X-Git-Url: http://xvm.mit.edu/gitweb/invirt/third/libt4.git/blobdiff_plain/24bebc0ecf83446c7371eff69042322aab34976a..refs/heads/iannucci:/rpc/connection.cc

diff --git a/rpc/connection.cc b/rpc/connection.cc
index 86d4ec5..7a6371a 100644
--- a/rpc/connection.cc
+++ b/rpc/connection.cc
@@ -1,209 +1,153 @@
-// std::bind and syscall bind have the same name, so don't use std::bind in this file
-#define LIBT4_NO_FUNCTIONAL
-#include "connection.h"
-#include <fcntl.h>
-#include <sys/types.h>
+#include "include/rpc/connection.h"
+#include "include/rpc/rpc_protocol.h"
+#include <cerrno>
+#include <csignal>
 #include <netinet/tcp.h>
-#include <errno.h>
-#include <signal.h>
 #include <unistd.h>
-#include <sys/socket.h>
+#include "include/rpc/marshall.h"
 
-#define MAX_PDU (10<<20) //maximum PDF is 10M
+connection_delegate::~connection_delegate() {}
 
-connection::connection(chanmgr *m1, int f1, int l1)
-: mgr_(m1), fd_(f1), lossy_(l1)
+connection::connection(connection_delegate * delegate, socket_t && f1, int l1)
+: fd(std::move(f1)), delegate_(delegate), lossy_(l1)
 {
-    int flags = fcntl(fd_, F_GETFL, NULL);
-    flags |= O_NONBLOCK;
-    fcntl(fd_, F_SETFL, flags);
+    fd.flags() |= O_NONBLOCK;
 
     signal(SIGPIPE, SIG_IGN);
 
-    create_time_ = steady_clock::now();
-
-    PollMgr::Instance()->add_callback(fd_, CB_RDONLY, this);
+    global->shared_mgr.add_callback(fd, CB_RDONLY, this);
 }
 
 connection::~connection() {
-    VERIFY(dead_);
-    if (rpdu_.buf)
-        free(rpdu_.buf);
-    VERIFY(!wpdu_.buf);
-    close(fd_);
-}
-
-void connection::incref() {
-    lock rl(ref_m_);
-    refno_++;
-}
-
-bool connection::isdead() {
-    lock ml(m_);
-    return dead_;
-}
-
-void connection::closeconn() {
     {
         lock ml(m_);
-        if (!dead_) {
-            dead_ = true;
-            shutdown(fd_,SHUT_RDWR);
-        } else {
+        if (dead_)
             return;
-        }
+        dead_ = true;
+        fd.shutdown(SHUT_RDWR);
     }
-    //after block_remove_fd, select will never wait on fd_
-    //and no callbacks will be active
-    PollMgr::Instance()->block_remove_fd(fd_);
+    // after block_remove_fd, select will never wait on fd and no callbacks
+    // will be active
+    global->shared_mgr.block_remove_fd(fd);
+    VERIFY(dead_ && wpdu_.status == unused);
 }
 
-void connection::decref() {
-    bool dead = false;
-    {
-        lock rl(ref_m_);
-        refno_--;
-        VERIFY(refno_>=0);
-        if (refno_==0) {
-            lock ml(m_);
-            dead = dead_;
-        }
+shared_ptr<connection> connection::to_dst(const sockaddr_in & dst, connection_delegate * delegate, int lossy) {
+    socket_t s = socket(AF_INET, SOCK_STREAM, 0);
+    s.setsockopt(IPPROTO_TCP, TCP_NODELAY, (int)1);
+    if (connect(s, (sockaddr*)&dst, sizeof(dst)) < 0) {
+        IF_LEVEL(1) LOG_NONMEMBER << "failed to " << inet_ntoa(dst.sin_addr) << ":" << ntoh(dst.sin_port);
+        close(s);
+        return nullptr;
     }
-    if (dead)
-        delete this;
-}
-
-int connection::compare(connection *another) {
-    if (create_time_ > another->create_time_)
-        return 1;
-    if (create_time_ < another->create_time_)
-        return -1;
-    return 0;
+    IF_LEVEL(2) LOG_NONMEMBER << "connection::to_dst fd=" << s << " to dst " << inet_ntoa(dst.sin_addr) << ":" << ntoh(dst.sin_port);
+    return std::make_shared<connection>(delegate, std::move(s), lossy);
 }
 
