contrib/bind-9.2.4rc7/lib/isc/unix/socket.c

   1 /*
   2  * Copyright (C) 2004  Internet Systems Consortium, Inc. ("ISC")
   3  * Copyright (C) 1998-2003  Internet Software Consortium.
   4  *
   5  * Permission to use, copy, modify, and distribute this software for any
   6  * purpose with or without fee is hereby granted, provided that the above
   7  * copyright notice and this permission notice appear in all copies.
   8  *
   9  * THE SOFTWARE IS PROVIDED "AS IS" AND ISC DISCLAIMS ALL WARRANTIES WITH
  10  * REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF MERCHANTABILITY
  11  * AND FITNESS.  IN NO EVENT SHALL ISC BE LIABLE FOR ANY SPECIAL, DIRECT,
  12  * INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING FROM
  13  * LOSS OF USE, DATA OR PROFITS, WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE
  14  * OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR
  15  * PERFORMANCE OF THIS SOFTWARE.
  16  */
  17
  18 /* $Id: socket.c,v 1.207.2.29 2004/07/01 04:51:44 marka Exp $ */
  19
  20 #include <config.h>
  21
  22 #include <sys/param.h>
  23 #include <sys/types.h>
  24 #include <sys/socket.h>
  25 #include <sys/time.h>
  26 #include <sys/uio.h>
  27
  28 #include <errno.h>
  29 #include <fcntl.h>
  30 #include <stddef.h>
  31 #include <stdlib.h>
  32 #include <string.h>
  33 #include <unistd.h>
  34
  35 #include <isc/buffer.h>
  36 #include <isc/bufferlist.h>
  37 #include <isc/condition.h>
  38 #include <isc/formatcheck.h>
  39 #include <isc/list.h>
  40 #include <isc/log.h>
  41 #include <isc/mem.h>
  42 #include <isc/msgs.h>
  43 #include <isc/mutex.h>
  44 #include <isc/net.h>
  45 #include <isc/platform.h>
  46 #include <isc/print.h>
  47 #include <isc/region.h>
  48 #include <isc/socket.h>
  49 #include <isc/strerror.h>
  50 #include <isc/task.h>
  51 #include <isc/thread.h>
  52 #include <isc/util.h>
  53
  54 #include "errno2result.h"
  55
  56 #ifndef ISC_PLATFORM_USETHREADS
  57 #include "socket_p.h"
  58 #endif /* ISC_PLATFORM_USETHREADS */
  59
  60 /*
  61  * Some systems define the socket length argument as an int, some as size_t,
  62  * some as socklen_t.  This is here so it can be easily changed if needed.
  63  */
  64 #ifndef ISC_SOCKADDR_LEN_T
  65 #ifdef _BSD_SOCKLEN_T_
  66 #define ISC_SOCKADDR_LEN_T _BSD_SOCKLEN_T_
  67 #else
  68 #define ISC_SOCKADDR_LEN_T unsigned int
  69 #endif
  70 #endif
  71
  72 /*
  73  * Define what the possible "soft" errors can be.  These are non-fatal returns
  74  * of various network related functions, like recv() and so on.
  75  *
  76  * For some reason, BSDI (and perhaps others) will sometimes return <0
  77  * from recv() but will have errno==0.  This is broken, but we have to
  78  * work around it here.
  79  */
  80 #define SOFT_ERROR(e)   ((e) == EAGAIN || \
  81                          (e) == EWOULDBLOCK || \
  82                          (e) == EINTR || \
  83                          (e) == 0)
  84
  85 #define DLVL(x) ISC_LOGCATEGORY_GENERAL, ISC_LOGMODULE_SOCKET, ISC_LOG_DEBUG(x)
  86
  87 /*
  88  * DLVL(90)  --  Function entry/exit and other tracing.
  89  * DLVL(70)  --  Socket "correctness" -- including returning of events, etc.
  90  * DLVL(60)  --  Socket data send/receive
  91  * DLVL(50)  --  Event tracing, including receiving/sending completion events.
  92  * DLVL(20)  --  Socket creation/destruction.
  93  */
  94 #define TRACE_LEVEL             90
  95 #define CORRECTNESS_LEVEL       70
  96 #define IOEVENT_LEVEL           60
  97 #define EVENT_LEVEL             50
  98 #define CREATION_LEVEL          20
  99
 100 #define TRACE           DLVL(TRACE_LEVEL)
 101 #define CORRECTNESS     DLVL(CORRECTNESS_LEVEL)
 102 #define IOEVENT         DLVL(IOEVENT_LEVEL)
 103 #define EVENT           DLVL(EVENT_LEVEL)
 104 #define CREATION        DLVL(CREATION_LEVEL)
 105
 106 typedef isc_event_t intev_t;
 107
 108 #define SOCKET_MAGIC            ISC_MAGIC('I', 'O', 'i', 'o')
 109 #define VALID_SOCKET(t)         ISC_MAGIC_VALID(t, SOCKET_MAGIC)
 110
 111 /*
 112  * IPv6 control information.  If the socket is an IPv6 socket we want
 113  * to collect the destination address and interface so the client can
 114  * set them on outgoing packets.
 115  */
 116 #ifdef ISC_PLATFORM_HAVEIPV6
 117 #ifndef USE_CMSG
 118 #define USE_CMSG        1
 119 #endif
 120 #endif
 121
 122 /*
 123  * NetBSD and FreeBSD can timestamp packets.  XXXMLG Should we have
 124  * a setsockopt() like interface to request timestamps, and if the OS
 125  * doesn't do it for us, call gettimeofday() on every UDP receive?
 126  */
 127 #ifdef SO_TIMESTAMP
 128 #ifndef USE_CMSG
 129 #define USE_CMSG        1
 130 #endif
 131 #endif
 132
 133 /*
 134  * The number of times a send operation is repeated if the result is EINTR.
 135  */
 136 #define NRETRIES 10
 137
 138 struct isc_socket {
 139         /* Not locked. */
 140         unsigned int            magic;
 141         isc_socketmgr_t        *manager;
 142         isc_mutex_t             lock;
 143         isc_sockettype_t        type;
 144
 145         /* Locked by socket lock. */
 146         ISC_LINK(isc_socket_t)  link;
 147         unsigned int            references;
 148         int                     fd;
 149         int                     pf;
 150
 151         ISC_LIST(isc_socketevent_t)             send_list;
 152         ISC_LIST(isc_socketevent_t)             recv_list;
 153         ISC_LIST(isc_socket_newconnev_t)        accept_list;
 154         isc_socket_connev_t                    *connect_ev;
 155
 156         /*
 157          * Internal events.  Posted when a descriptor is readable or
 158          * writable.  These are statically allocated and never freed.
 159          * They will be set to non-purgable before use.
 160          */
 161         intev_t                 readable_ev;
 162         intev_t                 writable_ev;
 163
 164         isc_sockaddr_t          address;  /* remote address */
 165
 166         unsigned int            pending_recv : 1,
 167                                 pending_send : 1,
 168                                 pending_accept : 1,
 169                                 listener : 1, /* listener socket */
 170                                 connected : 1,
 171                                 connecting : 1, /* connect pending */
 172                                 bound : 1; /* bound to local addr */
 173
 174 #ifdef ISC_NET_RECVOVERFLOW
 175         unsigned char           overflow; /* used for MSG_TRUNC fake */
 176 #endif
 177
 178         char                    *recvcmsgbuf;
 179         ISC_SOCKADDR_LEN_T      recvcmsgbuflen;
 180         char                    *sendcmsgbuf;
 181         ISC_SOCKADDR_LEN_T      sendcmsgbuflen;
 182 };
 183
 184 #define SOCKET_MANAGER_MAGIC    ISC_MAGIC('I', 'O', 'm', 'g')
 185 #define VALID_MANAGER(m)        ISC_MAGIC_VALID(m, SOCKET_MANAGER_MAGIC)
 186
 187 struct isc_socketmgr {
 188         /* Not locked. */
 189         unsigned int            magic;
 190         isc_mem_t              *mctx;
 191         isc_mutex_t             lock;
 192         /* Locked by manager lock. */
 193         ISC_LIST(isc_socket_t)  socklist;
 194         fd_set                  read_fds;
 195         fd_set                  write_fds;
 196         isc_socket_t           *fds[FD_SETSIZE];
 197         int                     fdstate[FD_SETSIZE];
 198         int                     maxfd;
 199 #ifdef ISC_PLATFORM_USETHREADS
 200         isc_thread_t            watcher;
 201         isc_condition_t         shutdown_ok;
 202         int                     pipe_fds[2];
 203 #else /* ISC_PLATFORM_USETHREADS */
 204         unsigned int            refs;
 205 #endif /* ISC_PLATFORM_USETHREADS */
 206 };
 207
 208 #ifndef ISC_PLATFORM_USETHREADS
 209 static isc_socketmgr_t *socketmgr = NULL;
 210 #endif /* ISC_PLATFORM_USETHREADS */
 211
 212 #define CLOSED          0       /* this one must be zero */
 213 #define MANAGED         1
 214 #define CLOSE_PENDING   2
 215
 216 /*
 217  * send() and recv() iovec counts
 218  */
 219 #define MAXSCATTERGATHER_SEND   (ISC_SOCKET_MAXSCATTERGATHER)
 220 #ifdef ISC_NET_RECVOVERFLOW
 221 # define MAXSCATTERGATHER_RECV  (ISC_SOCKET_MAXSCATTERGATHER + 1)
 222 #else
 223 # define MAXSCATTERGATHER_RECV  (ISC_SOCKET_MAXSCATTERGATHER)
 224 #endif
 225
 226 static void send_recvdone_event(isc_socket_t *, isc_socketevent_t **);
 227 static void send_senddone_event(isc_socket_t *, isc_socketevent_t **);
 228 static void free_socket(isc_socket_t **);
 229 static isc_result_t allocate_socket(isc_socketmgr_t *, isc_sockettype_t,
 230                                     isc_socket_t **);
 231 static void destroy(isc_socket_t **);
 232 static void internal_accept(isc_task_t *, isc_event_t *);
 233 static void internal_connect(isc_task_t *, isc_event_t *);
 234 static void internal_recv(isc_task_t *, isc_event_t *);
 235 static void internal_send(isc_task_t *, isc_event_t *);
 236 static void process_cmsg(isc_socket_t *, struct msghdr *, isc_socketevent_t *);
 237 static void build_msghdr_send(isc_socket_t *, isc_socketevent_t *,
 238                               struct msghdr *, struct iovec *, size_t *);
 239 static void build_msghdr_recv(isc_socket_t *, isc_socketevent_t *,
 240                               struct msghdr *, struct iovec *, size_t *);
 241
 242 #define SELECT_POKE_SHUTDOWN            (-1)
 243 #define SELECT_POKE_NOTHING             (-2)
 244 #define SELECT_POKE_READ                (-3)
 245 #define SELECT_POKE_ACCEPT              (-3) /* Same as _READ */
 246 #define SELECT_POKE_WRITE               (-4)
 247 #define SELECT_POKE_CONNECT             (-4) /* Same as _WRITE */
 248 #define SELECT_POKE_CLOSE               (-5)
 249
 250 #define SOCK_DEAD(s)                    ((s)->references == 0)
 251
 252 static void
 253 manager_log(isc_socketmgr_t *sockmgr,
 254             isc_logcategory_t *category, isc_logmodule_t *module, int level,
 255             const char *fmt, ...) ISC_FORMAT_PRINTF(5, 6);
 256 static void
 257 manager_log(isc_socketmgr_t *sockmgr,
 258             isc_logcategory_t *category, isc_logmodule_t *module, int level,
 259             const char *fmt, ...)
 260 {
 261         char msgbuf[2048];
 262         va_list ap;
 263
 264         if (! isc_log_wouldlog(isc_lctx, level))
 265                 return;
 266
 267         va_start(ap, fmt);
 268         vsnprintf(msgbuf, sizeof(msgbuf), fmt, ap);
 269         va_end(ap);
 270
 271         isc_log_write(isc_lctx, category, module, level,
 272                       "sockmgr %p: %s", sockmgr, msgbuf);
 273 }
 274
 275 static void
 276 socket_log(isc_socket_t *sock, isc_sockaddr_t *address,
 277            isc_logcategory_t *category, isc_logmodule_t *module, int level,
 278            isc_msgcat_t *msgcat, int msgset, int message,
 279            const char *fmt, ...) ISC_FORMAT_PRINTF(9, 10);
 280 static void
 281 socket_log(isc_socket_t *sock, isc_sockaddr_t *address,
 282            isc_logcategory_t *category, isc_logmodule_t *module, int level,
 283            isc_msgcat_t *msgcat, int msgset, int message,
 284            const char *fmt, ...)
 285 {
 286         char msgbuf[2048];
 287         char peerbuf[256];
 288         va_list ap;
 289
 290         if (! isc_log_wouldlog(isc_lctx, level))
 291                 return;
 292
 293         va_start(ap, fmt);
 294         vsnprintf(msgbuf, sizeof(msgbuf), fmt, ap);
 295         va_end(ap);
 296
 297         if (address == NULL) {
 298                 isc_log_iwrite(isc_lctx, category, module, level,
 299                                msgcat, msgset, message,
 300                                "socket %p: %s", sock, msgbuf);
 301         } else {
 302                 isc_sockaddr_format(address, peerbuf, sizeof(peerbuf));
 303                 isc_log_iwrite(isc_lctx, category, module, level,
 304                                msgcat, msgset, message,
 305                                "socket %p %s: %s", sock, peerbuf, msgbuf);
 306         }
 307 }
 308
 309 static void
 310 wakeup_socket(isc_socketmgr_t *manager, int fd, int msg) {
 311         isc_socket_t *sock;
 312
 313         /*
 314          * This is a wakeup on a socket.  If the socket is not in the
 315          * process of being closed, start watching it for either reads
 316          * or writes.
 317          */
 318
 319         INSIST(fd >= 0 && fd < (int)FD_SETSIZE);
 320
 321         if (manager->fdstate[fd] == CLOSE_PENDING) {
 322                 manager->fdstate[fd] = CLOSED;
 323                 FD_CLR(fd, &manager->read_fds);
 324                 FD_CLR(fd, &manager->write_fds);
 325                 close(fd);
 326                 return;
 327         }
 328         if (manager->fdstate[fd] != MANAGED)
 329                 return;
 330
 331         sock = manager->fds[fd];
 332
 333         /*
 334          * Set requested bit.
 335          */
 336         if (msg == SELECT_POKE_READ)
 337                 FD_SET(sock->fd, &manager->read_fds);
 338         if (msg == SELECT_POKE_WRITE)
 339                 FD_SET(sock->fd, &manager->write_fds);
 340 }
 341
 342 #ifdef ISC_PLATFORM_USETHREADS
 343 /*
 344  * Poke the select loop when there is something for us to do.
 345  * The write is required (by POSIX) to complete.  That is, we
 346  * will not get partial writes.
 347  */
 348 static void
 349 select_poke(isc_socketmgr_t *mgr, int fd, int msg) {
 350         int cc;
 351         int buf[2];
 352         char strbuf[ISC_STRERRORSIZE];
 353
 354         buf[0] = fd;
 355         buf[1] = msg;
 356
 357         do {
 358                 cc = write(mgr->pipe_fds[1], buf, sizeof(buf));
 359 #ifdef ENOSR
 360                 /*
 361                  * Treat ENOSR as EAGAIN but loop slowly as it is
 362                  * unlikely to clear fast.
 363                  */
 364                 if (cc < 0 && errno == ENOSR) {
 365                         sleep(1);
 366                         errno = EAGAIN;
 367                 }
 368 #endif
 369         } while (cc < 0 && SOFT_ERROR(errno));
 370
 371         if (cc < 0) {
 372                 isc__strerror(errno, strbuf, sizeof(strbuf));
 373                 FATAL_ERROR(__FILE__, __LINE__,
 374                             isc_msgcat_get(isc_msgcat, ISC_MSGSET_SOCKET,
 375                                            ISC_MSG_WRITEFAILED,
 376                                            "write() failed "
 377                                            "during watcher poke: %s"),
 378                             strbuf);
 379         }
 380
 381         INSIST(cc == sizeof(buf));
 382 }
 383
 384 /*
 385  * Read a message on the internal fd.
 386  */
 387 static void
 388 select_readmsg(isc_socketmgr_t *mgr, int *fd, int *msg) {
 389         int buf[2];
 390         int cc;
 391         char strbuf[ISC_STRERRORSIZE];
 392
 393         cc = read(mgr->pipe_fds[0], buf, sizeof(buf));
 394         if (cc < 0) {
 395                 *msg = SELECT_POKE_NOTHING;
 396                 if (SOFT_ERROR(errno))
 397                         return;
 398
 399                 isc__strerror(errno, strbuf, sizeof(strbuf));
 400                 FATAL_ERROR(__FILE__, __LINE__,
 401                             isc_msgcat_get(isc_msgcat, ISC_MSGSET_SOCKET,
 402                                            ISC_MSG_READFAILED,
 403                                            "read() failed "
 404                                            "during watcher poke: %s"),
 405                             strbuf);
 406
 407                 return;
 408         }
 409         INSIST(cc == sizeof(buf));
 410
 411         *fd = buf[0];
 412         *msg = buf[1];
 413 }
 414 #else /* ISC_PLATFORM_USETHREADS */
 415 /*
 416  * Update the state of the socketmgr when something changes.
