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.
16 #include "osd/OSDMap.h"
19 #include "mon/MonClient.h"
21 #include "msg/Messenger.h"
22 #include "msg/Message.h"
24 #include "messages/MPing.h"
25 #include "messages/MOSDOp.h"
26 #include "messages/MOSDOpReply.h"
27 #include "messages/MOSDBackoff.h"
28 #include "messages/MOSDMap.h"
30 #include "messages/MPoolOp.h"
31 #include "messages/MPoolOpReply.h"
33 #include "messages/MGetPoolStats.h"
34 #include "messages/MGetPoolStatsReply.h"
35 #include "messages/MStatfs.h"
36 #include "messages/MStatfsReply.h"
38 #include "messages/MMonCommand.h"
40 #include "messages/MCommand.h"
41 #include "messages/MCommandReply.h"
43 #include "messages/MWatchNotify.h"
47 #include "common/config.h"
48 #include "common/perf_counters.h"
49 #include "common/scrub_types.h"
50 #include "include/str_list.h"
51 #include "common/errno.h"
52 #include "common/EventTrace.h"
54 using ceph::real_time
;
55 using ceph::real_clock
;
57 using ceph::mono_clock
;
58 using ceph::mono_time
;
63 #define dout_subsys ceph_subsys_objecter
65 #define dout_prefix *_dout << messenger->get_myname() << ".objecter "
69 l_osdc_first
= 123200,
87 l_osdc_osdop_writefull
,
88 l_osdc_osdop_writesame
,
91 l_osdc_osdop_truncate
,
94 l_osdc_osdop_sparse_read
,
95 l_osdc_osdop_clonerange
,
96 l_osdc_osdop_getxattr
,
97 l_osdc_osdop_setxattr
,
98 l_osdc_osdop_cmpxattr
,
100 l_osdc_osdop_resetxattrs
,
101 l_osdc_osdop_tmap_up
,
102 l_osdc_osdop_tmap_put
,
103 l_osdc_osdop_tmap_get
,
107 l_osdc_osdop_src_cmpxattr
,
109 l_osdc_osdop_pgls_filter
,
112 l_osdc_linger_active
,
114 l_osdc_linger_resend
,
117 l_osdc_poolop_active
,
119 l_osdc_poolop_resend
,
121 l_osdc_poolstat_active
,
122 l_osdc_poolstat_send
,
123 l_osdc_poolstat_resend
,
125 l_osdc_statfs_active
,
127 l_osdc_statfs_resend
,
129 l_osdc_command_active
,
131 l_osdc_command_resend
,
138 l_osdc_osd_session_open
,
139 l_osdc_osd_session_close
,
142 l_osdc_osdop_omap_wr
,
143 l_osdc_osdop_omap_rd
,
144 l_osdc_osdop_omap_del
,
150 // config obs ----------------------------
152 static const char *config_keys
[] = {
157 class Objecter::RequestStateHook
: public AdminSocketHook
{
158 Objecter
*m_objecter
;
160 explicit RequestStateHook(Objecter
*objecter
);
161 bool call(std::string command
, cmdmap_t
& cmdmap
, std::string format
,
162 bufferlist
& out
) override
;
166 * This is a more limited form of C_Contexts, but that requires
167 * a ceph_context which we don't have here.
169 class ObjectOperation::C_TwoContexts
: public Context
{
173 C_TwoContexts(Context
*first
, Context
*second
) :
174 first(first
), second(second
) {}
175 void finish(int r
) override
{
182 ~C_TwoContexts() override
{
188 void ObjectOperation::add_handler(Context
*extra
) {
189 size_t last
= out_handler
.size() - 1;
190 Context
*orig
= out_handler
[last
];
192 Context
*wrapper
= new C_TwoContexts(orig
, extra
);
193 out_handler
[last
] = wrapper
;
195 out_handler
[last
] = extra
;
199 Objecter::OSDSession::unique_completion_lock
Objecter::OSDSession::get_lock(
202 if (oid
.name
.empty())
203 return unique_completion_lock();
205 static constexpr uint32_t HASH_PRIME
= 1021;
206 uint32_t h
= ceph_str_hash_linux(oid
.name
.c_str(), oid
.name
.size())
209 return unique_completion_lock(completion_locks
[h
% num_locks
],
213 const char** Objecter::get_tracked_conf_keys() const
219 void Objecter::handle_conf_change(const struct md_config_t
*conf
,
220 const std::set
<std::string
> &changed
)
222 if (changed
.count("crush_location")) {
223 update_crush_location();
227 void Objecter::update_crush_location()
229 unique_lock
wl(rwlock
);
230 crush_location
= cct
->crush_location
.get_location();
233 // messages ------------------------------
236 * initialize only internal data structures, don't initiate cluster interaction
238 void Objecter::init()
240 assert(!initialized
);
243 PerfCountersBuilder
pcb(cct
, "objecter", l_osdc_first
, l_osdc_last
);
245 pcb
.add_u64(l_osdc_op_active
, "op_active", "Operations active", "actv",
246 PerfCountersBuilder::PRIO_CRITICAL
);
247 pcb
.add_u64(l_osdc_op_laggy
, "op_laggy", "Laggy operations");
248 pcb
.add_u64_counter(l_osdc_op_send
, "op_send", "Sent operations");
249 pcb
.add_u64_counter(l_osdc_op_send_bytes
, "op_send_bytes", "Sent data", NULL
, 0, unit_t(BYTES
));
250 pcb
.add_u64_counter(l_osdc_op_resend
, "op_resend", "Resent operations");
251 pcb
.add_u64_counter(l_osdc_op_reply
, "op_reply", "Operation reply");
253 pcb
.add_u64_counter(l_osdc_op
, "op", "Operations");
254 pcb
.add_u64_counter(l_osdc_op_r
, "op_r", "Read operations", "rd",
255 PerfCountersBuilder::PRIO_CRITICAL
);
256 pcb
.add_u64_counter(l_osdc_op_w
, "op_w", "Write operations", "wr",
257 PerfCountersBuilder::PRIO_CRITICAL
);
258 pcb
.add_u64_counter(l_osdc_op_rmw
, "op_rmw", "Read-modify-write operations",
259 "rdwr", PerfCountersBuilder::PRIO_INTERESTING
);
260 pcb
.add_u64_counter(l_osdc_op_pg
, "op_pg", "PG operation");
262 pcb
.add_u64_counter(l_osdc_osdop_stat
, "osdop_stat", "Stat operations");
263 pcb
.add_u64_counter(l_osdc_osdop_create
, "osdop_create",
264 "Create object operations");
265 pcb
.add_u64_counter(l_osdc_osdop_read
, "osdop_read", "Read operations");
266 pcb
.add_u64_counter(l_osdc_osdop_write
, "osdop_write", "Write operations");
267 pcb
.add_u64_counter(l_osdc_osdop_writefull
, "osdop_writefull",
268 "Write full object operations");
269 pcb
.add_u64_counter(l_osdc_osdop_writesame
, "osdop_writesame",
270 "Write same operations");
271 pcb
.add_u64_counter(l_osdc_osdop_append
, "osdop_append",
273 pcb
.add_u64_counter(l_osdc_osdop_zero
, "osdop_zero",
274 "Set object to zero operations");
275 pcb
.add_u64_counter(l_osdc_osdop_truncate
, "osdop_truncate",
276 "Truncate object operations");
277 pcb
.add_u64_counter(l_osdc_osdop_delete
, "osdop_delete",
278 "Delete object operations");
279 pcb
.add_u64_counter(l_osdc_osdop_mapext
, "osdop_mapext",
280 "Map extent operations");
281 pcb
.add_u64_counter(l_osdc_osdop_sparse_read
, "osdop_sparse_read",
282 "Sparse read operations");
283 pcb
.add_u64_counter(l_osdc_osdop_clonerange
, "osdop_clonerange",
284 "Clone range operations");
285 pcb
.add_u64_counter(l_osdc_osdop_getxattr
, "osdop_getxattr",
286 "Get xattr operations");
287 pcb
.add_u64_counter(l_osdc_osdop_setxattr
, "osdop_setxattr",
288 "Set xattr operations");
289 pcb
.add_u64_counter(l_osdc_osdop_cmpxattr
, "osdop_cmpxattr",
290 "Xattr comparison operations");
291 pcb
.add_u64_counter(l_osdc_osdop_rmxattr
, "osdop_rmxattr",
292 "Remove xattr operations");
293 pcb
.add_u64_counter(l_osdc_osdop_resetxattrs
, "osdop_resetxattrs",
294 "Reset xattr operations");
295 pcb
.add_u64_counter(l_osdc_osdop_tmap_up
, "osdop_tmap_up",
296 "TMAP update operations");
297 pcb
.add_u64_counter(l_osdc_osdop_tmap_put
, "osdop_tmap_put",
298 "TMAP put operations");
299 pcb
.add_u64_counter(l_osdc_osdop_tmap_get
, "osdop_tmap_get",
300 "TMAP get operations");
301 pcb
.add_u64_counter(l_osdc_osdop_call
, "osdop_call",
302 "Call (execute) operations");
303 pcb
.add_u64_counter(l_osdc_osdop_watch
, "osdop_watch",
304 "Watch by object operations");
305 pcb
.add_u64_counter(l_osdc_osdop_notify
, "osdop_notify",
306 "Notify about object operations");
307 pcb
.add_u64_counter(l_osdc_osdop_src_cmpxattr
, "osdop_src_cmpxattr",
308 "Extended attribute comparison in multi operations");
309 pcb
.add_u64_counter(l_osdc_osdop_pgls
, "osdop_pgls");
310 pcb
.add_u64_counter(l_osdc_osdop_pgls_filter
, "osdop_pgls_filter");
311 pcb
.add_u64_counter(l_osdc_osdop_other
, "osdop_other", "Other operations");
313 pcb
.add_u64(l_osdc_linger_active
, "linger_active",
314 "Active lingering operations");
315 pcb
.add_u64_counter(l_osdc_linger_send
, "linger_send",
316 "Sent lingering operations");
317 pcb
.add_u64_counter(l_osdc_linger_resend
, "linger_resend",
318 "Resent lingering operations");
319 pcb
.add_u64_counter(l_osdc_linger_ping
, "linger_ping",
320 "Sent pings to lingering operations");
322 pcb
.add_u64(l_osdc_poolop_active
, "poolop_active",
323 "Active pool operations");
324 pcb
.add_u64_counter(l_osdc_poolop_send
, "poolop_send",
325 "Sent pool operations");
326 pcb
.add_u64_counter(l_osdc_poolop_resend
, "poolop_resend",
327 "Resent pool operations");
329 pcb
.add_u64(l_osdc_poolstat_active
, "poolstat_active",
330 "Active get pool stat operations");
331 pcb
.add_u64_counter(l_osdc_poolstat_send
, "poolstat_send",
332 "Pool stat operations sent");
333 pcb
.add_u64_counter(l_osdc_poolstat_resend
, "poolstat_resend",
334 "Resent pool stats");
336 pcb
.add_u64(l_osdc_statfs_active
, "statfs_active", "Statfs operations");
337 pcb
.add_u64_counter(l_osdc_statfs_send
, "statfs_send", "Sent FS stats");
338 pcb
.add_u64_counter(l_osdc_statfs_resend
, "statfs_resend",
341 pcb
.add_u64(l_osdc_command_active
, "command_active", "Active commands");
342 pcb
.add_u64_counter(l_osdc_command_send
, "command_send",
344 pcb
.add_u64_counter(l_osdc_command_resend
, "command_resend",
347 pcb
.add_u64(l_osdc_map_epoch
, "map_epoch", "OSD map epoch");
348 pcb
.add_u64_counter(l_osdc_map_full
, "map_full",
349 "Full OSD maps received");
350 pcb
.add_u64_counter(l_osdc_map_inc
, "map_inc",
351 "Incremental OSD maps received");
353 pcb
.add_u64(l_osdc_osd_sessions
, "osd_sessions",
354 "Open sessions"); // open sessions
355 pcb
.add_u64_counter(l_osdc_osd_session_open
, "osd_session_open",
357 pcb
.add_u64_counter(l_osdc_osd_session_close
, "osd_session_close",
359 pcb
.add_u64(l_osdc_osd_laggy
, "osd_laggy", "Laggy OSD sessions");
361 pcb
.add_u64_counter(l_osdc_osdop_omap_wr
, "omap_wr",
362 "OSD OMAP write operations");
363 pcb
.add_u64_counter(l_osdc_osdop_omap_rd
, "omap_rd",
364 "OSD OMAP read operations");
365 pcb
.add_u64_counter(l_osdc_osdop_omap_del
, "omap_del",
366 "OSD OMAP delete operations");
368 logger
= pcb
.create_perf_counters();
369 cct
->get_perfcounters_collection()->add(logger
);
372 m_request_state_hook
= new RequestStateHook(this);
373 AdminSocket
* admin_socket
= cct
->get_admin_socket();
374 int ret
= admin_socket
->register_command("objecter_requests",
376 m_request_state_hook
,
377 "show in-progress osd requests");
379 /* Don't warn on EEXIST, happens if multiple ceph clients
380 * are instantiated from one process */
381 if (ret
< 0 && ret
!= -EEXIST
) {
382 lderr(cct
) << "error registering admin socket command: "
383 << cpp_strerror(ret
) << dendl
;
386 update_crush_location();
388 cct
->_conf
->add_observer(this);
394 * ok, cluster interaction can happen
396 void Objecter::start(const OSDMap
* o
)
398 shared_lock
rl(rwlock
);
402 osdmap
->deepish_copy_from(*o
);
403 } else if (osdmap
->get_epoch() == 0) {
404 _maybe_request_map();
408 void Objecter::shutdown()
412 unique_lock
wl(rwlock
);
416 cct
->_conf
->remove_observer(this);
418 map
<int,OSDSession
*>::iterator p
;
419 while (!osd_sessions
.empty()) {
420 p
= osd_sessions
.begin();
421 close_session(p
->second
);
424 while(!check_latest_map_lingers
.empty()) {
425 map
<uint64_t, LingerOp
*>::iterator i
= check_latest_map_lingers
.begin();
427 check_latest_map_lingers
.erase(i
->first
);
430 while(!check_latest_map_ops
.empty()) {
431 map
<ceph_tid_t
, Op
*>::iterator i
= check_latest_map_ops
.begin();
433 check_latest_map_ops
.erase(i
->first
);
436 while(!check_latest_map_commands
.empty()) {
437 map
<ceph_tid_t
, CommandOp
*>::iterator i
438 = check_latest_map_commands
.begin();
440 check_latest_map_commands
.erase(i
->first
);
443 while(!poolstat_ops
.empty()) {
444 map
<ceph_tid_t
,PoolStatOp
*>::iterator i
= poolstat_ops
.begin();
446 poolstat_ops
.erase(i
->first
);
449 while(!statfs_ops
.empty()) {
450 map
<ceph_tid_t
, StatfsOp
*>::iterator i
= statfs_ops
.begin();
452 statfs_ops
.erase(i
->first
);
455 while(!pool_ops
.empty()) {
456 map
<ceph_tid_t
, PoolOp
*>::iterator i
= pool_ops
.begin();
458 pool_ops
.erase(i
->first
);
461 ldout(cct
, 20) << __func__
<< " clearing up homeless session..." << dendl
;
462 while(!homeless_session
->linger_ops
.empty()) {
463 std::map
<uint64_t, LingerOp
*>::iterator i
464 = homeless_session
->linger_ops
.begin();
465 ldout(cct
, 10) << " linger_op " << i
->first
<< dendl
;
466 LingerOp
*lop
= i
->second
;
468 OSDSession::unique_lock
swl(homeless_session
->lock
);
469 _session_linger_op_remove(homeless_session
, lop
);
471 linger_ops
.erase(lop
->linger_id
);
472 linger_ops_set
.erase(lop
);
476 while(!homeless_session
->ops
.empty()) {
477 std::map
<ceph_tid_t
, Op
*>::iterator i
= homeless_session
->ops
.begin();
478 ldout(cct
, 10) << " op " << i
->first
<< dendl
;
481 OSDSession::unique_lock
swl(homeless_session
->lock
);
482 _session_op_remove(homeless_session
, op
);
487 while(!homeless_session
->command_ops
.empty()) {
488 std::map
<ceph_tid_t
, CommandOp
*>::iterator i
489 = homeless_session
->command_ops
.begin();
490 ldout(cct
, 10) << " command_op " << i
->first
<< dendl
;
491 CommandOp
*cop
= i
->second
;
493 OSDSession::unique_lock
swl(homeless_session
->lock
);
494 _session_command_op_remove(homeless_session
, cop
);
500 if (timer
.cancel_event(tick_event
)) {
501 ldout(cct
, 10) << " successfully canceled tick" << dendl
;
507 cct
->get_perfcounters_collection()->remove(logger
);
512 // Let go of Objecter write lock so timer thread can shutdown
515 // Outside of lock to avoid cycle WRT calls to RequestStateHook
516 // This is safe because we guarantee no concurrent calls to
517 // shutdown() with the ::initialized check at start.
518 if (m_request_state_hook
) {
519 AdminSocket
* admin_socket
= cct
->get_admin_socket();
520 admin_socket
->unregister_command("objecter_requests");
521 delete m_request_state_hook
;
522 m_request_state_hook
= NULL
;
526 void Objecter::_send_linger(LingerOp
*info
,
529 assert(sul
.owns_lock() && sul
.mutex() == &rwlock
);
532 Context
*oncommit
= NULL
;
533 LingerOp::shared_lock
watchl(info
->watch_lock
);
534 bufferlist
*poutbl
= NULL
;
535 if (info
->registered
&& info
->is_watch
) {
536 ldout(cct
, 15) << "send_linger " << info
->linger_id
<< " reconnect"
538 opv
.push_back(OSDOp());
539 opv
.back().op
.op
= CEPH_OSD_OP_WATCH
;
540 opv
.back().op
.watch
.cookie
= info
->get_cookie();
541 opv
.back().op
.watch
.op
= CEPH_OSD_WATCH_OP_RECONNECT
;
542 opv
.back().op
.watch
.gen
= ++info
->register_gen
;
543 oncommit
= new C_Linger_Reconnect(this, info
);
545 ldout(cct
, 15) << "send_linger " << info
->linger_id
<< " register"
548 C_Linger_Commit
*c
= new C_Linger_Commit(this, info
);
549 if (!info
->is_watch
) {
556 Op
*o
= new Op(info
->target
.base_oid
, info
->target
.base_oloc
,
557 opv
, info
->target
.flags
| CEPH_OSD_FLAG_READ
,
558 oncommit
, info
->pobjver
);
560 o
->snapid
= info
->snap
;
561 o
->snapc
= info
->snapc
;
562 o
->mtime
= info
->mtime
;
564 o
->target
= info
->target
;
567 // do not resend this; we will send a new op to reregister
568 o
->should_resend
= false;
570 if (info
->register_tid
) {
571 // repeat send. cancel old registeration op, if any.
572 OSDSession::unique_lock
sl(info
->session
->lock
);
573 if (info
->session
->ops
.count(info
->register_tid
)) {
574 Op
*o
= info
->session
->ops
[info
->register_tid
];
575 _op_cancel_map_check(o
);
576 _cancel_linger_op(o
);
580 _op_submit(o
, sul
, &info
->register_tid
);
583 _op_submit_with_budget(o
, sul
, &info
->register_tid
);
586 logger
->inc(l_osdc_linger_send
);
589 void Objecter::_linger_commit(LingerOp
*info
, int r
, bufferlist
& outbl
)
591 LingerOp::unique_lock
wl(info
->watch_lock
);
592 ldout(cct
, 10) << "_linger_commit " << info
->linger_id
<< dendl
;
593 if (info
->on_reg_commit
) {
594 info
->on_reg_commit
->complete(r
);
595 info
->on_reg_commit
= NULL
;
597 if (r
< 0 && info
->on_notify_finish
) {
598 info
->on_notify_finish
->complete(r
);
599 info
->on_notify_finish
= nullptr;
602 // only tell the user the first time we do this
603 info
->registered
= true;
604 info
->pobjver
= NULL
;
606 if (!info
->is_watch
) {
607 // make note of the notify_id
608 bufferlist::iterator p
= outbl
.begin();
610 ::decode(info
->notify_id
, p
);
611 ldout(cct
, 10) << "_linger_commit notify_id=" << info
->notify_id
614 catch (buffer::error
& e
) {
619 struct C_DoWatchError
: public Context
{
621 Objecter::LingerOp
*info
;
623 C_DoWatchError(Objecter
*o
, Objecter::LingerOp
*i
, int r
)
624 : objecter(o
), info(i
), err(r
) {
626 info
->_queued_async();
628 void finish(int r
) override
{
629 Objecter::unique_lock
wl(objecter
->rwlock
);
630 bool canceled
= info
->canceled
;
634 info
->watch_context
->handle_error(info
->get_cookie(), err
);
637 info
->finished_async();
642 int Objecter::_normalize_watch_error(int r
)
644 // translate ENOENT -> ENOTCONN so that a delete->disconnection
645 // notification and a failure to reconnect becuase we raced with
646 // the delete appear the same to the user.
