1 /* 2 * util/netevent.c - event notification 3 * 4 * Copyright (c) 2007, NLnet Labs. All rights reserved. 5 * 6 * This software is open source. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 12 * Redistributions of source code must retain the above copyright notice, 13 * this list of conditions and the following disclaimer. 14 * 15 * Redistributions in binary form must reproduce the above copyright notice, 16 * this list of conditions and the following disclaimer in the documentation 17 * and/or other materials provided with the distribution. 18 * 19 * Neither the name of the NLNET LABS nor the names of its contributors may 20 * be used to endorse or promote products derived from this software without 21 * specific prior written permission. 22 * 23 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 24 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 25 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR 26 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 27 * HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 28 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED 29 * TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR 30 * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF 31 * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING 32 * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS 33 * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 34 */ 35 36 /** 37 * \file 38 * 39 * This file contains event notification functions. 40 */ 41 #include "config.h" 42 #include "util/netevent.h" 43 #include "util/ub_event.h" 44 #include "util/log.h" 45 #include "util/net_help.h" 46 #include "util/tcp_conn_limit.h" 47 #include "util/fptr_wlist.h" 48 #include "sldns/pkthdr.h" 49 #include "sldns/sbuffer.h" 50 #include "sldns/str2wire.h" 51 #include "dnstap/dnstap.h" 52 #include "dnscrypt/dnscrypt.h" 53 #include "services/listen_dnsport.h" 54 #ifdef HAVE_OPENSSL_SSL_H 55 #include <openssl/ssl.h> 56 #endif 57 #ifdef HAVE_OPENSSL_ERR_H 58 #include <openssl/err.h> 59 #endif 60 61 /* -------- Start of local definitions -------- */ 62 /** if CMSG_ALIGN is not defined on this platform, a workaround */ 63 #ifndef CMSG_ALIGN 64 # ifdef __CMSG_ALIGN 65 # define CMSG_ALIGN(n) __CMSG_ALIGN(n) 66 # elif defined(CMSG_DATA_ALIGN) 67 # define CMSG_ALIGN _CMSG_DATA_ALIGN 68 # else 69 # define CMSG_ALIGN(len) (((len)+sizeof(long)-1) & ~(sizeof(long)-1)) 70 # endif 71 #endif 72 73 /** if CMSG_LEN is not defined on this platform, a workaround */ 74 #ifndef CMSG_LEN 75 # define CMSG_LEN(len) (CMSG_ALIGN(sizeof(struct cmsghdr))+(len)) 76 #endif 77 78 /** if CMSG_SPACE is not defined on this platform, a workaround */ 79 #ifndef CMSG_SPACE 80 # ifdef _CMSG_HDR_ALIGN 81 # define CMSG_SPACE(l) (CMSG_ALIGN(l)+_CMSG_HDR_ALIGN(sizeof(struct cmsghdr))) 82 # else 83 # define CMSG_SPACE(l) (CMSG_ALIGN(l)+CMSG_ALIGN(sizeof(struct cmsghdr))) 84 # endif 85 #endif 86 87 /** The TCP writing query timeout in milliseconds */ 88 #define TCP_QUERY_TIMEOUT 120000 89 /** The minimum actual TCP timeout to use, regardless of what we advertise, 90 * in msec */ 91 #define TCP_QUERY_TIMEOUT_MINIMUM 200 92 93 #ifndef NONBLOCKING_IS_BROKEN 94 /** number of UDP reads to perform per read indication from select */ 95 #define NUM_UDP_PER_SELECT 100 96 #else 97 #define NUM_UDP_PER_SELECT 1 98 #endif 99 100 /** 101 * The internal event structure for keeping ub_event info for the event. 102 * Possibly other structures (list, tree) this is part of. 103 */ 104 struct internal_event { 105 /** the comm base */ 106 struct comm_base* base; 107 /** ub_event event type */ 108 struct ub_event* ev; 109 }; 110 111 /** 112 * Internal base structure, so that every thread has its own events. 113 */ 114 struct internal_base { 115 /** ub_event event_base type. */ 116 struct ub_event_base* base; 117 /** seconds time pointer points here */ 118 time_t secs; 119 /** timeval with current time */ 120 struct timeval now; 121 /** the event used for slow_accept timeouts */ 122 struct ub_event* slow_accept; 123 /** true if slow_accept is enabled */ 124 int slow_accept_enabled; 125 }; 126 127 /** 128 * Internal timer structure, to store timer event in. 129 */ 130 struct internal_timer { 131 /** the super struct from which derived */ 132 struct comm_timer super; 133 /** the comm base */ 134 struct comm_base* base; 135 /** ub_event event type */ 136 struct ub_event* ev; 137 /** is timer enabled */ 138 uint8_t enabled; 139 }; 140 141 /** 142 * Internal signal structure, to store signal event in. 143 */ 144 struct internal_signal { 145 /** ub_event event type */ 146 struct ub_event* ev; 147 /** next in signal list */ 148 struct internal_signal* next; 149 }; 150 151 /** create a tcp handler with a parent */ 152 static struct comm_point* comm_point_create_tcp_handler( 153 struct comm_base *base, struct comm_point* parent, size_t bufsize, 154 struct sldns_buffer* spoolbuf, comm_point_callback_type* callback, 155 void* callback_arg); 156 157 /* -------- End of local definitions -------- */ 158 159 struct comm_base* 160 comm_base_create(int sigs) 161 { 162 struct comm_base* b = (struct comm_base*)calloc(1, 163 sizeof(struct comm_base)); 164 const char *evnm="event", *evsys="", *evmethod=""; 165 166 if(!b) 167 return NULL; 168 b->eb = (struct internal_base*)calloc(1, sizeof(struct internal_base)); 169 if(!b->eb) { 170 free(b); 171 return NULL; 172 } 173 b->eb->base = ub_default_event_base(sigs, &b->eb->secs, &b->eb->now); 174 if(!b->eb->base) { 175 free(b->eb); 176 free(b); 177 return NULL; 178 } 179 ub_comm_base_now(b); 180 ub_get_event_sys(b->eb->base, &evnm, &evsys, &evmethod); 181 verbose(VERB_ALGO, "%s %s uses %s method.", evnm, evsys, evmethod); 182 return b; 183 } 184 185 struct comm_base* 186 comm_base_create_event(struct ub_event_base* base) 187 { 188 struct comm_base* b = (struct comm_base*)calloc(1, 189 sizeof(struct comm_base)); 190 if(!b) 191 return NULL; 192 b->eb = (struct internal_base*)calloc(1, sizeof(struct internal_base)); 193 if(!b->eb) { 194 free(b); 195 return NULL; 196 } 197 b->eb->base = base; 198 ub_comm_base_now(b); 199 return b; 200 } 201 202 void 203 comm_base_delete(struct comm_base* b) 204 { 205 if(!b) 206 return; 207 if(b->eb->slow_accept_enabled) { 208 if(ub_event_del(b->eb->slow_accept) != 0) { 209 log_err("could not event_del slow_accept"); 210 } 211 ub_event_free(b->eb->slow_accept); 212 } 213 ub_event_base_free(b->eb->base); 214 b->eb->base = NULL; 215 free(b->eb); 216 free(b); 217 } 218 219 void 220 comm_base_delete_no_base(struct comm_base* b) 221 { 222 if(!b) 223 return; 224 if(b->eb->slow_accept_enabled) { 225 if(ub_event_del(b->eb->slow_accept) != 0) { 226 log_err("could not event_del slow_accept"); 227 } 228 ub_event_free(b->eb->slow_accept); 229 } 230 b->eb->base = NULL; 231 free(b->eb); 232 free(b); 233 } 234 235 void 236 comm_base_timept(struct comm_base* b, time_t** tt, struct timeval** tv) 237 { 238 *tt = &b->eb->secs; 239 *tv = &b->eb->now; 240 } 241 242 void 243 comm_base_dispatch(struct comm_base* b) 244 { 245 int retval; 246 retval = ub_event_base_dispatch(b->eb->base); 247 if(retval < 0) { 248 fatal_exit("event_dispatch returned error %d, " 249 "errno is %s", retval, strerror(errno)); 250 } 251 } 252 253 void comm_base_exit(struct comm_base* b) 254 { 255 if(ub_event_base_loopexit(b->eb->base) != 0) { 256 log_err("Could not loopexit"); 257 } 258 } 259 260 void comm_base_set_slow_accept_handlers(struct comm_base* b, 261 void (*stop_acc)(void*), void (*start_acc)(void*), void* arg) 262 { 263 b->stop_accept = stop_acc; 264 b->start_accept = start_acc; 265 b->cb_arg = arg; 266 } 267 268 struct ub_event_base* comm_base_internal(struct comm_base* b) 269 { 270 return b->eb->base; 271 } 272 273 /** see if errno for udp has to be logged or not uses globals */ 274 static int 275 udp_send_errno_needs_log(struct sockaddr* addr, socklen_t addrlen) 276 { 277 /* do not log transient errors (unless high verbosity) */ 278 #if defined(ENETUNREACH) || defined(EHOSTDOWN) || defined(EHOSTUNREACH) || defined(ENETDOWN) 279 switch(errno) { 280 # ifdef ENETUNREACH 281 case ENETUNREACH: 282 # endif 283 # ifdef EHOSTDOWN 284 case EHOSTDOWN: 285 # endif 286 # ifdef EHOSTUNREACH 287 case EHOSTUNREACH: 288 # endif 289 # ifdef ENETDOWN 290 case ENETDOWN: 291 # endif 292 if(verbosity < VERB_ALGO) 293 return 0; 294 default: 295 break; 296 } 297 #endif 298 /* permission denied is gotten for every send if the 299 * network is disconnected (on some OS), squelch it */ 300 if( ((errno == EPERM) 301 # ifdef EADDRNOTAVAIL 302 /* 'Cannot assign requested address' also when disconnected */ 303 || (errno == EADDRNOTAVAIL) 304 # endif 305 ) && verbosity < VERB_DETAIL) 306 return 0; 307 # ifdef EADDRINUSE 308 /* If SO_REUSEADDR is set, we could try to connect to the same server 309 * from the same source port twice. */ 310 if(errno == EADDRINUSE && verbosity < VERB_DETAIL) 311 return 0; 312 # endif 313 /* squelch errors where people deploy AAAA ::ffff:bla for 314 * authority servers, which we try for intranets. */ 315 if(errno == EINVAL && addr_is_ip4mapped( 316 (struct sockaddr_storage*)addr, addrlen) && 317 verbosity < VERB_DETAIL) 318 return 0; 319 /* SO_BROADCAST sockopt can give access to 255.255.255.255, 320 * but a dns cache does not need it. */ 321 if(errno == EACCES && addr_is_broadcast( 322 (struct sockaddr_storage*)addr, addrlen) && 323 verbosity < VERB_DETAIL) 324 return 0; 325 return 1; 326 } 327 328 int tcp_connect_errno_needs_log(struct sockaddr* addr, socklen_t addrlen) 329 { 330 return udp_send_errno_needs_log(addr, addrlen); 331 } 332 333 /* send a UDP reply */ 334 int 335 comm_point_send_udp_msg(struct comm_point *c, sldns_buffer* packet, 336 struct sockaddr* addr, socklen_t addrlen) 337 { 338 ssize_t sent; 339 log_assert(c->fd != -1); 340 #ifdef UNBOUND_DEBUG 341 if(sldns_buffer_remaining(packet) == 0) 342 log_err("error: send empty UDP packet"); 343 #endif 344 log_assert(addr && addrlen > 0); 345 sent = sendto(c->fd, (void*)sldns_buffer_begin(packet), 346 sldns_buffer_remaining(packet), 0, 347 addr, addrlen); 348 if(sent == -1) { 349 /* try again and block, waiting for IO to complete, 350 * we want to send the answer, and we will wait for 351 * the ethernet interface buffer to have space. */ 352 #ifndef USE_WINSOCK 353 if(errno == EAGAIN || 354 # ifdef EWOULDBLOCK 355 errno == EWOULDBLOCK || 356 # endif 357 errno == ENOBUFS) { 358 #else 359 if(WSAGetLastError() == WSAEINPROGRESS || 360 WSAGetLastError() == WSAENOBUFS || 361 WSAGetLastError() == WSAEWOULDBLOCK) { 362 #endif 363 int e; 364 fd_set_block(c->fd); 365 sent = sendto(c->fd, (void*)sldns_buffer_begin(packet), 366 sldns_buffer_remaining(packet), 0, 367 addr, addrlen); 368 e = errno; 369 fd_set_nonblock(c->fd); 370 errno = e; 371 } 372 } 373 if(sent == -1) { 374 if(!udp_send_errno_needs_log(addr, addrlen)) 375 return 0; 376 #ifndef USE_WINSOCK 377 verbose(VERB_OPS, "sendto failed: %s", strerror(errno)); 378 #else 379 verbose(VERB_OPS, "sendto failed: %s", 380 wsa_strerror(WSAGetLastError())); 381 #endif 382 log_addr(VERB_OPS, "remote address is", 383 (struct sockaddr_storage*)addr, addrlen); 384 return 0; 385 } else if((size_t)sent != sldns_buffer_remaining(packet)) { 386 log_err("sent %d in place of %d bytes", 387 (int)sent, (int)sldns_buffer_remaining(packet)); 388 return 0; 389 } 390 return 1; 391 } 392 393 #if defined(AF_INET6) && defined(IPV6_PKTINFO) && (defined(HAVE_RECVMSG) || defined(HAVE_SENDMSG)) 394 /** print debug ancillary info */ 395 static void p_ancil(const char* str, struct comm_reply* r) 396 { 397 if(r->srctype != 4 && r->srctype != 6) { 398 log_info("%s: unknown srctype %d", str, r->srctype); 399 return; 400 } 401 if(r->srctype == 6) { 402 char buf[1024]; 403 if(inet_ntop(AF_INET6, &r->pktinfo.v6info.ipi6_addr, 404 buf, (socklen_t)sizeof(buf)) == 0) { 405 (void)strlcpy(buf, "(inet_ntop error)", sizeof(buf)); 406 } 407 buf[sizeof(buf)-1]=0; 408 log_info("%s: %s %d", str, buf, r->pktinfo.v6info.ipi6_ifindex); 409 } else if(r->srctype == 4) { 410 #ifdef IP_PKTINFO 411 char buf1[1024], buf2[1024]; 412 if(inet_ntop(AF_INET, &r->pktinfo.v4info.ipi_addr, 413 buf1, (socklen_t)sizeof(buf1)) == 0) { 414 (void)strlcpy(buf1, "(inet_ntop error)", sizeof(buf1)); 415 } 416 buf1[sizeof(buf1)-1]=0; 417 #ifdef HAVE_STRUCT_IN_PKTINFO_IPI_SPEC_DST 418 if(inet_ntop(AF_INET, &r->pktinfo.v4info.ipi_spec_dst, 419 buf2, (socklen_t)sizeof(buf2)) == 0) { 420 (void)strlcpy(buf2, "(inet_ntop error)", sizeof(buf2)); 421 } 422 buf2[sizeof(buf2)-1]=0; 423 #else 424 buf2[0]=0; 425 #endif 426 log_info("%s: %d %s %s", str, r->pktinfo.v4info.ipi_ifindex, 427 buf1, buf2); 428 #elif defined(IP_RECVDSTADDR) 429 char buf1[1024]; 430 if(inet_ntop(AF_INET, &r->pktinfo.v4addr, 431 buf1, (socklen_t)sizeof(buf1)) == 0) { 432 (void)strlcpy(buf1, "(inet_ntop error)", sizeof(buf1)); 433 } 434 buf1[sizeof(buf1)-1]=0; 435 log_info("%s: %s", str, buf1); 436 #endif /* IP_PKTINFO or PI_RECVDSTDADDR */ 437 } 438 } 439 #endif /* AF_INET6 && IPV6_PKTINFO && HAVE_RECVMSG||HAVE_SENDMSG */ 440 441 /** send a UDP reply over specified interface*/ 442 static int 443 comm_point_send_udp_msg_if(struct comm_point *c, sldns_buffer* packet, 444 struct sockaddr* addr, socklen_t addrlen, struct comm_reply* r) 445 { 446 #if defined(AF_INET6) && defined(IPV6_PKTINFO) && defined(HAVE_SENDMSG) 447 ssize_t sent; 448 struct msghdr msg; 449 struct iovec iov[1]; 450 char control[256]; 451 #ifndef S_SPLINT_S 452 struct cmsghdr *cmsg; 453 #endif /* S_SPLINT_S */ 454 455 log_assert(c->fd != -1); 456 #ifdef UNBOUND_DEBUG 457 if(sldns_buffer_remaining(packet) == 0) 458 log_err("error: send empty UDP packet"); 459 #endif 460 log_assert(addr && addrlen > 0); 461 462 msg.msg_name = addr; 463 msg.msg_namelen = addrlen; 464 iov[0].iov_base = sldns_buffer_begin(packet); 465 iov[0].iov_len = sldns_buffer_remaining(packet); 466 msg.msg_iov = iov; 467 msg.msg_iovlen = 1; 468 msg.msg_control = control; 469 #ifndef S_SPLINT_S 470 msg.msg_controllen = sizeof(control); 471 #endif /* S_SPLINT_S */ 472 msg.msg_flags = 0; 473 474 #ifndef S_SPLINT_S 475 cmsg = CMSG_FIRSTHDR(&msg); 476 if(r->srctype == 4) { 477 #ifdef IP_PKTINFO 478 void* cmsg_data; 479 msg.msg_controllen = CMSG_SPACE(sizeof(struct in_pktinfo)); 480 log_assert(msg.msg_controllen <= sizeof(control)); 481 cmsg->cmsg_level = IPPROTO_IP; 482 cmsg->cmsg_type = IP_PKTINFO; 483 memmove(CMSG_DATA(cmsg), &r->pktinfo.v4info, 484 sizeof(struct in_pktinfo)); 485 /* unset the ifindex to not bypass the routing tables */ 486 cmsg_data = CMSG_DATA(cmsg); 487 ((struct in_pktinfo *) cmsg_data)->ipi_ifindex = 0; 488 cmsg->cmsg_len = CMSG_LEN(sizeof(struct in_pktinfo)); 489 #elif defined(IP_SENDSRCADDR) 490 msg.msg_controllen = CMSG_SPACE(sizeof(struct in_addr)); 491 log_assert(msg.msg_controllen <= sizeof(control)); 492 cmsg->cmsg_level = IPPROTO_IP; 493 cmsg->cmsg_type = IP_SENDSRCADDR; 494 memmove(CMSG_DATA(cmsg), &r->pktinfo.v4addr, 495 sizeof(struct in_addr)); 496 cmsg->cmsg_len = CMSG_LEN(sizeof(struct in_addr)); 497 #else 498 verbose(VERB_ALGO, "no IP_PKTINFO or IP_SENDSRCADDR"); 499 msg.msg_control = NULL; 500 #endif /* IP_PKTINFO or IP_SENDSRCADDR */ 501 } else if(r->srctype == 6) { 502 void* cmsg_data; 503 msg.msg_controllen = CMSG_SPACE(sizeof(struct in6_pktinfo)); 504 log_assert(msg.msg_controllen <= sizeof(control)); 505 cmsg->cmsg_level = IPPROTO_IPV6; 506 cmsg->cmsg_type = IPV6_PKTINFO; 507 memmove(CMSG_DATA(cmsg), &r->pktinfo.v6info, 508 sizeof(struct in6_pktinfo)); 509 /* unset the ifindex to not bypass the routing tables */ 510 cmsg_data = CMSG_DATA(cmsg); 511 ((struct in6_pktinfo *) cmsg_data)->ipi6_ifindex = 0; 512 cmsg->cmsg_len = CMSG_LEN(sizeof(struct in6_pktinfo)); 513 } else { 514 /* try to pass all 0 to use default route */ 515 msg.msg_controllen = CMSG_SPACE(sizeof(struct in6_pktinfo)); 516 log_assert(msg.msg_controllen <= sizeof(control)); 517 cmsg->cmsg_level = IPPROTO_IPV6; 518 cmsg->cmsg_type = IPV6_PKTINFO; 519 memset(CMSG_DATA(cmsg), 0, sizeof(struct in6_pktinfo)); 520 cmsg->cmsg_len = CMSG_LEN(sizeof(struct in6_pktinfo)); 521 } 522 #endif /* S_SPLINT_S */ 523 if(verbosity >= VERB_ALGO) 524 p_ancil("send_udp over interface", r); 525 sent = sendmsg(c->fd, &msg, 0); 526 if(sent == -1) { 527 /* try again and block, waiting for IO to complete, 528 * we want to send the answer, and we will wait for 529 * the ethernet interface buffer to have space. */ 530 #ifndef USE_WINSOCK 531 if(errno == EAGAIN || 532 # ifdef EWOULDBLOCK 533 errno == EWOULDBLOCK || 534 # endif 535 errno == ENOBUFS) { 536 #else 537 if(WSAGetLastError() == WSAEINPROGRESS || 538 WSAGetLastError() == WSAENOBUFS || 539 WSAGetLastError() == WSAEWOULDBLOCK) { 540 #endif 541 int e; 542 fd_set_block(c->fd); 543 sent = sendmsg(c->fd, &msg, 0); 544 e = errno; 545 fd_set_nonblock(c->fd); 546 errno = e; 547 } 548 } 549 if(sent == -1) { 550 if(!udp_send_errno_needs_log(addr, addrlen)) 551 return 0; 552 verbose(VERB_OPS, "sendmsg failed: %s", strerror(errno)); 553 log_addr(VERB_OPS, "remote address is", 554 (struct sockaddr_storage*)addr, addrlen); 555 #ifdef __NetBSD__ 556 /* netbsd 7 has IP_PKTINFO for recv but not send */ 557 if(errno == EINVAL && r->srctype == 4) 558 log_err("sendmsg: No support for sendmsg(IP_PKTINFO). " 559 "Please disable interface-automatic"); 560 #endif 561 return 0; 562 } else if((size_t)sent != sldns_buffer_remaining(packet)) { 563 log_err("sent %d in place of %d bytes", 564 (int)sent, (int)sldns_buffer_remaining(packet)); 565 return 0; 566 } 567 return 1; 568 #else 569 (void)c; 570 (void)packet; 571 (void)addr; 572 (void)addrlen; 573 (void)r; 574 log_err("sendmsg: IPV6_PKTINFO not supported"); 575 return 0; 576 #endif /* AF_INET6 && IPV6_PKTINFO && HAVE_SENDMSG */ 577 } 578 579 void 580 comm_point_udp_ancil_callback(int fd, short event, void* arg) 581 { 582 #if defined(AF_INET6) && defined(IPV6_PKTINFO) && defined(HAVE_RECVMSG) 583 struct comm_reply rep; 584 struct msghdr msg; 585 struct iovec iov[1]; 586 ssize_t rcv; 587 char ancil[256]; 588 int i; 589 #ifndef S_SPLINT_S 590 struct cmsghdr* cmsg; 591 #endif /* S_SPLINT_S */ 592 593 rep.c = (struct comm_point*)arg; 594 log_assert(rep.c->type == comm_udp); 595 596 if(!(event&UB_EV_READ)) 597 return; 598 log_assert(rep.c && rep.c->buffer && rep.c->fd == fd); 599 ub_comm_base_now(rep.c->ev->base); 600 for(i=0; i<NUM_UDP_PER_SELECT; i++) { 601 sldns_buffer_clear(rep.c->buffer); 602 rep.addrlen = (socklen_t)sizeof(rep.addr); 603 log_assert(fd != -1); 604 log_assert(sldns_buffer_remaining(rep.c->buffer) > 0); 605 msg.msg_name = &rep.addr; 606 msg.msg_namelen = (socklen_t)sizeof(rep.addr); 607 iov[0].iov_base = sldns_buffer_begin(rep.c->buffer); 608 iov[0].iov_len = sldns_buffer_remaining(rep.c->buffer); 609 msg.msg_iov = iov; 610 msg.msg_iovlen = 1; 611 msg.msg_control = ancil; 612 #ifndef S_SPLINT_S 613 msg.msg_controllen = sizeof(ancil); 614 #endif /* S_SPLINT_S */ 615 msg.msg_flags = 0; 616 rcv = recvmsg(fd, &msg, 0); 617 if(rcv == -1) { 618 if(errno != EAGAIN && errno != EINTR) { 619 log_err("recvmsg failed: %s", strerror(errno)); 620 } 621 return; 622 } 623 rep.addrlen = msg.msg_namelen; 624 sldns_buffer_skip(rep.c->buffer, rcv); 625 sldns_buffer_flip(rep.c->buffer); 626 rep.srctype = 0; 627 #ifndef S_SPLINT_S 628 for(cmsg = CMSG_FIRSTHDR(&msg); cmsg != NULL; 629 cmsg = CMSG_NXTHDR(&msg, cmsg)) { 630 if( cmsg->cmsg_level == IPPROTO_IPV6 && 631 cmsg->cmsg_type == IPV6_PKTINFO) { 632 rep.srctype = 6; 633 memmove(&rep.pktinfo.v6info, CMSG_DATA(cmsg), 634 sizeof(struct in6_pktinfo)); 635 break; 636 #ifdef IP_PKTINFO 637 } else if( cmsg->cmsg_level == IPPROTO_IP && 638 cmsg->cmsg_type == IP_PKTINFO) { 639 rep.srctype = 4; 640 memmove(&rep.pktinfo.v4info, CMSG_DATA(cmsg), 641 sizeof(struct in_pktinfo)); 642 break; 643 #elif defined(IP_RECVDSTADDR) 644 } else if( cmsg->cmsg_level == IPPROTO_IP && 645 cmsg->cmsg_type == IP_RECVDSTADDR) { 646 rep.srctype = 4; 647 memmove(&rep.pktinfo.v4addr, CMSG_DATA(cmsg), 648 sizeof(struct in_addr)); 649 break; 650 #endif /* IP_PKTINFO or IP_RECVDSTADDR */ 651 } 652 } 653 if(verbosity >= VERB_ALGO) 654 p_ancil("receive_udp on interface", &rep); 655 #endif /* S_SPLINT_S */ 656 fptr_ok(fptr_whitelist_comm_point(rep.c->callback)); 657 if((*rep.c->callback)(rep.c, rep.c->cb_arg, NETEVENT_NOERROR, &rep)) { 658 /* send back immediate reply */ 659 (void)comm_point_send_udp_msg_if(rep.c, rep.c->buffer, 660 (struct sockaddr*)&rep.addr, rep.addrlen, &rep); 661 } 662 if(!rep.c || rep.c->fd == -1) /* commpoint closed */ 663 break; 664 } 665 #else 666 (void)fd; 667 (void)event; 668 (void)arg; 669 fatal_exit("recvmsg: No support for IPV6_PKTINFO; IP_PKTINFO or IP_RECVDSTADDR. " 670 "Please disable interface-automatic"); 671 #endif /* AF_INET6 && IPV6_PKTINFO && HAVE_RECVMSG */ 672 } 673 674 void 675 comm_point_udp_callback(int fd, short event, void* arg) 676 { 677 struct comm_reply rep; 678 ssize_t rcv; 679 int i; 680 struct sldns_buffer *buffer; 681 682 rep.c = (struct comm_point*)arg; 683 log_assert(rep.c->type == comm_udp); 684 685 if(!(event&UB_EV_READ)) 686 return; 687 log_assert(rep.c && rep.c->buffer && rep.c->fd == fd); 688 ub_comm_base_now(rep.c->ev->base); 689 for(i=0; i<NUM_UDP_PER_SELECT; i++) { 690 sldns_buffer_clear(rep.c->buffer); 691 rep.addrlen = (socklen_t)sizeof(rep.addr); 692 log_assert(fd != -1); 693 log_assert(sldns_buffer_remaining(rep.c->buffer) > 0); 694 rcv = recvfrom(fd, (void*)sldns_buffer_begin(rep.c->buffer), 695 sldns_buffer_remaining(rep.c->buffer), 0, 696 (struct sockaddr*)&rep.addr, &rep.addrlen); 697 if(rcv == -1) { 698 #ifndef USE_WINSOCK 699 if(errno != EAGAIN && errno != EINTR) 700 log_err("recvfrom %d failed: %s", 701 fd, strerror(errno)); 702 #else 703 if(WSAGetLastError() != WSAEINPROGRESS && 704 WSAGetLastError() != WSAECONNRESET && 705 WSAGetLastError()!= WSAEWOULDBLOCK) 706 log_err("recvfrom failed: %s", 707 wsa_strerror(WSAGetLastError())); 708 #endif 709 return; 710 } 711 sldns_buffer_skip(rep.c->buffer, rcv); 712 sldns_buffer_flip(rep.c->buffer); 713 rep.srctype = 0; 714 fptr_ok(fptr_whitelist_comm_point(rep.c->callback)); 715 if((*rep.c->callback)(rep.c, rep.c->cb_arg, NETEVENT_NOERROR, &rep)) { 716 /* send back immediate reply */ 717 #ifdef USE_DNSCRYPT 718 buffer = rep.c->dnscrypt_buffer; 719 #else 720 buffer = rep.c->buffer; 721 #endif 722 (void)comm_point_send_udp_msg(rep.c, buffer, 723 (struct sockaddr*)&rep.addr, rep.addrlen); 724 } 725 if(!rep.c || rep.c->fd != fd) /* commpoint closed to -1 or reused for 726 another UDP port. Note rep.c cannot be reused with TCP fd. */ 727 break; 728 } 729 } 730 731 /** Use a new tcp handler for new query fd, set to read query */ 732 static void 733 setup_tcp_handler(struct comm_point* c, int fd, int cur, int max) 734 { 735 int handler_usage; 736 log_assert(c->type == comm_tcp); 737 log_assert(c->fd == -1); 738 sldns_buffer_clear(c->buffer); 739 #ifdef USE_DNSCRYPT 740 if (c->dnscrypt) 741 sldns_buffer_clear(c->dnscrypt_buffer); 742 #endif 743 c->tcp_is_reading = 1; 744 c->tcp_byte_count = 0; 745 /* if more than half the tcp handlers are in use, use a shorter 746 * timeout for this TCP connection, we need to make space for 747 * other connections to be able to get attention */ 748 /* If > 50% TCP handler structures in use, set timeout to 1/100th 749 * configured value. 750 * If > 65%TCP handler structures in use, set to 1/500th configured 751 * value. 752 * If > 80% TCP handler structures in use, set to 0. 753 * 754 * If the timeout to use falls below 200 milliseconds, an actual 755 * timeout of 200ms is used. 756 */ 757 handler_usage = (cur * 100) / max; 758 if(handler_usage > 50 && handler_usage <= 65) 759 c->tcp_timeout_msec /= 100; 760 else if (handler_usage > 65 && handler_usage <= 80) 761 c->tcp_timeout_msec /= 500; 762 else if (handler_usage > 80) 763 c->tcp_timeout_msec = 0; 764 comm_point_start_listening(c, fd, 765 c->tcp_timeout_msec < TCP_QUERY_TIMEOUT_MINIMUM 766 ? TCP_QUERY_TIMEOUT_MINIMUM 767 : c->tcp_timeout_msec); 768 } 769 770 void comm_base_handle_slow_accept(int ATTR_UNUSED(fd), 771 short ATTR_UNUSED(event), void* arg) 772 { 773 struct comm_base* b = (struct comm_base*)arg; 774 /* timeout for the slow accept, re-enable accepts again */ 775 if(b->start_accept) { 776 verbose(VERB_ALGO, "wait is over, slow accept disabled"); 777 fptr_ok(fptr_whitelist_start_accept(b->start_accept)); 778 (*b->start_accept)(b->cb_arg); 779 b->eb->slow_accept_enabled = 0; 780 } 781 } 782 783 int comm_point_perform_accept(struct comm_point* c, 784 struct sockaddr_storage* addr, socklen_t* addrlen) 785 { 786 int new_fd; 787 *addrlen = (socklen_t)sizeof(*addr); 788 #ifndef HAVE_ACCEPT4 789 new_fd = accept(c->fd, (struct sockaddr*)addr, addrlen); 790 #else 791 /* SOCK_NONBLOCK saves extra calls to fcntl for the same result */ 792 new_fd = accept4(c->fd, (struct sockaddr*)addr, addrlen, SOCK_NONBLOCK); 793 #endif 794 if(new_fd == -1) { 795 #ifndef USE_WINSOCK 796 /* EINTR is signal interrupt. others are closed connection. */ 797 if( errno == EINTR || errno == EAGAIN 798 #ifdef EWOULDBLOCK 799 || errno == EWOULDBLOCK 800 #endif 801 #ifdef ECONNABORTED 802 || errno == ECONNABORTED 803 #endif 804 #ifdef EPROTO 805 || errno == EPROTO 806 #endif /* EPROTO */ 807 ) 808 return -1; 809 #if defined(ENFILE) && defined(EMFILE) 810 if(errno == ENFILE || errno == EMFILE) { 811 /* out of file descriptors, likely outside of our 812 * control. stop accept() calls for some time */ 813 if(c->ev->base->stop_accept) { 814 struct comm_base* b = c->ev->base; 815 struct timeval tv; 816 verbose(VERB_ALGO, "out of file descriptors: " 817 "slow accept"); 818 b->eb->slow_accept_enabled = 1; 819 fptr_ok(fptr_whitelist_stop_accept( 820 b->stop_accept)); 821 (*b->stop_accept)(b->cb_arg); 822 /* set timeout, no mallocs */ 823 tv.tv_sec = NETEVENT_SLOW_ACCEPT_TIME/1000; 824 tv.tv_usec = (NETEVENT_SLOW_ACCEPT_TIME%1000)*1000; 825 b->eb->slow_accept = ub_event_new(b->eb->base, 826 -1, UB_EV_TIMEOUT, 827 comm_base_handle_slow_accept, b); 828 if(b->eb->slow_accept == NULL) { 829 /* we do not want to log here, because 830 * that would spam the logfiles. 