-bool connection::send(char *b, size_t sz) {
+bool connection::send(const string & b) {
     lock ml(m_);
-    waiters_++;
-    while (!dead_ && wpdu_.buf) {
+
+    while (!dead_ && wpdu_.status != unused)
         send_wait_.wait(ml);
-    }
-    waiters_--;
-    if (dead_) {
+
+    if (dead_)
         return false;
-    }
-    wpdu_.buf = b;
-    wpdu_.sz = sz;
-    wpdu_.solong = 0;
-
-    if (lossy_) {
-        if ((random()%100) < lossy_) {
-            IF_LEVEL(1) LOG("connection::send LOSSY TEST shutdown fd_ " << fd_);
-            shutdown(fd_,SHUT_RDWR);
-        }
+
+    wpdu_ = {inflight, b, 0};
+
+    if (std::bernoulli_distribution(lossy_*.01)(global->random_generator)) {
+        IF_LEVEL(1) LOG << "send LOSSY TEST shutdown fd " << fd;
+        fd.shutdown(SHUT_RDWR);
     }
 
     if (!writepdu()) {
         dead_ = true;
         ml.unlock();
-        PollMgr::Instance()->block_remove_fd(fd_);
+        global->shared_mgr.block_remove_fd(fd);
         ml.lock();
-    } else {
-        if (wpdu_.solong == wpdu_.sz) {
-        } else {
-            //should be rare to need to explicitly add write callback
-            PollMgr::Instance()->add_callback(fd_, CB_WRONLY, this);
-            while (!dead_ && wpdu_.solong != size_t_max && wpdu_.solong < wpdu_.sz) {
-                send_complete_.wait(ml);
-            }
-        }
-    }
-    bool ret = (!dead_ && wpdu_.solong == wpdu_.sz);
-    wpdu_.solong = wpdu_.sz = 0;
-    wpdu_.buf = NULL;
-    if (waiters_ > 0)
-        send_wait_.notify_all();
+    } else if (wpdu_.status == inflight && wpdu_.cursor < b.size()) {
+        // should be rare to need to explicitly add write callback
+        global->shared_mgr.add_callback(fd, CB_WRONLY, this);
+        while (!dead_ && wpdu_.status == inflight && wpdu_.cursor < b.size())
+            send_complete_.wait(ml);
+    }
+    bool ret = (!dead_ && wpdu_.status == inflight && wpdu_.cursor == b.size());
+    wpdu_ = {unused, "", 0};
+    send_wait_.notify_all();
     return ret;
 }
 
-//fd_ is ready to be written
+// fd is ready to be written
 void connection::write_cb(int s) {
     lock ml(m_);
     VERIFY(!dead_);
-    VERIFY(fd_ == s);
-    if (wpdu_.sz == 0) {
-        PollMgr::Instance()->del_callback(fd_,CB_WRONLY);
+    VERIFY(fd == s);
+    if (wpdu_.status != inflight) {
+        global->shared_mgr.del_callback(fd, CB_WRONLY);
         return;
     }
     if (!writepdu()) {
-        PollMgr::Instance()->del_callback(fd_, CB_RDWR);
+        global->shared_mgr.del_callback(fd, CB_RDWR);
         dead_ = true;
     } else {
-        VERIFY(wpdu_.solong != size_t_max);
-        if (wpdu_.solong < wpdu_.sz) {
+        VERIFY(wpdu_.status != error);
+        if (wpdu_.cursor < wpdu_.buf.size())
             return;
-        }
     }
     send_complete_.notify_one();
 }
 
-//fd_ is ready to be read
+bool connection::writepdu() {
+    VERIFY(wpdu_.status == inflight);
+    if (wpdu_.cursor == wpdu_.buf.size())
+        return true;
+
+    ssize_t n = fd.write(&wpdu_.buf[wpdu_.cursor], (wpdu_.buf.size()-wpdu_.cursor));
+    if (n < 0) {
+        if (errno != EAGAIN) {
+            IF_LEVEL(1) LOG << "writepdu fd " << fd << " failure errno=" << errno;
+            wpdu_ = {error, "", 0};
+        }
+        return (errno == EAGAIN);
+    }
+    wpdu_.cursor += (size_t)n;
+    return true;
+}
+
+// fd is ready to be read
 void connection::read_cb(int s) {
     lock ml(m_);
-    VERIFY(fd_ == s);
-    if (dead_)  {
+    VERIFY(fd == s);
+    if (dead_)
         return;
-    }
 