652 void Objecter::_linger_reconnect(LingerOp
*info
, int r
)
654 ldout(cct
, 10) << __func__
<< " " << info
->linger_id
<< " = " << r
655 << " (last_error " << info
->last_error
<< ")" << dendl
;
657 LingerOp::unique_lock
wl(info
->watch_lock
);
658 if (!info
->last_error
) {
659 r
= _normalize_watch_error(r
);
660 info
->last_error
= r
;
661 if (info
->watch_context
) {
662 finisher
->queue(new C_DoWatchError(this, info
, r
));
669 void Objecter::_send_linger_ping(LingerOp
*info
)
671 // rwlock is locked unique
672 // info->session->lock is locked
674 if (cct
->_conf
->objecter_inject_no_watch_ping
) {
675 ldout(cct
, 10) << __func__
<< " " << info
->linger_id
<< " SKIPPING"
679 if (osdmap
->test_flag(CEPH_OSDMAP_PAUSERD
)) {
680 ldout(cct
, 10) << __func__
<< " PAUSERD" << dendl
;
684 ceph::mono_time now
= ceph::mono_clock::now();
685 ldout(cct
, 10) << __func__
<< " " << info
->linger_id
<< " now " << now
688 vector
<OSDOp
> opv(1);
689 opv
[0].op
.op
= CEPH_OSD_OP_WATCH
;
690 opv
[0].op
.watch
.cookie
= info
->get_cookie();
691 opv
[0].op
.watch
.op
= CEPH_OSD_WATCH_OP_PING
;
692 opv
[0].op
.watch
.gen
= info
->register_gen
;
693 C_Linger_Ping
*onack
= new C_Linger_Ping(this, info
);
694 Op
*o
= new Op(info
->target
.base_oid
, info
->target
.base_oloc
,
695 opv
, info
->target
.flags
| CEPH_OSD_FLAG_READ
,
697 o
->target
= info
->target
;
698 o
->should_resend
= false;
700 MOSDOp
*m
= _prepare_osd_op(o
);
702 _session_op_assign(info
->session
, o
);
704 info
->ping_tid
= o
->tid
;
707 logger
->inc(l_osdc_linger_ping
);
710 void Objecter::_linger_ping(LingerOp
*info
, int r
, mono_time sent
,
711 uint32_t register_gen
)
713 LingerOp::unique_lock
l(info
->watch_lock
);
714 ldout(cct
, 10) << __func__
<< " " << info
->linger_id
715 << " sent " << sent
<< " gen " << register_gen
<< " = " << r
716 << " (last_error " << info
->last_error
717 << " register_gen " << info
->register_gen
<< ")" << dendl
;
718 if (info
->register_gen
== register_gen
) {
720 info
->watch_valid_thru
= sent
;
721 } else if (r
< 0 && !info
->last_error
) {
722 r
= _normalize_watch_error(r
);
723 info
->last_error
= r
;
724 if (info
->watch_context
) {
725 finisher
->queue(new C_DoWatchError(this, info
, r
));
729 ldout(cct
, 20) << " ignoring old gen" << dendl
;
733 int Objecter::linger_check(LingerOp
*info
)
735 LingerOp::shared_lock
l(info
->watch_lock
);
737 mono_time stamp
= info
->watch_valid_thru
;
738 if (!info
->watch_pending_async
.empty())
739 stamp
= MIN(info
->watch_valid_thru
, info
->watch_pending_async
.front());
740 auto age
= mono_clock::now() - stamp
;
742 ldout(cct
, 10) << __func__
<< " " << info
->linger_id
743 << " err " << info
->last_error
744 << " age " << age
<< dendl
;
745 if (info
->last_error
)
746 return info
->last_error
;
747 // return a safe upper bound (we are truncating to ms)
749 1 + std::chrono::duration_cast
<std::chrono::milliseconds
>(age
).count();
752 void Objecter::linger_cancel(LingerOp
*info
)
754 unique_lock
wl(rwlock
);
755 _linger_cancel(info
);
759 void Objecter::_linger_cancel(LingerOp
*info
)
761 // rwlock is locked unique
762 ldout(cct
, 20) << __func__
<< " linger_id=" << info
->linger_id
<< dendl
;
763 if (!info
->canceled
) {
764 OSDSession
*s
= info
->session
;
765 OSDSession::unique_lock
sl(s
->lock
);
766 _session_linger_op_remove(s
, info
);
769 linger_ops
.erase(info
->linger_id
);
770 linger_ops_set
.erase(info
);
771 assert(linger_ops
.size() == linger_ops_set
.size());
773 info
->canceled
= true;
776 logger
->dec(l_osdc_linger_active
);
782 Objecter::LingerOp
*Objecter::linger_register(const object_t
& oid
,
783 const object_locator_t
& oloc
,
786 LingerOp
*info
= new LingerOp
;
787 info
->target
.base_oid
= oid
;
788 info
->target
.base_oloc
= oloc
;
789 if (info
->target
.base_oloc
.key
== oid
)
790 info
->target
.base_oloc
.key
.clear();
791 info
->target
.flags
= flags
;
792 info
->watch_valid_thru
= mono_clock::now();
794 unique_lock
l(rwlock
);
797 info
->linger_id
= ++max_linger_id
;
798 ldout(cct
, 10) << __func__
<< " info " << info
799 << " linger_id " << info
->linger_id
800 << " cookie " << info
->get_cookie()
802 linger_ops
[info
->linger_id
] = info
;
803 linger_ops_set
.insert(info
);
804 assert(linger_ops
.size() == linger_ops_set
.size());
806 info
->get(); // for the caller
810 ceph_tid_t
Objecter::linger_watch(LingerOp
*info
,
812 const SnapContext
& snapc
,
818 info
->is_watch
= true;
821 info
->target
.flags
|= CEPH_OSD_FLAG_WRITE
;
825 info
->pobjver
= objver
;
826 info
->on_reg_commit
= oncommit
;
828 shunique_lock
sul(rwlock
, ceph::acquire_unique
);
829 _linger_submit(info
, sul
);
830 logger
->inc(l_osdc_linger_active
);
832 return info
->linger_id
;
835 ceph_tid_t
Objecter::linger_notify(LingerOp
*info
,
837 snapid_t snap
, bufferlist
& inbl
,
843 info
->target
.flags
|= CEPH_OSD_FLAG_READ
;
846 info
->poutbl
= poutbl
;
847 info
->pobjver
= objver
;
848 info
->on_reg_commit
= onfinish
;
850 shunique_lock
sul(rwlock
, ceph::acquire_unique
);
851 _linger_submit(info
, sul
);
852 logger
->inc(l_osdc_linger_active
);
854 return info
->linger_id
;
857 void Objecter::_linger_submit(LingerOp
*info
, shunique_lock
& sul
)
859 assert(sul
.owns_lock() && sul
.mutex() == &rwlock
);
860 assert(info
->linger_id
);
862 // Populate Op::target
863 OSDSession
*s
= NULL
;
864 _calc_target(&info
->target
, nullptr);
866 // Create LingerOp<->OSDSession relation
867 int r
= _get_session(info
->target
.osd
, &s
, sul
);
869 OSDSession::unique_lock
sl(s
->lock
);
870 _session_linger_op_assign(s
, info
);
874 _send_linger(info
, sul
);
877 struct C_DoWatchNotify
: public Context
{
879 Objecter::LingerOp
*info
;
881 C_DoWatchNotify(Objecter
*o
, Objecter::LingerOp
*i
, MWatchNotify
*m
)
882 : objecter(o
), info(i
), msg(m
) {
884 info
->_queued_async();
887 void finish(int r
) override
{
888 objecter
->_do_watch_notify(info
, msg
);
892 void Objecter::handle_watch_notify(MWatchNotify
*m
)
894 shared_lock
l(rwlock
);
899 LingerOp
*info
= reinterpret_cast<LingerOp
*>(m
->cookie
);
900 if (linger_ops_set
.count(info
) == 0) {
901 ldout(cct
, 7) << __func__
<< " cookie " << m
->cookie
<< " dne" << dendl
;
904 LingerOp::unique_lock
wl(info
->watch_lock
);
905 if (m
->opcode
== CEPH_WATCH_EVENT_DISCONNECT
) {
906 if (!info
->last_error
) {
907 info
->last_error
= -ENOTCONN
;
908 if (info
->watch_context
) {
909 finisher
->queue(new C_DoWatchError(this, info
, -ENOTCONN
));
912 } else if (!info
->is_watch
) {
913 // we have CEPH_WATCH_EVENT_NOTIFY_COMPLETE; we can do this inline
914 // since we know the only user (librados) is safe to call in
915 // fast-dispatch context
916 if (info
->notify_id
&&
917 info
->notify_id
!= m
->notify_id
) {
918 ldout(cct
, 10) << __func__
<< " reply notify " << m
->notify_id
919 << " != " << info
->notify_id
<< ", ignoring" << dendl
;
920 } else if (info
->on_notify_finish
) {
921 info
->notify_result_bl
->claim(m
->get_data());
922 info
->on_notify_finish
->complete(m
->return_code
);
924 // if we race with reconnect we might get a second notify; only
925 // notify the caller once!
926 info
->on_notify_finish
= NULL
;
929 finisher
->queue(new C_DoWatchNotify(this, info
, m
));
933 void Objecter::_do_watch_notify(LingerOp
*info
, MWatchNotify
*m
)
935 ldout(cct
, 10) << __func__
<< " " << *m
<< dendl
;
937 shared_lock
l(rwlock
);
940 if (info
->canceled
) {
945 // notify completion?
946 assert(info
->is_watch
);
947 assert(info
->watch_context
);
948 assert(m
->opcode
!= CEPH_WATCH_EVENT_DISCONNECT
);
953 case CEPH_WATCH_EVENT_NOTIFY
:
954 info
->watch_context
->handle_notify(m
->notify_id
, m
->cookie
,
955 m
->notifier_gid
, m
->bl
);
960 info
->finished_async();
965 bool Objecter::ms_dispatch(Message
*m
)
967 ldout(cct
, 10) << __func__
<< " " << cct
<< " " << *m
<< dendl
;
971 switch (m
->get_type()) {
972 // these we exlusively handle
973 case CEPH_MSG_OSD_OPREPLY
:
974 handle_osd_op_reply(static_cast<MOSDOpReply
*>(m
));
977 case CEPH_MSG_OSD_BACKOFF
:
978 handle_osd_backoff(static_cast<MOSDBackoff
*>(m
));
981 case CEPH_MSG_WATCH_NOTIFY
:
982 handle_watch_notify(static_cast<MWatchNotify
*>(m
));
986 case MSG_COMMAND_REPLY
:
987 if (m
->get_source().type() == CEPH_ENTITY_TYPE_OSD
) {
988 handle_command_reply(static_cast<MCommandReply
*>(m
));
994 case MSG_GETPOOLSTATSREPLY
:
995 handle_get_pool_stats_reply(static_cast<MGetPoolStatsReply
*>(m
));
998 case CEPH_MSG_POOLOP_REPLY
:
999 handle_pool_op_reply(static_cast<MPoolOpReply
*>(m
));
1002 case CEPH_MSG_STATFS_REPLY
:
1003 handle_fs_stats_reply(static_cast<MStatfsReply
*>(m
));
1006 // these we give others a chance to inspect
1009 case CEPH_MSG_OSD_MAP
:
1010 handle_osd_map(static_cast<MOSDMap
*>(m
));
1016 void Objecter::_scan_requests(OSDSession
*s
,
1019 map
<int64_t, bool> *pool_full_map
,
1020 map
<ceph_tid_t
, Op
*>& need_resend
,
1021 list
<LingerOp
*>& need_resend_linger
,
1022 map
<ceph_tid_t
, CommandOp
*>& need_resend_command
,
1025 assert(sul
.owns_lock() && sul
.mutex() == &rwlock
);
1027 list
<LingerOp
*> unregister_lingers
;
1029 OSDSession::unique_lock
sl(s
->lock
);
1031 // check for changed linger mappings (_before_ regular ops)
1032 map
<ceph_tid_t
,LingerOp
*>::iterator lp
= s
->linger_ops
.begin();
1033 while (lp
!= s
->linger_ops
.end()) {
1034 LingerOp
*op
= lp
->second
;
1035 assert(op
->session
== s
);
1036 // check_linger_pool_dne() may touch linger_ops; prevent iterator
1039 ldout(cct
, 10) << " checking linger op " << op
->linger_id
<< dendl
;
1040 bool unregister
, force_resend_writes
= cluster_full
;
1041 int r
= _recalc_linger_op_target(op
, sul
);
1043 force_resend_writes
= force_resend_writes
||
1044 (*pool_full_map
)[op
->target
.base_oloc
.pool
];
1046 case RECALC_OP_TARGET_NO_ACTION
:
1047 if (!force_resend
&& !force_resend_writes
)
1050 case RECALC_OP_TARGET_NEED_RESEND
:
1051 need_resend_linger
.push_back(op
);
1052 _linger_cancel_map_check(op
);
1054 case RECALC_OP_TARGET_POOL_DNE
:
1055 _check_linger_pool_dne(op
, &unregister
);
1057 ldout(cct
, 10) << " need to unregister linger op "
1058 << op
->linger_id
<< dendl
;
1060 unregister_lingers
.push_back(op
);
1066 // check for changed request mappings
1067 map
<ceph_tid_t
,Op
*>::iterator p
= s
->ops
.begin();
1068 while (p
!= s
->ops
.end()) {
1070 ++p
; // check_op_pool_dne() may touch ops; prevent iterator invalidation
1071 ldout(cct
, 10) << " checking op " << op
->tid
<< dendl
;
1072 bool force_resend_writes
= cluster_full
;
1074 force_resend_writes
= force_resend_writes
||
1075 (*pool_full_map
)[op
->target
.base_oloc
.pool
];
1076 int r
= _calc_target(&op
->target
,
1077 op
->session
? op
->session
->con
.get() : nullptr);
1079 case RECALC_OP_TARGET_NO_ACTION
:
1080 if (!force_resend
&& !(force_resend_writes
&& op
->respects_full()))
1083 case RECALC_OP_TARGET_NEED_RESEND
:
1085 _session_op_remove(op
->session
, op
);
1087 need_resend
[op
->tid
] = op
;
1088 _op_cancel_map_check(op
);
1090 case RECALC_OP_TARGET_POOL_DNE
:
1091 _check_op_pool_dne(op
, &sl
);
1097 map
<ceph_tid_t
,CommandOp
*>::iterator cp
= s
->command_ops
.begin();
1098 while (cp
!= s
->command_ops
.end()) {
1099 CommandOp
*c
= cp
->second
;
1101 ldout(cct
, 10) << " checking command " << c
->tid
<< dendl
;
1102 bool force_resend_writes
= cluster_full
;
1104 force_resend_writes
= force_resend_writes
||
1105 (*pool_full_map
)[c
->target_pg
.pool()];
1106 int r
= _calc_command_target(c
, sul
);
1108 case RECALC_OP_TARGET_NO_ACTION
:
1109 // resend if skipped map; otherwise do nothing.
1110 if (!force_resend
&& !force_resend_writes
)
1113 case RECALC_OP_TARGET_NEED_RESEND
:
1114 need_resend_command
[c
->tid
] = c
;
1116 _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 (list
<LingerOp
*>::iterator iter
= unregister_lingers
.begin();
1131 iter
!= unregister_lingers
.end();
1133 _linger_cancel(*iter
);
1138 void Objecter::handle_osd_map(MOSDMap
*m
)
1140 shunique_lock
sul(rwlock
, acquire_unique
);
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 (map
<int64_t, pg_pool_t
>::const_iterator it
1158 = osdmap
->get_pools().begin();
1159 it
!= osdmap
->get_pools().end(); ++it
)
1160 pool_full_map
[it
->first
] = _osdmap_pool_full(it
->second
);
1163 list
<LingerOp
*> need_resend_linger
;
1164 map
<ceph_tid_t
, Op
*> need_resend
;
1165 map
<ceph_tid_t
, CommandOp
*> need_resend_command
;
1167 if (m
->get_last() <= osdmap
->get_epoch()) {
1168 ldout(cct
, 3) << "handle_osd_map ignoring epochs ["
1169 << m
->get_first() << "," << m
->get_last()
1170 << "] <= " << osdmap
->get_epoch() << dendl
;
1172 ldout(cct
, 3) << "handle_osd_map got epochs ["
1173 << m
->get_first() << "," << m
->get_last()
1174 << "] > " << osdmap
->get_epoch() << dendl
;
1176 if (osdmap
->get_epoch()) {
1177 bool skipped_map
= false;
1178 // we want incrementals
1179 for (epoch_t e
= osdmap
->get_epoch() + 1;
1183 if (osdmap
->get_epoch() == e
-1 &&
1184 m
->incremental_maps
.count(e
)) {
1185 ldout(cct
, 3) << "handle_osd_map decoding incremental epoch " << e
1187 OSDMap::Incremental
inc(m
->incremental_maps
[e
]);
1188 osdmap
->apply_incremental(inc
);
1190 emit_blacklist_events(inc
);
1192 logger
->inc(l_osdc_map_inc
);
1194 else if (m
->maps
.count(e
)) {
1195 ldout(cct
, 3) << "handle_osd_map decoding full epoch " << e
<< dendl
;
1196 OSDMap
*new_osdmap
= new OSDMap();
1197 new_osdmap
->decode(m
->maps
[e
]);
1199 emit_blacklist_events(*osdmap
, *new_osdmap
);
1201 osdmap
= new_osdmap
;
1203 logger
->inc(l_osdc_map_full
);
1206 if (e
>= m
->get_oldest()) {
1207 ldout(cct
, 3) << "handle_osd_map requesting missing epoch "
1208 << osdmap
->get_epoch()+1 << dendl
;
1209 _maybe_request_map();
1212 ldout(cct
, 3) << "handle_osd_map missing epoch "
1213 << osdmap
->get_epoch()+1
1214 << ", jumping to " << m
->get_oldest() << dendl
;
1215 e
= m
->get_oldest() - 1;
1219 logger
->set(l_osdc_map_epoch
, osdmap
->get_epoch());
1221 cluster_full
= cluster_full
|| _osdmap_full_flag();
1222 update_pool_full_map(pool_full_map
);
1224 // check all outstanding requests on every epoch
1225 _scan_requests(homeless_session
, skipped_map
, cluster_full
,
1226 &pool_full_map
, need_resend
,
1227 need_resend_linger
, need_resend_command
, sul
);
1228 for (map
<int,OSDSession
*>::iterator p
= osd_sessions
.begin();
1229 p
!= osd_sessions
.end(); ) {
1230 OSDSession
*s
= p
->second
;
1231 _scan_requests(s
, skipped_map
, cluster_full
,
1232 &pool_full_map
, need_resend
,
1233 need_resend_linger
, need_resend_command
, sul
);
1235 // osd down or addr change?
1236 if (!osdmap
->is_up(s
->osd
) ||
1238 s
->con
->get_peer_addr() != osdmap
->get_inst(s
->osd
).addr
)) {
1243 assert(e
== osdmap
->get_epoch());
1247 // first map. we want the full thing.