 417  */
 418 static void
 419 select_poke(isc_socketmgr_t *manager, int fd, int msg) {
 420         if (msg == SELECT_POKE_SHUTDOWN)
 421                 return;
 422         else if (fd >= 0)
 423                 wakeup_socket(manager, fd, msg);
 424         return;
 425 }
 426 #endif /* ISC_PLATFORM_USETHREADS */
 427
 428 /*
 429  * Make a fd non-blocking.
 430  */
 431 static isc_result_t
 432 make_nonblock(int fd) {
 433         int ret;
 434         int flags;
 435         char strbuf[ISC_STRERRORSIZE];
 436
 437         flags = fcntl(fd, F_GETFL, 0);
 438         flags |= O_NONBLOCK;
 439         ret = fcntl(fd, F_SETFL, flags);
 440
 441         if (ret == -1) {
 442                 isc__strerror(errno, strbuf, sizeof(strbuf));
 443                 UNEXPECTED_ERROR(__FILE__, __LINE__,
 444                                  "fcntl(%d, F_SETFL, %d): %s",
 445                                  fd, flags, strbuf);
 446
 447                 return (ISC_R_UNEXPECTED);
 448         }
 449
 450         return (ISC_R_SUCCESS);
 451 }
 452
 453 #ifdef USE_CMSG
 454 /*
 455  * Not all OSes support advanced CMSG macros: CMSG_LEN and CMSG_SPACE.
 456  * In order to ensure as much portability as possible, we provide wrapper
 457  * functions of these macros.
 458  * Note that cmsg_space() could run slow on OSes that do not have
 459  * CMSG_SPACE.
 460  */
 461 static inline ISC_SOCKADDR_LEN_T
 462 cmsg_len(ISC_SOCKADDR_LEN_T len) {
 463 #ifdef CMSG_LEN
 464         return (CMSG_LEN(len));
 465 #else
 466         ISC_SOCKADDR_LEN_T hdrlen;
 467
 468         hdrlen = (ISC_SOCKADDR_LEN_T)CMSG_DATA(NULL); /* XXX */
 469         return (hdrlen + len);
 470 #endif
 471 }
 472
 473 static inline ISC_SOCKADDR_LEN_T
 474 cmsg_space(ISC_SOCKADDR_LEN_T len) {
 475 #ifdef CMSG_SPACE
 476         return (CMSG_SPACE(len));
 477 #else
 478         struct msghdr msg;
 479         struct cmsghdr *cmsgp;
 480         /*
 481          * XXX: The buffer length is an ad-hoc value, but should be enough
 482          * in a practical sense.
 483          */
 484         char dummybuf[sizeof(struct cmsghdr) + 1024];
 485
 486         memset(&msg, 0, sizeof(msg));
 487         msg.msg_control = dummybuf;
 488         msg.msg_controllen = sizeof(dummybuf);
 489
 490         cmsgp = (struct cmsghdr *)dummybuf;
 491         cmsgp->cmsg_len = cmsg_len(len);
 492
 493         cmsgp = CMSG_NXTHDR(&msg, cmsgp);
 494         if (cmsgp != NULL)
 495                 return ((char *)cmsgp - (char *)msg.msg_control);
 496         else
 497                 return (0);
 498 #endif
 499 }
 500 #endif /* USE_CMSG */
 501
 502 /*
 503  * Process control messages received on a socket.
 504  */
 505 static void
 506 process_cmsg(isc_socket_t *sock, struct msghdr *msg, isc_socketevent_t *dev) {
 507 #ifdef USE_CMSG
 508         struct cmsghdr *cmsgp;
 509 #ifdef ISC_PLATFORM_HAVEIPV6
 510         struct in6_pktinfo *pktinfop;
 511 #endif
 512 #ifdef SO_TIMESTAMP
 513         struct timeval *timevalp;
 514 #endif
 515 #endif
 516
 517         /*
 518          * sock is used only when ISC_NET_BSD44MSGHDR and USE_CMSG are defined.
 519          * msg and dev are used only when ISC_NET_BSD44MSGHDR is defined.
 520          * They are all here, outside of the CPP tests, because it is
 521          * more consistent with the usual ISC coding style.
 522          */
 523         UNUSED(sock);
 524         UNUSED(msg);
 525         UNUSED(dev);
 526
 527 #ifdef ISC_NET_BSD44MSGHDR
 528
 529 #ifdef MSG_TRUNC
 530         if ((msg->msg_flags & MSG_TRUNC) == MSG_TRUNC)
 531                 dev->attributes |= ISC_SOCKEVENTATTR_TRUNC;
 532 #endif
 533
 534 #ifdef MSG_CTRUNC
 535         if ((msg->msg_flags & MSG_CTRUNC) == MSG_CTRUNC)
 536                 dev->attributes |= ISC_SOCKEVENTATTR_CTRUNC;
 537 #endif
 538
 539 #ifndef USE_CMSG
 540         return;
 541 #else
 542         if (msg->msg_controllen == 0U || msg->msg_control == NULL)
 543                 return;
 544
 545 #ifdef SO_TIMESTAMP
 546         timevalp = NULL;
 547 #endif
 548 #ifdef ISC_PLATFORM_HAVEIPV6
 549         pktinfop = NULL;
 550 #endif
 551
 552         cmsgp = CMSG_FIRSTHDR(msg);
 553         while (cmsgp != NULL) {
 554                 socket_log(sock, NULL, TRACE,
 555                            isc_msgcat, ISC_MSGSET_SOCKET, ISC_MSG_PROCESSCMSG,
 556                            "processing cmsg %p", cmsgp);
 557
 558 #ifdef ISC_PLATFORM_HAVEIPV6
 559                 if (cmsgp->cmsg_level == IPPROTO_IPV6
 560                     && cmsgp->cmsg_type == IPV6_PKTINFO) {
 561
 562                         pktinfop = (struct in6_pktinfo *)CMSG_DATA(cmsgp);
 563                         memcpy(&dev->pktinfo, pktinfop,
 564                                sizeof(struct in6_pktinfo));
 565                         dev->attributes |= ISC_SOCKEVENTATTR_PKTINFO;
 566                         socket_log(sock, NULL, TRACE,
 567                                    isc_msgcat, ISC_MSGSET_SOCKET,
 568                                    ISC_MSG_IFRECEIVED,
 569                                    "interface received on ifindex %u",
 570                                    dev->pktinfo.ipi6_ifindex);
 571                         if (IN6_IS_ADDR_MULTICAST(&pktinfop->ipi6_addr))
 572                                 dev->attributes |= ISC_SOCKEVENTATTR_MULTICAST;
 573                         goto next;
 574                 }
 575 #endif
 576
 577 #ifdef SO_TIMESTAMP
 578                 if (cmsgp->cmsg_level == SOL_SOCKET
 579                     && cmsgp->cmsg_type == SCM_TIMESTAMP) {
 580                         timevalp = (struct timeval *)CMSG_DATA(cmsgp);
 581                         dev->timestamp.seconds = timevalp->tv_sec;
 582                         dev->timestamp.nanoseconds = timevalp->tv_usec * 1000;
 583                         dev->attributes |= ISC_SOCKEVENTATTR_TIMESTAMP;
 584                         goto next;
 585                 }
 586 #endif
 587
 588         next:
 589                 cmsgp = CMSG_NXTHDR(msg, cmsgp);
 590         }
 591 #endif /* USE_CMSG */
 592
 593 #endif /* ISC_NET_BSD44MSGHDR */
 594 }
 595
 596 /*
 597  * Construct an iov array and attach it to the msghdr passed in.  This is
 598  * the SEND constructor, which will use the used region of the buffer
 599  * (if using a buffer list) or will use the internal region (if a single
 600  * buffer I/O is requested).
 601  *
 602  * Nothing can be NULL, and the done event must list at least one buffer
 603  * on the buffer linked list for this function to be meaningful.
 604  *
 605  * If write_countp != NULL, *write_countp will hold the number of bytes
 606  * this transaction can send.
 607  */
 608 static void
 609 build_msghdr_send(isc_socket_t *sock, isc_socketevent_t *dev,
 610                   struct msghdr *msg, struct iovec *iov, size_t *write_countp)
 611 {
 612         unsigned int iovcount;
 613         isc_buffer_t *buffer;
 614         isc_region_t used;
 615         size_t write_count;
 616         size_t skip_count;
 617
 618         memset(msg, 0, sizeof(*msg));
 619
 620         if (sock->type == isc_sockettype_udp) {
 621                 msg->msg_name = (void *)&dev->address.type.sa;
 622                 msg->msg_namelen = dev->address.length;
 623         } else {
 624                 msg->msg_name = NULL;
 625                 msg->msg_namelen = 0;
 626         }
 627
 628         buffer = ISC_LIST_HEAD(dev->bufferlist);
 629         write_count = 0;
 630         iovcount = 0;
 631
 632         /*
 633          * Single buffer I/O?  Skip what we've done so far in this region.
 634          */
 635         if (buffer == NULL) {
 636                 write_count = dev->region.length - dev->n;
 637                 iov[0].iov_base = (void *)(dev->region.base + dev->n);
 638                 iov[0].iov_len = write_count;
 639                 iovcount = 1;
 640
 641                 goto config;
 642         }
 643
 644         /*
 645          * Multibuffer I/O.
 646          * Skip the data in the buffer list that we have already written.
 647          */
 648         skip_count = dev->n;
 649         while (buffer != NULL) {
 650                 REQUIRE(ISC_BUFFER_VALID(buffer));
 651                 if (skip_count < isc_buffer_usedlength(buffer))
 652                         break;
 653                 skip_count -= isc_buffer_usedlength(buffer);
 654                 buffer = ISC_LIST_NEXT(buffer, link);
 655         }
 656
 657         while (buffer != NULL) {
 658                 INSIST(iovcount < MAXSCATTERGATHER_SEND);
 659
 660                 isc_buffer_usedregion(buffer, &used);
 661
 662                 if (used.length > 0) {
 663                         iov[iovcount].iov_base = (void *)(used.base
 664                                                           + skip_count);
 665                         iov[iovcount].iov_len = used.length - skip_count;
 666                         write_count += (used.length - skip_count);
 667                         skip_count = 0;
 668                         iovcount++;
 669                 }
 670                 buffer = ISC_LIST_NEXT(buffer, link);
 671         }
 672
 673         INSIST(skip_count == 0U);
 674
 675  config:
 676         msg->msg_iov = iov;
 677         msg->msg_iovlen = iovcount;
 678
 679 #ifdef ISC_NET_BSD44MSGHDR
 680         msg->msg_control = NULL;
 681         msg->msg_controllen = 0;
 682         msg->msg_flags = 0;
 683 #if defined(USE_CMSG) && defined(ISC_PLATFORM_HAVEIPV6)
 684         if ((sock->type == isc_sockettype_udp)
 685             && ((dev->attributes & ISC_SOCKEVENTATTR_PKTINFO) != 0)) {
 686                 struct cmsghdr *cmsgp;
 687                 struct in6_pktinfo *pktinfop;
 688
 689                 socket_log(sock, NULL, TRACE,
 690                            isc_msgcat, ISC_MSGSET_SOCKET, ISC_MSG_SENDTODATA,
 691                            "sendto pktinfo data, ifindex %u",
 692                            dev->pktinfo.ipi6_ifindex);
 693
 694                 msg->msg_controllen = cmsg_space(sizeof(struct in6_pktinfo));
 695                 INSIST(msg->msg_controllen <= sock->sendcmsgbuflen);
 696                 msg->msg_control = (void *)sock->sendcmsgbuf;
 697
 698                 cmsgp = (struct cmsghdr *)sock->sendcmsgbuf;
 699                 cmsgp->cmsg_level = IPPROTO_IPV6;
 700                 cmsgp->cmsg_type = IPV6_PKTINFO;
 701                 cmsgp->cmsg_len = cmsg_len(sizeof(struct in6_pktinfo));
 702                 pktinfop = (struct in6_pktinfo *)CMSG_DATA(cmsgp);
 703                 memcpy(pktinfop, &dev->pktinfo, sizeof(struct in6_pktinfo));
 704         }
 705 #endif /* USE_CMSG && ISC_PLATFORM_HAVEIPV6 */
 706 #else /* ISC_NET_BSD44MSGHDR */
 707         msg->msg_accrights = NULL;
 708         msg->msg_accrightslen = 0;
 709 #endif /* ISC_NET_BSD44MSGHDR */
 710
 711         if (write_countp != NULL)
 712                 *write_countp = write_count;
 713 }
 714
 715 /*
 716  * Construct an iov array and attach it to the msghdr passed in.  This is
 717  * the RECV constructor, which will use the avialable region of the buffer
 718  * (if using a buffer list) or will use the internal region (if a single
 719  * buffer I/O is requested).
 720  *
 721  * Nothing can be NULL, and the done event must list at least one buffer
 722  * on the buffer linked list for this function to be meaningful.
 723  *
 724  * If read_countp != NULL, *read_countp will hold the number of bytes
 725  * this transaction can receive.
 726  */
 727 static void
 728 build_msghdr_recv(isc_socket_t *sock, isc_socketevent_t *dev,
 729                   struct msghdr *msg, struct iovec *iov, size_t *read_countp)
 730 {
 731         unsigned int iovcount;
 732         isc_buffer_t *buffer;
 733         isc_region_t available;
 734         size_t read_count;
 735
 736         memset(msg, 0, sizeof(struct msghdr));
 737
 738         if (sock->type == isc_sockettype_udp) {
 739                 memset(&dev->address, 0, sizeof(dev->address));
 740                 msg->msg_name = (void *)&dev->address.type.sa;
 741                 msg->msg_namelen = sizeof(dev->address.type);
 742 #ifdef ISC_NET_RECVOVERFLOW
 743                 /* If needed, steal one iovec for overflow detection. */
 744                 maxiov--;
 745 #endif
 746         } else { /* TCP */
 747                 msg->msg_name = NULL;
 748                 msg->msg_namelen = 0;
 749                 dev->address = sock->address;
 750         }
 751
 752         buffer = ISC_LIST_HEAD(dev->bufferlist);
 753         read_count = 0;
 754
 755         /*
 756          * Single buffer I/O?  Skip what we've done so far in this region.
 757          */
 758         if (buffer == NULL) {
 759                 read_count = dev->region.length - dev->n;
 760                 iov[0].iov_base = (void *)(dev->region.base + dev->n);
 761                 iov[0].iov_len = read_count;
 762                 iovcount = 1;
 763
 764                 goto config;
 765         }
 766
 767         /*
 768          * Multibuffer I/O.
 769          * Skip empty buffers.
 770          */
 771         while (buffer != NULL) {
 772                 REQUIRE(ISC_BUFFER_VALID(buffer));
 773                 if (isc_buffer_availablelength(buffer) != 0)
 774                         break;
 775                 buffer = ISC_LIST_NEXT(buffer, link);
 776         }
 777
 778         iovcount = 0;
 779         while (buffer != NULL) {
 780                 INSIST(iovcount < MAXSCATTERGATHER_RECV);
 781
 782                 isc_buffer_availableregion(buffer, &available);
 783
 784                 if (available.length > 0) {
 785                         iov[iovcount].iov_base = (void *)(available.base);
 786                         iov[iovcount].iov_len = available.length;
 787                         read_count += available.length;
 788                         iovcount++;
 789                 }
 790                 buffer = ISC_LIST_NEXT(buffer, link);
 791         }
 792
 793  config:
 794
 795         /*
 796          * If needed, set up to receive that one extra byte.  Note that
 797          * we know there is at least one iov left, since we stole it
 798          * at the top of this function.
