src/comm.cc

   1 /*
   2  * Copyright (C) 1996-2016 The Squid Software Foundation and contributors
   3  *
   4  * Squid software is distributed under GPLv2+ license and includes
   5  * contributions from numerous individuals and organizations.
   6  * Please see the COPYING and CONTRIBUTORS files for details.
   7  */
   8
   9 /* DEBUG: section 05    Socket Functions */
  10
  11 #include "squid.h"
  12 #include "ClientInfo.h"
  13 #include "comm/AcceptLimiter.h"
  14 #include "comm/comm_internal.h"
  15 #include "comm/Connection.h"
  16 #include "comm/IoCallback.h"
  17 #include "comm/Loops.h"
  18 #include "comm/Read.h"
  19 #include "comm/TcpAcceptor.h"
  20 #include "comm/Write.h"
  21 #include "CommRead.h"
  22 #include "compat/cmsg.h"
  23 #include "DescriptorSet.h"
  24 #include "event.h"
  25 #include "fd.h"
  26 #include "fde.h"
  27 #include "globals.h"
  28 #include "icmp/net_db.h"
  29 #include "ip/Intercept.h"
  30 #include "ip/QosConfig.h"
  31 #include "ip/tools.h"
  32 #include "pconn.h"
  33 #include "profiler/Profiler.h"
  34 #include "sbuf/SBuf.h"
  35 #include "SquidConfig.h"
  36 #include "StatCounters.h"
  37 #include "StoreIOBuffer.h"
  38 #include "tools.h"
  39
  40 #if USE_OPENSSL
  41 #include "ssl/support.h"
  42 #endif
  43
  44 #include <cerrno>
  45 #include <cmath>
  46 #if _SQUID_CYGWIN_
  47 #include <sys/ioctl.h>
  48 #endif
  49 #ifdef HAVE_NETINET_TCP_H
  50 #include <netinet/tcp.h>
  51 #endif
  52 #if HAVE_SYS_UN_H
  53 #include <sys/un.h>
  54 #endif
  55
  56 /*
  57  * New C-like simple comm code. This stuff is a mess and doesn't really buy us anything.
  58  */
  59
  60 static IOCB commHalfClosedReader;
  61 static void comm_init_opened(const Comm::ConnectionPointer &conn, const char *note, struct addrinfo *AI);
  62 static int comm_apply_flags(int new_socket, Ip::Address &addr, int flags, struct addrinfo *AI);
  63
  64 #if USE_DELAY_POOLS
  65 CBDATA_CLASS_INIT(CommQuotaQueue);
  66
  67 static void commHandleWriteHelper(void * data);
  68 #endif
  69
  70 /* STATIC */
  71
  72 static DescriptorSet *TheHalfClosed = NULL; /// the set of half-closed FDs
  73 static bool WillCheckHalfClosed = false; /// true if check is scheduled
  74 static EVH commHalfClosedCheck;
  75 static void commPlanHalfClosedCheck();
  76
  77 static Comm::Flag commBind(int s, struct addrinfo &);
  78 static void commSetReuseAddr(int);
  79 static void commSetNoLinger(int);
  80 #ifdef TCP_NODELAY
  81 static void commSetTcpNoDelay(int);
  82 #endif
  83 static void commSetTcpRcvbuf(int, int);
  84
  85 fd_debug_t *fdd_table = NULL;
  86
  87 bool
  88 isOpen(const int fd)
  89 {
  90     return fd >= 0 && fd_table && fd_table[fd].flags.open != 0;
  91 }
  92
  93 /**
  94  * Empty the read buffers
  95  *
  96  * This is a magical routine that empties the read buffers.
  97  * Under some platforms (Linux) if a buffer has data in it before
  98  * you call close(), the socket will hang and take quite a while
  99  * to timeout.
 100  */
 101 static void
 102 comm_empty_os_read_buffers(int fd)
 103 {
 104 #if _SQUID_LINUX_
 105 #if USE_OPENSSL
 106     // Bug 4146: SSL-Bump BIO does not release sockets on close.
 107     if (fd_table[fd].ssl)
 108         return;
 109 #endif
 110
 111     /* prevent those nasty RST packets */
 112     char buf[SQUID_TCP_SO_RCVBUF];
 113     if (fd_table[fd].flags.nonblocking) {
 114         while (FD_READ_METHOD(fd, buf, SQUID_TCP_SO_RCVBUF) > 0) {};
 115     }
 116 #endif
 117 }
 118
 119 /**
 120  * synchronous wrapper around udp socket functions
 121  */
 122 int
 123 comm_udp_recvfrom(int fd, void *buf, size_t len, int flags, Ip::Address &from)
 124 {
 125     ++ statCounter.syscalls.sock.recvfroms;
 126     debugs(5,8, "comm_udp_recvfrom: FD " << fd << " from " << from);
 127     struct addrinfo *AI = NULL;
 128     Ip::Address::InitAddr(AI);
 129     int x = recvfrom(fd, buf, len, flags, AI->ai_addr, &AI->ai_addrlen);
 130     from = *AI;
 131     Ip::Address::FreeAddr(AI);
 132     return x;
 133 }
 134
 135 int
 136 comm_udp_recv(int fd, void *buf, size_t len, int flags)
 137 {
 138     Ip::Address nul;
 139     return comm_udp_recvfrom(fd, buf, len, flags, nul);
 140 }
 141
 142 ssize_t
 143 comm_udp_send(int s, const void *buf, size_t len, int flags)
 144 {
 145     return send(s, buf, len, flags);
 146 }
 147
 148 bool
 149 comm_has_incomplete_write(int fd)
 150 {
 151     assert(isOpen(fd) && COMMIO_FD_WRITECB(fd) != NULL);
 152     return COMMIO_FD_WRITECB(fd)->active();
 153 }
 154
 155 /**
 156  * Queue a write. handler/handler_data are called when the write fully
 157  * completes, on error, or on file descriptor close.
 158  */
 159
 160 /* Return the local port associated with fd. */
 161 unsigned short
 162 comm_local_port(int fd)
 163 {
 164     Ip::Address temp;
 165     struct addrinfo *addr = NULL;
 166     fde *F = &fd_table[fd];
 167
 168     /* If the fd is closed already, just return */
 169
 170     if (!F->flags.open) {
 171         debugs(5, 0, "comm_local_port: FD " << fd << " has been closed.");
 172         return 0;
 173     }
 174
 175     if (F->local_addr.port())
 176         return F->local_addr.port();
 177
 178     if (F->sock_family == AF_INET)
 179         temp.setIPv4();
 180
 181     Ip::Address::InitAddr(addr);
 182
 183     if (getsockname(fd, addr->ai_addr, &(addr->ai_addrlen)) ) {
 184         debugs(50, DBG_IMPORTANT, "comm_local_port: Failed to retrieve TCP/UDP port number for socket: FD " << fd << ": " << xstrerror());
 185         Ip::Address::FreeAddr(addr);
 186         return 0;
 187     }
 188     temp = *addr;
 189
 190     Ip::Address::FreeAddr(addr);
 191
 192     if (F->local_addr.isAnyAddr()) {
 193         /* save the whole local address, not just the port. */
 194         F->local_addr = temp;
 195     } else {
 196         F->local_addr.port(temp.port());
 197     }
 198
 199     debugs(5, 6, "comm_local_port: FD " << fd << ": port " << F->local_addr.port() << "(family=" << F->sock_family << ")");
 200     return F->local_addr.port();
 201 }
 202
 203 static Comm::Flag
 204 commBind(int s, struct addrinfo &inaddr)
 205 {
 206     ++ statCounter.syscalls.sock.binds;
 207
 208     if (bind(s, inaddr.ai_addr, inaddr.ai_addrlen) == 0) {
 209         debugs(50, 6, "commBind: bind socket FD " << s << " to " << fd_table[s].local_addr);
 210         return Comm::OK;
 211     }
 212
 213     debugs(50, 0, "commBind: Cannot bind socket FD " << s << " to " << fd_table[s].local_addr << ": " << xstrerror());
 214
 215     return Comm::COMM_ERROR;
 216 }
 217
 218 /**
 219  * Create a socket. Default is blocking, stream (TCP) socket.  IO_TYPE
 220  * is OR of flags specified in comm.h. Defaults TOS
 221  */
 222 int
 223 comm_open(int sock_type,
 224           int proto,
 225           Ip::Address &addr,
 226           int flags,
 227           const char *note)
 228 {
 229     return comm_openex(sock_type, proto, addr, flags, note);
 230 }
 231
 232 void
 233 comm_open_listener(int sock_type,
 234                    int proto,
 235                    Comm::ConnectionPointer &conn,
 236                    const char *note)
 237 {
 238     /* all listener sockets require bind() */
 239     conn->flags |= COMM_DOBIND;
 240
 241     /* attempt native enabled port. */
 242     conn->fd = comm_openex(sock_type, proto, conn->local, conn->flags, note);
 243 }
 244
 245 int
 246 comm_open_listener(int sock_type,
 247                    int proto,
 248                    Ip::Address &addr,
 249                    int flags,
 250                    const char *note)
 251 {
 252     int sock = -1;
 253
 254     /* all listener sockets require bind() */
 255     flags |= COMM_DOBIND;
 256
 257     /* attempt native enabled port. */
 258     sock = comm_openex(sock_type, proto, addr, flags, note);
 259
 260     return sock;
 261 }
 262
 263 static bool
 264 limitError(int const anErrno)
 265 {
 266     return anErrno == ENFILE || anErrno == EMFILE;
 267 }
 268
 269 void
 270 comm_set_v6only(int fd, int tos)
 271 {
 272 #ifdef IPV6_V6ONLY
 273     if (setsockopt(fd, IPPROTO_IPV6, IPV6_V6ONLY, (char *) &tos, sizeof(int)) < 0) {
 274         debugs(50, DBG_IMPORTANT, "comm_open: setsockopt(IPV6_V6ONLY) " << (tos?"ON":"OFF") << " for FD " << fd << ": " << xstrerror());
 275     }
 276 #else
 277     debugs(50, 0, "WARNING: comm_open: setsockopt(IPV6_V6ONLY) not supported on this platform");
 278 #endif /* sockopt */
 279 }
 280
 281 /**
 282  * Set the socket option required for TPROXY spoofing for:
 283  * - Linux TPROXY v4 support,
 284  * - OpenBSD divert-to support,
 285  * - FreeBSD IPFW TPROXY v4 support.