1248 if (m
->maps
.count(m
->get_last())) {
1249 for (map
<int,OSDSession
*>::iterator p
= osd_sessions
.begin();
1250 p
!= osd_sessions
.end(); ++p
) {
1251 OSDSession
*s
= p
->second
;
1252 _scan_requests(s
, false, false, NULL
, need_resend
,
1253 need_resend_linger
, need_resend_command
, sul
);
1255 ldout(cct
, 3) << "handle_osd_map decoding full epoch "
1256 << m
->get_last() << dendl
;
1257 osdmap
->decode(m
->maps
[m
->get_last()]);
1259 _scan_requests(homeless_session
, false, false, NULL
,
1260 need_resend
, need_resend_linger
,
1261 need_resend_command
, sul
);
1263 ldout(cct
, 3) << "handle_osd_map hmm, i want a full map, requesting"
1265 monc
->sub_want("osdmap", 0, CEPH_SUBSCRIBE_ONETIME
);
1271 // make sure need_resend targets reflect latest map
1272 for (auto p
= need_resend
.begin(); p
!= need_resend
.end(); ) {
1274 if (op
->target
.epoch
< osdmap
->get_epoch()) {
1275 ldout(cct
, 10) << __func__
<< " checking op " << p
->first
<< dendl
;
1276 int r
= _calc_target(&op
->target
, nullptr);
1277 if (r
== RECALC_OP_TARGET_POOL_DNE
) {
1278 p
= need_resend
.erase(p
);
1279 _check_op_pool_dne(op
, nullptr);
1288 bool pauserd
= osdmap
->test_flag(CEPH_OSDMAP_PAUSERD
);
1289 bool pausewr
= osdmap
->test_flag(CEPH_OSDMAP_PAUSEWR
) || _osdmap_full_flag()
1290 || _osdmap_has_pool_full();
1293 if (was_pauserd
|| was_pausewr
|| pauserd
|| pausewr
||
1294 osdmap
->get_epoch() < epoch_barrier
) {
1295 _maybe_request_map();
1299 for (map
<ceph_tid_t
, Op
*>::iterator p
= need_resend
.begin();
1300 p
!= need_resend
.end(); ++p
) {
1302 OSDSession
*s
= op
->session
;
1303 bool mapped_session
= false;
1305 int r
= _map_session(&op
->target
, &s
, sul
);
1307 mapped_session
= true;
1311 OSDSession::unique_lock
sl(s
->lock
);
1312 if (mapped_session
) {
1313 _session_op_assign(s
, op
);
1315 if (op
->should_resend
) {
1316 if (!op
->session
->is_homeless() && !op
->target
.paused
) {
1317 logger
->inc(l_osdc_op_resend
);
1321 _op_cancel_map_check(op
);
1322 _cancel_linger_op(op
);
1327 for (list
<LingerOp
*>::iterator p
= need_resend_linger
.begin();
1328 p
!= need_resend_linger
.end(); ++p
) {
1331 _calc_target(&op
->target
, nullptr);
1332 OSDSession
*s
= NULL
;
1333 const int r
= _get_session(op
->target
.osd
, &s
, sul
);
1339 if (!op
->session
->is_homeless()) {
1340 logger
->inc(l_osdc_linger_resend
);
1341 _send_linger(op
, sul
);
1344 for (map
<ceph_tid_t
,CommandOp
*>::iterator p
= need_resend_command
.begin();
1345 p
!= need_resend_command
.end(); ++p
) {
1346 CommandOp
*c
= p
->second
;
1347 if (c
->target
.osd
>= 0) {
1348 _assign_command_session(c
, sul
);
1349 if (c
->session
&& !c
->session
->is_homeless()) {
1357 // finish any Contexts that were waiting on a map update
1358 map
<epoch_t
,list
< pair
< Context
*, int > > >::iterator p
=
1359 waiting_for_map
.begin();
1360 while (p
!= waiting_for_map
.end() &&
1361 p
->first
<= osdmap
->get_epoch()) {
1362 //go through the list and call the onfinish methods
1363 for (list
<pair
<Context
*, int> >::iterator i
= p
->second
.begin();
1364 i
!= p
->second
.end(); ++i
) {
1365 i
->first
->complete(i
->second
);
1367 waiting_for_map
.erase(p
++);
1370 monc
->sub_got("osdmap", osdmap
->get_epoch());
1372 if (!waiting_for_map
.empty()) {
1373 _maybe_request_map();
1377 void Objecter::enable_blacklist_events()
1379 unique_lock
wl(rwlock
);
1381 blacklist_events_enabled
= true;
1384 void Objecter::consume_blacklist_events(std::set
<entity_addr_t
> *events
)
1386 unique_lock
wl(rwlock
);
1388 if (events
->empty()) {
1389 events
->swap(blacklist_events
);
1391 for (const auto &i
: blacklist_events
) {
1394 blacklist_events
.clear();
1398 void Objecter::emit_blacklist_events(const OSDMap::Incremental
&inc
)
1400 if (!blacklist_events_enabled
) {
1404 for (const auto &i
: inc
.new_blacklist
) {
1405 blacklist_events
.insert(i
.first
);
1409 void Objecter::emit_blacklist_events(const OSDMap
&old_osd_map
,
1410 const OSDMap
&new_osd_map
)
1412 if (!blacklist_events_enabled
) {
1416 std::set
<entity_addr_t
> old_set
;
1417 std::set
<entity_addr_t
> new_set
;
1419 old_osd_map
.get_blacklist(&old_set
);
1420 new_osd_map
.get_blacklist(&new_set
);
1422 std::set
<entity_addr_t
> delta_set
;
1423 std::set_difference(
1424 new_set
.begin(), new_set
.end(), old_set
.begin(), old_set
.end(),
1425 std::inserter(delta_set
, delta_set
.begin()));
1426 blacklist_events
.insert(delta_set
.begin(), delta_set
.end());
1431 void Objecter::C_Op_Map_Latest::finish(int r
)
1433 if (r
== -EAGAIN
|| r
== -ECANCELED
)
1436 lgeneric_subdout(objecter
->cct
, objecter
, 10)
1437 << "op_map_latest r=" << r
<< " tid=" << tid
1438 << " latest " << latest
<< dendl
;
1440 Objecter::unique_lock
wl(objecter
->rwlock
);
1442 map
<ceph_tid_t
, Op
*>::iterator iter
=
1443 objecter
->check_latest_map_ops
.find(tid
);
1444 if (iter
== objecter
->check_latest_map_ops
.end()) {
1445 lgeneric_subdout(objecter
->cct
, objecter
, 10)
1446 << "op_map_latest op "<< tid
<< " not found" << dendl
;
1450 Op
*op
= iter
->second
;
1451 objecter
->check_latest_map_ops
.erase(iter
);
1453 lgeneric_subdout(objecter
->cct
, objecter
, 20)
1454 << "op_map_latest op "<< op
<< dendl
;
1456 if (op
->map_dne_bound
== 0)
1457 op
->map_dne_bound
= latest
;
1459 OSDSession::unique_lock
sl(op
->session
->lock
, defer_lock
);
1460 objecter
->_check_op_pool_dne(op
, &sl
);
1465 int Objecter::pool_snap_by_name(int64_t poolid
, const char *snap_name
,
1466 snapid_t
*snap
) const
1468 shared_lock
rl(rwlock
);
1470 auto& pools
= osdmap
->get_pools();
1471 auto iter
= pools
.find(poolid
);
1472 if (iter
== pools
.end()) {
1475 const pg_pool_t
& pg_pool
= iter
->second
;
1476 for (auto p
= pg_pool
.snaps
.begin();
1477 p
!= pg_pool
.snaps
.end();
1479 if (p
->second
.name
== snap_name
) {
1487 int Objecter::pool_snap_get_info(int64_t poolid
, snapid_t snap
,
1488 pool_snap_info_t
*info
) const
1490 shared_lock
rl(rwlock
);
1492 auto& pools
= osdmap
->get_pools();
1493 auto iter
= pools
.find(poolid
);
1494 if (iter
== pools
.end()) {
1497 const pg_pool_t
& pg_pool
= iter
->second
;
1498 auto p
= pg_pool
.snaps
.find(snap
);
1499 if (p
== pg_pool
.snaps
.end())
1506 int Objecter::pool_snap_list(int64_t poolid
, vector
<uint64_t> *snaps
)
1508 shared_lock
rl(rwlock
);
1510 const pg_pool_t
*pi
= osdmap
->get_pg_pool(poolid
);
1513 for (map
<snapid_t
,pool_snap_info_t
>::const_iterator p
= pi
->snaps
.begin();
1514 p
!= pi
->snaps
.end();
1516 snaps
->push_back(p
->first
);
1521 // sl may be unlocked.
1522 void Objecter::_check_op_pool_dne(Op
*op
, unique_lock
*sl
)
1524 // rwlock is locked unique
1526 if (op
->target
.pool_ever_existed
) {
1527 // the pool previously existed and now it does not, which means it
1529 op
->map_dne_bound
= osdmap
->get_epoch();
1530 ldout(cct
, 10) << "check_op_pool_dne tid " << op
->tid
1531 << " pool previously exists but now does not"
1534 ldout(cct
, 10) << "check_op_pool_dne tid " << op
->tid
1535 << " current " << osdmap
->get_epoch()
1536 << " map_dne_bound " << op
->map_dne_bound
1539 if (op
->map_dne_bound
> 0) {
1540 if (osdmap
->get_epoch() >= op
->map_dne_bound
) {
1541 // we had a new enough map
1542 ldout(cct
, 10) << "check_op_pool_dne tid " << op
->tid
1543 << " concluding pool " << op
->target
.base_pgid
.pool()
1546 op
->onfinish
->complete(-ENOENT
);
1549 OSDSession
*s
= op
->session
;
1552 assert(sl
->mutex() == &s
->lock
);
1553 bool session_locked
= sl
->owns_lock();
1554 if (!session_locked
) {
1558 if (!session_locked
) {
1562 _finish_op(op
, 0); // no session
1566 _send_op_map_check(op
);
1570 void Objecter::_send_op_map_check(Op
*op
)
1572 // rwlock is locked unique
1574 if (check_latest_map_ops
.count(op
->tid
) == 0) {
1576 check_latest_map_ops
[op
->tid
] = op
;
1577 C_Op_Map_Latest
*c
= new C_Op_Map_Latest(this, op
->tid
);
1578 monc
->get_version("osdmap", &c
->latest
, NULL
, c
);
1582 void Objecter::_op_cancel_map_check(Op
*op
)
1584 // rwlock is locked unique
1585 map
<ceph_tid_t
, Op
*>::iterator iter
=
1586 check_latest_map_ops
.find(op
->tid
);
1587 if (iter
!= check_latest_map_ops
.end()) {
1588 Op
*op
= iter
->second
;
1590 check_latest_map_ops
.erase(iter
);
1594 // linger pool check
1596 void Objecter::C_Linger_Map_Latest::finish(int r
)
1598 if (r
== -EAGAIN
|| r
== -ECANCELED
) {
1599 // ignore callback; we will retry in resend_mon_ops()
1603 unique_lock
wl(objecter
->rwlock
);
1605 map
<uint64_t, LingerOp
*>::iterator iter
=
1606 objecter
->check_latest_map_lingers
.find(linger_id
);
1607 if (iter
== objecter
->check_latest_map_lingers
.end()) {
1611 LingerOp
*op
= iter
->second
;
1612 objecter
->check_latest_map_lingers
.erase(iter
);
1614 if (op
->map_dne_bound
== 0)
1615 op
->map_dne_bound
= latest
;
1618 objecter
->_check_linger_pool_dne(op
, &unregister
);
1621 objecter
->_linger_cancel(op
);
1627 void Objecter::_check_linger_pool_dne(LingerOp
*op
, bool *need_unregister
)
1629 // rwlock is locked unique
1631 *need_unregister
= false;
1633 if (op
->register_gen
> 0) {
1634 ldout(cct
, 10) << "_check_linger_pool_dne linger_id " << op
->linger_id
1635 << " pool previously existed but now does not"
1637 op
->map_dne_bound
= osdmap
->get_epoch();
1639 ldout(cct
, 10) << "_check_linger_pool_dne linger_id " << op
->linger_id
1640 << " current " << osdmap
->get_epoch()
1641 << " map_dne_bound " << op
->map_dne_bound
1644 if (op
->map_dne_bound
> 0) {
1645 if (osdmap
->get_epoch() >= op
->map_dne_bound
) {
1646 LingerOp::unique_lock wl
{op
->watch_lock
};
1647 if (op
->on_reg_commit
) {
1648 op
->on_reg_commit
->complete(-ENOENT
);
1649 op
->on_reg_commit
= nullptr;
1651 if (op
->on_notify_finish
) {
1652 op
->on_notify_finish
->complete(-ENOENT
);
1653 op
->on_notify_finish
= nullptr;
1655 *need_unregister
= true;
1658 _send_linger_map_check(op
);
1662 void Objecter::_send_linger_map_check(LingerOp
*op
)
1665 if (check_latest_map_lingers
.count(op
->linger_id
) == 0) {
1667 check_latest_map_lingers
[op
->linger_id
] = op
;
1668 C_Linger_Map_Latest
*c
= new C_Linger_Map_Latest(this, op
->linger_id
);
1669 monc
->get_version("osdmap", &c
->latest
, NULL
, c
);
1673 void Objecter::_linger_cancel_map_check(LingerOp
*op
)
1675 // rwlock is locked unique
1677 map
<uint64_t, LingerOp
*>::iterator iter
=
1678 check_latest_map_lingers
.find(op
->linger_id
);
1679 if (iter
!= check_latest_map_lingers
.end()) {
1680 LingerOp
*op
= iter
->second
;
1682 check_latest_map_lingers
.erase(iter
);
1686 // command pool check
1688 void Objecter::C_Command_Map_Latest::finish(int r
)
1690 if (r
== -EAGAIN
|| r
== -ECANCELED
) {
1691 // ignore callback; we will retry in resend_mon_ops()
1695 unique_lock
wl(objecter
->rwlock
);
1697 map
<uint64_t, CommandOp
*>::iterator iter
=
1698 objecter
->check_latest_map_commands
.find(tid
);
1699 if (iter
== objecter
->check_latest_map_commands
.end()) {
1703 CommandOp
*c
= iter
->second
;
1704 objecter
->check_latest_map_commands
.erase(iter
);
1706 if (c
->map_dne_bound
== 0)
1707 c
->map_dne_bound
= latest
;
1709 OSDSession::unique_lock
sul(c
->session
->lock
);
1710 objecter
->_check_command_map_dne(c
);
1716 void Objecter::_check_command_map_dne(CommandOp
*c
)
1718 // rwlock is locked unique
1719 // session is locked unique
1721 ldout(cct
, 10) << "_check_command_map_dne tid " << c
->tid
1722 << " current " << osdmap
->get_epoch()
1723 << " map_dne_bound " << c
->map_dne_bound
1725 if (c
->map_dne_bound
> 0) {
1726 if (osdmap
->get_epoch() >= c
->map_dne_bound
) {
1727 _finish_command(c
, c
->map_check_error
, c
->map_check_error_str
);
1730 _send_command_map_check(c
);
1734 void Objecter::_send_command_map_check(CommandOp
*c
)
1736 // rwlock is locked unique
1737 // session is locked unique
1740 if (check_latest_map_commands
.count(c
->tid
) == 0) {
1742 check_latest_map_commands
[c
->tid
] = c
;
1743 C_Command_Map_Latest
*f
= new C_Command_Map_Latest(this, c
->tid
);
1744 monc
->get_version("osdmap", &f
->latest
, NULL
, f
);
1748 void Objecter::_command_cancel_map_check(CommandOp
*c
)
1750 // rwlock is locked uniqe
1752 map
<uint64_t, CommandOp
*>::iterator iter
=
1753 check_latest_map_commands
.find(c
->tid
);
1754 if (iter
!= check_latest_map_commands
.end()) {
1755 CommandOp
*c
= iter
->second
;
1757 check_latest_map_commands
.erase(iter
);
1763 * Look up OSDSession by OSD id.
1765 * @returns 0 on success, or -EAGAIN if the lock context requires
1766 * promotion to write.
1768 int Objecter::_get_session(int osd
, OSDSession
**session
, shunique_lock
& sul
)
1770 assert(sul
&& sul
.mutex() == &rwlock
);
1773 *session
= homeless_session
;
1774 ldout(cct
, 20) << __func__
<< " osd=" << osd
<< " returning homeless"
1779 map
<int,OSDSession
*>::iterator p
= osd_sessions
.find(osd
);
1780 if (p
!= osd_sessions
.end()) {
1781 OSDSession
*s
= p
->second
;
1784 ldout(cct
, 20) << __func__
<< " s=" << s
<< " osd=" << osd
<< " "
1785 << s
->get_nref() << dendl
;
1788 if (!sul
.owns_lock()) {
1791 OSDSession
*s
= new OSDSession(cct
, osd
);
1792 osd_sessions
[osd
] = s
;
1793 s
->con
= messenger
->get_connection(osdmap
->get_inst(osd
));
1794 s
->con
->set_priv(s
->get());
1795 logger
->inc(l_osdc_osd_session_open
);
1796 logger
->set(l_osdc_osd_sessions
, osd_sessions
.size());
1799 ldout(cct
, 20) << __func__
<< " s=" << s
<< " osd=" << osd
<< " "
1800 << s
->get_nref() << dendl
;
1804 void Objecter::put_session(Objecter::OSDSession
*s
)
1806 if (s
&& !s
->is_homeless()) {
1807 ldout(cct
, 20) << __func__
<< " s=" << s
<< " osd=" << s
->osd
<< " "
1808 << s
->get_nref() << dendl
;
1813 void Objecter::get_session(Objecter::OSDSession
*s
)
1817 if (!s
->is_homeless()) {
1818 ldout(cct
, 20) << __func__
<< " s=" << s
<< " osd=" << s
->osd
<< " "
1819 << s
->get_nref() << dendl
;
1824 void Objecter::_reopen_session(OSDSession
*s
)
1826 // rwlock is locked unique
1827 // s->lock is locked
1829 entity_inst_t inst
= osdmap
->get_inst(s
->osd
);
1830 ldout(cct
, 10) << "reopen_session osd." << s
->osd
<< " session, addr now "
1833 s
->con
->set_priv(NULL
);
1834 s
->con
->mark_down();
1835 logger
->inc(l_osdc_osd_session_close
);
1837 s
->con
= messenger
->get_connection(inst
);
1838 s
->con
->set_priv(s
->get());
1840 logger
->inc(l_osdc_osd_session_open
);
1843 void Objecter::close_session(OSDSession
*s
)
1845 // rwlock is locked unique
1847 ldout(cct
, 10) << "close_session for osd." << s
->osd
<< dendl
;
1849 s
->con
->set_priv(NULL
);
1850 s
->con
->mark_down();
1851 logger
->inc(l_osdc_osd_session_close
);
1853 OSDSession::unique_lock
sl(s
->lock
);
1855 std::list
<LingerOp
*> homeless_lingers
;
1856 std::list
<CommandOp
*> homeless_commands
;
1857 std::list
<Op
*> homeless_ops
;
1859 while (!s
->linger_ops
.empty()) {
1860 std::map
<uint64_t, LingerOp
*>::iterator i
= s
->linger_ops
.begin();
1861 ldout(cct
, 10) << " linger_op " << i
->first
<< dendl
;
1862 homeless_lingers
.push_back(i
->second
);
1863 _session_linger_op_remove(s
, i
->second
);
1866 while (!s
->ops
.empty()) {
1867 std::map
<ceph_tid_t
, Op
*>::iterator i
= s
->ops
.begin();
1868 ldout(cct
, 10) << " op " << i
->first
<< dendl
;
1869 homeless_ops
.push_back(i
->second
);
1870 _session_op_remove(s
, i
->second
);
1873 while (!s
->command_ops
.empty()) {
1874 std::map
<ceph_tid_t
, CommandOp
*>::iterator i
= s
->command_ops
.begin();
1875 ldout(cct
, 10) << " command_op " << i
->first
<< dendl
;
1876 homeless_commands
.push_back(i
->second
);
1877 _session_command_op_remove(s
, i
->second
);
1880 osd_sessions
.erase(s
->osd
);
1884 // Assign any leftover ops to the homeless session
1886 OSDSession::unique_lock
hsl(homeless_session
->lock
);
1887 for (std::list
<LingerOp
*>::iterator i
= homeless_lingers
.begin();
1888 i
!= homeless_lingers
.end(); ++i
) {
1889 _session_linger_op_assign(homeless_session
, *i
);
1891 for (std::list
<Op
*>::iterator i
= homeless_ops
.begin();
1892 i
!= homeless_ops
.end(); ++i
) {
1893 _session_op_assign(homeless_session
, *i
);
1895 for (std::list
<CommandOp
*>::iterator i
= homeless_commands
.begin();
1896 i
!= homeless_commands
.end(); ++i
) {
1897 _session_command_op_assign(homeless_session
, *i
);
1901 logger
->set(l_osdc_osd_sessions
, osd_sessions
.size());
1904 void Objecter::wait_for_osd_map()
1906 unique_lock
l(rwlock
);
1907 if (osdmap
->get_epoch()) {
1912 // Leave this since it goes with C_SafeCond
1917 C_SafeCond
*context
= new C_SafeCond(&lock
, &cond
, &done
, NULL
);
1918 waiting_for_map
[0].push_back(pair
<Context
*, int>(context
, 0));
1925 struct C_Objecter_GetVersion
: public Context
{
1927 uint64_t oldest
, newest
;
1929 C_Objecter_GetVersion(Objecter
*o
, Context
*c
)
1930 : objecter(o
), oldest(0), newest(0), fin(c
) {}
1931 void finish(int r
) override
{
1933 objecter
->get_latest_version(oldest
, newest
, fin
);
1934 } else if (r
== -EAGAIN
) { // try again as instructed
1935 objecter
->wait_for_latest_osdmap(fin
);
1937 // it doesn't return any other error codes!
1943 void Objecter::wait_for_latest_osdmap(Context
*fin
)
1945 ldout(cct
, 10) << __func__
<< dendl
;
1946 C_Objecter_GetVersion
*c
= new C_Objecter_GetVersion(this, fin
);
1947 monc
->get_version("osdmap", &c
->newest
, &c
->oldest
, c
);
1950 void Objecter::get_latest_version(epoch_t oldest
, epoch_t newest
, Context
*fin
)
1952 unique_lock
wl(rwlock
);
1953 _get_latest_version(oldest
, newest
, fin
);
1956 void Objecter::_get_latest_version(epoch_t oldest
, epoch_t newest
,
1959 // rwlock is locked unique
1960 if (osdmap
->get_epoch() >= newest
) {
1961 ldout(cct
, 10) << __func__
<< " latest " << newest
<< ", have it" << dendl
;
1967 ldout(cct
, 10) << __func__
<< " latest " << newest
<< ", waiting" << dendl
;
1968 _wait_for_new_map(fin
, newest
, 0);
1971 void Objecter::maybe_request_map()
1973 shared_lock
rl(rwlock
);
1974 _maybe_request_map();
1977 void Objecter::_maybe_request_map()
1981 if (_osdmap_full_flag()
1982 || osdmap
->test_flag(CEPH_OSDMAP_PAUSERD
)
1983 || osdmap
->test_flag(CEPH_OSDMAP_PAUSEWR
)) {
1984 ldout(cct
, 10) << "_maybe_request_map subscribing (continuous) to next "
1985 "osd map (FULL flag is set)" << dendl
;
1988 << "_maybe_request_map subscribing (onetime) to next osd map" << dendl
;
1989 flag
= CEPH_SUBSCRIBE_ONETIME
;
1991 epoch_t epoch
= osdmap
->get_epoch() ? osdmap
->get_epoch()+1 : 0;
1992 if (monc
->sub_want("osdmap", epoch
, flag
)) {
1997 void Objecter::_wait_for_new_map(Context
*c
, epoch_t epoch
, int err
)
1999 // rwlock is locked unique
2000 waiting_for_map
[epoch
].push_back(pair
<Context
*, int>(c
, err
));
2001 _maybe_request_map();
2006 * Use this together with wait_for_map: this is a pre-check to avoid
2007 * allocating a Context for wait_for_map if we can see that we
2008 * definitely already have the epoch.