 799          */
 800 #ifdef ISC_NET_RECVOVERFLOW
 801         if (sock->type == isc_sockettype_udp) {
 802                 iov[iovcount].iov_base = (void *)(&sock->overflow);
 803                 iov[iovcount].iov_len = 1;
 804                 iovcount++;
 805         }
 806 #endif
 807
 808         msg->msg_iov = iov;
 809         msg->msg_iovlen = iovcount;
 810
 811 #ifdef ISC_NET_BSD44MSGHDR
 812         msg->msg_control = NULL;
 813         msg->msg_controllen = 0;
 814         msg->msg_flags = 0;
 815 #if defined(USE_CMSG)
 816         if (sock->type == isc_sockettype_udp) {
 817                 msg->msg_control = sock->recvcmsgbuf;
 818                 msg->msg_controllen = sock->recvcmsgbuflen;
 819         }
 820 #endif /* USE_CMSG */
 821 #else /* ISC_NET_BSD44MSGHDR */
 822         msg->msg_accrights = NULL;
 823         msg->msg_accrightslen = 0;
 824 #endif /* ISC_NET_BSD44MSGHDR */
 825
 826         if (read_countp != NULL)
 827                 *read_countp = read_count;
 828 }
 829
 830 static void
 831 set_dev_address(isc_sockaddr_t *address, isc_socket_t *sock,
 832                 isc_socketevent_t *dev)
 833 {
 834         if (sock->type == isc_sockettype_udp) {
 835                 if (address != NULL)
 836                         dev->address = *address;
 837                 else
 838                         dev->address = sock->address;
 839         } else if (sock->type == isc_sockettype_tcp) {
 840                 INSIST(address == NULL);
 841                 dev->address = sock->address;
 842         }
 843 }
 844
 845 static isc_socketevent_t *
 846 allocate_socketevent(isc_socket_t *sock, isc_eventtype_t eventtype,
 847                      isc_taskaction_t action, const void *arg)
 848 {
 849         isc_socketevent_t *ev;
 850
 851         ev = (isc_socketevent_t *)isc_event_allocate(sock->manager->mctx,
 852                                                      sock, eventtype,
 853                                                      action, arg,
 854                                                      sizeof(*ev));
 855
 856         if (ev == NULL)
 857                 return (NULL);
 858
 859         ev->result = ISC_R_UNEXPECTED;
 860         ISC_LINK_INIT(ev, ev_link);
 861         ISC_LIST_INIT(ev->bufferlist);
 862         ev->region.base = NULL;
 863         ev->n = 0;
 864         ev->offset = 0;
 865         ev->attributes = 0;
 866
 867         return (ev);
 868 }
 869
 870 #if defined(ISC_SOCKET_DEBUG)
 871 static void
 872 dump_msg(struct msghdr *msg) {
 873         unsigned int i;
 874
 875         printf("MSGHDR %p\n", msg);
 876         printf("\tname %p, namelen %d\n", msg->msg_name, msg->msg_namelen);
 877         printf("\tiov %p, iovlen %d\n", msg->msg_iov, msg->msg_iovlen);
 878         for (i = 0 ; i < (unsigned int)msg->msg_iovlen ; i++)
 879                 printf("\t\t%d\tbase %p, len %d\n", i,
 880                        msg->msg_iov[i].iov_base,
 881                        msg->msg_iov[i].iov_len);
 882 #ifdef ISC_NET_BSD44MSGHDR
 883         printf("\tcontrol %p, controllen %d\n", msg->msg_control,
 884                msg->msg_controllen);
 885 #endif
 886 }
 887 #endif
 888
 889 #define DOIO_SUCCESS            0       /* i/o ok, event sent */
 890 #define DOIO_SOFT               1       /* i/o ok, soft error, no event sent */
 891 #define DOIO_HARD               2       /* i/o error, event sent */
 892 #define DOIO_EOF                3       /* EOF, no event sent */
 893
 894 static int
 895 doio_recv(isc_socket_t *sock, isc_socketevent_t *dev) {
 896         int cc;
 897         struct iovec iov[MAXSCATTERGATHER_RECV];
 898         size_t read_count;
 899         size_t actual_count;
 900         struct msghdr msghdr;
 901         isc_buffer_t *buffer;
 902         int recv_errno;
 903         char strbuf[ISC_STRERRORSIZE];
 904
 905         build_msghdr_recv(sock, dev, &msghdr, iov, &read_count);
 906
 907 #if defined(ISC_SOCKET_DEBUG)
 908         dump_msg(&msghdr);
 909 #endif
 910
 911         cc = recvmsg(sock->fd, &msghdr, 0);
 912         recv_errno = errno;
 913
 914         if (cc < 0) {
 915                 if (SOFT_ERROR(recv_errno))
 916                         return (DOIO_SOFT);
 917
 918                 if (isc_log_wouldlog(isc_lctx, IOEVENT_LEVEL)) {
 919                         isc__strerror(recv_errno, strbuf, sizeof(strbuf));
 920                         socket_log(sock, NULL, IOEVENT,
 921                                    isc_msgcat, ISC_MSGSET_SOCKET,
 922                                    ISC_MSG_DOIORECV,
 923                                   "doio_recv: recvmsg(%d) %d bytes, err %d/%s",
 924                                    sock->fd, cc, recv_errno, strbuf);
 925                 }
 926
 927 #define SOFT_OR_HARD(_system, _isc) \
 928         if (recv_errno == _system) { \
 929                 if (sock->connected) { \
 930                         dev->result = _isc; \
 931                         return (DOIO_HARD); \
 932                 } \
 933                 return (DOIO_SOFT); \
 934         }
 935 #define ALWAYS_HARD(_system, _isc) \
 936         if (recv_errno == _system) { \
 937                 dev->result = _isc; \
 938                 return (DOIO_HARD); \
 939         }
 940
 941                 SOFT_OR_HARD(ECONNREFUSED, ISC_R_CONNREFUSED);
 942                 SOFT_OR_HARD(ENETUNREACH, ISC_R_NETUNREACH);
 943                 SOFT_OR_HARD(EHOSTUNREACH, ISC_R_HOSTUNREACH);
 944                 SOFT_OR_HARD(EHOSTDOWN, ISC_R_HOSTDOWN);
 945                 /* HPUX 11.11 can return EADDRNOTAVAIL. */
 946                 SOFT_OR_HARD(EADDRNOTAVAIL, ISC_R_ADDRNOTAVAIL);
 947                 ALWAYS_HARD(ENOBUFS, ISC_R_NORESOURCES);
 948
 949 #undef SOFT_OR_HARD
 950 #undef ALWAYS_HARD
 951
 952                 dev->result = isc__errno2result(recv_errno);
 953                 return (DOIO_HARD);
 954         }
 955
 956         /*
 957          * On TCP, zero length reads indicate EOF, while on
 958          * UDP, zero length reads are perfectly valid, although
 959          * strange.
 960          */
 961         if ((sock->type == isc_sockettype_tcp) && (cc == 0))
 962                 return (DOIO_EOF);
 963
 964         if (sock->type == isc_sockettype_udp) {
 965                 dev->address.length = msghdr.msg_namelen;
 966                 if (isc_sockaddr_getport(&dev->address) == 0) {
 967                         if (isc_log_wouldlog(isc_lctx, IOEVENT_LEVEL)) {
 968                                 socket_log(sock, &dev->address, IOEVENT,
 969                                            isc_msgcat, ISC_MSGSET_SOCKET,
 970                                            ISC_MSG_ZEROPORT,
 971                                            "dropping source port zero packet");
 972                         }
 973                         return (DOIO_SOFT);
 974                 }
 975         }
 976
 977         socket_log(sock, &dev->address, IOEVENT,
 978                    isc_msgcat, ISC_MSGSET_SOCKET, ISC_MSG_PKTRECV,
 979                    "packet received correctly");
 980
 981         /*
 982          * Overflow bit detection.  If we received MORE bytes than we should,
 983          * this indicates an overflow situation.  Set the flag in the
 984          * dev entry and adjust how much we read by one.
 985          */
 986 #ifdef ISC_NET_RECVOVERFLOW
 987         if ((sock->type == isc_sockettype_udp) && ((size_t)cc > read_count)) {
 988                 dev->attributes |= ISC_SOCKEVENTATTR_TRUNC;
 989                 cc--;
 990         }
 991 #endif
 992
 993         /*
 994          * If there are control messages attached, run through them and pull
 995          * out the interesting bits.
 996          */
 997         if (sock->type == isc_sockettype_udp)
 998                 process_cmsg(sock, &msghdr, dev);
 999
1000         /*
1001          * update the buffers (if any) and the i/o count
1002          */
1003         dev->n += cc;
1004         actual_count = cc;
1005         buffer = ISC_LIST_HEAD(dev->bufferlist);
1006         while (buffer != NULL && actual_count > 0U) {
1007                 REQUIRE(ISC_BUFFER_VALID(buffer));
1008                 if (isc_buffer_availablelength(buffer) <= actual_count) {
1009                         actual_count -= isc_buffer_availablelength(buffer);
1010                         isc_buffer_add(buffer,
1011                                        isc_buffer_availablelength(buffer));
1012                 } else {
1013                         isc_buffer_add(buffer, actual_count);
1014                         actual_count = 0;
1015                         break;
1016                 }
1017                 buffer = ISC_LIST_NEXT(buffer, link);
1018                 if (buffer == NULL) {
1019                         INSIST(actual_count == 0U);
1020                 }
1021         }
1022
1023         /*
1024          * If we read less than we expected, update counters,
1025          * and let the upper layer poke the descriptor.
1026          */
1027         if (((size_t)cc != read_count) && (dev->n < dev->minimum))
1028                 return (DOIO_SOFT);
1029
1030         /*
1031          * Full reads are posted, or partials if partials are ok.
1032          */
1033         dev->result = ISC_R_SUCCESS;
1034         return (DOIO_SUCCESS);
1035 }
1036
1037 /*
1038  * Returns:
1039  *      DOIO_SUCCESS    The operation succeeded.  dev->result contains
1040  *                      ISC_R_SUCCESS.
1041  *
1042  *      DOIO_HARD       A hard or unexpected I/O error was encountered.
1043  *                      dev->result contains the appropriate error.
1044  *
1045  *      DOIO_SOFT       A soft I/O error was encountered.  No senddone
1046  *                      event was sent.  The operation should be retried.
1047  *
1048  *      No other return values are possible.
1049  */
1050 static int
1051 doio_send(isc_socket_t *sock, isc_socketevent_t *dev) {
1052         int cc;
1053         struct iovec iov[MAXSCATTERGATHER_SEND];
1054         size_t write_count;
1055         struct msghdr msghdr;
1056         char addrbuf[ISC_SOCKADDR_FORMATSIZE];
1057         int attempts = 0;
1058         int send_errno;
1059         char strbuf[ISC_STRERRORSIZE];
1060
1061         build_msghdr_send(sock, dev, &msghdr, iov, &write_count);
1062
1063  resend:
1064         cc = sendmsg(sock->fd, &msghdr, 0);
1065         send_errno = errno;
1066
1067         /*
1068          * Check for error or block condition.
1069          */
1070         if (cc < 0) {
1071                 if (send_errno == EINTR && ++attempts < NRETRIES)
1072                         goto resend;
1073
1074                 if (SOFT_ERROR(send_errno))
1075                         return (DOIO_SOFT);
1076
1077 #define SOFT_OR_HARD(_system, _isc) \
1078         if (send_errno == _system) { \
1079                 if (sock->connected) { \
1080                         dev->result = _isc; \
1081                         return (DOIO_HARD); \
1082                 } \
1083                 return (DOIO_SOFT); \
1084         }
1085 #define ALWAYS_HARD(_system, _isc) \
1086         if (send_errno == _system) { \
1087                 dev->result = _isc; \
1088                 return (DOIO_HARD); \
1089         }
1090
1091                 SOFT_OR_HARD(ECONNREFUSED, ISC_R_CONNREFUSED);
1092                 ALWAYS_HARD(EACCES, ISC_R_NOPERM);
1093                 ALWAYS_HARD(EAFNOSUPPORT, ISC_R_ADDRNOTAVAIL);
1094                 ALWAYS_HARD(EADDRNOTAVAIL, ISC_R_ADDRNOTAVAIL);
1095                 ALWAYS_HARD(EHOSTUNREACH, ISC_R_HOSTUNREACH);
1096 #ifdef EHOSTDOWN
1097                 ALWAYS_HARD(EHOSTDOWN, ISC_R_HOSTUNREACH);
1098 #endif
1099                 ALWAYS_HARD(ENETUNREACH, ISC_R_NETUNREACH);
1100                 ALWAYS_HARD(ENOBUFS, ISC_R_NORESOURCES);
1101                 ALWAYS_HARD(EPERM, ISC_R_HOSTUNREACH);
1102                 ALWAYS_HARD(EPIPE, ISC_R_NOTCONNECTED);
1103                 ALWAYS_HARD(ECONNRESET, ISC_R_CONNECTIONRESET);
1104
1105 #undef SOFT_OR_HARD
1106 #undef ALWAYS_HARD
1107
1108                 /*
1109                  * The other error types depend on whether or not the
1110                  * socket is UDP or TCP.  If it is UDP, some errors
1111                  * that we expect to be fatal under TCP are merely
1112                  * annoying, and are really soft errors.
1113                  *
1114                  * However, these soft errors are still returned as
1115                  * a status.
1116                  */
1117                 isc_sockaddr_format(&dev->address, addrbuf, sizeof(addrbuf));
1118                 isc__strerror(send_errno, strbuf, sizeof(strbuf));
1119                 UNEXPECTED_ERROR(__FILE__, __LINE__, "internal_send: %s: %s",
1120                                  addrbuf, strbuf);
1121                 dev->result = isc__errno2result(send_errno);
1122                 return (DOIO_HARD);
1123         }
1124
1125         if (cc == 0)
1126                 UNEXPECTED_ERROR(__FILE__, __LINE__,
1127                                  "internal_send: send() %s 0",
1128                                  isc_msgcat_get(isc_msgcat, ISC_MSGSET_GENERAL,
1129                                                 ISC_MSG_RETURNED, "returned"));
1130
1131         /*
1132          * If we write less than we expected, update counters, poke.
1133          */
1134         dev->n += cc;
1135         if ((size_t)cc != write_count)
1136                 return (DOIO_SOFT);
1137
1138         /*
1139          * Exactly what we wanted to write.  We're done with this
1140          * entry.  Post its completion event.
1141          */
1142         dev->result = ISC_R_SUCCESS;
1143         return (DOIO_SUCCESS);
1144 }
1145
1146 /*
1147  * Kill.
1148  *
1149  * Caller must ensure that the socket is not locked and no external
1150  * references exist.
1151  */
1152 static void
1153 destroy(isc_socket_t **sockp) {
1154         isc_socket_t *sock = *sockp;
1155         isc_socketmgr_t *manager = sock->manager;
1156
1157         socket_log(sock, NULL, CREATION, isc_msgcat, ISC_MSGSET_SOCKET,
1158                    ISC_MSG_DESTROYING, "destroying");
1159
1160         INSIST(ISC_LIST_EMPTY(sock->accept_list));
1161         INSIST(ISC_LIST_EMPTY(sock->recv_list));
1162         INSIST(ISC_LIST_EMPTY(sock->send_list));
1163         INSIST(sock->connect_ev == NULL);
1164         REQUIRE(sock->fd >= 0 && sock->fd < (int)FD_SETSIZE);
1165
1166         LOCK(&manager->lock);
1167
1168         /*
1169          * No one has this socket open, so the watcher doesn't have to be
1170          * poked, and the socket doesn't have to be locked.