 286  */
 287 void
 288 comm_set_transparent(int fd)
 289 {
 290 #if _SQUID_LINUX_ && defined(IP_TRANSPARENT) // Linux
 291 # define soLevel SOL_IP
 292 # define soFlag  IP_TRANSPARENT
 293     bool doneSuid = false;
 294
 295 #elif defined(SO_BINDANY) // OpenBSD 4.7+ and NetBSD with PF
 296 # define soLevel SOL_SOCKET
 297 # define soFlag  SO_BINDANY
 298     enter_suid();
 299     bool doneSuid = true;
 300
 301 #elif defined(IP_BINDANY) // FreeBSD with IPFW
 302 # define soLevel IPPROTO_IP
 303 # define soFlag  IP_BINDANY
 304     enter_suid();
 305     bool doneSuid = true;
 306
 307 #else
 308     debugs(50, DBG_CRITICAL, "WARNING: comm_open: setsockopt(TPROXY) not supported on this platform");
 309 #endif /* sockopt */
 310
 311 #if defined(soLevel) && defined(soFlag)
 312     int tos = 1;
 313     if (setsockopt(fd, soLevel, soFlag, (char *) &tos, sizeof(int)) < 0) {
 314         debugs(50, DBG_IMPORTANT, "comm_open: setsockopt(TPROXY) on FD " << fd << ": " << xstrerror());
 315     } else {
 316         /* mark the socket as having transparent options */
 317         fd_table[fd].flags.transparent = true;
 318     }
 319     if (doneSuid)
 320         leave_suid();
 321 #endif
 322 }
 323
 324 /**
 325  * Create a socket. Default is blocking, stream (TCP) socket.  IO_TYPE
 326  * is OR of flags specified in defines.h:COMM_*
 327  */
 328 int
 329 comm_openex(int sock_type,
 330             int proto,
 331             Ip::Address &addr,
 332             int flags,
 333             const char *note)
 334 {
 335     int new_socket;
 336     struct addrinfo *AI = NULL;
 337
 338     PROF_start(comm_open);
 339     /* Create socket for accepting new connections. */
 340     ++ statCounter.syscalls.sock.sockets;
 341
 342     /* Setup the socket addrinfo details for use */
 343     addr.getAddrInfo(AI);
 344     AI->ai_socktype = sock_type;
 345     AI->ai_protocol = proto;
 346
 347     debugs(50, 3, "comm_openex: Attempt open socket for: " << addr );
 348
 349     new_socket = socket(AI->ai_family, AI->ai_socktype, AI->ai_protocol);
 350
 351     /* under IPv6 there is the possibility IPv6 is present but disabled. */
 352     /* try again as IPv4-native if possible */
 353     if ( new_socket < 0 && Ip::EnableIpv6 && addr.isIPv6() && addr.setIPv4() ) {
 354         /* attempt to open this IPv4-only. */
 355         Ip::Address::FreeAddr(AI);
 356         /* Setup the socket addrinfo details for use */
 357         addr.getAddrInfo(AI);
 358         AI->ai_socktype = sock_type;
 359         AI->ai_protocol = proto;
 360         debugs(50, 3, "comm_openex: Attempt fallback open socket for: " << addr );
 361         new_socket = socket(AI->ai_family, AI->ai_socktype, AI->ai_protocol);
 362         debugs(50, 2, HERE << "attempt open " << note << " socket on: " << addr);
 363     }
 364
 365     if (new_socket < 0) {
 366         /* Increase the number of reserved fd's if calls to socket()
 367          * are failing because the open file table is full.  This
 368          * limits the number of simultaneous clients */
 369
 370         if (limitError(errno)) {
 371             debugs(50, DBG_IMPORTANT, "comm_open: socket failure: " << xstrerror());
 372             fdAdjustReserved();
 373         } else {
 374             debugs(50, DBG_CRITICAL, "comm_open: socket failure: " << xstrerror());
 375         }
 376
 377         Ip::Address::FreeAddr(AI);
 378
 379         PROF_stop(comm_open);
 380         return -1;
 381     }
 382
 383     // XXX: temporary for the transition. comm_openex will eventually have a conn to play with.
 384     Comm::ConnectionPointer conn = new Comm::Connection;
 385     conn->local = addr;
 386     conn->fd = new_socket;
 387
 388     debugs(50, 3, "comm_openex: Opened socket " << conn << " : family=" << AI->ai_family << ", type=" << AI->ai_socktype << ", protocol=" << AI->ai_protocol );
 389
 390     if ( Ip::EnableIpv6&IPV6_SPECIAL_SPLITSTACK && addr.isIPv6() )
 391         comm_set_v6only(conn->fd, 1);
 392
 393     /* Windows Vista supports Dual-Sockets. BUT defaults them to V6ONLY. Turn it OFF. */
 394     /* Other OS may have this administratively disabled for general use. Same deal. */
 395     if ( Ip::EnableIpv6&IPV6_SPECIAL_V4MAPPING && addr.isIPv6() )
 396         comm_set_v6only(conn->fd, 0);
 397
 398     comm_init_opened(conn, note, AI);
 399     new_socket = comm_apply_flags(conn->fd, addr, flags, AI);
 400
 401     Ip::Address::FreeAddr(AI);
 402
 403     PROF_stop(comm_open);
 404
 405     // XXX transition only. prevent conn from closing the new FD on function exit.
 406     conn->fd = -1;
 407     return new_socket;
 408 }
 409
 410 /// update FD tables after a local or remote (IPC) comm_openex();
 411 void
 412 comm_init_opened(const Comm::ConnectionPointer &conn,
 413                  const char *note,
 414                  struct addrinfo *AI)
 415 {
 416     assert(Comm::IsConnOpen(conn));
 417     assert(AI);
 418
 419     /* update fdstat */
 420     debugs(5, 5, HERE << conn << " is a new socket");
 421
 422     assert(!isOpen(conn->fd)); // NP: global isOpen checks the fde entry for openness not the Comm::Connection
 423     fd_open(conn->fd, FD_SOCKET, note);
 424
 425     fdd_table[conn->fd].close_file = NULL;
 426     fdd_table[conn->fd].close_line = 0;
 427
 428     fde *F = &fd_table[conn->fd];
 429     F->local_addr = conn->local;
 430
 431     F->sock_family = AI->ai_family;
 432 }
 433
 434 /// apply flags after a local comm_open*() call;
 435 /// returns new_socket or -1 on error
 436 static int
 437 comm_apply_flags(int new_socket,
 438                  Ip::Address &addr,
 439                  int flags,
 440                  struct addrinfo *AI)
 441 {
 442     assert(new_socket >= 0);
 443     assert(AI);
 444     const int sock_type = AI->ai_socktype;
 445
 446     if (!(flags & COMM_NOCLOEXEC))
 447         commSetCloseOnExec(new_socket);
 448
 449     if ((flags & COMM_REUSEADDR))
 450         commSetReuseAddr(new_socket);
 451
 452     if (addr.port() > (unsigned short) 0) {
 453 #if _SQUID_WINDOWS_
 454         if (sock_type != SOCK_DGRAM)
 455 #endif
 456             commSetNoLinger(new_socket);
 457
 458         if (opt_reuseaddr)
 459             commSetReuseAddr(new_socket);
 460     }
 461
 462     /* MUST be done before binding or face OS Error: "(99) Cannot assign requested address"... */
 463     if ((flags & COMM_TRANSPARENT)) {
 464         comm_set_transparent(new_socket);
 465     }
 466
 467     if ( (flags & COMM_DOBIND) || addr.port() > 0 || !addr.isAnyAddr() ) {
 468         if ( !(flags & COMM_DOBIND) && addr.isAnyAddr() )
 469             debugs(5, DBG_IMPORTANT,"WARNING: Squid is attempting to bind() port " << addr << " without being a listener.");
 470         if ( addr.isNoAddr() )
 471             debugs(5,0,"CRITICAL: Squid is attempting to bind() port " << addr << "!!");
 472
 473         if (commBind(new_socket, *AI) != Comm::OK) {
 474             comm_close(new_socket);
 475             return -1;
 476         }
 477     }
 478
 479     if (flags & COMM_NONBLOCKING)
 480         if (commSetNonBlocking(new_socket) == Comm::COMM_ERROR) {
 481             comm_close(new_socket);
 482             return -1;
 483         }
 484
 485 #ifdef TCP_NODELAY
 486     if (sock_type == SOCK_STREAM)
 487         commSetTcpNoDelay(new_socket);
 488
 489 #endif
 490
 491     if (Config.tcpRcvBufsz > 0 && sock_type == SOCK_STREAM)
 492         commSetTcpRcvbuf(new_socket, Config.tcpRcvBufsz);
 493
 494     return new_socket;
 495 }
 496
 497 void
 498 comm_import_opened(const Comm::ConnectionPointer &conn,
 499                    const char *note,
 500                    struct addrinfo *AI)
 501 {
 502     debugs(5, 2, HERE << conn);
 503     assert(Comm::IsConnOpen(conn));
 504     assert(AI);
 505
 506     comm_init_opened(conn, note, AI);
 507
 508     if (!(conn->flags & COMM_NOCLOEXEC))
 509         fd_table[conn->fd].flags.close_on_exec = true;
 510
 511     if (conn->local.port() > (unsigned short) 0) {
 512 #if _SQUID_WINDOWS_
 513         if (AI->ai_socktype != SOCK_DGRAM)
 514 #endif
 515             fd_table[conn->fd].flags.nolinger = true;
 516     }
 517
 518     if ((conn->flags & COMM_TRANSPARENT))
 519         fd_table[conn->fd].flags.transparent = true;
 520
 521     if (conn->flags & COMM_NONBLOCKING)
 522         fd_table[conn->fd].flags.nonblocking = true;
 523
 524 #ifdef TCP_NODELAY
 525     if (AI->ai_socktype == SOCK_STREAM)
 526         fd_table[conn->fd].flags.nodelay = true;
 527 #endif
 528
 529     /* no fd_table[fd].flags. updates needed for these conditions:
 530      * if ((flags & COMM_REUSEADDR)) ...
