tid
<< dendl
;
1104 bool force_resend_writes
= cluster_full
;
1106 force_resend_writes
= force_resend_writes
||
1107 (*pool_full_map
)[c
->target_pg
.pool()];
1108 int r
= _calc_command_target(c
, sul
);
1110 case RECALC_OP_TARGET_NO_ACTION
:
1111 // resend if skipped map; otherwise do nothing.
1112 if (!skipped_map
&& !force_resend_writes
)
1115 case RECALC_OP_TARGET_NEED_RESEND
:
1116 need_resend_command
[c
->tid
] = c
;
1117 _session_command_op_remove(c
->session
, c
);
1118 _command_cancel_map_check(c
);
1120 case RECALC_OP_TARGET_POOL_DNE
:
1121 case RECALC_OP_TARGET_OSD_DNE
:
1122 case RECALC_OP_TARGET_OSD_DOWN
:
1123 _check_command_map_dne(c
);
1130 for (auto iter
= unregister_lingers
.begin();
1131 iter
!= unregister_lingers
.end();
1133 _linger_cancel(*iter
);
1138 void Objecter::handle_osd_map(MOSDMap
*m
)
1140 ceph::shunique_lock
sul(rwlock
, acquire_unique
);
1144 ceph_assert(osdmap
);
1146 if (m
->fsid
!= monc
->get_fsid()) {
1147 ldout(cct
, 0) << "handle_osd_map fsid " << m
->fsid
1148 << " != " << monc
->get_fsid() << dendl
;
1152 bool was_pauserd
= osdmap
->test_flag(CEPH_OSDMAP_PAUSERD
);
1153 bool cluster_full
= _osdmap_full_flag();
1154 bool was_pausewr
= osdmap
->test_flag(CEPH_OSDMAP_PAUSEWR
) || cluster_full
||
1155 _osdmap_has_pool_full();
1156 map
<int64_t, bool> pool_full_map
;
1157 for (auto it
= osdmap
->get_pools().begin();
1158 it
!= osdmap
->get_pools().end(); ++it
)
1159 pool_full_map
[it
->first
] = _osdmap_pool_full(it
->second
);
1162 list
<LingerOp
*> need_resend_linger
;
1163 map
<ceph_tid_t
, Op
*> need_resend
;
1164 map
<ceph_tid_t
, CommandOp
*> need_resend_command
;
1166 if (m
->get_last() <= osdmap
->get_epoch()) {
1167 ldout(cct
, 3) << "handle_osd_map ignoring epochs ["
1168 << m
->get_first() << "," << m
->get_last()
1169 << "] <= " << osdmap
->get_epoch() << dendl
;
1171 ldout(cct
, 3) << "handle_osd_map got epochs ["
1172 << m
->get_first() << "," << m
->get_last()
1173 << "] > " << osdmap
->get_epoch() << dendl
;
1175 if (osdmap
->get_epoch()) {
1176 bool skipped_map
= false;
1177 // we want incrementals
1178 for (epoch_t e
= osdmap
->get_epoch() + 1;
1182 if (osdmap
->get_epoch() == e
-1 &&
1183 m
->incremental_maps
.count(e
)) {
1184 ldout(cct
, 3) << "handle_osd_map decoding incremental epoch " << e
1186 OSDMap::Incremental
inc(m
->incremental_maps
[e
]);
1187 osdmap
->apply_incremental(inc
);
1189 emit_blocklist_events(inc
);
1191 logger
->inc(l_osdc_map_inc
);
1193 else if (m
->maps
.count(e
)) {
1194 ldout(cct
, 3) << "handle_osd_map decoding full epoch " << e
<< dendl
;
1195 auto new_osdmap
= std::make_unique
<OSDMap
>();
1196 new_osdmap
->decode(m
->maps
[e
]);
1198 emit_blocklist_events(*osdmap
, *new_osdmap
);
1199 osdmap
= std::move(new_osdmap
);
1201 logger
->inc(l_osdc_map_full
);
1204 if (e
>= m
->get_oldest()) {
1205 ldout(cct
, 3) << "handle_osd_map requesting missing epoch "
1206 << osdmap
->get_epoch()+1 << dendl
;
1207 _maybe_request_map();
1210 ldout(cct
, 3) << "handle_osd_map missing epoch "
1211 << osdmap
->get_epoch()+1
1212 << ", jumping to " << m
->get_oldest() << dendl
;
1213 e
= m
->get_oldest() - 1;
1217 logger
->set(l_osdc_map_epoch
, osdmap
->get_epoch());
1219 prune_pg_mapping(osdmap
->get_pools());
1220 cluster_full
= cluster_full
|| _osdmap_full_flag();
1221 update_pool_full_map(pool_full_map
);
1223 // check all outstanding requests on every epoch
1224 for (auto& i
: need_resend
) {
1225 _prune_snapc(osdmap
->get_new_removed_snaps(), i
.second
);
1227 _scan_requests(homeless_session
, skipped_map
, cluster_full
,
1228 &pool_full_map
, need_resend
,
1229 need_resend_linger
, need_resend_command
, sul
);
1230 for (auto p
= osd_sessions
.begin();
1231 p
!= osd_sessions
.end(); ) {
1233 _scan_requests(s
, skipped_map
, cluster_full
,
1234 &pool_full_map
, need_resend
,
1235 need_resend_linger
, need_resend_command
, sul
);
1237 // osd down or addr change?
1238 if (!osdmap
->is_up(s
->osd
) ||
1240 s
->con
->get_peer_addrs() != osdmap
->get_addrs(s
->osd
))) {
1245 ceph_assert(e
== osdmap
->get_epoch());
1249 // first map. we want the full thing.
1250 if (m
->maps
.count(m
->get_last())) {
1251 for (auto p
= osd_sessions
.begin();
1252 p
!= osd_sessions
.end(); ++p
) {
1253 OSDSession
*s
= p
->second
;
1254 _scan_requests(s
, false, false, NULL
, need_resend
,
1255 need_resend_linger
, need_resend_command
, sul
);
1257 ldout(cct
, 3) << "handle_osd_map decoding full epoch "
1258 << m
->get_last() << dendl
;
1259 osdmap
->decode(m
->maps
[m
->get_last()]);
1260 prune_pg_mapping(osdmap
->get_pools());
1262 _scan_requests(homeless_session
, false, false, NULL
,
1263 need_resend
, need_resend_linger
,
1264 need_resend_command
, sul
);
1266 ldout(cct
, 3) << "handle_osd_map hmm, i want a full map, requesting"
1268 monc
->sub_want("osdmap", 0, CEPH_SUBSCRIBE_ONETIME
);
1274 // make sure need_resend targets reflect latest map
1275 for (auto p
= need_resend
.begin(); p
!= need_resend
.end(); ) {
1277 if (op
->target
.epoch
< osdmap
->get_epoch()) {
1278 ldout(cct
, 10) << __func__
<< " checking op " << p
->first
<< dendl
;
1279 int r
= _calc_target(&op
->target
, nullptr);
1280 if (r
== RECALC_OP_TARGET_POOL_DNE
) {
1281 p
= need_resend
.erase(p
);
1282 _check_op_pool_dne(op
, nullptr);
1291 bool pauserd
= osdmap
->test_flag(CEPH_OSDMAP_PAUSERD
);
1292 bool pausewr
= osdmap
->test_flag(CEPH_OSDMAP_PAUSEWR
) || _osdmap_full_flag()
1293 || _osdmap_has_pool_full();
1296 if (was_pauserd
|| was_pausewr
|| pauserd
|| pausewr
||
1297 osdmap
->get_epoch() < epoch_barrier
) {
1298 _maybe_request_map();
1302 for (auto p
= need_resend
.begin();
1303 p
!= need_resend
.end(); ++p
) {
1304 auto op
= p
->second
;
1305 auto s
= op
->session
;
1306 bool mapped_session
= false;
1308 int r
= _map_session(&op
->target
, &s
, sul
);
1309 ceph_assert(r
== 0);
1310 mapped_session
= true;
1314 std::unique_lock
sl(s
->lock
);
1315 if (mapped_session
) {
1316 _session_op_assign(s
, op
);
1318 if (op
->should_resend
) {
1319 if (!op
->session
->is_homeless() && !op
->target
.paused
) {
1320 logger
->inc(l_osdc_op_resend
);
1324 _op_cancel_map_check(op
);
1325 _cancel_linger_op(op
);
1330 for (auto p
= need_resend_linger
.begin();
1331 p
!= need_resend_linger
.end(); ++p
) {
1333 ceph_assert(op
->session
);
1334 if (!op
->session
->is_homeless()) {
1335 logger
->inc(l_osdc_linger_resend
);
1336 _send_linger(op
, sul
);
1339 for (auto p
= need_resend_command
.begin();
1340 p
!= need_resend_command
.end(); ++p
) {
1342 if (c
->target
.osd
>= 0) {
1343 _assign_command_session(c
, sul
);
1344 if (c
->session
&& !c
->session
->is_homeless()) {
1352 // finish any Contexts that were waiting on a map update
1353 auto p
= waiting_for_map
.begin();
1354 while (p
!= waiting_for_map
.end() &&
1355 p
->first
<= osdmap
->get_epoch()) {
1356 //go through the list and call the onfinish methods
1357 for (auto& [c
, ec
] : p
->second
) {
1358 ca::post(std::move(c
), ec
);
1360 waiting_for_map
.erase(p
++);
1363 monc
->sub_got("osdmap", osdmap
->get_epoch());
1365 if (!waiting_for_map
.empty()) {
1366 _maybe_request_map();
1370 void Objecter::enable_blocklist_events()
1372 unique_lock
wl(rwlock
);
1374 blocklist_events_enabled
= true;
1377 void Objecter::consume_blocklist_events(std::set
<entity_addr_t
> *events
)
1379 unique_lock
wl(rwlock
);
1381 if (events
->empty()) {
1382 events
->swap(blocklist_events
);
1384 for (const auto &i
: blocklist_events
) {
1387 blocklist_events
.clear();
1391 void Objecter::emit_blocklist_events(const OSDMap::Incremental
&inc
)
1393 if (!blocklist_events_enabled
) {
1397 for (const auto &i
: inc
.new_blocklist
) {
1398 blocklist_events
.insert(i
.first
);
1402 void Objecter::emit_blocklist_events(const OSDMap
&old_osd_map
,
1403 const OSDMap
&new_osd_map
)
1405 if (!blocklist_events_enabled
) {
1409 std::set
<entity_addr_t
> old_set
;
1410 std::set
<entity_addr_t
> new_set
;
1412 old_osd_map
.get_blocklist(&old_set
);
1413 new_osd_map
.get_blocklist(&new_set
);
1415 std::set
<entity_addr_t
> delta_set
;
1416 std::set_difference(
1417 new_set
.begin(), new_set
.end(), old_set
.begin(), old_set
.end(),
1418 std::inserter(delta_set
, delta_set
.begin()));
1419 blocklist_events
.insert(delta_set
.begin(), delta_set
.end());
1424 void Objecter::CB_Op_Map_Latest::operator()(bs::error_code e
,
1425 version_t latest
, version_t
)
1427 if (e
== bs::errc::resource_unavailable_try_again
||
1428 e
== bs::errc::operation_canceled
)
1431 lgeneric_subdout(objecter
->cct
, objecter
, 10)
1432 << "op_map_latest r=" << e
<< " tid=" << tid
1433 << " latest " << latest
<< dendl
;
1435 unique_lock
wl(objecter
->rwlock
);
1437 auto iter
= objecter
->check_latest_map_ops
.find(tid
);
1438 if (iter
== objecter
->check_latest_map_ops
.end()) {
1439 lgeneric_subdout(objecter
->cct
, objecter
, 10)
1440 << "op_map_latest op "<< tid
<< " not found" << dendl
;
1444 Op
*op
= iter
->second
;
1445 objecter
->check_latest_map_ops
.erase(iter
);
1447 lgeneric_subdout(objecter
->cct
, objecter
, 20)
1448 << "op_map_latest op "<< op
<< dendl
;
1450 if (op
->map_dne_bound
== 0)
1451 op
->map_dne_bound
= latest
;
1453 unique_lock
sl(op
->session
->lock
, defer_lock
);
1454 objecter
->_check_op_pool_dne(op
, &sl
);
1459 int Objecter::pool_snap_by_name(int64_t poolid
, const char *snap_name
,
1460 snapid_t
*snap
) const
1462 shared_lock
rl(rwlock
);
1464 auto& pools
= osdmap
->get_pools();
1465 auto iter
= pools
.find(poolid
);
1466 if (iter
== pools
.end()) {
1469 const pg_pool_t
& pg_pool
= iter
->second
;
1470 for (auto p
= pg_pool
.snaps
.begin();
1471 p
!= pg_pool
.snaps
.end();
1473 if (p
->second
.name
== snap_name
) {
1481 int Objecter::pool_snap_get_info(int64_t poolid
, snapid_t snap
,
1482 pool_snap_info_t
*info
) const
1484 shared_lock
rl(rwlock
);
1486 auto& pools
= osdmap
->get_pools();
1487 auto iter
= pools
.find(poolid
);
1488 if (iter
== pools
.end()) {
1491 const pg_pool_t
& pg_pool
= iter
->second
;
1492 auto p
= pg_pool
.snaps
.find(snap
);
1493 if (p
== pg_pool
.snaps
.end())
1500 int Objecter::pool_snap_list(int64_t poolid
, vector
<uint64_t> *snaps
)
1502 shared_lock
rl(rwlock
);
1504 const pg_pool_t
*pi
= osdmap
->get_pg_pool(poolid
);
1507 for (auto p
= pi
->snaps
.begin();
1508 p
!= pi
->snaps
.end();
1510 snaps
->push_back(p
->first
);
1515 // sl may be unlocked.
1516 void Objecter::_check_op_pool_dne(Op
*op
, std::unique_lock
<std::shared_mutex
> *sl
)
1518 // rwlock is locked unique
1520 if (op
->target
.pool_ever_existed
) {
1521 // the pool previously existed and now it does not, which means it
1523 op
->map_dne_bound
= osdmap
->get_epoch();
1524 ldout(cct
, 10) << "check_op_pool_dne tid " << op
->tid
1525 << " pool previously exists but now does not"
1528 ldout(cct
, 10) << "check_op_pool_dne tid " << op
->tid
1529 << " current " << osdmap
->get_epoch()
1530 << " map_dne_bound " << op
->map_dne_bound
1533 if (op
->map_dne_bound
> 0) {
1534 if (osdmap
->get_epoch() >= op
->map_dne_bound
) {
1535 // we had a new enough map
1536 ldout(cct
, 10) << "check_op_pool_dne tid " << op
->tid
1537 << " concluding pool " << op
->target
.base_pgid
.pool()
1539 if (op
->has_completion()) {
1541 op
->complete(osdc_errc::pool_dne
, -ENOENT
);
1544 OSDSession
*s
= op
->session
;
1546 ceph_assert(s
!= NULL
);
1547 ceph_assert(sl
->mutex() == &s
->lock
);
1548 bool session_locked
= sl
->owns_lock();
1549 if (!session_locked
) {
1553 if (!session_locked
) {
1557 _finish_op(op
, 0); // no session
1561 _send_op_map_check(op
);
1565 void Objecter::_send_op_map_check(Op
*op
)
1567 // rwlock is locked unique
1569 if (check_latest_map_ops
.count(op
->tid
) == 0) {
1571 check_latest_map_ops
[op
->tid
] = op
;
1572 monc
->get_version("osdmap", CB_Op_Map_Latest(this, op
->tid
));
1576 void Objecter::_op_cancel_map_check(Op
*op
)
1578 // rwlock is locked unique
1579 auto iter
= check_latest_map_ops
.find(op
->tid
);
1580 if (iter
!= check_latest_map_ops
.end()) {
1581 Op
*op
= iter
->second
;
1583 check_latest_map_ops
.erase(iter
);
1587 // linger pool check
1589 void Objecter::CB_Linger_Map_Latest::operator()(bs::error_code e
,
1593 if (e
== bs::errc::resource_unavailable_try_again
||
1594 e
== bs::errc::operation_canceled
) {
1595 // ignore callback; we will retry in resend_mon_ops()
1599 unique_lock
wl(objecter
->rwlock
);
1601 auto iter
= objecter
->check_latest_map_lingers
.find(linger_id
);
1602 if (iter
== objecter
->check_latest_map_lingers
.end()) {
1606 auto op
= iter
->second
;
1607 objecter
->check_latest_map_lingers
.erase(iter
);
1609 if (op
->map_dne_bound
== 0)
1610 op
->map_dne_bound
= latest
;
1613 objecter
->_check_linger_pool_dne(op
, &unregister
);
1616 objecter
->_linger_cancel(op
);
1622 void Objecter::_check_linger_pool_dne(LingerOp
*op
, bool *need_unregister
)
1624 // rwlock is locked unique
1626 *need_unregister
= false;
1628 if (op
->register_gen
> 0) {
1629 ldout(cct
, 10) << "_check_linger_pool_dne linger_id " << op
->linger_id
1630 << " pool previously existed but now does not"
1632 op
->map_dne_bound
= osdmap
->get_epoch();
1634 ldout(cct
, 10) << "_check_linger_pool_dne linger_id " << op
->linger_id
1635 << " current " << osdmap
->get_epoch()
1636 << " map_dne_bound " << op
->map_dne_bound
1639 if (op
->map_dne_bound
> 0) {
1640 if (osdmap
->get_epoch() >= op
->map_dne_bound
) {
1641 std::unique_lock wl
{op
->watch_lock
};
1642 if (op
->on_reg_commit
) {
1643 op
->on_reg_commit
->defer(std::move(op
->on_reg_commit
),
1644 osdc_errc::pool_dne
, cb::list
{});
1645 op
->on_reg_commit
= nullptr;
1647 if (op
->on_notify_finish
) {
1648 op
->on_notify_finish
->defer(std::move(op
->on_notify_finish
),
1649 osdc_errc::pool_dne
, cb::list
{});
1650 op
->on_notify_finish
= nullptr;
1652 *need_unregister
= true;
1655 _send_linger_map_check(op
);
1659 void Objecter::_send_linger_map_check(LingerOp
*op
)
1662 if (check_latest_map_lingers
.count(op
->linger_id
) == 0) {
1664 check_latest_map_lingers
[op
->linger_id
] = op
;
1665 monc
->get_version("osdmap", CB_Linger_Map_Latest(this, op
->linger_id
));
1669 void Objecter::_linger_cancel_map_check(LingerOp
*op
)
1671 // rwlock is locked unique
1673 auto iter
= check_latest_map_lingers
.find(op
->linger_id
);
1674 if (iter
!= check_latest_map_lingers
.end()) {
1675 LingerOp
*op
= iter
->second
;
1677 check_latest_map_lingers
.erase(iter
);
1681 // command pool check
1683 void Objecter::CB_Command_Map_Latest::operator()(bs::error_code e
,
1684 version_t latest
, version_t
)
1686 if (e
== bs::errc::resource_unavailable_try_again
||
1687 e
== bs::errc::operation_canceled
) {
1688 // ignore callback; we will retry in resend_mon_ops()
1692 unique_lock
wl(objecter
->rwlock
);
1694 auto iter
= objecter
->check_latest_map_commands
.find(tid
);
1695 if (iter
== objecter
->check_latest_map_commands
.end()) {
1699 auto c
= iter
->second
;
1700 objecter
->check_latest_map_commands
.erase(iter
);
1702 if (c
->map_dne_bound
== 0)
1703 c
->map_dne_bound
= latest
;
1705 unique_lock
sul(c
->session
->lock
);
1706 objecter
->_check_command_map_dne(c
);
1712 void Objecter::_check_command_map_dne(CommandOp
*c
)
1714 // rwlock is locked unique
1715 // session is locked unique
1717 ldout(cct
, 10) << "_check_command_map_dne tid " << c
->tid
1718 << " current " << osdmap
->get_epoch()
1719 << " map_dne_bound " << c
->map_dne_bound
1721 if (c
->map_dne_bound
> 0) {
1722 if (osdmap
->get_epoch() >= c
->map_dne_bound
) {
1723 _finish_command(c
, osdcode(c
->map_check_error
),
1724 std::move(c
->map_check_error_str
), {});
1727 _send_command_map_check(c
);
1731 void Objecter::_send_command_map_check(CommandOp
*c
)
1733 // rwlock is locked unique
1734 // session is locked unique
1737 if (check_latest_map_commands
.count(c
->tid
) == 0) {
1739 check_latest_map_commands
[c
->tid
] = c
;
1740 monc
->get_version("osdmap", CB_Command_Map_Latest(this, c
->tid
));
1744 void Objecter::_command_cancel_map_check(CommandOp
*c
)
1746 // rwlock is locked uniqe
1748 auto iter
= check_latest_map_commands
.find(c
->tid
);
1749 if (iter
!= check_latest_map_commands
.end()) {
1750 auto c
= iter
->second
;
1752 check_latest_map_commands
.erase(iter
);
1758 * Look up OSDSession by OSD id.