1171          */
1172         manager->fds[sock->fd] = NULL;
1173         manager->fdstate[sock->fd] = CLOSE_PENDING;
1174         select_poke(manager, sock->fd, SELECT_POKE_CLOSE);
1175         ISC_LIST_UNLINK(manager->socklist, sock, link);
1176
1177 #ifdef ISC_PLATFORM_USETHREADS
1178         if (ISC_LIST_EMPTY(manager->socklist))
1179                 SIGNAL(&manager->shutdown_ok);
1180 #endif /* ISC_PLATFORM_USETHREADS */
1181
1182         /*
1183          * XXX should reset manager->maxfd here
1184          */
1185
1186         UNLOCK(&manager->lock);
1187
1188         free_socket(sockp);
1189 }
1190
1191 static isc_result_t
1192 allocate_socket(isc_socketmgr_t *manager, isc_sockettype_t type,
1193                 isc_socket_t **socketp)
1194 {
1195         isc_socket_t *sock;
1196         isc_result_t ret;
1197         ISC_SOCKADDR_LEN_T cmsgbuflen;
1198
1199         sock = isc_mem_get(manager->mctx, sizeof(*sock));
1200
1201         if (sock == NULL)
1202                 return (ISC_R_NOMEMORY);
1203
1204         ret = ISC_R_UNEXPECTED;
1205
1206         sock->magic = 0;
1207         sock->references = 0;
1208
1209         sock->manager = manager;
1210         sock->type = type;
1211         sock->fd = -1;
1212
1213         ISC_LINK_INIT(sock, link);
1214
1215         sock->recvcmsgbuf = NULL;
1216         sock->sendcmsgbuf = NULL;
1217
1218         /*
1219          * set up cmsg buffers
1220          */
1221         cmsgbuflen = 0;
1222 #if defined(USE_CMSG) && defined(ISC_PLATFORM_HAVEIPV6)
1223         cmsgbuflen = cmsg_space(sizeof(struct in6_pktinfo));
1224 #endif
1225 #if defined(USE_CMSG) && defined(SO_TIMESTAMP)
1226         cmsgbuflen += cmsg_space(sizeof(struct timeval));
1227 #endif
1228         sock->recvcmsgbuflen = cmsgbuflen;
1229         if (sock->recvcmsgbuflen != 0) {
1230                 sock->recvcmsgbuf = isc_mem_get(manager->mctx, cmsgbuflen);
1231                 if (sock->recvcmsgbuf == NULL)
1232                         goto error;
1233         }
1234
1235         cmsgbuflen = 0;
1236 #if defined(USE_CMSG) && defined(ISC_PLATFORM_HAVEIPV6)
1237         cmsgbuflen = cmsg_space(sizeof(struct in6_pktinfo));
1238 #endif
1239         sock->sendcmsgbuflen = cmsgbuflen;
1240         if (sock->sendcmsgbuflen != 0) {
1241                 sock->sendcmsgbuf = isc_mem_get(manager->mctx, cmsgbuflen);
1242                 if (sock->sendcmsgbuf == NULL)
1243                         goto error;
1244         }
1245
1246         /*
1247          * set up list of readers and writers to be initially empty
1248          */
1249         ISC_LIST_INIT(sock->recv_list);
1250         ISC_LIST_INIT(sock->send_list);
1251         ISC_LIST_INIT(sock->accept_list);
1252         sock->connect_ev = NULL;
1253         sock->pending_recv = 0;
1254         sock->pending_send = 0;
1255         sock->pending_accept = 0;
1256         sock->listener = 0;
1257         sock->connected = 0;
1258         sock->connecting = 0;
1259         sock->bound = 0;
1260
1261         /*
1262          * initialize the lock
1263          */
1264         if (isc_mutex_init(&sock->lock) != ISC_R_SUCCESS) {
1265                 sock->magic = 0;
1266                 UNEXPECTED_ERROR(__FILE__, __LINE__,
1267                                  "isc_mutex_init() %s",
1268                                  isc_msgcat_get(isc_msgcat, ISC_MSGSET_GENERAL,
1269                                                 ISC_MSG_FAILED, "failed"));
1270                 ret = ISC_R_UNEXPECTED;
1271                 goto error;
1272         }
1273
1274         /*
1275          * Initialize readable and writable events
1276          */
1277         ISC_EVENT_INIT(&sock->readable_ev, sizeof(intev_t),
1278                        ISC_EVENTATTR_NOPURGE, NULL, ISC_SOCKEVENT_INTR,
1279                        NULL, sock, sock, NULL, NULL);
1280         ISC_EVENT_INIT(&sock->writable_ev, sizeof(intev_t),
1281                        ISC_EVENTATTR_NOPURGE, NULL, ISC_SOCKEVENT_INTW,
1282                        NULL, sock, sock, NULL, NULL);
1283
1284         sock->magic = SOCKET_MAGIC;
1285         *socketp = sock;
1286
1287         return (ISC_R_SUCCESS);
1288
1289  error:
1290         if (sock->recvcmsgbuf != NULL)
1291                 isc_mem_put(manager->mctx, sock->recvcmsgbuf,
1292                             sock->recvcmsgbuflen);
1293         if (sock->sendcmsgbuf != NULL)
1294                 isc_mem_put(manager->mctx, sock->sendcmsgbuf,
1295                             sock->sendcmsgbuflen);
1296         isc_mem_put(manager->mctx, sock, sizeof(*sock));
1297
1298         return (ret);
1299 }
1300
1301 /*
1302  * This event requires that the various lists be empty, that the reference
1303  * count be 1, and that the magic number is valid.  The other socket bits,
1304  * like the lock, must be initialized as well.  The fd associated must be
1305  * marked as closed, by setting it to -1 on close, or this routine will
1306  * also close the socket.
1307  */
1308 static void
1309 free_socket(isc_socket_t **socketp) {
1310         isc_socket_t *sock = *socketp;
1311
1312         INSIST(sock->references == 0);
1313         INSIST(VALID_SOCKET(sock));
1314         INSIST(!sock->connecting);
1315         INSIST(!sock->pending_recv);
1316         INSIST(!sock->pending_send);
1317         INSIST(!sock->pending_accept);
1318         INSIST(ISC_LIST_EMPTY(sock->recv_list));
1319         INSIST(ISC_LIST_EMPTY(sock->send_list));
1320         INSIST(ISC_LIST_EMPTY(sock->accept_list));
1321         INSIST(!ISC_LINK_LINKED(sock, link));
1322
1323         if (sock->recvcmsgbuf != NULL)
1324                 isc_mem_put(sock->manager->mctx, sock->recvcmsgbuf,
1325                             sock->recvcmsgbuflen);
1326         if (sock->sendcmsgbuf != NULL)
1327                 isc_mem_put(sock->manager->mctx, sock->sendcmsgbuf,
1328                             sock->sendcmsgbuflen);
1329
1330         sock->magic = 0;
1331
1332         DESTROYLOCK(&sock->lock);
1333
1334         isc_mem_put(sock->manager->mctx, sock, sizeof(*sock));
1335
1336         *socketp = NULL;
1337 }
1338
1339 /*
1340  * Create a new 'type' socket managed by 'manager'.  Events
1341  * will be posted to 'task' and when dispatched 'action' will be
1342  * called with 'arg' as the arg value.  The new socket is returned
1343  * in 'socketp'.
1344  */
1345 isc_result_t
1346 isc_socket_create(isc_socketmgr_t *manager, int pf, isc_sockettype_t type,
1347                   isc_socket_t **socketp)
1348 {
1349         isc_socket_t *sock = NULL;
1350         isc_result_t ret;
1351 #if defined(USE_CMSG) || defined(SO_BSDCOMPAT)
1352         int on = 1;
1353 #endif
1354         char strbuf[ISC_STRERRORSIZE];
1355
1356         REQUIRE(VALID_MANAGER(manager));
1357         REQUIRE(socketp != NULL && *socketp == NULL);
1358
1359         ret = allocate_socket(manager, type, &sock);
1360         if (ret != ISC_R_SUCCESS)
1361                 return (ret);
1362
1363         sock->pf = pf;
1364         switch (type) {
1365         case isc_sockettype_udp:
1366                 sock->fd = socket(pf, SOCK_DGRAM, IPPROTO_UDP);
1367                 break;
1368         case isc_sockettype_tcp:
1369                 sock->fd = socket(pf, SOCK_STREAM, IPPROTO_TCP);
1370                 break;
1371         }
1372
1373 #ifdef F_DUPFD
1374         /*
1375          * Leave a space for stdio to work in.
1376          */
1377         if (sock->fd >= 0 && sock->fd < 20) {
1378                 int new, tmp;
1379                 new = fcntl(sock->fd, F_DUPFD, 20);
1380                 tmp = errno;
1381                 (void)close(sock->fd);
1382                 errno = tmp;
1383                 sock->fd = new;
1384         }
1385 #endif
1386
1387         if (sock->fd >= (int)FD_SETSIZE) {
1388                 (void)close(sock->fd);
1389                 isc_log_iwrite(isc_lctx, ISC_LOGCATEGORY_GENERAL,
1390                               ISC_LOGMODULE_SOCKET, ISC_LOG_ERROR,
1391                                isc_msgcat, ISC_MSGSET_SOCKET,
1392                                ISC_MSG_TOOMANYFDS,
1393                                "%s: too many open file descriptors", "socket");
1394                 free_socket(&sock);
1395                 return (ISC_R_NORESOURCES);
1396         }
1397
1398         if (sock->fd < 0) {
1399                 free_socket(&sock);
1400
1401                 switch (errno) {
1402                 case EMFILE:
1403                 case ENFILE:
1404                 case ENOBUFS:
1405                         return (ISC_R_NORESOURCES);
1406
1407                 case EPROTONOSUPPORT:
1408                 case EPFNOSUPPORT:
1409                 case EAFNOSUPPORT:
1410                 /*
1411                  * Linux 2.2 (and maybe others) return EINVAL instead of
1412                  * EAFNOSUPPORT.
1413                  */
1414                 case EINVAL:
1415                         return (ISC_R_FAMILYNOSUPPORT);
1416
1417                 default:
1418                         isc__strerror(errno, strbuf, sizeof(strbuf));
1419                         UNEXPECTED_ERROR(__FILE__, __LINE__,
1420                                          "socket() %s: %s",
1421                                          isc_msgcat_get(isc_msgcat,
1422                                                         ISC_MSGSET_GENERAL,
1423                                                         ISC_MSG_FAILED,
1424                                                         "failed"),
1425                                          strbuf);
1426                         return (ISC_R_UNEXPECTED);
1427                 }
1428         }
1429
1430         if (make_nonblock(sock->fd) != ISC_R_SUCCESS) {
1431                 (void)close(sock->fd);
1432                 free_socket(&sock);
1433                 return (ISC_R_UNEXPECTED);
1434         }
1435
1436 #ifdef SO_BSDCOMPAT
1437         if (setsockopt(sock->fd, SOL_SOCKET, SO_BSDCOMPAT,
1438                        (void *)&on, sizeof(on)) < 0) {
1439                 isc__strerror(errno, strbuf, sizeof(strbuf));
1440                 UNEXPECTED_ERROR(__FILE__, __LINE__,
1441                                  "setsockopt(%d, SO_BSDCOMPAT) %s: %s",
1442                                  sock->fd,
1443                                  isc_msgcat_get(isc_msgcat, ISC_MSGSET_GENERAL,
1444                                                 ISC_MSG_FAILED, "failed"),
1445                                  strbuf);
1446                 /* Press on... */
1447         }
1448 #endif
1449
1450 #if defined(USE_CMSG)
1451         if (type == isc_sockettype_udp) {
1452
1453 #if defined(SO_TIMESTAMP)
1454                 if (setsockopt(sock->fd, SOL_SOCKET, SO_TIMESTAMP,
1455                                (void *)&on, sizeof(on)) < 0
1456                     && errno != ENOPROTOOPT) {
1457                         isc__strerror(errno, strbuf, sizeof(strbuf));
1458                         UNEXPECTED_ERROR(__FILE__, __LINE__,
1459                                          "setsockopt(%d, SO_TIMESTAMP) %s: %s",
1460                                          sock->fd,
1461                                          isc_msgcat_get(isc_msgcat,
1462                                                         ISC_MSGSET_GENERAL,
1463                                                         ISC_MSG_FAILED,
1464                                                         "failed"),
1465                                          strbuf);
1466                         /* Press on... */
1467                 }
1468 #endif /* SO_TIMESTAMP */
1469
1470 #if defined(ISC_PLATFORM_HAVEIPV6)
1471                 if (pf == AF_INET6 && sock->recvcmsgbuflen == 0) {
1472                         /*
1473                          * Warn explicitly because this anomaly can be hidden
1474                          * in usual operation (and unexpectedly appear later).
1475                          */
1476                         UNEXPECTED_ERROR(__FILE__, __LINE__,
1477                                          "No buffer available to receive "
1478                                          "IPv6 destination");
1479                 }
1480 #ifdef IPV6_RECVPKTINFO
1481                 /* 2292bis */
1482                 if ((pf == AF_INET6)
1483                     && (setsockopt(sock->fd, IPPROTO_IPV6, IPV6_RECVPKTINFO,
1484                                    (void *)&on, sizeof(on)) < 0)) {
1485                         isc__strerror(errno, strbuf, sizeof(strbuf));
1486                         UNEXPECTED_ERROR(__FILE__, __LINE__,
1487                                          "setsockopt(%d, IPV6_RECVPKTINFO) "
1488                                          "%s: %s", sock->fd,
1489                                          isc_msgcat_get(isc_msgcat,
1490                                                         ISC_MSGSET_GENERAL,
1491                                                         ISC_MSG_FAILED,
1492                                                         "failed"),
1493                                          strbuf);
1494                 }
1495 #else
1496                 /* 2292 */
1497                 if ((pf == AF_INET6)
1498                     && (setsockopt(sock->fd, IPPROTO_IPV6, IPV6_PKTINFO,
1499                                    (void *)&on, sizeof(on)) < 0)) {
1500                         isc__strerror(errno, strbuf, sizeof(strbuf));
1501                         UNEXPECTED_ERROR(__FILE__, __LINE__,
1502                                          "setsockopt(%d, IPV6_PKTINFO) %s: %s",
1503                                          sock->fd,
1504                                          isc_msgcat_get(isc_msgcat,
1505                                                         ISC_MSGSET_GENERAL,
1506                                                         ISC_MSG_FAILED,
1507                                                         "failed"),
1508                                          strbuf);
1509                 }
1510 #endif /* IPV6_RECVPKTINFO */
1511 #ifdef IPV6_USE_MIN_MTU        /*2292bis, not too common yet*/
1512                 /* use minimum MTU */
1513                 if (pf == AF_INET6) {
1514                         (void)setsockopt(sock->fd, IPPROTO_IPV6,
1515                                          IPV6_USE_MIN_MTU,
1516                                          (void *)&on, sizeof(on));
1517                 }
1518 #endif
1519 #endif /* ISC_PLATFORM_HAVEIPV6 */
1520
1521         }
1522 #endif /* USE_CMSG */
1523
1524         sock->references = 1;
1525         *socketp = sock;
1526
1527         LOCK(&manager->lock);
1528
1529         /*
1530          * Note we don't have to lock the socket like we normally would because
1531          * there are no external references to it yet.
1532          */
1533
1534         manager->fds[sock->fd] = sock;
1535         manager->fdstate[sock->fd] = MANAGED;
1536         ISC_LIST_APPEND(manager->socklist, sock, link);
1537         if (manager->maxfd < sock->fd)
1538                 manager->maxfd = sock->fd;
1539
1540         UNLOCK(&manager->lock);
1541
1542         socket_log(sock, NULL, CREATION, isc_msgcat, ISC_MSGSET_SOCKET,
1543                    ISC_MSG_CREATED, "created");
1544
1545         return (ISC_R_SUCCESS);
1546 }
1547
1548 /*
1549  * Attach to a socket.  Caller must explicitly detach when it is done.
1550  */
1551 void
1552 isc_socket_attach(isc_socket_t *sock, isc_socket_t **socketp) {
1553         REQUIRE(VALID_SOCKET(sock));
1554         REQUIRE(socketp != NULL && *socketp == NULL);
1555
1556         LOCK(&sock->lock);
1557         sock->references++;
1558         UNLOCK(&sock->lock);
1559
1560         *socketp = sock;
1561 }
1562
1563 /*
1564  * Dereference a socket.  If this is the last reference to it, clean things
1565  * up by destroying the socket.
1566  */
1567 void
1568 isc_socket_detach(isc_socket_t **socketp) {
1569         isc_socket_t *sock;
1570         isc_boolean_t kill_socket = ISC_FALSE;
1571
1572         REQUIRE(socketp != NULL);
1573         sock = *socketp;
1574         REQUIRE(VALID_SOCKET(sock));
1575
1576         LOCK(&sock->lock);
1577         REQUIRE(sock->references > 0);
1578         sock->references--;
1579         if (sock->references == 0)
1580                 kill_socket = ISC_TRUE;
1581         UNLOCK(&sock->lock);
1582
1583         if (kill_socket)
1584                 destroy(&sock);
1585
1586         *socketp = NULL;
1587 }
1588
1589 /*
1590  * I/O is possible on a given socket.  Schedule an event to this task that
1591  * will call an internal function to do the I/O.  This will charge the
1592  * task with the I/O operation and let our select loop handler get back
1593  * to doing something real as fast as possible.
1594  *
1595  * The socket and manager must be locked before calling this function.
1596  */
1597 static void
1598 dispatch_recv(isc_socket_t *sock) {
1599         intev_t *iev;
1600         isc_socketevent_t *ev;
1601
1602         INSIST(!sock->pending_recv);
1603
1604         ev = ISC_LIST_HEAD(sock->recv_list);
1605         if (ev == NULL)
1606                 return;
1607
1608         sock->pending_recv = 1;
1609         iev = &sock->readable_ev;
1610
1611         socket_log(sock, NULL, EVENT, NULL, 0, 0,
1612                    "dispatch_recv:  event %p -> task %p", ev, ev->ev_sender);
1613
1614         sock->references++;
1615         iev->ev_sender = sock;
1616         iev->ev_action = internal_recv;
1617         iev->ev_arg = sock;
1618
1619         isc_task_send(ev->ev_sender, (isc_event_t **)&iev);
1620 }
1621
1622 static void
1623 dispatch_send(isc_socket_t *sock) {
1624         intev_t *iev;
1625         isc_socketevent_t *ev;
1626
1627         INSIST(!sock->pending_send);
1628
1629         ev = ISC_LIST_HEAD(sock->send_list);
1630         if (ev == NULL)
1631                 return;
1632
1633         sock->pending_send = 1;
1634         iev = &sock->writable_ev;
1635
1636         socket_log(sock, NULL, EVENT, NULL, 0, 0,
1637                    "dispatch_send:  event %p -> task %p", ev, ev->ev_sender);
1638
1639         sock->references++;
1640         iev->ev_sender = sock;
1641         iev->ev_action = internal_send;
1642         iev->ev_arg = sock;
1643
1644         isc_task_send(ev->ev_sender, (isc_event_t **)&iev);
1645 }
1646
1647 /*
1648  * Dispatch an internal accept event.
1649  */
1650 static void
1651 dispatch_accept(isc_socket_t *sock) {
1652         intev_t *iev;
1653         isc_socket_newconnev_t *ev;
1654
1655         INSIST(!sock->pending_accept);
1656
1657         /*
1658          * Are there any done events left, or were they all canceled
1659          * before the manager got the socket lock?