831 * error: "event_base_set failed." */ 832 } 833 else if(ub_event_add(b->eb->slow_accept, &tv) 834 != 0) { 835 /* we do not want to log here, 836 * error: "event_add failed." */ 837 } 838 } 839 return -1; 840 } 841 #endif 842 log_err_addr("accept failed", strerror(errno), addr, *addrlen); 843 #else /* USE_WINSOCK */ 844 if(WSAGetLastError() == WSAEINPROGRESS || 845 WSAGetLastError() == WSAECONNRESET) 846 return -1; 847 if(WSAGetLastError() == WSAEWOULDBLOCK) { 848 ub_winsock_tcp_wouldblock(c->ev->ev, UB_EV_READ); 849 return -1; 850 } 851 log_err_addr("accept failed", wsa_strerror(WSAGetLastError()), 852 addr, *addrlen); 853 #endif 854 return -1; 855 } 856 if(c->tcp_conn_limit && c->type == comm_tcp_accept) { 857 c->tcl_addr = tcl_addr_lookup(c->tcp_conn_limit, addr, *addrlen); 858 if(!tcl_new_connection(c->tcl_addr)) { 859 if(verbosity >= 3) 860 log_err_addr("accept rejected", 861 "connection limit exceeded", addr, *addrlen); 862 close(new_fd); 863 return -1; 864 } 865 } 866 #ifndef HAVE_ACCEPT4 867 fd_set_nonblock(new_fd); 868 #endif 869 return new_fd; 870 } 871 872 #ifdef USE_WINSOCK 873 static long win_bio_cb(BIO *b, int oper, const char* ATTR_UNUSED(argp), 874 int ATTR_UNUSED(argi), long argl, long retvalue) 875 { 876 int wsa_err = WSAGetLastError(); /* store errcode before it is gone */ 877 verbose(VERB_ALGO, "bio_cb %d, %s %s %s", oper, 878 (oper&BIO_CB_RETURN)?"return":"before", 879 (oper&BIO_CB_READ)?"read":((oper&BIO_CB_WRITE)?"write":"other"), 880 wsa_err==WSAEWOULDBLOCK?"wsawb":""); 881 /* on windows, check if previous operation caused EWOULDBLOCK */ 882 if( (oper == (BIO_CB_READ|BIO_CB_RETURN) && argl == 0) || 883 (oper == (BIO_CB_GETS|BIO_CB_RETURN) && argl == 0)) { 884 if(wsa_err == WSAEWOULDBLOCK) 885 ub_winsock_tcp_wouldblock((struct ub_event*) 886 BIO_get_callback_arg(b), UB_EV_READ); 887 } 888 if( (oper == (BIO_CB_WRITE|BIO_CB_RETURN) && argl == 0) || 889 (oper == (BIO_CB_PUTS|BIO_CB_RETURN) && argl == 0)) { 890 if(wsa_err == WSAEWOULDBLOCK) 891 ub_winsock_tcp_wouldblock((struct ub_event*) 892 BIO_get_callback_arg(b), UB_EV_WRITE); 893 } 894 /* return original return value */ 895 return retvalue; 896 } 897 898 /** set win bio callbacks for nonblocking operations */ 899 void 900 comm_point_tcp_win_bio_cb(struct comm_point* c, void* thessl) 901 { 902 SSL* ssl = (SSL*)thessl; 903 /* set them both just in case, but usually they are the same BIO */ 904 BIO_set_callback(SSL_get_rbio(ssl), &win_bio_cb); 905 BIO_set_callback_arg(SSL_get_rbio(ssl), (char*)c->ev->ev); 906 BIO_set_callback(SSL_get_wbio(ssl), &win_bio_cb); 907 BIO_set_callback_arg(SSL_get_wbio(ssl), (char*)c->ev->ev); 908 } 909 #endif 910 911 void 912 comm_point_tcp_accept_callback(int fd, short event, void* arg) 913 { 914 struct comm_point* c = (struct comm_point*)arg, *c_hdl; 915 int new_fd; 916 log_assert(c->type == comm_tcp_accept); 917 if(!(event & UB_EV_READ)) { 918 log_info("ignoring tcp accept event %d", (int)event); 919 return; 920 } 921 ub_comm_base_now(c->ev->base); 922 /* find free tcp handler. */ 923 if(!c->tcp_free) { 924 log_warn("accepted too many tcp, connections full"); 925 return; 926 } 927 /* accept incoming connection. */ 928 c_hdl = c->tcp_free; 929 /* clear leftover flags from previous use, and then set the 930 * correct event base for the event structure for libevent */ 931 ub_event_free(c_hdl->ev->ev); 932 c_hdl->ev->ev = ub_event_new(c_hdl->ev->base->eb->base, -1, UB_EV_PERSIST | UB_EV_READ | UB_EV_TIMEOUT, comm_point_tcp_handle_callback, c_hdl); 933 if(!c_hdl->ev->ev) { 934 log_warn("could not ub_event_new, dropped tcp"); 935 return; 936 } 937 log_assert(fd != -1); 938 (void)fd; 939 new_fd = comm_point_perform_accept(c, &c_hdl->repinfo.addr, 940 &c_hdl->repinfo.addrlen); 941 if(new_fd == -1) 942 return; 943 if(c->ssl) { 944 c_hdl->ssl = incoming_ssl_fd(c->ssl, new_fd); 945 if(!c_hdl->ssl) { 946 c_hdl->fd = new_fd; 947 comm_point_close(c_hdl); 948 return; 949 } 950 c_hdl->ssl_shake_state = comm_ssl_shake_read; 951 #ifdef USE_WINSOCK 952 comm_point_tcp_win_bio_cb(c_hdl, c_hdl->ssl); 953 #endif 954 } 955 956 /* grab the tcp handler buffers */ 957 c->cur_tcp_count++; 958 c->tcp_free = c_hdl->tcp_free; 959 if(!c->tcp_free) { 960 /* stop accepting incoming queries for now. */ 961 comm_point_stop_listening(c); 962 } 963 setup_tcp_handler(c_hdl, new_fd, c->cur_tcp_count, c->max_tcp_count); 964 } 965 966 /** Make tcp handler free for next assignment */ 967 static void 968 reclaim_tcp_handler(struct comm_point* c) 969 { 970 log_assert(c->type == comm_tcp); 971 if(c->ssl) { 972 #ifdef HAVE_SSL 973 SSL_shutdown(c->ssl); 974 SSL_free(c->ssl); 975 c->ssl = NULL; 976 #endif 977 } 978 comm_point_close(c); 979 if(c->tcp_parent) { 980 c->tcp_parent->cur_tcp_count--; 981 c->tcp_free = c->tcp_parent->tcp_free; 982 c->tcp_parent->tcp_free = c; 983 if(!c->tcp_free) { 984 /* re-enable listening on accept socket */ 985 comm_point_start_listening(c->tcp_parent, -1, -1); 986 } 987 } 988 } 989 990 /** do the callback when writing is done */ 991 static void 992 tcp_callback_writer(struct comm_point* c) 993 { 994 log_assert(c->type == comm_tcp); 995 sldns_buffer_clear(c->buffer); 996 if(c->tcp_do_toggle_rw) 997 c->tcp_is_reading = 1; 998 c->tcp_byte_count = 0; 999 /* switch from listening(write) to listening(read) */ 1000 if(c->tcp_req_info) { 1001 tcp_req_info_handle_writedone(c->tcp_req_info); 1002 } else { 1003 comm_point_stop_listening(c); 1004 comm_point_start_listening(c, -1, c->tcp_timeout_msec); 1005 } 1006 } 1007 1008 /** do the callback when reading is done */ 1009 static void 1010 tcp_callback_reader(struct comm_point* c) 1011 { 1012 log_assert(c->type == comm_tcp || c->type == comm_local); 1013 sldns_buffer_flip(c->buffer); 1014 if(c->tcp_do_toggle_rw) 1015 c->tcp_is_reading = 0; 1016 c->tcp_byte_count = 0; 1017 if(c->tcp_req_info) { 1018 tcp_req_info_handle_readdone(c->tcp_req_info); 1019 } else { 1020 if(c->type == comm_tcp) 1021 comm_point_stop_listening(c); 1022 fptr_ok(fptr_whitelist_comm_point(c->callback)); 1023 if( (*c->callback)(c, c->cb_arg, NETEVENT_NOERROR, &c->repinfo) ) { 1024 comm_point_start_listening(c, -1, c->tcp_timeout_msec); 1025 } 1026 } 1027 } 1028 1029 #ifdef HAVE_SSL 1030 /** log certificate details */ 1031 static void 1032 log_cert(unsigned level, const char* str, X509* cert) 1033 { 1034 BIO* bio; 1035 char nul = 0; 1036 char* pp = NULL; 1037 long len; 1038 if(verbosity < level) return; 1039 bio = BIO_new(BIO_s_mem()); 1040 if(!bio) return; 1041 X509_print_ex(bio, cert, 0, (unsigned long)-1 1042 ^(X509_FLAG_NO_SUBJECT 1043 |X509_FLAG_NO_ISSUER|X509_FLAG_NO_VALIDITY 1044 |X509_FLAG_NO_EXTENSIONS|X509_FLAG_NO_AUX 1045 |X509_FLAG_NO_ATTRIBUTES)); 1046 BIO_write(bio, &nul, (int)sizeof(nul)); 1047 len = BIO_get_mem_data(bio, &pp); 1048 if(len != 0 && pp) { 1049 verbose(level, "%s: \n%s", str, pp); 1050 } 1051 BIO_free(bio); 1052 } 1053 #endif /* HAVE_SSL */ 1054 1055 #ifdef HAVE_SSL 1056 /** true if the ssl handshake error has to be squelched from the logs */ 1057 static int 1058 squelch_err_ssl_handshake(unsigned long err) 1059 { 1060 if(verbosity >= VERB_QUERY) 1061 return 0; /* only squelch on low verbosity */ 1062 /* this is very specific, we could filter on ERR_GET_REASON() 1063 * (the third element in ERR_PACK) */ 1064 if(err == ERR_PACK(ERR_LIB_SSL, SSL_F_SSL3_GET_RECORD, SSL_R_HTTPS_PROXY_REQUEST) || 1065 err == ERR_PACK(ERR_LIB_SSL, SSL_F_SSL3_GET_RECORD, SSL_R_HTTP_REQUEST) || 1066 err == ERR_PACK(ERR_LIB_SSL, SSL_F_SSL3_GET_RECORD, SSL_R_WRONG_VERSION_NUMBER) || 1067 err == ERR_PACK(ERR_LIB_SSL, SSL_F_SSL3_READ_BYTES, SSL_R_SSLV3_ALERT_BAD_CERTIFICATE) 1068 #ifdef SSL_F_TLS_POST_PROCESS_CLIENT_HELLO 1069 || err == ERR_PACK(ERR_LIB_SSL, SSL_F_TLS_POST_PROCESS_CLIENT_HELLO, SSL_R_NO_SHARED_CIPHER) 1070 #endif 1071 #ifdef SSL_F_TLS_EARLY_POST_PROCESS_CLIENT_HELLO 1072 || err == ERR_PACK(ERR_LIB_SSL, SSL_F_TLS_EARLY_POST_PROCESS_CLIENT_HELLO, SSL_R_UNKNOWN_PROTOCOL) 1073 || err == ERR_PACK(ERR_LIB_SSL, SSL_F_TLS_EARLY_POST_PROCESS_CLIENT_HELLO, SSL_R_UNSUPPORTED_PROTOCOL) 1074 # ifdef SSL_R_VERSION_TOO_LOW 1075 || err == ERR_PACK(ERR_LIB_SSL, SSL_F_TLS_EARLY_POST_PROCESS_CLIENT_HELLO, SSL_R_VERSION_TOO_LOW) 1076 # endif 1077 #endif 1078 ) 1079 return 1; 1080 return 0; 1081 } 1082 #endif /* HAVE_SSL */ 1083 1084 /** continue ssl handshake */ 1085 #ifdef HAVE_SSL 1086 static int 1087 ssl_handshake(struct comm_point* c) 1088 { 1089 int r; 1090 if(c->ssl_shake_state == comm_ssl_shake_hs_read) { 1091 /* read condition satisfied back to writing */ 1092 comm_point_listen_for_rw(c, 1, 1); 1093 c->ssl_shake_state = comm_ssl_shake_none; 1094 return 1; 1095 } 1096 if(c->ssl_shake_state == comm_ssl_shake_hs_write) { 1097 /* write condition satisfied, back to reading */ 1098 comm_point_listen_for_rw(c, 1, 0); 1099 c->ssl_shake_state = comm_ssl_shake_none; 1100 return 1; 1101 } 1102 1103 ERR_clear_error(); 1104 r = SSL_do_handshake(c->ssl); 1105 if(r != 1) { 1106 int want = SSL_get_error(c->ssl, r); 1107 if(want == SSL_ERROR_WANT_READ) { 1108 if(c->ssl_shake_state == comm_ssl_shake_read) 1109 return 1; 1110 c->ssl_shake_state = comm_ssl_shake_read; 1111 comm_point_listen_for_rw(c, 1, 0); 1112 return 1; 1113 } else if(want == SSL_ERROR_WANT_WRITE) { 1114 if(c->ssl_shake_state == comm_ssl_shake_write) 1115 return 1; 1116 c->ssl_shake_state = comm_ssl_shake_write; 1117 comm_point_listen_for_rw(c, 0, 1); 1118 return 1; 1119 } else if(r == 0) { 1120 return 0; /* closed */ 1121 } else if(want == SSL_ERROR_SYSCALL) { 1122 /* SYSCALL and errno==0 means closed uncleanly */ 1123 if(errno != 0) 1124 log_err("SSL_handshake syscall: %s", 1125 strerror(errno)); 1126 return 0; 1127 } else { 1128 unsigned long err = ERR_get_error(); 1129 if(!squelch_err_ssl_handshake(err)) { 1130 log_crypto_err_code("ssl handshake failed", err); 1131 log_addr(VERB_OPS, "ssl handshake failed", &c->repinfo.addr, 1132 c->repinfo.addrlen); 1133 } 1134 return 0; 1135 } 1136 } 1137 /* this is where peer verification could take place */ 1138 if((SSL_get_verify_mode(c->ssl)&SSL_VERIFY_PEER)) { 1139 /* verification */ 1140 if(SSL_get_verify_result(c->ssl) == X509_V_OK) { 1141 X509* x = SSL_get_peer_certificate(c->ssl); 1142 if(!x) { 1143 log_addr(VERB_ALGO, "SSL connection failed: " 1144 "no certificate", 1145 &c->repinfo.addr, c->repinfo.addrlen); 1146 return 0; 1147 } 1148 log_cert(VERB_ALGO, "peer certificate", x); 1149 #ifdef HAVE_SSL_GET0_PEERNAME 1150 if(SSL_get0_peername(c->ssl)) { 1151 char buf[255]; 1152 snprintf(buf, sizeof(buf), "SSL connection " 1153 "to %s authenticated", 1154 SSL_get0_peername(c->ssl)); 1155 log_addr(VERB_ALGO, buf, &c->repinfo.addr, 1156 c->repinfo.addrlen); 1157 } else { 1158 #endif 1159 log_addr(VERB_ALGO, "SSL connection " 1160 "authenticated", &c->repinfo.addr, 1161 c->repinfo.addrlen); 1162 #ifdef HAVE_SSL_GET0_PEERNAME 1163 } 1164 #endif 1165 X509_free(x); 1166 } else { 1167 X509* x = SSL_get_peer_certificate(c->ssl); 1168 if(x) { 1169 log_cert(VERB_ALGO, "peer certificate", x); 1170 X509_free(x); 1171 } 1172 log_addr(VERB_ALGO, "SSL connection failed: " 1173 "failed to authenticate", 1174 &c->repinfo.addr, c->repinfo.addrlen); 1175 return 0; 1176 } 1177 } else { 1178 /* unauthenticated, the verify peer flag was not set 1179 * in c->ssl when the ssl object was created from ssl_ctx */ 1180 log_addr(VERB_ALGO, "SSL connection", &c->repinfo.addr, 1181 c->repinfo.addrlen); 1182 } 1183 1184 /* setup listen rw correctly */ 1185 if(c->tcp_is_reading) { 1186 if(c->ssl_shake_state != comm_ssl_shake_read) 1187 comm_point_listen_for_rw(c, 1, 0); 1188 } else { 1189 comm_point_listen_for_rw(c, 1, 1); 1190 } 1191 c->ssl_shake_state = comm_ssl_shake_none; 1192 return 1; 1193 } 1194 #endif /* HAVE_SSL */ 1195 1196 /** ssl read callback on TCP */ 1197 static int 1198 ssl_handle_read(struct comm_point* c) 1199 { 1200 #ifdef HAVE_SSL 1201 int r; 1202 if(c->ssl_shake_state != comm_ssl_shake_none) { 1203 if(!ssl_handshake(c)) 1204 return 0; 1205 if(c->ssl_shake_state != comm_ssl_shake_none) 1206 return 1; 1207 } 1208 if(c->tcp_byte_count < sizeof(uint16_t)) { 1209 /* read length bytes */ 1210 ERR_clear_error(); 1211 if((r=SSL_read(c->ssl, (void*)sldns_buffer_at(c->buffer, 1212 c->tcp_byte_count), (int)(sizeof(uint16_t) - 1213 c->tcp_byte_count))) <= 0) { 1214 int want = SSL_get_error(c->ssl, r); 1215 if(want == SSL_ERROR_ZERO_RETURN) { 1216 if(c->tcp_req_info) 1217 return tcp_req_info_handle_read_close(c->tcp_req_info); 1218 return 0; /* shutdown, closed */ 1219 } else if(want == SSL_ERROR_WANT_READ) { 1220 ub_winsock_tcp_wouldblock(c->ev->ev, UB_EV_READ); 1221 return 1; /* read more later */ 1222 } else if(want == SSL_ERROR_WANT_WRITE) { 1223 c->ssl_shake_state = comm_ssl_shake_hs_write; 1224 comm_point_listen_for_rw(c, 0, 1); 1225 return 1; 1226 } else if(want == SSL_ERROR_SYSCALL) { 1227 #ifdef ECONNRESET 1228 if(errno == ECONNRESET && verbosity < 2) 1229 return 0; /* silence reset by peer */ 1230 #endif 1231 if(errno != 0) 1232 log_err("SSL_read syscall: %s", 1233 strerror(errno)); 1234 return 0; 1235 } 1236 log_crypto_err("could not SSL_read"); 1237 return 0; 1238 } 1239 c->tcp_byte_count += r; 1240 if(c->tcp_byte_count < sizeof(uint16_t)) 1241 return 1; 1242 if(sldns_buffer_read_u16_at(c->buffer, 0) > 1243 sldns_buffer_capacity(c->buffer)) { 1244 verbose(VERB_QUERY, "ssl: dropped larger than buffer"); 1245 return 0; 1246 } 1247 sldns_buffer_set_limit(c->buffer, 