*
*/
+#include <algorithm>
#include <cerrno>
#include "Objecter.h"
#include "osd/OSDMap.h"
+#include "osd/error_code.h"
#include "Filer.h"
#include "mon/MonClient.h"
+#include "mon/error_code.h"
#include "msg/Messenger.h"
#include "msg/Message.h"
#include "messages/MWatchNotify.h"
+#include "common/Cond.h"
#include "common/config.h"
#include "common/perf_counters.h"
#include "common/scrub_types.h"
#include "include/str_list.h"
#include "common/errno.h"
#include "common/EventTrace.h"
+#include "common/async/waiter.h"
+#include "error_code.h"
+
using std::list;
using std::make_pair;
using ceph::Formatter;
using std::defer_lock;
+using std::scoped_lock;
+using std::shared_lock;
+using std::unique_lock;
using ceph::real_time;
using ceph::real_clock;
using ceph::acquire_shared;
using ceph::acquire_unique;
+namespace bc = boost::container;
+namespace bs = boost::system;
+namespace ca = ceph::async;
+namespace cb = ceph::buffer;
+
#define dout_subsys ceph_subsys_objecter
#undef dout_prefix
#define dout_prefix *_dout << messenger->get_myname() << ".objecter "
l_osdc_op_send_bytes,
l_osdc_op_resend,
l_osdc_op_reply,
+ l_osdc_oplen_avg,
l_osdc_op,
l_osdc_op_r,
l_osdc_last,
};
+namespace {
+inline bs::error_code osdcode(int r) {
+ return (r < 0) ? bs::error_code(-r, osd_category()) : bs::error_code();
+}
+}
// config obs ----------------------------
-static const char *config_keys[] = {
- "crush_location",
- NULL
-};
-
class Objecter::RequestStateHook : public AdminSocketHook {
Objecter *m_objecter;
public:
explicit RequestStateHook(Objecter *objecter);
int call(std::string_view command, const cmdmap_t& cmdmap,
+ const bufferlist&,
Formatter *f,
std::ostream& ss,
- ceph::buffer::list& out) override;
-};
-
-/**
- * This is a more limited form of C_Contexts, but that requires
- * a ceph_context which we don't have here.
- */
-class ObjectOperation::C_TwoContexts : public Context {
- Context *first;
- Context *second;
-public:
- C_TwoContexts(Context *first, Context *second) :
- first(first), second(second) {}
- void finish(int r) override {
- first->complete(r);
- second->complete(r);
- first = NULL;
- second = NULL;
- }
-
- ~C_TwoContexts() override {
- delete first;
- delete second;
- }
+ cb::list& out) override;
};
-void ObjectOperation::add_handler(Context *extra) {
- size_t last = out_handler.size() - 1;
- Context *orig = out_handler[last];
- if (orig) {
- Context *wrapper = new C_TwoContexts(orig, extra);
- out_handler[last] = wrapper;
- } else {
- out_handler[last] = extra;
- }
-}
-
-Objecter::OSDSession::unique_completion_lock Objecter::OSDSession::get_lock(
- object_t& oid)
+std::unique_lock<std::mutex> Objecter::OSDSession::get_lock(object_t& oid)
{
if (oid.name.empty())
- return unique_completion_lock();
+ return {};
static constexpr uint32_t HASH_PRIME = 1021;
uint32_t h = ceph_str_hash_linux(oid.name.c_str(), oid.name.size())
% HASH_PRIME;
- return unique_completion_lock(completion_locks[h % num_locks],
- std::defer_lock);
+ return {completion_locks[h % num_locks], std::defer_lock};
}
const char** Objecter::get_tracked_conf_keys() const
{
+ static const char *config_keys[] = {
+ "crush_location",
+ "rados_mon_op_timeout",
+ "rados_osd_op_timeout",
+ NULL
+ };
return config_keys;
}
if (changed.count("crush_location")) {
update_crush_location();
}
+ if (changed.count("rados_mon_op_timeout")) {
+ mon_timeout = conf.get_val<std::chrono::seconds>("rados_mon_op_timeout");
+ }
+ if (changed.count("rados_osd_op_timeout")) {
+ osd_timeout = conf.get_val<std::chrono::seconds>("rados_osd_op_timeout");
+ }
}
void Objecter::update_crush_location()
pcb.add_u64_counter(l_osdc_op_send_bytes, "op_send_bytes", "Sent data", NULL, 0, unit_t(UNIT_BYTES));
pcb.add_u64_counter(l_osdc_op_resend, "op_resend", "Resent operations");
pcb.add_u64_counter(l_osdc_op_reply, "op_reply", "Operation reply");
+ pcb.add_u64_avg(l_osdc_oplen_avg, "oplen_avg", "Average length of operation vector");
pcb.add_u64_counter(l_osdc_op, "op", "Operations");
pcb.add_u64_counter(l_osdc_op_r, "op_r", "Read operations", "rd",
}
m_request_state_hook = new RequestStateHook(this);
- AdminSocket* admin_socket = cct->get_admin_socket();
+ auto admin_socket = cct->get_admin_socket();
int ret = admin_socket->register_command("objecter_requests",
m_request_state_hook,
"show in-progress osd requests");
cct->_conf.remove_observer(this);
wl.lock();
- map<int,OSDSession*>::iterator p;
while (!osd_sessions.empty()) {
- p = osd_sessions.begin();
+ auto p = osd_sessions.begin();
close_session(p->second);
}
while(!check_latest_map_lingers.empty()) {
- map<uint64_t, LingerOp*>::iterator i = check_latest_map_lingers.begin();
+ auto i = check_latest_map_lingers.begin();
i->second->put();
check_latest_map_lingers.erase(i->first);
}
while(!check_latest_map_ops.empty()) {
- map<ceph_tid_t, Op*>::iterator i = check_latest_map_ops.begin();
+ auto i = check_latest_map_ops.begin();
i->second->put();
check_latest_map_ops.erase(i->first);
}
while(!check_latest_map_commands.empty()) {
- map<ceph_tid_t, CommandOp*>::iterator i
- = check_latest_map_commands.begin();
+ auto i = check_latest_map_commands.begin();
i->second->put();
check_latest_map_commands.erase(i->first);
}
while(!poolstat_ops.empty()) {
- map<ceph_tid_t,PoolStatOp*>::iterator i = poolstat_ops.begin();
+ auto i = poolstat_ops.begin();
delete i->second;
poolstat_ops.erase(i->first);
}
while(!statfs_ops.empty()) {
- map<ceph_tid_t, StatfsOp*>::iterator i = statfs_ops.begin();
+ auto i = statfs_ops.begin();
delete i->second;
statfs_ops.erase(i->first);
}
while(!pool_ops.empty()) {
- map<ceph_tid_t, PoolOp*>::iterator i = pool_ops.begin();
+ auto i = pool_ops.begin();
delete i->second;
pool_ops.erase(i->first);
}
ldout(cct, 20) << __func__ << " clearing up homeless session..." << dendl;
while(!homeless_session->linger_ops.empty()) {
- std::map<uint64_t, LingerOp*>::iterator i
- = homeless_session->linger_ops.begin();
+ auto i = homeless_session->linger_ops.begin();
ldout(cct, 10) << " linger_op " << i->first << dendl;
LingerOp *lop = i->second;
{
- OSDSession::unique_lock swl(homeless_session->lock);
+ std::unique_lock swl(homeless_session->lock);
_session_linger_op_remove(homeless_session, lop);
}
linger_ops.erase(lop->linger_id);
}
while(!homeless_session->ops.empty()) {
- std::map<ceph_tid_t, Op*>::iterator i = homeless_session->ops.begin();
+ auto i = homeless_session->ops.begin();
ldout(cct, 10) << " op " << i->first << dendl;
- Op *op = i->second;
+ auto op = i->second;
{
- OSDSession::unique_lock swl(homeless_session->lock);
+ std::unique_lock swl(homeless_session->lock);
_session_op_remove(homeless_session, op);
}
op->put();
}
while(!homeless_session->command_ops.empty()) {
- std::map<ceph_tid_t, CommandOp*>::iterator i
- = homeless_session->command_ops.begin();
+ auto i = homeless_session->command_ops.begin();
ldout(cct, 10) << " command_op " << i->first << dendl;
- CommandOp *cop = i->second;
+ auto cop = i->second;
{
- OSDSession::unique_lock swl(homeless_session->lock);
+ std::unique_lock swl(homeless_session->lock);
_session_command_op_remove(homeless_session, cop);
}
cop->put();
// This is safe because we guarantee no concurrent calls to
// shutdown() with the ::initialized check at start.
if (m_request_state_hook) {
- AdminSocket* admin_socket = cct->get_admin_socket();
+ auto admin_socket = cct->get_admin_socket();
admin_socket->unregister_commands(m_request_state_hook);
delete m_request_state_hook;
m_request_state_hook = NULL;
}
void Objecter::_send_linger(LingerOp *info,
- shunique_lock& sul)
+ ceph::shunique_lock<ceph::shared_mutex>& sul)
{
ceph_assert(sul.owns_lock() && sul.mutex() == &rwlock);
- vector<OSDOp> opv;
- Context *oncommit = NULL;
- LingerOp::shared_lock watchl(info->watch_lock);
- ceph::buffer::list *poutbl = NULL;
+ fu2::unique_function<Op::OpSig> oncommit;
+ osdc_opvec opv;
+ std::shared_lock watchl(info->watch_lock);
+ cb::list *poutbl = nullptr;
if (info->registered && info->is_watch) {
ldout(cct, 15) << "send_linger " << info->linger_id << " reconnect"
<< dendl;
opv.back().op.watch.cookie = info->get_cookie();
opv.back().op.watch.op = CEPH_OSD_WATCH_OP_RECONNECT;
opv.back().op.watch.gen = ++info->register_gen;
- oncommit = new C_Linger_Reconnect(this, info);
+ oncommit = CB_Linger_Reconnect(this, info);
} else {
ldout(cct, 15) << "send_linger " << info->linger_id << " register"
<< dendl;
opv = info->ops;
- C_Linger_Commit *c = new C_Linger_Commit(this, info);
+ // TODO Augment ca::Completion with an equivalent of
+ // target so we can handle these cases better.
+ auto c = std::make_unique<CB_Linger_Commit>(this, info);
if (!info->is_watch) {
info->notify_id = 0;
poutbl = &c->outbl;
}
- oncommit = c;
+ oncommit = [c = std::move(c)](bs::error_code ec) mutable {
+ std::move(*c)(ec);
+ };
}
watchl.unlock();
- Op *o = new Op(info->target.base_oid, info->target.base_oloc,
- opv, info->target.flags | CEPH_OSD_FLAG_READ,
- oncommit, info->pobjver);
+ auto o = new Op(info->target.base_oid, info->target.base_oloc,
+ std::move(opv), info->target.flags | CEPH_OSD_FLAG_READ,
+ std::move(oncommit), info->pobjver);
o->outbl = poutbl;
o->snapid = info->snap;
o->snapc = info->snapc;
if (info->register_tid) {
// repeat send. cancel old registration op, if any.
- OSDSession::unique_lock sl(info->session->lock);
+ std::unique_lock sl(info->session->lock);
if (info->session->ops.count(info->register_tid)) {
- Op *o = info->session->ops[info->register_tid];
+ auto o = info->session->ops[info->register_tid];
_op_cancel_map_check(o);
_cancel_linger_op(o);
}
logger->inc(l_osdc_linger_send);
}
-void Objecter::_linger_commit(LingerOp *info, int r, ceph::buffer::list& outbl)
+void Objecter::_linger_commit(LingerOp *info, bs::error_code ec,
+ cb::list& outbl)
{
- LingerOp::unique_lock wl(info->watch_lock);
+ std::unique_lock wl(info->watch_lock);
ldout(cct, 10) << "_linger_commit " << info->linger_id << dendl;
if (info->on_reg_commit) {
- info->on_reg_commit->complete(r);
- info->on_reg_commit = NULL;
+ info->on_reg_commit->defer(std::move(info->on_reg_commit),
+ ec, cb::list{});
+ info->on_reg_commit.reset();
}
- if (r < 0 && info->on_notify_finish) {
- info->on_notify_finish->complete(r);
- info->on_notify_finish = nullptr;
+ if (ec && info->on_notify_finish) {
+ info->on_notify_finish->defer(std::move(info->on_notify_finish),
+ ec, cb::list{});
+ info->on_notify_finish.reset();
}
// only tell the user the first time we do this
ldout(cct, 10) << "_linger_commit notify_id=" << info->notify_id
<< dendl;
}
- catch (ceph::buffer::error& e) {
+ catch (cb::error& e) {
}
}
}
-struct C_DoWatchError : public Context {
+class CB_DoWatchError {
Objecter *objecter;
- Objecter::LingerOp *info;
- int err;
- C_DoWatchError(Objecter *o, Objecter::LingerOp *i, int r)
- : objecter(o), info(i), err(r) {
- info->get();
+ boost::intrusive_ptr<Objecter::LingerOp> info;
+ bs::error_code ec;
+public:
+ CB_DoWatchError(Objecter *o, Objecter::LingerOp *i,
+ bs::error_code ec)
+ : objecter(o), info(i), ec(ec) {
info->_queued_async();
}
- void finish(int r) override {
- Objecter::unique_lock wl(objecter->rwlock);
+ void operator()() {
+ std::unique_lock wl(objecter->rwlock);
bool canceled = info->canceled;
wl.unlock();
if (!canceled) {
- info->watch_context->handle_error(info->get_cookie(), err);
+ info->handle(ec, 0, info->get_cookie(), 0, {});
}
info->finished_async();
- info->put();
}
};
-int Objecter::_normalize_watch_error(int r)
+bs::error_code Objecter::_normalize_watch_error(bs::error_code ec)
{
// translate ENOENT -> ENOTCONN so that a delete->disconnection
// notification and a failure to reconnect because we raced with
// the delete appear the same to the user.