2010 * This does *not* replace the need to handle the return value of
2011 * wait_for_map: just because we don't have it in this pre-check
2012 * doesn't mean we won't have it when calling back into wait_for_map,
2013 * since the objecter lock is dropped in between.
2015 bool Objecter::have_map(const epoch_t epoch
)
2017 shared_lock
rl(rwlock
);
2018 if (osdmap
->get_epoch() >= epoch
) {
2025 bool Objecter::wait_for_map(epoch_t epoch
, Context
*c
, int err
)
2027 unique_lock
wl(rwlock
);
2028 if (osdmap
->get_epoch() >= epoch
) {
2031 _wait_for_new_map(c
, epoch
, err
);
2035 void Objecter::kick_requests(OSDSession
*session
)
2037 ldout(cct
, 10) << "kick_requests for osd." << session
->osd
<< dendl
;
2039 map
<uint64_t, LingerOp
*> lresend
;
2040 unique_lock
wl(rwlock
);
2042 OSDSession::unique_lock
sl(session
->lock
);
2043 _kick_requests(session
, lresend
);
2046 _linger_ops_resend(lresend
, wl
);
2049 void Objecter::_kick_requests(OSDSession
*session
,
2050 map
<uint64_t, LingerOp
*>& lresend
)
2052 // rwlock is locked unique
2055 session
->backoffs
.clear();
2056 session
->backoffs_by_id
.clear();
2059 map
<ceph_tid_t
,Op
*> resend
; // resend in tid order
2060 for (map
<ceph_tid_t
, Op
*>::iterator p
= session
->ops
.begin();
2061 p
!= session
->ops
.end();) {
2064 logger
->inc(l_osdc_op_resend
);
2065 if (op
->should_resend
) {
2066 if (!op
->target
.paused
)
2067 resend
[op
->tid
] = op
;
2069 _op_cancel_map_check(op
);
2070 _cancel_linger_op(op
);
2074 while (!resend
.empty()) {
2075 _send_op(resend
.begin()->second
);
2076 resend
.erase(resend
.begin());
2080 for (map
<ceph_tid_t
, LingerOp
*>::iterator j
= session
->linger_ops
.begin();
2081 j
!= session
->linger_ops
.end(); ++j
) {
2082 LingerOp
*op
= j
->second
;
2084 logger
->inc(l_osdc_linger_resend
);
2085 assert(lresend
.count(j
->first
) == 0);
2086 lresend
[j
->first
] = op
;
2090 map
<uint64_t,CommandOp
*> cresend
; // resend in order
2091 for (map
<ceph_tid_t
, CommandOp
*>::iterator k
= session
->command_ops
.begin();
2092 k
!= session
->command_ops
.end(); ++k
) {
2093 logger
->inc(l_osdc_command_resend
);
2094 cresend
[k
->first
] = k
->second
;
2096 while (!cresend
.empty()) {
2097 _send_command(cresend
.begin()->second
);
2098 cresend
.erase(cresend
.begin());
2102 void Objecter::_linger_ops_resend(map
<uint64_t, LingerOp
*>& lresend
,
2105 assert(ul
.owns_lock());
2106 shunique_lock
sul(std::move(ul
));
2107 while (!lresend
.empty()) {
2108 LingerOp
*op
= lresend
.begin()->second
;
2109 if (!op
->canceled
) {
2110 _send_linger(op
, sul
);
2113 lresend
.erase(lresend
.begin());
2115 ul
= unique_lock(sul
.release_to_unique());
2118 void Objecter::start_tick()
2120 assert(tick_event
== 0);
2122 timer
.add_event(ceph::make_timespan(cct
->_conf
->objecter_tick_interval
),
2123 &Objecter::tick
, this);
2126 void Objecter::tick()
2128 shared_lock
rl(rwlock
);
2130 ldout(cct
, 10) << "tick" << dendl
;
2132 // we are only called by C_Tick
2136 // we raced with shutdown
2137 ldout(cct
, 10) << __func__
<< " raced with shutdown" << dendl
;
2141 set
<OSDSession
*> toping
;
2144 // look for laggy requests
2145 auto cutoff
= ceph::mono_clock::now();
2146 cutoff
-= ceph::make_timespan(cct
->_conf
->objecter_timeout
); // timeout
2148 unsigned laggy_ops
= 0;
2150 for (map
<int,OSDSession
*>::iterator siter
= osd_sessions
.begin();
2151 siter
!= osd_sessions
.end(); ++siter
) {
2152 OSDSession
*s
= siter
->second
;
2153 OSDSession::lock_guard
l(s
->lock
);
2155 for (map
<ceph_tid_t
,Op
*>::iterator p
= s
->ops
.begin();
2159 assert(op
->session
);
2160 if (op
->stamp
< cutoff
) {
2161 ldout(cct
, 2) << " tid " << p
->first
<< " on osd." << op
->session
->osd
2162 << " is laggy" << dendl
;
2167 for (map
<uint64_t,LingerOp
*>::iterator p
= s
->linger_ops
.begin();
2168 p
!= s
->linger_ops
.end();
2170 LingerOp
*op
= p
->second
;
2171 LingerOp::unique_lock
wl(op
->watch_lock
);
2172 assert(op
->session
);
2173 ldout(cct
, 10) << " pinging osd that serves lingering tid " << p
->first
2174 << " (osd." << op
->session
->osd
<< ")" << dendl
;
2176 if (op
->is_watch
&& op
->registered
&& !op
->last_error
)
2177 _send_linger_ping(op
);
2179 for (map
<uint64_t,CommandOp
*>::iterator p
= s
->command_ops
.begin();
2180 p
!= s
->command_ops
.end();
2182 CommandOp
*op
= p
->second
;
2183 assert(op
->session
);
2184 ldout(cct
, 10) << " pinging osd that serves command tid " << p
->first
2185 << " (osd." << op
->session
->osd
<< ")" << dendl
;
2191 if (num_homeless_ops
|| !toping
.empty()) {
2192 _maybe_request_map();
2195 logger
->set(l_osdc_op_laggy
, laggy_ops
);
2196 logger
->set(l_osdc_osd_laggy
, toping
.size());
2198 if (!toping
.empty()) {
2199 // send a ping to these osds, to ensure we detect any session resets
2200 // (osd reply message policy is lossy)
2201 for (set
<OSDSession
*>::const_iterator i
= toping
.begin();
2204 (*i
)->con
->send_message(new MPing
);
2208 // Make sure we don't resechedule if we wake up after shutdown
2210 tick_event
= timer
.reschedule_me(ceph::make_timespan(
2211 cct
->_conf
->objecter_tick_interval
));
2215 void Objecter::resend_mon_ops()
2217 unique_lock
wl(rwlock
);
2219 ldout(cct
, 10) << "resend_mon_ops" << dendl
;
2221 for (map
<ceph_tid_t
,PoolStatOp
*>::iterator p
= poolstat_ops
.begin();
2222 p
!= poolstat_ops
.end();
2224 _poolstat_submit(p
->second
);
2225 logger
->inc(l_osdc_poolstat_resend
);
2228 for (map
<ceph_tid_t
,StatfsOp
*>::iterator p
= statfs_ops
.begin();
2229 p
!= statfs_ops
.end();
2231 _fs_stats_submit(p
->second
);
2232 logger
->inc(l_osdc_statfs_resend
);
2235 for (map
<ceph_tid_t
,PoolOp
*>::iterator p
= pool_ops
.begin();
2236 p
!= pool_ops
.end();
2238 _pool_op_submit(p
->second
);
2239 logger
->inc(l_osdc_poolop_resend
);
2242 for (map
<ceph_tid_t
, Op
*>::iterator p
= check_latest_map_ops
.begin();
2243 p
!= check_latest_map_ops
.end();
2245 C_Op_Map_Latest
*c
= new C_Op_Map_Latest(this, p
->second
->tid
);
2246 monc
->get_version("osdmap", &c
->latest
, NULL
, c
);
2249 for (map
<uint64_t, LingerOp
*>::iterator p
= check_latest_map_lingers
.begin();
2250 p
!= check_latest_map_lingers
.end();
2252 C_Linger_Map_Latest
*c
2253 = new C_Linger_Map_Latest(this, p
->second
->linger_id
);
2254 monc
->get_version("osdmap", &c
->latest
, NULL
, c
);
2257 for (map
<uint64_t, CommandOp
*>::iterator p
2258 = check_latest_map_commands
.begin();
2259 p
!= check_latest_map_commands
.end();
2261 C_Command_Map_Latest
*c
= new C_Command_Map_Latest(this, p
->second
->tid
);
2262 monc
->get_version("osdmap", &c
->latest
, NULL
, c
);
2266 // read | write ---------------------------
2268 void Objecter::op_submit(Op
*op
, ceph_tid_t
*ptid
, int *ctx_budget
)
2270 shunique_lock
rl(rwlock
, ceph::acquire_shared
);
2274 op
->trace
.event("op submit");
2275 _op_submit_with_budget(op
, rl
, ptid
, ctx_budget
);
2278 void Objecter::_op_submit_with_budget(Op
*op
, shunique_lock
& sul
,
2282 assert(initialized
);
2284 assert(op
->ops
.size() == op
->out_bl
.size());
2285 assert(op
->ops
.size() == op
->out_rval
.size());
2286 assert(op
->ops
.size() == op
->out_handler
.size());
2288 // throttle. before we look at any state, because
2289 // _take_op_budget() may drop our lock while it blocks.
2290 if (!op
->ctx_budgeted
|| (ctx_budget
&& (*ctx_budget
== -1))) {
2291 int op_budget
= _take_op_budget(op
, sul
);
2292 // take and pass out the budget for the first OP
2293 // in the context session
2294 if (ctx_budget
&& (*ctx_budget
== -1)) {
2295 *ctx_budget
= op_budget
;
2299 if (osd_timeout
> timespan(0)) {
2301 op
->tid
= ++last_tid
;
2303 op
->ontimeout
= timer
.add_event(osd_timeout
,
2305 op_cancel(tid
, -ETIMEDOUT
); });
2308 _op_submit(op
, sul
, ptid
);
2311 void Objecter::_send_op_account(Op
*op
)
2315 // add to gather set(s)
2319 ldout(cct
, 20) << " note: not requesting reply" << dendl
;
2322 logger
->inc(l_osdc_op_active
);
2323 logger
->inc(l_osdc_op
);
2325 if ((op
->target
.flags
& (CEPH_OSD_FLAG_READ
| CEPH_OSD_FLAG_WRITE
)) ==
2326 (CEPH_OSD_FLAG_READ
|CEPH_OSD_FLAG_WRITE
))
2327 logger
->inc(l_osdc_op_rmw
);
2328 else if (op
->target
.flags
& CEPH_OSD_FLAG_WRITE
)
2329 logger
->inc(l_osdc_op_w
);
2330 else if (op
->target
.flags
& CEPH_OSD_FLAG_READ
)
2331 logger
->inc(l_osdc_op_r
);
2333 if (op
->target
.flags
& CEPH_OSD_FLAG_PGOP
)
2334 logger
->inc(l_osdc_op_pg
);
2336 for (vector
<OSDOp
>::iterator p
= op
->ops
.begin(); p
!= op
->ops
.end(); ++p
) {
2337 int code
= l_osdc_osdop_other
;
2339 case CEPH_OSD_OP_STAT
: code
= l_osdc_osdop_stat
; break;
2340 case CEPH_OSD_OP_CREATE
: code
= l_osdc_osdop_create
; break;
2341 case CEPH_OSD_OP_READ
: code
= l_osdc_osdop_read
; break;
2342 case CEPH_OSD_OP_WRITE
: code
= l_osdc_osdop_write
; break;
2343 case CEPH_OSD_OP_WRITEFULL
: code
= l_osdc_osdop_writefull
; break;
2344 case CEPH_OSD_OP_WRITESAME
: code
= l_osdc_osdop_writesame
; break;
2345 case CEPH_OSD_OP_APPEND
: code
= l_osdc_osdop_append
; break;
2346 case CEPH_OSD_OP_ZERO
: code
= l_osdc_osdop_zero
; break;
2347 case CEPH_OSD_OP_TRUNCATE
: code
= l_osdc_osdop_truncate
; break;
2348 case CEPH_OSD_OP_DELETE
: code
= l_osdc_osdop_delete
; break;
2349 case CEPH_OSD_OP_MAPEXT
: code
= l_osdc_osdop_mapext
; break;
2350 case CEPH_OSD_OP_SPARSE_READ
: code
= l_osdc_osdop_sparse_read
; break;
2351 case CEPH_OSD_OP_GETXATTR
: code
= l_osdc_osdop_getxattr
; break;
2352 case CEPH_OSD_OP_SETXATTR
: code
= l_osdc_osdop_setxattr
; break;
2353 case CEPH_OSD_OP_CMPXATTR
: code
= l_osdc_osdop_cmpxattr
; break;
2354 case CEPH_OSD_OP_RMXATTR
: code
= l_osdc_osdop_rmxattr
; break;
2355 case CEPH_OSD_OP_RESETXATTRS
: code
= l_osdc_osdop_resetxattrs
; break;
2356 case CEPH_OSD_OP_TMAPUP
: code
= l_osdc_osdop_tmap_up
; break;
2357 case CEPH_OSD_OP_TMAPPUT
: code
= l_osdc_osdop_tmap_put
; break;
2358 case CEPH_OSD_OP_TMAPGET
: code
= l_osdc_osdop_tmap_get
; break;
2360 // OMAP read operations
2361 case CEPH_OSD_OP_OMAPGETVALS
:
2362 case CEPH_OSD_OP_OMAPGETKEYS
:
2363 case CEPH_OSD_OP_OMAPGETHEADER
:
2364 case CEPH_OSD_OP_OMAPGETVALSBYKEYS
:
2365 case CEPH_OSD_OP_OMAP_CMP
: code
= l_osdc_osdop_omap_rd
; break;
2367 // OMAP write operations
2368 case CEPH_OSD_OP_OMAPSETVALS
:
2369 case CEPH_OSD_OP_OMAPSETHEADER
: code
= l_osdc_osdop_omap_wr
; break;
2371 // OMAP del operations
2372 case CEPH_OSD_OP_OMAPCLEAR
:
2373 case CEPH_OSD_OP_OMAPRMKEYS
: code
= l_osdc_osdop_omap_del
; break;
2375 case CEPH_OSD_OP_CALL
: code
= l_osdc_osdop_call
; break;
2376 case CEPH_OSD_OP_WATCH
: code
= l_osdc_osdop_watch
; break;
2377 case CEPH_OSD_OP_NOTIFY
: code
= l_osdc_osdop_notify
; break;
2384 void Objecter::_op_submit(Op
*op
, shunique_lock
& sul
, ceph_tid_t
*ptid
)
2388 ldout(cct
, 10) << __func__
<< " op " << op
<< dendl
;
2391 assert(op
->session
== NULL
);
2392 OSDSession
*s
= NULL
;
2394 bool check_for_latest_map
= _calc_target(&op
->target
, nullptr)
2395 == RECALC_OP_TARGET_POOL_DNE
;
2397 // Try to get a session, including a retry if we need to take write lock
2398 int r
= _get_session(op
->target
.osd
, &s
, sul
);
2400 (check_for_latest_map
&& sul
.owns_lock_shared())) {
2401 epoch_t orig_epoch
= osdmap
->get_epoch();
2403 if (cct
->_conf
->objecter_debug_inject_relock_delay
) {
2407 if (orig_epoch
!= osdmap
->get_epoch()) {
2408 // map changed; recalculate mapping
2409 ldout(cct
, 10) << __func__
<< " relock raced with osdmap, recalc target"
2411 check_for_latest_map
= _calc_target(&op
->target
, nullptr)
2412 == RECALC_OP_TARGET_POOL_DNE
;
2422 r
= _get_session(op
->target
.osd
, &s
, sul
);
2425 assert(s
); // may be homeless
2427 _send_op_account(op
);
2431 assert(op
->target
.flags
& (CEPH_OSD_FLAG_READ
|CEPH_OSD_FLAG_WRITE
));
2433 if (osdmap_full_try
) {
2434 op
->target
.flags
|= CEPH_OSD_FLAG_FULL_TRY
;
2437 bool need_send
= false;
2439 if (osdmap
->get_epoch() < epoch_barrier
) {
2440 ldout(cct
, 10) << " barrier, paused " << op
<< " tid " << op
->tid
2442 op
->target
.paused
= true;
2443 _maybe_request_map();
2444 } else if ((op
->target
.flags
& CEPH_OSD_FLAG_WRITE
) &&
2445 osdmap
->test_flag(CEPH_OSDMAP_PAUSEWR
)) {
2446 ldout(cct
, 10) << " paused modify " << op
<< " tid " << op
->tid
2448 op
->target
.paused
= true;
2449 _maybe_request_map();
2450 } else if ((op
->target
.flags
& CEPH_OSD_FLAG_READ
) &&
2451 osdmap
->test_flag(CEPH_OSDMAP_PAUSERD
)) {
2452 ldout(cct
, 10) << " paused read " << op
<< " tid " << op
->tid
2454 op
->target
.paused
= true;
2455 _maybe_request_map();
2456 } else if (op
->respects_full() &&
2457 (_osdmap_full_flag() ||
2458 _osdmap_pool_full(op
->target
.base_oloc
.pool
))) {
2459 ldout(cct
, 0) << " FULL, paused modify " << op
<< " tid "
2460 << op
->tid
<< dendl
;
2461 op
->target
.paused
= true;
2462 _maybe_request_map();
2463 } else if (!s
->is_homeless()) {
2466 _maybe_request_map();
2471 m
= _prepare_osd_op(op
);
2474 OSDSession::unique_lock
sl(s
->lock
);
2476 op
->tid
= ++last_tid
;
2478 ldout(cct
, 10) << "_op_submit oid " << op
->target
.base_oid
2479 << " '" << op
->target
.base_oloc
<< "' '"
2480 << op
->target
.target_oloc
<< "' " << op
->ops
<< " tid "
2481 << op
->tid
<< " osd." << (!s
->is_homeless() ? s
->osd
: -1)
2484 _session_op_assign(s
, op
);
2490 // Last chance to touch Op here, after giving up session lock it can
2491 // be freed at any time by response handler.