1760 * @returns 0 on success, or -EAGAIN if the lock context requires
1761 * promotion to write.
1763 int Objecter::_get_session(int osd
, OSDSession
**session
,
1764 shunique_lock
<ceph::shared_mutex
>& sul
)
1766 ceph_assert(sul
&& sul
.mutex() == &rwlock
);
1769 *session
= homeless_session
;
1770 ldout(cct
, 20) << __func__
<< " osd=" << osd
<< " returning homeless"
1775 auto p
= osd_sessions
.find(osd
);
1776 if (p
!= osd_sessions
.end()) {
1780 ldout(cct
, 20) << __func__
<< " s=" << s
<< " osd=" << osd
<< " "
1781 << s
->get_nref() << dendl
;
1784 if (!sul
.owns_lock()) {
1787 auto s
= new OSDSession(cct
, osd
);
1788 osd_sessions
[osd
] = s
;
1789 s
->con
= messenger
->connect_to_osd(osdmap
->get_addrs(osd
));
1790 s
->con
->set_priv(RefCountedPtr
{s
});
1791 logger
->inc(l_osdc_osd_session_open
);
1792 logger
->set(l_osdc_osd_sessions
, osd_sessions
.size());
1795 ldout(cct
, 20) << __func__
<< " s=" << s
<< " osd=" << osd
<< " "
1796 << s
->get_nref() << dendl
;
1800 void Objecter::put_session(Objecter::OSDSession
*s
)
1802 if (s
&& !s
->is_homeless()) {
1803 ldout(cct
, 20) << __func__
<< " s=" << s
<< " osd=" << s
->osd
<< " "
1804 << s
->get_nref() << dendl
;
1809 void Objecter::get_session(Objecter::OSDSession
*s
)
1811 ceph_assert(s
!= NULL
);
1813 if (!s
->is_homeless()) {
1814 ldout(cct
, 20) << __func__
<< " s=" << s
<< " osd=" << s
->osd
<< " "
1815 << s
->get_nref() << dendl
;
1820 void Objecter::_reopen_session(OSDSession
*s
)
1822 // rwlock is locked unique
1823 // s->lock is locked
1825 auto addrs
= osdmap
->get_addrs(s
->osd
);
1826 ldout(cct
, 10) << "reopen_session osd." << s
->osd
<< " session, addr now "
1829 s
->con
->set_priv(NULL
);
1830 s
->con
->mark_down();
1831 logger
->inc(l_osdc_osd_session_close
);
1833 s
->con
= messenger
->connect_to_osd(addrs
);
1834 s
->con
->set_priv(RefCountedPtr
{s
});
1836 logger
->inc(l_osdc_osd_session_open
);
1839 void Objecter::close_session(OSDSession
*s
)
1841 // rwlock is locked unique
1843 ldout(cct
, 10) << "close_session for osd." << s
->osd
<< dendl
;
1845 s
->con
->set_priv(NULL
);
1846 s
->con
->mark_down();
1847 logger
->inc(l_osdc_osd_session_close
);
1849 unique_lock
sl(s
->lock
);
1851 std::list
<LingerOp
*> homeless_lingers
;
1852 std::list
<CommandOp
*> homeless_commands
;
1853 std::list
<Op
*> homeless_ops
;
1855 while (!s
->linger_ops
.empty()) {
1856 auto i
= s
->linger_ops
.begin();
1857 ldout(cct
, 10) << " linger_op " << i
->first
<< dendl
;
1858 homeless_lingers
.push_back(i
->second
);
1859 _session_linger_op_remove(s
, i
->second
);
1862 while (!s
->ops
.empty()) {
1863 auto i
= s
->ops
.begin();
1864 ldout(cct
, 10) << " op " << i
->first
<< dendl
;
1865 homeless_ops
.push_back(i
->second
);
1866 _session_op_remove(s
, i
->second
);
1869 while (!s
->command_ops
.empty()) {
1870 auto i
= s
->command_ops
.begin();
1871 ldout(cct
, 10) << " command_op " << i
->first
<< dendl
;
1872 homeless_commands
.push_back(i
->second
);
1873 _session_command_op_remove(s
, i
->second
);
1876 osd_sessions
.erase(s
->osd
);
1880 // Assign any leftover ops to the homeless session
1882 unique_lock
hsl(homeless_session
->lock
);
1883 for (auto i
= homeless_lingers
.begin();
1884 i
!= homeless_lingers
.end(); ++i
) {
1885 _session_linger_op_assign(homeless_session
, *i
);
1887 for (auto i
= homeless_ops
.begin();
1888 i
!= homeless_ops
.end(); ++i
) {
1889 _session_op_assign(homeless_session
, *i
);
1891 for (auto i
= homeless_commands
.begin();
1892 i
!= homeless_commands
.end(); ++i
) {
1893 _session_command_op_assign(homeless_session
, *i
);
1897 logger
->set(l_osdc_osd_sessions
, osd_sessions
.size());
1900 void Objecter::wait_for_osd_map(epoch_t e
)
1902 unique_lock
l(rwlock
);
1903 if (osdmap
->get_epoch() >= e
) {
1908 ca::waiter
<bs::error_code
> w
;
1909 waiting_for_map
[e
].emplace_back(OpCompletion::create(
1910 service
.get_executor(),
1917 void Objecter::_get_latest_version(epoch_t oldest
, epoch_t newest
,
1918 std::unique_ptr
<OpCompletion
> fin
,
1919 std::unique_lock
<ceph::shared_mutex
>&& l
)
1922 if (osdmap
->get_epoch() >= newest
) {
1923 ldout(cct
, 10) << __func__
<< " latest " << newest
<< ", have it" << dendl
;
1925 ca::defer(std::move(fin
), bs::error_code
{});
1927 ldout(cct
, 10) << __func__
<< " latest " << newest
<< ", waiting" << dendl
;
1928 _wait_for_new_map(std::move(fin
), newest
, bs::error_code
{});
1933 void Objecter::maybe_request_map()
1935 shared_lock
rl(rwlock
);
1936 _maybe_request_map();
1939 void Objecter::_maybe_request_map()
1943 if (_osdmap_full_flag()
1944 || osdmap
->test_flag(CEPH_OSDMAP_PAUSERD
)
1945 || osdmap
->test_flag(CEPH_OSDMAP_PAUSEWR
)) {
1946 ldout(cct
, 10) << "_maybe_request_map subscribing (continuous) to next "
1947 "osd map (FULL flag is set)" << dendl
;
1950 << "_maybe_request_map subscribing (onetime) to next osd map" << dendl
;
1951 flag
= CEPH_SUBSCRIBE_ONETIME
;
1953 epoch_t epoch
= osdmap
->get_epoch() ? osdmap
->get_epoch()+1 : 0;
1954 if (monc
->sub_want("osdmap", epoch
, flag
)) {
1959 void Objecter::_wait_for_new_map(std::unique_ptr
<OpCompletion
> c
, epoch_t epoch
,
1962 // rwlock is locked unique
1963 waiting_for_map
[epoch
].emplace_back(std::move(c
), ec
);
1964 _maybe_request_map();
1969 * Use this together with wait_for_map: this is a pre-check to avoid
1970 * allocating a Context for wait_for_map if we can see that we
1971 * definitely already have the epoch.
1973 * This does *not* replace the need to handle the return value of
1974 * wait_for_map: just because we don't have it in this pre-check
1975 * doesn't mean we won't have it when calling back into wait_for_map,
1976 * since the objecter lock is dropped in between.
1978 bool Objecter::have_map(const epoch_t epoch
)
1980 shared_lock
rl(rwlock
);
1981 if (osdmap
->get_epoch() >= epoch
) {
1988 void Objecter::_kick_requests(OSDSession
*session
,
1989 map
<uint64_t, LingerOp
*>& lresend
)
1991 // rwlock is locked unique
1994 session
->backoffs
.clear();
1995 session
->backoffs_by_id
.clear();
1998 map
<ceph_tid_t
,Op
*> resend
; // resend in tid order
1999 for (auto p
= session
->ops
.begin(); p
!= session
->ops
.end();) {
2002 if (op
->should_resend
) {
2003 if (!op
->target
.paused
)
2004 resend
[op
->tid
] = op
;
2006 _op_cancel_map_check(op
);
2007 _cancel_linger_op(op
);
2011 logger
->inc(l_osdc_op_resend
, resend
.size());
2012 while (!resend
.empty()) {
2013 _send_op(resend
.begin()->second
);
2014 resend
.erase(resend
.begin());
2018 logger
->inc(l_osdc_linger_resend
, session
->linger_ops
.size());
2019 for (auto j
= session
->linger_ops
.begin();
2020 j
!= session
->linger_ops
.end(); ++j
) {
2021 LingerOp
*op
= j
->second
;
2023 ceph_assert(lresend
.count(j
->first
) == 0);
2024 lresend
[j
->first
] = op
;
2028 logger
->inc(l_osdc_command_resend
, session
->command_ops
.size());
2029 map
<uint64_t,CommandOp
*> cresend
; // resend in order
2030 for (auto k
= session
->command_ops
.begin();
2031 k
!= session
->command_ops
.end(); ++k
) {
2032 cresend
[k
->first
] = k
->second
;
2034 while (!cresend
.empty()) {
2035 _send_command(cresend
.begin()->second
);
2036 cresend
.erase(cresend
.begin());
2040 void Objecter::_linger_ops_resend(map
<uint64_t, LingerOp
*>& lresend
,
2041 unique_lock
<ceph::shared_mutex
>& ul
)
2043 ceph_assert(ul
.owns_lock());
2044 shunique_lock
sul(std::move(ul
));
2045 while (!lresend
.empty()) {
2046 LingerOp
*op
= lresend
.begin()->second
;
2047 if (!op
->canceled
) {
2048 _send_linger(op
, sul
);
2051 lresend
.erase(lresend
.begin());
2053 ul
= sul
.release_to_unique();
2056 void Objecter::start_tick()
2058 ceph_assert(tick_event
== 0);
2060 timer
.add_event(ceph::make_timespan(cct
->_conf
->objecter_tick_interval
),
2061 &Objecter::tick
, this);
2064 void Objecter::tick()
2066 shared_lock
rl(rwlock
);
2068 ldout(cct
, 10) << "tick" << dendl
;
2070 // we are only called by C_Tick
2074 // we raced with shutdown
2075 ldout(cct
, 10) << __func__
<< " raced with shutdown" << dendl
;
2079 set
<OSDSession
*> toping
;
2082 // look for laggy requests
2083 auto cutoff
= ceph::coarse_mono_clock::now();
2084 cutoff
-= ceph::make_timespan(cct
->_conf
->objecter_timeout
); // timeout
2086 unsigned laggy_ops
= 0;
2088 for (auto siter
= osd_sessions
.begin();
2089 siter
!= osd_sessions
.end(); ++siter
) {
2090 auto s
= siter
->second
;
2091 scoped_lock
l(s
->lock
);
2093 for (auto p
= s
->ops
.begin(); p
!= s
->ops
.end(); ++p
) {
2094 auto op
= p
->second
;
2095 ceph_assert(op
->session
);
2096 if (op
->stamp
< cutoff
) {
2097 ldout(cct
, 2) << " tid " << p
->first
<< " on osd." << op
->session
->osd
2098 << " is laggy" << dendl
;
2103 for (auto p
= s
->linger_ops
.begin();
2104 p
!= s
->linger_ops
.end();
2106 auto op
= p
->second
;
2107 std::unique_lock
wl(op
->watch_lock
);
2108 ceph_assert(op
->session
);
2109 ldout(cct
, 10) << " pinging osd that serves lingering tid " << p
->first
2110 << " (osd." << op
->session
->osd
<< ")" << dendl
;
2112 if (op
->is_watch
&& op
->registered
&& !op
->last_error
)
2113 _send_linger_ping(op
);
2115 for (auto p
= s
->command_ops
.begin();
2116 p
!= s
->command_ops
.end();
2118 auto op
= p
->second
;
2119 ceph_assert(op
->session
);
2120 ldout(cct
, 10) << " pinging osd that serves command tid " << p
->first
2121 << " (osd." << op
->session
->osd
<< ")" << dendl
;
2127 if (num_homeless_ops
|| !toping
.empty()) {
2128 _maybe_request_map();
2131 logger
->set(l_osdc_op_laggy
, laggy_ops
);
2132 logger
->set(l_osdc_osd_laggy
, toping
.size());
2134 if (!toping
.empty()) {
2135 // send a ping to these osds, to ensure we detect any session resets
2136 // (osd reply message policy is lossy)
2137 for (auto i
= toping
.begin(); i
!= toping
.end(); ++i
) {
2138 (*i
)->con
->send_message(new MPing
);
2142 // Make sure we don't reschedule if we wake up after shutdown
2144 tick_event
= timer
.reschedule_me(ceph::make_timespan(
2145 cct
->_conf
->objecter_tick_interval
));
2149 void Objecter::resend_mon_ops()
2151 unique_lock
wl(rwlock
);
2153 ldout(cct
, 10) << "resend_mon_ops" << dendl
;
2155 for (auto p
= poolstat_ops
.begin(); p
!= poolstat_ops
.end(); ++p
) {
2156 _poolstat_submit(p
->second
);
2157 logger
->inc(l_osdc_poolstat_resend
);
2160 for (auto p
= statfs_ops
.begin(); p
!= statfs_ops
.end(); ++p
) {
2161 _fs_stats_submit(p
->second
);
2162 logger
->inc(l_osdc_statfs_resend
);
2165 for (auto p
= pool_ops
.begin(); p
!= pool_ops
.end(); ++p
) {
2166 _pool_op_submit(p
->second
);
2167 logger
->inc(l_osdc_poolop_resend
);
2170 for (auto p
= check_latest_map_ops
.begin();
2171 p
!= check_latest_map_ops
.end();
2173 monc
->get_version("osdmap", CB_Op_Map_Latest(this, p
->second
->tid
));
2176 for (auto p
= check_latest_map_lingers
.begin();
2177 p
!= check_latest_map_lingers
.end();
2179 monc
->get_version("osdmap", CB_Linger_Map_Latest(this, p
->second
->linger_id
));
2182 for (auto p
= check_latest_map_commands
.begin();
2183 p
!= check_latest_map_commands
.end();
2185 monc
->get_version("osdmap", CB_Command_Map_Latest(this, p
->second
->tid
));
2189 // read | write ---------------------------
2191 void Objecter::op_submit(Op
*op
, ceph_tid_t
*ptid
, int *ctx_budget
)
2193 shunique_lock
rl(rwlock
, ceph::acquire_shared
);
2197 op
->trace
.event("op submit");
2198 _op_submit_with_budget(op
, rl
, ptid
, ctx_budget
);
2201 void Objecter::_op_submit_with_budget(Op
*op
,
2202 shunique_lock
<ceph::shared_mutex
>& sul
,
2206 ceph_assert(initialized
);
2208 ceph_assert(op
->ops
.size() == op
->out_bl
.size());
2209 ceph_assert(op
->ops
.size() == op
->out_rval
.size());
2210 ceph_assert(op
->ops
.size() == op
->out_handler
.size());
2212 // throttle. before we look at any state, because
2213 // _take_op_budget() may drop our lock while it blocks.