- if (r == -ENOENT)
- r = -ENOTCONN;
- return r;
+ if (ec == bs::errc::no_such_file_or_directory)
+ ec = bs::error_code(ENOTCONN, osd_category());
+ return ec;
}
-void Objecter::_linger_reconnect(LingerOp *info, int r)
+void Objecter::_linger_reconnect(LingerOp *info, bs::error_code ec)
{
- ldout(cct, 10) << __func__ << " " << info->linger_id << " = " << r
+ ldout(cct, 10) << __func__ << " " << info->linger_id << " = " << ec
<< " (last_error " << info->last_error << ")" << dendl;
- if (r < 0) {
- LingerOp::unique_lock wl(info->watch_lock);
+ std::unique_lock wl(info->watch_lock);
+ if (ec) {
if (!info->last_error) {
- r = _normalize_watch_error(r);
- info->last_error = r;
- if (info->watch_context) {
- finisher->queue(new C_DoWatchError(this, info, r));
+ ec = _normalize_watch_error(ec);
+ if (info->handle) {
+ boost::asio::defer(finish_strand, CB_DoWatchError(this, info, ec));
}
}
- wl.unlock();
}
+
+ info->last_error = ec;
}
void Objecter::_send_linger_ping(LingerOp *info)
ldout(cct, 10) << __func__ << " " << info->linger_id << " now " << now
<< dendl;
- vector<OSDOp> opv(1);
+ osdc_opvec opv(1);
opv[0].op.op = CEPH_OSD_OP_WATCH;
opv[0].op.watch.cookie = info->get_cookie();
opv[0].op.watch.op = CEPH_OSD_WATCH_OP_PING;
opv[0].op.watch.gen = info->register_gen;
- C_Linger_Ping *onack = new C_Linger_Ping(this, info);
+
Op *o = new Op(info->target.base_oid, info->target.base_oloc,
- opv, info->target.flags | CEPH_OSD_FLAG_READ,
- onack, NULL, NULL);
+ std::move(opv), info->target.flags | CEPH_OSD_FLAG_READ,
+ CB_Linger_Ping(this, info, now),
+ nullptr, nullptr);
o->target = info->target;
o->should_resend = false;
_send_op_account(o);
_send_op(o);
info->ping_tid = o->tid;
- onack->sent = now;
logger->inc(l_osdc_linger_ping);
}
-void Objecter::_linger_ping(LingerOp *info, int r, ceph::coarse_mono_time sent,
+void Objecter::_linger_ping(LingerOp *info, bs::error_code ec, ceph::coarse_mono_time sent,
uint32_t register_gen)
{
- LingerOp::unique_lock l(info->watch_lock);
+ std::unique_lock l(info->watch_lock);
ldout(cct, 10) << __func__ << " " << info->linger_id
- << " sent " << sent << " gen " << register_gen << " = " << r
+ << " sent " << sent << " gen " << register_gen << " = " << ec
<< " (last_error " << info->last_error
<< " register_gen " << info->register_gen << ")" << dendl;
if (info->register_gen == register_gen) {
- if (r == 0) {
+ if (!ec) {
info->watch_valid_thru = sent;
- } else if (r < 0 && !info->last_error) {
- r = _normalize_watch_error(r);
- info->last_error = r;
- if (info->watch_context) {
- finisher->queue(new C_DoWatchError(this, info, r));
+ } else if (ec && !info->last_error) {
+ ec = _normalize_watch_error(ec);
+ info->last_error = ec;
+ if (info->handle) {
+ boost::asio::defer(finish_strand, CB_DoWatchError(this, info, ec));
}
}
} else {
}
}
-int Objecter::linger_check(LingerOp *info)
+tl::expected<ceph::timespan,
+ bs::error_code> Objecter::linger_check(LingerOp *info)
{
- LingerOp::shared_lock l(info->watch_lock);
+ std::shared_lock l(info->watch_lock);
ceph::coarse_mono_time stamp = info->watch_valid_thru;
if (!info->watch_pending_async.empty())
<< " err " << info->last_error
<< " age " << age << dendl;
if (info->last_error)
- return info->last_error;
+ return tl::unexpected(info->last_error);
// return a safe upper bound (we are truncating to ms)
- return
- 1 + std::chrono::duration_cast<std::chrono::milliseconds>(age).count();
+ return age;
}
void Objecter::linger_cancel(LingerOp *info)
ldout(cct, 20) << __func__ << " linger_id=" << info->linger_id << dendl;
if (!info->canceled) {
OSDSession *s = info->session;
- OSDSession::unique_lock sl(s->lock);
+ std::unique_lock sl(s->lock);
_session_linger_op_remove(s, info);
sl.unlock();
const object_locator_t& oloc,
int flags)
{
- LingerOp *info = new LingerOp(this);
+ unique_lock l(rwlock);
+ // Acquire linger ID
+ auto info = new LingerOp(this, ++max_linger_id);
info->target.base_oid = oid;
info->target.base_oloc = oloc;
if (info->target.base_oloc.key == oid)
info->target.base_oloc.key.clear();
info->target.flags = flags;
info->watch_valid_thru = ceph::coarse_mono_clock::now();
-
- unique_lock l(rwlock);
-
- // Acquire linger ID
- info->linger_id = ++max_linger_id;
ldout(cct, 10) << __func__ << " info " << info
<< " linger_id " << info->linger_id
<< " cookie " << info->get_cookie()
ObjectOperation& op,
const SnapContext& snapc,
real_time mtime,
- ceph::buffer::list& inbl,
- Context *oncommit,
+ cb::list& inbl,
+ decltype(info->on_reg_commit)&& oncommit,
version_t *objver)
{
info->is_watch = true;
info->target.flags |= CEPH_OSD_FLAG_WRITE;
info->ops = op.ops;
info->inbl = inbl;
- info->poutbl = NULL;
info->pobjver = objver;
- info->on_reg_commit = oncommit;
+ info->on_reg_commit = std::move(oncommit);
info->ctx_budget = take_linger_budget(info);
_linger_submit(info, sul);
logger->inc(l_osdc_linger_active);
+ op.clear();
return info->linger_id;
}
ceph_tid_t Objecter::linger_notify(LingerOp *info,
ObjectOperation& op,
- snapid_t snap, ceph::buffer::list& inbl,
- ceph::buffer::list *poutbl,
- Context *onfinish,
+ snapid_t snap, cb::list& inbl,
+ decltype(LingerOp::on_reg_commit)&& onfinish,
version_t *objver)
{
info->snap = snap;
info->target.flags |= CEPH_OSD_FLAG_READ;
info->ops = op.ops;
info->inbl = inbl;
- info->poutbl = poutbl;
info->pobjver = objver;
- info->on_reg_commit = onfinish;
-
+ info->on_reg_commit = std::move(onfinish);
info->ctx_budget = take_linger_budget(info);
-
+
shunique_lock sul(rwlock, ceph::acquire_unique);
_linger_submit(info, sul);
logger->inc(l_osdc_linger_active);
+ op.clear();
return info->linger_id;
}
-void Objecter::_linger_submit(LingerOp *info, shunique_lock& sul)
+void Objecter::_linger_submit(LingerOp *info,
+ ceph::shunique_lock<ceph::shared_mutex>& sul)
{
ceph_assert(sul.owns_lock() && sul.mutex() == &rwlock);
ceph_assert(info->linger_id);
// Populate Op::target
OSDSession *s = NULL;
- _calc_target(&info->target, nullptr);
+ int r = _calc_target(&info->target, nullptr);
+ switch (r) {
+ case RECALC_OP_TARGET_POOL_EIO:
+ _check_linger_pool_eio(info);
+ return;
+ }
// Create LingerOp<->OSDSession relation
- int r = _get_session(info->target.osd, &s, sul);
+ r = _get_session(info->target.osd, &s, sul);
ceph_assert(r == 0);
- OSDSession::unique_lock sl(s->lock);
+ unique_lock sl(s->lock);
_session_linger_op_assign(s, info);
sl.unlock();
put_session(s);
_send_linger(info, sul);
}
-struct C_DoWatchNotify : public Context {
+struct CB_DoWatchNotify {
Objecter *objecter;
- Objecter::LingerOp *info;
- MWatchNotify *msg;
- C_DoWatchNotify(Objecter *o, Objecter::LingerOp *i, MWatchNotify *m)
+ boost::intrusive_ptr<Objecter::LingerOp> info;
+ boost::intrusive_ptr<MWatchNotify> msg;
+ CB_DoWatchNotify(Objecter *o, Objecter::LingerOp *i, MWatchNotify *m)
: objecter(o), info(i), msg(m) {
- info->get();
info->_queued_async();
- msg->get();
}
- void finish(int r) override {
- objecter->_do_watch_notify(info, msg);
+ void operator()() {
+ objecter->_do_watch_notify(std::move(info), std::move(msg));
}
};
ldout(cct, 7) << __func__ << " cookie " << m->cookie << " dne" << dendl;
return;
}
- LingerOp::unique_lock wl(info->watch_lock);
+ std::unique_lock wl(info->watch_lock);
if (m->opcode == CEPH_WATCH_EVENT_DISCONNECT) {
if (!info->last_error) {
- info->last_error = -ENOTCONN;
- if (info->watch_context) {
- finisher->queue(new C_DoWatchError(this, info, -ENOTCONN));
+ info->last_error = bs::error_code(ENOTCONN, osd_category());
+ if (info->handle) {
+ boost::asio::defer(finish_strand, CB_DoWatchError(this, info,
+ info->last_error));
}
}
} else if (!info->is_watch) {
ldout(cct, 10) << __func__ << " reply notify " << m->notify_id
<< " != " << info->notify_id << ", ignoring" << dendl;
} else if (info->on_notify_finish) {
- info->notify_result_bl->claim(m->get_data());
- info->on_notify_finish->complete(m->return_code);
+ info->on_notify_finish->defer(
+ std::move(info->on_notify_finish),
+ osdcode(m->return_code), std::move(m->get_data()));
// if we race with reconnect we might get a second notify; only
// notify the caller once!
- info->on_notify_finish = NULL;
+ info->on_notify_finish = nullptr;
}
} else {
- finisher->queue(new C_DoWatchNotify(this, info, m));
+ boost::asio::defer(finish_strand, CB_DoWatchNotify(this, info, m));
}
}
-void Objecter::_do_watch_notify(LingerOp *info, MWatchNotify *m)
+void Objecter::_do_watch_notify(boost::intrusive_ptr<LingerOp> info,
+ boost::intrusive_ptr<MWatchNotify> m)
{
ldout(cct, 10) << __func__ << " " << *m << dendl;
// notify completion?
ceph_assert(info->is_watch);
- ceph_assert(info->watch_context);
+ ceph_assert(info->handle);
ceph_assert(m->opcode != CEPH_WATCH_EVENT_DISCONNECT);
l.unlock();
switch (m->opcode) {
case CEPH_WATCH_EVENT_NOTIFY:
- info->watch_context->handle_notify(m->notify_id, m->cookie,
- m->notifier_gid, m->bl);
+ info->handle({}, m->notify_id, m->cookie, m->notifier_gid, std::move(m->bl));
break;
}
out:
info->finished_async();
- info->put();
- m->put();
}
bool Objecter::ms_dispatch(Message *m)
map<ceph_tid_t, Op*>& need_resend,
list<LingerOp*>& need_resend_linger,
map<ceph_tid_t, CommandOp*>& need_resend_command,
- shunique_lock& sul)
+ ceph::shunique_lock<ceph::shared_mutex>& sul)
{
ceph_assert(sul.owns_lock() && sul.mutex() == &rwlock);
list<LingerOp*> unregister_lingers;
- OSDSession::unique_lock sl(s->lock);
+ std::unique_lock sl(s->lock);
// check for changed linger mappings (_before_ regular ops)
- map<ceph_tid_t,LingerOp*>::iterator lp = s->linger_ops.begin();
+ auto lp = s->linger_ops.begin();
while (lp != s->linger_ops.end()) {
- LingerOp *op = lp->second;
+ auto op = lp->second;
ceph_assert(op->session == s);
// check_linger_pool_dne() may touch linger_ops; prevent iterator
// invalidation
unregister_lingers.push_back(op);
}
break;
+ case RECALC_OP_TARGET_POOL_EIO:
+ _check_linger_pool_eio(op);
+ ldout(cct, 10) << " need to unregister linger op "
+ << op->linger_id << dendl;
+ op->get();
+ unregister_lingers.push_back(op);
+ break;
}
}
// check for changed request mappings
- map<ceph_tid_t,Op*>::iterator p = s->ops.begin();
+ auto p = s->ops.begin();
while (p != s->ops.end()) {
Op *op = p->second;
++p; // check_op_pool_dne() may touch ops; prevent iterator invalidation
case RECALC_OP_TARGET_POOL_DNE:
_check_op_pool_dne(op, &sl);
break;
+ case RECALC_OP_TARGET_POOL_EIO:
+ _check_op_pool_eio(op, &sl);
+ break;
}
}
// commands
- map<ceph_tid_t,CommandOp*>::iterator cp = s->command_ops.begin();
+ auto cp = s->command_ops.begin();
while (cp != s->command_ops.end()) {
- CommandOp *c = cp->second;
+ auto c = cp->second;
++cp;
ldout(cct, 10) << " checking command " << c->tid << dendl;
bool force_resend_writes = cluster_full;
sl.unlock();
- for (list<LingerOp*>::iterator iter = unregister_lingers.begin();
+ for (auto iter = unregister_lingers.begin();
iter != unregister_lingers.end();
++iter) {
_linger_cancel(*iter);
void Objecter::handle_osd_map(MOSDMap *m)
{
- shunique_lock sul(rwlock, acquire_unique);
+ ceph::shunique_lock sul(rwlock, acquire_unique);
if (!initialized)
return;
bool was_pausewr = osdmap->test_flag(CEPH_OSDMAP_PAUSEWR) || cluster_full ||
_osdmap_has_pool_full();
map<int64_t, bool> pool_full_map;
- for (map<int64_t, pg_pool_t>::const_iterator it
- = osdmap->get_pools().begin();
+ for (auto it = osdmap->get_pools().begin();
it != osdmap->get_pools().end(); ++it)
pool_full_map[it->first] = _osdmap_pool_full(it->second);
OSDMap::Incremental inc(m->incremental_maps[e]);
osdmap->apply_incremental(inc);
- emit_blacklist_events(inc);
+ emit_blocklist_events(inc);
logger->inc(l_osdc_map_inc);
}
auto new_osdmap = std::make_unique<OSDMap>();
new_osdmap->decode(m->maps[e]);
- emit_blacklist_events(*osdmap, *new_osdmap);
+ emit_blocklist_events(*osdmap, *new_osdmap);
osdmap = std::move(new_osdmap);
logger->inc(l_osdc_map_full);
_scan_requests(homeless_session, skipped_map, cluster_full,
&pool_full_map, need_resend,
need_resend_linger, need_resend_command, sul);
- for (map<int,OSDSession*>::iterator p = osd_sessions.begin();
+ for (auto p = osd_sessions.begin();
p != osd_sessions.end(); ) {
- OSDSession *s = p->second;
+ auto s = p->second;
_scan_requests(s, skipped_map, cluster_full,
&pool_full_map, need_resend,
need_resend_linger, need_resend_command, sul);
} else {
// first map. we want the full thing.