-    bool succ = true;
-    if (!rpdu_.buf || rpdu_.solong < rpdu_.sz) {
-        succ = readpdu();
-    }
+    IF_LEVEL(5) LOG << "got data on fd " << s;
 
-    if (!succ) {
-        PollMgr::Instance()->del_callback(fd_,CB_RDWR);
-        dead_ = true;
-        send_complete_.notify_one();
-    }
-
-    if (rpdu_.buf && rpdu_.sz == rpdu_.solong) {
-        if (mgr_->got_pdu(this, rpdu_.buf, rpdu_.sz)) {
-            //chanmgr has successfully consumed the pdu
-            rpdu_.buf = NULL;
-            rpdu_.sz = rpdu_.solong = 0;
+    if (rpdu_.status == unused || rpdu_.cursor < rpdu_.buf.size()) {
+        if (!readpdu()) {
+            IF_LEVEL(5) LOG << "readpdu on fd " << s << " failed; dying";
+            global->shared_mgr.del_callback(fd, CB_RDWR);
+            dead_ = true;
+            send_complete_.notify_one();
         }
     }
-}
 
-bool connection::writepdu() {
-    VERIFY(wpdu_.solong != size_t_max);
-    if (wpdu_.solong == wpdu_.sz)
-        return true;
-
-    if (wpdu_.solong == 0) {
-        uint32_t sz = htonl((uint32_t)wpdu_.sz);
-        bcopy(&sz,wpdu_.buf,sizeof(sz));
-    }
-    ssize_t n = write(fd_, wpdu_.buf + wpdu_.solong, (wpdu_.sz-wpdu_.solong));
-    if (n < 0) {
-        if (errno != EAGAIN) {
-            IF_LEVEL(1) LOG("connection::writepdu fd_ " << fd_ << " failure errno=" << errno);
-            wpdu_.solong = size_t_max;
-            wpdu_.sz = 0;
+    if (rpdu_.status == inflight && rpdu_.buf.size() == rpdu_.cursor) {
+        if (delegate_->got_pdu(shared_from_this(), rpdu_.buf)) {
+            // connection_delegate has successfully consumed the pdu
+            rpdu_ = {unused, "", 0};
         }
-        return (errno == EAGAIN);
     }
-    wpdu_.solong += (size_t)n;
-    return true;
 }
 
 bool connection::readpdu() {
-    if (!rpdu_.sz) {
-        uint32_t sz1;
-        ssize_t n = read(fd_, &sz1, sizeof(sz1));
+    IF_LEVEL(5) LOG << "the receive buffer has length " << rpdu_.buf.size();
+    if (rpdu_.status == unused) {
+        rpc_protocol::rpc_sz_t sz1;
+        ssize_t n = fd.read(sz1);
 
-        if (n == 0) {
+        if (n == 0)
             return false;
-        }
 
         if (n < 0) {
             VERIFY(errno!=EAGAIN);
@@ -211,176 +155,89 @@ bool connection::readpdu() {
         }
 
         if (n > 0 && n != sizeof(sz1)) {
-            IF_LEVEL(0) LOG("connection::readpdu short read of sz");
+            IF_LEVEL(0) LOG << "short read of sz";
             return false;
         }
 
-        size_t sz = ntohl(sz1);
+        size_t sz = ntoh(sz1);
 
-        if (sz > MAX_PDU) {
-            IF_LEVEL(2) LOG("connection::readpdu read pdu TOO BIG " << sz << " network order=" << hex << sz1);
+        if (sz > rpc_protocol::MAX_PDU) {
+            IF_LEVEL(2) LOG << "read pdu TOO BIG " << sz << " network order=" << std::hex << sz1;
             return false;
         }
 