1248 sldns_buffer_read_u16_at(c->buffer, 0)); 1249 if(sldns_buffer_limit(c->buffer) < LDNS_HEADER_SIZE) { 1250 verbose(VERB_QUERY, "ssl: dropped bogus too short."); 1251 return 0; 1252 } 1253 sldns_buffer_skip(c->buffer, (ssize_t)(c->tcp_byte_count-sizeof(uint16_t))); 1254 verbose(VERB_ALGO, "Reading ssl tcp query of length %d", 1255 (int)sldns_buffer_limit(c->buffer)); 1256 } 1257 if(sldns_buffer_remaining(c->buffer) > 0) { 1258 ERR_clear_error(); 1259 r = SSL_read(c->ssl, (void*)sldns_buffer_current(c->buffer), 1260 (int)sldns_buffer_remaining(c->buffer)); 1261 if(r <= 0) { 1262 int want = SSL_get_error(c->ssl, r); 1263 if(want == SSL_ERROR_ZERO_RETURN) { 1264 if(c->tcp_req_info) 1265 return tcp_req_info_handle_read_close(c->tcp_req_info); 1266 return 0; /* shutdown, closed */ 1267 } else if(want == SSL_ERROR_WANT_READ) { 1268 ub_winsock_tcp_wouldblock(c->ev->ev, UB_EV_READ); 1269 return 1; /* read more later */ 1270 } else if(want == SSL_ERROR_WANT_WRITE) { 1271 c->ssl_shake_state = comm_ssl_shake_hs_write; 1272 comm_point_listen_for_rw(c, 0, 1); 1273 return 1; 1274 } else if(want == SSL_ERROR_SYSCALL) { 1275 #ifdef ECONNRESET 1276 if(errno == ECONNRESET && verbosity < 2) 1277 return 0; /* silence reset by peer */ 1278 #endif 1279 if(errno != 0) 1280 log_err("SSL_read syscall: %s", 1281 strerror(errno)); 1282 return 0; 1283 } 1284 log_crypto_err("could not SSL_read"); 1285 return 0; 1286 } 1287 sldns_buffer_skip(c->buffer, (ssize_t)r); 1288 } 1289 if(sldns_buffer_remaining(c->buffer) <= 0) { 1290 tcp_callback_reader(c); 1291 } 1292 return 1; 1293 #else 1294 (void)c; 1295 return 0; 1296 #endif /* HAVE_SSL */ 1297 } 1298 1299 /** ssl write callback on TCP */ 1300 static int 1301 ssl_handle_write(struct comm_point* c) 1302 { 1303 #ifdef HAVE_SSL 1304 int r; 1305 if(c->ssl_shake_state != comm_ssl_shake_none) { 1306 if(!ssl_handshake(c)) 1307 return 0; 1308 if(c->ssl_shake_state != comm_ssl_shake_none) 1309 return 1; 1310 } 1311 /* ignore return, if fails we may simply block */ 1312 (void)SSL_set_mode(c->ssl, (long)SSL_MODE_ENABLE_PARTIAL_WRITE); 1313 if(c->tcp_byte_count < sizeof(uint16_t)) { 1314 uint16_t len = htons(sldns_buffer_limit(c->buffer)); 1315 ERR_clear_error(); 1316 if(sizeof(uint16_t)+sldns_buffer_remaining(c->buffer) < 1317 LDNS_RR_BUF_SIZE) { 1318 /* combine the tcp length and the query for write, 1319 * this emulates writev */ 1320 uint8_t buf[LDNS_RR_BUF_SIZE]; 1321 memmove(buf, &len, sizeof(uint16_t)); 1322 memmove(buf+sizeof(uint16_t), 1323 sldns_buffer_current(c->buffer), 1324 sldns_buffer_remaining(c->buffer)); 1325 r = SSL_write(c->ssl, (void*)(buf+c->tcp_byte_count), 1326 (int)(sizeof(uint16_t)+ 1327 sldns_buffer_remaining(c->buffer) 1328 - c->tcp_byte_count)); 1329 } else { 1330 r = SSL_write(c->ssl, 1331 (void*)(((uint8_t*)&len)+c->tcp_byte_count), 1332 (int)(sizeof(uint16_t)-c->tcp_byte_count)); 1333 } 1334 if(r <= 0) { 1335 int want = SSL_get_error(c->ssl, r); 1336 if(want == SSL_ERROR_ZERO_RETURN) { 1337 return 0; /* closed */ 1338 } else if(want == SSL_ERROR_WANT_READ) { 1339 c->ssl_shake_state = comm_ssl_shake_hs_read; 1340 comm_point_listen_for_rw(c, 1, 0); 1341 return 1; /* wait for read condition */ 1342 } else if(want == SSL_ERROR_WANT_WRITE) { 1343 ub_winsock_tcp_wouldblock(c->ev->ev, UB_EV_WRITE); 1344 return 1; /* write more later */ 1345 } else if(want == SSL_ERROR_SYSCALL) { 1346 #ifdef EPIPE 1347 if(errno == EPIPE && verbosity < 2) 1348 return 0; /* silence 'broken pipe' */ 1349 #endif 1350 if(errno != 0) 1351 log_err("SSL_write syscall: %s", 1352 strerror(errno)); 1353 return 0; 1354 } 1355 log_crypto_err("could not SSL_write"); 1356 return 0; 1357 } 1358 c->tcp_byte_count += r; 1359 if(c->tcp_byte_count < sizeof(uint16_t)) 1360 return 1; 1361 sldns_buffer_set_position(c->buffer, c->tcp_byte_count - 1362 sizeof(uint16_t)); 1363 if(sldns_buffer_remaining(c->buffer) == 0) { 1364 tcp_callback_writer(c); 1365 return 1; 1366 } 1367 } 1368 log_assert(sldns_buffer_remaining(c->buffer) > 0); 1369 ERR_clear_error(); 1370 r = SSL_write(c->ssl, (void*)sldns_buffer_current(c->buffer), 1371 (int)sldns_buffer_remaining(c->buffer)); 1372 if(r <= 0) { 1373 int want = SSL_get_error(c->ssl, r); 1374 if(want == SSL_ERROR_ZERO_RETURN) { 1375 return 0; /* closed */ 1376 } else if(want == SSL_ERROR_WANT_READ) { 1377 c->ssl_shake_state = comm_ssl_shake_hs_read; 1378 comm_point_listen_for_rw(c, 1, 0); 1379 return 1; /* wait for read condition */ 1380 } else if(want == SSL_ERROR_WANT_WRITE) { 1381 ub_winsock_tcp_wouldblock(c->ev->ev, UB_EV_WRITE); 1382 return 1; /* write more later */ 1383 } else if(want == SSL_ERROR_SYSCALL) { 1384 #ifdef EPIPE 1385 if(errno == EPIPE && verbosity < 2) 1386 return 0; /* silence 'broken pipe' */ 1387 #endif 1388 if(errno != 0) 1389 log_err("SSL_write syscall: %s", 1390 strerror(errno)); 1391 return 0; 1392 } 1393 log_crypto_err("could not SSL_write"); 1394 return 0; 1395 } 1396 sldns_buffer_skip(c->buffer, (ssize_t)r); 1397 1398 if(sldns_buffer_remaining(c->buffer) == 0) { 1399 tcp_callback_writer(c); 1400 } 1401 return 1; 1402 #else 1403 (void)c; 1404 return 0; 1405 #endif /* HAVE_SSL */ 1406 } 1407 1408 /** handle ssl tcp connection with dns contents */ 1409 static int 1410 ssl_handle_it(struct comm_point* c) 1411 { 1412 if(c->tcp_is_reading) 1413 return ssl_handle_read(c); 1414 return ssl_handle_write(c); 1415 } 1416 1417 /** Handle tcp reading callback. 1418 * @param fd: file descriptor of socket. 1419 * @param c: comm point to read from into buffer. 1420 * @param short_ok: if true, very short packets are OK (for comm_local). 1421 * @return: 0 on error 1422 */ 1423 static int 1424 comm_point_tcp_handle_read(int fd, struct comm_point* c, int short_ok) 1425 { 1426 ssize_t r; 1427 log_assert(c->type == comm_tcp || c->type == comm_local); 1428 if(c->ssl) 1429 return ssl_handle_it(c); 1430 if(!c->tcp_is_reading) 1431 return 0; 1432 1433 log_assert(fd != -1); 1434 if(c->tcp_byte_count < sizeof(uint16_t)) { 1435 /* read length bytes */ 1436 r = recv(fd,(void*)sldns_buffer_at(c->buffer,c->tcp_byte_count), 1437 sizeof(uint16_t)-c->tcp_byte_count, 0); 1438 if(r == 0) { 1439 if(c->tcp_req_info) 1440 return tcp_req_info_handle_read_close(c->tcp_req_info); 1441 return 0; 1442 } else if(r == -1) { 1443 #ifndef USE_WINSOCK 1444 if(errno == EINTR || errno == EAGAIN) 1445 return 1; 1446 #ifdef ECONNRESET 1447 if(errno == ECONNRESET && verbosity < 2) 1448 return 0; /* silence reset by peer */ 1449 #endif 1450 log_err_addr("read (in tcp s)", strerror(errno), 1451 &c->repinfo.addr, c->repinfo.addrlen); 1452 #else /* USE_WINSOCK */ 1453 if(WSAGetLastError() == WSAECONNRESET) 1454 return 0; 1455 if(WSAGetLastError() == WSAEINPROGRESS) 1456 return 1; 1457 if(WSAGetLastError() == WSAEWOULDBLOCK) { 1458 ub_winsock_tcp_wouldblock(c->ev->ev, 1459 UB_EV_READ); 1460 return 1; 1461 } 1462 log_err_addr("read (in tcp s)", 1463 wsa_strerror(WSAGetLastError()), 1464 &c->repinfo.addr, c->repinfo.addrlen); 1465 #endif 1466 return 0; 1467 } 1468 c->tcp_byte_count += r; 1469 if(c->tcp_byte_count != sizeof(uint16_t)) 1470 return 1; 1471 if(sldns_buffer_read_u16_at(c->buffer, 0) > 1472 sldns_buffer_capacity(c->buffer)) { 1473 verbose(VERB_QUERY, "tcp: dropped larger than buffer"); 1474 return 0; 1475 } 1476 sldns_buffer_set_limit(c->buffer, 1477 sldns_buffer_read_u16_at(c->buffer, 0)); 1478 if(!short_ok && 1479 sldns_buffer_limit(c->buffer) < LDNS_HEADER_SIZE) { 1480 verbose(VERB_QUERY, "tcp: dropped bogus too short."); 1481 return 0; 1482 } 1483 verbose(VERB_ALGO, "Reading tcp query of length %d", 1484 (int)sldns_buffer_limit(c->buffer)); 1485 } 1486 1487 log_assert(sldns_buffer_remaining(c->buffer) > 0); 1488 r = recv(fd, (void*)sldns_buffer_current(c->buffer), 1489 sldns_buffer_remaining(c->buffer), 0); 1490 if(r == 0) { 1491 if(c->tcp_req_info) 1492 return tcp_req_info_handle_read_close(c->tcp_req_info); 1493 return 0; 1494 } else if(r == -1) { 1495 #ifndef USE_WINSOCK 1496 if(errno == EINTR || errno == EAGAIN) 1497 return 1; 1498 log_err_addr("read (in tcp r)", strerror(errno), 1499 &c->repinfo.addr, c->repinfo.addrlen); 1500 #else /* USE_WINSOCK */ 1501 if(WSAGetLastError() == WSAECONNRESET) 1502 return 0; 1503 if(WSAGetLastError() == WSAEINPROGRESS) 1504 return 1; 1505 if(WSAGetLastError() == WSAEWOULDBLOCK) { 1506 ub_winsock_tcp_wouldblock(c->ev->ev, UB_EV_READ); 1507 return 1; 1508 } 1509 log_err_addr("read (in tcp r)", 1510 wsa_strerror(WSAGetLastError()), 1511 &c->repinfo.addr, c->repinfo.addrlen); 1512 #endif 1513 return 0; 1514 } 1515 sldns_buffer_skip(c->buffer, r); 1516 if(sldns_buffer_remaining(c->buffer) <= 0) { 1517 tcp_callback_reader(c); 1518 } 1519 return 1; 1520 } 1521 1522 /** 1523 * Handle tcp writing callback. 1524 * @param fd: file descriptor of socket. 1525 * @param c: comm point to write buffer out of. 1526 * @return: 0 on error 1527 */ 1528 static int 1529 comm_point_tcp_handle_write(int fd, struct comm_point* c) 1530 { 1531 ssize_t r; 1532 struct sldns_buffer *buffer; 1533 log_assert(c->type == comm_tcp); 1534 #ifdef USE_DNSCRYPT 1535 buffer = c->dnscrypt_buffer; 1536 #else 1537 buffer = c->buffer; 1538 #endif 1539 if(c->tcp_is_reading && !c->ssl) 1540 return 0; 1541 log_assert(fd != -1); 1542 if(c->tcp_byte_count == 0 && c->tcp_check_nb_connect) { 1543 /* check for pending error from nonblocking connect */ 1544 /* from Stevens, unix network programming, vol1, 3rd ed, p450*/ 1545 int error = 0; 1546 socklen_t len = (socklen_t)sizeof(error); 1547 if(getsockopt(fd, SOL_SOCKET, SO_ERROR, (void*)&error, 1548 &len) < 0){ 1549 #ifndef USE_WINSOCK 1550 error = errno; /* on solaris errno is error */ 1551 #else /* USE_WINSOCK */ 1552 error = WSAGetLastError(); 1553 #endif 1554 } 1555 #ifndef USE_WINSOCK 1556 #if defined(EINPROGRESS) && defined(EWOULDBLOCK) 1557 if(error == EINPROGRESS || error == EWOULDBLOCK) 1558 return 1; /* try again later */ 1559 else 1560 #endif 1561 if(error != 0 && verbosity < 2) 1562 return 0; /* silence lots of chatter in the logs */ 1563 else if(error != 0) { 1564 log_err_addr("tcp connect", strerror(error), 1565 &c->repinfo.addr, c->repinfo.addrlen); 1566 #else /* USE_WINSOCK */ 1567 /* examine error */ 1568 if(error == WSAEINPROGRESS) 1569 return 1; 1570 else if(error == WSAEWOULDBLOCK) { 1571 ub_winsock_tcp_wouldblock(c->ev->ev, UB_EV_WRITE); 1572 return 1; 1573 } else if(error != 0 && verbosity < 2) 1574 return 0; 1575 else if(error != 0) { 1576 log_err_addr("tcp connect", wsa_strerror(error), 1577 &c->repinfo.addr, c->repinfo.addrlen); 1578 #endif /* USE_WINSOCK */ 1579 return 0; 1580 } 1581 } 1582 if(c->ssl) 1583 return ssl_handle_it(c); 1584 1585 #ifdef USE_MSG_FASTOPEN 1586 /* Only try this on first use of a connection that uses tfo, 1587 otherwise fall through to normal write */ 1588 /* Also, TFO support on WINDOWS not implemented at the moment */ 1589 if(c->tcp_do_fastopen == 1) { 1590 /* this form of sendmsg() does both a connect() and send() so need to 1591 look for various flavours of error*/ 1592 uint16_t len = htons(sldns_buffer_limit(buffer)); 1593 struct msghdr msg; 1594 struct iovec iov[2]; 1595 c->tcp_do_fastopen = 0; 1596 memset(&msg, 0, sizeof(msg)); 1597 iov[0].iov_base = (uint8_t*)&len + c->tcp_byte_count; 1598 iov[0].iov_len = sizeof(uint16_t) - c->tcp_byte_count; 1599 iov[1].iov_base = sldns_buffer_begin(buffer); 1600 iov[1].iov_len = sldns_buffer_limit(buffer); 1601 log_assert(iov[0].iov_len > 0); 1602 msg.msg_name = &c->repinfo.addr; 1603 msg.msg_namelen = c->repinfo.addrlen; 1604 msg.msg_iov = iov; 1605 msg.msg_iovlen = 2; 1606 r = sendmsg(fd, &msg, MSG_FASTOPEN); 1607 if (r == -1) { 1608 #if defined(EINPROGRESS) && defined(EWOULDBLOCK) 1609 /* Handshake is underway, maybe because no TFO cookie available. 1610 Come back to write the message*/ 1611 if(errno == EINPROGRESS || errno == EWOULDBLOCK) 1612 return 1; 1613 #endif 1614 if(errno == EINTR || errno == EAGAIN) 1615 return 1; 1616 /* Not handling EISCONN here as shouldn't ever hit that case.