if (m->maps.count(m->get_last())) {
- for (map<int,OSDSession*>::iterator p = osd_sessions.begin();
+ for (auto p = osd_sessions.begin();
p != osd_sessions.end(); ++p) {
OSDSession *s = p->second;
_scan_requests(s, false, false, NULL, need_resend,
}
// resend requests
- for (map<ceph_tid_t, Op*>::iterator p = need_resend.begin();
+ for (auto p = need_resend.begin();
p != need_resend.end(); ++p) {
- Op *op = p->second;
- OSDSession *s = op->session;
+ auto op = p->second;
+ auto s = op->session;
bool mapped_session = false;
if (!s) {
int r = _map_session(&op->target, &s, sul);
} else {
get_session(s);
}
- OSDSession::unique_lock sl(s->lock);
+ std::unique_lock sl(s->lock);
if (mapped_session) {
_session_op_assign(s, op);
}
sl.unlock();
put_session(s);
}
- for (list<LingerOp*>::iterator p = need_resend_linger.begin();
+ for (auto p = need_resend_linger.begin();
p != need_resend_linger.end(); ++p) {
LingerOp *op = *p;
ceph_assert(op->session);
_send_linger(op, sul);
}
}
- for (map<ceph_tid_t,CommandOp*>::iterator p = need_resend_command.begin();
+ for (auto p = need_resend_command.begin();
p != need_resend_command.end(); ++p) {
- CommandOp *c = p->second;
+ auto c = p->second;
if (c->target.osd >= 0) {
_assign_command_session(c, sul);
if (c->session && !c->session->is_homeless()) {
_dump_active();
// finish any Contexts that were waiting on a map update
- map<epoch_t,list< pair< Context*, int > > >::iterator p =
- waiting_for_map.begin();
+ auto p = waiting_for_map.begin();
while (p != waiting_for_map.end() &&
p->first <= osdmap->get_epoch()) {
//go through the list and call the onfinish methods
- for (list<pair<Context*, int> >::iterator i = p->second.begin();
- i != p->second.end(); ++i) {
- i->first->complete(i->second);
+ for (auto& [c, ec] : p->second) {
+ ca::post(std::move(c), ec);
}
waiting_for_map.erase(p++);
}
}
}
-void Objecter::enable_blacklist_events()
+void Objecter::enable_blocklist_events()
{
unique_lock wl(rwlock);
- blacklist_events_enabled = true;
+ blocklist_events_enabled = true;
}
-void Objecter::consume_blacklist_events(std::set<entity_addr_t> *events)
+void Objecter::consume_blocklist_events(std::set<entity_addr_t> *events)
{
unique_lock wl(rwlock);
if (events->empty()) {
- events->swap(blacklist_events);
+ events->swap(blocklist_events);
} else {
- for (const auto &i : blacklist_events) {
+ for (const auto &i : blocklist_events) {
events->insert(i);
}
- blacklist_events.clear();
+ blocklist_events.clear();
}
}
-void Objecter::emit_blacklist_events(const OSDMap::Incremental &inc)
+void Objecter::emit_blocklist_events(const OSDMap::Incremental &inc)
{
- if (!blacklist_events_enabled) {
+ if (!blocklist_events_enabled) {
return;
}
- for (const auto &i : inc.new_blacklist) {
- blacklist_events.insert(i.first);
+ for (const auto &i : inc.new_blocklist) {
+ blocklist_events.insert(i.first);
}
}
-void Objecter::emit_blacklist_events(const OSDMap &old_osd_map,
+void Objecter::emit_blocklist_events(const OSDMap &old_osd_map,
const OSDMap &new_osd_map)
{
- if (!blacklist_events_enabled) {
+ if (!blocklist_events_enabled) {
return;
}
std::set<entity_addr_t> old_set;
std::set<entity_addr_t> new_set;
+ std::set<entity_addr_t> old_range_set;
+ std::set<entity_addr_t> new_range_set;
- old_osd_map.get_blacklist(&old_set);
- new_osd_map.get_blacklist(&new_set);
+ old_osd_map.get_blocklist(&old_set, &old_range_set);
+ new_osd_map.get_blocklist(&new_set, &new_range_set);
std::set<entity_addr_t> delta_set;
std::set_difference(
new_set.begin(), new_set.end(), old_set.begin(), old_set.end(),
std::inserter(delta_set, delta_set.begin()));
- blacklist_events.insert(delta_set.begin(), delta_set.end());
+ std::set_difference(
+ new_range_set.begin(), new_range_set.end(),
+ old_range_set.begin(), old_range_set.end(),
+ std::inserter(delta_set, delta_set.begin()));
+ blocklist_events.insert(delta_set.begin(), delta_set.end());
}
// op pool check
-void Objecter::C_Op_Map_Latest::finish(int r)
+void Objecter::CB_Op_Map_Latest::operator()(bs::error_code e,
+ version_t latest, version_t)
{
- if (r == -EAGAIN || r == -ECANCELED)
+ if (e == bs::errc::resource_unavailable_try_again ||
+ e == bs::errc::operation_canceled)
return;
lgeneric_subdout(objecter->cct, objecter, 10)
- << "op_map_latest r=" << r << " tid=" << tid
+ << "op_map_latest r=" << e << " tid=" << tid
<< " latest " << latest << dendl;
- Objecter::unique_lock wl(objecter->rwlock);
+ unique_lock wl(objecter->rwlock);
- map<ceph_tid_t, Op*>::iterator iter =
- objecter->check_latest_map_ops.find(tid);
+ auto iter = objecter->check_latest_map_ops.find(tid);
if (iter == objecter->check_latest_map_ops.end()) {
lgeneric_subdout(objecter->cct, objecter, 10)
<< "op_map_latest op "<< tid << " not found" << dendl;
if (op->map_dne_bound == 0)
op->map_dne_bound = latest;
- OSDSession::unique_lock sl(op->session->lock, defer_lock);
+ unique_lock sl(op->session->lock, defer_lock);
objecter->_check_op_pool_dne(op, &sl);
op->put();
const pg_pool_t *pi = osdmap->get_pg_pool(poolid);
if (!pi)
return -ENOENT;
- for (map<snapid_t,pool_snap_info_t>::const_iterator p = pi->snaps.begin();
+ for (auto p = pi->snaps.begin();
p != pi->snaps.end();
++p) {
snaps->push_back(p->first);
}
// sl may be unlocked.
-void Objecter::_check_op_pool_dne(Op *op, unique_lock *sl)
+void Objecter::_check_op_pool_dne(Op *op, std::unique_lock<std::shared_mutex> *sl)
{
// rwlock is locked unique
ldout(cct, 10) << "check_op_pool_dne tid " << op->tid
<< " concluding pool " << op->target.base_pgid.pool()
<< " dne" << dendl;
- if (op->onfinish) {
+ if (op->has_completion()) {
num_in_flight--;
- op->onfinish->complete(-ENOENT);
+ op->complete(osdc_errc::pool_dne, -ENOENT);
}
OSDSession *s = op->session;
}
}
+// sl may be unlocked.
+void Objecter::_check_op_pool_eio(Op *op, std::unique_lock<std::shared_mutex> *sl)
+{
+ // rwlock is locked unique
+
+ // we had a new enough map
+ ldout(cct, 10) << "check_op_pool_eio tid " << op->tid
+ << " concluding pool " << op->target.base_pgid.pool()
+ << " has eio" << dendl;
+ if (op->has_completion()) {
+ num_in_flight--;
+ op->complete(osdc_errc::pool_eio, -EIO);
+ }
+
+ OSDSession *s = op->session;
+ if (s) {
+ ceph_assert(s != NULL);
+ ceph_assert(sl->mutex() == &s->lock);
+ bool session_locked = sl->owns_lock();
+ if (!session_locked) {
+ sl->lock();
+ }
+ _finish_op(op, 0);
+ if (!session_locked) {
+ sl->unlock();
+ }
+ } else {
+ _finish_op(op, 0); // no session
+ }
+}
+
void Objecter::_send_op_map_check(Op *op)
{
// rwlock is locked unique
if (check_latest_map_ops.count(op->tid) == 0) {
op->get();
check_latest_map_ops[op->tid] = op;
- C_Op_Map_Latest *c = new C_Op_Map_Latest(this, op->tid);
- monc->get_version("osdmap", &c->latest, NULL, c);
+ monc->get_version("osdmap", CB_Op_Map_Latest(this, op->tid));
}
}
void Objecter::_op_cancel_map_check(Op *op)
{
// rwlock is locked unique
- map<ceph_tid_t, Op*>::iterator iter =
- check_latest_map_ops.find(op->tid);
+ auto iter = check_latest_map_ops.find(op->tid);
if (iter != check_latest_map_ops.end()) {
Op *op = iter->second;
op->put();
// linger pool check
-void Objecter::C_Linger_Map_Latest::finish(int r)
+void Objecter::CB_Linger_Map_Latest::operator()(bs::error_code e,
+ version_t latest,
+ version_t)
{
- if (r == -EAGAIN || r == -ECANCELED) {
+ if (e == bs::errc::resource_unavailable_try_again ||
+ e == bs::errc::operation_canceled) {
// ignore callback; we will retry in resend_mon_ops()
return;
}
unique_lock wl(objecter->rwlock);
- map<uint64_t, LingerOp*>::iterator iter =
- objecter->check_latest_map_lingers.find(linger_id);
+ auto iter = objecter->check_latest_map_lingers.find(linger_id);
if (iter == objecter->check_latest_map_lingers.end()) {
return;
}
- LingerOp *op = iter->second;
+ auto op = iter->second;
objecter->check_latest_map_lingers.erase(iter);
if (op->map_dne_bound == 0)
}
if (op->map_dne_bound > 0) {
if (osdmap->get_epoch() >= op->map_dne_bound) {
- LingerOp::unique_lock wl{op->watch_lock};
+ std::unique_lock wl{op->watch_lock};
if (op->on_reg_commit) {
- op->on_reg_commit->complete(-ENOENT);
+ op->on_reg_commit->defer(std::move(op->on_reg_commit),
+ osdc_errc::pool_dne, cb::list{});
op->on_reg_commit = nullptr;
}
if (op->on_notify_finish) {
- op->on_notify_finish->complete(-ENOENT);
+ op->on_notify_finish->defer(std::move(op->on_notify_finish),
+ osdc_errc::pool_dne, cb::list{});
op->on_notify_finish = nullptr;
}
*need_unregister = true;
}
}
+void Objecter::_check_linger_pool_eio(LingerOp *op)
+{
+ // rwlock is locked unique
+
+ std::unique_lock wl{op->watch_lock};
+ if (op->on_reg_commit) {
+ op->on_reg_commit->defer(std::move(op->on_reg_commit),
+ osdc_errc::pool_dne, cb::list{});
+ op->on_reg_commit = nullptr;
+ }
+ if (op->on_notify_finish) {
+ op->on_notify_finish->defer(std::move(op->on_notify_finish),
+ osdc_errc::pool_dne, cb::list{});
+ op->on_notify_finish = nullptr;
+ }
+}
+
void Objecter::_send_linger_map_check(LingerOp *op)
{
// ask the monitor
if (check_latest_map_lingers.count(op->linger_id) == 0) {
op->get();
check_latest_map_lingers[op->linger_id] = op;
- C_Linger_Map_Latest *c = new C_Linger_Map_Latest(this, op->linger_id);
- monc->get_version("osdmap", &c->latest, NULL, c);
+ monc->get_version("osdmap", CB_Linger_Map_Latest(this, op->linger_id));
}
}
{
// rwlock is locked unique
- map<uint64_t, LingerOp*>::iterator iter =
- check_latest_map_lingers.find(op->linger_id);
+ auto iter = check_latest_map_lingers.find(op->linger_id);
if (iter != check_latest_map_lingers.end()) {
LingerOp *op = iter->second;
op->put();
// command pool check
-void Objecter::C_Command_Map_Latest::finish(int r)
+void Objecter::CB_Command_Map_Latest::operator()(bs::error_code e,
+ version_t latest, version_t)
{
- if (r == -EAGAIN || r == -ECANCELED) {
+ if (e == bs::errc::resource_unavailable_try_again ||
+ e == bs::errc::operation_canceled) {
// ignore callback; we will retry in resend_mon_ops()
return;
}
unique_lock wl(objecter->rwlock);
- map<uint64_t, CommandOp*>::iterator iter =
- objecter->check_latest_map_commands.find(tid);
+ auto iter = objecter->check_latest_map_commands.find(tid);
if (iter == objecter->check_latest_map_commands.end()) {
return;
}
- CommandOp *c = iter->second;
+ auto c = iter->second;
objecter->check_latest_map_commands.erase(iter);
if (c->map_dne_bound == 0)
c->map_dne_bound = latest;
- OSDSession::unique_lock sul(c->session->lock);
+ unique_lock sul(c->session->lock);
objecter->_check_command_map_dne(c);
sul.unlock();
<< dendl;
if (c->map_dne_bound > 0) {
if (osdmap->get_epoch() >= c->map_dne_bound) {
- _finish_command(c, c->map_check_error, c->map_check_error_str);
+ _finish_command(c, osdcode(c->map_check_error),
+ std::move(c->map_check_error_str), {});
}
} else {
_send_command_map_check(c);
if (check_latest_map_commands.count(c->tid) == 0) {
c->get();
check_latest_map_commands[c->tid] = c;
- C_Command_Map_Latest *f = new C_Command_Map_Latest(this, c->tid);
- monc->get_version("osdmap", &f->latest, NULL, f);
+ monc->get_version("osdmap", CB_Command_Map_Latest(this, c->tid));
}
}
{
// rwlock is locked uniqe
- map<uint64_t, CommandOp*>::iterator iter =
- check_latest_map_commands.find(c->tid);
+ auto iter = check_latest_map_commands.find(c->tid);
if (iter != check_latest_map_commands.end()) {
- CommandOp *c = iter->second;
+ auto c = iter->second;
c->put();
check_latest_map_commands.erase(iter);
}
* @returns 0 on success, or -EAGAIN if the lock context requires
* promotion to write.
*/
-int Objecter::_get_session(int osd, OSDSession **session, shunique_lock& sul)
+int Objecter::_get_session(int osd, OSDSession **session,
+ shunique_lock<ceph::shared_mutex>& sul)
{
ceph_assert(sul && sul.mutex() == &rwlock);
return 0;
}
- map<int,OSDSession*>::iterator p = osd_sessions.find(osd);
+ auto p = osd_sessions.find(osd);
if (p != osd_sessions.end()) {
- OSDSession *s = p->second;
+ auto s = p->second;
s->get();
*session = s;
ldout(cct, 20) << __func__ << " s=" << s << " osd=" << osd << " "
if (!sul.owns_lock()) {
return -EAGAIN;
}
- OSDSession *s = new OSDSession(cct, osd);
+ auto s = new OSDSession(cct, osd);
osd_sessions[osd] = s;
s->con = messenger->connect_to_osd(osdmap->get_addrs(osd));
s->con->set_priv(RefCountedPtr{s});
s->con->mark_down();
logger->inc(l_osdc_osd_session_close);
}
- OSDSession::unique_lock sl(s->lock);
+ unique_lock sl(s->lock);
std::list<LingerOp*> homeless_lingers;
std::list<CommandOp*> homeless_commands;
std::list<Op*> homeless_ops;
while (!s->linger_ops.empty()) {
- std::map<uint64_t, LingerOp*>::iterator i = s->linger_ops.begin();
+ auto i = s->linger_ops.begin();
ldout(cct, 10) << " linger_op " << i->first << dendl;
homeless_lingers.push_back(i->second);
_session_linger_op_remove(s, i->second);
}
while (!s->ops.empty()) {
- std::map<ceph_tid_t, Op*>::iterator i = s->ops.begin();
+ auto i = s->ops.begin();
ldout(cct, 10) << " op " << i->first << dendl;
homeless_ops.push_back(i->second);
_session_op_remove(s, i->second);
}
while (!s->command_ops.empty()) {
- std::map<ceph_tid_t, CommandOp*>::iterator i = s->command_ops.begin();
+ auto i = s->command_ops.begin();
ldout(cct, 10) << " command_op " << i->first << dendl;
homeless_commands.push_back(i->second);
_session_command_op_remove(s, i->second);
// Assign any leftover ops to the homeless session
{
- OSDSession::unique_lock hsl(homeless_session->lock);
- for (std::list<LingerOp*>::iterator i = homeless_lingers.begin();
+ unique_lock hsl(homeless_session->lock);
+ for (auto i = homeless_lingers.begin();
i != homeless_lingers.end(); ++i) {
_session_linger_op_assign(homeless_session, *i);
}
- for (std::list<Op*>::iterator i = homeless_ops.begin();
+ for (auto i = homeless_ops.begin();
i != homeless_ops.end(); ++i) {
_session_op_assign(homeless_session, *i);
}
- for (std::list<CommandOp*>::iterator i = homeless_commands.begin();
+ for (auto i = homeless_commands.begin();
i != homeless_commands.end(); ++i) {
_session_command_op_assign(homeless_session, *i);
}
return;
}
- // Leave this since it goes with C_SafeCond
- ceph::mutex lock = ceph::make_mutex("");
- ceph::condition_variable cond;
- bool done;
- std::unique_lock mlock{lock};
- C_SafeCond *context = new C_SafeCond(lock, cond, &done, NULL);
- waiting_for_map[e].push_back(pair<Context*, int>(context, 0));
+ ca::waiter<bs::error_code> w;
+ waiting_for_map[e].emplace_back(OpCompletion::create(
+ service.get_executor(),
+ w.ref()),
+ bs::error_code{});
l.unlock();
- cond.wait(mlock, [&done] { return done; });
-}
-
-struct C_Objecter_GetVersion : public Context {
- Objecter *objecter;
- uint64_t oldest, newest;
- Context *fin;
- C_Objecter_GetVersion(Objecter *o, Context *c)
- : objecter(o), oldest(0), newest(0), fin(c) {}
- void finish(int r) override {
- if (r >= 0) {
- objecter->get_latest_version(oldest, newest, fin);
- } else if (r == -EAGAIN) { // try again as instructed
- objecter->wait_for_latest_osdmap(fin);
- } else {
- // it doesn't return any other error codes!