2214 if (!op
->ctx_budgeted
|| (ctx_budget
&& (*ctx_budget
== -1))) {
2215 int op_budget
= _take_op_budget(op
, sul
);
2216 // take and pass out the budget for the first OP
2217 // in the context session
2218 if (ctx_budget
&& (*ctx_budget
== -1)) {
2219 *ctx_budget
= op_budget
;
2223 if (osd_timeout
> timespan(0)) {
2225 op
->tid
= ++last_tid
;
2227 op
->ontimeout
= timer
.add_event(osd_timeout
,
2229 op_cancel(tid
, -ETIMEDOUT
); });
2232 _op_submit(op
, sul
, ptid
);
2235 void Objecter::_send_op_account(Op
*op
)
2239 // add to gather set(s)
2240 if (op
->has_completion()) {
2243 ldout(cct
, 20) << " note: not requesting reply" << dendl
;
2246 logger
->inc(l_osdc_op_active
);
2247 logger
->inc(l_osdc_op
);
2248 logger
->inc(l_osdc_oplen_avg
, op
->ops
.size());
2250 if ((op
->target
.flags
& (CEPH_OSD_FLAG_READ
| CEPH_OSD_FLAG_WRITE
)) ==
2251 (CEPH_OSD_FLAG_READ
|CEPH_OSD_FLAG_WRITE
))
2252 logger
->inc(l_osdc_op_rmw
);
2253 else if (op
->target
.flags
& CEPH_OSD_FLAG_WRITE
)
2254 logger
->inc(l_osdc_op_w
);
2255 else if (op
->target
.flags
& CEPH_OSD_FLAG_READ
)
2256 logger
->inc(l_osdc_op_r
);
2258 if (op
->target
.flags
& CEPH_OSD_FLAG_PGOP
)
2259 logger
->inc(l_osdc_op_pg
);
2261 for (auto p
= op
->ops
.begin(); p
!= op
->ops
.end(); ++p
) {
2262 int code
= l_osdc_osdop_other
;
2264 case CEPH_OSD_OP_STAT
: code
= l_osdc_osdop_stat
; break;
2265 case CEPH_OSD_OP_CREATE
: code
= l_osdc_osdop_create
; break;
2266 case CEPH_OSD_OP_READ
: code
= l_osdc_osdop_read
; break;
2267 case CEPH_OSD_OP_WRITE
: code
= l_osdc_osdop_write
; break;
2268 case CEPH_OSD_OP_WRITEFULL
: code
= l_osdc_osdop_writefull
; break;
2269 case CEPH_OSD_OP_WRITESAME
: code
= l_osdc_osdop_writesame
; break;
2270 case CEPH_OSD_OP_APPEND
: code
= l_osdc_osdop_append
; break;
2271 case CEPH_OSD_OP_ZERO
: code
= l_osdc_osdop_zero
; break;
2272 case CEPH_OSD_OP_TRUNCATE
: code
= l_osdc_osdop_truncate
; break;
2273 case CEPH_OSD_OP_DELETE
: code
= l_osdc_osdop_delete
; break;
2274 case CEPH_OSD_OP_MAPEXT
: code
= l_osdc_osdop_mapext
; break;
2275 case CEPH_OSD_OP_SPARSE_READ
: code
= l_osdc_osdop_sparse_read
; break;
2276 case CEPH_OSD_OP_GETXATTR
: code
= l_osdc_osdop_getxattr
; break;
2277 case CEPH_OSD_OP_SETXATTR
: code
= l_osdc_osdop_setxattr
; break;
2278 case CEPH_OSD_OP_CMPXATTR
: code
= l_osdc_osdop_cmpxattr
; break;
2279 case CEPH_OSD_OP_RMXATTR
: code
= l_osdc_osdop_rmxattr
; break;
2280 case CEPH_OSD_OP_RESETXATTRS
: code
= l_osdc_osdop_resetxattrs
; break;
2282 // OMAP read operations
2283 case CEPH_OSD_OP_OMAPGETVALS
:
2284 case CEPH_OSD_OP_OMAPGETKEYS
:
2285 case CEPH_OSD_OP_OMAPGETHEADER
:
2286 case CEPH_OSD_OP_OMAPGETVALSBYKEYS
:
2287 case CEPH_OSD_OP_OMAP_CMP
: code
= l_osdc_osdop_omap_rd
; break;
2289 // OMAP write operations
2290 case CEPH_OSD_OP_OMAPSETVALS
:
2291 case CEPH_OSD_OP_OMAPSETHEADER
: code
= l_osdc_osdop_omap_wr
; break;
2293 // OMAP del operations
2294 case CEPH_OSD_OP_OMAPCLEAR
:
2295 case CEPH_OSD_OP_OMAPRMKEYS
: code
= l_osdc_osdop_omap_del
; break;
2297 case CEPH_OSD_OP_CALL
: code
= l_osdc_osdop_call
; break;
2298 case CEPH_OSD_OP_WATCH
: code
= l_osdc_osdop_watch
; break;
2299 case CEPH_OSD_OP_NOTIFY
: code
= l_osdc_osdop_notify
; break;
2306 void Objecter::_op_submit(Op
*op
, shunique_lock
<ceph::shared_mutex
>& sul
, ceph_tid_t
*ptid
)
2310 ldout(cct
, 10) << __func__
<< " op " << op
<< dendl
;
2313 ceph_assert(op
->session
== NULL
);
2314 OSDSession
*s
= NULL
;
2316 bool check_for_latest_map
= _calc_target(&op
->target
, nullptr)
2317 == RECALC_OP_TARGET_POOL_DNE
;
2319 // Try to get a session, including a retry if we need to take write lock
2320 int r
= _get_session(op
->target
.osd
, &s
, sul
);
2322 (check_for_latest_map
&& sul
.owns_lock_shared()) ||
2323 cct
->_conf
->objecter_debug_inject_relock_delay
) {
2324 epoch_t orig_epoch
= osdmap
->get_epoch();
2326 if (cct
->_conf
->objecter_debug_inject_relock_delay
) {
2330 if (orig_epoch
!= osdmap
->get_epoch()) {
2331 // map changed; recalculate mapping
2332 ldout(cct
, 10) << __func__
<< " relock raced with osdmap, recalc target"
2334 check_for_latest_map
= _calc_target(&op
->target
, nullptr)
2335 == RECALC_OP_TARGET_POOL_DNE
;
2344 ceph_assert(s
== NULL
);
2345 r
= _get_session(op
->target
.osd
, &s
, sul
);
2347 ceph_assert(r
== 0);
2348 ceph_assert(s
); // may be homeless
2350 _send_op_account(op
);
2354 ceph_assert(op
->target
.flags
& (CEPH_OSD_FLAG_READ
|CEPH_OSD_FLAG_WRITE
));
2356 bool need_send
= false;
2357 if (op
->target
.paused
) {
2358 ldout(cct
, 10) << " tid " << op
->tid
<< " op " << op
<< " is paused"
2360 _maybe_request_map();
2361 } else if (!s
->is_homeless()) {
2364 _maybe_request_map();
2367 unique_lock
sl(s
->lock
);
2369 op
->tid
= ++last_tid
;
2371 ldout(cct
, 10) << "_op_submit oid " << op
->target
.base_oid
2372 << " '" << op
->target
.base_oloc
<< "' '"
2373 << op
->target
.target_oloc
<< "' " << op
->ops
<< " tid "
2374 << op
->tid
<< " osd." << (!s
->is_homeless() ? s
->osd
: -1)
2377 _session_op_assign(s
, op
);
2383 // Last chance to touch Op here, after giving up session lock it can
2384 // be freed at any time by response handler.
2385 ceph_tid_t tid
= op
->tid
;
2386 if (check_for_latest_map
) {
2387 _send_op_map_check(op
);
2396 ldout(cct
, 5) << num_in_flight
<< " in flight" << dendl
;
2399 int Objecter::op_cancel(OSDSession
*s
, ceph_tid_t tid
, int r
)
2401 ceph_assert(initialized
);
2403 unique_lock
sl(s
->lock
);
2405 auto p
= s
->ops
.find(tid
);
2406 if (p
== s
->ops
.end()) {
2407 ldout(cct
, 10) << __func__
<< " tid " << tid
<< " dne in session "
2414 ldout(cct
, 20) << " revoking rx ceph::buffer for " << tid
2415 << " on " << s
->con
<< dendl
;
2416 s
->con
->revoke_rx_buffer(tid
);
2420 ldout(cct
, 10) << __func__
<< " tid " << tid
<< " in session " << s
->osd
2423 if (op
->has_completion()) {
2425 op
->complete(osdcode(r
), r
);
2427 _op_cancel_map_check(op
);
2434 int Objecter::op_cancel(ceph_tid_t tid
, int r
)
2438 unique_lock
wl(rwlock
);
2439 ret
= _op_cancel(tid
, r
);
2444 int Objecter::op_cancel(const vector
<ceph_tid_t
>& tids
, int r
)
2446 unique_lock
wl(rwlock
);
2447 ldout(cct
,10) << __func__
<< " " << tids
<< dendl
;
2448 for (auto tid
: tids
) {
2454 int Objecter::_op_cancel(ceph_tid_t tid
, int r
)
2458 ldout(cct
, 5) << __func__
<< ": cancelling tid " << tid
<< " r=" << r
2463 for (auto siter
= osd_sessions
.begin();
2464 siter
!= osd_sessions
.end(); ++siter
) {
2465 OSDSession
*s
= siter
->second
;
2466 shared_lock
sl(s
->lock
);
2467 if (s
->ops
.find(tid
) != s
->ops
.end()) {
2469 ret
= op_cancel(s
, tid
, r
);
2470 if (ret
== -ENOENT
) {
2471 /* oh no! raced, maybe tid moved to another session, restarting */
2478 ldout(cct
, 5) << __func__
<< ": tid " << tid
2479 << " not found in live sessions" << dendl
;
2481 // Handle case where the op is in homeless session
2482 shared_lock
sl(homeless_session
->lock
);
2483 if (homeless_session
->ops
.find(tid
) != homeless_session
->ops
.end()) {
2485 ret
= op_cancel(homeless_session
, tid
, r
);
2486 if (ret
== -ENOENT
) {
2487 /* oh no! raced, maybe tid moved to another session, restarting */
2496 ldout(cct
, 5) << __func__
<< ": tid " << tid
2497 << " not found in homeless session" << dendl
;
2503 epoch_t
Objecter::op_cancel_writes(int r
, int64_t pool
)
2505 unique_lock
wl(rwlock
);
2507 std::vector
<ceph_tid_t
> to_cancel
;
2510 for (auto siter
= osd_sessions
.begin();
2511 siter
!= osd_sessions
.end(); ++siter
) {
2512 OSDSession
*s
= siter
->second
;
2513 shared_lock
sl(s
->lock
);
2514 for (auto op_i
= s
->ops
.begin();
2515 op_i
!= s
->ops
.end(); ++op_i
) {
2516 if (op_i
->second
->target
.flags
& CEPH_OSD_FLAG_WRITE
2517 && (pool
== -1 || op_i
->second
->target
.target_oloc
.pool
== pool
)) {
2518 to_cancel
.push_back(op_i
->first
);
2523 for (auto titer
= to_cancel
.begin(); titer
!= to_cancel
.end(); ++titer
) {
2524 int cancel_result
= op_cancel(s
, *titer
, r
);
2525 // We hold rwlock across search and cancellation, so cancels
2526 // should always succeed
2527 ceph_assert(cancel_result
== 0);
2529 if (!found
&& to_cancel
.size())
2534 const epoch_t epoch
= osdmap
->get_epoch();
2545 bool Objecter::is_pg_changed(
2547 const vector
<int>& oldacting
,
2549 const vector
<int>& newacting
,
2552 if (OSDMap::primary_changed_broken( // https://tracker.ceph.com/issues/43213
2558 if (any_change
&& oldacting
!= newacting
)
2560 return false; // same primary (tho replicas may have changed)
2563 bool Objecter::target_should_be_paused(op_target_t
*t
)
2565 const pg_pool_t
*pi
= osdmap
->get_pg_pool(t
->base_oloc
.pool
);
2566 bool pauserd
= osdmap
->test_flag(CEPH_OSDMAP_PAUSERD
);
2567 bool pausewr
= osdmap
->test_flag(CEPH_OSDMAP_PAUSEWR
) ||
2568 (t
->respects_full() && (_osdmap_full_flag() || _osdmap_pool_full(*pi
)));
2570 return (t
->flags
& CEPH_OSD_FLAG_READ
&& pauserd
) ||
2571 (t
->flags
& CEPH_OSD_FLAG_WRITE
&& pausewr
) ||
2572 (osdmap
->get_epoch() < epoch_barrier
);
2576 * Locking public accessor for _osdmap_full_flag
2578 bool Objecter::osdmap_full_flag() const
2580 shared_lock
rl(rwlock
);
2582 return _osdmap_full_flag();
2585 bool Objecter::osdmap_pool_full(const int64_t pool_id
) const
2587 shared_lock
rl(rwlock
);
2589 if (_osdmap_full_flag()) {
2593 return _osdmap_pool_full(pool_id
);
2596 bool Objecter::_osdmap_pool_full(const int64_t pool_id
) const
2598 const pg_pool_t
*pool
= osdmap
->get_pg_pool(pool_id
);
2600 ldout(cct
, 4) << __func__
<< ": DNE pool " << pool_id
<< dendl
;
2604 return _osdmap_pool_full(*pool
);
2607 bool Objecter::_osdmap_has_pool_full() const
2609 for (auto it
= osdmap
->get_pools().begin();
2610 it
!= osdmap
->get_pools().end(); ++it
) {
2611 if (_osdmap_pool_full(it
->second
))
2618 * Wrapper around osdmap->test_flag for special handling of the FULL flag.