 531      * if ((flags & COMM_DOBIND) ...) ...
 532      */
 533 }
 534
 535 // XXX: now that raw-FD timeouts are only unset for pipes and files this SHOULD be a no-op.
 536 // With handler already unset. Leaving this present until that can be verified for all code paths.
 537 void
 538 commUnsetFdTimeout(int fd)
 539 {
 540     debugs(5, 3, HERE << "Remove timeout for FD " << fd);
 541     assert(fd >= 0);
 542     assert(fd < Squid_MaxFD);
 543     fde *F = &fd_table[fd];
 544     assert(F->flags.open);
 545
 546     F->timeoutHandler = NULL;
 547     F->timeout = 0;
 548 }
 549
 550 int
 551 commSetConnTimeout(const Comm::ConnectionPointer &conn, int timeout, AsyncCall::Pointer &callback)
 552 {
 553     debugs(5, 3, HERE << conn << " timeout " << timeout);
 554     assert(Comm::IsConnOpen(conn));
 555     assert(conn->fd < Squid_MaxFD);
 556     fde *F = &fd_table[conn->fd];
 557     assert(F->flags.open);
 558
 559     if (timeout < 0) {
 560         F->timeoutHandler = NULL;
 561         F->timeout = 0;
 562     } else {
 563         if (callback != NULL) {
 564             typedef CommTimeoutCbParams Params;
 565             Params &params = GetCommParams<Params>(callback);
 566             params.conn = conn;
 567             F->timeoutHandler = callback;
 568         }
 569
 570         F->timeout = squid_curtime + (time_t) timeout;
 571     }
 572
 573     return F->timeout;
 574 }
 575
 576 int
 577 commUnsetConnTimeout(const Comm::ConnectionPointer &conn)
 578 {
 579     debugs(5, 3, HERE << "Remove timeout for " << conn);
 580     AsyncCall::Pointer nil;
 581     return commSetConnTimeout(conn, -1, nil);
 582 }
 583
 584 /**
 585  * Connect socket FD to given remote address.
 586  * If return value is an error flag (COMM_ERROR, ERR_CONNECT, ERR_PROTOCOL, etc.),
 587  * then error code will also be returned in errno.
 588  */
 589 int
 590 comm_connect_addr(int sock, const Ip::Address &address)
 591 {
 592     Comm::Flag status = Comm::OK;
 593     fde *F = &fd_table[sock];
 594     int x = 0;
 595     int err = 0;
 596     socklen_t errlen;
 597     struct addrinfo *AI = NULL;
 598     PROF_start(comm_connect_addr);
 599
 600     assert(address.port() != 0);
 601
 602     debugs(5, 9, HERE << "connecting socket FD " << sock << " to " << address << " (want family: " << F->sock_family << ")");
 603
 604     /* Handle IPv6 over IPv4-only socket case.
 605      * this case must presently be handled here since the getAddrInfo asserts on bad mappings.
 606      * NP: because commResetFD is private to ConnStateData we have to return an error and
 607      *     trust its handled properly.
 608      */
 609     if (F->sock_family == AF_INET && !address.isIPv4()) {
 610         errno = ENETUNREACH;
 611         return Comm::ERR_PROTOCOL;
 612     }
 613
 614     /* Handle IPv4 over IPv6-only socket case.
 615      * This case is presently handled here as it's both a known case and it's
 616      * uncertain what error will be returned by the IPv6 stack in such case. It's
 617      * possible this will also be handled by the errno checks below after connect()
 618      * but needs carefull cross-platform verification, and verifying the address
 619      * condition here is simple.
 620      */
 621     if (!F->local_addr.isIPv4() && address.isIPv4()) {
 622         errno = ENETUNREACH;
 623         return Comm::ERR_PROTOCOL;
 624     }
 625
 626     address.getAddrInfo(AI, F->sock_family);
 627
 628     /* Establish connection. */
 629     int xerrno = 0;
 630
 631     if (!F->flags.called_connect) {
 632         F->flags.called_connect = true;
 633         ++ statCounter.syscalls.sock.connects;
 634
 635         errno = 0;
 636         if ((x = connect(sock, AI->ai_addr, AI->ai_addrlen)) < 0) {
 637             xerrno = errno;
 638             debugs(5,5, "sock=" << sock << ", addrinfo(" <<
 639                    " flags=" << AI->ai_flags <<
 640                    ", family=" << AI->ai_family <<
 641                    ", socktype=" << AI->ai_socktype <<
 642                    ", protocol=" << AI->ai_protocol <<
 643                    ", &addr=" << AI->ai_addr <<
 644                    ", addrlen=" << AI->ai_addrlen << " )");
 645             debugs(5, 9, "connect FD " << sock << ": (" << x << ") " << xstrerr(xerrno));
 646             debugs(14,9, "connecting to: " << address);
 647
 648         } else if (x == 0) {
 649             // XXX: ICAP code refuses callbacks during a pending comm_ call
 650             // Async calls development will fix this.