- ceph_abort();
- }
- }
-};
-
-void Objecter::wait_for_latest_osdmap(Context *fin)
-{
- ldout(cct, 10) << __func__ << dendl;
- C_Objecter_GetVersion *c = new C_Objecter_GetVersion(this, fin);
- monc->get_version("osdmap", &c->newest, &c->oldest, c);
+ w.wait();
}
-void Objecter::get_latest_version(epoch_t oldest, epoch_t newest, Context *fin)
+void Objecter::_get_latest_version(epoch_t oldest, epoch_t newest,
+ std::unique_ptr<OpCompletion> fin,
+ std::unique_lock<ceph::shared_mutex>&& l)
{
- unique_lock wl(rwlock);
+ ceph_assert(fin);
if (osdmap->get_epoch() >= newest) {
ldout(cct, 10) << __func__ << " latest " << newest << ", have it" << dendl;
- wl.unlock();
- if (fin)
- fin->complete(0);
- return;
+ l.unlock();
+ ca::defer(std::move(fin), bs::error_code{});
+ } else {
+ ldout(cct, 10) << __func__ << " latest " << newest << ", waiting" << dendl;
+ _wait_for_new_map(std::move(fin), newest, bs::error_code{});
+ l.unlock();
}
-
- ldout(cct, 10) << __func__ << " latest " << newest << ", waiting" << dendl;
- _wait_for_new_map(fin, newest, 0);
}
void Objecter::maybe_request_map()
}
}
-void Objecter::_wait_for_new_map(Context *c, epoch_t epoch, int err)
+void Objecter::_wait_for_new_map(std::unique_ptr<OpCompletion> c, epoch_t epoch,
+ bs::error_code ec)
{
// rwlock is locked unique
- waiting_for_map[epoch].push_back(pair<Context *, int>(c, err));
+ waiting_for_map[epoch].emplace_back(std::move(c), ec);
_maybe_request_map();
}
}
}
-bool Objecter::wait_for_map(epoch_t epoch, Context *c, int err)
-{
- unique_lock wl(rwlock);
- if (osdmap->get_epoch() >= epoch) {
- return true;
- }
- _wait_for_new_map(c, epoch, err);
- return false;
-}
-
void Objecter::_kick_requests(OSDSession *session,
map<uint64_t, LingerOp *>& lresend)
{
// resend ops
map<ceph_tid_t,Op*> resend; // resend in tid order
- for (map<ceph_tid_t, Op*>::iterator p = session->ops.begin();
- p != session->ops.end();) {
+ for (auto p = session->ops.begin(); p != session->ops.end();) {
Op *op = p->second;
++p;
if (op->should_resend) {
// resend lingers
logger->inc(l_osdc_linger_resend, session->linger_ops.size());
- for (map<ceph_tid_t, LingerOp*>::iterator j = session->linger_ops.begin();
+ for (auto j = session->linger_ops.begin();
j != session->linger_ops.end(); ++j) {
LingerOp *op = j->second;
op->get();
// resend commands
logger->inc(l_osdc_command_resend, session->command_ops.size());
map<uint64_t,CommandOp*> cresend; // resend in order
- for (map<ceph_tid_t, CommandOp*>::iterator k = session->command_ops.begin();
+ for (auto k = session->command_ops.begin();
k != session->command_ops.end(); ++k) {
cresend[k->first] = k->second;
}
}
void Objecter::_linger_ops_resend(map<uint64_t, LingerOp *>& lresend,
- unique_lock& ul)
+ unique_lock<ceph::shared_mutex>& ul)
{
ceph_assert(ul.owns_lock());
shunique_lock sul(std::move(ul));
unsigned laggy_ops = 0;
- for (map<int,OSDSession*>::iterator siter = osd_sessions.begin();
+ for (auto siter = osd_sessions.begin();
siter != osd_sessions.end(); ++siter) {
- OSDSession *s = siter->second;
- OSDSession::lock_guard l(s->lock);
+ auto s = siter->second;
+ scoped_lock l(s->lock);
bool found = false;
- for (map<ceph_tid_t,Op*>::iterator p = s->ops.begin();
- p != s->ops.end();
- ++p) {
- Op *op = p->second;
+ for (auto p = s->ops.begin(); p != s->ops.end(); ++p) {
+ auto op = p->second;
ceph_assert(op->session);
if (op->stamp < cutoff) {
ldout(cct, 2) << " tid " << p->first << " on osd." << op->session->osd
++laggy_ops;
}
}
- for (map<uint64_t,LingerOp*>::iterator p = s->linger_ops.begin();
+ for (auto p = s->linger_ops.begin();
p != s->linger_ops.end();
++p) {
- LingerOp *op = p->second;
- LingerOp::unique_lock wl(op->watch_lock);
+ auto op = p->second;
+ std::unique_lock wl(op->watch_lock);
ceph_assert(op->session);
ldout(cct, 10) << " pinging osd that serves lingering tid " << p->first
<< " (osd." << op->session->osd << ")" << dendl;
if (op->is_watch && op->registered && !op->last_error)
_send_linger_ping(op);
}
- for (map<uint64_t,CommandOp*>::iterator p = s->command_ops.begin();
+ for (auto p = s->command_ops.begin();
p != s->command_ops.end();
++p) {
- CommandOp *op = p->second;
+ auto op = p->second;
ceph_assert(op->session);
ldout(cct, 10) << " pinging osd that serves command tid " << p->first
<< " (osd." << op->session->osd << ")" << dendl;
if (!toping.empty()) {
// send a ping to these osds, to ensure we detect any session resets
// (osd reply message policy is lossy)
- for (set<OSDSession*>::const_iterator i = toping.begin();
- i != toping.end();
- ++i) {
+ for (auto i = toping.begin(); i != toping.end(); ++i) {
(*i)->con->send_message(new MPing);
}
}
ldout(cct, 10) << "resend_mon_ops" << dendl;
- for (map<ceph_tid_t,PoolStatOp*>::iterator p = poolstat_ops.begin();
- p != poolstat_ops.end();
- ++p) {
+ for (auto p = poolstat_ops.begin(); p != poolstat_ops.end(); ++p) {
_poolstat_submit(p->second);
logger->inc(l_osdc_poolstat_resend);
}
- for (map<ceph_tid_t,StatfsOp*>::iterator p = statfs_ops.begin();
- p != statfs_ops.end();
- ++p) {
+ for (auto p = statfs_ops.begin(); p != statfs_ops.end(); ++p) {
_fs_stats_submit(p->second);
logger->inc(l_osdc_statfs_resend);
}
- for (map<ceph_tid_t,PoolOp*>::iterator p = pool_ops.begin();
- p != pool_ops.end();
- ++p) {
+ for (auto p = pool_ops.begin(); p != pool_ops.end(); ++p) {
_pool_op_submit(p->second);
logger->inc(l_osdc_poolop_resend);
}
- for (map<ceph_tid_t, Op*>::iterator p = check_latest_map_ops.begin();
+ for (auto p = check_latest_map_ops.begin();
p != check_latest_map_ops.end();
++p) {
- C_Op_Map_Latest *c = new C_Op_Map_Latest(this, p->second->tid);
- monc->get_version("osdmap", &c->latest, NULL, c);
+ monc->get_version("osdmap", CB_Op_Map_Latest(this, p->second->tid));
}
- for (map<uint64_t, LingerOp*>::iterator p = check_latest_map_lingers.begin();
+ for (auto p = check_latest_map_lingers.begin();
p != check_latest_map_lingers.end();
++p) {
- C_Linger_Map_Latest *c
- = new C_Linger_Map_Latest(this, p->second->linger_id);
- monc->get_version("osdmap", &c->latest, NULL, c);
+ monc->get_version("osdmap", CB_Linger_Map_Latest(this, p->second->linger_id));
}
- for (map<uint64_t, CommandOp*>::iterator p
- = check_latest_map_commands.begin();
+ for (auto p = check_latest_map_commands.begin();
p != check_latest_map_commands.end();
++p) {
- C_Command_Map_Latest *c = new C_Command_Map_Latest(this, p->second->tid);
- monc->get_version("osdmap", &c->latest, NULL, c);
+ monc->get_version("osdmap", CB_Command_Map_Latest(this, p->second->tid));
}
}
_op_submit_with_budget(op, rl, ptid, ctx_budget);
}
-void Objecter::_op_submit_with_budget(Op *op, shunique_lock& sul,
+void Objecter::_op_submit_with_budget(Op *op,
+ shunique_lock<ceph::shared_mutex>& sul,
ceph_tid_t *ptid,
int *ctx_budget)
{
inflight_ops++;
// add to gather set(s)
- if (op->onfinish) {
+ if (op->has_completion()) {
num_in_flight++;
} else {
ldout(cct, 20) << " note: not requesting reply" << dendl;
logger->inc(l_osdc_op_active);
logger->inc(l_osdc_op);
+ logger->inc(l_osdc_oplen_avg, op->ops.size());
if ((op->target.flags & (CEPH_OSD_FLAG_READ | CEPH_OSD_FLAG_WRITE)) ==
(CEPH_OSD_FLAG_READ|CEPH_OSD_FLAG_WRITE))
if (op->target.flags & CEPH_OSD_FLAG_PGOP)
logger->inc(l_osdc_op_pg);
- for (vector<OSDOp>::iterator p = op->ops.begin(); p != op->ops.end(); ++p) {
+ for (auto p = op->ops.begin(); p != op->ops.end(); ++p) {
int code = l_osdc_osdop_other;
switch (p->op.op) {
case CEPH_OSD_OP_STAT: code = l_osdc_osdop_stat; break;
}
}
-void Objecter::_op_submit(Op *op, shunique_lock& sul, ceph_tid_t *ptid)
+void Objecter::_op_submit(Op *op, shunique_lock<ceph::shared_mutex>& sul, ceph_tid_t *ptid)
{
// rwlock is locked
ceph_assert(op->session == NULL);
OSDSession *s = NULL;
- bool check_for_latest_map = _calc_target(&op->target, nullptr)
- == RECALC_OP_TARGET_POOL_DNE;
+ bool check_for_latest_map = false;
+ int r = _calc_target(&op->target, nullptr);
+ switch(r) {
+ case RECALC_OP_TARGET_POOL_DNE:
+ check_for_latest_map = true;
+ break;
+ case RECALC_OP_TARGET_POOL_EIO:
+ if (op->has_completion()) {
+ op->complete(osdc_errc::pool_eio, -EIO);
+ }
+ return;
+ }
// Try to get a session, including a retry if we need to take write lock
- int r = _get_session(op->target.osd, &s, sul);
+ r = _get_session(op->target.osd, &s, sul);
if (r == -EAGAIN ||
(check_for_latest_map && sul.owns_lock_shared()) ||
cct->_conf->objecter_debug_inject_relock_delay) {
ceph_assert(op->target.flags & (CEPH_OSD_FLAG_READ|CEPH_OSD_FLAG_WRITE));
- if (pool_full_try) {
- op->target.flags |= CEPH_OSD_FLAG_FULL_TRY;
- }
-
bool need_send = false;
if (op->target.paused) {
ldout(cct, 10) << " tid " << op->tid << " op " << op << " is paused"
_maybe_request_map();
}
- OSDSession::unique_lock sl(s->lock);
+ unique_lock sl(s->lock);
if (op->tid == 0)
op->tid = ++last_tid;
{
ceph_assert(initialized);
- OSDSession::unique_lock sl(s->lock);
+ unique_lock sl(s->lock);
- map<ceph_tid_t, Op*>::iterator p = s->ops.find(tid);
+ auto p = s->ops.find(tid);
if (p == s->ops.end()) {
ldout(cct, 10) << __func__ << " tid " << tid << " dne in session "
<< s->osd << dendl;
ldout(cct, 10) << __func__ << " tid " << tid << " in session " << s->osd
<< dendl;
Op *op = p->second;
- if (op->onfinish) {
+ if (op->has_completion()) {
num_in_flight--;
- op->onfinish->complete(r);
- op->onfinish = NULL;
+ op->complete(osdcode(r), r);
}
_op_cancel_map_check(op);
_finish_op(op, r);
start:
- for (map<int, OSDSession *>::iterator siter = osd_sessions.begin();
+ for (auto siter = osd_sessions.begin();
siter != osd_sessions.end(); ++siter) {
OSDSession *s = siter->second;
- OSDSession::shared_lock sl(s->lock);
+ shared_lock sl(s->lock);
if (s->ops.find(tid) != s->ops.end()) {
sl.unlock();
ret = op_cancel(s, tid, r);
<< " not found in live sessions" << dendl;
// Handle case where the op is in homeless session
- OSDSession::shared_lock sl(homeless_session->lock);
+ shared_lock sl(homeless_session->lock);
if (homeless_session->ops.find(tid) != homeless_session->ops.end()) {
sl.unlock();
ret = op_cancel(homeless_session, tid, r);
std::vector<ceph_tid_t> to_cancel;
bool found = false;
- for (map<int, OSDSession *>::iterator siter = osd_sessions.begin();
+ for (auto siter = osd_sessions.begin();
siter != osd_sessions.end(); ++siter) {
OSDSession *s = siter->second;
- OSDSession::shared_lock sl(s->lock);
- for (map<ceph_tid_t, Op*>::iterator op_i = s->ops.begin();
+ shared_lock sl(s->lock);
+ for (auto op_i = s->ops.begin();
op_i != s->ops.end(); ++op_i) {
if (op_i->second->target.flags & CEPH_OSD_FLAG_WRITE
&& (pool == -1 || op_i->second->target.target_oloc.pool == pool)) {
}
sl.unlock();
- for (std::vector<ceph_tid_t>::iterator titer = to_cancel.begin();
- titer != to_cancel.end();
- ++titer) {
+ for (auto titer = to_cancel.begin(); titer != to_cancel.end(); ++titer) {
int cancel_result = op_cancel(s, *titer, r);
// We hold rwlock across search and cancellation, so cancels
// should always succeed
bool Objecter::_osdmap_has_pool_full() const
{
- for (map<int64_t, pg_pool_t>::const_iterator it
- = osdmap->get_pools().begin();
+ for (auto it = osdmap->get_pools().begin();
it != osdmap->get_pools().end(); ++it) {
if (_osdmap_pool_full(it->second))
return true;
t->osd = -1;
return RECALC_OP_TARGET_POOL_DNE;
}
+
+ if (pi->has_flag(pg_pool_t::FLAG_EIO)) {
+ return RECALC_OP_TARGET_POOL_EIO;
+ }
+
ldout(cct,30) << __func__ << " base pi " << pi
<< " pg_num " << pi->get_pg_num() << dendl;
vector<int> up, acting;
ps_t actual_ps = ceph_stable_mod(pgid.ps(), pg_num, pg_num_mask);
pg_t actual_pgid(actual_ps, pgid.pool());
- pg_mapping_t pg_mapping;
- pg_mapping.epoch = osdmap->get_epoch();
- if (lookup_pg_mapping(actual_pgid, &pg_mapping)) {
- up = pg_mapping.up;
- up_primary = pg_mapping.up_primary;
- acting = pg_mapping.acting;
- acting_primary = pg_mapping.acting_primary;
- } else {
+ if (!lookup_pg_mapping(actual_pgid, osdmap->get_epoch(), &up, &up_primary,
+ &acting, &acting_primary)) {
osdmap->pg_to_up_acting_osds(actual_pgid, &up, &up_primary,
&acting, &acting_primary);
pg_mapping_t pg_mapping(osdmap->get_epoch(),
sort_bitwise,
t->recovery_deletes,
recovery_deletes,
+ t->peering_crush_bucket_count,
+ pi->peering_crush_bucket_count,
+ t->peering_crush_bucket_target,
+ pi->peering_crush_bucket_target,
+ t->peering_crush_bucket_barrier,
+ pi->peering_crush_bucket_barrier,
+ t->peering_crush_mandatory_member,
+ pi->peering_crush_mandatory_member,
prev_pgid)) {
force_resend = true;
}
if (legacy_change || split_or_merge || force_resend) {
t->pgid = pgid;
- t->acting = acting;
+ t->acting = std::move(acting);
t->acting_primary = acting_primary;
t->up_primary = up_primary;
- t->up = up;
+ t->up = std::move(up);
t->size = size;
t->min_size = min_size;
t->pg_num = pg_num;
t->pg_num_pending = pg_num_pending;
spg_t spgid(actual_pgid);
if (pi->is_erasure()) {
- for (uint8_t i = 0; i < acting.size(); ++i) {
- if (acting[i] == acting_primary) {
+ for (uint8_t i = 0; i < t->acting.size(); ++i) {
+ if (t->acting[i] == acting_primary) {
spgid.reset_shard(shard_id_t(i));
break;
}
t->actual_pgid = spgid;
t->sort_bitwise = sort_bitwise;
t->recovery_deletes = recovery_deletes;
+ t->peering_crush_bucket_count = pi->peering_crush_bucket_count;
+ t->peering_crush_bucket_target = pi->peering_crush_bucket_target;
+ t->peering_crush_bucket_barrier = pi->peering_crush_bucket_barrier;
+ t->peering_crush_mandatory_member = pi->peering_crush_mandatory_member;
ldout(cct, 10) << __func__ << " "
<< " raw pgid " << pgid << " -> actual " << t->actual_pgid
- << " acting " << acting
+ << " acting " << t->acting
<< " primary " << acting_primary << dendl;
t->used_replica = false;
- if (acting_primary == -1) {
- t->osd = -1;
- } else {
+ if ((t->flags & (CEPH_OSD_FLAG_BALANCE_READS |
+ CEPH_OSD_FLAG_LOCALIZE_READS)) &&
+ !is_write && pi->is_replicated() && t->acting.size() > 1) {
int osd;
- bool read = is_read && !is_write;
- if (read && (t->flags & CEPH_OSD_FLAG_BALANCE_READS)) {
- int p = rand() % acting.size();
+ ceph_assert(is_read && t->acting[0] == acting_primary);
+ if (t->flags & CEPH_OSD_FLAG_BALANCE_READS) {
+ int p = rand() % t->acting.size();
if (p)
t->used_replica = true;
- osd = acting[p];
- ldout(cct, 10) << " chose random osd." << osd << " of " << acting
+ osd = t->acting[p];
+ ldout(cct, 10) << " chose random osd." << osd << " of " << t->acting
<< dendl;
- } else if (read && (t->flags & CEPH_OSD_FLAG_LOCALIZE_READS) &&
- acting.size() > 1) {
+ } else {
// look for a local replica. prefer the primary if the
// distance is the same.