2620 bool Objecter::_osdmap_full_flag() const
2622 // Ignore the FULL flag if the caller does not have honor_osdmap_full
2623 return osdmap
->test_flag(CEPH_OSDMAP_FULL
) && honor_pool_full
;
2626 void Objecter::update_pool_full_map(map
<int64_t, bool>& pool_full_map
)
2628 for (map
<int64_t, pg_pool_t
>::const_iterator it
2629 = osdmap
->get_pools().begin();
2630 it
!= osdmap
->get_pools().end(); ++it
) {
2631 if (pool_full_map
.find(it
->first
) == pool_full_map
.end()) {
2632 pool_full_map
[it
->first
] = _osdmap_pool_full(it
->second
);
2634 pool_full_map
[it
->first
] = _osdmap_pool_full(it
->second
) ||
2635 pool_full_map
[it
->first
];
2640 int64_t Objecter::get_object_hash_position(int64_t pool
, const string
& key
,
2643 shared_lock
rl(rwlock
);
2644 const pg_pool_t
*p
= osdmap
->get_pg_pool(pool
);
2647 return p
->hash_key(key
, ns
);
2650 int64_t Objecter::get_object_pg_hash_position(int64_t pool
, const string
& key
,
2653 shared_lock
rl(rwlock
);
2654 const pg_pool_t
*p
= osdmap
->get_pg_pool(pool
);
2657 return p
->raw_hash_to_pg(p
->hash_key(key
, ns
));
2660 void Objecter::_prune_snapc(
2661 const mempool::osdmap::map
<int64_t,
2662 snap_interval_set_t
>& new_removed_snaps
,
2666 auto i
= new_removed_snaps
.find(op
->target
.base_pgid
.pool());
2667 if (i
!= new_removed_snaps
.end()) {
2668 for (auto s
: op
->snapc
.snaps
) {
2669 if (i
->second
.contains(s
)) {
2675 vector
<snapid_t
> new_snaps
;
2676 for (auto s
: op
->snapc
.snaps
) {
2677 if (!i
->second
.contains(s
)) {
2678 new_snaps
.push_back(s
);
2681 op
->snapc
.snaps
.swap(new_snaps
);
2682 ldout(cct
,10) << __func__
<< " op " << op
->tid
<< " snapc " << op
->snapc
2683 << " (was " << new_snaps
<< ")" << dendl
;
2688 int Objecter::_calc_target(op_target_t
*t
, Connection
*con
, bool any_change
)
2691 bool is_read
= t
->flags
& CEPH_OSD_FLAG_READ
;
2692 bool is_write
= t
->flags
& CEPH_OSD_FLAG_WRITE
;
2693 t
->epoch
= osdmap
->get_epoch();
2694 ldout(cct
,20) << __func__
<< " epoch " << t
->epoch
2695 << " base " << t
->base_oid
<< " " << t
->base_oloc
2696 << " precalc_pgid " << (int)t
->precalc_pgid
2697 << " pgid " << t
->base_pgid
2698 << (is_read
? " is_read" : "")
2699 << (is_write
? " is_write" : "")
2702 const pg_pool_t
*pi
= osdmap
->get_pg_pool(t
->base_oloc
.pool
);
2705 return RECALC_OP_TARGET_POOL_DNE
;
2707 ldout(cct
,30) << __func__
<< " base pi " << pi
2708 << " pg_num " << pi
->get_pg_num() << dendl
;
2710 bool force_resend
= false;
2711 if (osdmap
->get_epoch() == pi
->last_force_op_resend
) {
2712 if (t
->last_force_resend
< pi
->last_force_op_resend
) {
2713 t
->last_force_resend
= pi
->last_force_op_resend
;
2714 force_resend
= true;
2715 } else if (t
->last_force_resend
== 0) {
2716 force_resend
= true;
2721 t
->target_oid
= t
->base_oid
;
2722 t
->target_oloc
= t
->base_oloc
;
2723 if ((t
->flags
& CEPH_OSD_FLAG_IGNORE_OVERLAY
) == 0) {
2724 if (is_read
&& pi
->has_read_tier())
2725 t
->target_oloc
.pool
= pi
->read_tier
;
2726 if (is_write
&& pi
->has_write_tier())
2727 t
->target_oloc
.pool
= pi
->write_tier
;
2728 pi
= osdmap
->get_pg_pool(t
->target_oloc
.pool
);
2731 return RECALC_OP_TARGET_POOL_DNE
;
2736 if (t
->precalc_pgid
) {
2737 ceph_assert(t
->flags
& CEPH_OSD_FLAG_IGNORE_OVERLAY
);
2738 ceph_assert(t
->base_oid
.name
.empty()); // make sure this is a pg op
2739 ceph_assert(t
->base_oloc
.pool
== (int64_t)t
->base_pgid
.pool());
2740 pgid
= t
->base_pgid
;
2742 int ret
= osdmap
->object_locator_to_pg(t
->target_oid
, t
->target_oloc
,
2744 if (ret
== -ENOENT
) {
2746 return RECALC_OP_TARGET_POOL_DNE
;
2749 ldout(cct
,20) << __func__
<< " target " << t
->target_oid
<< " "
2750 << t
->target_oloc
<< " -> pgid " << pgid
<< dendl
;
2751 ldout(cct
,30) << __func__
<< " target pi " << pi
2752 << " pg_num " << pi
->get_pg_num() << dendl
;
2753 t
->pool_ever_existed
= true;
2755 int size
= pi
->size
;
2756 int min_size
= pi
->min_size
;
2757 unsigned pg_num
= pi
->get_pg_num();
2758 unsigned pg_num_mask
= pi
->get_pg_num_mask();
2759 unsigned pg_num_pending
= pi
->get_pg_num_pending();
2760 int up_primary
, acting_primary
;
2761 vector
<int> up
, acting
;
2762 ps_t actual_ps
= ceph_stable_mod(pgid
.ps(), pg_num
, pg_num_mask
);
2763 pg_t
actual_pgid(actual_ps
, pgid
.pool());
2764 pg_mapping_t pg_mapping
;
2765 pg_mapping
.epoch
= osdmap
->get_epoch();
2766 if (lookup_pg_mapping(actual_pgid
, &pg_mapping
)) {
2768 up_primary
= pg_mapping
.up_primary
;
2769 acting
= pg_mapping
.acting
;
2770 acting_primary
= pg_mapping
.acting_primary
;
2772 osdmap
->pg_to_up_acting_osds(actual_pgid
, &up
, &up_primary
,
2773 &acting
, &acting_primary
);
2774 pg_mapping_t
pg_mapping(osdmap
->get_epoch(),
2775 up
, up_primary
, acting
, acting_primary
);
2776 update_pg_mapping(actual_pgid
, std::move(pg_mapping
));
2778 bool sort_bitwise
= osdmap
->test_flag(CEPH_OSDMAP_SORTBITWISE
);
2779 bool recovery_deletes
= osdmap
->test_flag(CEPH_OSDMAP_RECOVERY_DELETES
);
2780 unsigned prev_seed
= ceph_stable_mod(pgid
.ps(), t
->pg_num
, t
->pg_num_mask
);
2781 pg_t
prev_pgid(prev_seed
, pgid
.pool());
2782 if (any_change
&& PastIntervals::is_new_interval(
2801 t
->recovery_deletes
,
2803 t
->peering_crush_bucket_count
,
2804 pi
->peering_crush_bucket_count
,
2805 t
->peering_crush_bucket_target
,
2806 pi
->peering_crush_bucket_target
,
2807 t
->peering_crush_bucket_barrier
,
2808 pi
->peering_crush_bucket_barrier
,
2809 t
->peering_crush_mandatory_member
,
2810 pi
->peering_crush_mandatory_member
,
2812 force_resend
= true;
2815 bool unpaused
= false;
2816 bool should_be_paused
= target_should_be_paused(t
);
2817 if (t
->paused
&& !should_be_paused
) {
2820 if (t
->paused
!= should_be_paused
) {
2821 ldout(cct
, 10) << __func__
<< " paused " << t
->paused
2822 << " -> " << should_be_paused
<< dendl
;
2823 t
->paused
= should_be_paused
;
2826 bool legacy_change
=
2829 t
->acting_primary
, t
->acting
, acting_primary
, acting
,
2830 t
->used_replica
|| any_change
);
2831 bool split_or_merge
= false;
2834 prev_pgid
.is_split(t
->pg_num
, pg_num
, nullptr) ||
2835 prev_pgid
.is_merge_source(t
->pg_num
, pg_num
, nullptr) ||
2836 prev_pgid
.is_merge_target(t
->pg_num
, pg_num
);
2839 if (legacy_change
|| split_or_merge
|| force_resend
) {
2842 t
->acting_primary
= acting_primary
;
2843 t
->up_primary
= up_primary
;
2846 t
->min_size
= min_size
;
2848 t
->pg_num_mask
= pg_num_mask
;
2849 t
->pg_num_pending
= pg_num_pending
;
2850 spg_t
spgid(actual_pgid
);
2851 if (pi
->is_erasure()) {
2852 for (uint8_t i
= 0; i
< acting
.size(); ++i
) {
2853 if (acting
[i
] == acting_primary
) {
2854 spgid
.reset_shard(shard_id_t(i
));
2859 t
->actual_pgid
= spgid
;
2860 t
->sort_bitwise
= sort_bitwise
;
2861 t
->recovery_deletes
= recovery_deletes
;
2862 t
->peering_crush_bucket_count
= pi
->peering_crush_bucket_count
;
2863 t
->peering_crush_bucket_target
= pi
->peering_crush_bucket_target
;
2864 t
->peering_crush_bucket_barrier
= pi
->peering_crush_bucket_barrier
;
2865 t
->peering_crush_mandatory_member
= pi
->peering_crush_mandatory_member
;
2866 ldout(cct
, 10) << __func__
<< " "
2867 << " raw pgid " << pgid
<< " -> actual " << t
->actual_pgid
2868 << " acting " << acting
2869 << " primary " << acting_primary
<< dendl
;
2870 t
->used_replica
= false;
2871 if ((t
->flags
& (CEPH_OSD_FLAG_BALANCE_READS
|
2872 CEPH_OSD_FLAG_LOCALIZE_READS
)) &&
2873 !is_write
&& pi
->is_replicated() && acting
.size() > 1) {
2875 ceph_assert(is_read
&& acting
[0] == acting_primary
);
2876 if (t
->flags
& CEPH_OSD_FLAG_BALANCE_READS
) {
2877 int p
= rand() % acting
.size();
2879 t
->used_replica
= true;
2881 ldout(cct
, 10) << " chose random osd." << osd
<< " of " << acting
2884 // look for a local replica. prefer the primary if the
2885 // distance is the same.
2887 int best_locality
= 0;
2888 for (unsigned i
= 0; i
< acting
.size(); ++i
) {
2889 int locality
= osdmap
->crush
->get_common_ancestor_distance(
2890 cct
, acting
[i
], crush_location
);
2891 ldout(cct
, 20) << __func__
<< " localize: rank " << i
2892 << " osd." << acting
[i
]
2893 << " locality " << locality
<< dendl
;
2895 (locality
>= 0 && best_locality
>= 0 &&
2896 locality
< best_locality
) ||
2897 (best_locality
< 0 && locality
>= 0)) {
2899 best_locality
= locality
;
2901 t
->used_replica
= true;
2904 ceph_assert(best
>= 0);
2909 t
->osd
= acting_primary
;
2912 if (legacy_change
|| unpaused
|| force_resend
) {
2913 return RECALC_OP_TARGET_NEED_RESEND
;
2915 if (split_or_merge
&&
2916 (osdmap
->require_osd_release
>= ceph_release_t::luminous
||
2917 HAVE_FEATURE(osdmap
->get_xinfo(acting_primary
).features
,
2918 RESEND_ON_SPLIT
))) {
2919 return RECALC_OP_TARGET_NEED_RESEND
;
2921 return RECALC_OP_TARGET_NO_ACTION
;
2924 int Objecter::_map_session(op_target_t
*target
, OSDSession
**s
,
2925 shunique_lock
<ceph::shared_mutex
>& sul
)
2927 _calc_target(target
, nullptr);
2928 return _get_session(target
->osd
, s
, sul
);
2931 void Objecter::_session_op_assign(OSDSession
*to
, Op
*op
)
2933 // to->lock is locked
2934 ceph_assert(op
->session
== NULL
);
2935 ceph_assert(op
->tid
);
2939 to
->ops
[op
->tid
] = op
;
2941 if (to
->is_homeless()) {
2945 ldout(cct
, 15) << __func__
<< " " << to
->osd
<< " " << op
->tid
<< dendl
;
2948 void Objecter::_session_op_remove(OSDSession
*from
, Op
*op
)
2950 ceph_assert(op
->session
== from
);
2951 // from->lock is locked
2953 if (from
->is_homeless()) {
2957 from
->ops
.erase(op
->tid
);
2961 ldout(cct
, 15) << __func__
<< " " << from
->osd
<< " " << op
->tid
<< dendl
;
2964 void Objecter::_session_linger_op_assign(OSDSession
*to
, LingerOp
*op
)
2966 // to lock is locked unique
2967 ceph_assert(op
->session
== NULL
);
2969 if (to
->is_homeless()) {
2975 to
->linger_ops
[op
->linger_id
] = op
;
2977 ldout(cct
, 15) << __func__
<< " " << to
->osd
<< " " << op
->linger_id
2981 void Objecter::_session_linger_op_remove(OSDSession
*from
, LingerOp
*op
)
2983 ceph_assert(from
== op
->session
);
2984 // from->lock is locked unique
2986 if (from
->is_homeless()) {
2990 from
->linger_ops
.erase(op
->linger_id
);
2994 ldout(cct
, 15) << __func__
<< " " << from
->osd
<< " " << op
->linger_id
2998 void Objecter::_session_command_op_remove(OSDSession
*from
, CommandOp
*op
)
3000 ceph_assert(from
== op
->session
);
3001 // from->lock is locked
3003 if (from
->is_homeless()) {
3007 from
->command_ops
.erase(op
->tid
);
3011 ldout(cct
, 15) << __func__
<< " " << from
->osd
<< " " << op
->tid
<< dendl
;
3014 void Objecter::_session_command_op_assign(OSDSession
*to
, CommandOp
*op
)
3016 // to->lock is locked
3017 ceph_assert(op
->session
== NULL
);
3018 ceph_assert(op
->tid
);
3020 if (to
->is_homeless()) {
3026 to
->command_ops
[op
->tid
] = op
;
3028 ldout(cct
, 15) << __func__
<< " " << to
->osd
<< " " << op
->tid
<< dendl
;
3031 int Objecter::_recalc_linger_op_target(LingerOp
*linger_op
,
3032 shunique_lock
<ceph::shared_mutex
>& sul
)
3034 // rwlock is locked unique
3036 int r
= _calc_target(&linger_op
->target
, nullptr, true);
3037 if (r
== RECALC_OP_TARGET_NEED_RESEND
) {
3038 ldout(cct
, 10) << "recalc_linger_op_target tid " << linger_op
->linger_id
3039 << " pgid " << linger_op
->target
.pgid
3040 << " acting " << linger_op
->target
.acting
<< dendl
;
3042 OSDSession
*s
= NULL
;
3043 r
= _get_session(linger_op
->target
.osd
, &s
, sul
);
3044 ceph_assert(r
== 0);
3046 if (linger_op
->session
!= s
) {
3047 // NB locking two sessions (s and linger_op->session) at the
3048 // same time here is only safe because we are the only one that
3049 // takes two, and we are holding rwlock for write. We use
3050 // std::shared_mutex in OSDSession because lockdep doesn't know
3052 unique_lock
sl(s
->lock
);
3053 _session_linger_op_remove(linger_op
->session
, linger_op
);
3054 _session_linger_op_assign(s
, linger_op
);
3058 return RECALC_OP_TARGET_NEED_RESEND
;
3063 void Objecter::_cancel_linger_op(Op
*op
)
3065 ldout(cct
, 15) << "cancel_op " << op
->tid
<< dendl
;
3067 ceph_assert(!op
->should_resend
);
3068 if (op
->has_completion()) {
3069 op
->onfinish
= nullptr;
3076 void Objecter::_finish_op(Op
*op
, int r
)
3078 ldout(cct
, 15) << __func__
<< " " << op
->tid
<< dendl
;
3080 // op->session->lock is locked unique or op->session is null
3082 if (!op
->ctx_budgeted
&& op
->budget
>= 0) {
3083 put_op_budget_bytes(op
->budget
);
3087 if (op
->ontimeout
&& r
!= -ETIMEDOUT
)
3088 timer
.cancel_event(op
->ontimeout
);
3091 _session_op_remove(op
->session
, op
);
3094 logger
->dec(l_osdc_op_active
);
3096 ceph_assert(check_latest_map_ops
.find(op
->tid
) == check_latest_map_ops
.end());
3103 Objecter::MOSDOp
*Objecter::_prepare_osd_op(Op
*op
)
3107 int flags
= op
->target
.flags
;
3108 flags
|= CEPH_OSD_FLAG_KNOWN_REDIR
;
3110 // Nothing checks this any longer, but needed for compatibility with
3111 // pre-luminous osds
3112 flags
|= CEPH_OSD_FLAG_ONDISK
;
3114 if (!honor_pool_full
)
3115 flags
|= CEPH_OSD_FLAG_FULL_FORCE
;
3117 op
->target
.paused
= false;
3118 op
->stamp
= ceph::coarse_mono_clock::now();
3120 hobject_t hobj
= op
->target
.get_hobj();
3121 auto m
= new MOSDOp(client_inc
, op
->tid
,
3122 hobj
, op
->target
.actual_pgid
,
3123 osdmap
->get_epoch(),
3124 flags
, op
->features
);
3126 m
->set_snapid(op
->snapid
);
3127 m
->set_snap_seq(op
->snapc
.seq
);
3128 m
->set_snaps(op
->snapc
.snaps
);
3131 m
->set_mtime(op
->mtime
);
3132 m
->set_retry_attempt(op
->attempts
++);
3134 if (!op
->trace
.valid() && cct
->_conf
->osdc_blkin_trace_all
) {
3135 op
->trace
.init("op", &trace_endpoint
);
3139 m
->set_priority(op
->priority
);
3141 m
->set_priority(cct
->_conf
->osd_client_op_priority
);
3143 if (op
->reqid
!= osd_reqid_t()) {
3144 m
->set_reqid(op
->reqid
);
3147 logger
->inc(l_osdc_op_send
);
3149 for (unsigned i
= 0; i
< m
->ops
.size(); i
++) {
3150 sum
+= m
->ops
[i
].indata
.length();
3152 logger
->inc(l_osdc_op_send_bytes
, sum
);
3157 void Objecter::_send_op(Op
*op
)
3160 // op->session->lock is locked
3163 auto p
= op
->session
->backoffs
.find(op
->target
.actual_pgid
);
3164 if (p
!= op
->session
->backoffs
.end()) {
3165 hobject_t hoid
= op
->target
.get_hobj();
3166 auto q
= p
->second
.lower_bound(hoid
);
3167 if (q
!= p
->second
.begin()) {
3169 if (hoid
>= q
->second
.end
) {
3173 if (q
!= p
->second
.end()) {
3174 ldout(cct
, 20) << __func__
<< " ? " << q
->first
<< " [" << q
->second
.begin
3175 << "," << q
->second
.end
<< ")" << dendl
;
3176 int r
= cmp(hoid
, q
->second
.begin
);
3177 if (r
== 0 || (r
> 0 && hoid
< q
->second
.end
)) {
3178 ldout(cct
, 10) << __func__
<< " backoff " << op
->target
.actual_pgid
3179 << " id " << q
->second
.id
<< " on " << hoid
3180 << ", queuing " << op
<< " tid " << op
->tid
<< dendl
;
3186 ceph_assert(op
->tid
> 0);
3187 MOSDOp
*m
= _prepare_osd_op(op
);
3189 if (op
->target
.actual_pgid
!= m
->get_spg()) {
3190 ldout(cct
, 10) << __func__
<< " " << op
->tid
<< " pgid change from "
3191 << m
->get_spg() << " to " << op
->target
.actual_pgid
3192 << ", updating and reencoding" << dendl
;
3193 m
->set_spg(op
->target
.actual_pgid
);
3194 m
->clear_payload(); // reencode
3197 ldout(cct
, 15) << "_send_op " << op
->tid
<< " to "
3198 << op
->target
.actual_pgid
<< " on osd." << op
->session
->osd
3201 ConnectionRef con
= op
->session
->con
;
3205 // preallocated rx ceph::buffer?
