You cannot select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
srs/trunk/3rdparty/srt-1-fit/srtcore/epoll.cpp

728 lines
23 KiB
C++

/*
* SRT - Secure, Reliable, Transport
* Copyright (c) 2018 Haivision Systems Inc.
*
* This Source Code Form is subject to the terms of the Mozilla Public
* License, v. 2.0. If a copy of the MPL was not distributed with this
* file, You can obtain one at http://mozilla.org/MPL/2.0/.
*
*/
/*****************************************************************************
Copyright (c) 2001 - 2011, The Board of Trustees of the University of Illinois.
All rights reserved.
Redistribution and use in source and binary forms, with or without
modification, are permitted provided that the following conditions are
met:
* Redistributions of source code must retain the above
copyright notice, this list of conditions and the
following disclaimer.
* Redistributions in binary form must reproduce the
above copyright notice, this list of conditions
and the following disclaimer in the documentation
and/or other materials provided with the distribution.
* Neither the name of the University of Illinois
nor the names of its contributors may be used to
endorse or promote products derived from this
software without specific prior written permission.
THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS
IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO,
THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR
CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*****************************************************************************/
/*****************************************************************************
written by
Yunhong Gu, last updated 01/01/2011
modified by
Haivision Systems Inc.
*****************************************************************************/
#ifdef LINUX
#include <sys/epoll.h>
#include <unistd.h>
#endif
#if __APPLE__
#include "TargetConditionals.h"
#endif
#if defined(BSD) || defined(OSX) || (TARGET_OS_IOS == 1) || (TARGET_OS_TV == 1)
#include <sys/types.h>
#include <sys/event.h>
#include <sys/time.h>
#include <unistd.h>
#endif
#if defined(__ANDROID__) || defined(ANDROID)
#include <sys/select.h>
#endif
#include <algorithm>
#include <cerrno>
#include <cstring>
#include <iterator>
#include "common.h"
#include "epoll.h"
#include "logging.h"
#include "udt.h"
using namespace std;
namespace srt_logging
{
extern Logger mglog;
}
using namespace srt_logging;
#if ENABLE_HEAVY_LOGGING
#define IF_DIRNAME(tested, flag, name) (tested & flag ? name : "")
#endif
CEPoll::CEPoll():
m_iIDSeed(0)
{
CGuard::createMutex(m_EPollLock);
}
CEPoll::~CEPoll()
{
CGuard::releaseMutex(m_EPollLock);
}
int CEPoll::create()
{
CGuard pg(m_EPollLock);
if (++ m_iIDSeed >= 0x7FFFFFFF)
m_iIDSeed = 0;
// Check if an item already exists. Should not ever happen.
if (m_mPolls.find(m_iIDSeed) != m_mPolls.end())
throw CUDTException(MJ_SETUP, MN_NONE);
int localid = 0;
#ifdef LINUX
localid = epoll_create(1024);
/* Possible reasons of -1 error:
EMFILE: The per-user limit on the number of epoll instances imposed by /proc/sys/fs/epoll/max_user_instances was encountered.
ENFILE: The system limit on the total number of open files has been reached.
ENOMEM: There was insufficient memory to create the kernel object.
*/
if (localid < 0)
throw CUDTException(MJ_SETUP, MN_NONE, errno);
#elif defined(BSD) || defined(OSX) || (TARGET_OS_IOS == 1) || (TARGET_OS_TV == 1)
localid = kqueue();
if (localid < 0)
throw CUDTException(MJ_SETUP, MN_NONE, errno);
#else
// on Solaris, use /dev/poll
// on Windows, select
#endif
pair<map<int, CEPollDesc>::iterator, bool> res = m_mPolls.insert(make_pair(m_iIDSeed, CEPollDesc(m_iIDSeed, localid)));
if (!res.second) // Insertion failed (no memory?)