int best = -1;
int best_locality = 0;
- for (unsigned i = 0; i < acting.size(); ++i) {
+ for (unsigned i = 0; i < t->acting.size(); ++i) {
int locality = osdmap->crush->get_common_ancestor_distance(
- cct, acting[i], crush_location);
+ cct, t->acting[i], crush_location);
ldout(cct, 20) << __func__ << " localize: rank " << i
- << " osd." << acting[i]
+ << " osd." << t->acting[i]
<< " locality " << locality << dendl;
if (i == 0 ||
(locality >= 0 && best_locality >= 0 &&
}
}
ceph_assert(best >= 0);
- osd = acting[best];
- } else {
- osd = acting_primary;
+ osd = t->acting[best];
}
t->osd = osd;
+ } else {
+ t->osd = acting_primary;
}
}
if (legacy_change || unpaused || force_resend) {
}
int Objecter::_map_session(op_target_t *target, OSDSession **s,
- shunique_lock& sul)
+ shunique_lock<ceph::shared_mutex>& sul)
{
_calc_target(target, nullptr);
return _get_session(target->osd, s, sul);
}
int Objecter::_recalc_linger_op_target(LingerOp *linger_op,
- shunique_lock& sul)
+ shunique_lock<ceph::shared_mutex>& sul)
{
// rwlock is locked unique
if (linger_op->session != s) {
// NB locking two sessions (s and linger_op->session) at the
// same time here is only safe because we are the only one that
- // takes two, and we are holding rwlock for write. Disable
- // lockdep because it doesn't know that.
- OSDSession::unique_lock sl(s->lock);
+ // takes two, and we are holding rwlock for write. We use
+ // std::shared_mutex in OSDSession because lockdep doesn't know
+ // that.
+ unique_lock sl(s->lock);
_session_linger_op_remove(linger_op->session, linger_op);
_session_linger_op_assign(s, linger_op);
}
ldout(cct, 15) << "cancel_op " << op->tid << dendl;
ceph_assert(!op->should_resend);
- if (op->onfinish) {
- delete op->onfinish;
+ if (op->has_completion()) {
+ op->onfinish = nullptr;
num_in_flight--;
}
op->put();
}
-MOSDOp *Objecter::_prepare_osd_op(Op *op)
+Objecter::MOSDOp *Objecter::_prepare_osd_op(Op *op)
{
// rwlock is locked
int flags = op->target.flags;
flags |= CEPH_OSD_FLAG_KNOWN_REDIR;
+ flags |= CEPH_OSD_FLAG_SUPPORTSPOOLEIO;
// Nothing checks this any longer, but needed for compatibility with
// pre-luminous osds
op->stamp = ceph::coarse_mono_clock::now();
hobject_t hobj = op->target.get_hobj();
- MOSDOp *m = new MOSDOp(client_inc, op->tid,
- hobj, op->target.actual_pgid,
- osdmap->get_epoch(),
- flags, op->features);
+ auto m = new MOSDOp(client_inc, op->tid,
+ hobj, op->target.actual_pgid,
+ osdmap->get_epoch(),
+ flags, op->features);
m->set_snapid(op->snapid);
m->set_snap_seq(op->snapc.seq);
op->session->con->send_message(m);
}
-int Objecter::calc_op_budget(const vector<OSDOp>& ops)
+int Objecter::calc_op_budget(const bc::small_vector_base<OSDOp>& ops)
{
int op_budget = 0;
- for (vector<OSDOp>::const_iterator i = ops.begin();
- i != ops.end();
- ++i) {
+ for (auto i = ops.begin(); i != ops.end(); ++i) {
if (i->op.op & CEPH_OSD_OP_MODE_WR) {
op_budget += i->indata.length();
} else if (ceph_osd_op_mode_read(i->op.op)) {
}
void Objecter::_throttle_op(Op *op,
- shunique_lock& sul,
+ shunique_lock<ceph::shared_mutex>& sul,
int op_budget)
{
ceph_assert(sul && sul.mutex() == &rwlock);
return;
}
- OSDSession::unique_lock sl(s->lock);
+ unique_lock sl(s->lock);
map<ceph_tid_t, Op *>::iterator iter = s->ops.find(tid);
if (iter == s->ops.end()) {
if (retry_writes_after_first_reply && op->attempts == 1 &&
(op->target.flags & CEPH_OSD_FLAG_WRITE)) {
ldout(cct, 7) << "retrying write after first reply: " << tid << dendl;
- if (op->onfinish) {
+ if (op->has_completion()) {
num_in_flight--;
}
_session_op_remove(s, op);
// have, but that is better than doing callbacks out of order.
}
- Context *onfinish = 0;
+ decltype(op->onfinish) onfinish;
int rc = m->get_result();
if (m->is_redirect_reply()) {
ldout(cct, 5) << " got redirect reply; redirecting" << dendl;
- if (op->onfinish)
+ if (op->has_completion())
num_in_flight--;
_session_op_remove(s, op);
sl.unlock();
if (rc == -EAGAIN) {
ldout(cct, 7) << " got -EAGAIN, resubmitting" << dendl;
- if (op->onfinish)
+ if (op->has_completion())
num_in_flight--;
_session_op_remove(s, op);
sl.unlock();
ldout(cct,10) << __func__ << " copying resulting " << bl.length()
<< " into existing ceph::buffer of length " << op->outbl->length()
<< dendl;
- ceph::buffer::list t;
- t.claim(*op->outbl);
+ cb::list t;
+ t = std::move(*op->outbl);
t.invalidate_crc(); // we're overwriting the raw buffers via c_str()
bl.begin().copy(bl.length(), t.c_str());
op->outbl->substr_of(t, 0, bl.length());
<< " != request ops " << op->ops
<< " from " << m->get_source_inst() << dendl;
- vector<ceph::buffer::list*>::iterator pb = op->out_bl.begin();
- vector<int*>::iterator pr = op->out_rval.begin();
- vector<Context*>::iterator ph = op->out_handler.begin();
+ ceph_assert(op->ops.size() == op->out_bl.size());
+ ceph_assert(op->ops.size() == op->out_rval.size());
+ ceph_assert(op->ops.size() == op->out_ec.size());
+ ceph_assert(op->ops.size() == op->out_handler.size());
+ auto pb = op->out_bl.begin();
+ auto pr = op->out_rval.begin();
+ auto pe = op->out_ec.begin();
+ auto ph = op->out_handler.begin();
ceph_assert(op->out_bl.size() == op->out_rval.size());
ceph_assert(op->out_bl.size() == op->out_handler.size());
- vector<OSDOp>::iterator p = out_ops.begin();
+ auto p = out_ops.begin();
for (unsigned i = 0;
p != out_ops.end() && pb != op->out_bl.end();
- ++i, ++p, ++pb, ++pr, ++ph) {
+ ++i, ++p, ++pb, ++pr, ++pe, ++ph) {
ldout(cct, 10) << " op " << i << " rval " << p->rval
<< " len " << p->outdata.length() << dendl;
if (*pb)
// can change it if e.g. decoding fails
if (*pr)
**pr = ceph_to_hostos_errno(p->rval);
+ if (*pe)
+ **pe = p->rval < 0 ? bs::error_code(-p->rval, osd_category()) :
+ bs::error_code();
if (*ph) {
- ldout(cct, 10) << " op " << i << " handler " << *ph << dendl;
- (*ph)->complete(ceph_to_hostos_errno(p->rval));
- *ph = NULL;
+ std::move((*ph))(p->rval < 0 ?
+ bs::error_code(-p->rval, osd_category()) :
+ bs::error_code(),
+ p->rval, p->outdata);
}
}
// NOTE: we assume that since we only request ONDISK ever we will
// only ever get back one (type of) ack ever.
- if (op->onfinish) {
+ if (op->has_completion()) {
num_in_flight--;
- onfinish = op->onfinish;
- op->onfinish = NULL;
+ onfinish = std::move(op->onfinish);
+ op->onfinish = nullptr;
}
logger->inc(l_osdc_op_reply);
sl.unlock();
// do callbacks
- if (onfinish) {
- onfinish->complete(rc);
+ if (Op::has_completion(onfinish)) {
+ Op::complete(std::move(onfinish), osdcode(rc), rc);
}
if (completion_lock.mutex()) {
completion_lock.unlock();
get_session(s);
- OSDSession::unique_lock sl(s->lock);
+ unique_lock sl(s->lock);
switch (m->op) {
case CEPH_OSD_BACKOFF_OP_BLOCK:
// ack with original backoff's epoch so that the osd can discard this if
// there was a pg split.
- Message *r = new MOSDBackoff(m->pgid,
- m->map_epoch,
- CEPH_OSD_BACKOFF_OP_ACK_BLOCK,
- m->id, m->begin, m->end);
+ auto r = new MOSDBackoff(m->pgid, m->map_epoch,
+ CEPH_OSD_BACKOFF_OP_ACK_BLOCK,
+ m->id, m->begin, m->end);
// this priority must match the MOSDOps from _prepare_osd_op
r->set_priority(cct->_conf->osd_client_op_priority);
con->send_message(r);
op.pg_nls(list_context->max_entries, list_context->filter,
list_context->pos, osdmap->get_epoch());
list_context->bl.clear();
- C_NList *onack = new C_NList(list_context, onfinish, this);
+ auto onack = new C_NList(list_context, onfinish, this);
object_locator_t oloc(list_context->pool_id, list_context->nspace);
// note current_pg in case we don't have (or lose) SORTBITWISE
decode(response, iter);
if (!iter.end()) {
// we do this as legacy.