3207 ldout(cct
, 20) << " revoking rx ceph::buffer for " << op
->tid
<< " on "
3208 << op
->con
<< dendl
;
3209 op
->con
->revoke_rx_buffer(op
->tid
);
3212 op
->ontimeout
== 0 && // only post rx_buffer if no timeout; see #9582
3213 op
->outbl
->length()) {
3214 op
->outbl
->invalidate_crc(); // messenger writes through c_str()
3215 ldout(cct
, 20) << " posting rx ceph::buffer for " << op
->tid
<< " on " << con
3218 op
->con
->post_rx_buffer(op
->tid
, *op
->outbl
);
3222 op
->incarnation
= op
->session
->incarnation
;
3224 if (op
->trace
.valid()) {
3225 m
->trace
.init("op msg", nullptr, &op
->trace
);
3227 op
->session
->con
->send_message(m
);
3230 int Objecter::calc_op_budget(const bc::small_vector_base
<OSDOp
>& ops
)
3233 for (auto i
= ops
.begin(); i
!= ops
.end(); ++i
) {
3234 if (i
->op
.op
& CEPH_OSD_OP_MODE_WR
) {
3235 op_budget
+= i
->indata
.length();
3236 } else if (ceph_osd_op_mode_read(i
->op
.op
)) {
3237 if (ceph_osd_op_uses_extent(i
->op
.op
)) {
3238 if ((int64_t)i
->op
.extent
.length
> 0)
3239 op_budget
+= (int64_t)i
->op
.extent
.length
;
3240 } else if (ceph_osd_op_type_attr(i
->op
.op
)) {
3241 op_budget
+= i
->op
.xattr
.name_len
+ i
->op
.xattr
.value_len
;
3248 void Objecter::_throttle_op(Op
*op
,
3249 shunique_lock
<ceph::shared_mutex
>& sul
,
3252 ceph_assert(sul
&& sul
.mutex() == &rwlock
);
3253 bool locked_for_write
= sul
.owns_lock();
3256 op_budget
= calc_op_budget(op
->ops
);
3257 if (!op_throttle_bytes
.get_or_fail(op_budget
)) { //couldn't take right now
3259 op_throttle_bytes
.get(op_budget
);
3260 if (locked_for_write
)
3265 if (!op_throttle_ops
.get_or_fail(1)) { //couldn't take right now
3267 op_throttle_ops
.get(1);
3268 if (locked_for_write
)
3275 int Objecter::take_linger_budget(LingerOp
*info
)
3280 /* This function DOES put the passed message before returning */
3281 void Objecter::handle_osd_op_reply(MOSDOpReply
*m
)
3283 ldout(cct
, 10) << "in handle_osd_op_reply" << dendl
;
3286 ceph_tid_t tid
= m
->get_tid();
3288 shunique_lock
sul(rwlock
, ceph::acquire_shared
);
3294 ConnectionRef con
= m
->get_connection();
3295 auto priv
= con
->get_priv();
3296 auto s
= static_cast<OSDSession
*>(priv
.get());
3297 if (!s
|| s
->con
!= con
) {
3298 ldout(cct
, 7) << __func__
<< " no session on con " << con
<< dendl
;
3303 unique_lock
sl(s
->lock
);
3305 map
<ceph_tid_t
, Op
*>::iterator iter
= s
->ops
.find(tid
);
3306 if (iter
== s
->ops
.end()) {
3307 ldout(cct
, 7) << "handle_osd_op_reply " << tid
3308 << (m
->is_ondisk() ? " ondisk" : (m
->is_onnvram() ?
3309 " onnvram" : " ack"))
3310 << " ... stray" << dendl
;
3316 ldout(cct
, 7) << "handle_osd_op_reply " << tid
3317 << (m
->is_ondisk() ? " ondisk" :
3318 (m
->is_onnvram() ? " onnvram" : " ack"))
3319 << " uv " << m
->get_user_version()
3320 << " in " << m
->get_pg()
3321 << " attempt " << m
->get_retry_attempt()
3323 Op
*op
= iter
->second
;
3324 op
->trace
.event("osd op reply");
3326 if (retry_writes_after_first_reply
&& op
->attempts
== 1 &&
3327 (op
->target
.flags
& CEPH_OSD_FLAG_WRITE
)) {
3328 ldout(cct
, 7) << "retrying write after first reply: " << tid
<< dendl
;
3329 if (op
->has_completion()) {
3332 _session_op_remove(s
, op
);
3335 _op_submit(op
, sul
, NULL
);
3340 if (m
->get_retry_attempt() >= 0) {
3341 if (m
->get_retry_attempt() != (op
->attempts
- 1)) {
3342 ldout(cct
, 7) << " ignoring reply from attempt "
3343 << m
->get_retry_attempt()
3344 << " from " << m
->get_source_inst()
3345 << "; last attempt " << (op
->attempts
- 1) << " sent to "
3346 << op
->session
->con
->get_peer_addr() << dendl
;
3352 // we don't know the request attempt because the server is old, so
3353 // just accept this one. we may do ACK callbacks we shouldn't
3354 // have, but that is better than doing callbacks out of order.
3357 decltype(op
->onfinish
) onfinish
;
3359 int rc
= m
->get_result();
3361 if (m
->is_redirect_reply()) {
3362 ldout(cct
, 5) << " got redirect reply; redirecting" << dendl
;
3363 if (op
->has_completion())
3365 _session_op_remove(s
, op
);
3368 // FIXME: two redirects could race and reorder
3371 m
->get_redirect().combine_with_locator(op
->target
.target_oloc
,
3372 op
->target
.target_oid
.name
);
3373 op
->target
.flags
|= (CEPH_OSD_FLAG_REDIRECTED
|
3374 CEPH_OSD_FLAG_IGNORE_CACHE
|
3375 CEPH_OSD_FLAG_IGNORE_OVERLAY
);
3376 _op_submit(op
, sul
, NULL
);
3381 if (rc
== -EAGAIN
) {
3382 ldout(cct
, 7) << " got -EAGAIN, resubmitting" << dendl
;
3383 if (op
->has_completion())
3385 _session_op_remove(s
, op
);
3389 op
->target
.flags
&= ~(CEPH_OSD_FLAG_BALANCE_READS
|
3390 CEPH_OSD_FLAG_LOCALIZE_READS
);
3391 op
->target
.pgid
= pg_t();
3392 _op_submit(op
, sul
, NULL
);
3400 *op
->objver
= m
->get_user_version();
3401 if (op
->reply_epoch
)
3402 *op
->reply_epoch
= m
->get_map_epoch();
3403 if (op
->data_offset
)
3404 *op
->data_offset
= m
->get_header().data_off
;
3410 op
->con
->revoke_rx_buffer(op
->tid
);
3412 auto& bl
= m
->get_data();
3413 if (op
->outbl
->length() == bl
.length() &&
3414 bl
.get_num_buffers() <= 1) {
3415 // this is here to keep previous users to *relied* on getting data
3416 // read into existing buffers happy. Notably,
3417 // libradosstriper::RadosStriperImpl::aio_read().
3418 ldout(cct
,10) << __func__
<< " copying resulting " << bl
.length()
3419 << " into existing ceph::buffer of length " << op
->outbl
->length()
3422 t
= std::move(*op
->outbl
);
3423 t
.invalidate_crc(); // we're overwriting the raw buffers via c_str()
3424 bl
.begin().copy(bl
.length(), t
.c_str());
3425 op
->outbl
->substr_of(t
, 0, bl
.length());
3427 m
->claim_data(*op
->outbl
);
3432 // per-op result demuxing
3433 vector
<OSDOp
> out_ops
;
3434 m
->claim_ops(out_ops
);
3436 if (out_ops
.size() != op
->ops
.size())
3437 ldout(cct
, 0) << "WARNING: tid " << op
->tid
<< " reply ops " << out_ops
3438 << " != request ops " << op
->ops
3439 << " from " << m
->get_source_inst() << dendl
;
3441 ceph_assert(op
->ops
.size() == op
->out_bl
.size());
3442 ceph_assert(op
->ops
.size() == op
->out_rval
.size());
3443 ceph_assert(op
->ops
.size() == op
->out_ec
.size());
3444 ceph_assert(op
->ops
.size() == op
->out_handler
.size());
3445 auto pb
= op
->out_bl
.begin();
3446 auto pr
= op
->out_rval
.begin();
3447 auto pe
= op
->out_ec
.begin();
3448 auto ph
= op
->out_handler
.begin();
3449 ceph_assert(op
->out_bl
.size() == op
->out_rval
.size());
3450 ceph_assert(op
->out_bl
.size() == op
->out_handler
.size());
3451 auto p
= out_ops
.begin();
3452 for (unsigned i
= 0;
3453 p
!= out_ops
.end() && pb
!= op
->out_bl
.end();
3454 ++i
, ++p
, ++pb
, ++pr
, ++pe
, ++ph
) {
3455 ldout(cct
, 10) << " op " << i
<< " rval " << p
->rval
3456 << " len " << p
->outdata
.length() << dendl
;
3459 // set rval before running handlers so that handlers
3460 // can change it if e.g. decoding fails
3462 **pr
= ceph_to_hostos_errno(p
->rval
);
3464 **pe
= p
->rval
< 0 ? bs::error_code(-p
->rval
, osd_category()) :
3467 std::move((*ph
))(p
->rval
< 0 ?
3468 bs::error_code(-p
->rval
, osd_category()) :
3470 p
->rval
, p
->outdata
);
3474 // NOTE: we assume that since we only request ONDISK ever we will
3475 // only ever get back one (type of) ack ever.
3477 if (op
->has_completion()) {
3479 onfinish
= std::move(op
->onfinish
);
3480 op
->onfinish
= nullptr;
3482 logger
->inc(l_osdc_op_reply
);
3484 /* get it before we call _finish_op() */
3485 auto completion_lock
= s
->get_lock(op
->target
.base_oid
);
3487 ldout(cct
, 15) << "handle_osd_op_reply completed tid " << tid
<< dendl
;
3490 ldout(cct
, 5) << num_in_flight
<< " in flight" << dendl
;
3492 // serialize completions
3493 if (completion_lock
.mutex()) {
3494 completion_lock
.lock();
3499 if (Op::has_completion(onfinish
)) {
3500 Op::complete(std::move(onfinish
), osdcode(rc
), rc
);
3502 if (completion_lock
.mutex()) {
3503 completion_lock
.unlock();
3509 void Objecter::handle_osd_backoff(MOSDBackoff
*m
)
3511 ldout(cct
, 10) << __func__
<< " " << *m
<< dendl
;
3512 shunique_lock
sul(rwlock
, ceph::acquire_shared
);
3518 ConnectionRef con
= m
->get_connection();
3519 auto priv
= con
->get_priv();
3520 auto s
= static_cast<OSDSession
*>(priv
.get());
3521 if (!s
|| s
->con
!= con
) {
3522 ldout(cct
, 7) << __func__
<< " no session on con " << con
<< dendl
;
3529 unique_lock
sl(s
->lock
);
3532 case CEPH_OSD_BACKOFF_OP_BLOCK
:
3535 OSDBackoff
& b
= s
->backoffs
[m
->pgid
][m
->begin
];
3536 s
->backoffs_by_id
.insert(make_pair(m
->id
, &b
));
3542 // ack with original backoff's epoch so that the osd can discard this if
3543 // there was a pg split.
3544 auto r
= new MOSDBackoff(m
->pgid
, m
->map_epoch
,
3545 CEPH_OSD_BACKOFF_OP_ACK_BLOCK
,
3546 m
->id
, m
->begin
, m
->end
);
3547 // this priority must match the MOSDOps from _prepare_osd_op
3548 r
->set_priority(cct
->_conf
->osd_client_op_priority
);
3549 con
->send_message(r
);
3553 case CEPH_OSD_BACKOFF_OP_UNBLOCK
:
3555 auto p
= s
->backoffs_by_id
.find(m
->id
);
3556 if (p
!= s
->backoffs_by_id
.end()) {
3557 OSDBackoff
*b
= p
->second
;
3558 if (b
->begin
!= m
->begin
&&
3560 lderr(cct
) << __func__
<< " got " << m
->pgid
<< " id " << m
->id
3562 << m
->begin
<< "," << m
->end
<< ") but backoff is ["
3563 << b
->begin
<< "," << b
->end
<< ")" << dendl
;
3564 // hrmpf, unblock it anyway.
3566 ldout(cct
, 10) << __func__
<< " unblock backoff " << b
->pgid
3568 << " [" << b
->begin
<< "," << b
->end
3570 auto spgp
= s
->backoffs
.find(b
->pgid
);
3571 ceph_assert(spgp
!= s
->backoffs
.end());
3572 spgp
->second
.erase(b
->begin
);
3573 if (spgp
->second
.empty()) {
3574 s
->backoffs
.erase(spgp
);
3576 s
->backoffs_by_id
.erase(p
);
3578 // check for any ops to resend
3579 for (auto& q
: s
->ops
) {
3580 if (q
.second
->target
.actual_pgid
== m
->pgid
) {
3581 int r
= q
.second
->target
.contained_by(m
->begin
, m
->end
);
3582 ldout(cct
, 20) << __func__
<< " contained_by " << r
<< " on "
3583 << q
.second
->target
.get_hobj() << dendl
;
3590 lderr(cct
) << __func__
<< " " << m
->pgid
<< " id " << m
->id
3592 << m
->begin
<< "," << m
->end
<< ") but backoff dne" << dendl
;
3598 ldout(cct
, 10) << __func__
<< " unrecognized op " << (int)m
->op
<< dendl
;
3608 uint32_t Objecter::list_nobjects_seek(NListContext
*list_context
,
3611 shared_lock
rl(rwlock
);
3612 list_context
->pos
= hobject_t(object_t(), string(), CEPH_NOSNAP
,
3613 pos
, list_context
->pool_id
, string());
3614 ldout(cct
, 10) << __func__
<< " " << list_context
3615 << " pos " << pos
<< " -> " << list_context
->pos
<< dendl
;
3616 pg_t actual
= osdmap
->raw_pg_to_pg(pg_t(pos
, list_context
->pool_id
));
3617 list_context
->current_pg
= actual
.ps();
3618 list_context
->at_end_of_pool
= false;
3622 uint32_t Objecter::list_nobjects_seek(NListContext
*list_context
,
3623 const hobject_t
& cursor
)
3625 shared_lock
rl(rwlock
);
3626 ldout(cct
, 10) << "list_nobjects_seek " << list_context
<< dendl
;
3627 list_context
->pos
= cursor
;
3628 list_context
->at_end_of_pool
= false;
3629 pg_t actual
= osdmap
->raw_pg_to_pg(pg_t(cursor
.get_hash(), list_context
->pool_id
));
3630 list_context
->current_pg
= actual
.ps();
3631 list_context
->sort_bitwise
= true;
3632 return list_context
->current_pg
;
3635 void Objecter::list_nobjects_get_cursor(NListContext
*list_context
,
3638 shared_lock
rl(rwlock
);
3639 if (list_context
->list
.empty()) {
3640 *cursor
= list_context
->pos
;
3642 const librados::ListObjectImpl
& entry
= list_context
->list
.front();
3643 const string
*key
= (entry
.locator
.empty() ? &entry
.oid
: &entry
.locator
);
3644 uint32_t h
= osdmap
->get_pg_pool(list_context
->pool_id
)->hash_key(*key
, entry
.nspace
);
3645 *cursor
= hobject_t(entry
.oid
, entry
.locator
, list_context
->pool_snap_seq
, h
, list_context
->pool_id
, entry
.nspace
);
3649 void Objecter::list_nobjects(NListContext
*list_context
, Context
*onfinish
)
3651 ldout(cct
, 10) << __func__
<< " pool_id " << list_context
->pool_id
3652 << " pool_snap_seq " << list_context
->pool_snap_seq
3653 << " max_entries " << list_context
->max_entries
3654 << " list_context " << list_context
3655 << " onfinish " << onfinish
3656 << " current_pg " << list_context
->current_pg
3657 << " pos " << list_context
->pos
<< dendl
;
3659 shared_lock
rl(rwlock
);
3660 const pg_pool_t
*pool
= osdmap
->get_pg_pool(list_context
->pool_id
);
3661 if (!pool
) { // pool is gone
3663 put_nlist_context_budget(list_context
);
3664 onfinish
->complete(-ENOENT
);
3667 int pg_num
= pool
->get_pg_num();
3668 bool sort_bitwise
= osdmap
->test_flag(CEPH_OSDMAP_SORTBITWISE
);
3670 if (list_context
->pos
.is_min()) {
3671 list_context
->starting_pg_num
= 0;
3672 list_context
->sort_bitwise
= sort_bitwise
;
3673 list_context
->starting_pg_num
= pg_num
;
3675 if (list_context
->sort_bitwise
!= sort_bitwise
) {
3676 list_context
->pos
= hobject_t(
3677 object_t(), string(), CEPH_NOSNAP
,
3678 list_context
->current_pg
, list_context
->pool_id
, string());
3679 list_context
->sort_bitwise
= sort_bitwise
;
3680 ldout(cct
, 10) << " hobject sort order changed, restarting this pg at "
3681 << list_context
->pos
<< dendl
;
3683 if (list_context
->starting_pg_num
!= pg_num
) {
3684 if (!sort_bitwise
) {
3685 // start reading from the beginning; the pgs have changed
3686 ldout(cct
, 10) << " pg_num changed; restarting with " << pg_num
<< dendl
;
3687 list_context
->pos
= collection_list_handle_t();
3689 list_context
->starting_pg_num
= pg_num
;
3692 if (list_context
->pos
.is_max()) {
3693 ldout(cct
, 20) << __func__
<< " end of pool, list "
3694 << list_context
->list
<< dendl
;
3695 if (list_context
->list
.empty()) {
3696 list_context
->at_end_of_pool
= true;
3698 // release the listing context's budget once all
3699 // OPs (in the session) are finished
3700 put_nlist_context_budget(list_context
);
3701 onfinish
->complete(0);
3706 op
.pg_nls(list_context
->max_entries
, list_context
->filter
,
3707 list_context
->pos
, osdmap
->get_epoch());
3708 list_context
->bl
.clear();
3709 auto onack
= new C_NList(list_context
, onfinish
, this);
3710 object_locator_t
oloc(list_context
->pool_id
, list_context
->nspace
);
3712 // note current_pg in case we don't have (or lose) SORTBITWISE
3713 list_context
->current_pg
= pool
->raw_hash_to_pg(list_context
->pos
.get_hash());
3716 pg_read(list_context
->current_pg
, oloc
, op
,
3717 &list_context
->bl
, 0, onack
, &onack
->epoch
,
3718 &list_context
->ctx_budget
);
3721 void Objecter::_nlist_reply(NListContext
*list_context
, int r
,
3722 Context
*final_finish
, epoch_t reply_epoch
)
3724 ldout(cct
, 10) << __func__
<< " " << list_context
<< dendl
;
3726 auto iter
= list_context
->bl
.cbegin();
3727 pg_nls_response_t response
;
3728 decode(response
, iter
);
3730 // we do this as legacy.