throw CUDTException(MJ_SETUP, MN_NONE);
return m_iIDSeed;
}
int CEPoll::add_ssock(const int eid, const SYSSOCKET& s, const int* events)
{
CGuard pg(m_EPollLock);
map<int, CEPollDesc>::iterator p = m_mPolls.find(eid);
if (p == m_mPolls.end())
throw CUDTException(MJ_NOTSUP, MN_EIDINVAL);
#ifdef LINUX
epoll_event ev;
memset(&ev, 0, sizeof(epoll_event));
if (NULL == events)
ev.events = EPOLLIN | EPOLLOUT | EPOLLERR;
else
{
ev.events = 0;
if (*events & UDT_EPOLL_IN)
ev.events |= EPOLLIN;
if (*events & UDT_EPOLL_OUT)
ev.events |= EPOLLOUT;
if (*events & UDT_EPOLL_ERR)
ev.events |= EPOLLERR;
}
ev.data.fd = s;
if (::epoll_ctl(p->second.m_iLocalID, EPOLL_CTL_ADD, s, &ev) < 0)
throw CUDTException();
#elif defined(BSD) || defined(OSX) || (TARGET_OS_IOS == 1) || (TARGET_OS_TV == 1)
struct kevent ke[2];
int num = 0;
if (NULL == events)
{
EV_SET(&ke[num++], s, EVFILT_READ, EV_ADD, 0, 0, NULL);
EV_SET(&ke[num++], s, EVFILT_WRITE, EV_ADD, 0, 0, NULL);
}
else
{
if (*events & UDT_EPOLL_IN)
{
EV_SET(&ke[num++], s, EVFILT_READ, EV_ADD, 0, 0, NULL);
}
if (*events & UDT_EPOLL_OUT)
{
EV_SET(&ke[num++], s, EVFILT_WRITE, EV_ADD, 0, 0, NULL);
}
}
if (kevent(p->second.m_iLocalID, ke, num, NULL, 0, NULL) < 0)
throw CUDTException();
#else
#ifdef _MSC_VER
// Microsoft Visual Studio doesn't support the #warning directive - nonstandard anyway.
// Use #pragma message with the same text.
// All other compilers should be ok :)
#pragma message("WARNING: Unsupported system for epoll. The epoll_add_ssock() API call won't work on this platform.")
#else
#warning "Unsupported system for epoll. The epoll_add_ssock() API call won't work on this platform."
#endif
#endif
p->second.m_sLocals.insert(s);
return 0;
}
int CEPoll::remove_ssock(const int eid, const SYSSOCKET& s)
{
CGuard pg(m_EPollLock);
map<int, CEPollDesc>::iterator p = m_mPolls.find(eid);
if (p == m_mPolls.end())
throw CUDTException(MJ_NOTSUP, MN_EIDINVAL);
#ifdef LINUX
epoll_event ev; // ev is ignored, for compatibility with old Linux kernel only.
if (::epoll_ctl(p->second.m_iLocalID, EPOLL_CTL_DEL, s, &ev) < 0)
throw CUDTException();
#elif defined(BSD) || defined(OSX) || (TARGET_OS_IOS == 1) || (TARGET_OS_TV == 1)
struct kevent ke;
//
// Since I don't know what was set before
// Just clear out both read and write
//
EV_SET(&ke, s, EVFILT_READ, EV_DELETE, 0, 0, NULL);
kevent(p->second.m_iLocalID, &ke, 1, NULL, 0, NULL);
EV_SET(&ke, s, EVFILT_WRITE, EV_DELETE, 0, 0, NULL);
kevent(p->second.m_iLocalID, &ke, 1, NULL, 0, NULL);
#endif
p->second.m_sLocals.erase(s);
return 0;
}
// Need this to atomically modify polled events (ex: remove write/keep read)
int CEPoll::update_usock(const int eid, const SRTSOCKET& u, const int* events)
{
CGuard pg(m_EPollLock);
map<int, CEPollDesc>::iterator p = m_mPolls.find(eid);
if (p == m_mPolls.end())
throw CUDTException(MJ_NOTSUP, MN_EIDINVAL);
CEPollDesc& d = p->second;
int32_t evts = events ? *events : uint32_t(SRT_EPOLL_IN | SRT_EPOLL_OUT | SRT_EPOLL_ERR);
bool edgeTriggered = evts & SRT_EPOLL_ET;
evts &= ~SRT_EPOLL_ET;
if (evts)
{
pair<CEPollDesc::ewatch_t::iterator, bool> iter_new = d.addWatch(u, evts, edgeTriggered);
CEPollDesc::Wait& wait = iter_new.first->second;
if (!iter_new.second)
{
// The object exists. We only are certain about the `u`
// parameter, but others are probably unchanged. Change them
// forcefully and take out notices that are no longer valid.