- ceph::buffer::list legacy_extra_info;
+ cb::list legacy_extra_info;
decode(legacy_extra_info, iter);
}
<< ", handle " << response.handle
<< ", tentative new pos " << list_context->pos << dendl;
if (response_size) {
- list_context->list.splice(list_context->list.end(), response.entries);
+ std::move(response.entries.begin(), response.entries.end(),
+ std::back_inserter(list_context->list));
+ response.entries.clear();
}
if (list_context->list.size() >= list_context->max_entries) {
// snapshots
-int Objecter::create_pool_snap(int64_t pool, string& snap_name,
- Context *onfinish)
+void Objecter::create_pool_snap(int64_t pool, std::string_view snap_name,
+ decltype(PoolOp::onfinish)&& onfinish)
{
unique_lock wl(rwlock);
ldout(cct, 10) << "create_pool_snap; pool: " << pool << "; snap: "
<< snap_name << dendl;
const pg_pool_t *p = osdmap->get_pg_pool(pool);
- if (!p)
- return -EINVAL;
- if (p->snap_exists(snap_name.c_str()))
- return -EEXIST;
+ if (!p) {
+ onfinish->defer(std::move(onfinish), osdc_errc::pool_dne, cb::list{});
+ return;
+ }
+ if (p->snap_exists(snap_name)) {
+ onfinish->defer(std::move(onfinish), osdc_errc::snapshot_exists,
+ cb::list{});
+ return;
+ }
- PoolOp *op = new PoolOp;
- if (!op)
- return -ENOMEM;
+ auto op = new PoolOp;
op->tid = ++last_tid;
op->pool = pool;
op->name = snap_name;
- op->onfinish = onfinish;
+ op->onfinish = std::move(onfinish);
op->pool_op = POOL_OP_CREATE_SNAP;
pool_ops[op->tid] = op;
pool_op_submit(op);
-
- return 0;
}
-struct C_SelfmanagedSnap : public Context {
- ceph::buffer::list bl;
- snapid_t *psnapid;
- Context *fin;
- C_SelfmanagedSnap(snapid_t *ps, Context *f) : psnapid(ps), fin(f) {}
- void finish(int r) override {
- if (r == 0) {
+struct CB_SelfmanagedSnap {
+ std::unique_ptr<ca::Completion<void(bs::error_code, snapid_t)>> fin;
+ CB_SelfmanagedSnap(decltype(fin)&& fin)
+ : fin(std::move(fin)) {}
+ void operator()(bs::error_code ec, const cb::list& bl) {
+ snapid_t snapid = 0;
+ if (!ec) {
try {
- auto p = bl.cbegin();
- decode(*psnapid, p);
- } catch (ceph::buffer::error&) {
- r = -EIO;
+ auto p = bl.cbegin();
+ decode(snapid, p);
+ } catch (const cb::error& e) {
+ ec = e.code();
}
}
- fin->complete(r);
+ fin->defer(std::move(fin), ec, snapid);
}
};
-int Objecter::allocate_selfmanaged_snap(int64_t pool, snapid_t *psnapid,
- Context *onfinish)
+void Objecter::allocate_selfmanaged_snap(
+ int64_t pool,
+ std::unique_ptr<ca::Completion<void(bs::error_code, snapid_t)>> onfinish)
{
unique_lock wl(rwlock);
ldout(cct, 10) << "allocate_selfmanaged_snap; pool: " << pool << dendl;
- PoolOp *op = new PoolOp;
- if (!op) return -ENOMEM;
+ auto op = new PoolOp;
op->tid = ++last_tid;
op->pool = pool;
- C_SelfmanagedSnap *fin = new C_SelfmanagedSnap(psnapid, onfinish);
- op->onfinish = fin;
- op->blp = &fin->bl;
+ op->onfinish = PoolOp::OpComp::create(
+ service.get_executor(),
+ CB_SelfmanagedSnap(std::move(onfinish)));
op->pool_op = POOL_OP_CREATE_UNMANAGED_SNAP;
pool_ops[op->tid] = op;
pool_op_submit(op);
- return 0;
}
-int Objecter::delete_pool_snap(int64_t pool, string& snap_name,
- Context *onfinish)
+void Objecter::delete_pool_snap(
+ int64_t pool, std::string_view snap_name,
+ decltype(PoolOp::onfinish)&& onfinish)
{
unique_lock wl(rwlock);
ldout(cct, 10) << "delete_pool_snap; pool: " << pool << "; snap: "
<< snap_name << dendl;
const pg_pool_t *p = osdmap->get_pg_pool(pool);
- if (!p)
- return -EINVAL;
- if (!p->snap_exists(snap_name.c_str()))
- return -ENOENT;
+ if (!p) {
+ onfinish->defer(std::move(onfinish), osdc_errc::pool_dne, cb::list{});
+ return;
+ }
+
+ if (!p->snap_exists(snap_name)) {
+ onfinish->defer(std::move(onfinish), osdc_errc::snapshot_dne, cb::list{});
+ return;
+ }
- PoolOp *op = new PoolOp;
- if (!op)
- return -ENOMEM;
+ auto op = new PoolOp;
op->tid = ++last_tid;
op->pool = pool;
op->name = snap_name;
- op->onfinish = onfinish;
+ op->onfinish = std::move(onfinish);
op->pool_op = POOL_OP_DELETE_SNAP;
pool_ops[op->tid] = op;
pool_op_submit(op);
-
- return 0;
}
-int Objecter::delete_selfmanaged_snap(int64_t pool, snapid_t snap,
- Context *onfinish)
+void Objecter::delete_selfmanaged_snap(int64_t pool, snapid_t snap,
+ decltype(PoolOp::onfinish)&& onfinish)
{
unique_lock wl(rwlock);
ldout(cct, 10) << "delete_selfmanaged_snap; pool: " << pool << "; snap: "
<< snap << dendl;
- PoolOp *op = new PoolOp;
- if (!op) return -ENOMEM;
+ auto op = new PoolOp;
op->tid = ++last_tid;
op->pool = pool;
- op->onfinish = onfinish;
+ op->onfinish = std::move(onfinish);
op->pool_op = POOL_OP_DELETE_UNMANAGED_SNAP;
op->snapid = snap;
pool_ops[op->tid] = op;
pool_op_submit(op);
-
- return 0;
}
-int Objecter::create_pool(string& name, Context *onfinish,
- int crush_rule)
+void Objecter::create_pool(std::string_view name,
+ decltype(PoolOp::onfinish)&& onfinish,
+ int crush_rule)
{
unique_lock wl(rwlock);
ldout(cct, 10) << "create_pool name=" << name << dendl;
- if (osdmap->lookup_pg_pool_name(name) >= 0)
- return -EEXIST;
+ if (osdmap->lookup_pg_pool_name(name) >= 0) {
+ onfinish->defer(std::move(onfinish), osdc_errc::pool_exists, cb::list{});
+ return;
+ }
- PoolOp *op = new PoolOp;
- if (!op)
- return -ENOMEM;
+ auto op = new PoolOp;
op->tid = ++last_tid;
op->pool = 0;
op->name = name;
- op->onfinish = onfinish;
+ op->onfinish = std::move(onfinish);
op->pool_op = POOL_OP_CREATE;
pool_ops[op->tid] = op;
op->crush_rule = crush_rule;
pool_op_submit(op);
-
- return 0;
}
-int Objecter::delete_pool(int64_t pool, Context *onfinish)
+void Objecter::delete_pool(int64_t pool,
+ decltype(PoolOp::onfinish)&& onfinish)
{
unique_lock wl(rwlock);
ldout(cct, 10) << "delete_pool " << pool << dendl;
if (!osdmap->have_pg_pool(pool))
- return -ENOENT;
-
- _do_delete_pool(pool, onfinish);
- return 0;
+ onfinish->defer(std::move(onfinish), osdc_errc::pool_dne, cb::list{});
+ else
+ _do_delete_pool(pool, std::move(onfinish));
}
-int Objecter::delete_pool(const string &pool_name, Context *onfinish)
+void Objecter::delete_pool(std::string_view pool_name,
+ decltype(PoolOp::onfinish)&& onfinish)
{
unique_lock wl(rwlock);
ldout(cct, 10) << "delete_pool " << pool_name << dendl;
int64_t pool = osdmap->lookup_pg_pool_name(pool_name);
if (pool < 0)
- return pool;
-
- _do_delete_pool(pool, onfinish);
- return 0;
+ // This only returns one error: -ENOENT.
+ onfinish->defer(std::move(onfinish), osdc_errc::pool_dne, cb::list{});
+ else
+ _do_delete_pool(pool, std::move(onfinish));
}
-void Objecter::_do_delete_pool(int64_t pool, Context *onfinish)
+void Objecter::_do_delete_pool(int64_t pool,
+ decltype(PoolOp::onfinish)&& onfinish)
+
{
- PoolOp *op = new PoolOp;
+ auto op = new PoolOp;
op->tid = ++last_tid;
op->pool = pool;
op->name = "delete";
- op->onfinish = onfinish;
+ op->onfinish = std::move(onfinish);
op->pool_op = POOL_OP_DELETE;
pool_ops[op->tid] = op;
pool_op_submit(op);
// rwlock is locked unique
ldout(cct, 10) << "pool_op_submit " << op->tid << dendl;
- MPoolOp *m = new MPoolOp(monc->get_fsid(), op->tid, op->pool,
- op->name, op->pool_op,
- last_seen_osdmap_version);
+ auto m = new MPoolOp(monc->get_fsid(), op->tid, op->pool,
+ op->name, op->pool_op,
+ last_seen_osdmap_version);
if (op->snapid) m->snapid = op->snapid;
if (op->crush_rule) m->crush_rule = op->crush_rule;
monc->send_mon_message(m);
/**
* Handle a reply to a PoolOp message. Check that we sent the message
- * and give the caller responsibility for the returned ceph::buffer::list.
+ * and give the caller responsibility for the returned cb::list.
* Then either call the finisher or stash the PoolOp, depending on if we
* have a new enough map.
* Lastly, clean up the message and PoolOp.
*/
void Objecter::handle_pool_op_reply(MPoolOpReply *m)
{
+ int rc = m->replyCode;
+ auto ec = rc < 0 ? bs::error_code(-rc, mon_category()) : bs::error_code();
FUNCTRACE(cct);
shunique_lock sul(rwlock, acquire_shared);
if (!initialized) {
ldout(cct, 10) << "handle_pool_op_reply " << *m << dendl;
ceph_tid_t tid = m->get_tid();
- map<ceph_tid_t, PoolOp *>::iterator iter = pool_ops.find(tid);
+ auto iter = pool_ops.find(tid);
if (iter != pool_ops.end()) {
PoolOp *op = iter->second;
ldout(cct, 10) << "have request " << tid << " at " << op << " Op: "
<< ceph_pool_op_name(op->pool_op) << dendl;
- if (op->blp)
- op->blp->claim(m->response_data);
+ cb::list bl{std::move(m->response_data)};
if (m->version > last_seen_osdmap_version)
last_seen_osdmap_version = m->version;
if (osdmap->get_epoch() < m->epoch) {
if (osdmap->get_epoch() < m->epoch) {
ldout(cct, 20) << "waiting for client to reach epoch " << m->epoch
<< " before calling back" << dendl;
- _wait_for_new_map(op->onfinish, m->epoch, m->replyCode);
+ _wait_for_new_map(OpCompletion::create(
+ service.get_executor(),
+ [o = std::move(op->onfinish),
+ bl = std::move(bl)](
+ bs::error_code ec) mutable {
+ o->defer(std::move(o), ec, bl);
+ }),
+ m->epoch,
+ ec);
} else {
// map epoch changed, probably because a MOSDMap message
// sneaked in. Do caller-specified callback now or else
// we lose it forever.
ceph_assert(op->onfinish);
- op->onfinish->complete(m->replyCode);
+ op->onfinish->defer(std::move(op->onfinish), ec, std::move(bl));
}
} else {
ceph_assert(op->onfinish);
- op->onfinish->complete(m->replyCode);
+ op->onfinish->defer(std::move(op->onfinish), ec, std::move(bl));
}
- op->onfinish = NULL;
+ op->onfinish = nullptr;
if (!sul.owns_lock()) {
sul.unlock();
sul.lock();
unique_lock wl(rwlock);
- map<ceph_tid_t, PoolOp*>::iterator it = pool_ops.find(tid);
+ auto it = pool_ops.find(tid);
if (it == pool_ops.end()) {
ldout(cct, 10) << __func__ << " tid " << tid << " dne" << dendl;
return -ENOENT;
PoolOp *op = it->second;
if (op->onfinish)
- op->onfinish->complete(r);
+ op->onfinish->defer(std::move(op->onfinish), osdcode(r), cb::list{});
_finish_pool_op(op, r);
return 0;
// pool stats
-void Objecter::get_pool_stats(list<string>& pools,
- map<string,pool_stat_t> *result,
- bool *per_pool,
- Context *onfinish)
+void Objecter::get_pool_stats(
+ const std::vector<std::string>& pools,
+ decltype(PoolStatOp::onfinish)&& onfinish)
{
ldout(cct, 10) << "get_pool_stats " << pools << dendl;
- PoolStatOp *op = new PoolStatOp;
+ auto op = new PoolStatOp;
op->tid = ++last_tid;
op->pools = pools;
- op->pool_stats = result;
- op->per_pool = per_pool;
- op->onfinish = onfinish;
+ op->onfinish = std::move(onfinish);
if (mon_timeout > timespan(0)) {
op->ontimeout = timer.add_event(mon_timeout,
[this, op]() {
return;
}
- map<ceph_tid_t, PoolStatOp *>::iterator iter = poolstat_ops.find(tid);
+ auto iter = poolstat_ops.find(tid);
if (iter != poolstat_ops.end()) {
PoolStatOp *op = poolstat_ops[tid];
ldout(cct, 10) << "have request " << tid << " at " << op << dendl;
- *op->pool_stats = m->pool_stats;
- *op->per_pool = m->per_pool;
if (m->version > last_seen_pgmap_version) {
last_seen_pgmap_version = m->version;
}
- op->onfinish->complete(0);
+ op->onfinish->defer(std::move(op->onfinish), bs::error_code{},
+ std::move(m->pool_stats), m->per_pool);
_finish_pool_stat_op(op, 0);
} else {
ldout(cct, 10) << "unknown request " << tid << dendl;
unique_lock wl(rwlock);
- map<ceph_tid_t, PoolStatOp*>::iterator it = poolstat_ops.find(tid);
+ auto it = poolstat_ops.find(tid);
if (it == poolstat_ops.end()) {
ldout(cct, 10) << __func__ << " tid " << tid << " dne" << dendl;
return -ENOENT;
ldout(cct, 10) << __func__ << " tid " << tid << dendl;
- PoolStatOp *op = it->second;
+ auto op = it->second;
if (op->onfinish)
- op->onfinish->complete(r);
+ op->onfinish->defer(std::move(op->onfinish), osdcode(r),
+ bc::flat_map<std::string, pool_stat_t>{}, false);
_finish_pool_stat_op(op, r);
return 0;
}
delete op;
}
-void Objecter::get_fs_stats(ceph_statfs& result,
- boost::optional<int64_t> data_pool,
- Context *onfinish)
+void Objecter::get_fs_stats(std::optional<int64_t> poolid,
+ decltype(StatfsOp::onfinish)&& onfinish)
{
ldout(cct, 10) << "get_fs_stats" << dendl;
unique_lock l(rwlock);
- StatfsOp *op = new StatfsOp;
+ auto op = new StatfsOp;
op->tid = ++last_tid;
- op->stats = &result;
- op->data_pool = data_pool;
- op->onfinish = onfinish;
+ op->data_pool = poolid;
+ op->onfinish = std::move(onfinish);
if (mon_timeout > timespan(0)) {
op->ontimeout = timer.add_event(mon_timeout,
[this, op]() {
if (statfs_ops.count(tid)) {
StatfsOp *op = statfs_ops[tid];
ldout(cct, 10) << "have request " << tid << " at " << op << dendl;
- *(op->stats) = m->h.st;
if (m->h.version > last_seen_pgmap_version)
last_seen_pgmap_version = m->h.version;
- op->onfinish->complete(0);
+ op->onfinish->defer(std::move(op->onfinish), bs::error_code{}, m->h.st);
_finish_statfs_op(op, 0);
} else {
ldout(cct, 10) << "unknown request " << tid << dendl;
unique_lock wl(rwlock);
- map<ceph_tid_t, StatfsOp*>::iterator it = statfs_ops.find(tid);
+ auto it = statfs_ops.find(tid);
if (it == statfs_ops.end()) {
ldout(cct, 10) << __func__ << " tid " << tid << " dne" << dendl;
return -ENOENT;
ldout(cct, 10) << __func__ << " tid " << tid << dendl;
- StatfsOp *op = it->second;
+ auto op = it->second;
if (op->onfinish)
- op->onfinish->complete(r);
