]>
Commit | Line | Data |
---|---|---|
30a4f2a8 | 1 | /* |
b8ae064d | 2 | * Copyright (C) 1996-2023 The Squid Software Foundation and contributors |
30a4f2a8 | 3 | * |
bbc27441 AJ |
4 | * Squid software is distributed under GPLv2+ license and includes |
5 | * contributions from numerous individuals and organizations. | |
6 | * Please see the COPYING and CONTRIBUTORS files for details. | |
30a4f2a8 | 7 | */ |
090089c4 | 8 | |
bbc27441 AJ |
9 | /* DEBUG: section 05 Socket Functions */ |
10 | ||
582c2af2 | 11 | #include "squid.h" |
2949dd02 | 12 | #include "base/AsyncFunCalls.h" |
582c2af2 | 13 | #include "ClientInfo.h" |
04f55905 AJ |
14 | #include "comm/AcceptLimiter.h" |
15 | #include "comm/comm_internal.h" | |
cfd66529 | 16 | #include "comm/Connection.h" |
ec41b64c | 17 | #include "comm/IoCallback.h" |
d841c88d | 18 | #include "comm/Loops.h" |
7e66d5e2 | 19 | #include "comm/Read.h" |
cbff89ba | 20 | #include "comm/TcpAcceptor.h" |
602d9612 | 21 | #include "comm/Write.h" |
582c2af2 | 22 | #include "compat/cmsg.h" |
74257126 | 23 | #include "DescriptorSet.h" |
582c2af2 | 24 | #include "event.h" |
c4ad1349 | 25 | #include "fd.h" |
582c2af2 FC |
26 | #include "fde.h" |
27 | #include "globals.h" | |
9b5c4a9a | 28 | #include "icmp/net_db.h" |
96d89ea0 | 29 | #include "ip/Intercept.h" |
425de4c8 | 30 | #include "ip/QosConfig.h" |
055421ee | 31 | #include "ip/tools.h" |
582c2af2 | 32 | #include "pconn.h" |
65e41a45 | 33 | #include "sbuf/SBuf.h" |
7d17a6a6 | 34 | #include "sbuf/Stream.h" |
4d5904f7 | 35 | #include "SquidConfig.h" |
e4f1fdae | 36 | #include "StatCounters.h" |
582c2af2 | 37 | #include "StoreIOBuffer.h" |
1125ea7b MM |
38 | #include "tools.h" |
39 | ||
cb4f4424 | 40 | #if USE_OPENSSL |
4db984be CT |
41 | #include "ssl/support.h" |
42 | #endif | |
090089c4 | 43 | |
074d6a40 AJ |
44 | #include <cerrno> |
45 | #include <cmath> | |
be266cb2 | 46 | #if _SQUID_CYGWIN_ |
b671cc68 | 47 | #include <sys/ioctl.h> |
48 | #endif | |
30a4f2a8 | 49 | #ifdef HAVE_NETINET_TCP_H |
50 | #include <netinet/tcp.h> | |
51 | #endif | |
582c2af2 FC |
52 | #if HAVE_SYS_UN_H |
53 | #include <sys/un.h> | |
54 | #endif | |
090089c4 | 55 | |
2b663917 | 56 | /* |
57 | * New C-like simple comm code. This stuff is a mess and doesn't really buy us anything. | |
58 | */ | |
59 | ||
82ec8dfc | 60 | static IOCB commHalfClosedReader; |
74d7d196 | 61 | static int comm_openex(int sock_type, int proto, Ip::Address &, int flags, const char *note); |
c6f168c1 | 62 | static void comm_init_opened(const Comm::ConnectionPointer &conn, const char *note, struct addrinfo *AI); |
a67d2b2e | 63 | static int comm_apply_flags(int new_socket, Ip::Address &addr, int flags, struct addrinfo *AI); |
82ec8dfc | 64 | |
9a0a18de | 65 | #if USE_DELAY_POOLS |
b4cd430a CT |
66 | CBDATA_CLASS_INIT(CommQuotaQueue); |
67 | ||
68 | static void commHandleWriteHelper(void * data); | |
69 | #endif | |
70 | ||
090089c4 | 71 | /* STATIC */ |
62e76326 | 72 | |
aee3523a | 73 | static DescriptorSet *TheHalfClosed = nullptr; /// the set of half-closed FDs |
74257126 AR |
74 | static bool WillCheckHalfClosed = false; /// true if check is scheduled |
75 | static EVH commHalfClosedCheck; | |
76 | static void commPlanHalfClosedCheck(); | |
77 | ||
c8407295 | 78 | static Comm::Flag commBind(int s, struct addrinfo &); |
74d7d196 | 79 | static void commSetBindAddressNoPort(int); |