const int removable = wait.watch & ~evts;
// Check if there are any events that would be removed.
// If there are no removed events watched (for example, when
// only new events are being added to existing socket),
// there's nothing to remove, but might be something to update.
if (removable)
{
d.removeExcessEvents(wait, evts);
}
// Update the watch configuration, including edge
wait.watch = evts;
if (edgeTriggered)
wait.edge = evts;
// Now it should look exactly like newly added
// and the state is also updated
}
const int newstate = wait.watch & wait.state;
if (newstate)
{
d.addEventNotice(wait, u, newstate);
}
}
else if (edgeTriggered)
{
// Specified only SRT_EPOLL_ET flag, but no event flag. Error.
throw CUDTException(MJ_NOTSUP, MN_INVAL);
}
else
{
// Update with no events means to remove subscription
d.removeSubscription(u);
}
return 0;
}
int CEPoll::update_ssock(const int eid, const SYSSOCKET& s, const int* events)
{
CGuard pg(m_EPollLock);
map<int, CEPollDesc>::iterator p = m_mPolls.find(eid);
if (p == m_mPolls.end())
throw CUDTException(MJ_NOTSUP, MN_EIDINVAL);
#ifdef LINUX
epoll_event ev;
memset(&ev, 0, sizeof(epoll_event));
if (NULL == events)
ev.events = EPOLLIN | EPOLLOUT | EPOLLERR;
else
{
ev.events = 0;
if (*events & UDT_EPOLL_IN)
ev.events |= EPOLLIN;
if (*events & UDT_EPOLL_OUT)
ev.events |= EPOLLOUT;
if (*events & UDT_EPOLL_ERR)
ev.events |= EPOLLERR;
}
ev.data.fd = s;
if (::epoll_ctl(p->second.m_iLocalID, EPOLL_CTL_MOD, s, &ev) < 0)
throw CUDTException();
#elif defined(BSD) || defined(OSX) || (TARGET_OS_IOS == 1) || (TARGET_OS_TV == 1)
struct kevent ke[2];
int num = 0;
//
// Since I don't know what was set before
// Just clear out both read and write
//
EV_SET(&ke[0], s, EVFILT_READ, EV_DELETE, 0, 0, NULL);
kevent(p->second.m_iLocalID, ke, 1, NULL, 0, NULL);
EV_SET(&ke[0], s, EVFILT_WRITE, EV_DELETE, 0, 0, NULL);
kevent(p->second.m_iLocalID, ke, 1, NULL, 0, NULL);
if (NULL == events)
{
EV_SET(&ke[num++], s, EVFILT_READ, EV_ADD, 0, 0, NULL);
EV_SET(&ke[num++], s, EVFILT_WRITE, EV_ADD, 0, 0, NULL);
}
else
{
if (*events & UDT_EPOLL_IN)
{
EV_SET(&ke[num++], s, EVFILT_READ, EV_ADD, 0, 0, NULL);
}
if (*events & UDT_EPOLL_OUT)
{
EV_SET(&ke[num++], s, EVFILT_WRITE, EV_ADD, 0, 0, NULL);
}
}
if (kevent(p->second.m_iLocalID, ke, num, NULL, 0, NULL) < 0)
throw CUDTException();
#endif
// Assuming add is used if not inserted
// p->second.m_sLocals.insert(s);
return 0;
}
int CEPoll::setflags(const int eid, int32_t flags)
{
CGuard pg(m_EPollLock);
map<int, CEPollDesc>::iterator p = m_mPolls.find(eid);
if (p == m_mPolls.end())
throw CUDTException(MJ_NOTSUP, MN_EIDINVAL);
CEPollDesc& ed = p->second;
int32_t oflags = ed.flags();
if (flags == -1)
return oflags;
if (flags == 0)
{
ed.clr_flags(~int32_t());
}
else
{
ed.set_flags(flags);
}
return oflags;
}
int CEPoll::uwait(const int eid, SRT_EPOLL_EVENT* fdsSet, int fdsSize, int64_t msTimeOut)
{
// It is allowed to call this function witn fdsSize == 0
// and therefore also NULL fdsSet. This will then only report
// the number of ready sockets, just without information which.