+ op->onfinish->defer(std::move(op->onfinish), osdcode(r), ceph_statfs{});
_finish_statfs_op(op, r);
return 0;
}
// scatter/gather
void Objecter::_sg_read_finish(vector<ObjectExtent>& extents,
- vector<ceph::buffer::list>& resultbl,
- ceph::buffer::list *bl, Context *onfinish)
+ vector<cb::list>& resultbl,
+ cb::list *bl, Context *onfinish)
{
// all done
ldout(cct, 15) << "_sg_read_finish" << dendl;
if (extents.size() > 1) {
Striper::StripedReadResult r;
- vector<ceph::buffer::list>::iterator bit = resultbl.begin();
- for (vector<ObjectExtent>::iterator eit = extents.begin();
+ auto bit = resultbl.begin();
+ for (auto eit = extents.begin();
eit != extents.end();
++eit, ++bit) {
r.add_partial_result(cct, *bit, eit->buffer_extents);
r.assemble_result(cct, *bl, false);
} else {
ldout(cct, 15) << " only one frag" << dendl;
- bl->claim(resultbl[0]);
+ *bl = std::move(resultbl[0]);
}
// done
return false;
}
map<uint64_t, LingerOp *> lresend;
- OSDSession::unique_lock sl(session->lock);
+ unique_lock sl(session->lock);
_reopen_session(session);
_kick_requests(session, lresend);
sl.unlock();
void Objecter::_dump_active(OSDSession *s)
{
- for (map<ceph_tid_t,Op*>::iterator p = s->ops.begin();
- p != s->ops.end();
- ++p) {
+ for (auto p = s->ops.begin(); p != s->ops.end(); ++p) {
Op *op = p->second;
ldout(cct, 20) << op->tid << "\t" << op->target.pgid
<< "\tosd." << (op->session ? op->session->osd : -1)
{
ldout(cct, 20) << "dump_active .. " << num_homeless_ops << " homeless"
<< dendl;
- for (map<int, OSDSession *>::iterator siter = osd_sessions.begin();
+ for (auto siter = osd_sessions.begin();
siter != osd_sessions.end(); ++siter) {
- OSDSession *s = siter->second;
- OSDSession::shared_lock sl(s->lock);
+ auto s = siter->second;
+ shared_lock sl(s->lock);
_dump_active(s);
sl.unlock();
}
void Objecter::_dump_ops(const OSDSession *s, Formatter *fmt)
{
- for (map<ceph_tid_t,Op*>::const_iterator p = s->ops.begin();
- p != s->ops.end();
- ++p) {
+ for (auto p = s->ops.begin(); p != s->ops.end(); ++p) {
Op *op = p->second;
- auto age = std::chrono::duration<double>(coarse_mono_clock::now() - op->stamp);
+ auto age = std::chrono::duration<double>(ceph::coarse_mono_clock::now() - op->stamp);
fmt->open_object_section("op");
fmt->dump_unsigned("tid", op->tid);
op->target.dump(fmt);
fmt->dump_stream("mtime") << op->mtime;
fmt->open_array_section("osd_ops");
- for (vector<OSDOp>::const_iterator it = op->ops.begin();
- it != op->ops.end();
- ++it) {
+ for (auto it = op->ops.begin(); it != op->ops.end(); ++it) {
fmt->dump_stream("osd_op") << *it;
}
fmt->close_section(); // osd_ops array
{
// Read-lock on Objecter held
fmt->open_array_section("ops");
- for (map<int, OSDSession *>::const_iterator siter = osd_sessions.begin();
+ for (auto siter = osd_sessions.begin();
siter != osd_sessions.end(); ++siter) {
OSDSession *s = siter->second;
- OSDSession::shared_lock sl(s->lock);
+ shared_lock sl(s->lock);
_dump_ops(s, fmt);
sl.unlock();
}
void Objecter::_dump_linger_ops(const OSDSession *s, Formatter *fmt)
{
- for (map<uint64_t, LingerOp*>::const_iterator p = s->linger_ops.begin();
- p != s->linger_ops.end();
- ++p) {
- LingerOp *op = p->second;
+ for (auto p = s->linger_ops.begin(); p != s->linger_ops.end(); ++p) {
+ auto op = p->second;
fmt->open_object_section("linger_op");
fmt->dump_unsigned("linger_id", op->linger_id);
op->target.dump(fmt);
{
// We have a read-lock on the objecter
fmt->open_array_section("linger_ops");
- for (map<int, OSDSession *>::const_iterator siter = osd_sessions.begin();
+ for (auto siter = osd_sessions.begin();
siter != osd_sessions.end(); ++siter) {
- OSDSession *s = siter->second;
- OSDSession::shared_lock sl(s->lock);
+ auto s = siter->second;
+ shared_lock sl(s->lock);
_dump_linger_ops(s, fmt);
sl.unlock();
}
void Objecter::_dump_command_ops(const OSDSession *s, Formatter *fmt)
{
- for (map<uint64_t, CommandOp*>::const_iterator p = s->command_ops.begin();
- p != s->command_ops.end();
- ++p) {
- CommandOp *op = p->second;
+ for (auto p = s->command_ops.begin(); p != s->command_ops.end(); ++p) {
+ auto op = p->second;
fmt->open_object_section("command_op");
fmt->dump_unsigned("command_id", op->tid);
fmt->dump_int("osd", op->session ? op->session->osd : -1);
fmt->open_array_section("command");
- for (vector<string>::const_iterator q = op->cmd.begin();
- q != op->cmd.end(); ++q)
+ for (auto q = op->cmd.begin(); q != op->cmd.end(); ++q)
fmt->dump_string("word", *q);
fmt->close_section();
if (op->target_osd >= 0)
{
// We have a read-lock on the Objecter here
fmt->open_array_section("command_ops");
- for (map<int, OSDSession *>::const_iterator siter = osd_sessions.begin();
+ for (auto siter = osd_sessions.begin();
siter != osd_sessions.end(); ++siter) {
- OSDSession *s = siter->second;
- OSDSession::shared_lock sl(s->lock);
+ auto s = siter->second;
+ shared_lock sl(s->lock);
_dump_command_ops(s, fmt);
sl.unlock();
}
void Objecter::dump_pool_ops(Formatter *fmt) const
{
fmt->open_array_section("pool_ops");
- for (map<ceph_tid_t, PoolOp*>::const_iterator p = pool_ops.begin();
- p != pool_ops.end();
- ++p) {
- PoolOp *op = p->second;
+ for (auto p = pool_ops.begin(); p != pool_ops.end(); ++p) {
+ auto op = p->second;
fmt->open_object_section("pool_op");
fmt->dump_unsigned("tid", op->tid);
fmt->dump_int("pool", op->pool);
void Objecter::dump_pool_stat_ops(Formatter *fmt) const
{
fmt->open_array_section("pool_stat_ops");
- for (map<ceph_tid_t, PoolStatOp*>::const_iterator p = poolstat_ops.begin();
+ for (auto p = poolstat_ops.begin();
p != poolstat_ops.end();
++p) {
PoolStatOp *op = p->second;
fmt->dump_stream("last_sent") << op->last_submit;
fmt->open_array_section("pools");
- for (list<string>::const_iterator it = op->pools.begin();
- it != op->pools.end();
- ++it) {
- fmt->dump_string("pool", *it);
+ for (const auto& it : op->pools) {
+ fmt->dump_string("pool", it);
}
fmt->close_section(); // pools array
void Objecter::dump_statfs_ops(Formatter *fmt) const
{
fmt->open_array_section("statfs_ops");
- for (map<ceph_tid_t, StatfsOp*>::const_iterator p = statfs_ops.begin();
- p != statfs_ops.end();
- ++p) {
- StatfsOp *op = p->second;
+ for (auto p = statfs_ops.begin(); p != statfs_ops.end(); ++p) {
+ auto op = p->second;
fmt->open_object_section("statfs_op");
fmt->dump_unsigned("tid", op->tid);
fmt->dump_stream("last_sent") << op->last_submit;
int Objecter::RequestStateHook::call(std::string_view command,
const cmdmap_t& cmdmap,
+ const bufferlist&,
Formatter *f,
std::ostream& ss,
- ceph::buffer::list& out)
+ cb::list& out)
{
shared_lock rl(m_objecter->rwlock);
m_objecter->dump_requests(f);
return 0;
}
-void Objecter::blacklist_self(bool set)
+void Objecter::blocklist_self(bool set)
{
- ldout(cct, 10) << "blacklist_self " << (set ? "add" : "rm") << dendl;
+ ldout(cct, 10) << "blocklist_self " << (set ? "add" : "rm") << dendl;
vector<string> cmd;
- cmd.push_back("{\"prefix\":\"osd blacklist\", ");
+ cmd.push_back("{\"prefix\":\"osd blocklist\", ");
if (set)
- cmd.push_back("\"blacklistop\":\"add\",");
+ cmd.push_back("\"blocklistop\":\"add\",");
else
- cmd.push_back("\"blacklistop\":\"rm\",");
+ cmd.push_back("\"blocklistop\":\"rm\",");
stringstream ss;
- // this is somewhat imprecise in that we are blacklisting our first addr only
+ // this is somewhat imprecise in that we are blocklisting our first addr only
ss << messenger->get_myaddrs().front().get_legacy_str();
cmd.push_back("\"addr\":\"" + ss.str() + "\"");
- MMonCommand *m = new MMonCommand(monc->get_fsid());
+ auto m = new MMonCommand(monc->get_fsid());
m->cmd = cmd;
+ // NOTE: no fallback to legacy blacklist command implemented here
+ // since this is only used for test code.
+
monc->send_mon_message(m);
}
return;
}
- OSDSession::shared_lock sl(s->lock);
- map<ceph_tid_t,CommandOp*>::iterator p = s->command_ops.find(m->get_tid());
+ shared_lock sl(s->lock);
+ auto p = s->command_ops.find(m->get_tid());
if (p == s->command_ops.end()) {
ldout(cct, 10) << "handle_command_reply tid " << m->get_tid()
<< " not found" << dendl;
sl.unlock();
return;
}
+
if (m->r == -EAGAIN) {
ldout(cct,10) << __func__ << " tid " << m->get_tid()
<< " got EAGAIN, requesting map and resending" << dendl;
return;
}
- if (c->poutbl) {
- c->poutbl->claim(m->get_data());
- }
-
sl.unlock();
- OSDSession::unique_lock sul(s->lock);
- _finish_command(c, m->r, m->rs);
+ unique_lock sul(s->lock);
+ _finish_command(c, m->r < 0 ? bs::error_code(-m->r, osd_category()) :
+ bs::error_code(), std::move(m->rs),
+ std::move(m->get_data()));
sul.unlock();
m->put();
}
+Objecter::LingerOp::LingerOp(Objecter *o, uint64_t linger_id)
+ : objecter(o),
+ linger_id(linger_id),
+ watch_lock(ceph::make_shared_mutex(
+ fmt::format("LingerOp::watch_lock #{}", linger_id)))
+{}
+
void Objecter::submit_command(CommandOp *c, ceph_tid_t *ptid)
{
shunique_lock sul(rwlock, ceph::acquire_unique);
c->tid = tid;
{
- OSDSession::unique_lock hs_wl(homeless_session->lock);
+ unique_lock hs_wl(homeless_session->lock);
_session_command_op_assign(homeless_session, c);
}
if (osd_timeout > timespan(0)) {
c->ontimeout = timer.add_event(osd_timeout,
[this, c, tid]() {
- command_op_cancel(c->session, tid,
- -ETIMEDOUT); });
+ command_op_cancel(
+ c->session, tid,
+ osdc_errc::timed_out); });
}
if (!c->session->is_homeless()) {
}
if (c->map_check_error)
_send_command_map_check(c);
- *ptid = tid;
+ if (ptid)
+ *ptid = tid;
logger->inc(l_osdc_command_active);
}
-int Objecter::_calc_command_target(CommandOp *c, shunique_lock& sul)
+int Objecter::_calc_command_target(CommandOp *c,
+ shunique_lock<ceph::shared_mutex>& sul)
{
ceph_assert(sul.owns_lock() && sul.mutex() == &rwlock);
}
void Objecter::_assign_command_session(CommandOp *c,
- shunique_lock& sul)
+ shunique_lock<ceph::shared_mutex>& sul)
{
ceph_assert(sul.owns_lock() && sul.mutex() == &rwlock);
if (c->session != s) {
if (c->session) {
OSDSession *cs = c->session;
- OSDSession::unique_lock csl(cs->lock);
+ unique_lock csl(cs->lock);
_session_command_op_remove(c->session, c);
csl.unlock();
}
- OSDSession::unique_lock sl(s->lock);
+ unique_lock sl(s->lock);
_session_command_op_assign(s, c);
}
ldout(cct, 10) << "_send_command " << c->tid << dendl;
ceph_assert(c->session);
ceph_assert(c->session->con);
- MCommand *m = new MCommand(monc->monmap.fsid);
+ auto m = new MCommand(monc->monmap.fsid);
m->cmd = c->cmd;
m->set_data(c->inbl);
m->set_tid(c->tid);
logger->inc(l_osdc_command_send);
}
-int Objecter::command_op_cancel(OSDSession *s, ceph_tid_t tid, int r)
+int Objecter::command_op_cancel(OSDSession *s, ceph_tid_t tid,
+ bs::error_code ec)
{
ceph_assert(initialized);
unique_lock wl(rwlock);
- map<ceph_tid_t, CommandOp*>::iterator it = s->command_ops.find(tid);
+ auto it = s->command_ops.find(tid);
if (it == s->command_ops.end()) {
ldout(cct, 10) << __func__ << " tid " << tid << " dne" << dendl;
return -ENOENT;
CommandOp *op = it->second;
_command_cancel_map_check(op);
- OSDSession::unique_lock sl(op->session->lock);
- _finish_command(op, r, "");
+ unique_lock sl(op->session->lock);
+ _finish_command(op, ec, {}, {});
sl.unlock();
return 0;
}
-void Objecter::_finish_command(CommandOp *c, int r, string rs)
+void Objecter::_finish_command(CommandOp *c, bs::error_code ec,
+ string&& rs, cb::list&& bl)
{
// rwlock is locked unique
// session lock is locked
- ldout(cct, 10) << "_finish_command " << c->tid << " = " << r << " "
+ ldout(cct, 10) << "_finish_command " << c->tid << " = " << ec << " "
<< rs << dendl;
- if (c->prs)
- *c->prs = rs;
+
if (c->onfinish)
- c->onfinish->complete(r);
+ c->onfinish->defer(std::move(c->onfinish), ec, std::move(rs), std::move(bl));
- if (c->ontimeout && r != -ETIMEDOUT)
+ if (c->ontimeout && ec != bs::errc::timed_out)
timer.cancel_event(c->ontimeout);
_session_command_op_remove(c->session, c);
ceph_assert(command_ops.empty());
}
-Objecter::Objecter(CephContext *cct_, Messenger *m, MonClient *mc,
- Finisher *fin,
- double mon_timeout,
- double osd_timeout) :
- Dispatcher(cct_), messenger(m), monc(mc), finisher(fin),
- trace_endpoint("0.0.0.0", 0, "Objecter"),
- osdmap{std::make_unique<OSDMap>()},
- homeless_session(new OSDSession(cct, -1)),
- mon_timeout(ceph::make_timespan(mon_timeout)),
- osd_timeout(ceph::make_timespan(osd_timeout)),
- op_throttle_bytes(cct, "objecter_bytes",
- cct->_conf->objecter_inflight_op_bytes),
- op_throttle_ops(cct, "objecter_ops", cct->_conf->objecter_inflight_ops),
- retry_writes_after_first_reply(cct->_conf->objecter_retry_writes_after_first_reply)
-{}
+Objecter::Objecter(CephContext *cct,
+ Messenger *m, MonClient *mc,
+ boost::asio::io_context& service) :
+ Dispatcher(cct), messenger(m), monc(mc), service(service)
+{
+ mon_timeout = cct->_conf.get_val<std::chrono::seconds>("rados_mon_op_timeout");
+ osd_timeout = cct->_conf.get_val<std::chrono::seconds>("rados_osd_op_timeout");
+}
Objecter::~Objecter()
{
* sending any more operations to OSDs. Use this
* when it is known that the client can't trust
* anything from before this epoch (e.g. due to
- * client blacklist at this epoch).