2492 ceph_tid_t tid
= op
->tid
;
2493 if (check_for_latest_map
) {
2494 _send_op_map_check(op
);
2503 ldout(cct
, 5) << num_in_flight
<< " in flight" << dendl
;
2506 int Objecter::op_cancel(OSDSession
*s
, ceph_tid_t tid
, int r
)
2508 assert(initialized
);
2510 OSDSession::unique_lock
sl(s
->lock
);
2512 map
<ceph_tid_t
, Op
*>::iterator p
= s
->ops
.find(tid
);
2513 if (p
== s
->ops
.end()) {
2514 ldout(cct
, 10) << __func__
<< " tid " << tid
<< " dne in session "
2520 ldout(cct
, 20) << " revoking rx buffer for " << tid
2521 << " on " << s
->con
<< dendl
;
2522 s
->con
->revoke_rx_buffer(tid
);
2525 ldout(cct
, 10) << __func__
<< " tid " << tid
<< " in session " << s
->osd
2530 op
->onfinish
->complete(r
);
2531 op
->onfinish
= NULL
;
2533 _op_cancel_map_check(op
);
2540 int Objecter::op_cancel(ceph_tid_t tid
, int r
)
2544 unique_lock
wl(rwlock
);
2545 ret
= _op_cancel(tid
, r
);
2550 int Objecter::op_cancel(const vector
<ceph_tid_t
>& tids
, int r
)
2552 unique_lock
wl(rwlock
);
2553 ldout(cct
,10) << __func__
<< " " << tids
<< dendl
;
2554 for (auto tid
: tids
) {
2560 int Objecter::_op_cancel(ceph_tid_t tid
, int r
)
2564 ldout(cct
, 5) << __func__
<< ": cancelling tid " << tid
<< " r=" << r
2569 for (map
<int, OSDSession
*>::iterator siter
= osd_sessions
.begin();
2570 siter
!= osd_sessions
.end(); ++siter
) {
2571 OSDSession
*s
= siter
->second
;
2572 OSDSession::shared_lock
sl(s
->lock
);
2573 if (s
->ops
.find(tid
) != s
->ops
.end()) {
2575 ret
= op_cancel(s
, tid
, r
);
2576 if (ret
== -ENOENT
) {
2577 /* oh no! raced, maybe tid moved to another session, restarting */
2584 ldout(cct
, 5) << __func__
<< ": tid " << tid
2585 << " not found in live sessions" << dendl
;
2587 // Handle case where the op is in homeless session
2588 OSDSession::shared_lock
sl(homeless_session
->lock
);
2589 if (homeless_session
->ops
.find(tid
) != homeless_session
->ops
.end()) {
2591 ret
= op_cancel(homeless_session
, tid
, r
);
2592 if (ret
== -ENOENT
) {
2593 /* oh no! raced, maybe tid moved to another session, restarting */
2602 ldout(cct
, 5) << __func__
<< ": tid " << tid
2603 << " not found in homeless session" << dendl
;
2609 epoch_t
Objecter::op_cancel_writes(int r
, int64_t pool
)
2611 unique_lock
wl(rwlock
);
2613 std::vector
<ceph_tid_t
> to_cancel
;
2616 for (map
<int, OSDSession
*>::iterator siter
= osd_sessions
.begin();
2617 siter
!= osd_sessions
.end(); ++siter
) {
2618 OSDSession
*s
= siter
->second
;
2619 OSDSession::shared_lock
sl(s
->lock
);
2620 for (map
<ceph_tid_t
, Op
*>::iterator op_i
= s
->ops
.begin();
2621 op_i
!= s
->ops
.end(); ++op_i
) {
2622 if (op_i
->second
->target
.flags
& CEPH_OSD_FLAG_WRITE
2623 && (pool
== -1 || op_i
->second
->target
.target_oloc
.pool
== pool
)) {
2624 to_cancel
.push_back(op_i
->first
);
2629 for (std::vector
<ceph_tid_t
>::iterator titer
= to_cancel
.begin();
2630 titer
!= to_cancel
.end();
2632 int cancel_result
= op_cancel(s
, *titer
, r
);
2633 // We hold rwlock across search and cancellation, so cancels
2634 // should always succeed
2635 assert(cancel_result
== 0);
2637 if (!found
&& to_cancel
.size())
2642 const epoch_t epoch
= osdmap
->get_epoch();
2653 bool Objecter::is_pg_changed(
2655 const vector
<int>& oldacting
,
2657 const vector
<int>& newacting
,
2660 if (OSDMap::primary_changed(
2666 if (any_change
&& oldacting
!= newacting
)
2668 return false; // same primary (tho replicas may have changed)
2671 bool Objecter::target_should_be_paused(op_target_t
*t
)
2673 const pg_pool_t
*pi
= osdmap
->get_pg_pool(t
->base_oloc
.pool
);
2674 bool pauserd
= osdmap
->test_flag(CEPH_OSDMAP_PAUSERD
);
2675 bool pausewr
= osdmap
->test_flag(CEPH_OSDMAP_PAUSEWR
) ||
2676 _osdmap_full_flag() || _osdmap_pool_full(*pi
);
2678 return (t
->flags
& CEPH_OSD_FLAG_READ
&& pauserd
) ||
2679 (t
->flags
& CEPH_OSD_FLAG_WRITE
&& pausewr
) ||
2680 (osdmap
->get_epoch() < epoch_barrier
);
2684 * Locking public accessor for _osdmap_full_flag
2686 bool Objecter::osdmap_full_flag() const
2688 shared_lock
rl(rwlock
);
2690 return _osdmap_full_flag();
2693 bool Objecter::osdmap_pool_full(const int64_t pool_id
) const
2695 shared_lock
rl(rwlock
);
2697 if (_osdmap_full_flag()) {
2701 return _osdmap_pool_full(pool_id
);
2704 bool Objecter::_osdmap_pool_full(const int64_t pool_id
) const
2706 const pg_pool_t
*pool
= osdmap
->get_pg_pool(pool_id
);
2708 ldout(cct
, 4) << __func__
<< ": DNE pool " << pool_id
<< dendl
;
2712 return _osdmap_pool_full(*pool
);
2715 bool Objecter::_osdmap_has_pool_full() const
2717 for (map
<int64_t, pg_pool_t
>::const_iterator it
2718 = osdmap
->get_pools().begin();
2719 it
!= osdmap
->get_pools().end(); ++it
) {
2720 if (_osdmap_pool_full(it
->second
))
2726 bool Objecter::_osdmap_pool_full(const pg_pool_t
&p
) const
2728 return p
.has_flag(pg_pool_t::FLAG_FULL
) && honor_osdmap_full
;
2732 * Wrapper around osdmap->test_flag for special handling of the FULL flag.
2734 bool Objecter::_osdmap_full_flag() const
2736 // Ignore the FULL flag if the caller has honor_osdmap_full
2737 return osdmap
->test_flag(CEPH_OSDMAP_FULL
) && honor_osdmap_full
;
2740 void Objecter::update_pool_full_map(map
<int64_t, bool>& pool_full_map
)
2742 for (map
<int64_t, pg_pool_t
>::const_iterator it
2743 = osdmap
->get_pools().begin();
2744 it
!= osdmap
->get_pools().end(); ++it
) {
2745 if (pool_full_map
.find(it
->first
) == pool_full_map
.end()) {
2746 pool_full_map
[it
->first
] = _osdmap_pool_full(it
->second
);
2748 pool_full_map
[it
->first
] = _osdmap_pool_full(it
->second
) ||
2749 pool_full_map
[it
->first
];
2754 int64_t Objecter::get_object_hash_position(int64_t pool
, const string
& key
,
2757 shared_lock
rl(rwlock
);
2758 const pg_pool_t
*p
= osdmap
->get_pg_pool(pool
);
2761 return p
->hash_key(key
, ns
);
2764 int64_t Objecter::get_object_pg_hash_position(int64_t pool
, const string
& key
,
2767 shared_lock
rl(rwlock
);
2768 const pg_pool_t
*p
= osdmap
->get_pg_pool(pool
);
2771 return p
->raw_hash_to_pg(p
->hash_key(key
, ns
));
2774 int Objecter::_calc_target(op_target_t
*t
, Connection
*con
, bool any_change
)
2777 bool is_read
= t
->flags
& CEPH_OSD_FLAG_READ
;
2778 bool is_write
= t
->flags
& CEPH_OSD_FLAG_WRITE
;
2779 t
->epoch
= osdmap
->get_epoch();
2780 ldout(cct
,20) << __func__
<< " epoch " << t
->epoch
2781 << " base " << t
->base_oid
<< " " << t
->base_oloc
2782 << " precalc_pgid " << (int)t
->precalc_pgid
2783 << " pgid " << t
->base_pgid
2784 << (is_read
? " is_read" : "")
2785 << (is_write
? " is_write" : "")
2788 const pg_pool_t
*pi
= osdmap
->get_pg_pool(t
->base_oloc
.pool
);
2791 return RECALC_OP_TARGET_POOL_DNE
;
2793 ldout(cct
,30) << __func__
<< " base pi " << pi
2794 << " pg_num " << pi
->get_pg_num() << dendl
;
2796 bool force_resend
= false;
2797 if (osdmap
->get_epoch() == pi
->last_force_op_resend
) {
2798 if (t
->last_force_resend
< pi
->last_force_op_resend
) {
2799 t
->last_force_resend
= pi
->last_force_op_resend
;
2800 force_resend
= true;
2801 } else if (t
->last_force_resend
== 0) {
2802 force_resend
= true;
2807 t
->target_oid
= t
->base_oid
;
2808 t
->target_oloc
= t
->base_oloc
;
2809 if ((t
->flags
& CEPH_OSD_FLAG_IGNORE_OVERLAY
) == 0) {
2810 if (is_read
&& pi
->has_read_tier())
2811 t
->target_oloc
.pool
= pi
->read_tier
;
2812 if (is_write
&& pi
->has_write_tier())
2813 t
->target_oloc
.pool
= pi
->write_tier
;
2814 pi
= osdmap
->get_pg_pool(t
->target_oloc
.pool
);
2817 return RECALC_OP_TARGET_POOL_DNE
;
2822 if (t
->precalc_pgid
) {
2823 assert(t
->flags
& CEPH_OSD_FLAG_IGNORE_OVERLAY
);
2824 assert(t
->base_oid
.name
.empty()); // make sure this is a pg op
2825 assert(t
->base_oloc
.pool
== (int64_t)t
->base_pgid
.pool());
2826 pgid
= t
->base_pgid
;
2828 int ret
= osdmap
->object_locator_to_pg(t
->target_oid
, t
->target_oloc
,
2830 if (ret
== -ENOENT
) {
2832 return RECALC_OP_TARGET_POOL_DNE
;
2835 ldout(cct
,20) << __func__
<< " target " << t
->target_oid
<< " "
2836 << t
->target_oloc
<< " -> pgid " << pgid
<< dendl
;
2837 ldout(cct
,30) << __func__
<< " target pi " << pi
2838 << " pg_num " << pi
->get_pg_num() << dendl
;
2839 t
->pool_ever_existed
= true;
2841 int size
= pi
->size
;
2842 int min_size
= pi
->min_size
;
2843 unsigned pg_num
= pi
->get_pg_num();
2844 int up_primary
, acting_primary
;
2845 vector
<int> up
, acting
;
2846 osdmap
->pg_to_up_acting_osds(pgid
, &up
, &up_primary
,
2847 &acting
, &acting_primary
);
2848 bool sort_bitwise
= osdmap
->test_flag(CEPH_OSDMAP_SORTBITWISE
);
2849 bool recovery_deletes
= osdmap
->test_flag(CEPH_OSDMAP_RECOVERY_DELETES
);
2850 unsigned prev_seed
= ceph_stable_mod(pgid
.ps(), t
->pg_num
, t
->pg_num_mask
);
2851 pg_t
prev_pgid(prev_seed
, pgid
.pool());
2852 if (any_change
&& PastIntervals::is_new_interval(
2869 t
->recovery_deletes
,
2872 force_resend
= true;
2875 bool unpaused
= false;
2876 if (t
->paused
&& !target_should_be_paused(t
)) {
2881 bool legacy_change
=
2884 t
->acting_primary
, t
->acting
, acting_primary
, acting
,
2885 t
->used_replica
|| any_change
);
2888 split
= prev_pgid
.is_split(t
->pg_num
, pg_num
, nullptr);
2891 if (legacy_change
|| split
|| force_resend
) {
2894 t
->acting_primary
= acting_primary
;
2895 t
->up_primary
= up_primary
;
2898 t
->min_size
= min_size
;
2900 t
->pg_num_mask
= pi
->get_pg_num_mask();
2901 osdmap
->get_primary_shard(
2902 pg_t(ceph_stable_mod(pgid
.ps(), t
->pg_num
, t
->pg_num_mask
), pgid
.pool()),
2904 t
->sort_bitwise
= sort_bitwise
;
2905 t
->recovery_deletes
= recovery_deletes
;
2906 ldout(cct
, 10) << __func__
<< " "
2907 << " raw pgid " << pgid
<< " -> actual " << t
->actual_pgid
2908 << " acting " << acting
2909 << " primary " << acting_primary
<< dendl
;
2910 t
->used_replica
= false;
2911 if (acting_primary
== -1) {
2915 bool read
= is_read
&& !is_write
;
2916 if (read
&& (t
->flags
& CEPH_OSD_FLAG_BALANCE_READS
)) {
2917 int p
= rand() % acting
.size();
2919 t
->used_replica
= true;
2921 ldout(cct
, 10) << " chose random osd." << osd
<< " of " << acting
2923 } else if (read
&& (t
->flags
& CEPH_OSD_FLAG_LOCALIZE_READS
) &&
2924 acting
.size() > 1) {
2925 // look for a local replica. prefer the primary if the
2926 // distance is the same.
2928 int best_locality
= 0;
2929 for (unsigned i
= 0; i
< acting
.size(); ++i
) {
2930 int locality
= osdmap
->crush
->get_common_ancestor_distance(
2931 cct
, acting
[i
], crush_location
);
2932 ldout(cct
, 20) << __func__
<< " localize: rank " << i
2933 << " osd." << acting
[i
]
2934 << " locality " << locality
<< dendl
;
2936 (locality
>= 0 && best_locality
>= 0 &&
2937 locality
< best_locality
) ||
2938 (best_locality
< 0 && locality
>= 0)) {
2940 best_locality
= locality
;
2942 t
->used_replica
= true;
2948 osd
= acting_primary
;
2953 if (legacy_change
|| unpaused
|| force_resend
) {
2954 return RECALC_OP_TARGET_NEED_RESEND
;
2957 (osdmap
->require_osd_release
>= CEPH_RELEASE_LUMINOUS
||
2958 HAVE_FEATURE(osdmap
->get_xinfo(acting_primary
).features
,
2959 RESEND_ON_SPLIT
))) {
2960 return RECALC_OP_TARGET_NEED_RESEND
;
2962 return RECALC_OP_TARGET_NO_ACTION
;
2965 int Objecter::_map_session(op_target_t
*target
, OSDSession
**s
,
2968 _calc_target(target
, nullptr);
2969 return _get_session(target
->osd
, s
, sul
);
2972 void Objecter::_session_op_assign(OSDSession
*to
, Op
*op
)
2974 // to->lock is locked
2975 assert(op
->session
== NULL
);
2980 to
->ops
[op
->tid
] = op
;
2982 if (to
->is_homeless()) {
2986 ldout(cct
, 15) << __func__
<< " " << to
->osd
<< " " << op
->tid
<< dendl
;
2989 void Objecter::_session_op_remove(OSDSession
*from
, Op
*op
)
2991 assert(op
->session
== from
);
2992 // from->lock is locked
2994 if (from
->is_homeless()) {
2998 from
->ops
.erase(op
->tid
);
3002 ldout(cct
, 15) << __func__
<< " " << from
->osd
<< " " << op
->tid
<< dendl
;
3005 void Objecter::_session_linger_op_assign(OSDSession
*to
, LingerOp
*op
)
3007 // to lock is locked unique
3008 assert(op
->session
== NULL
);
3010 if (to
->is_homeless()) {
3016 to
->linger_ops
[op
->linger_id
] = op
;
3018 ldout(cct
, 15) << __func__
<< " " << to
->osd
<< " " << op
->linger_id
3022 void Objecter::_session_linger_op_remove(OSDSession
*from
, LingerOp
*op
)
3024 assert(from
== op
->session
);
3025 // from->lock is locked unique
3027 if (from
->is_homeless()) {
3031 from
->linger_ops
.erase(op
->linger_id
);
3035 ldout(cct
, 15) << __func__
<< " " << from
->osd
<< " " << op
->linger_id
3039 void Objecter::_session_command_op_remove(OSDSession
*from
, CommandOp
*op
)
3041 assert(from
== op
->session
);
3042 // from->lock is locked
3044 if (from
->is_homeless()) {
3048 from
->command_ops
.erase(op
->tid
);
3052 ldout(cct
, 15) << __func__
<< " " << from
->osd
<< " " << op
->tid
<< dendl
;
3055 void Objecter::_session_command_op_assign(OSDSession
*to
, CommandOp
*op
)
3057 // to->lock is locked
3058 assert(op
->session
== NULL
);
3061 if (to
->is_homeless()) {
3067 to
->command_ops
[op
->tid
] = op
;
3069 ldout(cct
, 15) << __func__
<< " " << to
->osd
<< " " << op
->tid
<< dendl
;
3072 int Objecter::_recalc_linger_op_target(LingerOp
*linger_op
,
3075 // rwlock is locked unique
3077 int r
= _calc_target(&linger_op
->target
, nullptr, true);
3078 if (r
== RECALC_OP_TARGET_NEED_RESEND
) {
3079 ldout(cct
, 10) << "recalc_linger_op_target tid " << linger_op
->linger_id
3080 << " pgid " << linger_op
->target
.pgid
3081 << " acting " << linger_op
->target
.acting
<< dendl
;
3083 OSDSession
*s
= NULL
;
3084 r
= _get_session(linger_op
->target
.osd
, &s
, sul
);
3087 if (linger_op
->session
!= s
) {
3088 // NB locking two sessions (s and linger_op->session) at the
3089 // same time here is only safe because we are the only one that
3090 // takes two, and we are holding rwlock for write. Disable
3091 // lockdep because it doesn't know that.
3092 OSDSession::unique_lock
sl(s
->lock
);
3093 _session_linger_op_remove(linger_op
->session
, linger_op
);
3094 _session_linger_op_assign(s
, linger_op
);
3098 return RECALC_OP_TARGET_NEED_RESEND
;
3103 void Objecter::_cancel_linger_op(Op
*op
)
3105 ldout(cct
, 15) << "cancel_op " << op
->tid
<< dendl
;
3107 assert(!op
->should_resend
);
3109 delete op
->onfinish
;
3116 void Objecter::_finish_op(Op
*op
, int r
)
3118 ldout(cct
, 15) << "finish_op " << op
->tid
<< dendl
;
3120 // op->session->lock is locked unique or op->session is null
3122 if (!op
->ctx_budgeted
&& op
->budgeted
)
3125 if (op
->ontimeout
&& r
!= -ETIMEDOUT
)
3126 timer
.cancel_event(op
->ontimeout
);
3129 _session_op_remove(op
->session
, op
);
3132 logger
->dec(l_osdc_op_active
);
3134 assert(check_latest_map_ops
.find(op
->tid
) == check_latest_map_ops
.end());
3141 void Objecter::finish_op(OSDSession
*session
, ceph_tid_t tid
)
3143 ldout(cct
, 15) << "finish_op " << tid
<< dendl
;
3144 shared_lock
rl(rwlock
);
3146 OSDSession::unique_lock
wl(session
->lock
);
3148 map
<ceph_tid_t
, Op
*>::iterator iter
= session
->ops
.find(tid
);
3149 if (iter
== session
->ops
.end())
3152 Op
*op
= iter
->second
;
3157 MOSDOp
*Objecter::_prepare_osd_op(Op
*op
)
3161 int flags
= op
->target
.flags
;
3162 flags
|= CEPH_OSD_FLAG_KNOWN_REDIR
;
3164 // Nothing checks this any longer, but needed for compatibility with
3165 // pre-luminous osds
3166 flags
|= CEPH_OSD_FLAG_ONDISK
;
3168 if (!honor_osdmap_full
)
3169 flags
|= CEPH_OSD_FLAG_FULL_FORCE
;
3171 op
->target
.paused
= false;
3172 op
->stamp
= ceph::mono_clock::now();
3174 hobject_t hobj
= op
->target
.get_hobj();
3175 MOSDOp
*m
= new MOSDOp(client_inc
, op
->tid
,
3176 hobj
, op
->target
.actual_pgid
,
3177 osdmap
->get_epoch(),
3178 flags
, op
->features
);
3180 m
->set_snapid(op
->snapid
);
3181 m
->set_snap_seq(op
->snapc
.seq
);
3182 m
->set_snaps(op
->snapc
.snaps
);
3185 m
->set_mtime(op
->mtime
);
3186 m
->set_retry_attempt(op
->attempts
++);
3188 if (!op
->trace
.valid() && cct
->_conf
->osdc_blkin_trace_all
) {
3189 op
->trace
.init("op", &trace_endpoint
);
3193 m
->set_priority(op
->priority
);
3195 m
->set_priority(cct
->_conf
->osd_client_op_priority
);
3197 if (op
->reqid
!= osd_reqid_t()) {
3198 m
->set_reqid(op
->reqid
);
3201 logger
->inc(l_osdc_op_send
);
3203 for (unsigned i
= 0; i
< m
->ops
.size(); i
++) {
3204 sum
+= m
->ops
[i
].indata
.length();
3206 logger
->inc(l_osdc_op_send_bytes
, sum
);
3211 void Objecter::_send_op(Op
*op
, MOSDOp
*m
)
3214 // op->session->lock is locked
3217 auto p
= op
->session
->backoffs
.find(op
->target
.actual_pgid
);
3218 if (p
!= op
->session
->backoffs
.end()) {
3219 hobject_t hoid
= op
->target
.get_hobj();
3220 auto q
= p
->second
.lower_bound(hoid
);
3221 if (q
!= p
->second
.begin()) {
3223 if (hoid
>= q
->second
.end
) {
3227 if (q
!= p
->second
.end()) {
3228 ldout(cct
, 20) << __func__
<< " ? " << q
->first
<< " [" << q
->second
.begin
3229 << "," << q
->second
.end
<< ")" << dendl
;
3230 int r
= cmp(hoid
, q
->second
.begin
);
3231 if (r
== 0 || (r
> 0 && hoid
< q
->second
.end
)) {
3232 ldout(cct
, 10) << __func__
<< " backoff " << op
->target
.actual_pgid
3233 << " id " << q
->second
.id
<< " on " << hoid
3234 << ", queuing " << op
<< " tid " << op
->tid
<< dendl
;
3241 assert(op
->tid
> 0);
3242 m
= _prepare_osd_op(op
);
3245 if (op
->target
.actual_pgid
!= m
->get_spg()) {
3246 ldout(cct
, 10) << __func__
<< " " << op
->tid
<< " pgid change from "
3247 << m
->get_spg() << " to " << op
->target
.actual_pgid
3248 << ", updating and reencoding" << dendl
;
3249 m
->set_spg(op
->target
.actual_pgid
);
3250 m
->clear_payload(); // reencode
3253 ldout(cct
, 15) << "_send_op " << op
->tid
<< " to "
3254 << op
->target
.actual_pgid
<< " on osd." << op
->session
->osd
3257 ConnectionRef con
= op
->session
->con
;
3260 // preallocated rx buffer?