 651             x = -1;
 652             xerrno = EINPROGRESS;
 653         }
 654
 655     } else {
 656         errno = 0;
 657 #if _SQUID_NEWSOS6_
 658         /* Makoto MATSUSHITA <matusita@ics.es.osaka-u.ac.jp> */
 659         if (connect(sock, AI->ai_addr, AI->ai_addrlen) < 0)
 660             xerrno = errno;
 661
 662         if (xerrno == EINVAL) {
 663             errlen = sizeof(err);
 664             x = getsockopt(sock, SOL_SOCKET, SO_ERROR, &err, &errlen);
 665             if (x >= 0)
 666                 xerrno = x;
 667         }
 668 #else
 669         errlen = sizeof(err);
 670         x = getsockopt(sock, SOL_SOCKET, SO_ERROR, &err, &errlen);
 671         if (x == 0)
 672             xerrno = err;
 673
 674 #if _SQUID_SOLARIS_
 675         /*
 676         * Solaris 2.4's socket emulation doesn't allow you
 677         * to determine the error from a failed non-blocking
 678         * connect and just returns EPIPE.  Create a fake
 679         * error message for connect.   -- fenner@parc.xerox.com
 680         */
 681         if (x < 0 && xerrno == EPIPE)
 682             xerrno = ENOTCONN;
 683         else
 684             xerrno = errno;
 685 #endif
 686 #endif
 687     }
 688
 689     Ip::Address::FreeAddr(AI);
 690
 691     PROF_stop(comm_connect_addr);
 692
 693     errno = xerrno;
 694     if (xerrno == 0 || xerrno == EISCONN)
 695         status = Comm::OK;
 696     else if (ignoreErrno(xerrno))
 697         status = Comm::INPROGRESS;
 698     else if (xerrno == EAFNOSUPPORT || xerrno == EINVAL)
 699         return Comm::ERR_PROTOCOL;
 700     else
 701         return Comm::COMM_ERROR;
 702
 703     address.toStr(F->ipaddr, MAX_IPSTRLEN);
 704
 705     F->remote_port = address.port(); /* remote_port is HS */
 706
 707     if (status == Comm::OK) {
 708         debugs(5, DBG_DATA, "comm_connect_addr: FD " << sock << " connected to " << address);
 709     } else if (status == Comm::INPROGRESS) {
 710         debugs(5, DBG_DATA, "comm_connect_addr: FD " << sock << " connection pending");
 711     }
 712
 713     errno = xerrno;
 714     return status;
 715 }
 716
 717 void
 718 commCallCloseHandlers(int fd)
 719 {
 720     fde *F = &fd_table[fd];
 721     debugs(5, 5, "commCallCloseHandlers: FD " << fd);
 722
 723     while (F->closeHandler != NULL) {
 724         AsyncCall::Pointer call = F->closeHandler;
 725         F->closeHandler = call->Next();
 726         call->setNext(NULL);
 727         // If call is not canceled schedule it for execution else ignore it
 728         if (!call->canceled()) {
 729             debugs(5, 5, "commCallCloseHandlers: ch->handler=" << call);
 730             ScheduleCallHere(call);
 731         }
 732     }
 733 }
 734
 735 #if LINGERING_CLOSE
 736 static void
 737 commLingerClose(int fd, void *unused)
 738 {
 739     LOCAL_ARRAY(char, buf, 1024);
 740     int n;
 741     n = FD_READ_METHOD(fd, buf, 1024);
 742
 743     if (n < 0)
 744         debugs(5, 3, "commLingerClose: FD " << fd << " read: " << xstrerror());
 745
 746     comm_close(fd);
 747 }
 748
 749 static void
 750 commLingerTimeout(const FdeCbParams &params)
 751 {
 752     debugs(5, 3, "commLingerTimeout: FD " << params.fd);
 753     comm_close(params.fd);
 754 }
 755
 756 /*
 757  * Inspired by apache
 758  */
 759 void
 760 comm_lingering_close(int fd)
 761 {
 762 #if USE_OPENSSL
 763     if (fd_table[fd].ssl)
 764         ssl_shutdown_method(fd_table[fd].ssl);
 765 #endif
 766
 767     if (shutdown(fd, 1) < 0) {
 768         comm_close(fd);
 769         return;
 770     }
 771
 772     fd_note(fd, "lingering close");
 773     AsyncCall::Pointer call = commCbCall(5,4, "commLingerTimeout", FdeCbPtrFun(commLingerTimeout, NULL));
 774
 775     debugs(5, 3, HERE << "FD " << fd << " timeout " << timeout);
 776     assert(fd_table[fd].flags.open);
 777     if (callback != NULL) {
 778         typedef FdeCbParams Params;
 779         Params &params = GetCommParams<Params>(callback);
 780         params.fd = fd;
 781         fd_table[fd].timeoutHandler = callback;
 782         fd_table[fd].timeout = squid_curtime + static_cast<time_t>(10);
 783     }
 784
 785     Comm::SetSelect(fd, COMM_SELECT_READ, commLingerClose, NULL, 0);
 786 }
 787
 788 #endif
 789
 790 /**
 791  * enable linger with time of 0 so that when the socket is
 792  * closed, TCP generates a RESET
 793  */
 794 void
 795 comm_reset_close(const Comm::ConnectionPointer &conn)
 796 {
 797     struct linger L;
 798     L.l_onoff = 1;
 799     L.l_linger = 0;
 800
 801     if (setsockopt(conn->fd, SOL_SOCKET, SO_LINGER, (char *) &L, sizeof(L)) < 0)
 802         debugs(50, DBG_CRITICAL, "ERROR: Closing " << conn << " with TCP RST: " << xstrerror());
 803
 804     conn->close();
 805 }
 806
 807 // Legacy close function.
 808 void
 809 old_comm_reset_close(int fd)
 810 {
 811     struct linger L;
 812     L.l_onoff = 1;
 813     L.l_linger = 0;
 814
 815     if (setsockopt(fd, SOL_SOCKET, SO_LINGER, (char *) &L, sizeof(L)) < 0)
 816         debugs(50, DBG_CRITICAL, "ERROR: Closing FD " << fd << " with TCP RST: " << xstrerror());
 817
 818     comm_close(fd);
 819 }
 820
 821 #if USE_OPENSSL
 822 void
 823 commStartSslClose(const FdeCbParams &params)
 824 {
 825     assert(fd_table[params.fd].ssl);
 826     ssl_shutdown_method(fd_table[params.fd].ssl.get());
 827 }
 828 #endif
 829
 830 void
 831 comm_close_complete(const FdeCbParams &params)
 832 {
 833     fde *F = &fd_table[params.fd];
 834     F->ssl.reset(nullptr);
 835
 836 #if USE_OPENSSL
 837     if (F->dynamicSslContext) {
 838         SSL_CTX_free(F->dynamicSslContext);
 839         F->dynamicSslContext = NULL;
 840     }
 841 #endif
 842     fd_close(params.fd);        /* update fdstat */
 843     close(params.fd);
 844
 845     ++ statCounter.syscalls.sock.closes;
 846
 847     /* When one connection closes, give accept() a chance, if need be */
 848     Comm::AcceptLimiter::Instance().kick();
 849 }
 850
 851 /*
 852  * Close the socket fd.
 853  *
 854  * + call write handlers with ERR_CLOSING
 855  * + call read handlers with ERR_CLOSING
 856  * + call closing handlers
 857  *
 858  * NOTE: Comm::ERR_CLOSING will NOT be called for CommReads' sitting in a
 859  * DeferredReadManager.
 860  */
 861 void
 862 _comm_close(int fd, char const *file, int line)
 863 {
 864     debugs(5, 3, "comm_close: start closing FD " << fd);
 865     assert(fd >= 0);
 866     assert(fd < Squid_MaxFD);
 867
 868     fde *F = &fd_table[fd];
 869     fdd_table[fd].close_file = file;
 870     fdd_table[fd].close_line = line;
 871
 872     if (F->closing())
 873         return;
 874
 875     /* XXX: is this obsolete behind F->closing() ? */
 876     if ( (shutting_down || reconfiguring) && (!F->flags.open || F->type == FD_FILE))
 877         return;
 878
 879     /* The following fails because ipc.c is doing calls to pipe() to create sockets! */
 880     if (!isOpen(fd)) {
 881         debugs(50, DBG_IMPORTANT, HERE << "BUG 3556: FD " << fd << " is not an open socket.");
 882         // XXX: do we need to run close(fd) or fd_close(fd) here?
 883         return;
 884     }
 885
 886     assert(F->type != FD_FILE);
 887
 888     PROF_start(comm_close);
 889
 890     F->flags.close_request = true;
 891
 892 #if USE_OPENSSL
 893     if (F->ssl) {
 894         AsyncCall::Pointer startCall=commCbCall(5,4, "commStartSslClose",
 895                                                 FdeCbPtrFun(commStartSslClose, NULL));
 896         FdeCbParams &startParams = GetCommParams<FdeCbParams>(startCall);
 897         startParams.fd = fd;
 898         ScheduleCallHere(startCall);
 899     }
 900 #endif
 901
 902     // a half-closed fd may lack a reader, so we stop monitoring explicitly
 903     if (commHasHalfClosedMonitor(fd))
 904         commStopHalfClosedMonitor(fd);
 905     commUnsetFdTimeout(fd);
 906
 907     // notify read/write handlers after canceling select reservations, if any
 908     if (COMMIO_FD_WRITECB(fd)->active()) {
 909         Comm::SetSelect(fd, COMM_SELECT_WRITE, NULL, NULL, 0);
 910         COMMIO_FD_WRITECB(fd)->finish(Comm::ERR_CLOSING, errno);
 911     }
 912     if (COMMIO_FD_READCB(fd)->active()) {
 913         Comm::SetSelect(fd, COMM_SELECT_READ, NULL, NULL, 0);
 914         COMMIO_FD_READCB(fd)->finish(Comm::ERR_CLOSING, errno);
 915     }
 916
 917 #if USE_DELAY_POOLS
 918     if (ClientInfo *clientInfo = F->clientInfo) {
 919         if (clientInfo->selectWaiting) {
 920             clientInfo->selectWaiting = false;
 921             // kick queue or it will get stuck as commWriteHandle is not called
 922             clientInfo->kickQuotaQueue();
 923         }
 924     }
 925 #endif
 926
 927     commCallCloseHandlers(fd);
 928
 929     comm_empty_os_read_buffers(fd);
 930
 931     AsyncCall::Pointer completeCall=commCbCall(5,4, "comm_close_complete",
 932                                     FdeCbPtrFun(comm_close_complete, NULL));
 933     FdeCbParams &completeParams = GetCommParams<FdeCbParams>(completeCall);
 934     completeParams.fd = fd;
 935     // must use async call to wait for all callbacks
 936     // scheduled before comm_close() to finish
 937     ScheduleCallHere(completeCall);
 938
 939     PROF_stop(comm_close);
 940 }
 941
 942 /* Send a udp datagram to specified TO_ADDR. */
 943 int