+ * client blocklist at this epoch).
*/
void Objecter::set_epoch_barrier(epoch_t epoch)
{
return hobject_t::get_max();
}
-struct C_EnumerateReply : public Context {
- ceph::buffer::list bl;
-
- Objecter *objecter;
- hobject_t *next;
- std::list<librados::ListObjectImpl> *result;
+template<typename T>
+struct EnumerationContext {
+ Objecter* objecter;
const hobject_t end;
- const int64_t pool_id;
- Context *on_finish;
+ const cb::list filter;
+ uint32_t max;
+ const object_locator_t oloc;
+ std::vector<T> ls;
+private:
+ fu2::unique_function<void(bs::error_code,
+ std::vector<T>,
+ hobject_t) &&> on_finish;
+public:
+ epoch_t epoch = 0;
+ int budget = -1;
+
+ EnumerationContext(Objecter* objecter,
+ hobject_t end, cb::list filter,
+ uint32_t max, object_locator_t oloc,
+ decltype(on_finish) on_finish)
+ : objecter(objecter), end(std::move(end)), filter(std::move(filter)),
+ max(max), oloc(std::move(oloc)), on_finish(std::move(on_finish)) {}
+
+ void operator()(bs::error_code ec,
+ std::vector<T> v,
+ hobject_t h) && {
+ if (budget >= 0) {
+ objecter->put_op_budget_bytes(budget);
+ budget = -1;
+ }
+
+ std::move(on_finish)(ec, std::move(v), std::move(h));
+ }
+};
- epoch_t epoch;
- int budget;
+template<typename T>
+struct CB_EnumerateReply {
+ cb::list bl;
- C_EnumerateReply(Objecter *objecter_, hobject_t *next_,
- std::list<librados::ListObjectImpl> *result_,
- const hobject_t end_, const int64_t pool_id_, Context *on_finish_) :
- objecter(objecter_), next(next_), result(result_),
- end(end_), pool_id(pool_id_), on_finish(on_finish_),
- epoch(0), budget(-1)
- {}
+ Objecter* objecter;
+ std::unique_ptr<EnumerationContext<T>> ctx;
- void finish(int r) override {
- objecter->_enumerate_reply(
- bl, r, end, pool_id, budget, epoch, result, next, on_finish);
+ CB_EnumerateReply(Objecter* objecter,
+ std::unique_ptr<EnumerationContext<T>>&& ctx) :
+ objecter(objecter), ctx(std::move(ctx)) {}
+
+ void operator()(bs::error_code ec) {
+ objecter->_enumerate_reply(std::move(bl), ec, std::move(ctx));
}
};
+template<typename T>
void Objecter::enumerate_objects(
- int64_t pool_id,
- const std::string &ns,
- const hobject_t &start,
- const hobject_t &end,
- const uint32_t max,
- const ceph::buffer::list &filter_bl,
- std::list<librados::ListObjectImpl> *result,
- hobject_t *next,
- Context *on_finish)
-{
- ceph_assert(result);
-
+ int64_t pool_id,
+ std::string_view ns,
+ hobject_t start,
+ hobject_t end,
+ const uint32_t max,
+ const cb::list& filter_bl,
+ fu2::unique_function<void(bs::error_code,
+ std::vector<T>,
+ hobject_t) &&> on_finish) {
if (!end.is_max() && start > end) {
lderr(cct) << __func__ << ": start " << start << " > end " << end << dendl;
- on_finish->complete(-EINVAL);
+ std::move(on_finish)(osdc_errc::precondition_violated, {}, {});
return;
}
if (max < 1) {
lderr(cct) << __func__ << ": result size may not be zero" << dendl;
- on_finish->complete(-EINVAL);
+ std::move(on_finish)(osdc_errc::precondition_violated, {}, {});
return;
}
if (start.is_max()) {
- on_finish->complete(0);
+ std::move(on_finish)({}, {}, {});
return;
}
if (!osdmap->test_flag(CEPH_OSDMAP_SORTBITWISE)) {
rl.unlock();
lderr(cct) << __func__ << ": SORTBITWISE cluster flag not set" << dendl;
- on_finish->complete(-EOPNOTSUPP);
+ std::move(on_finish)(osdc_errc::not_supported, {}, {});
return;
}
- const pg_pool_t *p = osdmap->get_pg_pool(pool_id);
+ const pg_pool_t* p = osdmap->get_pg_pool(pool_id);
if (!p) {
lderr(cct) << __func__ << ": pool " << pool_id << " DNE in osd epoch "
<< osdmap->get_epoch() << dendl;
rl.unlock();
- on_finish->complete(-ENOENT);
+ std::move(on_finish)(osdc_errc::pool_dne, {}, {});
return;
} else {
rl.unlock();
}
- ldout(cct, 20) << __func__ << ": start=" << start << " end=" << end << dendl;
-
- // Stash completion state
- C_EnumerateReply *on_ack = new C_EnumerateReply(
- this, next, result, end, pool_id, on_finish);
-
+ _issue_enumerate(start,
+ std::make_unique<EnumerationContext<T>>(
+ this, std::move(end), filter_bl,
+ max, object_locator_t{pool_id, ns},
+ std::move(on_finish)));
+}
+
+template
+void Objecter::enumerate_objects<librados::ListObjectImpl>(
+ int64_t pool_id,
+ std::string_view ns,
+ hobject_t start,
+ hobject_t end,
+ const uint32_t max,
+ const cb::list& filter_bl,
+ fu2::unique_function<void(bs::error_code,
+ std::vector<librados::ListObjectImpl>,
+ hobject_t) &&> on_finish);
+
+template
+void Objecter::enumerate_objects<neorados::Entry>(
+ int64_t pool_id,
+ std::string_view ns,
+ hobject_t start,
+ hobject_t end,
+ const uint32_t max,
+ const cb::list& filter_bl,
+ fu2::unique_function<void(bs::error_code,
+ std::vector<neorados::Entry>,
+ hobject_t) &&> on_finish);
+
+
+
+template<typename T>
+void Objecter::_issue_enumerate(hobject_t start,
+ std::unique_ptr<EnumerationContext<T>> ctx) {
ObjectOperation op;
- op.pg_nls(max, filter_bl, start, 0);
+ auto c = ctx.get();
+ op.pg_nls(c->max, c->filter, start, osdmap->get_epoch());
+ auto on_ack = std::make_unique<CB_EnumerateReply<T>>(this, std::move(ctx));
+ // I hate having to do this. Try to find a cleaner way
+ // later.
+ auto epoch = &c->epoch;
+ auto budget = &c->budget;
+ auto pbl = &on_ack->bl;
// Issue. See you later in _enumerate_reply
- object_locator_t oloc(pool_id, ns);
- pg_read(start.get_hash(), oloc, op,
- &on_ack->bl, 0, on_ack, &on_ack->epoch, &on_ack->budget);
-}
-
+ pg_read(start.get_hash(),
+ c->oloc, op, pbl, 0,
+ Op::OpComp::create(service.get_executor(),
+ [c = std::move(on_ack)]
+ (bs::error_code ec) mutable {
+ (*c)(ec);
+ }), epoch, budget);
+}
+
+template
+void Objecter::_issue_enumerate<librados::ListObjectImpl>(
+ hobject_t start,
+ std::unique_ptr<EnumerationContext<librados::ListObjectImpl>> ctx);
+template
+void Objecter::_issue_enumerate<neorados::Entry>(
+ hobject_t start, std::unique_ptr<EnumerationContext<neorados::Entry>> ctx);
+
+template<typename T>
void Objecter::_enumerate_reply(
- ceph::buffer::list &bl,
- int r,
- const hobject_t &end,
- const int64_t pool_id,
- int budget,
- epoch_t reply_epoch,
- std::list<librados::ListObjectImpl> *result,
- hobject_t *next,
- Context *on_finish)
-{
- if (budget >= 0) {
- put_op_budget_bytes(budget);
- }
-
- if (r < 0) {
- ldout(cct, 4) << __func__ << ": remote error " << r << dendl;
- on_finish->complete(r);
+ cb::list&& bl,
+ bs::error_code ec,
+ std::unique_ptr<EnumerationContext<T>>&& ctx)
+{
+ if (ec) {
+ std::move(*ctx)(ec, {}, {});
return;
}
- ceph_assert(next != NULL);
-
// Decode the results
auto iter = bl.cbegin();
- pg_nls_response_t response;
+ pg_nls_response_template<T> response;
+
+ try {
+ response.decode(iter);
+ if (!iter.end()) {
+ // extra_info isn't used anywhere. We do this solely to preserve
+ // backward compatibility
+ cb::list legacy_extra_info;
+ decode(legacy_extra_info, iter);
+ }
+ } catch (const bs::system_error& e) {
+ std::move(*ctx)(e.code(), {}, {});
+ return;
+ }
- decode(response, iter);
- if (!iter.end()) {
- // extra_info isn't used anywhere. We do this solely to preserve
- // backward compatibility
- ceph::buffer::list legacy_extra_info;
- decode(legacy_extra_info, iter);
+ shared_lock rl(rwlock);
+ auto pool = osdmap->get_pg_pool(ctx->oloc.get_pool());
+ rl.unlock();
+ if (!pool) {
+ // pool is gone, drop any results which are now meaningless.
+ std::move(*ctx)(osdc_errc::pool_dne, {}, {});
+ return;
}
- ldout(cct, 10) << __func__ << ": got " << response.entries.size()
- << " handle " << response.handle
- << " reply_epoch " << reply_epoch << dendl;
- ldout(cct, 20) << __func__ << ": response.entries.size "
- << response.entries.size() << ", response.entries "
- << response.entries << dendl;
- if (response.handle <= end) {
- *next = response.handle;
+ hobject_t next;
+ if ((response.handle <= ctx->end)) {
+ next = response.handle;
} else {
- ldout(cct, 10) << __func__ << ": adjusted next down to end " << end
- << dendl;
- *next = end;
+ next = ctx->end;
// drop anything after 'end'
- shared_lock rl(rwlock);
- const pg_pool_t *pool = osdmap->get_pg_pool(pool_id);
- if (!pool) {
- // pool is gone, drop any results which are now meaningless.
- rl.unlock();
- on_finish->complete(-ENOENT);
- return;
- }
while (!response.entries.empty()) {
uint32_t hash = response.entries.back().locator.empty() ?
pool->hash_key(response.entries.back().oid,
response.entries.back().locator,
CEPH_NOSNAP,
hash,
- pool_id,
+ ctx->oloc.get_pool(),
response.entries.back().nspace);
- if (last < end)
+ if (last < ctx->end)
break;
- ldout(cct, 20) << __func__ << " dropping item " << last
- << " >= end " << end << dendl;
response.entries.pop_back();
}
- rl.unlock();
}
- if (!response.entries.empty()) {
- result->merge(response.entries);
+
+ if (response.entries.size() <= ctx->max) {
+ ctx->max -= response.entries.size();
+ std::move(response.entries.begin(), response.entries.end(),
+ std::back_inserter(ctx->ls));
+ } else {
+ auto i = response.entries.begin();
+ while (ctx->max > 0) {
+ ctx->ls.push_back(std::move(*i));
+ --(ctx->max);
+ ++i;
+ }
+ uint32_t hash =
+ i->locator.empty() ?
+ pool->hash_key(i->oid, i->nspace) :
+ pool->hash_key(i->locator, i->nspace);
+
+ next = hobject_t{i->oid, i->locator,
+ CEPH_NOSNAP,
+ hash,
+ ctx->oloc.get_pool(),
+ i->nspace};
}
- // release the listing context's budget once all
- // OPs (in the session) are finished
-#if 0
- put_nlist_context_budget(list_context);
-#endif
- on_finish->complete(r);
- return;
+ if (next == ctx->end || ctx->max == 0) {
+ std::move(*ctx)(ec, std::move(ctx->ls), std::move(next));
+ } else {
+ _issue_enumerate(next, std::move(ctx));
+ }
}
+template
+void Objecter::_enumerate_reply<librados::ListObjectImpl>(
+ cb::list&& bl,
+ bs::error_code ec,
+ std::unique_ptr<EnumerationContext<librados::ListObjectImpl>>&& ctx);
+
+template
+void Objecter::_enumerate_reply<neorados::Entry>(
+ cb::list&& bl,
+ bs::error_code ec,
+ std::unique_ptr<EnumerationContext<neorados::Entry>>&& ctx);
+
namespace {
using namespace librados;
template <typename T>
- void do_decode(std::vector<T>& items, std::vector<ceph::buffer::list>& bls)
+ void do_decode(std::vector<T>& items, std::vector<cb::list>& bls)
{
for (auto bl : bls) {
auto p = bl.cbegin();
}
struct C_ObjectOperation_scrub_ls : public Context {
- ceph::buffer::list bl;
- uint32_t *interval;
+ cb::list bl;
+ uint32_t* interval;
std::vector<inconsistent_obj_t> *objects = nullptr;
std::vector<inconsistent_snapset_t> *snapsets = nullptr;
- int *rval;
+ int* rval;
- C_ObjectOperation_scrub_ls(uint32_t *interval,
- std::vector<inconsistent_obj_t> *objects,
- int *rval)
+ C_ObjectOperation_scrub_ls(uint32_t* interval,
+ std::vector<inconsistent_obj_t>* objects,
+ int* rval)
: interval(interval), objects(objects), rval(rval) {}
- C_ObjectOperation_scrub_ls(uint32_t *interval,
- std::vector<inconsistent_snapset_t> *snapsets,
- int *rval)
+ C_ObjectOperation_scrub_ls(uint32_t* interval,
+ std::vector<inconsistent_snapset_t>* snapsets,
+ int* rval)
: interval(interval), snapsets(snapsets), rval(rval) {}
void finish(int r) override {
if (r < 0 && r != -EAGAIN) {
try {
decode();
- } catch (ceph::buffer::error&) {
+ } catch (cb::error&) {
if (rval)
*rval = -EIO;
}
};
template <typename T>
- void do_scrub_ls(::ObjectOperation *op,
+ void do_scrub_ls(::ObjectOperation* op,
const scrub_ls_arg_t& arg,
std::vector<T> *items,
- uint32_t *interval,
- int *rval)
+ uint32_t* interval,
+ int* rval)
{
OSDOp& osd_op = op->add_op(CEPH_OSD_OP_SCRUBLS);
op->flags |= CEPH_OSD_FLAG_PGOP;
ceph_assert(interval);
arg.encode(osd_op.indata);
unsigned p = op->ops.size() - 1;
- auto *h = new C_ObjectOperation_scrub_ls{interval, items, rval};
- op->out_handler[p] = h;
+ auto h = new C_ObjectOperation_scrub_ls{interval, items, rval};
+ op->set_handler(h);
op->out_bl[p] = &h->bl;
op->out_rval[p] = rval;
}
void ::ObjectOperation::scrub_ls(const librados::object_id_t& start_after,
uint64_t max_to_get,
- std::vector<librados::inconsistent_obj_t> *objects,
- uint32_t *interval,
- int *rval)
+ std::vector<librados::inconsistent_obj_t>* objects,
+ uint32_t* interval,
+ int* rval)
{
scrub_ls_arg_t arg = {*interval, 0, start_after, max_to_get};
do_scrub_ls(this, arg, objects, interval, rval);