f5b8bbc4 | 80 | static void commSetReuseAddr(int); |
81 | static void commSetNoLinger(int); | |
30a4f2a8 | 82 | #ifdef TCP_NODELAY |
f5b8bbc4 | 83 | static void commSetTcpNoDelay(int); |
30a4f2a8 | 84 | #endif |
f5b8bbc4 | 85 | static void commSetTcpRcvbuf(int, int); |
723123a9 | 86 | |
04f55905 | 87 | bool |
b0469965 | 88 | isOpen(const int fd) |
b300c36d | 89 | { |
9a5ac93d | 90 | return fd >= 0 && fd_table && fd_table[fd].flags.open != 0; |
b300c36d | 91 | } |
92 | ||
e1a88700 | 93 | /** |
c4b7a5a9 | 94 | * Empty the read buffers |
95 | * | |
96 | * This is a magical routine that empties the read buffers. | |
97 | * Under some platforms (Linux) if a buffer has data in it before | |
98 | * you call close(), the socket will hang and take quite a while | |
99 | * to timeout. | |
100 | */ | |
101 | static void | |
102 | comm_empty_os_read_buffers(int fd) | |
103 | { | |
1191b93b | 104 | #if _SQUID_LINUX_ |
b41a5a1b AJ |
105 | #if USE_OPENSSL |
106 | // Bug 4146: SSL-Bump BIO does not release sockets on close. | |
107 | if (fd_table[fd].ssl) | |
108 | return; | |
109 | #endif | |
110 | ||
c4b7a5a9 | 111 | /* prevent those nasty RST packets */ |
112 | char buf[SQUID_TCP_SO_RCVBUF]; | |
e3c37fc2 | 113 | if (fd_table[fd].flags.nonblocking && fd_table[fd].type != FD_MSGHDR) { |
cc192b50 | 114 | while (FD_READ_METHOD(fd, buf, SQUID_TCP_SO_RCVBUF) > 0) {}; |
115 | } | |
2081cefa FC |
116 | #else |
117 | (void)fd; | |
c4b7a5a9 | 118 | #endif |
119 | } | |
120 | ||
e1a88700 | 121 | /** |
ce767c23 | 122 | * synchronous wrapper around udp socket functions |
123 | */ | |
ce767c23 | 124 | int |
b7ac5457 | 125 | comm_udp_recvfrom(int fd, void *buf, size_t len, int flags, Ip::Address &from) |
ce767c23 | 126 | { |
95dc7ff4 | 127 | ++ statCounter.syscalls.sock.recvfroms; |
cc192b50 | 128 | debugs(5,8, "comm_udp_recvfrom: FD " << fd << " from " << from); |
aee3523a | 129 | struct addrinfo *AI = nullptr; |
851614a8 | 130 | Ip::Address::InitAddr(AI); |
4dd643d5 | 131 | int x = recvfrom(fd, buf, len, flags, AI->ai_addr, &AI->ai_addrlen); |
cc192b50 | 132 | from = *AI; |
851614a8 | 133 | Ip::Address::FreeAddr(AI); |
cc192b50 | 134 | return x; |
ce767c23 | 135 | } |
136 | ||
365f12a9 | 137 | int |
7d21986b | 138 | comm_udp_recv(int fd, void *buf, size_t len, int flags) |
365f12a9 | 139 | { |
b7ac5457 | 140 | Ip::Address nul; |
cc192b50 | 141 | return comm_udp_recvfrom(fd, buf, len, flags, nul); |
365f12a9 | 142 | } |
143 | ||
f71da12c | 144 | ssize_t |
7d21986b | 145 | comm_udp_send(int s, const void *buf, size_t len, int flags) |
f71da12c | 146 | { |
62e76326 | 147 | return send(s, buf, len, flags); |
f71da12c | 148 | } |
ce767c23 | 149 | |
545d554b | 150 | bool |
151 | comm_has_incomplete_write(int fd) | |
152 | { | |
aee3523a | 153 | assert(isOpen(fd) && COMMIO_FD_WRITECB(fd) != nullptr); |
b0469965 | 154 | return COMMIO_FD_WRITECB(fd)->active(); |
d4cb310b | 155 | } |
156 | ||
e1a88700 | 157 | /** |
cf3c0ee3 | 158 | * Queue a write. handler/handler_data are called when the write fully |
159 | * completes, on error, or on file descriptor close. | |
160 | */ | |
9864ee44 | 161 | |
090089c4 | 162 | /* Return the local port associated with fd. */ |
f45dd259 | 163 | unsigned short |
b8d8561b | 164 | comm_local_port(int fd) |
090089c4 | 165 | { |
b7ac5457 | 166 | Ip::Address temp; |