if (fdsSize < 0 || (fdsSize > 0 && !fdsSet))
throw CUDTException(MJ_NOTSUP, MN_INVAL);
int64_t entertime = CTimer::getTime();
while (true)
{
{
CGuard pg(m_EPollLock);
map<int, CEPollDesc>::iterator p = m_mPolls.find(eid);
if (p == m_mPolls.end())
throw CUDTException(MJ_NOTSUP, MN_EIDINVAL);
CEPollDesc& ed = p->second;
if (!ed.flags(SRT_EPOLL_ENABLE_EMPTY) && ed.watch_empty())
{
// Empty EID is not allowed, report error.
throw CUDTException(MJ_NOTSUP, MN_INVAL);
}
if (ed.flags(SRT_EPOLL_ENABLE_OUTPUTCHECK) && (fdsSet == NULL || fdsSize == 0))
{
// Empty EID is not allowed, report error.
throw CUDTException(MJ_NOTSUP, MN_INVAL);
}
if (!ed.m_sLocals.empty())
{
// XXX Add error log
// uwait should not be used with EIDs subscribed to system sockets
throw CUDTException(MJ_NOTSUP, MN_INVAL);
}
int total = 0; // This is a list, so count it during iteration
CEPollDesc::enotice_t::iterator i = ed.enotice_begin();
while (i != ed.enotice_end())
{
int pos = total; // previous past-the-end position
++total;
if (total > fdsSize)
break;
fdsSet[pos] = *i;
ed.checkEdge(i++); // NOTE: potentially deletes `i`
}
if (total)
return total;
}
if ((msTimeOut >= 0) && (int64_t(CTimer::getTime() - entertime) >= msTimeOut * int64_t(1000)))
break; // official wait does: throw CUDTException(MJ_AGAIN, MN_XMTIMEOUT, 0);
CTimer::waitForEvent();
}
return 0;
}
int CEPoll::wait(const int eid, set<SRTSOCKET>* readfds, set<SRTSOCKET>* writefds, int64_t msTimeOut, set<SYSSOCKET>* lrfds, set<SYSSOCKET>* lwfds)
{
// if all fields is NULL and waiting time is infinite, then this would be a deadlock
if (!readfds && !writefds && !lrfds && !lwfds && (msTimeOut < 0))
throw CUDTException(MJ_NOTSUP, MN_INVAL, 0);
// Clear these sets in case the app forget to do it.
if (readfds) readfds->clear();
if (writefds) writefds->clear();
if (lrfds) lrfds->clear();
if (lwfds) lwfds->clear();
int total = 0;
int64_t entertime = CTimer::getTime();
HLOGC(mglog.Debug, log << "CEPoll::wait: START for eid=" << eid);
while (true)
{
{
CGuard epollock(m_EPollLock);
map<int, CEPollDesc>::iterator p = m_mPolls.find(eid);
if (p == m_mPolls.end())
{
throw CUDTException(MJ_NOTSUP, MN_EIDINVAL);
}
CEPollDesc& ed = p->second;
if (!ed.flags(SRT_EPOLL_ENABLE_EMPTY) && ed.watch_empty() && ed.m_sLocals.empty())
{
// Empty EID is not allowed, report error.
throw CUDTException(MJ_NOTSUP, MN_INVAL);
}
if (ed.flags(SRT_EPOLL_ENABLE_OUTPUTCHECK))
{
// Empty report is not allowed, report error.