 944 comm_udp_sendto(int fd,
 945                 const Ip::Address &to_addr,
 946                 const void *buf,
 947                 int len)
 948 {
 949     PROF_start(comm_udp_sendto);
 950     ++ statCounter.syscalls.sock.sendtos;
 951
 952     debugs(50, 3, "comm_udp_sendto: Attempt to send UDP packet to " << to_addr <<
 953            " using FD " << fd << " using Port " << comm_local_port(fd) );
 954
 955     struct addrinfo *AI = NULL;
 956     to_addr.getAddrInfo(AI, fd_table[fd].sock_family);
 957     int x = sendto(fd, buf, len, 0, AI->ai_addr, AI->ai_addrlen);
 958     Ip::Address::FreeAddr(AI);
 959
 960     PROF_stop(comm_udp_sendto);
 961
 962     if (x >= 0)
 963         return x;
 964
 965 #if _SQUID_LINUX_
 966
 967     if (ECONNREFUSED != errno)
 968 #endif
 969
 970         debugs(50, DBG_IMPORTANT, "comm_udp_sendto: FD " << fd << ", (family=" << fd_table[fd].sock_family << ") " << to_addr << ": " << xstrerror());
 971
 972     return Comm::COMM_ERROR;
 973 }
 974
 975 AsyncCall::Pointer
 976 comm_add_close_handler(int fd, CLCB * handler, void *data)
 977 {
 978     debugs(5, 5, "comm_add_close_handler: FD " << fd << ", handler=" <<
 979            handler << ", data=" << data);
 980
 981     AsyncCall::Pointer call=commCbCall(5,4, "SomeCloseHandler",
 982                                        CommCloseCbPtrFun(handler, data));
 983     comm_add_close_handler(fd, call);
 984     return call;
 985 }
 986
 987 void
 988 comm_add_close_handler(int fd, AsyncCall::Pointer &call)
 989 {
 990     debugs(5, 5, "comm_add_close_handler: FD " << fd << ", AsyncCall=" << call);
 991
 992     /*TODO:Check for a similar scheduled AsyncCall*/
 993 //    for (c = fd_table[fd].closeHandler; c; c = c->next)
 994 //        assert(c->handler != handler || c->data != data);
 995
 996     call->setNext(fd_table[fd].closeHandler);
 997
 998     fd_table[fd].closeHandler = call;
 999 }
1000
1001 // remove function-based close handler
1002 void
1003 comm_remove_close_handler(int fd, CLCB * handler, void *data)
1004 {
1005     assert(isOpen(fd));
1006     /* Find handler in list */
1007     debugs(5, 5, "comm_remove_close_handler: FD " << fd << ", handler=" <<
1008            handler << ", data=" << data);
1009
1010     AsyncCall::Pointer p, prev = NULL;
1011     for (p = fd_table[fd].closeHandler; p != NULL; prev = p, p = p->Next()) {
1012         typedef CommCbFunPtrCallT<CommCloseCbPtrFun> Call;
1013         const Call *call = dynamic_cast<const Call*>(p.getRaw());
1014         if (!call) // method callbacks have their own comm_remove_close_handler
1015             continue;
1016
1017         typedef CommCloseCbParams Params;
1018         const Params &params = GetCommParams<Params>(p);
1019         if (call->dialer.handler == handler && params.data == data)
1020             break;      /* This is our handler */
1021     }
1022
1023     // comm_close removes all close handlers so our handler may be gone
1024     if (p != NULL) {
1025         p->dequeue(fd_table[fd].closeHandler, prev);
1026         p->cancel("comm_remove_close_handler");
1027     }
1028 }
1029
1030 // remove method-based close handler
1031 void
1032 comm_remove_close_handler(int fd, AsyncCall::Pointer &call)
1033 {
1034     assert(isOpen(fd));
1035     debugs(5, 5, "comm_remove_close_handler: FD " << fd << ", AsyncCall=" << call);
1036
1037     // comm_close removes all close handlers so our handler may be gone
1038     AsyncCall::Pointer p, prev = NULL;
1039     for (p = fd_table[fd].closeHandler; p != NULL && p != call; prev = p, p = p->Next());
1040
1041     if (p != NULL)
1042         p->dequeue(fd_table[fd].closeHandler, prev);
1043     call->cancel("comm_remove_close_handler");
1044 }
1045
1046 static void
1047 commSetNoLinger(int fd)
1048 {
1049
1050     struct linger L;
1051     L.l_onoff = 0;      /* off */
1052     L.l_linger = 0;
1053
1054     if (setsockopt(fd, SOL_SOCKET, SO_LINGER, (char *) &L, sizeof(L)) < 0)
1055         debugs(50, 0, "commSetNoLinger: FD " << fd << ": " << xstrerror());
1056
1057     fd_table[fd].flags.nolinger = true;
1058 }
1059
1060 static void
1061 commSetReuseAddr(int fd)
1062 {
1063     int on = 1;
1064
1065     if (setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, (char *) &on, sizeof(on)) < 0)
1066         debugs(50, DBG_IMPORTANT, "commSetReuseAddr: FD " << fd << ": " << xstrerror());
1067 }
1068
1069 static void
1070 commSetTcpRcvbuf(int fd, int size)
1071 {
1072     if (setsockopt(fd, SOL_SOCKET, SO_RCVBUF, (char *) &size, sizeof(size)) < 0)
1073         debugs(50, DBG_IMPORTANT, "commSetTcpRcvbuf: FD " << fd << ", SIZE " << size << ": " << xstrerror());
1074     if (setsockopt(fd, SOL_SOCKET, SO_SNDBUF, (char *) &size, sizeof(size)) < 0)
1075         debugs(50, DBG_IMPORTANT, "commSetTcpRcvbuf: FD " << fd << ", SIZE " << size << ": " << xstrerror());
1076 #ifdef TCP_WINDOW_CLAMP
1077     if (setsockopt(fd, SOL_TCP, TCP_WINDOW_CLAMP, (char *) &size, sizeof(size)) < 0)
1078         debugs(50, DBG_IMPORTANT, "commSetTcpRcvbuf: FD " << fd << ", SIZE " << size << ": " << xstrerror());
1079 #endif
1080 }
1081
1082 int
1083 commSetNonBlocking(int fd)
1084 {
1085 #if _SQUID_WINDOWS_
1086     int nonblocking = TRUE;
1087
1088     if (ioctl(fd, FIONBIO, &nonblocking) < 0) {
1089         debugs(50, 0, "commSetNonBlocking: FD " << fd << ": " << xstrerror() << " " << fd_table[fd].type);
1090         return Comm::COMM_ERROR;
1091     }
1092
1093 #else
1094     int flags;
1095     int dummy = 0;
1096
1097     if ((flags = fcntl(fd, F_GETFL, dummy)) < 0) {
1098         debugs(50, 0, "FD " << fd << ": fcntl F_GETFL: " << xstrerror());
1099         return Comm::COMM_ERROR;
1100     }
1101
1102     if (fcntl(fd, F_SETFL, flags | SQUID_NONBLOCK) < 0) {
1103         debugs(50, 0, "commSetNonBlocking: FD " << fd << ": " << xstrerror());
1104         return Comm::COMM_ERROR;
1105     }
1106 #endif
1107
1108     fd_table[fd].flags.nonblocking = true;
1109     return 0;
1110 }
1111
1112 int
1113 commUnsetNonBlocking(int fd)
1114 {
1115 #if _SQUID_WINDOWS_
1116     int nonblocking = FALSE;
1117
1118     if (ioctlsocket(fd, FIONBIO, (unsigned long *) &nonblocking) < 0) {
1119 #else
1120     int flags;
1121     int dummy = 0;
1122
1123     if ((flags = fcntl(fd, F_GETFL, dummy)) < 0) {
1124         debugs(50, 0, "FD " << fd << ": fcntl F_GETFL: " << xstrerror());
1125         return Comm::COMM_ERROR;
1126     }
1127
1128     if (fcntl(fd, F_SETFL, flags & (~SQUID_NONBLOCK)) < 0) {
1129 #endif
1130         debugs(50, 0, "commUnsetNonBlocking: FD " << fd << ": " << xstrerror());
1131         return Comm::COMM_ERROR;
1132     }
1133
1134     fd_table[fd].flags.nonblocking = false;
1135     return 0;
1136 }
1137
1138 void
1139 commSetCloseOnExec(int fd)
1140 {
1141 #ifdef FD_CLOEXEC
1142     int flags;
1143     int dummy = 0;
1144
1145     if ((flags = fcntl(fd, F_GETFD, dummy)) < 0) {
1146         debugs(50, 0, "FD " << fd << ": fcntl F_GETFD: " << xstrerror());
1147         return;
1148     }
1149
1150     if (fcntl(fd, F_SETFD, flags | FD_CLOEXEC) < 0)
1151         debugs(50, 0, "FD " << fd << ": set close-on-exec failed: " << xstrerror());
1152
1153     fd_table[fd].flags.close_on_exec = true;
1154
1155 #endif
1156 }
1157
1158 #ifdef TCP_NODELAY
1159 static void
1160 commSetTcpNoDelay(int fd)
1161 {
1162     int on = 1;
1163
1164     if (setsockopt(fd, IPPROTO_TCP, TCP_NODELAY, (char *) &on, sizeof(on)) < 0)
1165         debugs(50, DBG_IMPORTANT, "commSetTcpNoDelay: FD " << fd << ": " << xstrerror());
1166
1167     fd_table[fd].flags.nodelay = true;
1168 }
1169
1170 #endif
1171
1172 void
1173 commSetTcpKeepalive(int fd, int idle, int interval, int timeout)
1174 {
1175     int on = 1;
1176 #ifdef TCP_KEEPCNT
1177     if (timeout && interval) {
1178         int count = (timeout + interval - 1) / interval;
1179         if (setsockopt(fd, IPPROTO_TCP, TCP_KEEPCNT, &count, sizeof(on)) < 0)
1180             debugs(5, DBG_IMPORTANT, "commSetKeepalive: FD " << fd << ": " << xstrerror());
1181     }
1182 #endif
1183 #ifdef TCP_KEEPIDLE
1184     if (idle) {
1185         if (setsockopt(fd, IPPROTO_TCP, TCP_KEEPIDLE, &idle, sizeof(on)) < 0)
1186             debugs(5, DBG_IMPORTANT, "commSetKeepalive: FD " << fd << ": " << xstrerror());
1187     }
1188 #endif
1189 #ifdef TCP_KEEPINTVL
1190     if (interval) {
1191         if (setsockopt(fd, IPPROTO_TCP, TCP_KEEPINTVL, &interval, sizeof(on)) < 0)
1192             debugs(5, DBG_IMPORTANT, "commSetKeepalive: FD " << fd << ": " << xstrerror());
1193     }
1194 #endif
1195     if (setsockopt(fd, SOL_SOCKET, SO_KEEPALIVE, (char *) &on, sizeof(on)) < 0)
1196         debugs(5, DBG_IMPORTANT, "commSetKeepalive: FD " << fd << ": " << xstrerror());
1197 }
1198
1199 void
1200 comm_init(void)
1201 {
1202     fd_table =(fde *) xcalloc(Squid_MaxFD, sizeof(fde));
1203     fdd_table = (fd_debug_t *)xcalloc(Squid_MaxFD, sizeof(fd_debug_t));
1204
1205     /* make sure the accept() socket FIFO delay queue exists */
1206     Comm::AcceptLimiter::Instance();
1207
1208     // make sure the IO pending callback table exists
1209     Comm::CallbackTableInit();
1210
1211     /* XXX account fd_table */
1212     /* Keep a few file descriptors free so that we don't run out of FD's
1213      * after accepting a client but before it opens a socket or a file.