1660          */
1661         ev = ISC_LIST_HEAD(sock->accept_list);
1662         if (ev == NULL)
1663                 return;
1664
1665         sock->pending_accept = 1;
1666         iev = &sock->readable_ev;
1667
1668         sock->references++;  /* keep socket around for this internal event */
1669         iev->ev_sender = sock;
1670         iev->ev_action = internal_accept;
1671         iev->ev_arg = sock;
1672
1673         isc_task_send(ev->ev_sender, (isc_event_t **)&iev);
1674 }
1675
1676 static void
1677 dispatch_connect(isc_socket_t *sock) {
1678         intev_t *iev;
1679         isc_socket_connev_t *ev;
1680
1681         iev = &sock->writable_ev;
1682
1683         ev = sock->connect_ev;
1684         INSIST(ev != NULL); /* XXX */
1685
1686         INSIST(sock->connecting);
1687
1688         sock->references++;  /* keep socket around for this internal event */
1689         iev->ev_sender = sock;
1690         iev->ev_action = internal_connect;
1691         iev->ev_arg = sock;
1692
1693         isc_task_send(ev->ev_sender, (isc_event_t **)&iev);
1694 }
1695
1696 /*
1697  * Dequeue an item off the given socket's read queue, set the result code
1698  * in the done event to the one provided, and send it to the task it was
1699  * destined for.
1700  *
1701  * If the event to be sent is on a list, remove it before sending.  If
1702  * asked to, send and detach from the socket as well.
1703  *
1704  * Caller must have the socket locked if the event is attached to the socket.
1705  */
1706 static void
1707 send_recvdone_event(isc_socket_t *sock, isc_socketevent_t **dev) {
1708         isc_task_t *task;
1709
1710         task = (*dev)->ev_sender;
1711
1712         (*dev)->ev_sender = sock;
1713
1714         if (ISC_LINK_LINKED(*dev, ev_link))
1715                 ISC_LIST_DEQUEUE(sock->recv_list, *dev, ev_link);
1716
1717         if (((*dev)->attributes & ISC_SOCKEVENTATTR_ATTACHED)
1718             == ISC_SOCKEVENTATTR_ATTACHED)
1719                 isc_task_sendanddetach(&task, (isc_event_t **)dev);
1720         else
1721                 isc_task_send(task, (isc_event_t **)dev);
1722 }
1723
1724 /*
1725  * See comments for send_recvdone_event() above.
1726  *
1727  * Caller must have the socket locked if the event is attached to the socket.
1728  */
1729 static void
1730 send_senddone_event(isc_socket_t *sock, isc_socketevent_t **dev) {
1731         isc_task_t *task;
1732
1733         INSIST(dev != NULL && *dev != NULL);
1734
1735         task = (*dev)->ev_sender;
1736         (*dev)->ev_sender = sock;
1737
1738         if (ISC_LINK_LINKED(*dev, ev_link))
1739                 ISC_LIST_DEQUEUE(sock->send_list, *dev, ev_link);
1740
1741         if (((*dev)->attributes & ISC_SOCKEVENTATTR_ATTACHED)
1742             == ISC_SOCKEVENTATTR_ATTACHED)
1743                 isc_task_sendanddetach(&task, (isc_event_t **)dev);
1744         else
1745                 isc_task_send(task, (isc_event_t **)dev);
1746 }
1747
1748 /*
1749  * Call accept() on a socket, to get the new file descriptor.  The listen
1750  * socket is used as a prototype to create a new isc_socket_t.  The new
1751  * socket has one outstanding reference.  The task receiving the event
1752  * will be detached from just after the event is delivered.
1753  *
1754  * On entry to this function, the event delivered is the internal
1755  * readable event, and the first item on the accept_list should be
1756  * the done event we want to send.  If the list is empty, this is a no-op,
1757  * so just unlock and return.
1758  */
1759 static void
1760 internal_accept(isc_task_t *me, isc_event_t *ev) {
1761         isc_socket_t *sock;
1762         isc_socketmgr_t *manager;
1763         isc_socket_newconnev_t *dev;
1764         isc_task_t *task;
1765         ISC_SOCKADDR_LEN_T addrlen;
1766         int fd;
1767         isc_result_t result = ISC_R_SUCCESS;
1768         char strbuf[ISC_STRERRORSIZE];
1769
1770         UNUSED(me);
1771
1772         sock = ev->ev_sender;
1773         INSIST(VALID_SOCKET(sock));
1774
1775         LOCK(&sock->lock);
1776         socket_log(sock, NULL, TRACE,
1777                    isc_msgcat, ISC_MSGSET_SOCKET, ISC_MSG_ACCEPTLOCK,
1778                    "internal_accept called, locked socket");
1779
1780         manager = sock->manager;
1781         INSIST(VALID_MANAGER(manager));
1782
1783         INSIST(sock->listener);
1784         INSIST(sock->pending_accept == 1);
1785         sock->pending_accept = 0;
1786
1787         INSIST(sock->references > 0);
1788         sock->references--;  /* the internal event is done with this socket */
1789         if (sock->references == 0) {
1790                 UNLOCK(&sock->lock);
1791                 destroy(&sock);
1792                 return;
1793         }
1794
1795         /*
1796          * Get the first item off the accept list.
1797          * If it is empty, unlock the socket and return.
1798          */
1799         dev = ISC_LIST_HEAD(sock->accept_list);
1800         if (dev == NULL) {
1801                 UNLOCK(&sock->lock);
1802                 return;
1803         }
1804
1805         /*
1806          * Try to accept the new connection.  If the accept fails with
1807          * EAGAIN or EINTR, simply poke the watcher to watch this socket
1808          * again.  Also ignore ECONNRESET, which has been reported to
1809          * be spuriously returned on Linux 2.2.19 although it is not
1810          * a documented error for accept().  ECONNABORTED has been
1811          * reported for Solaris 8.  The rest are thrown in not because
1812          * we have seen them but because they are ignored by other
1813          * deamons such as BIND 8 and Apache.
1814          */
1815
1816         addrlen = sizeof(dev->newsocket->address.type);
1817         memset(&dev->newsocket->address.type.sa, 0, addrlen);
1818         fd = accept(sock->fd, &dev->newsocket->address.type.sa,
1819                     (void *)&addrlen);
1820
1821 #ifdef F_DUPFD
1822         /*
1823          * Leave a space for stdio to work in.
1824          */
1825         if (fd >= 0 && fd < 20) {
1826                 int new, tmp;
1827                 new = fcntl(fd, F_DUPFD, 20);
1828                 tmp = errno;
1829                 (void)close(fd);
1830                 errno = tmp;
1831                 fd = new;
1832         }
1833 #endif
1834
1835         if (fd < 0) {
1836                 if (SOFT_ERROR(errno))
1837                         goto soft_error;
1838                 switch (errno) {
1839                 case ENOBUFS:
1840                 case ENFILE:
1841                 case ENOMEM:
1842                 case ECONNRESET:
1843                 case ECONNABORTED:
1844                 case EHOSTUNREACH:
1845                 case EHOSTDOWN:
1846                 case ENETUNREACH:
1847                 case ENETDOWN:
1848                 case ECONNREFUSED:
1849 #ifdef EPROTO
1850                 case EPROTO:
1851 #endif
1852 #ifdef ENONET
1853                 case ENONET:
1854 #endif
1855                         goto soft_error;
1856                 default:
1857                         break;
1858                 }
1859                 isc__strerror(errno, strbuf, sizeof(strbuf));
1860                 UNEXPECTED_ERROR(__FILE__, __LINE__,
1861                                  "internal_accept: accept() %s: %s",
1862                                  isc_msgcat_get(isc_msgcat,
1863                                                 ISC_MSGSET_GENERAL,
1864                                                 ISC_MSG_FAILED,
1865                                                 "failed"),
1866                                  strbuf);
1867                 fd = -1;
1868                 result = ISC_R_UNEXPECTED;
1869         } else {
1870                 if (addrlen == 0) {
1871                         UNEXPECTED_ERROR(__FILE__, __LINE__,
1872                                          "internal_accept(): "
1873                                          "accept() failed to return "
1874                                          "remote address");
1875
1876                         (void)close(fd);
1877                         goto soft_error;
1878                 } else if (dev->newsocket->address.type.sa.sa_family !=
1879                            sock->pf)
1880                 {
1881                         UNEXPECTED_ERROR(__FILE__, __LINE__,
1882                                          "internal_accept(): "
1883                                          "accept() returned peer address "
1884                                          "family %u (expected %u)",
1885                                          dev->newsocket->address.
1886                                          type.sa.sa_family,
1887                                          sock->pf);
1888                         (void)close(fd);
1889                         goto soft_error;
1890                 } else if (fd >= (int)FD_SETSIZE) {
1891                         isc_log_iwrite(isc_lctx, ISC_LOGCATEGORY_GENERAL,
1892                                        ISC_LOGMODULE_SOCKET, ISC_LOG_ERROR,
1893                                        isc_msgcat, ISC_MSGSET_SOCKET,
1894                                        ISC_MSG_TOOMANYFDS,
1895                                        "%s: too many open file descriptors",
1896                                        "accept");
1897                         (void)close(fd);
1898                         goto soft_error;
1899                 }
1900         }
1901
1902         if (fd != -1) {
1903                 dev->newsocket->address.length = addrlen;
1904                 dev->newsocket->pf = sock->pf;
1905         }
1906
1907         /*
1908          * Pull off the done event.
1909          */
1910         ISC_LIST_UNLINK(sock->accept_list, dev, ev_link);
1911
1912         /*
1913          * Poke watcher if there are more pending accepts.
1914          */
1915         if (!ISC_LIST_EMPTY(sock->accept_list))
1916                 select_poke(sock->manager, sock->fd, SELECT_POKE_ACCEPT);
1917
1918         UNLOCK(&sock->lock);
1919
1920         if (fd != -1 && (make_nonblock(fd) != ISC_R_SUCCESS)) {
1921                 close(fd);
1922                 fd = -1;
1923                 result = ISC_R_UNEXPECTED;
1924         }
1925
1926         /*
1927          * -1 means the new socket didn't happen.
1928          */
1929         if (fd != -1) {
1930                 LOCK(&manager->lock);
1931                 ISC_LIST_APPEND(manager->socklist, dev->newsocket, link);
1932
1933                 dev->newsocket->fd = fd;
1934                 dev->newsocket->bound = 1;
1935                 dev->newsocket->connected = 1;
1936
1937                 /*
1938                  * Save away the remote address
1939                  */
1940                 dev->address = dev->newsocket->address;
1941
1942                 manager->fds[fd] = dev->newsocket;
1943                 manager->fdstate[fd] = MANAGED;
1944                 if (manager->maxfd < fd)
1945                         manager->maxfd = fd;
1946
1947                 socket_log(sock, &dev->newsocket->address, CREATION,
1948                            isc_msgcat, ISC_MSGSET_SOCKET, ISC_MSG_ACCEPTEDCXN,
1949                            "accepted connection, new socket %p",
1950                            dev->newsocket);
1951
1952                 UNLOCK(&manager->lock);
1953         } else {
1954                 dev->newsocket->references--;
1955                 free_socket(&dev->newsocket);
1956         }
1957
1958         /*
1959          * Fill in the done event details and send it off.
1960          */
1961         dev->result = result;
1962         task = dev->ev_sender;
1963         dev->ev_sender = sock;
1964
1965         isc_task_sendanddetach(&task, ISC_EVENT_PTR(&dev));
1966         return;
1967
1968  soft_error:
1969         select_poke(sock->manager, sock->fd, SELECT_POKE_ACCEPT);
1970         UNLOCK(&sock->lock);
1971         return;
1972 }
1973
1974 static void
1975 internal_recv(isc_task_t *me, isc_event_t *ev) {
1976         isc_socketevent_t *dev;
1977         isc_socket_t *sock;
1978
1979         INSIST(ev->ev_type == ISC_SOCKEVENT_INTR);
1980
1981         sock = ev->ev_sender;
1982         INSIST(VALID_SOCKET(sock));
1983
1984         LOCK(&sock->lock);
1985         socket_log(sock, NULL, IOEVENT,
1986                    isc_msgcat, ISC_MSGSET_SOCKET, ISC_MSG_INTERNALRECV,
1987                    "internal_recv: task %p got event %p", me, ev);
1988
1989         INSIST(sock->pending_recv == 1);
1990         sock->pending_recv = 0;
1991
1992         INSIST(sock->references > 0);
1993         sock->references--;  /* the internal event is done with this socket */
1994         if (sock->references == 0) {
1995                 UNLOCK(&sock->lock);
1996                 destroy(&sock);
1997                 return;
1998         }
1999
2000         /*
2001          * Try to do as much I/O as possible on this socket.  There are no
2002          * limits here, currently.
2003          */
2004         dev = ISC_LIST_HEAD(sock->recv_list);
2005         while (dev != NULL) {
2006                 switch (doio_recv(sock, dev)) {
2007                 case DOIO_SOFT:
2008                         goto poke;
2009
2010                 case DOIO_EOF:
2011                         /*
2012                          * read of 0 means the remote end was closed.
2013                          * Run through the event queue and dispatch all
2014                          * the events with an EOF result code.
2015                          */
2016                         do {
2017                                 dev->result = ISC_R_EOF;
2018                                 send_recvdone_event(sock, &dev);
2019                                 dev = ISC_LIST_HEAD(sock->recv_list);
2020                         } while (dev != NULL);
2021                         goto poke;
2022
2023                 case DOIO_SUCCESS:
2024                 case DOIO_HARD:
2025                         send_recvdone_event(sock, &dev);
2026                         break;
2027                 }
2028
2029                 dev = ISC_LIST_HEAD(sock->recv_list);
2030         }
2031
2032  poke:
2033         if (!ISC_LIST_EMPTY(sock->recv_list))
2034                 select_poke(sock->manager, sock->fd, SELECT_POKE_READ);
2035
2036         UNLOCK(&sock->lock);
2037 }
2038
2039 static void
2040 internal_send(isc_task_t *me, isc_event_t *ev) {
2041         isc_socketevent_t *dev;
2042         isc_socket_t *sock;
2043
2044         INSIST(ev->ev_type == ISC_SOCKEVENT_INTW);
2045
2046         /*
2047          * Find out what socket this is and lock it.
2048          */
2049         sock = (isc_socket_t *)ev->ev_sender;
2050         INSIST(VALID_SOCKET(sock));
2051
2052         LOCK(&sock->lock);
2053         socket_log(sock, NULL, IOEVENT,
2054                    isc_msgcat, ISC_MSGSET_SOCKET, ISC_MSG_INTERNALSEND,
2055                    "internal_send: task %p got event %p", me, ev);
2056
2057         INSIST(sock->pending_send == 1);
2058         sock->pending_send = 0;
2059
2060         INSIST(sock->references > 0);
2061         sock->references--;  /* the internal event is done with this socket */
2062         if (sock->references == 0) {
2063                 UNLOCK(&sock->lock);
2064                 destroy(&sock);
2065                 return;
2066         }
2067
2068         /*
2069          * Try to do as much I/O as possible on this socket.  There are no
2070          * limits here, currently.
2071          */
2072         dev = ISC_LIST_HEAD(sock->send_list);
2073         while (dev != NULL) {
2074                 switch (doio_send(sock, dev)) {
2075                 case DOIO_SOFT:
2076                         goto poke;
2077
2078                 case DOIO_HARD:
2079                 case DOIO_SUCCESS:
2080                         send_senddone_event(sock, &dev);
2081                         break;
2082                 }
2083
2084                 dev = ISC_LIST_HEAD(sock->send_list);
2085         }
2086
2087  poke:
2088         if (!ISC_LIST_EMPTY(sock->send_list))
2089                 select_poke(sock->manager, sock->fd, SELECT_POKE_WRITE);
2090
2091         UNLOCK(&sock->lock);
2092 }
2093
2094 static void
2095 process_fds(isc_socketmgr_t *manager, int maxfd,
2096             fd_set *readfds, fd_set *writefds)
2097 {
2098         int i;
2099         isc_socket_t *sock;
2100         isc_boolean_t unlock_sock;
2101
2102         REQUIRE(maxfd <= (int)FD_SETSIZE);
2103
2104         /*
2105          * Process read/writes on other fds here.  Avoid locking
2106          * and unlocking twice if both reads and writes are possible.