*/ 1617 if(errno != EPIPE && errno != 0 && verbosity < 2) 1618 return 0; /* silence lots of chatter in the logs */ 1619 if(errno != EPIPE && errno != 0) { 1620 log_err_addr("tcp sendmsg", strerror(errno), 1621 &c->repinfo.addr, c->repinfo.addrlen); 1622 return 0; 1623 } 1624 /* fallthrough to nonFASTOPEN 1625 * (MSG_FASTOPEN on Linux 3 produces EPIPE) 1626 * we need to perform connect() */ 1627 if(connect(fd, (struct sockaddr *)&c->repinfo.addr, c->repinfo.addrlen) == -1) { 1628 #ifdef EINPROGRESS 1629 if(errno == EINPROGRESS) 1630 return 1; /* wait until connect done*/ 1631 #endif 1632 #ifdef USE_WINSOCK 1633 if(WSAGetLastError() == WSAEINPROGRESS || 1634 WSAGetLastError() == WSAEWOULDBLOCK) 1635 return 1; /* wait until connect done*/ 1636 #endif 1637 if(tcp_connect_errno_needs_log( 1638 (struct sockaddr *)&c->repinfo.addr, c->repinfo.addrlen)) { 1639 log_err_addr("outgoing tcp: connect after EPIPE for fastopen", 1640 strerror(errno), &c->repinfo.addr, c->repinfo.addrlen); 1641 } 1642 return 0; 1643 } 1644 1645 } else { 1646 c->tcp_byte_count += r; 1647 if(c->tcp_byte_count < sizeof(uint16_t)) 1648 return 1; 1649 sldns_buffer_set_position(buffer, c->tcp_byte_count - 1650 sizeof(uint16_t)); 1651 if(sldns_buffer_remaining(buffer) == 0) { 1652 tcp_callback_writer(c); 1653 return 1; 1654 } 1655 } 1656 } 1657 #endif /* USE_MSG_FASTOPEN */ 1658 1659 if(c->tcp_byte_count < sizeof(uint16_t)) { 1660 uint16_t len = htons(sldns_buffer_limit(buffer)); 1661 #ifdef HAVE_WRITEV 1662 struct iovec iov[2]; 1663 iov[0].iov_base = (uint8_t*)&len + c->tcp_byte_count; 1664 iov[0].iov_len = sizeof(uint16_t) - c->tcp_byte_count; 1665 iov[1].iov_base = sldns_buffer_begin(buffer); 1666 iov[1].iov_len = sldns_buffer_limit(buffer); 1667 log_assert(iov[0].iov_len > 0); 1668 r = writev(fd, iov, 2); 1669 #else /* HAVE_WRITEV */ 1670 r = send(fd, (void*)(((uint8_t*)&len)+c->tcp_byte_count), 1671 sizeof(uint16_t)-c->tcp_byte_count, 0); 1672 #endif /* HAVE_WRITEV */ 1673 if(r == -1) { 1674 #ifndef USE_WINSOCK 1675 # ifdef EPIPE 1676 if(errno == EPIPE && verbosity < 2) 1677 return 0; /* silence 'broken pipe' */ 1678 #endif 1679 if(errno == EINTR || errno == EAGAIN) 1680 return 1; 1681 #ifdef ECONNRESET 1682 if(errno == ECONNRESET && verbosity < 2) 1683 return 0; /* silence reset by peer */ 1684 #endif 1685 # ifdef HAVE_WRITEV 1686 log_err_addr("tcp writev", strerror(errno), 1687 &c->repinfo.addr, c->repinfo.addrlen); 1688 # else /* HAVE_WRITEV */ 1689 log_err_addr("tcp send s", strerror(errno), 1690 &c->repinfo.addr, c->repinfo.addrlen); 1691 # endif /* HAVE_WRITEV */ 1692 #else 1693 if(WSAGetLastError() == WSAENOTCONN) 1694 return 1; 1695 if(WSAGetLastError() == WSAEINPROGRESS) 1696 return 1; 1697 if(WSAGetLastError() == WSAEWOULDBLOCK) { 1698 ub_winsock_tcp_wouldblock(c->ev->ev, 1699 UB_EV_WRITE); 1700 return 1; 1701 } 1702 if(WSAGetLastError() == WSAECONNRESET && verbosity < 2) 1703 return 0; /* silence reset by peer */ 1704 log_err_addr("tcp send s", 1705 wsa_strerror(WSAGetLastError()), 1706 &c->repinfo.addr, c->repinfo.addrlen); 1707 #endif 1708 return 0; 1709 } 1710 c->tcp_byte_count += r; 1711 if(c->tcp_byte_count < sizeof(uint16_t)) 1712 return 1; 1713 sldns_buffer_set_position(buffer, c->tcp_byte_count - 1714 sizeof(uint16_t)); 1715 if(sldns_buffer_remaining(buffer) == 0) { 1716 tcp_callback_writer(c); 1717 return 1; 1718 } 1719 } 1720 log_assert(sldns_buffer_remaining(buffer) > 0); 1721 r = send(fd, (void*)sldns_buffer_current(buffer), 1722 sldns_buffer_remaining(buffer), 0); 1723 if(r == -1) { 1724 #ifndef USE_WINSOCK 1725 if(errno == EINTR || errno == EAGAIN) 1726 return 1; 1727 #ifdef ECONNRESET 1728 if(errno == ECONNRESET && verbosity < 2) 1729 return 0; /* silence reset by peer */ 1730 #endif 1731 log_err_addr("tcp send r", strerror(errno), 1732 &c->repinfo.addr, c->repinfo.addrlen); 1733 #else 1734 if(WSAGetLastError() == WSAEINPROGRESS) 1735 return 1; 1736 if(WSAGetLastError() == WSAEWOULDBLOCK) { 1737 ub_winsock_tcp_wouldblock(c->ev->ev, UB_EV_WRITE); 1738 return 1; 1739 } 1740 if(WSAGetLastError() == WSAECONNRESET && verbosity < 2) 1741 return 0; /* silence reset by peer */ 1742 log_err_addr("tcp send r", wsa_strerror(WSAGetLastError()), 1743 &c->repinfo.addr, c->repinfo.addrlen); 1744 #endif 1745 return 0; 1746 } 1747 sldns_buffer_skip(buffer, r); 1748 1749 if(sldns_buffer_remaining(buffer) == 0) { 1750 tcp_callback_writer(c); 1751 } 1752 1753 return 1; 1754 } 1755 1756 /** read again to drain buffers when there could be more to read */ 1757 static void 1758 tcp_req_info_read_again(int fd, struct comm_point* c) 1759 { 1760 while(c->tcp_req_info->read_again) { 1761 int r; 1762 c->tcp_req_info->read_again = 0; 1763 if(c->tcp_is_reading) 1764 r = comm_point_tcp_handle_read(fd, c, 0); 1765 else r = comm_point_tcp_handle_write(fd, c); 1766 if(!r) { 1767 reclaim_tcp_handler(c); 1768 if(!c->tcp_do_close) { 1769 fptr_ok(fptr_whitelist_comm_point( 1770 c->callback)); 1771 (void)(*c->callback)(c, c->cb_arg, 1772 NETEVENT_CLOSED, NULL); 1773 } 1774 return; 1775 } 1776 } 1777 } 1778 1779 void 1780 comm_point_tcp_handle_callback(int fd, short event, void* arg) 1781 { 1782 struct comm_point* c = (struct comm_point*)arg; 1783 log_assert(c->type == comm_tcp); 1784 ub_comm_base_now(c->ev->base); 1785 1786 #ifdef USE_DNSCRYPT 1787 /* Initialize if this is a dnscrypt socket */ 1788 if(c->tcp_parent) { 1789 c->dnscrypt = c->tcp_parent->dnscrypt; 1790 } 1791 if(c->dnscrypt && c->dnscrypt_buffer == c->buffer) { 1792 c->dnscrypt_buffer = sldns_buffer_new(sldns_buffer_capacity(c->buffer)); 1793 if(!c->dnscrypt_buffer) { 1794 log_err("Could not allocate dnscrypt buffer"); 1795 reclaim_tcp_handler(c); 1796 if(!c->tcp_do_close) { 1797 fptr_ok(fptr_whitelist_comm_point( 1798 c->callback)); 1799 (void)(*c->callback)(c, c->cb_arg, 1800 NETEVENT_CLOSED, NULL); 1801 } 1802 return; 1803 } 1804 } 1805 #endif 1806 1807 if(event&UB_EV_TIMEOUT) { 1808 verbose(VERB_QUERY, "tcp took too long, dropped"); 1809 reclaim_tcp_handler(c); 1810 if(!c->tcp_do_close) { 1811 fptr_ok(fptr_whitelist_comm_point(c->callback)); 1812 (void)(*c->callback)(c, c->cb_arg, 1813 NETEVENT_TIMEOUT, NULL); 1814 } 1815 return; 1816 } 1817 if(event&UB_EV_READ) { 1818 int has_tcpq = (c->tcp_req_info != NULL); 1819 if(!comm_point_tcp_handle_read(fd, c, 0)) { 1820 reclaim_tcp_handler(c); 1821 if(!c->tcp_do_close) { 1822 fptr_ok(fptr_whitelist_comm_point( 1823 c->callback)); 1824 (void)(*c->callback)(c, c->cb_arg, 1825 NETEVENT_CLOSED, NULL); 1826 } 1827 } 1828 if(has_tcpq && c->tcp_req_info && c->tcp_req_info->read_again) 1829 tcp_req_info_read_again(fd, c); 1830 return; 1831 } 1832 if(event&UB_EV_WRITE) { 1833 int has_tcpq = (c->tcp_req_info != NULL); 1834 if(!comm_point_tcp_handle_write(fd, c)) { 1835 reclaim_tcp_handler(c); 1836 if(!c->tcp_do_close) { 1837 fptr_ok(fptr_whitelist_comm_point( 1838 c->callback)); 1839 (void)(*c->callback)(c, c->cb_arg, 1840 NETEVENT_CLOSED, NULL); 1841 } 1842 } 1843 if(has_tcpq && c->tcp_req_info && c->tcp_req_info->read_again) 1844 tcp_req_info_read_again(fd, c); 1845 return; 1846 } 1847 log_err("Ignored event %d for tcphdl.", event); 1848 } 1849 1850 /** Make http handler free for next assignment */ 1851 static void 1852 reclaim_http_handler(struct comm_point* c) 1853 { 1854 log_assert(c->type == comm_http); 1855 if(c->ssl) { 1856 #ifdef HAVE_SSL 1857 SSL_shutdown(c->ssl); 1858 SSL_free(c->ssl); 1859 c->ssl = NULL; 1860 #endif 1861 } 1862 comm_point_close(c); 1863 if(c->tcp_parent) { 1864 c->tcp_parent->cur_tcp_count--; 1865 c->tcp_free = c->tcp_parent->tcp_free; 1866 c->tcp_parent->tcp_free = c; 1867 if(!c->tcp_free) { 1868 /* re-enable listening on accept socket */ 1869 comm_point_start_listening(c->tcp_parent, -1, -1); 1870 } 1871 } 1872 } 1873 1874 /** read more data for http (with ssl) */ 1875 static int 1876 ssl_http_read_more(struct comm_point* c) 1877 { 1878 #ifdef HAVE_SSL 1879 int r; 1880 log_assert(sldns_buffer_remaining(c->buffer) > 0); 1881 ERR_clear_error(); 1882 r = SSL_read(c->ssl, (void*)sldns_buffer_current(c->buffer), 1883 (int)sldns_buffer_remaining(c->buffer)); 1884 if(r <= 0) { 1885 int want = SSL_get_error(c->ssl, r); 1886 if(want == SSL_ERROR_ZERO_RETURN) { 1887 return 0; /* shutdown, closed */ 1888 } else if(want == SSL_ERROR_WANT_READ) { 1889 return 1; /* read more later */ 1890 } else if(want == SSL_ERROR_WANT_WRITE) { 1891 c->ssl_shake_state = comm_ssl_shake_hs_write; 1892 comm_point_listen_for_rw(c, 0, 1); 1893 return 1; 1894 } else if(want == SSL_ERROR_SYSCALL) { 1895 #ifdef ECONNRESET 1896 if(errno == ECONNRESET && verbosity < 2) 1897 return 0; /* silence reset by peer */ 1898 #endif 1899 if(errno != 0) 1900 log_err("SSL_read syscall: %s", 1901 strerror(errno)); 1902 return 0; 1903 } 1904 log_crypto_err("could not SSL_read"); 1905 return 0; 1906 } 1907 sldns_buffer_skip(c->buffer, (ssize_t)r); 1908 return 1; 1909 #else 1910 (void)c; 1911 return 0; 1912 #endif /* HAVE_SSL */ 1913 } 1914 1915 /** read more data for http */ 1916 static int 1917 http_read_more(int fd, struct comm_point* c) 1918 { 1919 ssize_t r; 1920 log_assert(sldns_buffer_remaining(c->buffer) > 0); 1921 r = recv(fd, (void*)sldns_buffer_current(c->buffer), 1922 sldns_buffer_remaining(c->buffer), 0); 1923 if(r == 0) { 1924 return 0; 1925 } else if(r == -1) { 1926 #ifndef USE_WINSOCK 1927 if(errno == EINTR || errno == EAGAIN) 1928 return 1; 1929 log_err_addr("read (in http r)", strerror(errno), 1930 &c->repinfo.addr, c->repinfo.addrlen); 1931 #else /* USE_WINSOCK */ 1932 if(WSAGetLastError() == WSAECONNRESET) 1933 return 0; 1934 if(WSAGetLastError() == WSAEINPROGRESS) 1935 return 1; 1936 if(WSAGetLastError() == WSAEWOULDBLOCK) { 1937 ub_winsock_tcp_wouldblock(c->ev->ev, UB_EV_READ); 1938 return 1; 1939 } 1940 log_err_addr("read (in http r)", 1941 wsa_strerror(WSAGetLastError()), 1942 &c->repinfo.addr, c->repinfo.addrlen); 1943 #endif 1944 return 0; 1945 } 1946 sldns_buffer_skip(c->buffer, r); 1947 return 1; 1948 } 1949 1950 /** return true if http header has been read (one line complete) */ 1951 static int 1952 http_header_done(sldns_buffer* buf) 1953 { 1954 size_t i; 1955 for(i=sldns_buffer_position(buf); i<sldns_buffer_limit(buf); i++) { 1956 /* there was a \r before the \n, but we ignore that */ 1957 if((char)sldns_buffer_read_u8_at(buf, i) == '\n') 1958 return 1; 1959 } 1960 return 0; 1961 } 1962 1963 /** return character string into buffer for header line, moves buffer 1964 * past that line and puts zero terminator into linefeed-newline */ 1965 static char* 1966 http_header_line(sldns_buffer* buf) 1967 { 1968 char* result = (char*)sldns_buffer_current(buf); 1969 size_t i; 1970 for(i=sldns_buffer_position(buf); i<sldns_buffer_limit(buf); i++) { 1971 /* terminate the string on the \r */ 1972 if((char)sldns_buffer_read_u8_at(buf, i) == '\r') 1973 sldns_buffer_write_u8_at(buf, i, 0); 1974 /* terminate on the \n and skip past the it and done */ 1975 if((char)sldns_buffer_read_u8_at(buf, i) == '\n') { 1976 sldns_buffer_write_u8_at(buf, i, 0); 1977 sldns_buffer_set_position(buf, i+1); 1978 return result; 1979 } 1980 } 1981 return NULL; 1982 } 1983 1984 /** move unread buffer to start and clear rest for putting the rest into it */ 1985 static void 1986 http_moveover_buffer(sldns_buffer* buf) 1987 { 1988 size_t pos = sldns_buffer_position(buf); 1989 size_t len = sldns_buffer_remaining(buf); 1990 sldns_buffer_clear(buf); 1991 memmove(sldns_buffer_begin(buf), sldns_buffer_at(buf, pos), len); 1992 sldns_buffer_set_position(buf, len); 1993 } 1994 1995 /** a http header is complete, process it */ 1996 static int 1997 http_process_initial_header(struct comm_point* c) 1998 { 1999 char* line = http_header_line(c->buffer); 2000 if(!line) return 1; 2001 verbose(VERB_ALGO, "http header: %s", line); 2002 if(strncasecmp(line, "HTTP/1.1 ", 9) == 0) { 2003 /* check returncode */ 2004 if(line[9] != '2') { 2005 verbose(VERB_ALGO, "http bad status %s", line+9); 2006 return 0; 2007 } 2008 } else if(strncasecmp(line, "Content-Length: ", 16) == 0) { 2009 if(!c->http_is_chunked) 2010 c->tcp_byte_count = (size_t)atoi(line+16); 2011 } else if(strncasecmp(line, "Transfer-Encoding: chunked", 19+7) == 0) { 2012 c->tcp_byte_count = 0; 2013 c->http_is_chunked = 1; 2014 } else if(line[0] == 0) { 2015 /* end of initial headers */ 2016 c->http_in_headers = 0; 2017 if(c->http_is_chunked) 2018 c->http_in_chunk_headers = 1; 2019 /* remove header text from front of buffer 2020 * the buffer is going to be used to return the data segment 2021 * itself and we don't want the header to get returned 2022 * prepended with it */ 2023 http_moveover_buffer(c->buffer); 2024 sldns_buffer_flip(c->buffer); 2025 return 1; 2026 } 2027 /* ignore other headers */ 2028 return 1; 2029 } 2030 2031 /** a chunk header is complete, process it, return 0=fail, 1=continue next 2032 * header line, 2=done with chunked transfer*/ 2033 static int 2034 http_process_chunk_header(struct comm_point* c) 2035 { 2036 char* line = http_header_line(c->buffer); 2037 if(!line) return 1; 2038 if(c->http_in_chunk_headers == 3) { 2039 verbose(VERB_ALGO, "http chunk trailer: %s", line); 2040 /* are we done ? */ 2041 if(line[0] == 0 && c->tcp_byte_count == 0) { 2042 /* callback of http reader when NETEVENT_DONE, 2043 * end of data, with no data in buffer */ 2044 sldns_buffer_set_position(c->buffer, 0); 2045 sldns_buffer_set_limit(c->buffer, 0); 2046 fptr_ok(fptr_whitelist_comm_point(c->callback)); 2047 (void)(*c->callback)(c, c->cb_arg, NETEVENT_DONE, NULL); 2048 /* return that we are done */ 2049 return 2; 2050 } 2051 if(line[0] == 0) { 2052 /* continue with header of the next chunk */ 2053 c->http_in_chunk_headers = 1; 2054 /* remove header text from front of buffer */ 2055 http_moveover_buffer(c->buffer); 2056 sldns_buffer_flip(c->buffer); 2057 return 1; 2058 } 2059 /* ignore further trail headers */ 2060 return 1; 2061 } 2062 verbose(VERB_ALGO, "http chunk header: %s", line); 2063 if(c->http_in_chunk_headers == 1) { 2064 /* read chunked start line */ 2065 char* end = NULL; 2066 c->tcp_byte_count = (size_t)strtol(line, &end, 16); 2067 if(end == line) 2068 return 0; 2069 c->http_in_chunk_headers = 0; 2070 /* remove header text from front of buffer */ 2071 http_moveover_buffer(c->buffer); 2072 sldns_buffer_flip(c->buffer); 2073 if(c->tcp_byte_count == 0) { 2074 /* done with chunks, process chunk_trailer lines */ 