1214      * Since Squid_MaxFD can be as high as several thousand, don't waste them */
1215     RESERVED_FD = min(100, Squid_MaxFD / 4);
1216
1217     TheHalfClosed = new DescriptorSet;
1218
1219     /* setup the select loop module */
1220     Comm::SelectLoopInit();
1221 }
1222
1223 void
1224 comm_exit(void)
1225 {
1226     delete TheHalfClosed;
1227     TheHalfClosed = NULL;
1228
1229     safe_free(fd_table);
1230     safe_free(fdd_table);
1231     Comm::CallbackTableDestruct();
1232 }
1233
1234 #if USE_DELAY_POOLS
1235 // called when the queue is done waiting for the client bucket to fill
1236 void
1237 commHandleWriteHelper(void * data)
1238 {
1239     CommQuotaQueue *queue = static_cast<CommQuotaQueue*>(data);
1240     assert(queue);
1241
1242     ClientInfo *clientInfo = queue->clientInfo;
1243     // ClientInfo invalidates queue if freed, so if we got here through,
1244     // evenAdd cbdata protections, everything should be valid and consistent
1245     assert(clientInfo);
1246     assert(clientInfo->hasQueue());
1247     assert(clientInfo->hasQueue(queue));
1248     assert(!clientInfo->selectWaiting);
1249     assert(clientInfo->eventWaiting);
1250     clientInfo->eventWaiting = false;
1251
1252     do {
1253         // check that the head descriptor is still relevant
1254         const int head = clientInfo->quotaPeekFd();
1255         Comm::IoCallback *ccb = COMMIO_FD_WRITECB(head);
1256
1257         if (fd_table[head].clientInfo == clientInfo &&
1258                 clientInfo->quotaPeekReserv() == ccb->quotaQueueReserv &&
1259                 !fd_table[head].closing()) {
1260
1261             // wait for the head descriptor to become ready for writing
1262             Comm::SetSelect(head, COMM_SELECT_WRITE, Comm::HandleWrite, ccb, 0);
1263             clientInfo->selectWaiting = true;
1264             return;
1265         }
1266
1267         clientInfo->quotaDequeue(); // remove the no longer relevant descriptor
1268         // and continue looking for a relevant one
1269     } while (clientInfo->hasQueue());
1270
1271     debugs(77,3, HERE << "emptied queue");
1272 }
1273
1274 bool
1275 ClientInfo::hasQueue() const
1276 {
1277     assert(quotaQueue);
1278     return !quotaQueue->empty();
1279 }
1280
1281 bool
1282 ClientInfo::hasQueue(const CommQuotaQueue *q) const
1283 {
1284     assert(quotaQueue);
1285     return quotaQueue == q;
1286 }
1287
1288 /// returns the first descriptor to be dequeued
1289 int
1290 ClientInfo::quotaPeekFd() const
1291 {
1292     assert(quotaQueue);
1293     return quotaQueue->front();
1294 }
1295
1296 /// returns the reservation ID of the first descriptor to be dequeued
1297 unsigned int
1298 ClientInfo::quotaPeekReserv() const
1299 {
1300     assert(quotaQueue);
1301     return quotaQueue->outs + 1;
1302 }
1303
1304 /// queues a given fd, creating the queue if necessary; returns reservation ID
1305 unsigned int
1306 ClientInfo::quotaEnqueue(int fd)
1307 {
1308     assert(quotaQueue);
1309     return quotaQueue->enqueue(fd);
1310 }
1311
1312 /// removes queue head
1313 void
1314 ClientInfo::quotaDequeue()
1315 {
1316     assert(quotaQueue);
1317     quotaQueue->dequeue();
1318 }
1319
1320 void
1321 ClientInfo::kickQuotaQueue()
1322 {
1323     if (!eventWaiting && !selectWaiting && hasQueue()) {
1324         // wait at least a second if the bucket is empty
1325         const double delay = (bucketSize < 1.0) ? 1.0 : 0.0;
1326         eventAdd("commHandleWriteHelper", &commHandleWriteHelper,
1327                  quotaQueue, delay, 0, true);
1328         eventWaiting = true;
1329     }
1330 }
1331
1332 /// calculates how much to write for a single dequeued client
1333 int
1334 ClientInfo::quotaForDequed()
1335 {
1336     /* If we have multiple clients and give full bucketSize to each client then
1337      * clt1 may often get a lot more because clt1->clt2 time distance in the
1338      * select(2) callback order may be a lot smaller than cltN->clt1 distance.
1339      * We divide quota evenly to be more fair. */
1340
1341     if (!rationedCount) {
1342         rationedCount = quotaQueue->size() + 1;
1343
1344         // The delay in ration recalculation _temporary_ deprives clients from
1345         // bytes that should have trickled in while rationedCount was positive.
1346         refillBucket();
1347
1348         // Rounding errors do not accumulate here, but we round down to avoid
1349         // negative bucket sizes after write with rationedCount=1.
1350         rationedQuota = static_cast<int>(floor(bucketSize/rationedCount));
1351         debugs(77,5, HERE << "new rationedQuota: " << rationedQuota <<
1352                '*' << rationedCount);
1353     }
1354
1355     --rationedCount;
1356     debugs(77,7, HERE << "rationedQuota: " << rationedQuota <<
1357            " rations remaining: " << rationedCount);
1358
1359     // update 'last seen' time to prevent clientdb GC from dropping us
1360     last_seen = squid_curtime;
1361     return rationedQuota;
1362 }
1363
1364 ///< adds bytes to the quota bucket based on the rate and passed time
1365 void
1366 ClientInfo::refillBucket()
1367 {
1368     // all these times are in seconds, with double precision
1369     const double currTime = current_dtime;
1370     const double timePassed = currTime - prevTime;
1371
1372     // Calculate allowance for the time passed. Use double to avoid
1373     // accumulating rounding errors for small intervals. For example, always
1374     // adding 1 byte instead of 1.4 results in 29% bandwidth allocation error.
1375     const double gain = timePassed * writeSpeedLimit;
1376
1377     debugs(77,5, HERE << currTime << " clt" << (const char*)hash.key << ": " <<
1378            bucketSize << " + (" << timePassed << " * " << writeSpeedLimit <<
1379            " = " << gain << ')');
1380
1381     // to further combat error accumulation during micro updates,
1382     // quit before updating time if we cannot add at least one byte
1383     if (gain < 1.0)
1384         return;
1385
1386     prevTime = currTime;
1387
1388     // for "first" connections, drain initial fat before refilling but keep
1389     // updating prevTime to avoid bursts after the fat is gone
1390     if (bucketSize > bucketSizeLimit) {
1391         debugs(77,4, HERE << "not refilling while draining initial fat");
1392         return;
1393     }
1394
1395     bucketSize += gain;
1396
1397     // obey quota limits
1398     if (bucketSize > bucketSizeLimit)
1399         bucketSize = bucketSizeLimit;
1400 }
1401
1402 void
1403 ClientInfo::setWriteLimiter(const int aWriteSpeedLimit, const double anInitialBurst, const double aHighWatermark)
1404 {
1405     debugs(77,5, HERE << "Write limits for " << (const char*)hash.key <<
1406            " speed=" << aWriteSpeedLimit << " burst=" << anInitialBurst <<
1407            " highwatermark=" << aHighWatermark);
1408
1409     // set or possibly update traffic shaping parameters
1410     writeLimitingActive = true;
1411     writeSpeedLimit = aWriteSpeedLimit;
1412     bucketSizeLimit = aHighWatermark;
1413
1414     // but some members should only be set once for a newly activated bucket
1415     if (firstTimeConnection) {
1416         firstTimeConnection = false;
1417
1418         assert(!selectWaiting);
1419         assert(!quotaQueue);
1420         quotaQueue = new CommQuotaQueue(this);
1421
1422         bucketSize = anInitialBurst;
1423         prevTime = current_dtime;
1424     }
1425 }
1426
1427 CommQuotaQueue::CommQuotaQueue(ClientInfo *info): clientInfo(info),
1428     ins(0), outs(0)
1429 {
1430     assert(clientInfo);
1431 }
1432
1433 CommQuotaQueue::~CommQuotaQueue()
1434 {
1435     assert(!clientInfo); // ClientInfo should clear this before destroying us
1436 }
1437
1438 /// places the given fd at the end of the queue; returns reservation ID
1439 unsigned int
1440 CommQuotaQueue::enqueue(int fd)
1441 {
1442     debugs(77,5, HERE << "clt" << (const char*)clientInfo->hash.key <<
1443            ": FD " << fd << " with qqid" << (ins+1) << ' ' << fds.size());
1444     fds.push_back(fd);
1445     return ++ins;
1446 }
1447
1448 /// removes queue head
1449 void
1450 CommQuotaQueue::dequeue()
1451 {
1452     assert(!fds.empty());
1453     debugs(77,5, HERE << "clt" << (const char*)clientInfo->hash.key <<
1454            ": FD " << fds.front() << " with qqid" << (outs+1) << ' ' <<
1455            fds.size());
1456     fds.pop_front();
1457     ++outs;
1458 }
1459 #endif
1460
1461 /*
1462  * hm, this might be too general-purpose for all the places we'd
1463  * like to use it.