3262 ldout(cct
, 20) << " revoking rx buffer for " << op
->tid
<< " on "
3263 << op
->con
<< dendl
;
3264 op
->con
->revoke_rx_buffer(op
->tid
);
3267 op
->ontimeout
== 0 && // only post rx_buffer if no timeout; see #9582
3268 op
->outbl
->length()) {
3269 ldout(cct
, 20) << " posting rx buffer for " << op
->tid
<< " on " << con
3272 op
->con
->post_rx_buffer(op
->tid
, *op
->outbl
);
3275 op
->incarnation
= op
->session
->incarnation
;
3277 m
->set_tid(op
->tid
);
3279 if (op
->trace
.valid()) {
3280 m
->trace
.init("op msg", nullptr, &op
->trace
);
3282 op
->session
->con
->send_message(m
);
3285 int Objecter::calc_op_budget(Op
*op
)
3288 for (vector
<OSDOp
>::iterator i
= op
->ops
.begin();
3291 if (i
->op
.op
& CEPH_OSD_OP_MODE_WR
) {
3292 op_budget
+= i
->indata
.length();
3293 } else if (ceph_osd_op_mode_read(i
->op
.op
)) {
3294 if (ceph_osd_op_type_data(i
->op
.op
)) {
3295 if ((int64_t)i
->op
.extent
.length
> 0)
3296 op_budget
+= (int64_t)i
->op
.extent
.length
;
3297 } else if (ceph_osd_op_type_attr(i
->op
.op
)) {
3298 op_budget
+= i
->op
.xattr
.name_len
+ i
->op
.xattr
.value_len
;
3305 void Objecter::_throttle_op(Op
*op
,
3309 assert(sul
&& sul
.mutex() == &rwlock
);
3310 bool locked_for_write
= sul
.owns_lock();
3313 op_budget
= calc_op_budget(op
);
3314 if (!op_throttle_bytes
.get_or_fail(op_budget
)) { //couldn't take right now
3316 op_throttle_bytes
.get(op_budget
);
3317 if (locked_for_write
)
3322 if (!op_throttle_ops
.get_or_fail(1)) { //couldn't take right now
3324 op_throttle_ops
.get(1);
3325 if (locked_for_write
)
3332 void Objecter::unregister_op(Op
*op
)
3334 OSDSession::unique_lock
sl(op
->session
->lock
);
3335 op
->session
->ops
.erase(op
->tid
);
3337 put_session(op
->session
);
3343 /* This function DOES put the passed message before returning */
3344 void Objecter::handle_osd_op_reply(MOSDOpReply
*m
)
3346 ldout(cct
, 10) << "in handle_osd_op_reply" << dendl
;
3349 ceph_tid_t tid
= m
->get_tid();
3351 shunique_lock
sul(rwlock
, ceph::acquire_shared
);
3357 ConnectionRef con
= m
->get_connection();
3358 OSDSession
*s
= static_cast<OSDSession
*>(con
->get_priv());
3359 if (!s
|| s
->con
!= con
) {
3360 ldout(cct
, 7) << __func__
<< " no session on con " << con
<< dendl
;
3368 OSDSession::unique_lock
sl(s
->lock
);
3370 map
<ceph_tid_t
, Op
*>::iterator iter
= s
->ops
.find(tid
);
3371 if (iter
== s
->ops
.end()) {
3372 ldout(cct
, 7) << "handle_osd_op_reply " << tid
3373 << (m
->is_ondisk() ? " ondisk" : (m
->is_onnvram() ?
3374 " onnvram" : " ack"))
3375 << " ... stray" << dendl
;
3382 ldout(cct
, 7) << "handle_osd_op_reply " << tid
3383 << (m
->is_ondisk() ? " ondisk" :
3384 (m
->is_onnvram() ? " onnvram" : " ack"))
3385 << " uv " << m
->get_user_version()
3386 << " in " << m
->get_pg()
3387 << " attempt " << m
->get_retry_attempt()
3389 Op
*op
= iter
->second
;
3390 op
->trace
.event("osd op reply");
3392 if (retry_writes_after_first_reply
&& op
->attempts
== 1 &&
3393 (op
->target
.flags
& CEPH_OSD_FLAG_WRITE
)) {
3394 ldout(cct
, 7) << "retrying write after first reply: " << tid
<< dendl
;
3398 _session_op_remove(s
, op
);
3402 _op_submit(op
, sul
, NULL
);
3407 if (m
->get_retry_attempt() >= 0) {
3408 if (m
->get_retry_attempt() != (op
->attempts
- 1)) {
3409 ldout(cct
, 7) << " ignoring reply from attempt "
3410 << m
->get_retry_attempt()
3411 << " from " << m
->get_source_inst()
3412 << "; last attempt " << (op
->attempts
- 1) << " sent to "
3413 << op
->session
->con
->get_peer_addr() << dendl
;
3420 // we don't know the request attempt because the server is old, so
3421 // just accept this one. we may do ACK callbacks we shouldn't
3422 // have, but that is better than doing callbacks out of order.
3425 Context
*onfinish
= 0;
3427 int rc
= m
->get_result();
3429 if (m
->is_redirect_reply()) {
3430 ldout(cct
, 5) << " got redirect reply; redirecting" << dendl
;
3433 _session_op_remove(s
, op
);
3437 // FIXME: two redirects could race and reorder
3440 m
->get_redirect().combine_with_locator(op
->target
.target_oloc
,
3441 op
->target
.target_oid
.name
);
3442 op
->target
.flags
|= (CEPH_OSD_FLAG_REDIRECTED
| CEPH_OSD_FLAG_IGNORE_OVERLAY
);
3443 _op_submit(op
, sul
, NULL
);
3448 if (rc
== -EAGAIN
) {
3449 ldout(cct
, 7) << " got -EAGAIN, resubmitting" << dendl
;
3452 _session_op_remove(s
, op
);
3457 op
->target
.flags
&= ~(CEPH_OSD_FLAG_BALANCE_READS
|
3458 CEPH_OSD_FLAG_LOCALIZE_READS
);
3459 op
->target
.pgid
= pg_t();
3460 _op_submit(op
, sul
, NULL
);
3468 *op
->objver
= m
->get_user_version();
3469 if (op
->reply_epoch
)
3470 *op
->reply_epoch
= m
->get_map_epoch();
3471 if (op
->data_offset
)
3472 *op
->data_offset
= m
->get_header().data_off
;
3477 op
->con
->revoke_rx_buffer(op
->tid
);
3478 m
->claim_data(*op
->outbl
);
3482 // per-op result demuxing
3483 vector
<OSDOp
> out_ops
;
3484 m
->claim_ops(out_ops
);
3486 if (out_ops
.size() != op
->ops
.size())
3487 ldout(cct
, 0) << "WARNING: tid " << op
->tid
<< " reply ops " << out_ops
3488 << " != request ops " << op
->ops
3489 << " from " << m
->get_source_inst() << dendl
;
3491 vector
<bufferlist
*>::iterator pb
= op
->out_bl
.begin();
3492 vector
<int*>::iterator pr
= op
->out_rval
.begin();
3493 vector
<Context
*>::iterator ph
= op
->out_handler
.begin();
3494 assert(op
->out_bl
.size() == op
->out_rval
.size());
3495 assert(op
->out_bl
.size() == op
->out_handler
.size());
3496 vector
<OSDOp
>::iterator p
= out_ops
.begin();
3497 for (unsigned i
= 0;
3498 p
!= out_ops
.end() && pb
!= op
->out_bl
.end();
3499 ++i
, ++p
, ++pb
, ++pr
, ++ph
) {
3500 ldout(cct
, 10) << " op " << i
<< " rval " << p
->rval
3501 << " len " << p
->outdata
.length() << dendl
;
3504 // set rval before running handlers so that handlers
3505 // can change it if e.g. decoding fails
3507 **pr
= ceph_to_hostos_errno(p
->rval
);
3509 ldout(cct
, 10) << " op " << i
<< " handler " << *ph
<< dendl
;
3510 (*ph
)->complete(ceph_to_hostos_errno(p
->rval
));
3515 // NOTE: we assume that since we only request ONDISK ever we will
3516 // only ever get back one (type of) ack ever.
3520 onfinish
= op
->onfinish
;
3521 op
->onfinish
= NULL
;
3523 logger
->inc(l_osdc_op_reply
);
3525 /* get it before we call _finish_op() */
3526 auto completion_lock
= s
->get_lock(op
->target
.base_oid
);
3528 ldout(cct
, 15) << "handle_osd_op_reply completed tid " << tid
<< dendl
;
3531 ldout(cct
, 5) << num_in_flight
<< " in flight" << dendl
;
3533 // serialize completions
3534 if (completion_lock
.mutex()) {
3535 completion_lock
.lock();
3541 onfinish
->complete(rc
);
3543 if (completion_lock
.mutex()) {
3544 completion_lock
.unlock();
3551 void Objecter::handle_osd_backoff(MOSDBackoff
*m
)
3553 ldout(cct
, 10) << __func__
<< " " << *m
<< dendl
;
3554 shunique_lock
sul(rwlock
, ceph::acquire_shared
);
3560 ConnectionRef con
= m
->get_connection();
3561 OSDSession
*s
= static_cast<OSDSession
*>(con
->get_priv());
3562 if (!s
|| s
->con
!= con
) {
3563 ldout(cct
, 7) << __func__
<< " no session on con " << con
<< dendl
;
3571 s
->put(); // from get_priv() above
3573 OSDSession::unique_lock
sl(s
->lock
);
3576 case CEPH_OSD_BACKOFF_OP_BLOCK
:
3579 OSDBackoff
& b
= s
->backoffs
[m
->pgid
][m
->begin
];
3580 s
->backoffs_by_id
.insert(make_pair(m
->id
, &b
));
3586 // ack with original backoff's epoch so that the osd can discard this if
3587 // there was a pg split.
3588 Message
*r
= new MOSDBackoff(m
->pgid
,
3590 CEPH_OSD_BACKOFF_OP_ACK_BLOCK
,
3591 m
->id
, m
->begin
, m
->end
);
3592 // this priority must match the MOSDOps from _prepare_osd_op
3593 r
->set_priority(cct
->_conf
->osd_client_op_priority
);
3594 con
->send_message(r
);
3598 case CEPH_OSD_BACKOFF_OP_UNBLOCK
:
3600 auto p
= s
->backoffs_by_id
.find(m
->id
);
3601 if (p
!= s
->backoffs_by_id
.end()) {
3602 OSDBackoff
*b
= p
->second
;
3603 if (b
->begin
!= m
->begin
&&
3605 lderr(cct
) << __func__
<< " got " << m
->pgid
<< " id " << m
->id
3607 << m
->begin
<< "," << m
->end
<< ") but backoff is ["
3608 << b
->begin
<< "," << b
->end
<< ")" << dendl
;
3609 // hrmpf, unblock it anyway.
3611 ldout(cct
, 10) << __func__
<< " unblock backoff " << b
->pgid
3613 << " [" << b
->begin
<< "," << b
->end
3615 auto spgp
= s
->backoffs
.find(b
->pgid
);
3616 assert(spgp
!= s
->backoffs
.end());
3617 spgp
->second
.erase(b
->begin
);
3618 if (spgp
->second
.empty()) {
3619 s
->backoffs
.erase(spgp
);
3621 s
->backoffs_by_id
.erase(p
);
3623 // check for any ops to resend
3624 for (auto& q
: s
->ops
) {
3625 if (q
.second
->target
.actual_pgid
== m
->pgid
) {
3626 int r
= q
.second
->target
.contained_by(m
->begin
, m
->end
);
3627 ldout(cct
, 20) << __func__
<< " contained_by " << r
<< " on "
3628 << q
.second
->target
.get_hobj() << dendl
;
3635 lderr(cct
) << __func__
<< " " << m
->pgid
<< " id " << m
->id
3637 << m
->begin
<< "," << m
->end
<< ") but backoff dne" << dendl
;
3643 ldout(cct
, 10) << __func__
<< " unrecognized op " << (int)m
->op
<< dendl
;
3653 uint32_t Objecter::list_nobjects_seek(NListContext
*list_context
,
3656 shared_lock
rl(rwlock
);
3657 list_context
->pos
= hobject_t(object_t(), string(), CEPH_NOSNAP
,
3658 pos
, list_context
->pool_id
, string());
3659 ldout(cct
, 10) << __func__
<< list_context
3660 << " pos " << pos
<< " -> " << list_context
->pos
<< dendl
;
3661 pg_t actual
= osdmap
->raw_pg_to_pg(pg_t(pos
, list_context
->pool_id
));
3662 list_context
->current_pg
= actual
.ps();
3663 list_context
->at_end_of_pool
= false;
3667 uint32_t Objecter::list_nobjects_seek(NListContext
*list_context
,
3668 const hobject_t
& cursor
)
3670 shared_lock
rl(rwlock
);
3671 ldout(cct
, 10) << "list_nobjects_seek " << list_context
<< dendl
;
3672 list_context
->pos
= cursor
;
3673 list_context
->at_end_of_pool
= false;
3674 pg_t actual
= osdmap
->raw_pg_to_pg(pg_t(cursor
.get_hash(), list_context
->pool_id
));
3675 list_context
->current_pg
= actual
.ps();
3676 list_context
->sort_bitwise
= true;
3677 return list_context
->current_pg
;
3680 void Objecter::list_nobjects_get_cursor(NListContext
*list_context
,
3683 shared_lock
rl(rwlock
);
3684 if (list_context
->list
.empty()) {
3685 *cursor
= list_context
->pos
;
3687 const librados::ListObjectImpl
& entry
= list_context
->list
.front();
3688 const string
*key
= (entry
.locator
.empty() ? &entry
.oid
: &entry
.locator
);
3689 uint32_t h
= osdmap
->get_pg_pool(list_context
->pool_id
)->hash_key(*key
, entry
.nspace
);
3690 *cursor
= hobject_t(entry
.oid
, entry
.locator
, list_context
->pool_snap_seq
, h
, list_context
->pool_id
, entry
.nspace
);
3694 void Objecter::list_nobjects(NListContext
*list_context
, Context
*onfinish
)
3696 ldout(cct
, 10) << __func__
<< " pool_id " << list_context
->pool_id
3697 << " pool_snap_seq " << list_context
->pool_snap_seq
3698 << " max_entries " << list_context
->max_entries
3699 << " list_context " << list_context
3700 << " onfinish " << onfinish
3701 << " current_pg " << list_context
->current_pg
3702 << " pos " << list_context
->pos
<< dendl
;
3704 shared_lock
rl(rwlock
);
3705 const pg_pool_t
*pool
= osdmap
->get_pg_pool(list_context
->pool_id
);
3706 if (!pool
) { // pool is gone
3708 put_nlist_context_budget(list_context
);
3709 onfinish
->complete(-ENOENT
);
3712 int pg_num
= pool
->get_pg_num();
3713 bool sort_bitwise
= osdmap
->test_flag(CEPH_OSDMAP_SORTBITWISE
);
3715 if (list_context
->pos
.is_min()) {
3716 list_context
->starting_pg_num
= 0;
3717 list_context
->sort_bitwise
= sort_bitwise
;
3718 list_context
->starting_pg_num
= pg_num
;
3720 if (list_context
->sort_bitwise
!= sort_bitwise
) {
3721 list_context
->pos
= hobject_t(
3722 object_t(), string(), CEPH_NOSNAP
,
3723 list_context
->current_pg
, list_context
->pool_id
, string());
3724 list_context
->sort_bitwise
= sort_bitwise
;
3725 ldout(cct
, 10) << " hobject sort order changed, restarting this pg at "
3726 << list_context
->pos
<< dendl
;
3728 if (list_context
->starting_pg_num
!= pg_num
) {
3729 if (!sort_bitwise
) {
3730 // start reading from the beginning; the pgs have changed
3731 ldout(cct
, 10) << " pg_num changed; restarting with " << pg_num
<< dendl
;
3732 list_context
->pos
= collection_list_handle_t();
3734 list_context
->starting_pg_num
= pg_num
;
3737 if (list_context
->pos
.is_max()) {
3738 ldout(cct
, 20) << __func__
<< " end of pool, list "
3739 << list_context
->list
<< dendl
;
3740 if (list_context
->list
.empty()) {
3741 list_context
->at_end_of_pool
= true;
3743 // release the listing context's budget once all
3744 // OPs (in the session) are finished
3745 put_nlist_context_budget(list_context
);
3746 onfinish
->complete(0);
3751 op
.pg_nls(list_context
->max_entries
, list_context
->filter
,
3752 list_context
->pos
, osdmap
->get_epoch());
3753 list_context
->bl
.clear();
3754 C_NList
*onack
= new C_NList(list_context
, onfinish
, this);
3755 object_locator_t
oloc(list_context
->pool_id
, list_context
->nspace
);
3757 // note current_pg in case we don't have (or lose) SORTBITWISE
3758 list_context
->current_pg
= pool
->raw_hash_to_pg(list_context
->pos
.get_hash());
3761 pg_read(list_context
->current_pg
, oloc
, op
,
3762 &list_context
->bl
, 0, onack
, &onack
->epoch
,
3763 &list_context
->ctx_budget
);
3766 void Objecter::_nlist_reply(NListContext
*list_context
, int r
,
3767 Context
*final_finish
, epoch_t reply_epoch
)
3769 ldout(cct
, 10) << __func__
<< " " << list_context
<< dendl
;
3771 bufferlist::iterator iter
= list_context
->bl
.begin();
3772 pg_nls_response_t response
;
3773 bufferlist extra_info
;
3774 ::decode(response
, iter
);
3776 ::decode(extra_info
, iter
);
3779 // if the osd returns 1 (newer code), or handle MAX, it means we
3780 // hit the end of the pg.
3781 if ((response
.handle
.is_max() || r
== 1) &&
3782 !list_context
->sort_bitwise
) {
3783 // legacy OSD and !sortbitwise, figure out the next PG on our own
3784 ++list_context
->current_pg
;
3785 if (list_context
->current_pg
== list_context
->starting_pg_num
) {
3787 list_context
->pos
= hobject_t::get_max();
3790 list_context
->pos
= hobject_t(object_t(), string(), CEPH_NOSNAP
,
3791 list_context
->current_pg
,
3792 list_context
->pool_id
, string());
3795 list_context
->pos
= response
.handle
;
3798 int response_size
= response
.entries
.size();
3799 ldout(cct
, 20) << " response.entries.size " << response_size
3800 << ", response.entries " << response
.entries
3801 << ", handle " << response
.handle
3802 << ", tentative new pos " << list_context
->pos
<< dendl
;
3803 list_context
->extra_info
.append(extra_info
);
3804 if (response_size
) {
3805 list_context
->list
.splice(list_context
->list
.end(), response
.entries
);
3808 if (list_context
->list
.size() >= list_context
->max_entries
) {
3809 ldout(cct
, 20) << " hit max, returning results so far, "
3810 << list_context
->list
<< dendl
;
3811 // release the listing context's budget once all
3812 // OPs (in the session) are finished
3813 put_nlist_context_budget(list_context
);
3814 final_finish
->complete(0);
3819 list_nobjects(list_context
, final_finish
);
3822 void Objecter::put_nlist_context_budget(NListContext
*list_context
)
3824 if (list_context
->ctx_budget
>= 0) {
3825 ldout(cct
, 10) << " release listing context's budget " <<
3826 list_context
->ctx_budget
<< dendl
;
3827 put_op_budget_bytes(list_context
->ctx_budget
);
3828 list_context
->ctx_budget
= -1;
3834 int Objecter::create_pool_snap(int64_t pool
, string
& snap_name
,
3837 unique_lock
wl(rwlock
);
3838 ldout(cct
, 10) << "create_pool_snap; pool: " << pool
<< "; snap: "
3839 << snap_name
<< dendl
;
3841 const pg_pool_t
*p
= osdmap
->get_pg_pool(pool
);
3844 if (p
->snap_exists(snap_name
.c_str()))
3847 PoolOp
*op
= new PoolOp
;
3850 op
->tid
= ++last_tid
;
3852 op
->name
= snap_name
;
3853 op
->onfinish
= onfinish
;
3854 op
->pool_op
= POOL_OP_CREATE_SNAP
;
3855 pool_ops
[op
->tid
] = op
;
3862 struct C_SelfmanagedSnap
: public Context
{
3866 C_SelfmanagedSnap(snapid_t
*ps
, Context
*f
) : psnapid(ps
), fin(f
) {}
3867 void finish(int r
) override
{
3869 bufferlist::iterator p
= bl
.begin();
3870 ::decode(*psnapid
, p
);
3876 int Objecter::allocate_selfmanaged_snap(int64_t pool
, snapid_t
*psnapid
,
3879 unique_lock
wl(rwlock
);
3880 ldout(cct
, 10) << "allocate_selfmanaged_snap; pool: " << pool
<< dendl
;
3881 PoolOp
*op
= new PoolOp
;
3882 if (!op
) return -ENOMEM
;
3883 op
->tid
= ++last_tid
;
3885 C_SelfmanagedSnap
*fin
= new C_SelfmanagedSnap(psnapid
, onfinish
);
3888 op
->pool_op
= POOL_OP_CREATE_UNMANAGED_SNAP
;
3889 pool_ops
[op
->tid
] = op
;
3895 int Objecter::delete_pool_snap(int64_t pool
, string
& snap_name
,
3898 unique_lock
wl(rwlock
);
3899 ldout(cct
, 10) << "delete_pool_snap; pool: " << pool
<< "; snap: "
3900 << snap_name
<< dendl
;
3902 const pg_pool_t
*p
= osdmap
->get_pg_pool(pool
);
3905 if (!p
->snap_exists(snap_name
.c_str()))
3908 PoolOp
*op
= new PoolOp
;
3911 op
->tid
= ++last_tid
;
3913 op
->name
= snap_name
;
3914 op
->onfinish
= onfinish
;
3915 op
->pool_op
= POOL_OP_DELETE_SNAP
;
3916 pool_ops
[op
->tid
] = op
;
3923 int Objecter::delete_selfmanaged_snap(int64_t pool
, snapid_t snap
,
3926 unique_lock
wl(rwlock
);
3927 ldout(cct
, 10) << "delete_selfmanaged_snap; pool: " << pool
<< "; snap: "
3929 PoolOp
*op
= new PoolOp
;
3930 if (!op
) return -ENOMEM
;
3931 op
->tid
= ++last_tid
;
3933 op
->onfinish
= onfinish
;
3934 op
->pool_op
= POOL_OP_DELETE_UNMANAGED_SNAP
;
3936 pool_ops
[op
->tid
] = op
;
3943 int Objecter::create_pool(string
& name
, Context
*onfinish
, uint64_t auid
,
3946 unique_lock
wl(rwlock
);
3947 ldout(cct
, 10) << "create_pool name=" << name
<< dendl
;
3949 if (osdmap
->lookup_pg_pool_name(name
) >= 0)
3952 PoolOp
*op
= new PoolOp
;
3955 op
->tid
= ++last_tid
;
3958 op
->onfinish
= onfinish
;
3959 op
->pool_op
= POOL_OP_CREATE
;
3960 pool_ops
[op
->tid
] = op
;
3962 op
->crush_rule
= crush_rule
;
3969 int Objecter::delete_pool(int64_t pool
, Context
*onfinish
)
3971 unique_lock
wl(rwlock
);
3972 ldout(cct
, 10) << "delete_pool " << pool
<< dendl
;
3974 if (!osdmap
->have_pg_pool(pool
))
3977 _do_delete_pool(pool
, onfinish
);
3981 int Objecter::delete_pool(const string
&pool_name
, Context
*onfinish
)
3983 unique_lock
wl(rwlock
);
3984 ldout(cct
, 10) << "delete_pool " << pool_name
<< dendl
;
3986 int64_t pool
= osdmap
->lookup_pg_pool_name(pool_name
);
3990 _do_delete_pool(pool
, onfinish
);
3994 void Objecter::_do_delete_pool(int64_t pool
, Context
*onfinish
)
3996 PoolOp
*op
= new PoolOp
;
3997 op
->tid
= ++last_tid
;
3999 op
->name
= "delete";
4000 op
->onfinish
= onfinish
;
4001 op
->pool_op
= POOL_OP_DELETE
;
4002 pool_ops
[op
->tid
] = op
;
4007 * change the auid owner of a pool by contacting the monitor.