2107          */
2108         for (i = 0 ; i < maxfd ; i++) {
2109 #ifdef ISC_PLATFORM_USETHREADS
2110                 if (i == manager->pipe_fds[0] || i == manager->pipe_fds[1])
2111                         continue;
2112 #endif /* ISC_PLATFORM_USETHREADS */
2113
2114                 if (manager->fdstate[i] == CLOSE_PENDING) {
2115                         manager->fdstate[i] = CLOSED;
2116                         FD_CLR(i, &manager->read_fds);
2117                         FD_CLR(i, &manager->write_fds);
2118
2119                         close(i);
2120
2121                         continue;
2122                 }
2123
2124                 sock = manager->fds[i];
2125                 unlock_sock = ISC_FALSE;
2126                 if (FD_ISSET(i, readfds)) {
2127                         if (sock == NULL) {
2128                                 FD_CLR(i, &manager->read_fds);
2129                                 goto check_write;
2130                         }
2131                         unlock_sock = ISC_TRUE;
2132                         LOCK(&sock->lock);
2133                         if (!SOCK_DEAD(sock)) {
2134                                 if (sock->listener)
2135                                         dispatch_accept(sock);
2136                                 else
2137                                         dispatch_recv(sock);
2138                         }
2139                         FD_CLR(i, &manager->read_fds);
2140                 }
2141         check_write:
2142                 if (FD_ISSET(i, writefds)) {
2143                         if (sock == NULL) {
2144                                 FD_CLR(i, &manager->write_fds);
2145                                 continue;
2146                         }
2147                         if (!unlock_sock) {
2148                                 unlock_sock = ISC_TRUE;
2149                                 LOCK(&sock->lock);
2150                         }
2151                         if (!SOCK_DEAD(sock)) {
2152                                 if (sock->connecting)
2153                                         dispatch_connect(sock);
2154                                 else
2155                                         dispatch_send(sock);
2156                         }
2157                         FD_CLR(i, &manager->write_fds);
2158                 }
2159                 if (unlock_sock)
2160                         UNLOCK(&sock->lock);
2161         }
2162 }
2163
2164 #ifdef ISC_PLATFORM_USETHREADS
2165 /*
2166  * This is the thread that will loop forever, always in a select or poll
2167  * call.
2168  *
2169  * When select returns something to do, track down what thread gets to do
2170  * this I/O and post the event to it.
2171  */
2172 static isc_threadresult_t
2173 watcher(void *uap) {
2174         isc_socketmgr_t *manager = uap;
2175         isc_boolean_t done;
2176         int ctlfd;
2177         int cc;
2178         fd_set readfds;
2179         fd_set writefds;
2180         int msg, fd;
2181         int maxfd;
2182         char strbuf[ISC_STRERRORSIZE];
2183
2184         /*
2185          * Get the control fd here.  This will never change.
2186          */
2187         LOCK(&manager->lock);
2188         ctlfd = manager->pipe_fds[0];
2189
2190         done = ISC_FALSE;
2191         while (!done) {
2192                 do {
2193                         readfds = manager->read_fds;
2194                         writefds = manager->write_fds;
2195                         maxfd = manager->maxfd + 1;
2196
2197                         UNLOCK(&manager->lock);
2198
2199                         cc = select(maxfd, &readfds, &writefds, NULL, NULL);
2200                         if (cc < 0) {
2201                                 if (!SOFT_ERROR(errno)) {
2202                                         isc__strerror(errno, strbuf,
2203                                                       sizeof(strbuf));
2204                                         FATAL_ERROR(__FILE__, __LINE__,
2205                                                     "select() %s: %s",
2206                                                     isc_msgcat_get(isc_msgcat,
2207                                                             ISC_MSGSET_GENERAL,
2208                                                             ISC_MSG_FAILED,
2209                                                             "failed"),
2210                                                     strbuf);
2211                                 }
2212                         }
2213
2214                         LOCK(&manager->lock);
2215                 } while (cc < 0);
2216
2217
2218                 /*
2219                  * Process reads on internal, control fd.
2220                  */
2221                 if (FD_ISSET(ctlfd, &readfds)) {
2222                         for (;;) {
2223                                 select_readmsg(manager, &fd, &msg);
2224
2225                                 manager_log(manager, IOEVENT,
2226                                             isc_msgcat_get(isc_msgcat,
2227                                                      ISC_MSGSET_SOCKET,
2228                                                      ISC_MSG_WATCHERMSG,
2229                                                      "watcher got message %d"),
2230                                                      msg);
2231
2232                                 /*
2233                                  * Nothing to read?
2234                                  */
2235                                 if (msg == SELECT_POKE_NOTHING)
2236                                         break;
2237
2238                                 /*
2239                                  * Handle shutdown message.  We really should
2240                                  * jump out of this loop right away, but
2241                                  * it doesn't matter if we have to do a little
2242                                  * more work first.
2243                                  */
2244                                 if (msg == SELECT_POKE_SHUTDOWN) {
2245                                         done = ISC_TRUE;
2246
2247                                         break;
2248                                 }
2249
2250                                 /*
2251                                  * This is a wakeup on a socket.  Look
2252                                  * at the event queue for both read and write,
2253                                  * and decide if we need to watch on it now
2254                                  * or not.
2255                                  */
2256                                 wakeup_socket(manager, fd, msg);
2257                         }
2258                 }
2259
2260                 process_fds(manager, maxfd, &readfds, &writefds);
2261         }
2262
2263         manager_log(manager, TRACE,
2264                     isc_msgcat_get(isc_msgcat, ISC_MSGSET_GENERAL,
2265                                    ISC_MSG_EXITING, "watcher exiting"));
2266
2267         UNLOCK(&manager->lock);
2268         return ((isc_threadresult_t)0);
2269 }
2270 #endif /* ISC_PLATFORM_USETHREADS */
2271
2272 /*
2273  * Create a new socket manager.
2274  */
2275 isc_result_t
2276 isc_socketmgr_create(isc_mem_t *mctx, isc_socketmgr_t **managerp) {
2277         isc_socketmgr_t *manager;
2278 #ifdef ISC_PLATFORM_USETHREADS
2279         char strbuf[ISC_STRERRORSIZE];
2280 #endif
2281
2282         REQUIRE(managerp != NULL && *managerp == NULL);
2283
2284 #ifndef ISC_PLATFORM_USETHREADS
2285         if (socketmgr != NULL) {
2286                 socketmgr->refs++;
2287                 *managerp = socketmgr;
2288                 return (ISC_R_SUCCESS);
2289         }
2290 #endif /* ISC_PLATFORM_USETHREADS */
2291
2292         manager = isc_mem_get(mctx, sizeof(*manager));
2293         if (manager == NULL)
2294                 return (ISC_R_NOMEMORY);
2295
2296         manager->magic = SOCKET_MANAGER_MAGIC;
2297         manager->mctx = NULL;
2298         memset(manager->fds, 0, sizeof(manager->fds));
2299         ISC_LIST_INIT(manager->socklist);
2300         if (isc_mutex_init(&manager->lock) != ISC_R_SUCCESS) {
2301                 isc_mem_put(mctx, manager, sizeof(*manager));
2302                 UNEXPECTED_ERROR(__FILE__, __LINE__,
2303                                  "isc_mutex_init() %s",
2304                                  isc_msgcat_get(isc_msgcat, ISC_MSGSET_GENERAL,
2305                                                 ISC_MSG_FAILED, "failed"));
2306                 return (ISC_R_UNEXPECTED);
2307         }
2308 #ifdef ISC_PLATFORM_USETHREADS
2309         if (isc_condition_init(&manager->shutdown_ok) != ISC_R_SUCCESS) {
2310                 DESTROYLOCK(&manager->lock);
2311                 isc_mem_put(mctx, manager, sizeof(*manager));
2312                 UNEXPECTED_ERROR(__FILE__, __LINE__,
2313                                  "isc_condition_init() %s",
2314                                  isc_msgcat_get(isc_msgcat, ISC_MSGSET_GENERAL,
2315                                                 ISC_MSG_FAILED, "failed"));
2316                 return (ISC_R_UNEXPECTED);
2317         }
2318
2319         /*
2320          * Create the special fds that will be used to wake up the
2321          * select/poll loop when something internal needs to be done.
2322          */
2323         if (pipe(manager->pipe_fds) != 0) {
2324                 DESTROYLOCK(&manager->lock);
2325                 isc_mem_put(mctx, manager, sizeof(*manager));
2326                 isc__strerror(errno, strbuf, sizeof(strbuf));
2327                 UNEXPECTED_ERROR(__FILE__, __LINE__,
2328                                  "pipe() %s: %s",
2329                                  isc_msgcat_get(isc_msgcat, ISC_MSGSET_GENERAL,
2330                                                 ISC_MSG_FAILED, "failed"),
2331                                  strbuf);
2332
2333                 return (ISC_R_UNEXPECTED);
2334         }
2335
2336         RUNTIME_CHECK(make_nonblock(manager->pipe_fds[0]) == ISC_R_SUCCESS);
2337 #if 0
2338         RUNTIME_CHECK(make_nonblock(manager->pipe_fds[1]) == ISC_R_SUCCESS);
2339 #endif
2340 #else /* ISC_PLATFORM_USETHREADS */
2341         manager->refs = 1;
2342 #endif /* ISC_PLATFORM_USETHREADS */
2343
2344         /*
2345          * Set up initial state for the select loop
2346          */
2347         FD_ZERO(&manager->read_fds);
2348         FD_ZERO(&manager->write_fds);
2349 #ifdef ISC_PLATFORM_USETHREADS
2350         FD_SET(manager->pipe_fds[0], &manager->read_fds);
2351         manager->maxfd = manager->pipe_fds[0];
2352 #else /* ISC_PLATFORM_USETHREADS */
2353         manager->maxfd = 0;
2354 #endif /* ISC_PLATFORM_USETHREADS */
2355         memset(manager->fdstate, 0, sizeof(manager->fdstate));
2356
2357 #ifdef ISC_PLATFORM_USETHREADS
2358         /*
2359          * Start up the select/poll thread.
2360          */
2361         if (isc_thread_create(watcher, manager, &manager->watcher) !=
2362             ISC_R_SUCCESS) {
2363                 close(manager->pipe_fds[0]);
2364                 close(manager->pipe_fds[1]);
2365                 DESTROYLOCK(&manager->lock);
2366                 isc_mem_put(mctx, manager, sizeof(*manager));
2367                 UNEXPECTED_ERROR(__FILE__, __LINE__,
2368                                  "isc_thread_create() %s",
2369                                  isc_msgcat_get(isc_msgcat, ISC_MSGSET_GENERAL,
2370                                                 ISC_MSG_FAILED, "failed"));
2371                 return (ISC_R_UNEXPECTED);
2372         }
2373 #endif /* ISC_PLATFORM_USETHREADS */
2374         isc_mem_attach(mctx, &manager->mctx);
2375
2376 #ifndef ISC_PLATFORM_USETHREADS
2377         socketmgr = manager;
2378 #endif /* ISC_PLATFORM_USETHREADS */
2379         *managerp = manager;
2380
2381         return (ISC_R_SUCCESS);
2382 }
2383
2384 void
2385 isc_socketmgr_destroy(isc_socketmgr_t **managerp) {
2386         isc_socketmgr_t *manager;
2387         int i;
2388         isc_mem_t *mctx;
2389
2390         /*
2391          * Destroy a socket manager.
2392          */
2393
2394         REQUIRE(managerp != NULL);
2395         manager = *managerp;
2396         REQUIRE(VALID_MANAGER(manager));
2397
2398 #ifndef ISC_PLATFORM_USETHREADS
2399         if (manager->refs > 1) {
2400                 manager->refs--;
2401                 *managerp = NULL;
2402                 return;
2403         }
2404 #endif /* ISC_PLATFORM_USETHREADS */
2405
2406         LOCK(&manager->lock);
2407
2408 #ifdef ISC_PLATFORM_USETHREADS
2409         /*
2410          * Wait for all sockets to be destroyed.
2411          */
2412         while (!ISC_LIST_EMPTY(manager->socklist)) {
2413                 manager_log(manager, CREATION,
2414                             isc_msgcat_get(isc_msgcat, ISC_MSGSET_SOCKET,
2415                                            ISC_MSG_SOCKETSREMAIN,
2416                                            "sockets exist"));
2417                 WAIT(&manager->shutdown_ok, &manager->lock);
2418         }
2419 #else /* ISC_PLATFORM_USETHREADS */
2420         /*
2421          * Hope all sockets have been destroyed.
2422          */
2423         if (!ISC_LIST_EMPTY(manager->socklist)) {
2424                 manager_log(manager, CREATION,
2425                             isc_msgcat_get(isc_msgcat, ISC_MSGSET_SOCKET,
2426                                            ISC_MSG_SOCKETSREMAIN,
2427                                            "sockets exist"));
2428                 INSIST(0);
2429         }
2430 #endif /* ISC_PLATFORM_USETHREADS */
2431
2432         UNLOCK(&manager->lock);
2433
2434         /*
2435          * Here, poke our select/poll thread.  Do this by closing the write
2436          * half of the pipe, which will send EOF to the read half.
2437          * This is currently a no-op in the non-threaded case.
2438          */
2439         select_poke(manager, 0, SELECT_POKE_SHUTDOWN);
2440
2441 #ifdef ISC_PLATFORM_USETHREADS
2442         /*
2443          * Wait for thread to exit.
2444          */
2445         if (isc_thread_join(manager->watcher, NULL) != ISC_R_SUCCESS)
2446                 UNEXPECTED_ERROR(__FILE__, __LINE__,
2447                                  "isc_thread_join() %s",
2448                                  isc_msgcat_get(isc_msgcat, ISC_MSGSET_GENERAL,
2449                                                 ISC_MSG_FAILED, "failed"));
2450 #endif /* ISC_PLATFORM_USETHREADS */
2451
2452         /*
2453          * Clean up.
2454          */
2455 #ifdef ISC_PLATFORM_USETHREADS
2456         close(manager->pipe_fds[0]);
2457         close(manager->pipe_fds[1]);
2458         (void)isc_condition_destroy(&manager->shutdown_ok);
2459 #endif /* ISC_PLATFORM_USETHREADS */
2460
2461         for (i = 0 ; i < (int)FD_SETSIZE ; i++)
2462                 if (manager->fdstate[i] == CLOSE_PENDING)
2463                         close(i);
2464
2465         DESTROYLOCK(&manager->lock);
2466         manager->magic = 0;
2467         mctx= manager->mctx;
2468         isc_mem_put(mctx, manager, sizeof(*manager));
2469
2470         isc_mem_detach(&mctx);
2471
2472         *managerp = NULL;
2473 }
2474
2475 static isc_result_t
2476 socket_recv(isc_socket_t *sock, isc_socketevent_t *dev, isc_task_t *task,
2477             unsigned int flags)
2478 {
2479         int io_state;
2480         isc_boolean_t have_lock = ISC_FALSE;
2481         isc_task_t *ntask = NULL;
2482         isc_result_t result = ISC_R_SUCCESS;
2483
2484         dev->ev_sender = task;
2485
2486         if (sock->type == isc_sockettype_udp) {
2487                 io_state = doio_recv(sock, dev);
2488         } else {
2489                 LOCK(&sock->lock);
2490                 have_lock = ISC_TRUE;
2491
2492                 if (ISC_LIST_EMPTY(sock->recv_list))
2493                         io_state = doio_recv(sock, dev);
2494                 else
2495                         io_state = DOIO_SOFT;
2496         }
2497
2498         switch (io_state) {
2499         case DOIO_SOFT:
2500                 /*
2501                  * We couldn't read all or part of the request right now, so
2502                  * queue it.
2503                  *
2504                  * Attach to socket and to task
2505                  */
2506                 isc_task_attach(task, &ntask);
2507                 dev->attributes |= ISC_SOCKEVENTATTR_ATTACHED;
2508
2509                 if (!have_lock) {
2510                         LOCK(&sock->lock);
2511                         have_lock = ISC_TRUE;
2512                 }
2513
2514                 /*
2515                  * Enqueue the request.  If the socket was previously not being
2516                  * watched, poke the watcher to start paying attention to it.
2517                  */
2518                 if (ISC_LIST_EMPTY(sock->recv_list))
2519                         select_poke(sock->manager, sock->fd, SELECT_POKE_READ);
2520                 ISC_LIST_ENQUEUE(sock->recv_list, dev, ev_link);
2521
2522                 socket_log(sock, NULL, EVENT, NULL, 0, 0,
2523                            "socket_recv: event %p -> task %p",
2524                            dev, ntask);
2525
2526                 if ((flags & ISC_SOCKFLAG_IMMEDIATE) != 0)
2527                         result = ISC_R_INPROGRESS;
2528                 break;
2529
2530         case DOIO_EOF:
2531                 dev->result = ISC_R_EOF;
2532                 /* fallthrough */
2533
2534         case DOIO_HARD:
2535         case DOIO_SUCCESS:
2536                 if ((flags & ISC_SOCKFLAG_IMMEDIATE) == 0)
2537                         send_recvdone_event(sock, &dev);
2538                 break;
2539         }
2540
2541         if (have_lock)
2542                 UNLOCK(&sock->lock);
2543
2544         return (result);
2545 }
2546
2547 isc_result_t
2548 isc_socket_recvv(isc_socket_t *sock, isc_bufferlist_t *buflist,
2549                  unsigned int minimum, isc_task_t *task,
2550                  isc_taskaction_t action, const void *arg)
2551 {
2552         isc_socketevent_t *dev;
2553         isc_socketmgr_t *manager;
2554         unsigned int iocount;
2555         isc_buffer_t *buffer;
2556
2557         REQUIRE(VALID_SOCKET(sock));
2558         REQUIRE(buflist != NULL);
2559         REQUIRE(!ISC_LIST_EMPTY(*buflist));
2560         REQUIRE(task != NULL);
2561         REQUIRE(action != NULL);
2562
2563         manager = sock->manager;
2564         REQUIRE(VALID_MANAGER(manager));
2565
2566         iocount = isc_bufferlist_availablecount(buflist);
2567         REQUIRE(iocount > 0);
2568
2569         INSIST(sock->bound);
2570
2571         dev = allocate_socketevent(sock, ISC_SOCKEVENT_RECVDONE, action, arg);
2572         if (dev == NULL) {
2573                 return (ISC_R_NOMEMORY);
2574         }
2575
2576         /*
2577          * UDP sockets are always partial read
2578          */
2579         if (sock->type == isc_sockettype_udp)
2580                 dev->minimum = 1;
2581         else {
2582                 if (minimum == 0)
2583                         dev->minimum = iocount;
2584                 else
2585                         dev->minimum = minimum;
2586         }
2587
2588         /*
2589          * Move each buffer from the passed in list to our internal one.