2075 c->http_in_chunk_headers = 3; 2076 } 2077 return 1; 2078 } 2079 /* ignore other headers */ 2080 return 1; 2081 } 2082 2083 /** handle nonchunked data segment */ 2084 static int 2085 http_nonchunk_segment(struct comm_point* c) 2086 { 2087 /* c->buffer at position..limit has new data we read in. 2088 * the buffer itself is full of nonchunked data. 2089 * we are looking to read tcp_byte_count more data 2090 * and then the transfer is done. */ 2091 size_t remainbufferlen; 2092 size_t got_now = sldns_buffer_limit(c->buffer) - c->http_stored; 2093 if(c->tcp_byte_count <= got_now) { 2094 /* done, this is the last data fragment */ 2095 c->http_stored = 0; 2096 sldns_buffer_set_position(c->buffer, 0); 2097 fptr_ok(fptr_whitelist_comm_point(c->callback)); 2098 (void)(*c->callback)(c, c->cb_arg, NETEVENT_DONE, NULL); 2099 return 1; 2100 } 2101 c->tcp_byte_count -= got_now; 2102 /* if we have the buffer space, 2103 * read more data collected into the buffer */ 2104 remainbufferlen = sldns_buffer_capacity(c->buffer) - 2105 sldns_buffer_limit(c->buffer); 2106 if(remainbufferlen >= c->tcp_byte_count || 2107 remainbufferlen >= 2048) { 2108 size_t total = sldns_buffer_limit(c->buffer); 2109 sldns_buffer_clear(c->buffer); 2110 sldns_buffer_set_position(c->buffer, total); 2111 c->http_stored = total; 2112 /* return and wait to read more */ 2113 return 1; 2114 } 2115 /* call callback with this data amount, then 2116 * wait for more */ 2117 c->http_stored = 0; 2118 sldns_buffer_set_position(c->buffer, 0); 2119 fptr_ok(fptr_whitelist_comm_point(c->callback)); 2120 (void)(*c->callback)(c, c->cb_arg, NETEVENT_NOERROR, NULL); 2121 /* c->callback has to buffer_clear(c->buffer). */ 2122 /* return and wait to read more */ 2123 return 1; 2124 } 2125 2126 /** handle nonchunked data segment, return 0=fail, 1=wait, 2=process more */ 2127 static int 2128 http_chunked_segment(struct comm_point* c) 2129 { 2130 /* the c->buffer has from position..limit new data we read. */ 2131 /* the current chunk has length tcp_byte_count. 2132 * once we read that read more chunk headers. 2133 */ 2134 size_t remainbufferlen; 2135 size_t got_now = sldns_buffer_limit(c->buffer) - c->http_stored; 2136 if(c->tcp_byte_count <= got_now) { 2137 /* the chunk has completed (with perhaps some extra data 2138 * from next chunk header and next chunk) */ 2139 /* save too much info into temp buffer */ 2140 size_t fraglen; 2141 struct comm_reply repinfo; 2142 c->http_stored = 0; 2143 sldns_buffer_skip(c->buffer, (ssize_t)c->tcp_byte_count); 2144 sldns_buffer_clear(c->http_temp); 2145 sldns_buffer_write(c->http_temp, 2146 sldns_buffer_current(c->buffer), 2147 sldns_buffer_remaining(c->buffer)); 2148 sldns_buffer_flip(c->http_temp); 2149 2150 /* callback with this fragment */ 2151 fraglen = sldns_buffer_position(c->buffer); 2152 sldns_buffer_set_position(c->buffer, 0); 2153 sldns_buffer_set_limit(c->buffer, fraglen); 2154 repinfo = c->repinfo; 2155 fptr_ok(fptr_whitelist_comm_point(c->callback)); 2156 (void)(*c->callback)(c, c->cb_arg, NETEVENT_NOERROR, &repinfo); 2157 /* c->callback has to buffer_clear(). */ 2158 2159 /* is commpoint deleted? */ 2160 if(!repinfo.c) { 2161 return 1; 2162 } 2163 /* copy waiting info */ 2164 sldns_buffer_clear(c->buffer); 2165 sldns_buffer_write(c->buffer, 2166 sldns_buffer_begin(c->http_temp), 2167 sldns_buffer_remaining(c->http_temp)); 2168 sldns_buffer_flip(c->buffer); 2169 /* process end of chunk trailer header lines, until 2170 * an empty line */ 2171 c->http_in_chunk_headers = 3; 2172 /* process more data in buffer (if any) */ 2173 return 2; 2174 } 2175 c->tcp_byte_count -= got_now; 2176 2177 /* if we have the buffer space, 2178 * read more data collected into the buffer */ 2179 remainbufferlen = sldns_buffer_capacity(c->buffer) - 2180 sldns_buffer_limit(c->buffer); 2181 if(remainbufferlen >= c->tcp_byte_count || 2182 remainbufferlen >= 2048) { 2183 size_t total = sldns_buffer_limit(c->buffer); 2184 sldns_buffer_clear(c->buffer); 2185 sldns_buffer_set_position(c->buffer, total); 2186 c->http_stored = total; 2187 /* return and wait to read more */ 2188 return 1; 2189 } 2190 2191 /* callback of http reader for a new part of the data */ 2192 c->http_stored = 0; 2193 sldns_buffer_set_position(c->buffer, 0); 2194 fptr_ok(fptr_whitelist_comm_point(c->callback)); 2195 (void)(*c->callback)(c, c->cb_arg, NETEVENT_NOERROR, NULL); 2196 /* c->callback has to buffer_clear(c->buffer). */ 2197 /* return and wait to read more */ 2198 return 1; 2199 } 2200 2201 /** 2202 * Handle http reading callback. 2203 * @param fd: file descriptor of socket. 2204 * @param c: comm point to read from into buffer. 2205 * @return: 0 on error 2206 */ 2207 static int 2208 comm_point_http_handle_read(int fd, struct comm_point* c) 2209 { 2210 log_assert(c->type == comm_http); 2211 log_assert(fd != -1); 2212 2213 /* if we are in ssl handshake, handle SSL handshake */ 2214 #ifdef HAVE_SSL 2215 if(c->ssl && c->ssl_shake_state != comm_ssl_shake_none) { 2216 if(!ssl_handshake(c)) 2217 return 0; 2218 if(c->ssl_shake_state != comm_ssl_shake_none) 2219 return 1; 2220 } 2221 #endif /* HAVE_SSL */ 2222 2223 if(!c->tcp_is_reading) 2224 return 1; 2225 /* read more data */ 2226 if(c->ssl) { 2227 if(!ssl_http_read_more(c)) 2228 return 0; 2229 } else { 2230 if(!http_read_more(fd, c)) 2231 return 0; 2232 } 2233 2234 sldns_buffer_flip(c->buffer); 2235 while(sldns_buffer_remaining(c->buffer) > 0) { 2236 /* if we are reading headers, read more headers */ 2237 if(c->http_in_headers || c->http_in_chunk_headers) { 2238 /* if header is done, process the header */ 2239 if(!http_header_done(c->buffer)) { 2240 /* copy remaining data to front of buffer 2241 * and set rest for writing into it */ 2242 http_moveover_buffer(c->buffer); 2243 /* return and wait to read more */ 2244 return 1; 2245 } 2246 if(!c->http_in_chunk_headers) { 2247 /* process initial headers */ 2248 if(!http_process_initial_header(c)) 2249 return 0; 2250 } else { 2251 /* process chunk headers */ 2252 int r = http_process_chunk_header(c); 2253 if(r == 0) return 0; 2254 if(r == 2) return 1; /* done */ 2255 /* r == 1, continue */ 2256 } 2257 /* see if we have more to process */ 2258 continue; 2259 } 2260 2261 if(!c->http_is_chunked) { 2262 /* if we are reading nonchunks, process that*/ 2263 return http_nonchunk_segment(c); 2264 } else { 2265 /* if we are reading chunks, read the chunk */ 2266 int r = http_chunked_segment(c); 2267 if(r == 0) return 0; 2268 if(r == 1) return 1; 2269 continue; 2270 } 2271 } 2272 /* broke out of the loop; could not process header instead need 2273 * to read more */ 2274 /* moveover any remaining data and read more data */ 2275 http_moveover_buffer(c->buffer); 2276 /* return and wait to read more */ 2277 return 1; 2278 } 2279 2280 /** check pending connect for http */ 2281 static int 2282 http_check_connect(int fd, struct comm_point* c) 2283 { 2284 /* check for pending error from nonblocking connect */ 2285 /* from Stevens, unix network programming, vol1, 3rd ed, p450*/ 2286 int error = 0; 2287 socklen_t len = (socklen_t)sizeof(error); 2288 if(getsockopt(fd, SOL_SOCKET, SO_ERROR, (void*)&error, 2289 &len) < 0){ 2290 #ifndef USE_WINSOCK 2291 error = errno; /* on solaris errno is error */ 2292 #else /* USE_WINSOCK */ 2293 error = WSAGetLastError(); 2294 #endif 2295 } 2296 #ifndef USE_WINSOCK 2297 #if defined(EINPROGRESS) && defined(EWOULDBLOCK) 2298 if(error == EINPROGRESS || error == EWOULDBLOCK) 2299 return 1; /* try again later */ 2300 else 2301 #endif 2302 if(error != 0 && verbosity < 2) 2303 return 0; /* silence lots of chatter in the logs */ 2304 else if(error != 0) { 2305 log_err_addr("http connect", strerror(error), 2306 &c->repinfo.addr, c->repinfo.addrlen); 2307 #else /* USE_WINSOCK */ 2308 /* examine error */ 2309 if(error == WSAEINPROGRESS) 2310 return 1; 2311 else if(error == WSAEWOULDBLOCK) { 2312 ub_winsock_tcp_wouldblock(c->ev->ev, UB_EV_WRITE); 2313 return 1; 2314 } else if(error != 0 && verbosity < 2) 2315 return 0; 2316 else if(error != 0) { 2317 log_err_addr("http connect", wsa_strerror(error), 2318 &c->repinfo.addr, c->repinfo.addrlen); 2319 #endif /* USE_WINSOCK */ 2320 return 0; 2321 } 2322 /* keep on processing this socket */ 2323 return 2; 2324 } 2325 2326 /** write more data for http (with ssl) */ 2327 static int 2328 ssl_http_write_more(struct comm_point* c) 2329 { 2330 #ifdef HAVE_SSL 2331 int r; 2332 log_assert(sldns_buffer_remaining(c->buffer) > 0); 2333 ERR_clear_error(); 2334 r = SSL_write(c->ssl, (void*)sldns_buffer_current(c->buffer), 2335 (int)sldns_buffer_remaining(c->buffer)); 2336 if(r <= 0) { 2337 int want = SSL_get_error(c->ssl, r); 2338 if(want == SSL_ERROR_ZERO_RETURN) { 2339 return 0; /* closed */ 2340 } else if(want == SSL_ERROR_WANT_READ) { 2341 c->ssl_shake_state = comm_ssl_shake_hs_read; 2342 comm_point_listen_for_rw(c, 1, 0); 2343 return 1; /* wait for read condition */ 2344 } else if(want == SSL_ERROR_WANT_WRITE) { 2345 return 1; /* write more later */ 2346 } else if(want == SSL_ERROR_SYSCALL) { 2347 #ifdef EPIPE 2348 if(errno == EPIPE && verbosity < 2) 2349 return 0; /* silence 'broken pipe' */ 2350 #endif 2351 if(errno != 0) 2352 log_err("SSL_write syscall: %s", 2353 strerror(errno)); 2354 return 0; 2355 } 2356 log_crypto_err("could not SSL_write"); 2357 return 0; 2358 } 2359 sldns_buffer_skip(c->buffer, (ssize_t)r); 2360 return 1; 2361 #else 2362 (void)c; 2363 return 0; 2364 #endif /* HAVE_SSL */ 2365 } 2366 2367 /** write more data for http */ 2368 static int 2369 http_write_more(int fd, struct comm_point* c) 2370 { 2371 ssize_t r; 2372 log_assert(sldns_buffer_remaining(c->buffer) > 0); 2373 r = send(fd, (void*)sldns_buffer_current(c->buffer), 2374 sldns_buffer_remaining(c->buffer), 0); 2375 if(r == -1) { 2376 #ifndef USE_WINSOCK 2377 if(errno == EINTR || errno == EAGAIN) 2378 return 1; 2379 log_err_addr("http send r", strerror(errno), 2380 &c->repinfo.addr, c->repinfo.addrlen); 2381 #else 2382 if(WSAGetLastError() == WSAEINPROGRESS) 2383 return 1; 2384 if(WSAGetLastError() == WSAEWOULDBLOCK) { 2385 ub_winsock_tcp_wouldblock(c->ev->ev, UB_EV_WRITE); 2386 return 1; 2387 } 2388 log_err_addr("http send r", wsa_strerror(WSAGetLastError()), 2389 &c->repinfo.addr, c->repinfo.addrlen); 2390 #endif 2391 return 0; 2392 } 2393 sldns_buffer_skip(c->buffer, r); 2394 return 1; 2395 } 2396 2397 /** 2398 * Handle http writing callback. 2399 * @param fd: file descriptor of socket. 2400 * @param c: comm point to write buffer out of. 2401 * @return: 0 on error 2402 */ 2403 static int 2404 comm_point_http_handle_write(int fd, struct comm_point* c) 2405 { 2406 log_assert(c->type == comm_http); 2407 log_assert(fd != -1); 2408 2409 /* check pending connect errors, if that fails, we wait for more, 2410 * or we can continue to write contents */ 2411 if(c->tcp_check_nb_connect) { 2412 int r = http_check_connect(fd, c); 2413 if(r == 0) return 0; 2414 if(r == 1) return 1; 2415 c->tcp_check_nb_connect = 0; 2416 } 2417 /* if we are in ssl handshake, handle SSL handshake */ 2418 #ifdef HAVE_SSL 2419 if(c->ssl && c->ssl_shake_state != comm_ssl_shake_none) { 2420 if(!ssl_handshake(c)) 2421 return 0; 2422 if(c->ssl_shake_state != comm_ssl_shake_none) 2423 return 1; 2424 } 2425 #endif /* HAVE_SSL */ 2426 if(c->tcp_is_reading) 2427 return 1; 2428 /* if we are writing, write more */ 2429 if(c->ssl) { 2430 if(!ssl_http_write_more(c)) 2431 return 0; 2432 } else { 2433 if(!http_write_more(fd, c)) 2434 return 0; 2435 } 2436 2437 /* we write a single buffer contents, that can contain 2438 * the http request, and then flip to read the results */ 2439 /* see if write is done */ 2440 if(sldns_buffer_remaining(c->buffer) == 0) { 2441 sldns_buffer_clear(c->buffer); 2442 if(c->tcp_do_toggle_rw) 2443 c->tcp_is_reading = 1; 2444 c->tcp_byte_count = 0; 2445 /* switch from listening(write) to listening(read) */ 2446 comm_point_stop_listening(c); 2447 comm_point_start_listening(c, -1, -1); 2448 } 2449 return 1; 2450 } 2451 2452 void 2453 comm_point_http_handle_callback(int fd, short event, void* arg) 2454 { 2455 struct comm_point* c = (struct comm_point*)arg; 2456 log_assert(c->type == comm_http); 2457 ub_comm_base_now(c->ev->base); 2458 2459 if(event&UB_EV_TIMEOUT) { 2460 verbose(VERB_QUERY, "http took too long, dropped"); 2461 reclaim_http_handler(c); 2462 if(!c->tcp_do_close) { 2463 fptr_ok(fptr_whitelist_comm_point(c->callback)); 2464 (void)(*c->callback)(c, c->cb_arg, 2465 NETEVENT_TIMEOUT, NULL); 2466 } 2467 return; 2468 } 2469 if(event&UB_EV_READ) { 2470 if(!comm_point_http_handle_read(fd, c)) { 2471 reclaim_http_handler(c); 2472 if(!c->tcp_do_close) { 2473 fptr_ok(fptr_whitelist_comm_point( 2474 c->callback)); 2475 (void)(*c->callback)(c, c->cb_arg, 2476 NETEVENT_CLOSED, NULL); 2477 } 2478 } 2479 return; 2480 } 2481 if(event&UB_EV_WRITE) { 2482 if(!comm_point_http_handle_write(fd, c)) { 2483 reclaim_http_handler(c); 2484 if(!c->tcp_do_close) { 2485 fptr_ok(fptr_whitelist_comm_point( 2486 c->callback)); 2487 (void)(*c->callback)(c, c->cb_arg, 2488 NETEVENT_CLOSED, NULL); 2489 } 2490 } 2491 return; 2492 } 2493 log_err("Ignored event %d for httphdl.", event); 2494 } 2495 2496 void