1464  */
1465 int
1466 ignoreErrno(int ierrno)
1467 {
1468     switch (ierrno) {
1469
1470     case EINPROGRESS:
1471
1472     case EWOULDBLOCK:
1473 #if EAGAIN != EWOULDBLOCK
1474
1475     case EAGAIN:
1476 #endif
1477
1478     case EALREADY:
1479
1480     case EINTR:
1481 #ifdef ERESTART
1482
1483     case ERESTART:
1484 #endif
1485
1486         return 1;
1487
1488     default:
1489         return 0;
1490     }
1491
1492     /* NOTREACHED */
1493 }
1494
1495 void
1496 commCloseAllSockets(void)
1497 {
1498     int fd;
1499     fde *F = NULL;
1500
1501     for (fd = 0; fd <= Biggest_FD; ++fd) {
1502         F = &fd_table[fd];
1503
1504         if (!F->flags.open)
1505             continue;
1506
1507         if (F->type != FD_SOCKET)
1508             continue;
1509
1510         if (F->flags.ipc)   /* don't close inter-process sockets */
1511             continue;
1512
1513         if (F->timeoutHandler != NULL) {
1514             AsyncCall::Pointer callback = F->timeoutHandler;
1515             F->timeoutHandler = NULL;
1516             debugs(5, 5, "commCloseAllSockets: FD " << fd << ": Calling timeout handler");
1517             ScheduleCallHere(callback);
1518         } else {
1519             debugs(5, 5, "commCloseAllSockets: FD " << fd << ": calling comm_reset_close()");
1520             old_comm_reset_close(fd);
1521         }
1522     }
1523 }
1524
1525 static bool
1526 AlreadyTimedOut(fde *F)
1527 {
1528     if (!F->flags.open)
1529         return true;
1530
1531     if (F->timeout == 0)
1532         return true;
1533
1534     if (F->timeout > squid_curtime)
1535         return true;
1536
1537     return false;
1538 }
1539
1540 static bool
1541 writeTimedOut(int fd)
1542 {
1543     if (!COMMIO_FD_WRITECB(fd)->active())
1544         return false;
1545
1546     if ((squid_curtime - fd_table[fd].writeStart) < Config.Timeout.write)
1547         return false;
1548
1549     return true;
1550 }
1551
1552 void
1553 checkTimeouts(void)
1554 {
1555     int fd;
1556     fde *F = NULL;
1557     AsyncCall::Pointer callback;
1558
1559     for (fd = 0; fd <= Biggest_FD; ++fd) {
1560         F = &fd_table[fd];
1561
1562         if (writeTimedOut(fd)) {
1563             // We have an active write callback and we are timed out
1564             debugs(5, 5, "checkTimeouts: FD " << fd << " auto write timeout");
1565             Comm::SetSelect(fd, COMM_SELECT_WRITE, NULL, NULL, 0);
1566             COMMIO_FD_WRITECB(fd)->finish(Comm::COMM_ERROR, ETIMEDOUT);
1567         } else if (AlreadyTimedOut(F))
1568             continue;
1569
1570         debugs(5, 5, "checkTimeouts: FD " << fd << " Expired");
1571
1572         if (F->timeoutHandler != NULL) {
1573             debugs(5, 5, "checkTimeouts: FD " << fd << ": Call timeout handler");
1574             callback = F->timeoutHandler;
1575             F->timeoutHandler = NULL;
1576             ScheduleCallHere(callback);
1577         } else {
1578             debugs(5, 5, "checkTimeouts: FD " << fd << ": Forcing comm_close()");
1579             comm_close(fd);
1580         }
1581     }
1582 }
1583
1584 /// Start waiting for a possibly half-closed connection to close
1585 // by scheduling a read callback to a monitoring handler that
1586 // will close the connection on read errors.
1587 void
1588 commStartHalfClosedMonitor(int fd)
1589 {
1590     debugs(5, 5, HERE << "adding FD " << fd << " to " << *TheHalfClosed);
1591     assert(isOpen(fd) && !commHasHalfClosedMonitor(fd));
1592     (void)TheHalfClosed->add(fd); // could also assert the result
1593     commPlanHalfClosedCheck(); // may schedule check if we added the first FD
1594 }
1595
1596 static
1597 void
1598 commPlanHalfClosedCheck()
1599 {
1600     if (!WillCheckHalfClosed && !TheHalfClosed->empty()) {
1601         eventAdd("commHalfClosedCheck", &commHalfClosedCheck, NULL, 1.0, 1);
1602         WillCheckHalfClosed = true;
1603     }
1604 }
1605
1606 /// iterates over all descriptors that may need half-closed tests and
1607 /// calls comm_read for those that do; re-schedules the check if needed
1608 static
1609 void
1610 commHalfClosedCheck(void *)
1611 {
1612     debugs(5, 5, HERE << "checking " << *TheHalfClosed);
1613
1614     typedef DescriptorSet::const_iterator DSCI;
1615     const DSCI end = TheHalfClosed->end();
1616     for (DSCI i = TheHalfClosed->begin(); i != end; ++i) {
1617         Comm::ConnectionPointer c = new Comm::Connection; // XXX: temporary. make HalfClosed a list of these.
1618         c->fd = *i;
1619         if (!fd_table[c->fd].halfClosedReader) { // not reading already
1620             AsyncCall::Pointer call = commCbCall(5,4, "commHalfClosedReader",
1621                                                  CommIoCbPtrFun(&commHalfClosedReader, NULL));
1622             Comm::Read(c, call);
1623             fd_table[c->fd].halfClosedReader = call;
1624         } else
1625             c->fd = -1; // XXX: temporary. prevent c replacement erase closing listed FD
1626     }
1627
1628     WillCheckHalfClosed = false; // as far as we know
1629     commPlanHalfClosedCheck(); // may need to check again
1630 }
1631
1632 /// checks whether we are waiting for possibly half-closed connection to close
1633 // We are monitoring if the read handler for the fd is the monitoring handler.
1634 bool
1635 commHasHalfClosedMonitor(int fd)
1636 {
1637     return TheHalfClosed->has(fd);
1638 }
1639
1640 /// stop waiting for possibly half-closed connection to close
1641 void
1642 commStopHalfClosedMonitor(int const fd)
1643 {
1644     debugs(5, 5, HERE << "removing FD " << fd << " from " << *TheHalfClosed);
1645
1646     // cancel the read if one was scheduled
1647     AsyncCall::Pointer reader = fd_table[fd].halfClosedReader;
1648     if (reader != NULL)
1649         Comm::ReadCancel(fd, reader);
1650     fd_table[fd].halfClosedReader = NULL;
1651
1652     TheHalfClosed->del(fd);
1653 }
1654
1655 /// I/O handler for the possibly half-closed connection monitoring code
1656 static void
1657 commHalfClosedReader(const Comm::ConnectionPointer &conn, char *, size_t size, Comm::Flag flag, int, void *)
1658 {
1659     // there cannot be more data coming in on half-closed connections
1660     assert(size == 0);
1661     assert(conn != NULL);
1662     assert(commHasHalfClosedMonitor(conn->fd)); // or we would have canceled the read
1663
1664     fd_table[conn->fd].halfClosedReader = NULL; // done reading, for now
1665
1666     // nothing to do if fd is being closed
1667     if (flag == Comm::ERR_CLOSING)
1668         return;
1669
1670     // if read failed, close the connection
1671     if (flag != Comm::OK) {
1672         debugs(5, 3, HERE << "closing " << conn);
1673         conn->close();
1674         return;
1675     }
1676
1677     // continue waiting for close or error
1678     commPlanHalfClosedCheck(); // make sure this fd will be checked again
1679 }
1680
1681 CommRead::CommRead() : conn(NULL), buf(NULL), len(0), callback(NULL) {}
1682
1683 CommRead::CommRead(const Comm::ConnectionPointer &c, char *buf_, int len_, AsyncCall::Pointer &callback_)
1684     : conn(c), buf(buf_), len(len_), callback(callback_) {}
1685
1686 DeferredRead::DeferredRead () : theReader(NULL), theContext(NULL), theRead(), cancelled(false) {}
1687
1688 DeferredRead::DeferredRead (DeferrableRead *aReader, void *data, CommRead const &aRead) : theReader(aReader), theContext (data), theRead(aRead), cancelled(false) {}
1689
1690 DeferredReadManager::~DeferredReadManager()
1691 {
1692     flushReads();
1693     assert (deferredReads.empty());
1694 }
1695
1696 /* explicit instantiation required for some systems */
1697
1698 /// \cond AUTODOCS_IGNORE
1699 template cbdata_type CbDataList<DeferredRead>::CBDATA_CbDataList;
1700 /// \endcond
1701
1702 void
1703 DeferredReadManager::delayRead(DeferredRead const &aRead)
1704 {
1705     debugs(5, 3, "Adding deferred read on " << aRead.theRead.conn);
1706     CbDataList<DeferredRead> *temp = deferredReads.push_back(aRead);
1707
1708     // We have to use a global function as a closer and point to temp
1709     // instead of "this" because DeferredReadManager is not a job and
1710     // is not even cbdata protected
1711     // XXX: and yet we use cbdata protection functions on it??