4008 * This requires the current connection to have write permissions
4009 * on both the pool's current auid and the new (parameter) auid.
4010 * Uses the standard Context callback when done.
4012 int Objecter::change_pool_auid(int64_t pool
, Context
*onfinish
, uint64_t auid
)
4014 unique_lock
wl(rwlock
);
4015 ldout(cct
, 10) << "change_pool_auid " << pool
<< " to " << auid
<< dendl
;
4016 PoolOp
*op
= new PoolOp
;
4017 if (!op
) return -ENOMEM
;
4018 op
->tid
= ++last_tid
;
4020 op
->name
= "change_pool_auid";
4021 op
->onfinish
= onfinish
;
4022 op
->pool_op
= POOL_OP_AUID_CHANGE
;
4024 pool_ops
[op
->tid
] = op
;
4026 logger
->set(l_osdc_poolop_active
, pool_ops
.size());
4032 void Objecter::pool_op_submit(PoolOp
*op
)
4035 if (mon_timeout
> timespan(0)) {
4036 op
->ontimeout
= timer
.add_event(mon_timeout
,
4038 pool_op_cancel(op
->tid
, -ETIMEDOUT
); });
4040 _pool_op_submit(op
);
4043 void Objecter::_pool_op_submit(PoolOp
*op
)
4045 // rwlock is locked unique
4047 ldout(cct
, 10) << "pool_op_submit " << op
->tid
<< dendl
;
4048 MPoolOp
*m
= new MPoolOp(monc
->get_fsid(), op
->tid
, op
->pool
,
4049 op
->name
, op
->pool_op
,
4050 op
->auid
, last_seen_osdmap_version
);
4051 if (op
->snapid
) m
->snapid
= op
->snapid
;
4052 if (op
->crush_rule
) m
->crush_rule
= op
->crush_rule
;
4053 monc
->send_mon_message(m
);
4054 op
->last_submit
= ceph::mono_clock::now();
4056 logger
->inc(l_osdc_poolop_send
);
4060 * Handle a reply to a PoolOp message. Check that we sent the message
4061 * and give the caller responsibility for the returned bufferlist.
4062 * Then either call the finisher or stash the PoolOp, depending on if we
4063 * have a new enough map.
4064 * Lastly, clean up the message and PoolOp.
4066 void Objecter::handle_pool_op_reply(MPoolOpReply
*m
)
4069 shunique_lock
sul(rwlock
, acquire_shared
);
4076 ldout(cct
, 10) << "handle_pool_op_reply " << *m
<< dendl
;
4077 ceph_tid_t tid
= m
->get_tid();
4078 map
<ceph_tid_t
, PoolOp
*>::iterator iter
= pool_ops
.find(tid
);
4079 if (iter
!= pool_ops
.end()) {
4080 PoolOp
*op
= iter
->second
;
4081 ldout(cct
, 10) << "have request " << tid
<< " at " << op
<< " Op: "
4082 << ceph_pool_op_name(op
->pool_op
) << dendl
;
4084 op
->blp
->claim(m
->response_data
);
4085 if (m
->version
> last_seen_osdmap_version
)
4086 last_seen_osdmap_version
= m
->version
;
4087 if (osdmap
->get_epoch() < m
->epoch
) {
4090 // recheck op existence since we have let go of rwlock
4091 // (for promotion) above.
4092 iter
= pool_ops
.find(tid
);
4093 if (iter
== pool_ops
.end())
4094 goto done
; // op is gone.
4095 if (osdmap
->get_epoch() < m
->epoch
) {
4096 ldout(cct
, 20) << "waiting for client to reach epoch " << m
->epoch
4097 << " before calling back" << dendl
;
4098 _wait_for_new_map(op
->onfinish
, m
->epoch
, m
->replyCode
);
4100 // map epoch changed, probably because a MOSDMap message
4101 // sneaked in. Do caller-specified callback now or else
4102 // we lose it forever.
4103 assert(op
->onfinish
);
4104 op
->onfinish
->complete(m
->replyCode
);
4107 assert(op
->onfinish
);
4108 op
->onfinish
->complete(m
->replyCode
);
4110 op
->onfinish
= NULL
;
4111 if (!sul
.owns_lock()) {
4115 iter
= pool_ops
.find(tid
);
4116 if (iter
!= pool_ops
.end()) {
4117 _finish_pool_op(op
, 0);
4120 ldout(cct
, 10) << "unknown request " << tid
<< dendl
;
4124 // Not strictly necessary, since we'll release it on return.
4127 ldout(cct
, 10) << "done" << dendl
;
4131 int Objecter::pool_op_cancel(ceph_tid_t tid
, int r
)
4133 assert(initialized
);
4135 unique_lock
wl(rwlock
);
4137 map
<ceph_tid_t
, PoolOp
*>::iterator it
= pool_ops
.find(tid
);
4138 if (it
== pool_ops
.end()) {
4139 ldout(cct
, 10) << __func__
<< " tid " << tid
<< " dne" << dendl
;
4143 ldout(cct
, 10) << __func__
<< " tid " << tid
<< dendl
;
4145 PoolOp
*op
= it
->second
;
4147 op
->onfinish
->complete(r
);
4149 _finish_pool_op(op
, r
);
4153 void Objecter::_finish_pool_op(PoolOp
*op
, int r
)
4155 // rwlock is locked unique
4156 pool_ops
.erase(op
->tid
);
4157 logger
->set(l_osdc_poolop_active
, pool_ops
.size());
4159 if (op
->ontimeout
&& r
!= -ETIMEDOUT
) {
4160 timer
.cancel_event(op
->ontimeout
);
4168 void Objecter::get_pool_stats(list
<string
>& pools
,
4169 map
<string
,pool_stat_t
> *result
,
4172 ldout(cct
, 10) << "get_pool_stats " << pools
<< dendl
;
4174 PoolStatOp
*op
= new PoolStatOp
;
4175 op
->tid
= ++last_tid
;
4177 op
->pool_stats
= result
;
4178 op
->onfinish
= onfinish
;
4179 if (mon_timeout
> timespan(0)) {
4180 op
->ontimeout
= timer
.add_event(mon_timeout
,
4182 pool_stat_op_cancel(op
->tid
,
4188 unique_lock
wl(rwlock
);
4190 poolstat_ops
[op
->tid
] = op
;
4192 logger
->set(l_osdc_poolstat_active
, poolstat_ops
.size());
4194 _poolstat_submit(op
);
4197 void Objecter::_poolstat_submit(PoolStatOp
*op
)
4199 ldout(cct
, 10) << "_poolstat_submit " << op
->tid
<< dendl
;
4200 monc
->send_mon_message(new MGetPoolStats(monc
->get_fsid(), op
->tid
,
4202 last_seen_pgmap_version
));
4203 op
->last_submit
= ceph::mono_clock::now();
4205 logger
->inc(l_osdc_poolstat_send
);
4208 void Objecter::handle_get_pool_stats_reply(MGetPoolStatsReply
*m
)
4210 ldout(cct
, 10) << "handle_get_pool_stats_reply " << *m
<< dendl
;
4211 ceph_tid_t tid
= m
->get_tid();
4213 unique_lock
wl(rwlock
);
4219 map
<ceph_tid_t
, PoolStatOp
*>::iterator iter
= poolstat_ops
.find(tid
);
4220 if (iter
!= poolstat_ops
.end()) {
4221 PoolStatOp
*op
= poolstat_ops
[tid
];
4222 ldout(cct
, 10) << "have request " << tid
<< " at " << op
<< dendl
;
4223 *op
->pool_stats
= m
->pool_stats
;
4224 if (m
->version
> last_seen_pgmap_version
) {
4225 last_seen_pgmap_version
= m
->version
;
4227 op
->onfinish
->complete(0);
4228 _finish_pool_stat_op(op
, 0);
4230 ldout(cct
, 10) << "unknown request " << tid
<< dendl
;
4232 ldout(cct
, 10) << "done" << dendl
;
4236 int Objecter::pool_stat_op_cancel(ceph_tid_t tid
, int r
)
4238 assert(initialized
);
4240 unique_lock
wl(rwlock
);
4242 map
<ceph_tid_t
, PoolStatOp
*>::iterator it
= poolstat_ops
.find(tid
);
4243 if (it
== poolstat_ops
.end()) {
4244 ldout(cct
, 10) << __func__
<< " tid " << tid
<< " dne" << dendl
;
4248 ldout(cct
, 10) << __func__
<< " tid " << tid
<< dendl
;
4250 PoolStatOp
*op
= it
->second
;
4252 op
->onfinish
->complete(r
);
4253 _finish_pool_stat_op(op
, r
);
4257 void Objecter::_finish_pool_stat_op(PoolStatOp
*op
, int r
)
4259 // rwlock is locked unique
4261 poolstat_ops
.erase(op
->tid
);
4262 logger
->set(l_osdc_poolstat_active
, poolstat_ops
.size());
4264 if (op
->ontimeout
&& r
!= -ETIMEDOUT
)
4265 timer
.cancel_event(op
->ontimeout
);
4270 void Objecter::get_fs_stats(ceph_statfs
& result
,
4271 boost::optional
<int64_t> data_pool
,
4274 ldout(cct
, 10) << "get_fs_stats" << dendl
;
4275 unique_lock
l(rwlock
);
4277 StatfsOp
*op
= new StatfsOp
;
4278 op
->tid
= ++last_tid
;
4279 op
->stats
= &result
;
4280 op
->data_pool
= data_pool
;
4281 op
->onfinish
= onfinish
;
4282 if (mon_timeout
> timespan(0)) {
4283 op
->ontimeout
= timer
.add_event(mon_timeout
,
4285 statfs_op_cancel(op
->tid
,
4290 statfs_ops
[op
->tid
] = op
;
4292 logger
->set(l_osdc_statfs_active
, statfs_ops
.size());
4294 _fs_stats_submit(op
);
4297 void Objecter::_fs_stats_submit(StatfsOp
*op
)
4299 // rwlock is locked unique
4301 ldout(cct
, 10) << "fs_stats_submit" << op
->tid
<< dendl
;
4302 monc
->send_mon_message(new MStatfs(monc
->get_fsid(), op
->tid
,
4304 last_seen_pgmap_version
));
4305 op
->last_submit
= ceph::mono_clock::now();
4307 logger
->inc(l_osdc_statfs_send
);
4310 void Objecter::handle_fs_stats_reply(MStatfsReply
*m
)
4312 unique_lock
wl(rwlock
);
4318 ldout(cct
, 10) << "handle_fs_stats_reply " << *m
<< dendl
;
4319 ceph_tid_t tid
= m
->get_tid();
4321 if (statfs_ops
.count(tid
)) {
4322 StatfsOp
*op
= statfs_ops
[tid
];
4323 ldout(cct
, 10) << "have request " << tid
<< " at " << op
<< dendl
;
4324 *(op
->stats
) = m
->h
.st
;
4325 if (m
->h
.version
> last_seen_pgmap_version
)
4326 last_seen_pgmap_version
= m
->h
.version
;
4327 op
->onfinish
->complete(0);
4328 _finish_statfs_op(op
, 0);
4330 ldout(cct
, 10) << "unknown request " << tid
<< dendl
;
4333 ldout(cct
, 10) << "done" << dendl
;
4336 int Objecter::statfs_op_cancel(ceph_tid_t tid
, int r
)
4338 assert(initialized
);
4340 unique_lock
wl(rwlock
);
4342 map
<ceph_tid_t
, StatfsOp
*>::iterator it
= statfs_ops
.find(tid
);
4343 if (it
== statfs_ops
.end()) {
4344 ldout(cct
, 10) << __func__
<< " tid " << tid
<< " dne" << dendl
;
4348 ldout(cct
, 10) << __func__
<< " tid " << tid
<< dendl
;
4350 StatfsOp
*op
= it
->second
;
4352 op
->onfinish
->complete(r
);
4353 _finish_statfs_op(op
, r
);
4357 void Objecter::_finish_statfs_op(StatfsOp
*op
, int r
)
4359 // rwlock is locked unique
4361 statfs_ops
.erase(op
->tid
);
4362 logger
->set(l_osdc_statfs_active
, statfs_ops
.size());
4364 if (op
->ontimeout
&& r
!= -ETIMEDOUT
)
4365 timer
.cancel_event(op
->ontimeout
);
4372 void Objecter::_sg_read_finish(vector
<ObjectExtent
>& extents
,
4373 vector
<bufferlist
>& resultbl
,
4374 bufferlist
*bl
, Context
*onfinish
)
4377 ldout(cct
, 15) << "_sg_read_finish" << dendl
;
4379 if (extents
.size() > 1) {
4380 Striper::StripedReadResult r
;
4381 vector
<bufferlist
>::iterator bit
= resultbl
.begin();
4382 for (vector
<ObjectExtent
>::iterator eit
= extents
.begin();
4383 eit
!= extents
.end();
4385 r
.add_partial_result(cct
, *bit
, eit
->buffer_extents
);
4388 r
.assemble_result(cct
, *bl
, false);
4390 ldout(cct
, 15) << " only one frag" << dendl
;
4391 bl
->claim(resultbl
[0]);
4395 uint64_t bytes_read
= bl
->length();
4396 ldout(cct
, 7) << "_sg_read_finish " << bytes_read
<< " bytes" << dendl
;
4399 onfinish
->complete(bytes_read
);// > 0 ? bytes_read:m->get_result());
4404 void Objecter::ms_handle_connect(Connection
*con
)
4406 ldout(cct
, 10) << "ms_handle_connect " << con
<< dendl
;
4410 if (con
->get_peer_type() == CEPH_ENTITY_TYPE_MON
)
4414 bool Objecter::ms_handle_reset(Connection
*con
)
4418 if (con
->get_peer_type() == CEPH_ENTITY_TYPE_OSD
) {
4419 unique_lock
wl(rwlock
);
4421 OSDSession
*session
= static_cast<OSDSession
*>(con
->get_priv());
4423 ldout(cct
, 1) << "ms_handle_reset " << con
<< " session " << session
4424 << " osd." << session
->osd
<< dendl
;
4429 map
<uint64_t, LingerOp
*> lresend
;
4430 OSDSession::unique_lock
sl(session
->lock
);
4431 _reopen_session(session
);
4432 _kick_requests(session
, lresend
);
4434 _linger_ops_resend(lresend
, wl
);
4436 maybe_request_map();
4444 void Objecter::ms_handle_remote_reset(Connection
*con
)
4447 * treat these the same.