if (!ed.m_sLocals.empty() && (!lrfds || !lwfds))
throw CUDTException(MJ_NOTSUP, MN_INVAL);
if (!ed.watch_empty() && (!readfds || !writefds))
throw CUDTException(MJ_NOTSUP, MN_INVAL);
}
IF_HEAVY_LOGGING(int total_noticed = 0);
IF_HEAVY_LOGGING(ostringstream debug_sockets);
// Sockets with exceptions are returned to both read and write sets.
for (CEPollDesc::enotice_t::iterator it = ed.enotice_begin(), it_next = it; it != ed.enotice_end(); it = it_next)
{
++it_next;
IF_HEAVY_LOGGING(++total_noticed);
if (readfds && ((it->events & UDT_EPOLL_IN) || (it->events & UDT_EPOLL_ERR)))
{
if (readfds->insert(it->fd).second)
++total;
}
if (writefds && ((it->events & UDT_EPOLL_OUT) || (it->events & UDT_EPOLL_ERR)))
{
if (writefds->insert(it->fd).second)
++total;
}
IF_HEAVY_LOGGING(debug_sockets << " " << it->fd << ":"
<< IF_DIRNAME(it->events, SRT_EPOLL_IN, "R")
<< IF_DIRNAME(it->events, SRT_EPOLL_OUT, "W")
<< IF_DIRNAME(it->events, SRT_EPOLL_ERR, "E"));
if (ed.checkEdge(it)) // NOTE: potentially erases 'it'.
{
IF_HEAVY_LOGGING(debug_sockets << "!");
}
}
HLOGC(mglog.Debug, log << "CEPoll::wait: REPORTED " << total << "/" << total_noticed
<< debug_sockets.str());
if (lrfds || lwfds)
{
#ifdef LINUX
const int max_events = ed.m_sLocals.size();
epoll_event ev[max_events];
int nfds = ::epoll_wait(ed.m_iLocalID, ev, max_events, 0);
IF_HEAVY_LOGGING(const int prev_total = total);
for (int i = 0; i < nfds; ++ i)
{
if ((NULL != lrfds) && (ev[i].events & EPOLLIN))
{
lrfds->insert(ev[i].data.fd);
++ total;
}
if ((NULL != lwfds) && (ev[i].events & EPOLLOUT))
{
lwfds->insert(ev[i].data.fd);
++ total;
}
}
HLOGC(mglog.Debug, log << "CEPoll::wait: LINUX: picking up " << (total - prev_total) << " ready fds.");
#elif defined(BSD) || defined(OSX) || (TARGET_OS_IOS == 1) || (TARGET_OS_TV == 1)
struct timespec tmout = {0, 0};
const int max_events = ed.m_sLocals.size();
struct kevent ke[max_events];
int nfds = kevent(ed.m_iLocalID, NULL, 0, ke, max_events, &tmout);
IF_HEAVY_LOGGING(const int prev_total = total);
for (int i = 0; i < nfds; ++ i)
{
if ((NULL != lrfds) && (ke[i].filter == EVFILT_READ))
{
lrfds->insert(ke[i].ident);
++ total;
}
if ((NULL != lwfds) && (ke[i].filter == EVFILT_WRITE))
{
lwfds->insert(ke[i].ident);
++ total;
}
}
HLOGC(mglog.Debug, log << "CEPoll::wait: Darwin/BSD: picking up " << (total - prev_total) << " ready fds.");
#else
//currently "select" is used for all non-Linux platforms.
//faster approaches can be applied for specific systems in the future.