1712     AsyncCall::Pointer closer = commCbCall(5,4,
1713                                            "DeferredReadManager::CloseHandler",
1714                                            CommCloseCbPtrFun(&CloseHandler, temp));
1715     comm_add_close_handler(aRead.theRead.conn->fd, closer);
1716     temp->element.closer = closer; // remeber so that we can cancel
1717 }
1718
1719 void
1720 DeferredReadManager::CloseHandler(const CommCloseCbParams &params)
1721 {
1722     if (!cbdataReferenceValid(params.data))
1723         return;
1724
1725     CbDataList<DeferredRead> *temp = (CbDataList<DeferredRead> *)params.data;
1726
1727     temp->element.closer = NULL;
1728     temp->element.markCancelled();
1729 }
1730
1731 DeferredRead
1732 DeferredReadManager::popHead(CbDataListContainer<DeferredRead> &deferredReads)
1733 {
1734     assert (!deferredReads.empty());
1735
1736     DeferredRead &read = deferredReads.head->element;
1737
1738     // NOTE: at this point the connection has been paused/stalled for an unknown
1739     //       amount of time. We must re-validate that it is active and usable.
1740
1741     // If the connection has been closed already. Cancel this read.
1742     if (!Comm::IsConnOpen(read.theRead.conn)) {
1743         if (read.closer != NULL) {
1744             read.closer->cancel("Connection closed before.");
1745             read.closer = NULL;
1746         }
1747         read.markCancelled();
1748     }
1749
1750     if (!read.cancelled) {
1751         comm_remove_close_handler(read.theRead.conn->fd, read.closer);
1752         read.closer = NULL;
1753     }
1754
1755     DeferredRead result = deferredReads.pop_front();
1756
1757     return result;
1758 }
1759
1760 void
1761 DeferredReadManager::kickReads(int const count)
1762 {
1763     /* if we had CbDataList::size() we could consolidate this and flushReads */
1764
1765     if (count < 1) {
1766         flushReads();
1767         return;
1768     }
1769
1770     size_t remaining = count;
1771
1772     while (!deferredReads.empty() && remaining) {
1773         DeferredRead aRead = popHead(deferredReads);
1774         kickARead(aRead);
1775
1776         if (!aRead.cancelled)
1777             --remaining;
1778     }
1779 }
1780
1781 void
1782 DeferredReadManager::flushReads()
1783 {
1784     CbDataListContainer<DeferredRead> reads;
1785     reads = deferredReads;
1786     deferredReads = CbDataListContainer<DeferredRead>();
1787
1788     // XXX: For fairness this SHOULD randomize the order
1789     while (!reads.empty()) {
1790         DeferredRead aRead = popHead(reads);
1791         kickARead(aRead);
1792     }
1793 }
1794
1795 void
1796 DeferredReadManager::kickARead(DeferredRead const &aRead)
1797 {
1798     if (aRead.cancelled)
1799         return;
1800
1801     if (Comm::IsConnOpen(aRead.theRead.conn) && fd_table[aRead.theRead.conn->fd].closing())
1802         return;
1803
1804     debugs(5, 3, "Kicking deferred read on " << aRead.theRead.conn);
1805
1806     aRead.theReader(aRead.theContext, aRead.theRead);
1807 }
1808
1809 void
1810 DeferredRead::markCancelled()
1811 {
1812     cancelled = true;
1813 }
1814
1815 int
1816 CommSelectEngine::checkEvents(int timeout)
1817 {
1818     static time_t last_timeout = 0;
1819
1820     /* No, this shouldn't be here. But it shouldn't be in each comm handler. -adrian */
1821     if (squid_curtime > last_timeout) {
1822         last_timeout = squid_curtime;
1823         checkTimeouts();
1824     }
1825
1826     switch (Comm::DoSelect(timeout)) {
1827
1828     case Comm::OK:
1829
1830     case Comm::TIMEOUT:
1831         return 0;
1832
1833     case Comm::IDLE:
1834
1835     case Comm::SHUTDOWN:
1836         return EVENT_IDLE;
1837
1838     case Comm::COMM_ERROR:
1839         return EVENT_ERROR;
1840
1841     default:
1842         fatal_dump("comm.cc: Internal error -- this should never happen.");
1843         return EVENT_ERROR;
1844     };
1845 }
1846
1847 /// Create a unix-domain socket (UDS) that only supports FD_MSGHDR I/O.
1848 int
1849 comm_open_uds(int sock_type,
1850               int proto,
1851               struct sockaddr_un* addr,
1852               int flags)
1853 {
1854     // TODO: merge with comm_openex() when Ip::Address becomes NetAddress
1855
1856     int new_socket;
1857
1858     PROF_start(comm_open);
1859     /* Create socket for accepting new connections. */
1860     ++ statCounter.syscalls.sock.sockets;
1861
1862     /* Setup the socket addrinfo details for use */
1863     struct addrinfo AI;
1864     AI.ai_flags = 0;
1865     AI.ai_family = PF_UNIX;
1866     AI.ai_socktype = sock_type;
1867     AI.ai_protocol = proto;
1868     AI.ai_addrlen = SUN_LEN(addr);
1869     AI.ai_addr = (sockaddr*)addr;
1870     AI.ai_canonname = NULL;
1871     AI.ai_next = NULL;
1872
1873     debugs(50, 3, HERE << "Attempt open socket for: " << addr->sun_path);
1874
1875     if ((new_socket = socket(AI.ai_family, AI.ai_socktype, AI.ai_protocol)) < 0) {
1876         /* Increase the number of reserved fd's if calls to socket()
1877          * are failing because the open file table is full.  This
1878          * limits the number of simultaneous clients */
1879
1880         if (limitError(errno)) {
1881             debugs(50, DBG_IMPORTANT, HERE << "socket failure: " << xstrerror());
1882             fdAdjustReserved();
1883         } else {
1884             debugs(50, DBG_CRITICAL, HERE << "socket failure: " << xstrerror());
1885         }
1886
1887         PROF_stop(comm_open);
1888         return -1;
1889     }
1890
1891     debugs(50, 3, "Opened UDS FD " << new_socket << " : family=" << AI.ai_family << ", type=" << AI.ai_socktype << ", protocol=" << AI.ai_protocol);
1892
1893     /* update fdstat */
1894     debugs(50, 5, HERE << "FD " << new_socket << " is a new socket");
1895
1896     assert(!isOpen(new_socket));
1897     fd_open(new_socket, FD_MSGHDR, addr->sun_path);
1898
1899     fdd_table[new_socket].close_file = NULL;
1900
1901     fdd_table[new_socket].close_line = 0;
1902
1903     fd_table[new_socket].sock_family = AI.ai_family;
1904
1905     if (!(flags & COMM_NOCLOEXEC))
1906         commSetCloseOnExec(new_socket);
1907
1908     if (flags & COMM_REUSEADDR)
1909         commSetReuseAddr(new_socket);
1910
1911     if (flags & COMM_NONBLOCKING) {
1912         if (commSetNonBlocking(new_socket) != Comm::OK) {
1913             comm_close(new_socket);
1914             PROF_stop(comm_open);
1915             return -1;
1916         }
1917     }
1918
1919     if (flags & COMM_DOBIND) {
1920         if (commBind(new_socket, AI) != Comm::OK) {
1921             comm_close(new_socket);
1922             PROF_stop(comm_open);
1923             return -1;
1924         }
1925     }
1926
1927 #ifdef TCP_NODELAY
1928     if (sock_type == SOCK_STREAM)
1929         commSetTcpNoDelay(new_socket);
1930
1931 #endif
1932
1933     if (Config.tcpRcvBufsz > 0 && sock_type == SOCK_STREAM)
1934         commSetTcpRcvbuf(new_socket, Config.tcpRcvBufsz);
1935
1936     PROF_stop(comm_open);
1937
1938     return new_socket;
1939 }
1940