-        rpdu_.sz = sz;
-        VERIFY(rpdu_.buf == NULL);
-        rpdu_.buf = (char *)malloc(sz+sizeof(sz1));
-        VERIFY(rpdu_.buf);
-        bcopy(&sz1,rpdu_.buf,sizeof(sz1));
-        rpdu_.solong = sizeof(sz1);
+        IF_LEVEL(5) LOG << "read size of datagram = " << sz;
+
+        rpdu_ = {inflight, string(sz+sizeof(sz1), 0), sizeof(sz1)};
     }
 
-    ssize_t n = read(fd_, rpdu_.buf + rpdu_.solong, rpdu_.sz - rpdu_.solong);
+    ssize_t n = fd.read(&rpdu_.buf[rpdu_.cursor], rpdu_.buf.size() - rpdu_.cursor);
+
+    IF_LEVEL(5) LOG << "read " << n << " bytes";
+
     if (n <= 0) {
         if (errno == EAGAIN)
             return true;
-        if (rpdu_.buf)
-            free(rpdu_.buf);
-        rpdu_.buf = NULL;
-        rpdu_.sz = rpdu_.solong = 0;
-        return (errno == EAGAIN);
+        rpdu_ = {unused, "", 0};
+        return false;
     }
-    rpdu_.solong += (size_t)n;
+    rpdu_.cursor += (size_t)n;
     return true;
 }
 
-tcpsconn::tcpsconn(chanmgr *m1, unsigned int port, int lossytest)
-: mgr_(m1), lossy_(lossytest)
+connection_listener::connection_listener(connection_delegate * delegate, in_port_t port, int lossytest)
+: tcp_(socket(AF_INET, SOCK_STREAM, 0)), delegate_(delegate), lossy_(lossytest)
 {
-    struct sockaddr_in sin;
-    memset(&sin, 0, sizeof(sin));
-    sin.sin_family = AF_INET;
-    sin.sin_port = htons(port);
+    tcp_.setsockopt(SOL_SOCKET, SO_REUSEADDR, int{1});
+    tcp_.setsockopt(IPPROTO_TCP, TCP_NODELAY, int{1});
+    tcp_.setsockopt(SOL_SOCKET, SO_RCVTIMEO, timeval{0, 50000});
+    tcp_.setsockopt(SOL_SOCKET, SO_SNDTIMEO, timeval{0, 50000});
 
-    tcp_ = socket(AF_INET, SOCK_STREAM, 0);
-    if (tcp_ < 0) {
-        perror("tcpsconn::tcpsconn accept_loop socket:");
-        VERIFY(0);
-    }
-
-    int yes = 1;
-    setsockopt(tcp_, SOL_SOCKET, SO_REUSEADDR, &yes, sizeof(yes));
-    setsockopt(tcp_, IPPROTO_TCP, TCP_NODELAY, &yes, sizeof(yes));
+    sockaddr_in sin = sockaddr_in(); // zero initialize
+    sin.sin_family = AF_INET;
+    sin.sin_port = hton(port);
 
     if (bind(tcp_, (sockaddr *)&sin, sizeof(sin)) < 0) {
-        perror("accept_loop tcp bind:");
+        perror("accept_loop bind");
         VERIFY(0);
     }
 
     if (listen(tcp_, 1000) < 0) {
-        perror("tcpsconn::tcpsconn listen:");
+        perror("accept_loop listen");
         VERIFY(0);
     }
 
     socklen_t addrlen = sizeof(sin);
     VERIFY(getsockname(tcp_, (sockaddr *)&sin, &addrlen) == 0);
-    port_ = ntohs(sin.sin_port);
-
-    IF_LEVEL(2) LOG("tcpsconn::tcpsconn listen on " << port_ << " " << sin.sin_port);
-
-    if (pipe(pipe_) < 0) {
-        perror("accept_loop pipe:");
-        VERIFY(0);
-    }
+    port_ = ntoh(sin.sin_port);
 
-    int flags = fcntl(pipe_[0], F_GETFL, NULL);
-    flags |= O_NONBLOCK;
-    fcntl(pipe_[0], F_SETFL, flags);
+    IF_LEVEL(2) LOG << "listen on " << port_ << " " << sin.sin_port;
 
-    th_ = thread(&tcpsconn::accept_conn, this);
+    global->shared_mgr.add_callback(tcp_, CB_RDONLY, this);
 }
 