2590          */
2591         buffer = ISC_LIST_HEAD(*buflist);
2592         while (buffer != NULL) {
2593                 ISC_LIST_DEQUEUE(*buflist, buffer, link);
2594                 ISC_LIST_ENQUEUE(dev->bufferlist, buffer, link);
2595                 buffer = ISC_LIST_HEAD(*buflist);
2596         }
2597
2598         return (socket_recv(sock, dev, task, 0));
2599 }
2600
2601 isc_result_t
2602 isc_socket_recv(isc_socket_t *sock, isc_region_t *region, unsigned int minimum,
2603                 isc_task_t *task, isc_taskaction_t action, const void *arg)
2604 {
2605         isc_socketevent_t *dev;
2606         isc_socketmgr_t *manager;
2607
2608         REQUIRE(VALID_SOCKET(sock));
2609         REQUIRE(action != NULL);
2610
2611         manager = sock->manager;
2612         REQUIRE(VALID_MANAGER(manager));
2613
2614         INSIST(sock->bound);
2615
2616         dev = allocate_socketevent(sock, ISC_SOCKEVENT_RECVDONE, action, arg);
2617         if (dev == NULL)
2618                 return (ISC_R_NOMEMORY);
2619
2620         return (isc_socket_recv2(sock, region, minimum, task, dev, 0));
2621 }
2622
2623 isc_result_t
2624 isc_socket_recv2(isc_socket_t *sock, isc_region_t *region,
2625                  unsigned int minimum, isc_task_t *task,
2626                  isc_socketevent_t *event, unsigned int flags)
2627 {
2628         event->ev_sender = sock;
2629         event->result = ISC_R_UNEXPECTED;
2630         ISC_LIST_INIT(event->bufferlist);
2631         event->region = *region;
2632         event->n = 0;
2633         event->offset = 0;
2634         event->attributes = 0;
2635
2636         /*
2637          * UDP sockets are always partial read.
2638          */
2639         if (sock->type == isc_sockettype_udp)
2640                 event->minimum = 1;
2641         else {
2642                 if (minimum == 0)
2643                         event->minimum = region->length;
2644                 else
2645                         event->minimum = minimum;
2646         }
2647
2648         return (socket_recv(sock, event, task, flags));
2649 }
2650
2651 static isc_result_t
2652 socket_send(isc_socket_t *sock, isc_socketevent_t *dev, isc_task_t *task,
2653             isc_sockaddr_t *address, struct in6_pktinfo *pktinfo,
2654             unsigned int flags)
2655 {
2656         int io_state;
2657         isc_boolean_t have_lock = ISC_FALSE;
2658         isc_task_t *ntask = NULL;
2659         isc_result_t result = ISC_R_SUCCESS;
2660
2661         dev->ev_sender = task;
2662
2663         set_dev_address(address, sock, dev);
2664         if (pktinfo != NULL) {
2665                 socket_log(sock, NULL, TRACE, isc_msgcat, ISC_MSGSET_SOCKET,
2666                            ISC_MSG_PKTINFOPROVIDED,
2667                            "pktinfo structure provided, ifindex %u (set to 0)",
2668                            pktinfo->ipi6_ifindex);
2669
2670                 dev->attributes |= ISC_SOCKEVENTATTR_PKTINFO;
2671                 dev->pktinfo = *pktinfo;
2672                 /*
2673                  * Set the pktinfo index to 0 here, to let the kernel decide
2674                  * what interface it should send on.
2675                  */
2676                 dev->pktinfo.ipi6_ifindex = 0;
2677         }
2678
2679         if (sock->type == isc_sockettype_udp)
2680                 io_state = doio_send(sock, dev);
2681         else {
2682                 LOCK(&sock->lock);
2683                 have_lock = ISC_TRUE;
2684
2685                 if (ISC_LIST_EMPTY(sock->send_list))
2686                         io_state = doio_send(sock, dev);
2687                 else
2688                         io_state = DOIO_SOFT;
2689         }
2690
2691         switch (io_state) {
2692         case DOIO_SOFT:
2693                 /*
2694                  * We couldn't send all or part of the request right now, so
2695                  * queue it unless ISC_SOCKFLAG_NORETRY is set.
2696                  */
2697                 if ((flags & ISC_SOCKFLAG_NORETRY) == 0) {
2698                         isc_task_attach(task, &ntask);
2699                         dev->attributes |= ISC_SOCKEVENTATTR_ATTACHED;
2700
2701                         if (!have_lock) {
2702                                 LOCK(&sock->lock);
2703                                 have_lock = ISC_TRUE;
2704                         }
2705
2706                         /*
2707                          * Enqueue the request.  If the socket was previously
2708                          * not being watched, poke the watcher to start
2709                          * paying attention to it.
2710                          */
2711                         if (ISC_LIST_EMPTY(sock->send_list))
2712                                 select_poke(sock->manager, sock->fd,
2713                                             SELECT_POKE_WRITE);
2714                         ISC_LIST_ENQUEUE(sock->send_list, dev, ev_link);
2715
2716                         socket_log(sock, NULL, EVENT, NULL, 0, 0,
2717                                    "socket_send: event %p -> task %p",
2718                                    dev, ntask);
2719
2720                         if ((flags & ISC_SOCKFLAG_IMMEDIATE) != 0)
2721                                 result = ISC_R_INPROGRESS;
2722                         break;
2723                 }
2724
2725         case DOIO_HARD:
2726         case DOIO_SUCCESS:
2727                 if ((flags & ISC_SOCKFLAG_IMMEDIATE) == 0)
2728                         send_senddone_event(sock, &dev);
2729                 break;
2730         }
2731
2732         if (have_lock)
2733                 UNLOCK(&sock->lock);
2734
2735         return (result);
2736 }
2737
2738 isc_result_t
2739 isc_socket_send(isc_socket_t *sock, isc_region_t *region,
2740                 isc_task_t *task, isc_taskaction_t action, const void *arg)
2741 {
2742         /*
2743          * REQUIRE() checking is performed in isc_socket_sendto().
2744          */
2745         return (isc_socket_sendto(sock, region, task, action, arg, NULL,
2746                                   NULL));
2747 }
2748
2749 isc_result_t
2750 isc_socket_sendto(isc_socket_t *sock, isc_region_t *region,
2751                   isc_task_t *task, isc_taskaction_t action, const void *arg,
2752                   isc_sockaddr_t *address, struct in6_pktinfo *pktinfo)
2753 {
2754         isc_socketevent_t *dev;
2755         isc_socketmgr_t *manager;
2756
2757         REQUIRE(VALID_SOCKET(sock));
2758         REQUIRE(region != NULL);
2759         REQUIRE(task != NULL);
2760         REQUIRE(action != NULL);
2761
2762         manager = sock->manager;
2763         REQUIRE(VALID_MANAGER(manager));
2764
2765         INSIST(sock->bound);
2766
2767         dev = allocate_socketevent(sock, ISC_SOCKEVENT_SENDDONE, action, arg);
2768         if (dev == NULL) {
2769                 return (ISC_R_NOMEMORY);
2770         }
2771
2772         dev->region = *region;
2773
2774         return (socket_send(sock, dev, task, address, pktinfo, 0));
2775 }
2776
2777 isc_result_t
2778 isc_socket_sendv(isc_socket_t *sock, isc_bufferlist_t *buflist,
2779                  isc_task_t *task, isc_taskaction_t action, const void *arg)
2780 {
2781         return (isc_socket_sendtov(sock, buflist, task, action, arg, NULL,
2782                                    NULL));
2783 }
2784
2785 isc_result_t
2786 isc_socket_sendtov(isc_socket_t *sock, isc_bufferlist_t *buflist,
2787                    isc_task_t *task, isc_taskaction_t action, const void *arg,
2788                    isc_sockaddr_t *address, struct in6_pktinfo *pktinfo)
2789 {
2790         isc_socketevent_t *dev;
2791         isc_socketmgr_t *manager;
2792         unsigned int iocount;
2793         isc_buffer_t *buffer;
2794
2795         REQUIRE(VALID_SOCKET(sock));
2796         REQUIRE(buflist != NULL);
2797         REQUIRE(!ISC_LIST_EMPTY(*buflist));
2798         REQUIRE(task != NULL);
2799         REQUIRE(action != NULL);
2800
2801         manager = sock->manager;
2802         REQUIRE(VALID_MANAGER(manager));
2803
2804         iocount = isc_bufferlist_usedcount(buflist);
2805         REQUIRE(iocount > 0);
2806
2807         dev = allocate_socketevent(sock, ISC_SOCKEVENT_SENDDONE, action, arg);
2808         if (dev == NULL) {
2809                 return (ISC_R_NOMEMORY);
2810         }
2811
2812         /*
2813          * Move each buffer from the passed in list to our internal one.
2814          */
2815         buffer = ISC_LIST_HEAD(*buflist);
2816         while (buffer != NULL) {
2817                 ISC_LIST_DEQUEUE(*buflist, buffer, link);
2818                 ISC_LIST_ENQUEUE(dev->bufferlist, buffer, link);
2819                 buffer = ISC_LIST_HEAD(*buflist);
2820         }
2821
2822         return (socket_send(sock, dev, task, address, pktinfo, 0));
2823 }
2824
2825 isc_result_t
2826 isc_socket_sendto2(isc_socket_t *sock, isc_region_t *region,
2827                    isc_task_t *task,
2828                    isc_sockaddr_t *address, struct in6_pktinfo *pktinfo,
2829                    isc_socketevent_t *event, unsigned int flags)
2830 {
2831         REQUIRE((flags & ~(ISC_SOCKFLAG_IMMEDIATE|ISC_SOCKFLAG_NORETRY)) == 0);
2832         if ((flags & ISC_SOCKFLAG_NORETRY) != 0)
2833                 REQUIRE(sock->type == isc_sockettype_udp);
2834         event->ev_sender = sock;
2835         event->result = ISC_R_UNEXPECTED;
2836         ISC_LIST_INIT(event->bufferlist);
2837         event->region = *region;
2838         event->n = 0;
2839         event->offset = 0;
2840         event->attributes = 0;
2841
2842         return (socket_send(sock, event, task, address, pktinfo, flags));
2843 }
2844
2845 isc_result_t
2846 isc_socket_bind(isc_socket_t *sock, isc_sockaddr_t *sockaddr) {
2847         char strbuf[ISC_STRERRORSIZE];
2848         int on = 1;
2849
2850         LOCK(&sock->lock);
2851
2852         INSIST(!sock->bound);
2853
2854         if (sock->pf != sockaddr->type.sa.sa_family) {
2855                 UNLOCK(&sock->lock);
2856                 return (ISC_R_FAMILYMISMATCH);
2857         }
2858         /*
2859          * Only set SO_REUSEADDR when we want a specific port.
2860          */
2861         if (isc_sockaddr_getport(sockaddr) != (in_port_t)0 &&
2862             setsockopt(sock->fd, SOL_SOCKET, SO_REUSEADDR, (void *)&on,
2863                        sizeof(on)) < 0) {
2864                 UNEXPECTED_ERROR(__FILE__, __LINE__,
2865                                  "setsockopt(%d) %s", sock->fd,
2866                                  isc_msgcat_get(isc_msgcat, ISC_MSGSET_GENERAL,
2867                                                 ISC_MSG_FAILED, "failed"));
2868                 /* Press on... */
2869         }
2870         if (bind(sock->fd, &sockaddr->type.sa, sockaddr->length) < 0) {
2871                 UNLOCK(&sock->lock);
2872                 switch (errno) {
2873                 case EACCES:
2874                         return (ISC_R_NOPERM);
2875                 case EADDRNOTAVAIL:
2876                         return (ISC_R_ADDRNOTAVAIL);
2877                 case EADDRINUSE:
2878                         return (ISC_R_ADDRINUSE);
2879                 case EINVAL:
2880                         return (ISC_R_BOUND);
2881                 default:
2882                         isc__strerror(errno, strbuf, sizeof(strbuf));
2883                         UNEXPECTED_ERROR(__FILE__, __LINE__, "bind: %s",
2884                                          strbuf);
2885                         return (ISC_R_UNEXPECTED);
2886                 }
2887         }
2888
2889         socket_log(sock, sockaddr, TRACE,
2890                    isc_msgcat, ISC_MSGSET_SOCKET, ISC_MSG_BOUND, "bound");
2891         sock->bound = 1;
2892
2893         UNLOCK(&sock->lock);
2894         return (ISC_R_SUCCESS);
2895 }
2896
2897 /*
2898  * Set up to listen on a given socket.  We do this by creating an internal
2899  * event that will be dispatched when the socket has read activity.  The
2900  * watcher will send the internal event to the task when there is a new
2901  * connection.
2902  *
2903  * Unlike in read, we don't preallocate a done event here.  Every time there
2904  * is a new connection we'll have to allocate a new one anyway, so we might
2905  * as well keep things simple rather than having to track them.
2906  */
2907 isc_result_t
2908 isc_socket_listen(isc_socket_t *sock, unsigned int backlog) {
2909         char strbuf[ISC_STRERRORSIZE];
2910
2911         REQUIRE(VALID_SOCKET(sock));
2912
2913         LOCK(&sock->lock);
2914
2915         REQUIRE(!sock->listener);
2916         REQUIRE(sock->bound);
2917         REQUIRE(sock->type == isc_sockettype_tcp);
2918
2919         if (backlog == 0)
2920                 backlog = SOMAXCONN;
2921
2922         if (listen(sock->fd, (int)backlog) < 0) {
2923                 UNLOCK(&sock->lock);
2924                 isc__strerror(errno, strbuf, sizeof(strbuf));
2925
2926                 UNEXPECTED_ERROR(__FILE__, __LINE__, "listen: %s", strbuf);
2927
2928                 return (ISC_R_UNEXPECTED);
2929         }
2930
2931         sock->listener = 1;
2932
2933         UNLOCK(&sock->lock);
2934         return (ISC_R_SUCCESS);
2935 }
2936
2937 /*
2938  * This should try to do agressive accept() XXXMLG
2939  */
2940 isc_result_t
2941 isc_socket_accept(isc_socket_t *sock,
2942                   isc_task_t *task, isc_taskaction_t action, const void *arg)
2943 {
2944         isc_socket_newconnev_t *dev;
2945         isc_socketmgr_t *manager;
2946         isc_task_t *ntask = NULL;
2947         isc_socket_t *nsock;
2948         isc_result_t ret;
2949         isc_boolean_t do_poke = ISC_FALSE;
2950
2951         REQUIRE(VALID_SOCKET(sock));
2952         manager = sock->manager;
2953         REQUIRE(VALID_MANAGER(manager));
2954
2955         LOCK(&sock->lock);
2956
2957         REQUIRE(sock->listener);
2958
2959         /*
2960          * Sender field is overloaded here with the task we will be sending
2961          * this event to.  Just before the actual event is delivered the
2962          * actual ev_sender will be touched up to be the socket.
2963          */
2964         dev = (isc_socket_newconnev_t *)
2965                 isc_event_allocate(manager->mctx, task, ISC_SOCKEVENT_NEWCONN,
2966                                    action, arg, sizeof(*dev));
2967         if (dev == NULL) {
2968                 UNLOCK(&sock->lock);
2969                 return (ISC_R_NOMEMORY);
2970         }
2971         ISC_LINK_INIT(dev, ev_link);
2972
2973         ret = allocate_socket(manager, sock->type, &nsock);
2974         if (ret != ISC_R_SUCCESS) {
2975                 isc_event_free(ISC_EVENT_PTR(&dev));
2976                 UNLOCK(&sock->lock);
2977                 return (ret);
2978         }
2979
2980         /*
2981          * Attach to socket and to task.
2982          */
2983         isc_task_attach(task, &ntask);
2984         nsock->references++;
2985
2986         dev->ev_sender = ntask;
2987         dev->newsocket = nsock;
2988
2989         /*
2990          * Poke watcher here.  We still have the socket locked, so there
2991          * is no race condition.  We will keep the lock for such a short
2992          * bit of time waking it up now or later won't matter all that much.