3731 cb::list legacy_extra_info
;
3732 decode(legacy_extra_info
, iter
);
3735 // if the osd returns 1 (newer code), or handle MAX, it means we
3736 // hit the end of the pg.
3737 if ((response
.handle
.is_max() || r
== 1) &&
3738 !list_context
->sort_bitwise
) {
3739 // legacy OSD and !sortbitwise, figure out the next PG on our own
3740 ++list_context
->current_pg
;
3741 if (list_context
->current_pg
== list_context
->starting_pg_num
) {
3743 list_context
->pos
= hobject_t::get_max();
3746 list_context
->pos
= hobject_t(object_t(), string(), CEPH_NOSNAP
,
3747 list_context
->current_pg
,
3748 list_context
->pool_id
, string());
3751 list_context
->pos
= response
.handle
;
3754 int response_size
= response
.entries
.size();
3755 ldout(cct
, 20) << " response.entries.size " << response_size
3756 << ", response.entries " << response
.entries
3757 << ", handle " << response
.handle
3758 << ", tentative new pos " << list_context
->pos
<< dendl
;
3759 if (response_size
) {
3760 std::move(response
.entries
.begin(), response
.entries
.end(),
3761 std::back_inserter(list_context
->list
));
3762 response
.entries
.clear();
3765 if (list_context
->list
.size() >= list_context
->max_entries
) {
3766 ldout(cct
, 20) << " hit max, returning results so far, "
3767 << list_context
->list
<< dendl
;
3768 // release the listing context's budget once all
3769 // OPs (in the session) are finished
3770 put_nlist_context_budget(list_context
);
3771 final_finish
->complete(0);
3776 list_nobjects(list_context
, final_finish
);
3779 void Objecter::put_nlist_context_budget(NListContext
*list_context
)
3781 if (list_context
->ctx_budget
>= 0) {
3782 ldout(cct
, 10) << " release listing context's budget " <<
3783 list_context
->ctx_budget
<< dendl
;
3784 put_op_budget_bytes(list_context
->ctx_budget
);
3785 list_context
->ctx_budget
= -1;
3791 void Objecter::create_pool_snap(int64_t pool
, std::string_view snap_name
,
3792 decltype(PoolOp::onfinish
)&& onfinish
)
3794 unique_lock
wl(rwlock
);
3795 ldout(cct
, 10) << "create_pool_snap; pool: " << pool
<< "; snap: "
3796 << snap_name
<< dendl
;
3798 const pg_pool_t
*p
= osdmap
->get_pg_pool(pool
);
3800 onfinish
->defer(std::move(onfinish
), osdc_errc::pool_dne
, cb::list
{});
3803 if (p
->snap_exists(snap_name
)) {
3804 onfinish
->defer(std::move(onfinish
), osdc_errc::snapshot_exists
,
3809 auto op
= new PoolOp
;
3810 op
->tid
= ++last_tid
;
3812 op
->name
= snap_name
;
3813 op
->onfinish
= std::move(onfinish
);
3814 op
->pool_op
= POOL_OP_CREATE_SNAP
;
3815 pool_ops
[op
->tid
] = op
;
3820 struct CB_SelfmanagedSnap
{
3821 std::unique_ptr
<ca::Completion
<void(bs::error_code
, snapid_t
)>> fin
;
3822 CB_SelfmanagedSnap(decltype(fin
)&& fin
)
3823 : fin(std::move(fin
)) {}
3824 void operator()(bs::error_code ec
, const cb::list
& bl
) {
3825 snapid_t snapid
= 0;
3828 auto p
= bl
.cbegin();
3830 } catch (const cb::error
& e
) {
3834 fin
->defer(std::move(fin
), ec
, snapid
);
3838 void Objecter::allocate_selfmanaged_snap(
3840 std::unique_ptr
<ca::Completion
<void(bs::error_code
, snapid_t
)>> onfinish
)
3842 unique_lock
wl(rwlock
);
3843 ldout(cct
, 10) << "allocate_selfmanaged_snap; pool: " << pool
<< dendl
;
3844 auto op
= new PoolOp
;
3845 op
->tid
= ++last_tid
;
3847 op
->onfinish
= PoolOp::OpComp::create(
3848 service
.get_executor(),
3849 CB_SelfmanagedSnap(std::move(onfinish
)));
3850 op
->pool_op
= POOL_OP_CREATE_UNMANAGED_SNAP
;
3851 pool_ops
[op
->tid
] = op
;
3856 void Objecter::delete_pool_snap(
3857 int64_t pool
, std::string_view snap_name
,
3858 decltype(PoolOp::onfinish
)&& onfinish
)
3860 unique_lock
wl(rwlock
);
3861 ldout(cct
, 10) << "delete_pool_snap; pool: " << pool
<< "; snap: "
3862 << snap_name
<< dendl
;
3864 const pg_pool_t
*p
= osdmap
->get_pg_pool(pool
);
3866 onfinish
->defer(std::move(onfinish
), osdc_errc::pool_dne
, cb::list
{});
3870 if (!p
->snap_exists(snap_name
)) {
3871 onfinish
->defer(std::move(onfinish
), osdc_errc::snapshot_dne
, cb::list
{});
3875 auto op
= new PoolOp
;
3876 op
->tid
= ++last_tid
;
3878 op
->name
= snap_name
;
3879 op
->onfinish
= std::move(onfinish
);
3880 op
->pool_op
= POOL_OP_DELETE_SNAP
;
3881 pool_ops
[op
->tid
] = op
;
3886 void Objecter::delete_selfmanaged_snap(int64_t pool
, snapid_t snap
,
3887 decltype(PoolOp::onfinish
)&& onfinish
)
3889 unique_lock
wl(rwlock
);
3890 ldout(cct
, 10) << "delete_selfmanaged_snap; pool: " << pool
<< "; snap: "
3892 auto op
= new PoolOp
;
3893 op
->tid
= ++last_tid
;
3895 op
->onfinish
= std::move(onfinish
);
3896 op
->pool_op
= POOL_OP_DELETE_UNMANAGED_SNAP
;
3898 pool_ops
[op
->tid
] = op
;
3903 void Objecter::create_pool(std::string_view name
,
3904 decltype(PoolOp::onfinish
)&& onfinish
,
3907 unique_lock
wl(rwlock
);
3908 ldout(cct
, 10) << "create_pool name=" << name
<< dendl
;
3910 if (osdmap
->lookup_pg_pool_name(name
) >= 0) {
3911 onfinish
->defer(std::move(onfinish
), osdc_errc::pool_exists
, cb::list
{});
3915 auto op
= new PoolOp
;
3916 op
->tid
= ++last_tid
;
3919 op
->onfinish
= std::move(onfinish
);
3920 op
->pool_op
= POOL_OP_CREATE
;
3921 pool_ops
[op
->tid
] = op
;
3922 op
->crush_rule
= crush_rule
;
3927 void Objecter::delete_pool(int64_t pool
,
3928 decltype(PoolOp::onfinish
)&& onfinish
)
3930 unique_lock
wl(rwlock
);
3931 ldout(cct
, 10) << "delete_pool " << pool
<< dendl
;
3933 if (!osdmap
->have_pg_pool(pool
))
3934 onfinish
->defer(std::move(onfinish
), osdc_errc::pool_dne
, cb::list
{});
3936 _do_delete_pool(pool
, std::move(onfinish
));
3939 void Objecter::delete_pool(std::string_view pool_name
,
3940 decltype(PoolOp::onfinish
)&& onfinish
)
3942 unique_lock
wl(rwlock
);
3943 ldout(cct
, 10) << "delete_pool " << pool_name
<< dendl
;
3945 int64_t pool
= osdmap
->lookup_pg_pool_name(pool_name
);
3947 // This only returns one error: -ENOENT.
3948 onfinish
->defer(std::move(onfinish
), osdc_errc::pool_dne
, cb::list
{});
3950 _do_delete_pool(pool
, std::move(onfinish
));
3953 void Objecter::_do_delete_pool(int64_t pool
,
3954 decltype(PoolOp::onfinish
)&& onfinish
)
3957 auto op
= new PoolOp
;
3958 op
->tid
= ++last_tid
;
3960 op
->name
= "delete";
3961 op
->onfinish
= std::move(onfinish
);
3962 op
->pool_op
= POOL_OP_DELETE
;
3963 pool_ops
[op
->tid
] = op
;
3967 void Objecter::pool_op_submit(PoolOp
*op
)
3970 if (mon_timeout
> timespan(0)) {
3971 op
->ontimeout
= timer
.add_event(mon_timeout
,
3973 pool_op_cancel(op
->tid
, -ETIMEDOUT
); });
3975 _pool_op_submit(op
);
3978 void Objecter::_pool_op_submit(PoolOp
*op
)
3980 // rwlock is locked unique
3982 ldout(cct
, 10) << "pool_op_submit " << op
->tid
<< dendl
;
3983 auto m
= new MPoolOp(monc
->get_fsid(), op
->tid
, op
->pool
,
3984 op
->name
, op
->pool_op
,
3985 last_seen_osdmap_version
);
3986 if (op
->snapid
) m
->snapid
= op
->snapid
;
3987 if (op
->crush_rule
) m
->crush_rule
= op
->crush_rule
;
3988 monc
->send_mon_message(m
);
3989 op
->last_submit
= ceph::coarse_mono_clock::now();
3991 logger
->inc(l_osdc_poolop_send
);
3995 * Handle a reply to a PoolOp message. Check that we sent the message
3996 * and give the caller responsibility for the returned cb::list.
3997 * Then either call the finisher or stash the PoolOp, depending on if we
3998 * have a new enough map.
3999 * Lastly, clean up the message and PoolOp.
4001 void Objecter::handle_pool_op_reply(MPoolOpReply
*m
)
4003 int rc
= m
->replyCode
;
4004 auto ec
= rc
< 0 ? bs::error_code(-rc
, mon_category()) : bs::error_code();
4006 shunique_lock
sul(rwlock
, acquire_shared
);
4013 ldout(cct
, 10) << "handle_pool_op_reply " << *m
<< dendl
;
4014 ceph_tid_t tid
= m
->get_tid();
4015 auto iter
= pool_ops
.find(tid
);
4016 if (iter
!= pool_ops
.end()) {
4017 PoolOp
*op
= iter
->second
;
4018 ldout(cct
, 10) << "have request " << tid
<< " at " << op
<< " Op: "
4019 << ceph_pool_op_name(op
->pool_op
) << dendl
;
4020 cb::list bl
{std::move(m
->response_data
)};
4021 if (m
->version
> last_seen_osdmap_version
)
4022 last_seen_osdmap_version
= m
->version
;
4023 if (osdmap
->get_epoch() < m
->epoch
) {
4026 // recheck op existence since we have let go of rwlock
4027 // (for promotion) above.
4028 iter
= pool_ops
.find(tid
);
4029 if (iter
== pool_ops
.end())
4030 goto done
; // op is gone.
4031 if (osdmap
->get_epoch() < m
->epoch
) {
4032 ldout(cct
, 20) << "waiting for client to reach epoch " << m
->epoch
4033 << " before calling back" << dendl
;
4034 _wait_for_new_map(OpCompletion::create(
4035 service
.get_executor(),
4036 [o
= std::move(op
->onfinish
),
4037 bl
= std::move(bl
)](
4038 bs::error_code ec
) mutable {
4039 o
->defer(std::move(o
), ec
, bl
);
4044 // map epoch changed, probably because a MOSDMap message
4045 // sneaked in. Do caller-specified callback now or else
4046 // we lose it forever.
4047 ceph_assert(op
->onfinish
);
4048 op
->onfinish
->defer(std::move(op
->onfinish
), ec
, std::move(bl
));
4051 ceph_assert(op
->onfinish
);
4052 op
->onfinish
->defer(std::move(op
->onfinish
), ec
, std::move(bl
));
4054 op
->onfinish
= nullptr;
4055 if (!sul
.owns_lock()) {
4059 iter
= pool_ops
.find(tid
);
4060 if (iter
!= pool_ops
.end()) {
4061 _finish_pool_op(op
, 0);
4064 ldout(cct
, 10) << "unknown request " << tid
<< dendl
;
4068 // Not strictly necessary, since we'll release it on return.
4071 ldout(cct
, 10) << "done" << dendl
;
4075 int Objecter::pool_op_cancel(ceph_tid_t tid
, int r
)
4077 ceph_assert(initialized
);
4079 unique_lock
wl(rwlock
);
4081 auto it
= pool_ops
.find(tid
);
4082 if (it
== pool_ops
.end()) {
4083 ldout(cct
, 10) << __func__
<< " tid " << tid
<< " dne" << dendl
;
4087 ldout(cct
, 10) << __func__
<< " tid " << tid
<< dendl
;
4089 PoolOp
*op
= it
->second
;
4091 op
->onfinish
->defer(std::move(op
->onfinish
), osdcode(r
), cb::list
{});
4093 _finish_pool_op(op
, r
);
4097 void Objecter::_finish_pool_op(PoolOp
*op
, int r
)
4099 // rwlock is locked unique
4100 pool_ops
.erase(op
->tid
);
4101 logger
->set(l_osdc_poolop_active
, pool_ops
.size());
4103 if (op
->ontimeout
&& r
!= -ETIMEDOUT
) {
4104 timer
.cancel_event(op
->ontimeout
);
4112 void Objecter::get_pool_stats(
4113 const std::vector
<std::string
>& pools
,
4114 decltype(PoolStatOp::onfinish
)&& onfinish
)
4116 ldout(cct
, 10) << "get_pool_stats " << pools
<< dendl
;
4118 auto op
= new PoolStatOp
;
4119 op
->tid
= ++last_tid
;
4121 op
->onfinish
= std::move(onfinish
);
4122 if (mon_timeout
> timespan(0)) {
4123 op
->ontimeout
= timer
.add_event(mon_timeout
,
4125 pool_stat_op_cancel(op
->tid
,
4131 unique_lock
wl(rwlock
);
4133 poolstat_ops
[op
->tid
] = op
;
4135 logger
->set(l_osdc_poolstat_active
, poolstat_ops
.size());
4137 _poolstat_submit(op
);
4140 void Objecter::_poolstat_submit(PoolStatOp
*op
)
4142 ldout(cct
, 10) << "_poolstat_submit " << op
->tid
<< dendl
;
4143 monc
->send_mon_message(new MGetPoolStats(monc
->get_fsid(), op
->tid
,
4145 last_seen_pgmap_version
));
4146 op
->last_submit
= ceph::coarse_mono_clock::now();
4148 logger
->inc(l_osdc_poolstat_send
);
4151 void Objecter::handle_get_pool_stats_reply(MGetPoolStatsReply
*m
)
4153 ldout(cct
, 10) << "handle_get_pool_stats_reply " << *m
<< dendl
;
4154 ceph_tid_t tid
= m
->get_tid();
4156 unique_lock
wl(rwlock
);
4162 auto iter
= poolstat_ops
.find(tid
);
4163 if (iter
!= poolstat_ops
.end()) {
4164 PoolStatOp
*op
= poolstat_ops
[tid
];
4165 ldout(cct
, 10) << "have request " << tid
<< " at " << op
<< dendl
;
4166 if (m
->version
> last_seen_pgmap_version
) {
4167 last_seen_pgmap_version
= m
->version
;
4169 op
->onfinish
->defer(std::move(op
->onfinish
), bs::error_code
{},
4170 std::move(m
->pool_stats
), m
->per_pool
);
4171 _finish_pool_stat_op(op
, 0);
4173 ldout(cct
, 10) << "unknown request " << tid
<< dendl
;
4175 ldout(cct
, 10) << "done" << dendl
;
4179 int Objecter::pool_stat_op_cancel(ceph_tid_t tid
, int r
)
4181 ceph_assert(initialized
);
4183 unique_lock
wl(rwlock
);
4185 auto it
= poolstat_ops
.find(tid
);
4186 if (it
== poolstat_ops
.end()) {
4187 ldout(cct
, 10) << __func__
<< " tid " << tid
<< " dne" << dendl
;
4191 ldout(cct
, 10) << __func__
<< " tid " << tid
<< dendl
;
4193 auto op
= it
->second
;
4195 op
->onfinish
->defer(std::move(op
->onfinish
), osdcode(r
),
4196 bc::flat_map
<std::string
, pool_stat_t
>{}, false);
4197 _finish_pool_stat_op(op
, r
);
4201 void Objecter::_finish_pool_stat_op(PoolStatOp
*op
, int r
)
4203 // rwlock is locked unique
4205 poolstat_ops
.erase(op
->tid
);
4206 logger
->set(l_osdc_poolstat_active
, poolstat_ops
.size());
4208 if (op
->ontimeout
&& r
!= -ETIMEDOUT
)
4209 timer
.cancel_event(op
->ontimeout
);
4214 void Objecter::get_fs_stats(boost::optional
<int64_t> poolid
,
4215 decltype(StatfsOp::onfinish
)&& onfinish
)
4217 ldout(cct
, 10) << "get_fs_stats" << dendl
;
4218 unique_lock
l(rwlock
);
4220 auto op
= new StatfsOp
;
4221 op
->tid
= ++last_tid
;
4222 op
->data_pool
= poolid
;
4223 op
->onfinish
= std::move(onfinish
);
4224 if (mon_timeout
> timespan(0)) {
4225 op
->ontimeout
= timer
.add_event(mon_timeout
,
4227 statfs_op_cancel(op
->tid
,
4232 statfs_ops
[op
->tid
] = op
;
4234 logger
->set(l_osdc_statfs_active
, statfs_ops
.size());
4236 _fs_stats_submit(op
);
4239 void Objecter::_fs_stats_submit(StatfsOp
*op
)
4241 // rwlock is locked unique
4243 ldout(cct
, 10) << "fs_stats_submit" << op
->tid
<< dendl
;
4244 monc
->send_mon_message(new MStatfs(monc
->get_fsid(), op
->tid
,
4246 last_seen_pgmap_version
));
4247 op
->last_submit
= ceph::coarse_mono_clock::now();
4249 logger
->inc(l_osdc_statfs_send
);
4252 void Objecter::handle_fs_stats_reply(MStatfsReply
*m
)
4254 unique_lock
wl(rwlock
);
4260 ldout(cct
, 10) << "handle_fs_stats_reply " << *m
<< dendl
;
4261 ceph_tid_t tid
= m
->get_tid();
4263 if (statfs_ops
.count(tid
)) {
4264 StatfsOp
*op
= statfs_ops
[tid
];
4265 ldout(cct
, 10) << "have request " << tid
<< " at " << op
<< dendl
;
4266 if (m
->h
.version
> last_seen_pgmap_version
)
4267 last_seen_pgmap_version
= m
->h
.version
;
4268 op
->onfinish
->defer(std::move(op
->onfinish
), bs::error_code
{}, m
->h
.st
);
4269 _finish_statfs_op(op
, 0);
4271 ldout(cct
, 10) << "unknown request " << tid
<< dendl
;
4274 ldout(cct
, 10) << "done" << dendl
;
4277 int Objecter::statfs_op_cancel(ceph_tid_t tid
, int r
)
4279 ceph_assert(initialized
);
4281 unique_lock
wl(rwlock
);
4283 auto it
= statfs_ops
.find(tid
);
4284 if (it
== statfs_ops
.end()) {
4285 ldout(cct
, 10) << __func__
<< " tid " << tid
<< " dne" << dendl
;
4289 ldout(cct
, 10) << __func__
<< " tid " << tid
<< dendl
;
4291 auto op
= it
->second
;
4293 op
->onfinish
->defer(std::move(op
->onfinish
), osdcode(r
), ceph_statfs
{});
4294 _finish_statfs_op(op
, r
);
4298 void Objecter::_finish_statfs_op(StatfsOp
*op
, int r
)
4300 // rwlock is locked unique
4302 statfs_ops
.erase(op
->tid
);
4303 logger
->set(l_osdc_statfs_active
, statfs_ops
.size());
4305 if (op
->ontimeout
&& r
!= -ETIMEDOUT
)
4306 timer
.cancel_event(op
->ontimeout
);
4313 void Objecter::_sg_read_finish(vector
<ObjectExtent
>& extents
,
4314 vector
<cb::list
>& resultbl
,
4315 cb::list
*bl
, Context
*onfinish
)
4318 ldout(cct
, 15) << "_sg_read_finish" << dendl
;
4320 if (extents
.size() > 1) {
4321 Striper::StripedReadResult r
;
4322 auto bit
= resultbl
.begin();
4323 for (auto eit
= extents
.begin();
4324 eit
!= extents
.end();
4326 r
.add_partial_result(cct
, *bit
, eit
->buffer_extents
);
4329 r
.assemble_result(cct
, *bl
, false);
4331 ldout(cct
, 15) << " only one frag" << dendl
;
4332 *bl
= std::move(resultbl
[0]);
4336 uint64_t bytes_read
= bl
->length();
4337 ldout(cct
, 7) << "_sg_read_finish " << bytes_read
<< " bytes" << dendl
;
4340 onfinish
->complete(bytes_read
);// > 0 ? bytes_read:m->get_result());
4345 void Objecter::ms_handle_connect(Connection
*con
)
4347 ldout(cct
, 10) << "ms_handle_connect " << con
<< dendl
;
4351 if (con
->get_peer_type() == CEPH_ENTITY_TYPE_MON
)
4355 bool Objecter::ms_handle_reset(Connection
*con
)
4359 if (con
->get_peer_type() == CEPH_ENTITY_TYPE_OSD
) {
4360 unique_lock
wl(rwlock
);
4362 auto priv
= con
->get_priv();
4363 auto session
= static_cast<OSDSession
*>(priv
.get());
4365 ldout(cct
, 1) << "ms_handle_reset " << con
<< " session " << session
4366 << " osd." << session
->osd
<< dendl
;
4367 // the session maybe had been closed if new osdmap just handled
4368 // says the osd down
4369 if (!(initialized
&& osdmap
->is_up(session
->osd
))) {
4370 ldout(cct
, 1) << "ms_handle_reset aborted,initialized=" << initialized
<< dendl
;
4374 map
<uint64_t, LingerOp
*> lresend
;
4375 unique_lock
sl(session
->lock
);
4376 _reopen_session(session
);
4377 _kick_requests(session
, lresend
);
4379 _linger_ops_resend(lresend
, wl
);
4381 maybe_request_map();
4388 void Objecter::ms_handle_remote_reset(Connection
*con
)
4391 * treat these the same.