4449 ms_handle_reset(con
);
4452 bool Objecter::ms_handle_refused(Connection
*con
)
4455 if (osdmap
&& (con
->get_peer_type() == CEPH_ENTITY_TYPE_OSD
)) {
4456 int osd
= osdmap
->identify_osd(con
->get_peer_addr());
4458 ldout(cct
, 1) << "ms_handle_refused on osd." << osd
<< dendl
;
4464 bool Objecter::ms_get_authorizer(int dest_type
,
4465 AuthAuthorizer
**authorizer
,
4470 if (dest_type
== CEPH_ENTITY_TYPE_MON
)
4472 *authorizer
= monc
->build_authorizer(dest_type
);
4473 return *authorizer
!= NULL
;
4476 void Objecter::op_target_t::dump(Formatter
*f
) const
4478 f
->dump_stream("pg") << pgid
;
4479 f
->dump_int("osd", osd
);
4480 f
->dump_stream("object_id") << base_oid
;
4481 f
->dump_stream("object_locator") << base_oloc
;
4482 f
->dump_stream("target_object_id") << target_oid
;
4483 f
->dump_stream("target_object_locator") << target_oloc
;
4484 f
->dump_int("paused", (int)paused
);
4485 f
->dump_int("used_replica", (int)used_replica
);
4486 f
->dump_int("precalc_pgid", (int)precalc_pgid
);
4489 void Objecter::_dump_active(OSDSession
*s
)
4491 for (map
<ceph_tid_t
,Op
*>::iterator p
= s
->ops
.begin();
4495 ldout(cct
, 20) << op
->tid
<< "\t" << op
->target
.pgid
4496 << "\tosd." << (op
->session
? op
->session
->osd
: -1)
4497 << "\t" << op
->target
.base_oid
4498 << "\t" << op
->ops
<< dendl
;
4502 void Objecter::_dump_active()
4504 ldout(cct
, 20) << "dump_active .. " << num_homeless_ops
<< " homeless"
4506 for (map
<int, OSDSession
*>::iterator siter
= osd_sessions
.begin();
4507 siter
!= osd_sessions
.end(); ++siter
) {
4508 OSDSession
*s
= siter
->second
;
4509 OSDSession::shared_lock
sl(s
->lock
);
4513 _dump_active(homeless_session
);
4516 void Objecter::dump_active()
4518 shared_lock
rl(rwlock
);
4523 void Objecter::dump_requests(Formatter
*fmt
)
4525 // Read-lock on Objecter held here
4526 fmt
->open_object_section("requests");
4528 dump_linger_ops(fmt
);
4530 dump_pool_stat_ops(fmt
);
4531 dump_statfs_ops(fmt
);
4532 dump_command_ops(fmt
);
4533 fmt
->close_section(); // requests object
4536 void Objecter::_dump_ops(const OSDSession
*s
, Formatter
*fmt
)
4538 for (map
<ceph_tid_t
,Op
*>::const_iterator p
= s
->ops
.begin();
4542 fmt
->open_object_section("op");
4543 fmt
->dump_unsigned("tid", op
->tid
);
4544 op
->target
.dump(fmt
);
4545 fmt
->dump_stream("last_sent") << op
->stamp
;
4546 fmt
->dump_int("attempts", op
->attempts
);
4547 fmt
->dump_stream("snapid") << op
->snapid
;
4548 fmt
->dump_stream("snap_context") << op
->snapc
;
4549 fmt
->dump_stream("mtime") << op
->mtime
;
4551 fmt
->open_array_section("osd_ops");
4552 for (vector
<OSDOp
>::const_iterator it
= op
->ops
.begin();
4553 it
!= op
->ops
.end();
4555 fmt
->dump_stream("osd_op") << *it
;
4557 fmt
->close_section(); // osd_ops array
4559 fmt
->close_section(); // op object
4563 void Objecter::dump_ops(Formatter
*fmt
)
4565 // Read-lock on Objecter held
4566 fmt
->open_array_section("ops");
4567 for (map
<int, OSDSession
*>::const_iterator siter
= osd_sessions
.begin();
4568 siter
!= osd_sessions
.end(); ++siter
) {
4569 OSDSession
*s
= siter
->second
;
4570 OSDSession::shared_lock
sl(s
->lock
);
4574 _dump_ops(homeless_session
, fmt
);
4575 fmt
->close_section(); // ops array
4578 void Objecter::_dump_linger_ops(const OSDSession
*s
, Formatter
*fmt
)
4580 for (map
<uint64_t, LingerOp
*>::const_iterator p
= s
->linger_ops
.begin();
4581 p
!= s
->linger_ops
.end();
4583 LingerOp
*op
= p
->second
;
4584 fmt
->open_object_section("linger_op");
4585 fmt
->dump_unsigned("linger_id", op
->linger_id
);
4586 op
->target
.dump(fmt
);
4587 fmt
->dump_stream("snapid") << op
->snap
;
4588 fmt
->dump_stream("registered") << op
->registered
;
4589 fmt
->close_section(); // linger_op object
4593 void Objecter::dump_linger_ops(Formatter
*fmt
)
4595 // We have a read-lock on the objecter
4596 fmt
->open_array_section("linger_ops");
4597 for (map
<int, OSDSession
*>::const_iterator siter
= osd_sessions
.begin();
4598 siter
!= osd_sessions
.end(); ++siter
) {
4599 OSDSession
*s
= siter
->second
;
4600 OSDSession::shared_lock
sl(s
->lock
);
4601 _dump_linger_ops(s
, fmt
);
4604 _dump_linger_ops(homeless_session
, fmt
);
4605 fmt
->close_section(); // linger_ops array
4608 void Objecter::_dump_command_ops(const OSDSession
*s
, Formatter
*fmt
)
4610 for (map
<uint64_t, CommandOp
*>::const_iterator p
= s
->command_ops
.begin();
4611 p
!= s
->command_ops
.end();
4613 CommandOp
*op
= p
->second
;
4614 fmt
->open_object_section("command_op");
4615 fmt
->dump_unsigned("command_id", op
->tid
);
4616 fmt
->dump_int("osd", op
->session
? op
->session
->osd
: -1);
4617 fmt
->open_array_section("command");
4618 for (vector
<string
>::const_iterator q
= op
->cmd
.begin();
4619 q
!= op
->cmd
.end(); ++q
)
4620 fmt
->dump_string("word", *q
);
4621 fmt
->close_section();
4622 if (op
->target_osd
>= 0)
4623 fmt
->dump_int("target_osd", op
->target_osd
);
4625 fmt
->dump_stream("target_pg") << op
->target_pg
;
4626 fmt
->close_section(); // command_op object
4630 void Objecter::dump_command_ops(Formatter
*fmt
)
4632 // We have a read-lock on the Objecter here
4633 fmt
->open_array_section("command_ops");
4634 for (map
<int, OSDSession
*>::const_iterator siter
= osd_sessions
.begin();
4635 siter
!= osd_sessions
.end(); ++siter
) {
4636 OSDSession
*s
= siter
->second
;
4637 OSDSession::shared_lock
sl(s
->lock
);
4638 _dump_command_ops(s
, fmt
);
4641 _dump_command_ops(homeless_session
, fmt
);
4642 fmt
->close_section(); // command_ops array
4645 void Objecter::dump_pool_ops(Formatter
*fmt
) const
4647 fmt
->open_array_section("pool_ops");
4648 for (map
<ceph_tid_t
, PoolOp
*>::const_iterator p
= pool_ops
.begin();
4649 p
!= pool_ops
.end();
4651 PoolOp
*op
= p
->second
;
4652 fmt
->open_object_section("pool_op");
4653 fmt
->dump_unsigned("tid", op
->tid
);
4654 fmt
->dump_int("pool", op
->pool
);
4655 fmt
->dump_string("name", op
->name
);
4656 fmt
->dump_int("operation_type", op
->pool_op
);
4657 fmt
->dump_unsigned("auid", op
->auid
);
4658 fmt
->dump_unsigned("crush_rule", op
->crush_rule
);
4659 fmt
->dump_stream("snapid") << op
->snapid
;
4660 fmt
->dump_stream("last_sent") << op
->last_submit
;
4661 fmt
->close_section(); // pool_op object
4663 fmt
->close_section(); // pool_ops array
4666 void Objecter::dump_pool_stat_ops(Formatter
*fmt
) const
4668 fmt
->open_array_section("pool_stat_ops");
4669 for (map
<ceph_tid_t
, PoolStatOp
*>::const_iterator p
= poolstat_ops
.begin();
4670 p
!= poolstat_ops
.end();
4672 PoolStatOp
*op
= p
->second
;
4673 fmt
->open_object_section("pool_stat_op");
4674 fmt
->dump_unsigned("tid", op
->tid
);
4675 fmt
->dump_stream("last_sent") << op
->last_submit
;
4677 fmt
->open_array_section("pools");
4678 for (list
<string
>::const_iterator it
= op
->pools
.begin();
4679 it
!= op
->pools
.end();
4681 fmt
->dump_string("pool", *it
);
4683 fmt
->close_section(); // pools array
4685 fmt
->close_section(); // pool_stat_op object
4687 fmt
->close_section(); // pool_stat_ops array
4690 void Objecter::dump_statfs_ops(Formatter
*fmt
) const
4692 fmt
->open_array_section("statfs_ops");
4693 for (map
<ceph_tid_t
, StatfsOp
*>::const_iterator p
= statfs_ops
.begin();
4694 p
!= statfs_ops
.end();
4696 StatfsOp
*op
= p
->second
;
4697 fmt
->open_object_section("statfs_op");
4698 fmt
->dump_unsigned("tid", op
->tid
);
4699 fmt
->dump_stream("last_sent") << op
->last_submit
;
4700 fmt
->close_section(); // statfs_op object
4702 fmt
->close_section(); // statfs_ops array
4705 Objecter::RequestStateHook::RequestStateHook(Objecter
*objecter
) :
4706 m_objecter(objecter
)
4710 bool Objecter::RequestStateHook::call(std::string command
, cmdmap_t
& cmdmap
,
4711 std::string format
, bufferlist
& out
)
4713 Formatter
*f
= Formatter::create(format
, "json-pretty", "json-pretty");
4714 shared_lock
rl(m_objecter
->rwlock
);
4715 m_objecter
->dump_requests(f
);
4721 void Objecter::blacklist_self(bool set
)
4723 ldout(cct
, 10) << "blacklist_self " << (set
? "add" : "rm") << dendl
;
4726 cmd
.push_back("{\"prefix\":\"osd blacklist\", ");
4728 cmd
.push_back("\"blacklistop\":\"add\",");
4730 cmd
.push_back("\"blacklistop\":\"rm\",");
4732 ss
<< messenger
->get_myaddr();
4733 cmd
.push_back("\"addr\":\"" + ss
.str() + "\"");
4735 MMonCommand
*m
= new MMonCommand(monc
->get_fsid());
4738 monc
->send_mon_message(m
);
4743 void Objecter::handle_command_reply(MCommandReply
*m
)
4745 unique_lock
wl(rwlock
);
4751 ConnectionRef con
= m
->get_connection();
4752 OSDSession
*s
= static_cast<OSDSession
*>(con
->get_priv());
4753 if (!s
|| s
->con
!= con
) {
4754 ldout(cct
, 7) << __func__
<< " no session on con " << con
<< dendl
;
4761 OSDSession::shared_lock
sl(s
->lock
);
4762 map
<ceph_tid_t
,CommandOp
*>::iterator p
= s
->command_ops
.find(m
->get_tid());
4763 if (p
== s
->command_ops
.end()) {
4764 ldout(cct
, 10) << "handle_command_reply tid " << m
->get_tid()
4765 << " not found" << dendl
;
4773 CommandOp
*c
= p
->second
;
4775 m
->get_connection() != c
->session
->con
) {
4776 ldout(cct
, 10) << "handle_command_reply tid " << m
->get_tid()
4777 << " got reply from wrong connection "
4778 << m
->get_connection() << " " << m
->get_source_inst()
4787 c
->poutbl
->claim(m
->get_data());
4792 OSDSession::unique_lock
sul(s
->lock
);
4793 _finish_command(c
, m
->r
, m
->rs
);
4801 void Objecter::submit_command(CommandOp
*c
, ceph_tid_t
*ptid
)
4803 shunique_lock
sul(rwlock
, ceph::acquire_unique
);
4805 ceph_tid_t tid
= ++last_tid
;
4806 ldout(cct
, 10) << "_submit_command " << tid
<< " " << c
->cmd
<< dendl
;
4810 OSDSession::unique_lock
hs_wl(homeless_session
->lock
);
4811 _session_command_op_assign(homeless_session
, c
);
4814 _calc_command_target(c
, sul
);
4815 _assign_command_session(c
, sul
);
4816 if (osd_timeout
> timespan(0)) {
4817 c
->ontimeout
= timer
.add_event(osd_timeout
,
4819 command_op_cancel(c
->session
, tid
,
4823 if (!c
->session
->is_homeless()) {
4826 _maybe_request_map();
4828 if (c
->map_check_error
)
4829 _send_command_map_check(c
);
4832 logger
->inc(l_osdc_command_active
);
4835 int Objecter::_calc_command_target(CommandOp
*c
, shunique_lock
& sul
)
4837 assert(sul
.owns_lock() && sul
.mutex() == &rwlock
);
4839 c
->map_check_error
= 0;
4841 // ignore overlays, just like we do with pg ops
4842 c
->target
.flags
|= CEPH_OSD_FLAG_IGNORE_OVERLAY
;
4844 if (c
->target_osd
>= 0) {
4845 if (!osdmap
->exists(c
->target_osd
)) {
4846 c
->map_check_error
= -ENOENT
;
4847 c
->map_check_error_str
= "osd dne";
4849 return RECALC_OP_TARGET_OSD_DNE
;
4851 if (osdmap
->is_down(c
->target_osd
)) {
4852 c
->map_check_error
= -ENXIO
;
4853 c
->map_check_error_str
= "osd down";
4855 return RECALC_OP_TARGET_OSD_DOWN
;
4857 c
->target
.osd
= c
->target_osd
;
4859 int ret
= _calc_target(&(c
->target
), nullptr, true);
4860 if (ret
== RECALC_OP_TARGET_POOL_DNE
) {
4861 c
->map_check_error
= -ENOENT
;
4862 c
->map_check_error_str
= "pool dne";
4865 } else if (ret
== RECALC_OP_TARGET_OSD_DOWN
) {
4866 c
->map_check_error
= -ENXIO
;
4867 c
->map_check_error_str
= "osd down";
4874 int r
= _get_session(c
->target
.osd
, &s
, sul
);
4875 assert(r
!= -EAGAIN
); /* shouldn't happen as we're holding the write lock */
4877 if (c
->session
!= s
) {
4879 return RECALC_OP_TARGET_NEED_RESEND
;
4884 ldout(cct
, 20) << "_recalc_command_target " << c
->tid
<< " no change, "
4885 << c
->session
<< dendl
;
4887 return RECALC_OP_TARGET_NO_ACTION
;
4890 void Objecter::_assign_command_session(CommandOp
*c
,
4893 assert(sul
.owns_lock() && sul
.mutex() == &rwlock
);
4896 int r
= _get_session(c
->target
.osd
, &s
, sul
);
4897 assert(r
!= -EAGAIN
); /* shouldn't happen as we're holding the write lock */
4899 if (c
->session
!= s
) {
4901 OSDSession
*cs
= c
->session
;
4902 OSDSession::unique_lock
csl(cs
->lock
);
4903 _session_command_op_remove(c
->session
, c
);
4906 OSDSession::unique_lock
sl(s
->lock
);
4907 _session_command_op_assign(s
, c
);
4913 void Objecter::_send_command(CommandOp
*c
)
4915 ldout(cct
, 10) << "_send_command " << c
->tid
<< dendl
;
4917 assert(c
->session
->con
);
4918 MCommand
*m
= new MCommand(monc
->monmap
.fsid
);
4920 m
->set_data(c
->inbl
);
4922 c
->session
->con
->send_message(m
);
4923 logger
->inc(l_osdc_command_send
);
4926 int Objecter::command_op_cancel(OSDSession
*s
, ceph_tid_t tid
, int r
)
4928 assert(initialized
);
4930 unique_lock
wl(rwlock
);
4932 map
<ceph_tid_t
, CommandOp
*>::iterator it
= s
->command_ops
.find(tid
);
4933 if (it
== s
->command_ops
.end()) {
4934 ldout(cct
, 10) << __func__
<< " tid " << tid
<< " dne" << dendl
;
4938 ldout(cct
, 10) << __func__
<< " tid " << tid
<< dendl
;
4940 CommandOp
*op
= it
->second
;
4941 _command_cancel_map_check(op
);
4942 OSDSession::unique_lock
sl(op
->session
->lock
);
4943 _finish_command(op
, r
, "");
4948 void Objecter::_finish_command(CommandOp
*c
, int r
, string rs
)
4950 // rwlock is locked unique
4951 // session lock is locked
4953 ldout(cct
, 10) << "_finish_command " << c
->tid
<< " = " << r
<< " "
4958 c
->onfinish
->complete(r
);
4960 if (c
->ontimeout
&& r
!= -ETIMEDOUT
)
4961 timer
.cancel_event(c
->ontimeout
);
4963 OSDSession
*s
= c
->session
;
4964 _session_command_op_remove(c
->session
, c
);
4968 logger
->dec(l_osdc_command_active
);
4971 Objecter::OSDSession::~OSDSession()
4973 // Caller is responsible for re-assigning or
4974 // destroying any ops that were assigned to us
4975 assert(ops
.empty());
4976 assert(linger_ops
.empty());
4977 assert(command_ops
.empty());
4980 Objecter::~Objecter()
4984 assert(homeless_session
->get_nref() == 1);
4985 assert(num_homeless_ops
== 0);
4986 homeless_session
->put();
4988 assert(osd_sessions
.empty());
4989 assert(poolstat_ops
.empty());
4990 assert(statfs_ops
.empty());
4991 assert(pool_ops
.empty());
4992 assert(waiting_for_map
.empty());
4993 assert(linger_ops
.empty());
4994 assert(check_latest_map_lingers
.empty());
4995 assert(check_latest_map_ops
.empty());
4996 assert(check_latest_map_commands
.empty());
4998 assert(!m_request_state_hook
);
5003 * Wait until this OSD map epoch is received before
5004 * sending any more operations to OSDs. Use this
5005 * when it is known that the client can't trust
5006 * anything from before this epoch (e.g. due to
5007 * client blacklist at this epoch).
5009 void Objecter::set_epoch_barrier(epoch_t epoch
)
5011 unique_lock
wl(rwlock
);
5013 ldout(cct
, 7) << __func__
<< ": barrier " << epoch
<< " (was "
5014 << epoch_barrier
<< ") current epoch " << osdmap
->get_epoch()
5016 if (epoch
> epoch_barrier
) {
5017 epoch_barrier
= epoch
;
5018 _maybe_request_map();
5024 hobject_t
Objecter::enumerate_objects_begin()
5029 hobject_t
Objecter::enumerate_objects_end()
5031 return hobject_t::get_max();
5034 struct C_EnumerateReply
: public Context
{
5039 std::list
<librados::ListObjectImpl
> *result
;
5040 const hobject_t end
;
5041 const int64_t pool_id
;
5047 C_EnumerateReply(Objecter
*objecter_
, hobject_t
*next_
,
5048 std::list
<librados::ListObjectImpl
> *result_
,
5049 const hobject_t end_
, const int64_t pool_id_
, Context
*on_finish_
) :
5050 objecter(objecter_
), next(next_
), result(result_
),
5051 end(end_
), pool_id(pool_id_
), on_finish(on_finish_
),
5055 void finish(int r
) override
{
5056 objecter
->_enumerate_reply(
5057 bl
, r
, end
, pool_id
, budget
, epoch
, result
, next
, on_finish
);
5061 void Objecter::enumerate_objects(
5063 const std::string
&ns
,
5064 const hobject_t
&start
,
5065 const hobject_t
&end
,
5067 const bufferlist
&filter_bl
,
5068 std::list
<librados::ListObjectImpl
> *result
,
5074 if (!end
.is_max() && start
> end
) {
5075 lderr(cct
) << __func__
<< ": start " << start
<< " > end " << end
<< dendl
;
5076 on_finish
->complete(-EINVAL
);
5081 lderr(cct
) << __func__
<< ": result size may not be zero" << dendl
;
5082 on_finish
->complete(-EINVAL
);
5086 if (start
.is_max()) {
5087 on_finish
->complete(0);
5091 shared_lock
rl(rwlock
);
5092 assert(osdmap
->get_epoch());
5093 if (!osdmap
->test_flag(CEPH_OSDMAP_SORTBITWISE
)) {
5095 lderr(cct
) << __func__
<< ": SORTBITWISE cluster flag not set" << dendl
;
5096 on_finish
->complete(-EOPNOTSUPP
);
5099 const pg_pool_t
*p
= osdmap
->get_pg_pool(pool_id
);
5101 lderr(cct
) << __func__
<< ": pool " << pool_id
<< " DNE in osd epoch "
5102 << osdmap
->get_epoch() << dendl
;
5104 on_finish
->complete(-ENOENT
);
5110 ldout(cct
, 20) << __func__
<< ": start=" << start
<< " end=" << end
<< dendl
;
5112 // Stash completion state
5113 C_EnumerateReply
*on_ack
= new C_EnumerateReply(
5114 this, next
, result
, end
, pool_id
, on_finish
);
5117 op
.pg_nls(max
, filter_bl
, start
, 0);
5119 // Issue. See you later in _enumerate_reply
5120 object_locator_t
oloc(pool_id
, ns
);
5121 pg_read(start
.get_hash(), oloc
, op
,
5122 &on_ack
->bl
, 0, on_ack
, &on_ack
->epoch
, &on_ack
->budget
);
5125 void Objecter::_enumerate_reply(
5128 const hobject_t
&end
,
5129 const int64_t pool_id
,
5131 epoch_t reply_epoch
,
5132 std::list
<librados::ListObjectImpl
> *result
,
5137 put_op_budget_bytes(budget
);
5141 ldout(cct
, 4) << __func__
<< ": remote error " << r
<< dendl
;
5142 on_finish
->complete(r
);
5146 assert(next
!= NULL
);
5148 // Decode the results
5149 bufferlist::iterator iter
= bl
.begin();
5150 pg_nls_response_t response
;
5152 // XXX extra_info doesn't seem used anywhere?
5153 bufferlist extra_info
;
5154 ::decode(response
, iter
);
5156 ::decode(extra_info
, iter
);
5159 ldout(cct
, 10) << __func__
<< ": got " << response
.entries
.size()
5160 << " handle " << response
.handle
5161 << " reply_epoch " << reply_epoch
<< dendl
;
5162 ldout(cct
, 20) << __func__
<< ": response.entries.size "
5163 << response
.entries
.size() << ", response.entries "
5164 << response
.entries
<< dendl
;
5165 if (response
.handle
<= end
) {
5166 *next
= response
.handle
;
5168 ldout(cct
, 10) << __func__
<< ": adjusted next down to end " << end
5172 // drop anything after 'end'
5173 shared_lock
rl(rwlock
);
5174 const pg_pool_t
*pool
= osdmap
->get_pg_pool(pool_id
);
5176 // pool is gone, drop any results which are now meaningless.
5178 on_finish
->complete(-ENOENT
);
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
,
5192 response
.entries
.back().nspace
);
5195 ldout(cct
, 20) << __func__
<< " dropping item " << last
5196 << " >= end " << end
<< dendl
;
5197 response
.entries
.pop_back();
5201 if (!response
.entries
.empty()) {
5202 result
->merge(response
.entries
);
5205 // release the listing context's budget once all
5206 // OPs (in the session) are finished
5208 put_nlist_context_budget(list_context
);
5210 on_finish
->complete(r
);
5215 using namespace librados
;
5217 template <typename T
>
5218 void do_decode(std::vector
<T
>& items
, std::vector
<bufferlist
>& bls
)
5220 for (auto bl
: bls
) {
5221 auto p
= bl
.begin();
5228 struct C_ObjectOperation_scrub_ls
: public Context
{
5231 std::vector
<inconsistent_obj_t
> *objects
= nullptr;
5232 std::vector
<inconsistent_snapset_t
> *snapsets
= nullptr;
5235 C_ObjectOperation_scrub_ls(uint32_t *interval
,
5236 std::vector
<inconsistent_obj_t
> *objects
,
5238 : interval(interval
), objects(objects
), rval(rval
) {}
5239 C_ObjectOperation_scrub_ls(uint32_t *interval
,
5240 std::vector
<inconsistent_snapset_t
> *snapsets
,
5242 : interval(interval
), snapsets(snapsets
), rval(rval
) {}
5243 void finish(int r
) override
{
5244 if (r
< 0 && r
!= -EAGAIN
) {
5255 } catch (buffer::error
&) {
5262 scrub_ls_result_t result
;
5263 auto p
= bl
.begin();
5265 *interval
= result
.interval
;
5267 do_decode(*objects
, result
.vals
);
5269 do_decode(*snapsets
, result
.vals
);
5274 template <typename T
>
5275 void do_scrub_ls(::ObjectOperation
*op
,
5276 const scrub_ls_arg_t
& arg
,
5277 std::vector
<T
> *items
,
5281 OSDOp
& osd_op
= op
->add_op(CEPH_OSD_OP_SCRUBLS
);
5282 op
->flags
|= CEPH_OSD_FLAG_PGOP
;
5284 arg
.encode(osd_op
.indata
);
5285 unsigned p
= op
->ops
.size() - 1;
5286 auto *h
= new C_ObjectOperation_scrub_ls
{interval
, items
, rval
};
5287 op
->out_handler
[p
] = h
;
5288 op
->out_bl
[p
] = &h
->bl
;
5289 op
->out_rval
[p
] = rval
;
5293 void ::ObjectOperation::scrub_ls(const librados::object_id_t
& start_after
,
5294 uint64_t max_to_get
,
5295 std::vector
<librados::inconsistent_obj_t
> *objects
,
5299 scrub_ls_arg_t arg
= {*interval
, 0, start_after
, max_to_get
};
5300 do_scrub_ls(this, arg
, objects
, interval
, rval
);
5303 void ::ObjectOperation::scrub_ls(const librados::object_id_t
& start_after
,
5304 uint64_t max_to_get
,
5305 std::vector
<librados::inconsistent_snapset_t
> *snapsets
,
5309 scrub_ls_arg_t arg
= {*interval
, 1, start_after
, max_to_get
};
5310 do_scrub_ls(this, arg
, snapsets
, interval
, rval
);