2993          */
2994         if (ISC_LIST_EMPTY(sock->accept_list))
2995                 do_poke = ISC_TRUE;
2996
2997         ISC_LIST_ENQUEUE(sock->accept_list, dev, ev_link);
2998
2999         if (do_poke)
3000                 select_poke(manager, sock->fd, SELECT_POKE_ACCEPT);
3001
3002         UNLOCK(&sock->lock);
3003         return (ISC_R_SUCCESS);
3004 }
3005
3006 isc_result_t
3007 isc_socket_connect(isc_socket_t *sock, isc_sockaddr_t *addr,
3008                    isc_task_t *task, isc_taskaction_t action, const void *arg)
3009 {
3010         isc_socket_connev_t *dev;
3011         isc_task_t *ntask = NULL;
3012         isc_socketmgr_t *manager;
3013         int cc;
3014         char strbuf[ISC_STRERRORSIZE];
3015
3016         REQUIRE(VALID_SOCKET(sock));
3017         REQUIRE(addr != NULL);
3018         REQUIRE(task != NULL);
3019         REQUIRE(action != NULL);
3020
3021         manager = sock->manager;
3022         REQUIRE(VALID_MANAGER(manager));
3023         REQUIRE(addr != NULL);
3024
3025         if (isc_sockaddr_ismulticast(addr))
3026                 return (ISC_R_MULTICAST);
3027
3028         LOCK(&sock->lock);
3029
3030         REQUIRE(!sock->connecting);
3031
3032         dev = (isc_socket_connev_t *)isc_event_allocate(manager->mctx, sock,
3033                                                         ISC_SOCKEVENT_CONNECT,
3034                                                         action, arg,
3035                                                         sizeof(*dev));
3036         if (dev == NULL) {
3037                 UNLOCK(&sock->lock);
3038                 return (ISC_R_NOMEMORY);
3039         }
3040         ISC_LINK_INIT(dev, ev_link);
3041
3042         /*
3043          * Try to do the connect right away, as there can be only one
3044          * outstanding, and it might happen to complete.
3045          */
3046         sock->address = *addr;
3047         cc = connect(sock->fd, &addr->type.sa, addr->length);
3048         if (cc < 0) {
3049                 if (SOFT_ERROR(errno) || errno == EINPROGRESS)
3050                         goto queue;
3051
3052                 switch (errno) {
3053 #define ERROR_MATCH(a, b) case a: dev->result = b; goto err_exit;
3054                         ERROR_MATCH(EACCES, ISC_R_NOPERM);
3055                         ERROR_MATCH(EADDRNOTAVAIL, ISC_R_ADDRNOTAVAIL);
3056                         ERROR_MATCH(EAFNOSUPPORT, ISC_R_ADDRNOTAVAIL);
3057                         ERROR_MATCH(ECONNREFUSED, ISC_R_CONNREFUSED);
3058                         ERROR_MATCH(EHOSTUNREACH, ISC_R_HOSTUNREACH);
3059 #ifdef EHOSTDOWN
3060                         ERROR_MATCH(EHOSTDOWN, ISC_R_HOSTUNREACH);
3061 #endif
3062                         ERROR_MATCH(ENETUNREACH, ISC_R_NETUNREACH);
3063                         ERROR_MATCH(ENOBUFS, ISC_R_NORESOURCES);
3064                         ERROR_MATCH(EPERM, ISC_R_HOSTUNREACH);
3065                         ERROR_MATCH(EPIPE, ISC_R_NOTCONNECTED);
3066 #undef ERROR_MATCH
3067                 }
3068
3069                 sock->connected = 0;
3070
3071                 isc__strerror(errno, strbuf, sizeof(strbuf));
3072                 UNEXPECTED_ERROR(__FILE__, __LINE__, "%d/%s", errno, strbuf);
3073
3074                 UNLOCK(&sock->lock);
3075                 isc_event_free(ISC_EVENT_PTR(&dev));
3076                 return (ISC_R_UNEXPECTED);
3077
3078         err_exit:
3079                 sock->connected = 0;
3080                 isc_task_send(task, ISC_EVENT_PTR(&dev));
3081
3082                 UNLOCK(&sock->lock);
3083                 return (ISC_R_SUCCESS);
3084         }
3085
3086         /*
3087          * If connect completed, fire off the done event.
3088          */
3089         if (cc == 0) {
3090                 sock->connected = 1;
3091                 sock->bound = 1;
3092                 dev->result = ISC_R_SUCCESS;
3093                 isc_task_send(task, ISC_EVENT_PTR(&dev));
3094
3095                 UNLOCK(&sock->lock);
3096                 return (ISC_R_SUCCESS);
3097         }
3098
3099  queue:
3100
3101         /*
3102          * Attach to task.
3103          */
3104         isc_task_attach(task, &ntask);
3105
3106         sock->connecting = 1;
3107
3108         dev->ev_sender = ntask;
3109
3110         /*
3111          * Poke watcher here.  We still have the socket locked, so there
3112          * is no race condition.  We will keep the lock for such a short
3113          * bit of time waking it up now or later won't matter all that much.
3114          */
3115         if (sock->connect_ev == NULL)
3116                 select_poke(manager, sock->fd, SELECT_POKE_CONNECT);
3117
3118         sock->connect_ev = dev;
3119
3120         UNLOCK(&sock->lock);
3121         return (ISC_R_SUCCESS);
3122 }
3123
3124 /*
3125  * Called when a socket with a pending connect() finishes.
3126  */
3127 static void
3128 internal_connect(isc_task_t *me, isc_event_t *ev) {
3129         isc_socket_t *sock;
3130         isc_socket_connev_t *dev;
3131         isc_task_t *task;
3132         int cc;
3133         ISC_SOCKADDR_LEN_T optlen;
3134         char strbuf[ISC_STRERRORSIZE];
3135
3136         UNUSED(me);
3137         INSIST(ev->ev_type == ISC_SOCKEVENT_INTW);
3138
3139         sock = ev->ev_sender;
3140         INSIST(VALID_SOCKET(sock));
3141
3142         LOCK(&sock->lock);
3143
3144         /*
3145          * When the internal event was sent the reference count was bumped
3146          * to keep the socket around for us.  Decrement the count here.
3147          */
3148         INSIST(sock->references > 0);
3149         sock->references--;
3150         if (sock->references == 0) {
3151                 UNLOCK(&sock->lock);
3152                 destroy(&sock);
3153                 return;
3154         }
3155
3156         /*
3157          * Has this event been canceled?
3158          */
3159         dev = sock->connect_ev;
3160         if (dev == NULL) {
3161                 INSIST(!sock->connecting);
3162                 UNLOCK(&sock->lock);
3163                 return;
3164         }
3165
3166         INSIST(sock->connecting);
3167         sock->connecting = 0;
3168
3169         /*
3170          * Get any possible error status here.
3171          */
3172         optlen = sizeof(cc);
3173         if (getsockopt(sock->fd, SOL_SOCKET, SO_ERROR,
3174                        (void *)&cc, (void *)&optlen) < 0)
3175                 cc = errno;
3176         else
3177                 errno = cc;
3178
3179         if (errno != 0) {
3180                 /*
3181                  * If the error is EAGAIN, just re-select on this
3182                  * fd and pretend nothing strange happened.
3183                  */
3184                 if (SOFT_ERROR(errno) || errno == EINPROGRESS) {
3185                         sock->connecting = 1;
3186                         select_poke(sock->manager, sock->fd,
3187                                     SELECT_POKE_CONNECT);
3188                         UNLOCK(&sock->lock);
3189
3190                         return;
3191                 }
3192
3193                 /*
3194                  * Translate other errors into ISC_R_* flavors.
3195                  */
3196                 switch (errno) {
3197 #define ERROR_MATCH(a, b) case a: dev->result = b; break;
3198                         ERROR_MATCH(EACCES, ISC_R_NOPERM);
3199                         ERROR_MATCH(EADDRNOTAVAIL, ISC_R_ADDRNOTAVAIL);
3200                         ERROR_MATCH(EAFNOSUPPORT, ISC_R_ADDRNOTAVAIL);
3201                         ERROR_MATCH(ECONNREFUSED, ISC_R_CONNREFUSED);
3202                         ERROR_MATCH(EHOSTUNREACH, ISC_R_HOSTUNREACH);
3203 #ifdef EHOSTDOWN
3204                         ERROR_MATCH(EHOSTDOWN, ISC_R_HOSTUNREACH);
3205 #endif
3206                         ERROR_MATCH(ENETUNREACH, ISC_R_NETUNREACH);
3207                         ERROR_MATCH(ENOBUFS, ISC_R_NORESOURCES);
3208                         ERROR_MATCH(EPERM, ISC_R_HOSTUNREACH);
3209                         ERROR_MATCH(EPIPE, ISC_R_NOTCONNECTED);
3210                         ERROR_MATCH(ETIMEDOUT, ISC_R_TIMEDOUT);
3211 #undef ERROR_MATCH
3212                 default:
3213                         dev->result = ISC_R_UNEXPECTED;
3214                         isc__strerror(errno, strbuf, sizeof(strbuf));
3215                         UNEXPECTED_ERROR(__FILE__, __LINE__,
3216                                          "internal_connect: connect() %s",
3217                                          strbuf);
3218                 }
3219         } else {
3220                 dev->result = ISC_R_SUCCESS;
3221                 sock->connected = 1;
3222                 sock->bound = 1;
3223         }
3224
3225         sock->connect_ev = NULL;
3226
3227         UNLOCK(&sock->lock);
3228
3229         task = dev->ev_sender;
3230         dev->ev_sender = sock;
3231         isc_task_sendanddetach(&task, ISC_EVENT_PTR(&dev));
3232 }
3233
3234 isc_result_t
3235 isc_socket_getpeername(isc_socket_t *sock, isc_sockaddr_t *addressp) {
3236         isc_result_t ret;
3237
3238         REQUIRE(VALID_SOCKET(sock));
3239         REQUIRE(addressp != NULL);
3240
3241         LOCK(&sock->lock);
3242
3243         if (sock->connected) {
3244                 *addressp = sock->address;
3245                 ret = ISC_R_SUCCESS;
3246         } else {
3247                 ret = ISC_R_NOTCONNECTED;
3248         }
3249
3250         UNLOCK(&sock->lock);
3251
3252         return (ret);
3253 }
3254
3255 isc_result_t
3256 isc_socket_getsockname(isc_socket_t *sock, isc_sockaddr_t *addressp) {
3257         ISC_SOCKADDR_LEN_T len;
3258         isc_result_t ret;
3259         char strbuf[ISC_STRERRORSIZE];
3260
3261         REQUIRE(VALID_SOCKET(sock));
3262         REQUIRE(addressp != NULL);
3263
3264         LOCK(&sock->lock);
3265
3266         if (!sock->bound) {
3267                 ret = ISC_R_NOTBOUND;
3268                 goto out;
3269         }
3270
3271         ret = ISC_R_SUCCESS;
3272
3273         len = sizeof(addressp->type);
3274         if (getsockname(sock->fd, &addressp->type.sa, (void *)&len) < 0) {
3275                 isc__strerror(errno, strbuf, sizeof(strbuf));
3276                 UNEXPECTED_ERROR(__FILE__, __LINE__, "getsockname: %s",
3277                                  strbuf);
3278                 ret = ISC_R_UNEXPECTED;
3279                 goto out;
3280         }
3281         addressp->length = (unsigned int)len;
3282
3283  out:
3284         UNLOCK(&sock->lock);
3285
3286         return (ret);
3287 }
3288
3289 /*
3290  * Run through the list of events on this socket, and cancel the ones
3291  * queued for task "task" of type "how".  "how" is a bitmask.
3292  */
3293 void
3294 isc_socket_cancel(isc_socket_t *sock, isc_task_t *task, unsigned int how) {
3295
3296         REQUIRE(VALID_SOCKET(sock));
3297
3298         /*
3299          * Quick exit if there is nothing to do.  Don't even bother locking
3300          * in this case.
3301          */
3302         if (how == 0)
3303                 return;
3304
3305         LOCK(&sock->lock);
3306
3307         /*
3308          * All of these do the same thing, more or less.
3309          * Each will:
3310          *      o If the internal event is marked as "posted" try to
3311          *        remove it from the task's queue.  If this fails, mark it
3312          *        as canceled instead, and let the task clean it up later.
3313          *      o For each I/O request for that task of that type, post
3314          *        its done event with status of "ISC_R_CANCELED".
3315          *      o Reset any state needed.
3316          */
3317         if (((how & ISC_SOCKCANCEL_RECV) == ISC_SOCKCANCEL_RECV)
3318             && !ISC_LIST_EMPTY(sock->recv_list)) {
3319                 isc_socketevent_t      *dev;
3320                 isc_socketevent_t      *next;
3321                 isc_task_t             *current_task;
3322
3323                 dev = ISC_LIST_HEAD(sock->recv_list);
3324
3325                 while (dev != NULL) {
3326                         current_task = dev->ev_sender;
3327                         next = ISC_LIST_NEXT(dev, ev_link);
3328
3329                         if ((task == NULL) || (task == current_task)) {
3330                                 dev->result = ISC_R_CANCELED;
3331                                 send_recvdone_event(sock, &dev);
3332                         }
3333                         dev = next;
3334                 }
3335         }
3336
3337         if (((how & ISC_SOCKCANCEL_SEND) == ISC_SOCKCANCEL_SEND)
3338             && !ISC_LIST_EMPTY(sock->send_list)) {
3339                 isc_socketevent_t      *dev;
3340                 isc_socketevent_t      *next;
3341                 isc_task_t             *current_task;
3342
3343                 dev = ISC_LIST_HEAD(sock->send_list);
3344
3345                 while (dev != NULL) {
3346                         current_task = dev->ev_sender;
3347                         next = ISC_LIST_NEXT(dev, ev_link);
3348
3349                         if ((task == NULL) || (task == current_task)) {
3350                                 dev->result = ISC_R_CANCELED;
3351                                 send_senddone_event(sock, &dev);
3352                         }
3353                         dev = next;
3354                 }
3355         }
3356
3357         if (((how & ISC_SOCKCANCEL_ACCEPT) == ISC_SOCKCANCEL_ACCEPT)
3358             && !ISC_LIST_EMPTY(sock->accept_list)) {
3359                 isc_socket_newconnev_t *dev;
3360                 isc_socket_newconnev_t *next;
3361                 isc_task_t             *current_task;
3362
3363                 dev = ISC_LIST_HEAD(sock->accept_list);
3364                 while (dev != NULL) {
3365                         current_task = dev->ev_sender;
3366                         next = ISC_LIST_NEXT(dev, ev_link);
3367
3368                         if ((task == NULL) || (task == current_task)) {
3369
3370                                 ISC_LIST_UNLINK(sock->accept_list, dev,
3371                                                 ev_link);
3372
3373                                 dev->newsocket->references--;
3374                                 free_socket(&dev->newsocket);
3375
3376                                 dev->result = ISC_R_CANCELED;
3377                                 dev->ev_sender = sock;
3378                                 isc_task_sendanddetach(&current_task,
3379                                                 ISC_EVENT_PTR(&dev));
3380                         }
3381
3382                         dev = next;
3383                 }
3384         }
3385
3386         /*
3387          * Connecting is not a list.
3388          */
3389         if (((how & ISC_SOCKCANCEL_CONNECT) == ISC_SOCKCANCEL_CONNECT)
3390             && sock->connect_ev != NULL) {
3391                 isc_socket_connev_t    *dev;
3392                 isc_task_t             *current_task;
3393
3394                 INSIST(sock->connecting);
3395                 sock->connecting = 0;
3396
3397                 dev = sock->connect_ev;
3398                 current_task = dev->ev_sender;
3399
3400                 if ((task == NULL) || (task == current_task)) {
3401                         sock->connect_ev = NULL;
3402
3403                         dev->result = ISC_R_CANCELED;
3404                         dev->ev_sender = sock;
3405                         isc_task_sendanddetach(&current_task,
3406                                                ISC_EVENT_PTR(&dev));
3407                 }
3408         }
3409
3410         UNLOCK(&sock->lock);
3411 }
3412
3413 isc_sockettype_t
3414 isc_socket_gettype(isc_socket_t *sock) {
3415         REQUIRE(VALID_SOCKET(sock));
3416
3417         return (sock->type);
3418 }
3419
3420 isc_boolean_t
3421 isc_socket_isbound(isc_socket_t *sock) {
3422         isc_boolean_t val;
3423
3424         LOCK(&sock->lock);
3425         val = ((sock->bound) ? ISC_TRUE : ISC_FALSE);
3426         UNLOCK(&sock->lock);
3427
3428         return (val);
3429 }
3430
3431 #ifndef ISC_PLATFORM_USETHREADS
3432 void
3433 isc__socketmgr_getfdsets(fd_set *readset, fd_set *writeset, int *maxfd) {
3434         if (socketmgr == NULL)
3435                 *maxfd = 0;
3436         else {
3437                 *readset = socketmgr->read_fds;
3438                 *writeset = socketmgr->write_fds;
3439                 *maxfd = socketmgr->maxfd + 1;
3440         }
3441 }
3442
3443 isc_result_t
3444 isc__socketmgr_dispatch(fd_set *readset, fd_set *writeset, int maxfd) {
3445         isc_socketmgr_t *manager = socketmgr;
3446
3447         if (manager == NULL)
3448                 return (ISC_R_NOTFOUND);
3449
3450         process_fds(manager, maxfd, readset, writeset);
3451         return (ISC_R_SUCCESS);
3452 }
3453 #endif /* ISC_PLATFORM_USETHREADS */