//"select" has a limitation on the number of sockets
int max_fd = 0;
fd_set rqreadfds;
fd_set rqwritefds;
FD_ZERO(&rqreadfds);
FD_ZERO(&rqwritefds);
for (set<SYSSOCKET>::const_iterator i = ed.m_sLocals.begin(); i != ed.m_sLocals.end(); ++ i)
{
if (lrfds)
FD_SET(*i, &rqreadfds);
if (lwfds)
FD_SET(*i, &rqwritefds);
if ((int)*i > max_fd)
max_fd = *i;
}
IF_HEAVY_LOGGING(const int prev_total = total);
timeval tv;
tv.tv_sec = 0;
tv.tv_usec = 0;
if (::select(max_fd + 1, &rqreadfds, &rqwritefds, NULL, &tv) > 0)
{
for (set<SYSSOCKET>::const_iterator i = ed.m_sLocals.begin(); i != ed.m_sLocals.end(); ++ i)
{
if (lrfds && FD_ISSET(*i, &rqreadfds))
{
lrfds->insert(*i);
++ total;
}
if (lwfds && FD_ISSET(*i, &rqwritefds))
{
lwfds->insert(*i);
++ total;
}
}
}
HLOGC(mglog.Debug, log << "CEPoll::wait: select(otherSYS): picking up " << (total - prev_total) << " ready fds.");
#endif
}
} // END-LOCK: m_EPollLock
HLOGC(mglog.Debug, log << "CEPoll::wait: Total of " << total << " READY SOCKETS");
if (total > 0)
return total;
if ((msTimeOut >= 0) && (int64_t(CTimer::getTime() - entertime) >= msTimeOut * int64_t(1000)))
{
HLOGP(mglog.Debug, "... not waiting longer - timeout");
throw CUDTException(MJ_AGAIN, MN_XMTIMEOUT, 0);
}
CTimer::EWait wt ATR_UNUSED = CTimer::waitForEvent();
HLOGC(mglog.Debug, log << "CEPoll::wait: EVENT WAITING: "
<< (wt == CTimer::WT_TIMEOUT ? "CHECKPOINT" : wt == CTimer::WT_EVENT ? "TRIGGERED" : "ERROR"));
}
return 0;
}
int CEPoll::release(const int eid)
{
CGuard pg(m_EPollLock);
map<int, CEPollDesc>::iterator i = m_mPolls.find(eid);
if (i == m_mPolls.end())
throw CUDTException(MJ_NOTSUP, MN_EIDINVAL);
#ifdef LINUX
// release local/system epoll descriptor
::close(i->second.m_iLocalID);
#elif defined(BSD) || defined(OSX) || (TARGET_OS_IOS == 1) || (TARGET_OS_TV == 1)
::close(i->second.m_iLocalID);
#endif
m_mPolls.erase(i);
return 0;
}
int CEPoll::update_events(const SRTSOCKET& uid, std::set<int>& eids, const int events, const bool enable)
{
vector<int> lost;
CGuard pg(m_EPollLock);
for (set<int>::iterator i = eids.begin(); i != eids.end(); ++ i)
{
map<int, CEPollDesc>::iterator p = m_mPolls.find(*i);
if (p == m_mPolls.end())
{
// EID invalid, though still present in the socket's subscriber list
// (dangling in the socket). Postpone to fix the subscruption and continue.
lost.push_back(*i);
continue;
}
CEPollDesc& ed = p->second;
// Check if this EID is subscribed for this socket.
CEPollDesc::Wait* pwait = ed.watch_find(uid);
if (!pwait)
{
// As this is mapped in the socket's data, it should be impossible.
continue;
}
// compute new states
// New state to be set into the permanent state
const int newstate = enable ? pwait->state | events // SET event bits if enable
: pwait->state & (~events); // CLEAR event bits
// compute states changes!
int changes = pwait->state ^ newstate; // oldState XOR newState
if (!changes)
continue; // no changes!
// assign new state
pwait->state = newstate;
// filter change relating what is watching
changes &= pwait->watch;
if (!changes)
continue; // no change watching
// set events changes!
// This function will update the notice object associated with
// the given events, that is:
// - if enable, it will set event flags, possibly in a new notice object
// - if !enable, it will clear event flags, possibly remove notice if resulted in 0
ed.updateEventNotice(*pwait, uid, events, enable);
}
for (vector<int>::iterator i = lost.begin(); i != lost.end(); ++ i)
eids.erase(*i);
return 0;
}