4393 ms_handle_reset(con
);
4396 bool Objecter::ms_handle_refused(Connection
*con
)
4399 if (osdmap
&& (con
->get_peer_type() == CEPH_ENTITY_TYPE_OSD
)) {
4400 int osd
= osdmap
->identify_osd(con
->get_peer_addr());
4402 ldout(cct
, 1) << "ms_handle_refused on osd." << osd
<< dendl
;
4408 void Objecter::op_target_t::dump(Formatter
*f
) const
4410 f
->dump_stream("pg") << pgid
;
4411 f
->dump_int("osd", osd
);
4412 f
->dump_stream("object_id") << base_oid
;
4413 f
->dump_stream("object_locator") << base_oloc
;
4414 f
->dump_stream("target_object_id") << target_oid
;
4415 f
->dump_stream("target_object_locator") << target_oloc
;
4416 f
->dump_int("paused", (int)paused
);
4417 f
->dump_int("used_replica", (int)used_replica
);
4418 f
->dump_int("precalc_pgid", (int)precalc_pgid
);
4421 void Objecter::_dump_active(OSDSession
*s
)
4423 for (auto p
= s
->ops
.begin(); p
!= s
->ops
.end(); ++p
) {
4425 ldout(cct
, 20) << op
->tid
<< "\t" << op
->target
.pgid
4426 << "\tosd." << (op
->session
? op
->session
->osd
: -1)
4427 << "\t" << op
->target
.base_oid
4428 << "\t" << op
->ops
<< dendl
;
4432 void Objecter::_dump_active()
4434 ldout(cct
, 20) << "dump_active .. " << num_homeless_ops
<< " homeless"
4436 for (auto siter
= osd_sessions
.begin();
4437 siter
!= osd_sessions
.end(); ++siter
) {
4438 auto s
= siter
->second
;
4439 shared_lock
sl(s
->lock
);
4443 _dump_active(homeless_session
);
4446 void Objecter::dump_active()
4448 shared_lock
rl(rwlock
);
4453 void Objecter::dump_requests(Formatter
*fmt
)
4455 // Read-lock on Objecter held here
4456 fmt
->open_object_section("requests");
4458 dump_linger_ops(fmt
);
4460 dump_pool_stat_ops(fmt
);
4461 dump_statfs_ops(fmt
);
4462 dump_command_ops(fmt
);
4463 fmt
->close_section(); // requests object
4466 void Objecter::_dump_ops(const OSDSession
*s
, Formatter
*fmt
)
4468 for (auto p
= s
->ops
.begin(); p
!= s
->ops
.end(); ++p
) {
4470 auto age
= std::chrono::duration
<double>(ceph::coarse_mono_clock::now() - op
->stamp
);
4471 fmt
->open_object_section("op");
4472 fmt
->dump_unsigned("tid", op
->tid
);
4473 op
->target
.dump(fmt
);
4474 fmt
->dump_stream("last_sent") << op
->stamp
;
4475 fmt
->dump_float("age", age
.count());
4476 fmt
->dump_int("attempts", op
->attempts
);
4477 fmt
->dump_stream("snapid") << op
->snapid
;
4478 fmt
->dump_stream("snap_context") << op
->snapc
;
4479 fmt
->dump_stream("mtime") << op
->mtime
;
4481 fmt
->open_array_section("osd_ops");
4482 for (auto it
= op
->ops
.begin(); it
!= op
->ops
.end(); ++it
) {
4483 fmt
->dump_stream("osd_op") << *it
;
4485 fmt
->close_section(); // osd_ops array
4487 fmt
->close_section(); // op object
4491 void Objecter::dump_ops(Formatter
*fmt
)
4493 // Read-lock on Objecter held
4494 fmt
->open_array_section("ops");
4495 for (auto siter
= osd_sessions
.begin();
4496 siter
!= osd_sessions
.end(); ++siter
) {
4497 OSDSession
*s
= siter
->second
;
4498 shared_lock
sl(s
->lock
);
4502 _dump_ops(homeless_session
, fmt
);
4503 fmt
->close_section(); // ops array
4506 void Objecter::_dump_linger_ops(const OSDSession
*s
, Formatter
*fmt
)
4508 for (auto p
= s
->linger_ops
.begin(); p
!= s
->linger_ops
.end(); ++p
) {
4509 auto op
= p
->second
;
4510 fmt
->open_object_section("linger_op");
4511 fmt
->dump_unsigned("linger_id", op
->linger_id
);
4512 op
->target
.dump(fmt
);
4513 fmt
->dump_stream("snapid") << op
->snap
;
4514 fmt
->dump_stream("registered") << op
->registered
;
4515 fmt
->close_section(); // linger_op object
4519 void Objecter::dump_linger_ops(Formatter
*fmt
)
4521 // We have a read-lock on the objecter
4522 fmt
->open_array_section("linger_ops");
4523 for (auto siter
= osd_sessions
.begin();
4524 siter
!= osd_sessions
.end(); ++siter
) {
4525 auto s
= siter
->second
;
4526 shared_lock
sl(s
->lock
);
4527 _dump_linger_ops(s
, fmt
);
4530 _dump_linger_ops(homeless_session
, fmt
);
4531 fmt
->close_section(); // linger_ops array
4534 void Objecter::_dump_command_ops(const OSDSession
*s
, Formatter
*fmt
)
4536 for (auto p
= s
->command_ops
.begin(); p
!= s
->command_ops
.end(); ++p
) {
4537 auto op
= p
->second
;
4538 fmt
->open_object_section("command_op");
4539 fmt
->dump_unsigned("command_id", op
->tid
);
4540 fmt
->dump_int("osd", op
->session
? op
->session
->osd
: -1);
4541 fmt
->open_array_section("command");
4542 for (auto q
= op
->cmd
.begin(); q
!= op
->cmd
.end(); ++q
)
4543 fmt
->dump_string("word", *q
);
4544 fmt
->close_section();
4545 if (op
->target_osd
>= 0)
4546 fmt
->dump_int("target_osd", op
->target_osd
);
4548 fmt
->dump_stream("target_pg") << op
->target_pg
;
4549 fmt
->close_section(); // command_op object
4553 void Objecter::dump_command_ops(Formatter
*fmt
)
4555 // We have a read-lock on the Objecter here
4556 fmt
->open_array_section("command_ops");
4557 for (auto siter
= osd_sessions
.begin();
4558 siter
!= osd_sessions
.end(); ++siter
) {
4559 auto s
= siter
->second
;
4560 shared_lock
sl(s
->lock
);
4561 _dump_command_ops(s
, fmt
);
4564 _dump_command_ops(homeless_session
, fmt
);
4565 fmt
->close_section(); // command_ops array
4568 void Objecter::dump_pool_ops(Formatter
*fmt
) const
4570 fmt
->open_array_section("pool_ops");
4571 for (auto p
= pool_ops
.begin(); p
!= pool_ops
.end(); ++p
) {
4572 auto op
= p
->second
;
4573 fmt
->open_object_section("pool_op");
4574 fmt
->dump_unsigned("tid", op
->tid
);
4575 fmt
->dump_int("pool", op
->pool
);
4576 fmt
->dump_string("name", op
->name
);
4577 fmt
->dump_int("operation_type", op
->pool_op
);
4578 fmt
->dump_unsigned("crush_rule", op
->crush_rule
);
4579 fmt
->dump_stream("snapid") << op
->snapid
;
4580 fmt
->dump_stream("last_sent") << op
->last_submit
;
4581 fmt
->close_section(); // pool_op object
4583 fmt
->close_section(); // pool_ops array
4586 void Objecter::dump_pool_stat_ops(Formatter
*fmt
) const
4588 fmt
->open_array_section("pool_stat_ops");
4589 for (auto p
= poolstat_ops
.begin();
4590 p
!= poolstat_ops
.end();
4592 PoolStatOp
*op
= p
->second
;
4593 fmt
->open_object_section("pool_stat_op");
4594 fmt
->dump_unsigned("tid", op
->tid
);
4595 fmt
->dump_stream("last_sent") << op
->last_submit
;
4597 fmt
->open_array_section("pools");
4598 for (const auto& it
: op
->pools
) {
4599 fmt
->dump_string("pool", it
);
4601 fmt
->close_section(); // pools array
4603 fmt
->close_section(); // pool_stat_op object
4605 fmt
->close_section(); // pool_stat_ops array
4608 void Objecter::dump_statfs_ops(Formatter
*fmt
) const
4610 fmt
->open_array_section("statfs_ops");
4611 for (auto p
= statfs_ops
.begin(); p
!= statfs_ops
.end(); ++p
) {
4612 auto op
= p
->second
;
4613 fmt
->open_object_section("statfs_op");
4614 fmt
->dump_unsigned("tid", op
->tid
);
4615 fmt
->dump_stream("last_sent") << op
->last_submit
;
4616 fmt
->close_section(); // statfs_op object
4618 fmt
->close_section(); // statfs_ops array
4621 Objecter::RequestStateHook::RequestStateHook(Objecter
*objecter
) :
4622 m_objecter(objecter
)
4626 int Objecter::RequestStateHook::call(std::string_view command
,
4627 const cmdmap_t
& cmdmap
,
4632 shared_lock
rl(m_objecter
->rwlock
);
4633 m_objecter
->dump_requests(f
);
4637 void Objecter::blocklist_self(bool set
)
4639 ldout(cct
, 10) << "blocklist_self " << (set
? "add" : "rm") << dendl
;
4642 cmd
.push_back("{\"prefix\":\"osd blocklist\", ");
4644 cmd
.push_back("\"blocklistop\":\"add\",");
4646 cmd
.push_back("\"blocklistop\":\"rm\",");
4648 // this is somewhat imprecise in that we are blocklisting our first addr only
4649 ss
<< messenger
->get_myaddrs().front().get_legacy_str();
4650 cmd
.push_back("\"addr\":\"" + ss
.str() + "\"");
4652 auto m
= new MMonCommand(monc
->get_fsid());
4655 // NOTE: no fallback to legacy blacklist command implemented here
4656 // since this is only used for test code.
4658 monc
->send_mon_message(m
);
4663 void Objecter::handle_command_reply(MCommandReply
*m
)
4665 unique_lock
wl(rwlock
);
4671 ConnectionRef con
= m
->get_connection();
4672 auto priv
= con
->get_priv();
4673 auto s
= static_cast<OSDSession
*>(priv
.get());
4674 if (!s
|| s
->con
!= con
) {
4675 ldout(cct
, 7) << __func__
<< " no session on con " << con
<< dendl
;
4680 shared_lock
sl(s
->lock
);
4681 auto p
= s
->command_ops
.find(m
->get_tid());
4682 if (p
== s
->command_ops
.end()) {
4683 ldout(cct
, 10) << "handle_command_reply tid " << m
->get_tid()
4684 << " not found" << dendl
;
4690 CommandOp
*c
= p
->second
;
4692 m
->get_connection() != c
->session
->con
) {
4693 ldout(cct
, 10) << "handle_command_reply tid " << m
->get_tid()
4694 << " got reply from wrong connection "
4695 << m
->get_connection() << " " << m
->get_source_inst()
4702 if (m
->r
== -EAGAIN
) {
4703 ldout(cct
,10) << __func__
<< " tid " << m
->get_tid()
4704 << " got EAGAIN, requesting map and resending" << dendl
;
4705 // NOTE: This might resend twice... once now, and once again when
4706 // we get an updated osdmap and the PG is found to have moved.