comm_point_local_handle_callback(int fd, short event, void* arg) 2497 { 2498 struct comm_point* c = (struct comm_point*)arg; 2499 log_assert(c->type == comm_local); 2500 ub_comm_base_now(c->ev->base); 2501 2502 if(event&UB_EV_READ) { 2503 if(!comm_point_tcp_handle_read(fd, c, 1)) { 2504 fptr_ok(fptr_whitelist_comm_point(c->callback)); 2505 (void)(*c->callback)(c, c->cb_arg, NETEVENT_CLOSED, 2506 NULL); 2507 } 2508 return; 2509 } 2510 log_err("Ignored event %d for localhdl.", event); 2511 } 2512 2513 void comm_point_raw_handle_callback(int ATTR_UNUSED(fd), 2514 short event, void* arg) 2515 { 2516 struct comm_point* c = (struct comm_point*)arg; 2517 int err = NETEVENT_NOERROR; 2518 log_assert(c->type == comm_raw); 2519 ub_comm_base_now(c->ev->base); 2520 2521 if(event&UB_EV_TIMEOUT) 2522 err = NETEVENT_TIMEOUT; 2523 fptr_ok(fptr_whitelist_comm_point_raw(c->callback)); 2524 (void)(*c->callback)(c, c->cb_arg, err, NULL); 2525 } 2526 2527 struct comm_point* 2528 comm_point_create_udp(struct comm_base *base, int fd, sldns_buffer* buffer, 2529 comm_point_callback_type* callback, void* callback_arg) 2530 { 2531 struct comm_point* c = (struct comm_point*)calloc(1, 2532 sizeof(struct comm_point)); 2533 short evbits; 2534 if(!c) 2535 return NULL; 2536 c->ev = (struct internal_event*)calloc(1, 2537 sizeof(struct internal_event)); 2538 if(!c->ev) { 2539 free(c); 2540 return NULL; 2541 } 2542 c->ev->base = base; 2543 c->fd = fd; 2544 c->buffer = buffer; 2545 c->timeout = NULL; 2546 c->tcp_is_reading = 0; 2547 c->tcp_byte_count = 0; 2548 c->tcp_parent = NULL; 2549 c->max_tcp_count = 0; 2550 c->cur_tcp_count = 0; 2551 c->tcp_handlers = NULL; 2552 c->tcp_free = NULL; 2553 c->type = comm_udp; 2554 c->tcp_do_close = 0; 2555 c->do_not_close = 0; 2556 c->tcp_do_toggle_rw = 0; 2557 c->tcp_check_nb_connect = 0; 2558 #ifdef USE_MSG_FASTOPEN 2559 c->tcp_do_fastopen = 0; 2560 #endif 2561 #ifdef USE_DNSCRYPT 2562 c->dnscrypt = 0; 2563 c->dnscrypt_buffer = buffer; 2564 #endif 2565 c->inuse = 0; 2566 c->callback = callback; 2567 c->cb_arg = callback_arg; 2568 evbits = UB_EV_READ | UB_EV_PERSIST; 2569 /* ub_event stuff */ 2570 c->ev->ev = ub_event_new(base->eb->base, c->fd, evbits, 2571 comm_point_udp_callback, c); 2572 if(c->ev->ev == NULL) { 2573 log_err("could not baseset udp event"); 2574 comm_point_delete(c); 2575 return NULL; 2576 } 2577 if(fd!=-1 && ub_event_add(c->ev->ev, c->timeout) != 0 ) { 2578 log_err("could not add udp event"); 2579 comm_point_delete(c); 2580 return NULL; 2581 } 2582 return c; 2583 } 2584 2585 struct comm_point* 2586 comm_point_create_udp_ancil(struct comm_base *base, int fd, 2587 sldns_buffer* buffer, 2588 comm_point_callback_type* callback, void* callback_arg) 2589 { 2590 struct comm_point* c = (struct comm_point*)calloc(1, 2591 sizeof(struct comm_point)); 2592 short evbits; 2593 if(!c) 2594 return NULL; 2595 c->ev = (struct internal_event*)calloc(1, 2596 sizeof(struct internal_event)); 2597 if(!c->ev) { 2598 free(c); 2599 return NULL; 2600 } 2601 c->ev->base = base; 2602 c->fd = fd; 2603 c->buffer = buffer; 2604 c->timeout = NULL; 2605 c->tcp_is_reading = 0; 2606 c->tcp_byte_count = 0; 2607 c->tcp_parent = NULL; 2608 c->max_tcp_count = 0; 2609 c->cur_tcp_count = 0; 2610 c->tcp_handlers = NULL; 2611 c->tcp_free = NULL; 2612 c->type = comm_udp; 2613 c->tcp_do_close = 0; 2614 c->do_not_close = 0; 2615 #ifdef USE_DNSCRYPT 2616 c->dnscrypt = 0; 2617 c->dnscrypt_buffer = buffer; 2618 #endif 2619 c->inuse = 0; 2620 c->tcp_do_toggle_rw = 0; 2621 c->tcp_check_nb_connect = 0; 2622 #ifdef USE_MSG_FASTOPEN 2623 c->tcp_do_fastopen = 0; 2624 #endif 2625 c->callback = callback; 2626 c->cb_arg = callback_arg; 2627 evbits = UB_EV_READ | UB_EV_PERSIST; 2628 /* ub_event stuff */ 2629 c->ev->ev = ub_event_new(base->eb->base, c->fd, evbits, 2630 comm_point_udp_ancil_callback, c); 2631 if(c->ev->ev == NULL) { 2632 log_err("could not baseset udp event"); 2633 comm_point_delete(c); 2634 return NULL; 2635 } 2636 if(fd!=-1 && ub_event_add(c->ev->ev, c->timeout) != 0 ) { 2637 log_err("could not add udp event"); 2638 comm_point_delete(c); 2639 return NULL; 2640 } 2641 return c; 2642 } 2643 2644 static struct comm_point* 2645 comm_point_create_tcp_handler(struct comm_base *base, 2646 struct comm_point* parent, size_t bufsize, 2647 struct sldns_buffer* spoolbuf, comm_point_callback_type* callback, 2648 void* callback_arg) 2649 { 2650 struct comm_point* c = (struct comm_point*)calloc(1, 2651 sizeof(struct comm_point)); 2652 short evbits; 2653 if(!c) 2654 return NULL; 2655 c->ev = (struct internal_event*)calloc(1, 2656 sizeof(struct internal_event)); 2657 if(!c->ev) { 2658 free(c); 2659 return NULL; 2660 } 2661 c->ev->base = base; 2662 c->fd = -1; 2663 c->buffer = sldns_buffer_new(bufsize); 2664 if(!c->buffer) { 2665 free(c->ev); 2666 free(c); 2667 return NULL; 2668 } 2669 c->timeout = (struct timeval*)malloc(sizeof(struct timeval)); 2670 if(!c->timeout) { 2671 sldns_buffer_free(c->buffer); 2672 free(c->ev); 2673 free(c); 2674 return NULL; 2675 } 2676 c->tcp_is_reading = 0; 2677 c->tcp_byte_count = 0; 2678 c->tcp_parent = parent; 2679 c->tcp_timeout_msec = parent->tcp_timeout_msec; 2680 c->tcp_conn_limit = parent->tcp_conn_limit; 2681 c->tcl_addr = NULL; 2682 c->tcp_keepalive = 0; 2683 c->max_tcp_count = 0; 2684 c->cur_tcp_count = 0; 2685 c->tcp_handlers = NULL; 2686 c->tcp_free = NULL; 2687 c->type = comm_tcp; 2688 c->tcp_do_close = 0; 2689 c->do_not_close = 0; 2690 c->tcp_do_toggle_rw = 1; 2691 c->tcp_check_nb_connect = 0; 2692 #ifdef USE_MSG_FASTOPEN 2693 c->tcp_do_fastopen = 0; 2694 #endif 2695 #ifdef USE_DNSCRYPT 2696 c->dnscrypt = 0; 2697 /* We don't know just yet if this is a dnscrypt channel. Allocation 2698 * will be done when handling the callback. */ 2699 c->dnscrypt_buffer = c->buffer; 2700 #endif 2701 c->repinfo.c = c; 2702 c->callback = callback; 2703 c->cb_arg = callback_arg; 2704 if(spoolbuf) { 2705 c->tcp_req_info = tcp_req_info_create(spoolbuf); 2706 if(!c->tcp_req_info) { 2707 log_err("could not create tcp commpoint"); 2708 sldns_buffer_free(c->buffer); 2709 free(c->timeout); 2710 free(c->ev); 2711 free(c); 2712 return NULL; 2713 } 2714 c->tcp_req_info->cp = c; 2715 c->tcp_do_close = 1; 2716 c->tcp_do_toggle_rw = 0; 2717 } 2718 /* add to parent free list */ 2719 c->tcp_free = parent->tcp_free; 2720 parent->tcp_free = c; 2721 /* ub_event stuff */ 2722 evbits = UB_EV_PERSIST | UB_EV_READ | UB_EV_TIMEOUT; 2723 c->ev->ev = ub_event_new(base->eb->base, c->fd, evbits, 2724 comm_point_tcp_handle_callback, c); 2725 if(c->ev->ev == NULL) 2726 { 2727 log_err("could not basetset tcphdl event"); 2728 parent->tcp_free = c->tcp_free; 2729 tcp_req_info_delete(c->tcp_req_info); 2730 sldns_buffer_free(c->buffer); 2731 free(c->timeout); 2732 free(c->ev); 2733 free(c); 2734 return NULL; 2735 } 2736 return c; 2737 } 2738 2739 struct comm_point* 2740 comm_point_create_tcp(struct comm_base *base, int fd, int num, 2741 int idle_timeout, struct tcl_list* tcp_conn_limit, size_t bufsize, 2742 struct sldns_buffer* spoolbuf, comm_point_callback_type* callback, 2743 void* callback_arg) 2744 { 2745 struct comm_point* c = (struct comm_point*)calloc(1, 2746 sizeof(struct comm_point)); 2747 short evbits; 2748 int i; 2749 /* first allocate the TCP accept listener */ 2750 if(!c) 2751 return NULL; 2752 c->ev = (struct internal_event*)calloc(1, 2753 sizeof(struct internal_event)); 2754 if(!c->ev) { 2755 free(c); 2756 return NULL; 2757 } 2758 c->ev->base = base; 2759 c->fd = fd; 2760 c->buffer = NULL; 2761 c->timeout = NULL; 2762 c->tcp_is_reading = 0; 2763 c->tcp_byte_count = 0; 2764 c->tcp_timeout_msec = idle_timeout; 2765 c->tcp_conn_limit = tcp_conn_limit; 2766 c->tcl_addr = NULL; 2767 c->tcp_keepalive = 0; 2768 c->tcp_parent = NULL; 2769 c->max_tcp_count = num; 2770 c->cur_tcp_count = 0; 2771 c->tcp_handlers = (struct comm_point**)calloc((size_t)num, 2772 sizeof(struct comm_point*)); 2773 if(!c->tcp_handlers) { 2774 free(c->ev); 2775 free(c); 2776 return NULL; 2777 } 2778 c->tcp_free = NULL; 2779 c->type = comm_tcp_accept; 2780 c->tcp_do_close = 0; 2781 c->do_not_close = 0; 2782 c->tcp_do_toggle_rw = 0; 2783 c->tcp_check_nb_connect = 0; 2784 #ifdef USE_MSG_FASTOPEN 2785 c->tcp_do_fastopen = 0; 2786 #endif 2787 #ifdef USE_DNSCRYPT 2788 c->dnscrypt = 0; 2789 c->dnscrypt_buffer = NULL; 2790 #endif 2791 c->callback = NULL; 2792 c->cb_arg = NULL; 2793 evbits = UB_EV_READ | UB_EV_PERSIST; 2794 /* ub_event stuff */ 2795 c->ev->ev = ub_event_new(base->eb->base, c->fd, evbits, 2796 comm_point_tcp_accept_callback, c); 2797 if(c->ev->ev == NULL) { 2798 log_err("could not baseset tcpacc event"); 2799 comm_point_delete(c); 2800 return NULL; 2801 } 2802 if (ub_event_add(c->ev->ev, c->timeout) != 0) { 2803 log_err("could not add tcpacc event"); 2804 comm_point_delete(c); 2805 return NULL; 2806 } 2807 /* now prealloc the tcp handlers */ 2808 for(i=0; i<num; i++) { 2809 c->tcp_handlers[i] = comm_point_create_tcp_handler(base, 2810 c, bufsize, spoolbuf, callback, callback_arg); 2811 if(!c->tcp_handlers[i]) { 2812 comm_point_delete(c); 2813 return NULL; 2814 } 2815 } 2816 2817 return c; 2818 } 2819 2820 struct comm_point* 2821 comm_point_create_tcp_out(struct comm_base *base, size_t bufsize, 2822 comm_point_callback_type* callback, void* callback_arg) 2823 { 2824 struct comm_point* c = (struct comm_point*)calloc(1, 2825 sizeof(struct comm_point)); 2826 short evbits; 2827 if(!c) 2828 return NULL; 2829 c->ev = (struct internal_event*)calloc(1, 2830 sizeof(struct internal_event)); 2831 if(!c->ev) { 2832 free(c); 2833 return NULL; 2834 } 2835 c->ev->base = base; 2836 c->fd = -1; 2837 c->buffer = sldns_buffer_new(bufsize); 2838 if(!c->buffer) { 2839 free(c->ev); 2840 free(c); 2841 return NULL; 2842 } 2843 c->timeout = NULL; 2844 c->tcp_is_reading = 0; 2845 c->tcp_byte_count = 0; 2846 c->tcp_timeout_msec = TCP_QUERY_TIMEOUT; 2847 c->tcp_conn_limit = NULL; 2848 c->tcl_addr = NULL; 2849 c->tcp_keepalive = 0; 2850 c->tcp_parent = NULL; 2851 c->max_tcp_count = 0; 2852 c->cur_tcp_count = 0; 2853 c->tcp_handlers = NULL; 2854 c->tcp_free = NULL; 2855 c->type = comm_tcp; 2856 c->tcp_do_close = 0; 2857 c->do_not_close = 0; 2858 c->tcp_do_toggle_rw = 1; 2859 c->tcp_check_nb_connect = 1; 2860 #ifdef USE_MSG_FASTOPEN 2861 c->tcp_do_fastopen = 1; 2862 #endif 2863 #ifdef USE_DNSCRYPT 2864 c->dnscrypt = 0; 2865 c->dnscrypt_buffer = c->buffer; 2866 #endif 2867 c->repinfo.c = c; 2868 c->callback = callback; 2869 c->cb_arg = callback_arg; 2870 evbits = UB_EV_PERSIST | UB_EV_WRITE; 2871 c->ev->ev = ub_event_new(base->eb->base, c->fd, evbits, 2872 comm_point_tcp_handle_callback, c); 2873 if(c->ev->ev == NULL) 2874 { 2875 log_err("could not baseset tcpout event"); 2876 sldns_buffer_free(c->buffer); 2877 free(c->ev); 2878 free(c); 2879 return NULL; 2880 } 2881 2882 return c; 2883 } 2884 2885 struct comm_point* 2886 comm_point_create_http_out(struct comm_base *base, size_t bufsize, 2887 comm_point_callback_type* callback, void* callback_arg, 2888 sldns_buffer* temp) 2889 { 2890 struct comm_point* c = (struct comm_point*)calloc(1, 2891 sizeof(struct comm_point)); 2892 short evbits; 2893 if(!c) 2894 return NULL; 2895 c->ev = (struct internal_event*)calloc(1, 2896 sizeof(struct internal_event)); 2897 if(!c->ev) { 2898 free(c); 2899 return NULL; 2900 } 2901 c->ev->base = base; 2902 c->fd = -1; 2903 c->buffer = sldns_buffer_new(bufsize); 2904 if(!c->buffer) { 2905 free(c->ev); 2906 free(c); 2907 return NULL; 2908 } 2909 c->timeout = NULL; 2910 c->tcp_is_reading = 0; 2911 c->tcp_byte_count = 0; 2912 c->tcp_parent = NULL; 2913 c->max_tcp_count = 0; 2914 c->cur_tcp_count = 0; 2915 c->tcp_handlers = NULL; 2916 c->tcp_free = NULL; 2917 c->type = comm_http; 2918 c->tcp_do_close = 0; 2919 c->do_not_close = 0; 2920 c->tcp_do_toggle_rw = 1; 2921 c->tcp_check_nb_connect = 1; 2922 c->http_in_headers = 1; 2923 c->http_in_chunk_headers = 0; 2924 c->http_is_chunked = 0; 2925 c->http_temp = temp; 2926 #ifdef USE_MSG_FASTOPEN 2927 c->tcp_do_fastopen = 1; 2928 #endif 2929 #ifdef USE_DNSCRYPT 2930 c->dnscrypt = 0; 2931 c->dnscrypt_buffer = c->buffer; 2932 #endif 2933 c->repinfo.c = c; 2934 c->callback = callback; 2935 c->cb_arg = callback_arg; 2936 evbits = UB_EV_PERSIST | UB_EV_WRITE; 2937 c->ev->ev = ub_event_new(base->eb->base, c->fd, evbits, 2938 comm_point_http_handle_callback, c); 2939 if(c->ev->ev == NULL) 2940 { 2941 log_err("could not baseset tcpout event"); 2942 #ifdef HAVE_SSL 2943 SSL_free(c->ssl); 2944 #endif 2945 sldns_buffer_free(c->buffer); 2946 free(c->ev); 2947 free(c); 2948 return NULL; 2949 } 2950 2951 return c; 2952 } 2953 2954 struct comm_point* 2955 comm_point_create_local(struct comm_base *base, int fd, size_t bufsize, 2956 comm_point_callback_type* callback, void* callback_arg) 2957 { 2958 struct comm_point* c = (struct comm_point*)calloc(1, 2959 sizeof(struct comm_point)); 2960 short evbits; 2961 if(!c) 2962 return NULL; 2963 c->ev = (struct internal_event*)calloc(1, 2964 sizeof(struct internal_event)); 2965 if(!c->ev) { 2966 free(c); 2967 return NULL; 2968 } 2969 c->ev->base = base; 2970 c->fd = fd; 2971 c->buffer = sldns_buffer_new(bufsize); 2972 if(!c->buffer) { 2973 free(c->ev); 2974 free(c); 