-tcpsconn::~tcpsconn()
-{
-    VERIFY(close(pipe_[1]) == 0);
-    th_.join();
-
-    //close all the active connections
-    map<int, connection *>::iterator i;
-    for (i = conns_.begin(); i != conns_.end(); i++) {
-        i->second->closeconn();
-        i->second->decref();
-    }
+connection_listener::~connection_listener() {
+    global->shared_mgr.block_remove_fd(tcp_);
 }
 
-void tcpsconn::process_accept() {
+void connection_listener::read_cb(int) {
     sockaddr_in sin;
     socklen_t slen = sizeof(sin);
     int s1 = accept(tcp_, (sockaddr *)&sin, &slen);
     if (s1 < 0) {
-        perror("tcpsconn::accept_conn error");
-        throw thread_exit_exception();
+        perror("connection_listener::accept_conn error");
+        throw std::runtime_error("connection listener failure");
     }
 
-    IF_LEVEL(2) LOG("accept_loop got connection fd=" << s1 << " " << inet_ntoa(sin.sin_addr) << ":" << ntohs(sin.sin_port));
-    connection *ch = new connection(mgr_, s1, lossy_);
+    IF_LEVEL(2) LOG << "accept_loop got connection fd=" << s1 << " " << inet_ntoa(sin.sin_addr) << ":" << ntoh(sin.sin_port);
 
-    // garbage collect all dead connections with refcount of 1
+    // garbage collect dead connections
     for (auto i = conns_.begin(); i != conns_.end();) {
-        if (i->second->isdead() && i->second->ref() == 1) {
-            IF_LEVEL(2) LOG("accept_loop garbage collected fd=" << i->second->channo());
-            i->second->decref();
-            // Careful not to reuse i right after erase. (i++) will
-            // be evaluated before the erase call because in C++,
-            // there is a sequence point before a function call.
-            // See http://en.wikipedia.org/wiki/Sequence_point.
+        if (i->second->isdead())
             conns_.erase(i++);
-        } else
+        else
             ++i;
     }
 
-    conns_[ch->channo()] = ch;
-}
-
-void tcpsconn::accept_conn() {
-    fd_set rfds;
-    int max_fd = pipe_[0] > tcp_ ? pipe_[0] : tcp_;
-
-    try {
-        while (1) {
-            FD_ZERO(&rfds);
-            FD_SET(pipe_[0], &rfds);
-            FD_SET(tcp_, &rfds);
-
-            int ret = select(max_fd+1, &rfds, NULL, NULL, NULL);
-
-            if (ret < 0) {
-                if (errno == EINTR) {
-                    continue;
-                } else {
-                    perror("accept_conn select:");
-                    IF_LEVEL(0) LOG("tcpsconn::accept_conn failure errno " << errno);
-                    VERIFY(0);
-                }
-            }
-
-            if (FD_ISSET(pipe_[0], &rfds)) {
-                close(pipe_[0]);
-                close(tcp_);
-                return;
-            }
-            else if (FD_ISSET(tcp_, &rfds)) {
-                process_accept();
-            } else {
-                VERIFY(0);
-            }
-        }
-    }
-    catch (thread_exit_exception e)
-    {
-    }
-}
-
-connection * connect_to_dst(const sockaddr_in &dst, chanmgr *mgr, int lossy) {
-    int s = socket(AF_INET, SOCK_STREAM, 0);
-    int yes = 1;
-    setsockopt(s, IPPROTO_TCP, TCP_NODELAY, &yes, sizeof(yes));
-    if (connect(s, (sockaddr*)&dst, sizeof(dst)) < 0) {
-        IF_LEVEL(1) LOG_NONMEMBER("rpcc::connect_to_dst failed to " << inet_ntoa(dst.sin_addr) << ":" << ntohs(dst.sin_port));
-        close(s);
-        return NULL;
-    }
-    IF_LEVEL(2) LOG_NONMEMBER("connect_to_dst fd=" << s << " to dst " << inet_ntoa(dst.sin_addr) << ":" << ntohs(dst.sin_port));
-    return new connection(mgr, s, lossy);
+    conns_[s1] = std::make_shared<connection>(delegate_, s1, lossy_);
 }
-