4707 _maybe_request_map();
4716 unique_lock
sul(s
->lock
);
4717 _finish_command(c
, m
->r
< 0 ? bs::error_code(-m
->r
, osd_category()) :
4718 bs::error_code(), std::move(m
->rs
),
4719 std::move(m
->get_data()));
4725 Objecter::LingerOp::LingerOp(Objecter
*o
, uint64_t linger_id
)
4727 linger_id(linger_id
),
4728 watch_lock(ceph::make_shared_mutex(
4729 fmt::format("LingerOp::watch_lock #{}", linger_id
)))
4732 void Objecter::submit_command(CommandOp
*c
, ceph_tid_t
*ptid
)
4734 shunique_lock
sul(rwlock
, ceph::acquire_unique
);
4736 ceph_tid_t tid
= ++last_tid
;
4737 ldout(cct
, 10) << "_submit_command " << tid
<< " " << c
->cmd
<< dendl
;
4741 unique_lock
hs_wl(homeless_session
->lock
);
4742 _session_command_op_assign(homeless_session
, c
);
4745 _calc_command_target(c
, sul
);
4746 _assign_command_session(c
, sul
);
4747 if (osd_timeout
> timespan(0)) {
4748 c
->ontimeout
= timer
.add_event(osd_timeout
,
4752 osdc_errc::timed_out
); });
4755 if (!c
->session
->is_homeless()) {
4758 _maybe_request_map();
4760 if (c
->map_check_error
)
4761 _send_command_map_check(c
);
4765 logger
->inc(l_osdc_command_active
);
4768 int Objecter::_calc_command_target(CommandOp
*c
,
4769 shunique_lock
<ceph::shared_mutex
>& sul
)
4771 ceph_assert(sul
.owns_lock() && sul
.mutex() == &rwlock
);
4773 c
->map_check_error
= 0;
4775 // ignore overlays, just like we do with pg ops
4776 c
->target
.flags
|= CEPH_OSD_FLAG_IGNORE_OVERLAY
;
4778 if (c
->target_osd
>= 0) {
4779 if (!osdmap
->exists(c
->target_osd
)) {
4780 c
->map_check_error
= -ENOENT
;
4781 c
->map_check_error_str
= "osd dne";
4783 return RECALC_OP_TARGET_OSD_DNE
;
4785 if (osdmap
->is_down(c
->target_osd
)) {
4786 c
->map_check_error
= -ENXIO
;
4787 c
->map_check_error_str
= "osd down";
4789 return RECALC_OP_TARGET_OSD_DOWN
;
4791 c
->target
.osd
= c
->target_osd
;
4793 int ret
= _calc_target(&(c
->target
), nullptr, true);
4794 if (ret
== RECALC_OP_TARGET_POOL_DNE
) {
4795 c
->map_check_error
= -ENOENT
;
4796 c
->map_check_error_str
= "pool dne";
4799 } else if (ret
== RECALC_OP_TARGET_OSD_DOWN
) {
4800 c
->map_check_error
= -ENXIO
;
4801 c
->map_check_error_str
= "osd down";
4808 int r
= _get_session(c
->target
.osd
, &s
, sul
);
4809 ceph_assert(r
!= -EAGAIN
); /* shouldn't happen as we're holding the write lock */
4811 if (c
->session
!= s
) {
4813 return RECALC_OP_TARGET_NEED_RESEND
;
4818 ldout(cct
, 20) << "_recalc_command_target " << c
->tid
<< " no change, "
4819 << c
->session
<< dendl
;
4821 return RECALC_OP_TARGET_NO_ACTION
;
4824 void Objecter::_assign_command_session(CommandOp
*c
,
4825 shunique_lock
<ceph::shared_mutex
>& sul
)
4827 ceph_assert(sul
.owns_lock() && sul
.mutex() == &rwlock
);
4830 int r
= _get_session(c
->target
.osd
, &s
, sul
);
4831 ceph_assert(r
!= -EAGAIN
); /* shouldn't happen as we're holding the write lock */
4833 if (c
->session
!= s
) {
4835 OSDSession
*cs
= c
->session
;
4836 unique_lock
csl(cs
->lock
);
4837 _session_command_op_remove(c
->session
, c
);
4840 unique_lock
sl(s
->lock
);
4841 _session_command_op_assign(s
, c
);
4847 void Objecter::_send_command(CommandOp
*c
)
4849 ldout(cct
, 10) << "_send_command " << c
->tid
<< dendl
;
4850 ceph_assert(c
->session
);
4851 ceph_assert(c
->session
->con
);
4852 auto m
= new MCommand(monc
->monmap
.fsid
);
4854 m
->set_data(c
->inbl
);
4856 c
->session
->con
->send_message(m
);
4857 logger
->inc(l_osdc_command_send
);
4860 int Objecter::command_op_cancel(OSDSession
*s
, ceph_tid_t tid
,
4863 ceph_assert(initialized
);
4865 unique_lock
wl(rwlock
);
4867 auto it
= s
->command_ops
.find(tid
);
4868 if (it
== s
->command_ops
.end()) {
4869 ldout(cct
, 10) << __func__
<< " tid " << tid
<< " dne" << dendl
;
4873 ldout(cct
, 10) << __func__
<< " tid " << tid
<< dendl
;
4875 CommandOp
*op
= it
->second
;
4876 _command_cancel_map_check(op
);
4877 unique_lock
sl(op
->session
->lock
);
4878 _finish_command(op
, ec
, {}, {});
4883 void Objecter::_finish_command(CommandOp
*c
, bs::error_code ec
,
4884 string
&& rs
, cb::list
&& bl
)
4886 // rwlock is locked unique
4887 // session lock is locked
4889 ldout(cct
, 10) << "_finish_command " << c
->tid
<< " = " << ec
<< " "
4893 c
->onfinish
->defer(std::move(c
->onfinish
), ec
, std::move(rs
), std::move(bl
));
4895 if (c
->ontimeout
&& ec
!= bs::errc::timed_out
)
4896 timer
.cancel_event(c
->ontimeout
);
4898 _session_command_op_remove(c
->session
, c
);
4902 logger
->dec(l_osdc_command_active
);
4905 Objecter::OSDSession::~OSDSession()
4907 // Caller is responsible for re-assigning or
4908 // destroying any ops that were assigned to us
4909 ceph_assert(ops
.empty());
4910 ceph_assert(linger_ops
.empty());
4911 ceph_assert(command_ops
.empty());
4914 Objecter::Objecter(CephContext
*cct
,
4915 Messenger
*m
, MonClient
*mc
,
4916 boost::asio::io_context
& service
) :
4917 Dispatcher(cct
), messenger(m
), monc(mc
), service(service
)
4919 mon_timeout
= cct
->_conf
.get_val
<std::chrono::seconds
>("rados_mon_op_timeout");
4920 osd_timeout
= cct
->_conf
.get_val
<std::chrono::seconds
>("rados_osd_op_timeout");
4923 Objecter::~Objecter()
4925 ceph_assert(homeless_session
->get_nref() == 1);
4926 ceph_assert(num_homeless_ops
== 0);
4927 homeless_session
->put();
4929 ceph_assert(osd_sessions
.empty());
4930 ceph_assert(poolstat_ops
.empty());
4931 ceph_assert(statfs_ops
.empty());
4932 ceph_assert(pool_ops
.empty());
4933 ceph_assert(waiting_for_map
.empty());
4934 ceph_assert(linger_ops
.empty());
4935 ceph_assert(check_latest_map_lingers
.empty());
4936 ceph_assert(check_latest_map_ops
.empty());
4937 ceph_assert(check_latest_map_commands
.empty());
4939 ceph_assert(!m_request_state_hook
);
4940 ceph_assert(!logger
);
4944 * Wait until this OSD map epoch is received before
4945 * sending any more operations to OSDs. Use this
4946 * when it is known that the client can't trust
4947 * anything from before this epoch (e.g. due to
4948 * client blocklist at this epoch).
4950 void Objecter::set_epoch_barrier(epoch_t epoch
)
4952 unique_lock
wl(rwlock
);
4954 ldout(cct
, 7) << __func__
<< ": barrier " << epoch
<< " (was "
4955 << epoch_barrier
<< ") current epoch " << osdmap
->get_epoch()
4957 if (epoch
> epoch_barrier
) {
4958 epoch_barrier
= epoch
;
4959 _maybe_request_map();
4965 hobject_t
Objecter::enumerate_objects_begin()
4970 hobject_t
Objecter::enumerate_objects_end()
4972 return hobject_t::get_max();
4975 template<typename T
>
4976 struct EnumerationContext
{
4978 const hobject_t end
;
4979 const cb::list filter
;
4981 const object_locator_t oloc
;
4984 fu2::unique_function
<void(bs::error_code
,
4986 hobject_t
) &&> on_finish
;
4991 EnumerationContext(Objecter
* objecter
,
4992 hobject_t end
, cb::list filter
,
4993 uint32_t max
, object_locator_t oloc
,
4994 decltype(on_finish
) on_finish
)
4995 : objecter(objecter
), end(std::move(end
)), filter(std::move(filter
)),
4996 max(max
), oloc(std::move(oloc
)), on_finish(std::move(on_finish
)) {}
4998 void operator()(bs::error_code ec
,
5002 objecter
->put_op_budget_bytes(budget
);
5006 std::move(on_finish
)(ec
, std::move(v
), std::move(h
));
5010 template<typename T
>
5011 struct CB_EnumerateReply
{
5015 std::unique_ptr
<EnumerationContext
<T
>> ctx
;
5017 CB_EnumerateReply(Objecter
* objecter
,
5018 std::unique_ptr
<EnumerationContext
<T
>>&& ctx
) :
5019 objecter(objecter
), ctx(std::move(ctx
)) {}
5021 void operator()(bs::error_code ec
) {
5022 objecter
->_enumerate_reply(std::move(bl
), ec
, std::move(ctx
));
5026 template<typename T
>
5027 void Objecter::enumerate_objects(
5029 std::string_view ns
,
5033 const cb::list
& filter_bl
,
5034 fu2::unique_function
<void(bs::error_code
,
5036 hobject_t
) &&> on_finish
) {
5037 if (!end
.is_max() && start
> end
) {
5038 lderr(cct
) << __func__
<< ": start " << start
<< " > end " << end
<< dendl
;
5039 std::move(on_finish
)(osdc_errc::precondition_violated
, {}, {});
5044 lderr(cct
) << __func__
<< ": result size may not be zero" << dendl
;
5045 std::move(on_finish
)(osdc_errc::precondition_violated
, {}, {});
5049 if (start
.is_max()) {
5050 std::move(on_finish
)({}, {}, {});
5054 shared_lock
rl(rwlock
);
5055 ceph_assert(osdmap
->get_epoch());
5056 if (!osdmap
->test_flag(CEPH_OSDMAP_SORTBITWISE
)) {
5058 lderr(cct
) << __func__
<< ": SORTBITWISE cluster flag not set" << dendl
;
5059 std::move(on_finish
)(osdc_errc::not_supported
, {}, {});
5062 const pg_pool_t
* p
= osdmap
->get_pg_pool(pool_id
);
5064 lderr(cct
) << __func__
<< ": pool " << pool_id
<< " DNE in osd epoch "
5065 << osdmap
->get_epoch() << dendl
;
5067 std::move(on_finish
)(osdc_errc::pool_dne
, {}, {});
5073 _issue_enumerate(start
,
5074 std::make_unique
<EnumerationContext
<T
>>(
5075 this, std::move(end
), filter_bl
,
5076 max
, object_locator_t
{pool_id
, ns
},
5077 std::move(on_finish
)));
5081 void Objecter::enumerate_objects
<librados::ListObjectImpl
>(
5083 std::string_view ns
,
5087 const cb::list
& filter_bl
,
5088 fu2::unique_function
<void(bs::error_code
,
5089 std::vector
<librados::ListObjectImpl
>,
5090 hobject_t
) &&> on_finish
);
5093 void Objecter::enumerate_objects
<neorados::Entry
>(
5095 std::string_view ns
,
5099 const cb::list
& filter_bl
,
5100 fu2::unique_function
<void(bs::error_code
,
5101 std::vector
<neorados::Entry
>,
5102 hobject_t
) &&> on_finish
);
5106 template<typename T
>
5107 void Objecter::_issue_enumerate(hobject_t start
,
5108 std::unique_ptr
<EnumerationContext
<T
>> ctx
) {
5111 op
.pg_nls(c
->max
, c
->filter
, start
, osdmap
->get_epoch());
5112 auto on_ack
= std::make_unique
<CB_EnumerateReply
<T
>>(this, std::move(ctx
));
5113 // I hate having to do this. Try to find a cleaner way
5115 auto epoch
= &c
->epoch
;
5116 auto budget
= &c
->budget
;
5117 auto pbl
= &on_ack
->bl
;
5119 // Issue. See you later in _enumerate_reply
5120 pg_read(start
.get_hash(),
5121 c
->oloc
, op
, pbl
, 0,
5122 Op::OpComp::create(service
.get_executor(),
5123 [c
= std::move(on_ack
)]
5124 (bs::error_code ec
) mutable {
5130 void Objecter::_issue_enumerate
<librados::ListObjectImpl
>(
5132 std::unique_ptr
<EnumerationContext
<librados::ListObjectImpl
>> ctx
);
5134 void Objecter::_issue_enumerate
<neorados::Entry
>(
5135 hobject_t start
, std::unique_ptr
<EnumerationContext
<neorados::Entry
>> ctx
);
5137 template<typename T
>
5138 void Objecter::_enumerate_reply(
5141 std::unique_ptr
<EnumerationContext
<T
>>&& ctx
)
5144 std::move(*ctx
)(ec
, {}, {});
5148 // Decode the results
5149 auto iter
= bl
.cbegin();
5150 pg_nls_response_template
<T
> response
;
5153 response
.decode(iter
);
5155 // extra_info isn't used anywhere. We do this solely to preserve
5156 // backward compatibility
5157 cb::list legacy_extra_info
;
5158 decode(legacy_extra_info
, iter
);
5160 } catch (const bs::system_error
& e
) {
5161 std::move(*ctx
)(e
.code(), {}, {});
5165 shared_lock
rl(rwlock
);
5166 auto pool
= osdmap
->get_pg_pool(ctx
->oloc
.get_pool());
5169 // pool is gone, drop any results which are now meaningless.
5170 std::move(*ctx
)(osdc_errc::pool_dne
, {}, {});
5175 if ((response
.handle
<= ctx
->end
)) {
5176 next
= response
.handle
;
5180 // drop anything after 'end'
5181 while (!response
.entries
.empty()) {
5182 uint32_t hash
= response
.entries
.back().locator
.empty() ?
5183 pool
->hash_key(response
.entries
.back().oid
,
5184 response
.entries
.back().nspace
) :
5185 pool
->hash_key(response
.entries
.back().locator
,
5186 response
.entries
.back().nspace
);
5187 hobject_t
last(response
.entries
.back().oid
,
5188 response
.entries
.back().locator
,
5191 ctx
->oloc
.get_pool(),
5192 response
.entries
.back().nspace
);
5193 if (last
< ctx
->end
)
5195 response
.entries
.pop_back();
5199 if (response
.entries
.size() <= ctx
->max
) {
5200 ctx
->max
-= response
.entries
.size();
5201 std::move(response
.entries
.begin(), response
.entries
.end(),
5202 std::back_inserter(ctx
->ls
));
5204 auto i
= response
.entries
.begin();
5205 while (ctx
->max
> 0) {
5206 ctx
->ls
.push_back(std::move(*i
));
5211 i
->locator
.empty() ?
5212 pool
->hash_key(i
->oid
, i
->nspace
) :
5213 pool
->hash_key(i
->locator
, i
->nspace
);
5215 next
= hobject_t
{i
->oid
, i
->locator
,
5218 ctx
->oloc
.get_pool(),
5222 if (next
== ctx
->end
|| ctx
->max
== 0) {
5223 std::move(*ctx
)(ec
, std::move(ctx
->ls
), std::move(next
));
5225 _issue_enumerate(next
, std::move(ctx
));
5230 void Objecter::_enumerate_reply
<librados::ListObjectImpl
>(
5233 std::unique_ptr
<EnumerationContext
<librados::ListObjectImpl
>>&& ctx
);
5236 void Objecter::_enumerate_reply
<neorados::Entry
>(
5239 std::unique_ptr
<EnumerationContext
<neorados::Entry
>>&& ctx
);
5242 using namespace librados
;
5244 template <typename T
>
5245 void do_decode(std::vector
<T
>& items
, std::vector
<cb::list
>& bls
)
5247 for (auto bl
: bls
) {
5248 auto p
= bl
.cbegin();
5255 struct C_ObjectOperation_scrub_ls
: public Context
{
5258 std::vector
<inconsistent_obj_t
> *objects
= nullptr;
5259 std::vector
<inconsistent_snapset_t
> *snapsets
= nullptr;
5262 C_ObjectOperation_scrub_ls(uint32_t* interval
,
5263 std::vector
<inconsistent_obj_t
>* objects
,
5265 : interval(interval
), objects(objects
), rval(rval
) {}
5266 C_ObjectOperation_scrub_ls(uint32_t* interval
,
5267 std::vector
<inconsistent_snapset_t
>* snapsets
,
5269 : interval(interval
), snapsets(snapsets
), rval(rval
) {}
5270 void finish(int r
) override
{
5271 if (r
< 0 && r
!= -EAGAIN
) {
5282 } catch (cb::error
&) {
5289 scrub_ls_result_t result
;
5290 auto p
= bl
.cbegin();
5292 *interval
= result
.interval
;
5294 do_decode(*objects
, result
.vals
);
5296 do_decode(*snapsets
, result
.vals
);
5301 template <typename T
>
5302 void do_scrub_ls(::ObjectOperation
* op
,
5303 const scrub_ls_arg_t
& arg
,
5304 std::vector
<T
> *items
,
5308 OSDOp
& osd_op
= op
->add_op(CEPH_OSD_OP_SCRUBLS
);
5309 op
->flags
|= CEPH_OSD_FLAG_PGOP
;
5310 ceph_assert(interval
);
5311 arg
.encode(osd_op
.indata
);
5312 unsigned p
= op
->ops
.size() - 1;
5313 auto h
= new C_ObjectOperation_scrub_ls
{interval
, items
, rval
};
5315 op
->out_bl
[p
] = &h
->bl
;
5316 op
->out_rval
[p
] = rval
;
5320 void ::ObjectOperation::scrub_ls(const librados::object_id_t
& start_after
,
5321 uint64_t max_to_get
,
5322 std::vector
<librados::inconsistent_obj_t
>* objects
,
5326 scrub_ls_arg_t arg
= {*interval
, 0, start_after
, max_to_get
};
5327 do_scrub_ls(this, arg
, objects
, interval
, rval
);
5330 void ::ObjectOperation::scrub_ls(const librados::object_id_t
& start_after
,
5331 uint64_t max_to_get
,
5332 std::vector
<librados::inconsistent_snapset_t
> *snapsets
,
5336 scrub_ls_arg_t arg
= {*interval
, 1, start_after
, max_to_get
};
5337 do_scrub_ls(this, arg
, snapsets
, interval
, rval
);