2975 return NULL; 2976 } 2977 c->timeout = NULL; 2978 c->tcp_is_reading = 1; 2979 c->tcp_byte_count = 0; 2980 c->tcp_parent = NULL; 2981 c->max_tcp_count = 0; 2982 c->cur_tcp_count = 0; 2983 c->tcp_handlers = NULL; 2984 c->tcp_free = NULL; 2985 c->type = comm_local; 2986 c->tcp_do_close = 0; 2987 c->do_not_close = 1; 2988 c->tcp_do_toggle_rw = 0; 2989 c->tcp_check_nb_connect = 0; 2990 #ifdef USE_MSG_FASTOPEN 2991 c->tcp_do_fastopen = 0; 2992 #endif 2993 #ifdef USE_DNSCRYPT 2994 c->dnscrypt = 0; 2995 c->dnscrypt_buffer = c->buffer; 2996 #endif 2997 c->callback = callback; 2998 c->cb_arg = callback_arg; 2999 /* ub_event stuff */ 3000 evbits = UB_EV_PERSIST | UB_EV_READ; 3001 c->ev->ev = ub_event_new(base->eb->base, c->fd, evbits, 3002 comm_point_local_handle_callback, c); 3003 if(c->ev->ev == NULL) { 3004 log_err("could not baseset localhdl event"); 3005 free(c->ev); 3006 free(c); 3007 return NULL; 3008 } 3009 if (ub_event_add(c->ev->ev, c->timeout) != 0) { 3010 log_err("could not add localhdl event"); 3011 ub_event_free(c->ev->ev); 3012 free(c->ev); 3013 free(c); 3014 return NULL; 3015 } 3016 return c; 3017 } 3018 3019 struct comm_point* 3020 comm_point_create_raw(struct comm_base* base, int fd, int writing, 3021 comm_point_callback_type* callback, void* callback_arg) 3022 { 3023 struct comm_point* c = (struct comm_point*)calloc(1, 3024 sizeof(struct comm_point)); 3025 short evbits; 3026 if(!c) 3027 return NULL; 3028 c->ev = (struct internal_event*)calloc(1, 3029 sizeof(struct internal_event)); 3030 if(!c->ev) { 3031 free(c); 3032 return NULL; 3033 } 3034 c->ev->base = base; 3035 c->fd = fd; 3036 c->buffer = NULL; 3037 c->timeout = NULL; 3038 c->tcp_is_reading = 0; 3039 c->tcp_byte_count = 0; 3040 c->tcp_parent = NULL; 3041 c->max_tcp_count = 0; 3042 c->cur_tcp_count = 0; 3043 c->tcp_handlers = NULL; 3044 c->tcp_free = NULL; 3045 c->type = comm_raw; 3046 c->tcp_do_close = 0; 3047 c->do_not_close = 1; 3048 c->tcp_do_toggle_rw = 0; 3049 c->tcp_check_nb_connect = 0; 3050 #ifdef USE_MSG_FASTOPEN 3051 c->tcp_do_fastopen = 0; 3052 #endif 3053 #ifdef USE_DNSCRYPT 3054 c->dnscrypt = 0; 3055 c->dnscrypt_buffer = c->buffer; 3056 #endif 3057 c->callback = callback; 3058 c->cb_arg = callback_arg; 3059 /* ub_event stuff */ 3060 if(writing) 3061 evbits = UB_EV_PERSIST | UB_EV_WRITE; 3062 else evbits = UB_EV_PERSIST | UB_EV_READ; 3063 c->ev->ev = ub_event_new(base->eb->base, c->fd, evbits, 3064 comm_point_raw_handle_callback, c); 3065 if(c->ev->ev == NULL) { 3066 log_err("could not baseset rawhdl event"); 3067 free(c->ev); 3068 free(c); 3069 return NULL; 3070 } 3071 if (ub_event_add(c->ev->ev, c->timeout) != 0) { 3072 log_err("could not add rawhdl event"); 3073 ub_event_free(c->ev->ev); 3074 free(c->ev); 3075 free(c); 3076 return NULL; 3077 } 3078 return c; 3079 } 3080 3081 void 3082 comm_point_close(struct comm_point* c) 3083 { 3084 if(!c) 3085 return; 3086 if(c->fd != -1) { 3087 if(ub_event_del(c->ev->ev) != 0) { 3088 log_err("could not event_del on close"); 3089 } 3090 } 3091 tcl_close_connection(c->tcl_addr); 3092 if(c->tcp_req_info) 3093 tcp_req_info_clear(c->tcp_req_info); 3094 /* close fd after removing from event lists, or epoll.. is messed up */ 3095 if(c->fd != -1 && !c->do_not_close) { 3096 if(c->type == comm_tcp || c->type == comm_http) { 3097 /* delete sticky events for the fd, it gets closed */ 3098 ub_winsock_tcp_wouldblock(c->ev->ev, UB_EV_READ); 3099 ub_winsock_tcp_wouldblock(c->ev->ev, UB_EV_WRITE); 3100 } 3101 verbose(VERB_ALGO, "close fd %d", c->fd); 3102 #ifndef USE_WINSOCK 3103 close(c->fd); 3104 #else 3105 closesocket(c->fd); 3106 #endif 3107 } 3108 c->fd = -1; 3109 } 3110 3111 void 3112 comm_point_delete(struct comm_point* c) 3113 { 3114 if(!c) 3115 return; 3116 if((c->type == comm_tcp || c->type == comm_http) && c->ssl) { 3117 #ifdef HAVE_SSL 3118 SSL_shutdown(c->ssl); 3119 SSL_free(c->ssl); 3120 #endif 3121 } 3122 comm_point_close(c); 3123 if(c->tcp_handlers) { 3124 int i; 3125 for(i=0; i<c->max_tcp_count; i++) 3126 comm_point_delete(c->tcp_handlers[i]); 3127 free(c->tcp_handlers); 3128 } 3129 free(c->timeout); 3130 if(c->type == comm_tcp || c->type == comm_local || c->type == comm_http) { 3131 sldns_buffer_free(c->buffer); 3132 #ifdef USE_DNSCRYPT 3133 if(c->dnscrypt && c->dnscrypt_buffer != c->buffer) { 3134 sldns_buffer_free(c->dnscrypt_buffer); 3135 } 3136 #endif 3137 if(c->tcp_req_info) { 3138 tcp_req_info_delete(c->tcp_req_info); 3139 } 3140 } 3141 ub_event_free(c->ev->ev); 3142 free(c->ev); 3143 free(c); 3144 } 3145 3146 void 3147 comm_point_send_reply(struct comm_reply *repinfo) 3148 { 3149 struct sldns_buffer* buffer; 3150 log_assert(repinfo && repinfo->c); 3151 #ifdef USE_DNSCRYPT 3152 buffer = repinfo->c->dnscrypt_buffer; 3153 if(!dnsc_handle_uncurved_request(repinfo)) { 3154 return; 3155 } 3156 #else 3157 buffer = repinfo->c->buffer; 3158 #endif 3159 if(repinfo->c->type == comm_udp) { 3160 if(repinfo->srctype) 3161 comm_point_send_udp_msg_if(repinfo->c, 3162 buffer, (struct sockaddr*)&repinfo->addr, 3163 repinfo->addrlen, repinfo); 3164 else 3165 comm_point_send_udp_msg(repinfo->c, buffer, 3166 (struct sockaddr*)&repinfo->addr, repinfo->addrlen); 3167 #ifdef USE_DNSTAP 3168 if(repinfo->c->dtenv != NULL && 3169 repinfo->c->dtenv->log_client_response_messages) 3170 dt_msg_send_client_response(repinfo->c->dtenv, 3171 &repinfo->addr, repinfo->c->type, repinfo->c->buffer); 3172 #endif 3173 } else { 3174 #ifdef USE_DNSTAP 3175 if(repinfo->c->tcp_parent->dtenv != NULL && 3176 repinfo->c->tcp_parent->dtenv->log_client_response_messages) 3177 dt_msg_send_client_response(repinfo->c->tcp_parent->dtenv, 3178 &repinfo->addr, repinfo->c->type, repinfo->c->buffer); 3179 #endif 3180 if(repinfo->c->tcp_req_info) { 3181 tcp_req_info_send_reply(repinfo->c->tcp_req_info); 3182 } else { 3183 comm_point_start_listening(repinfo->c, -1, 3184 repinfo->c->tcp_timeout_msec); 3185 } 3186 } 3187 } 3188 3189 void 3190 comm_point_drop_reply(struct comm_reply* repinfo) 3191 { 3192 if(!repinfo) 3193 return; 3194 log_assert(repinfo->c); 3195 log_assert(repinfo->c->type != comm_tcp_accept); 3196 if(repinfo->c->type == comm_udp) 3197 return; 3198 if(repinfo->c->tcp_req_info) 3199 repinfo->c->tcp_req_info->is_drop = 1; 3200 reclaim_tcp_handler(repinfo->c); 3201 } 3202 3203 void 3204 comm_point_stop_listening(struct comm_point* c) 3205 { 3206 verbose(VERB_ALGO, "comm point stop listening %d", c->fd); 3207 if(ub_event_del(c->ev->ev) != 0) { 3208 log_err("event_del error to stoplisten"); 3209 } 3210 } 3211 3212 void 3213 comm_point_start_listening(struct comm_point* c, int newfd, int msec) 3214 { 3215 verbose(VERB_ALGO, "comm point start listening %d (%d msec)", 3216 c->fd==-1?newfd:c->fd, msec); 3217 if(c->type == comm_tcp_accept && !c->tcp_free) { 3218 /* no use to start listening no free slots. */ 3219 return; 3220 } 3221 if(msec != -1 && msec != 0) { 3222 if(!c->timeout) { 3223 c->timeout = (struct timeval*)malloc(sizeof( 3224 struct timeval)); 3225 if(!c->timeout) { 3226 log_err("cpsl: malloc failed. No net read."); 3227 return; 3228 } 3229 } 3230 ub_event_add_bits(c->ev->ev, UB_EV_TIMEOUT); 3231 #ifndef S_SPLINT_S /* splint fails on struct timeval. */ 3232 c->timeout->tv_sec = msec/1000; 3233 c->timeout->tv_usec = (msec%1000)*1000; 3234 #endif /* S_SPLINT_S */ 3235 } 3236 if(c->type == comm_tcp || c->type == comm_http) { 3237 ub_event_del_bits(c->ev->ev, UB_EV_READ|UB_EV_WRITE); 3238 if(c->tcp_is_reading) 3239 ub_event_add_bits(c->ev->ev, UB_EV_READ); 3240 else ub_event_add_bits(c->ev->ev, UB_EV_WRITE); 3241 } 3242 if(newfd != -1) { 3243 if(c->fd != -1) { 3244 #ifndef USE_WINSOCK 3245 close(c->fd); 3246 #else 3247 closesocket(c->fd); 3248 #endif 3249 } 3250 c->fd = newfd; 3251 ub_event_set_fd(c->ev->ev, c->fd); 3252 } 3253 if(ub_event_add(c->ev->ev, msec==0?NULL:c->timeout) != 0) { 3254 log_err("event_add failed. in cpsl."); 3255 } 3256 } 3257 3258 void comm_point_listen_for_rw(struct comm_point* c, int rd, int wr) 3259 { 3260 verbose(VERB_ALGO, "comm point listen_for_rw %d %d", c->fd, wr); 3261 if(ub_event_del(c->ev->ev) != 0) { 3262 log_err("event_del error to cplf"); 3263 } 3264 ub_event_del_bits(c->ev->ev, UB_EV_READ|UB_EV_WRITE); 3265 if(rd) ub_event_add_bits(c->ev->ev, UB_EV_READ); 3266 if(wr) ub_event_add_bits(c->ev->ev, UB_EV_WRITE); 3267 if(ub_event_add(c->ev->ev, c->timeout) != 0) { 3268 log_err("event_add failed. in cplf."); 3269 } 3270 } 3271 3272 size_t comm_point_get_mem(struct comm_point* c) 3273 { 3274 size_t s; 3275 if(!c) 3276 return 0; 3277 s = sizeof(*c) + sizeof(*c->ev); 3278 if(c->timeout) 3279 s += sizeof(*c->timeout); 3280 if(c->type == comm_tcp || c->type == comm_local) { 3281 s += sizeof(*c->buffer) + sldns_buffer_capacity(c->buffer); 3282 #ifdef USE_DNSCRYPT 3283 s += sizeof(*c->dnscrypt_buffer); 3284 if(c->buffer != c->dnscrypt_buffer) { 3285 s += sldns_buffer_capacity(c->dnscrypt_buffer); 3286 } 3287 #endif 3288 } 3289 if(c->type == comm_tcp_accept) { 3290 int i; 3291 for(i=0; i<c->max_tcp_count; i++) 3292 s += comm_point_get_mem(c->tcp_handlers[i]); 3293 } 3294 return s; 3295 } 3296 3297 struct comm_timer* 3298 comm_timer_create(struct comm_base* base, void (*cb)(void*), void* cb_arg) 3299 { 3300 struct internal_timer *tm = (struct internal_timer*)calloc(1, 3301 sizeof(struct internal_timer)); 3302 if(!tm) { 3303 log_err("malloc failed"); 3304 return NULL; 3305 } 3306 tm->super.ev_timer = tm; 3307 tm->base = base; 3308 tm->super.callback = cb; 3309 tm->super.cb_arg = cb_arg; 3310 tm->ev = ub_event_new(base->eb->base, -1, UB_EV_TIMEOUT, 3311 comm_timer_callback, &tm->super); 3312 if(tm->ev == NULL) { 3313 log_err("timer_create: event_base_set failed."); 3314 free(tm); 3315 return NULL; 3316 } 3317 return &tm->super; 3318 } 3319 3320 void 3321 comm_timer_disable(struct comm_timer* timer) 3322 { 3323 if(!timer) 3324 return; 3325 ub_timer_del(timer->ev_timer->ev); 3326 timer->ev_timer->enabled = 0; 3327 } 3328 3329 void 3330 comm_timer_set(struct comm_timer* timer, struct timeval* tv) 3331 { 3332 log_assert(tv); 3333 if(timer->ev_timer->enabled) 3334 comm_timer_disable(timer); 3335 if(ub_timer_add(timer->ev_timer->ev, timer->ev_timer->base->eb->base, 3336 comm_timer_callback, timer, tv) != 0) 3337 log_err("comm_timer_set: evtimer_add failed."); 3338 timer->ev_timer->enabled = 1; 3339 } 3340 3341 void 3342 comm_timer_delete(struct comm_timer* timer) 3343 { 3344 if(!timer) 3345 return; 3346 comm_timer_disable(timer); 3347 /* Free the sub struct timer->ev_timer derived from the super struct timer. 3348 * i.e. assert(timer == timer->ev_timer) 3349 */ 3350 ub_event_free(timer->ev_timer->ev); 3351 free(timer->ev_timer); 3352 } 3353 3354 void 3355 comm_timer_callback(int ATTR_UNUSED(fd), short event, void* arg) 3356 { 3357 struct comm_timer* tm = (struct comm_timer*)arg; 3358 if(!(event&UB_EV_TIMEOUT)) 3359 return; 3360 ub_comm_base_now(tm->ev_timer->base); 3361 tm->ev_timer->enabled = 0; 3362 fptr_ok(fptr_whitelist_comm_timer(tm->callback)); 3363 (*tm->callback)(tm->cb_arg); 3364 } 3365 3366 int 3367 comm_timer_is_set(struct comm_timer* timer) 3368 { 3369 return (int)timer->ev_timer->enabled; 3370 } 3371 3372 size_t 3373 comm_timer_get_mem(struct comm_timer* ATTR_UNUSED(timer)) 3374 { 3375 return sizeof(struct internal_timer); 3376 } 3377 3378 struct comm_signal* 3379 comm_signal_create(struct comm_base* base, 3380 void (*callback)(int, void*), void* cb_arg) 3381 { 3382 struct comm_signal* com = (struct comm_signal*)malloc( 3383 sizeof(struct comm_signal)); 3384 if(!com) { 3385 log_err("malloc failed"); 3386 return NULL; 3387 } 3388 com->base = base; 3389 com->callback = callback; 3390 com->cb_arg = cb_arg; 3391 com->ev_signal = NULL; 3392 return com; 3393 } 3394 3395 void 3396 comm_signal_callback(int sig, short event, void* arg) 3397 { 3398 struct comm_signal* comsig = (struct comm_signal*)arg; 3399 if(!(event & UB_EV_SIGNAL)) 3400 return; 3401 ub_comm_base_now(comsig->base); 3402 fptr_ok(fptr_whitelist_comm_signal(comsig->callback)); 3403 (*comsig->callback)(sig, comsig->cb_arg); 3404 } 3405 3406 int 3407 comm_signal_bind(struct comm_signal* comsig, int sig) 3408 { 3409 struct internal_signal* entry = (struct internal_signal*)calloc(1, 3410 sizeof(struct internal_signal)); 3411 if(!entry) { 3412 log_err("malloc failed"); 3413 return 0; 3414 } 3415 log_assert(comsig); 3416 /* add signal event */ 3417 entry->ev = ub_signal_new(comsig->base->eb->base, sig, 3418 comm_signal_callback, comsig); 3419 if(entry->ev == NULL) { 3420 log_err("Could not create signal event"); 3421 free(entry); 3422 return 0; 3423 } 3424 if(ub_signal_add(entry->ev, NULL) != 0) { 3425 log_err("Could not add signal handler"); 3426 ub_event_free(entry->ev); 3427 free(entry); 3428 return 0; 3429 } 3430 /* link into list */ 3431 entry->next = comsig->ev_signal; 3432 comsig->ev_signal = entry; 3433 return 1; 3434 } 3435 3436 void 3437 comm_signal_delete(struct comm_signal* comsig) 3438 { 3439 struct internal_signal* p, *np; 3440 if(!comsig) 3441 return; 3442 p=comsig->ev_signal; 3443 while(p) { 3444 np = p->next; 3445 ub_signal_del(p->ev); 3446 ub_event_free(p->ev); 3447 free(p); 3448 p = np; 3449 } 3450 free(comsig); 3451 } 3452