aee3523a | 167 | struct addrinfo *addr = nullptr; |
76f87348 | 168 | fde *F = &fd_table[fd]; |
090089c4 | 169 | |
090089c4 | 170 | /* If the fd is closed already, just return */ |
62e76326 | 171 | |
60c0b5a2 | 172 | if (!F->flags.open) { |
bf8fe701 | 173 | debugs(5, 0, "comm_local_port: FD " << fd << " has been closed."); |
62e76326 | 174 | return 0; |
090089c4 | 175 | } |
62e76326 | 176 | |
4dd643d5 AJ |
177 | if (F->local_addr.port()) |
178 | return F->local_addr.port(); | |
62e76326 | 179 | |
6084c0b6 | 180 | if (F->sock_family == AF_INET) |
4dd643d5 | 181 | temp.setIPv4(); |
6084c0b6 | 182 | |
851614a8 | 183 | Ip::Address::InitAddr(addr); |
62e76326 | 184 | |
cc192b50 | 185 | if (getsockname(fd, addr->ai_addr, &(addr->ai_addrlen)) ) { |
b69e9ffa | 186 | int xerrno = errno; |
d816f28d | 187 | debugs(50, DBG_IMPORTANT, "ERROR: " << MYNAME << "Failed to retrieve TCP/UDP port number for socket: FD " << fd << ": " << xstrerr(xerrno)); |
851614a8 | 188 | Ip::Address::FreeAddr(addr); |
62e76326 | 189 | return 0; |
090089c4 | 190 | } |
cc192b50 | 191 | temp = *addr; |
192 | ||
851614a8 | 193 | Ip::Address::FreeAddr(addr); |
cc192b50 | 194 | |
4dd643d5 | 195 | if (F->local_addr.isAnyAddr()) { |
3d031c35 AJ |
196 | /* save the whole local address, not just the port. */ |
197 | F->local_addr = temp; | |
198 | } else { | |
4dd643d5 | 199 | F->local_addr.port(temp.port()); |
3d031c35 | 200 | } |
62e76326 | 201 | |
4dd643d5 AJ |
202 | debugs(5, 6, "comm_local_port: FD " << fd << ": port " << F->local_addr.port() << "(family=" << F->sock_family << ")"); |
203 | return F->local_addr.port(); | |
090089c4 | 204 | } |
205 | ||
74d7d196 | 206 | /// sets the IP_BIND_ADDRESS_NO_PORT socket option to optimize ephemeral port |
207 | /// reuse by outgoing TCP connections that must bind(2) to a source IP address | |
208 | static void | |
209 | commSetBindAddressNoPort(const int fd) | |
210 | { | |
211 | #if defined(IP_BIND_ADDRESS_NO_PORT) | |
212 | int flag = 1; | |
213 | if (setsockopt(fd, IPPROTO_IP, IP_BIND_ADDRESS_NO_PORT, reinterpret_cast<char*>(&flag), sizeof(flag)) < 0) { | |
214 | const auto savedErrno = errno; | |
215 | debugs(50, DBG_IMPORTANT, "ERROR: setsockopt(IP_BIND_ADDRESS_NO_PORT) failure: " << xstrerr(savedErrno)); | |
216 | } | |
217 | #else | |
218 | (void)fd; | |
219 | #endif | |
220 | } | |
221 | ||
c8407295 | 222 | static Comm::Flag |
cc192b50 | 223 | commBind(int s, struct addrinfo &inaddr) |
090089c4 | 224 | { |
95dc7ff4 | 225 | ++ statCounter.syscalls.sock.binds; |
62e76326 | 226 | |
ac760b5e | 227 | if (bind(s, inaddr.ai_addr, inaddr.ai_addrlen) == 0) { |
b69e9ffa | 228 | debugs(50, 6, "bind socket FD " << s << " to " << fd_table[s].local_addr); |
c8407295 | 229 | return Comm::OK; |
ac760b5e | 230 | } |
b69e9ffa | 231 | int xerrno = errno; |
d816f28d | 232 | debugs(50, DBG_CRITICAL, "ERROR: " << MYNAME << "Cannot bind socket FD " << s << " to " << fd_table[s].local_addr << ": " << xstrerr(xerrno)); |
62e76326 | 233 | |
4ee57cbe | 234 | return Comm::COMM_ERROR; |
090089c4 | 235 | } |
236 | ||
e1a88700 | 237 | /** |
238 | * Create a socket. Default is blocking, stream (TCP) socket. IO_TYPE | |
239 | * is OR of flags specified in comm.h. Defaults TOS | |
240 | */ | |
b8d8561b | 241 | int |
16b204c4 | 242 | comm_open(int sock_type, |
62e76326 | 243 | int proto, |
b7ac5457 | 244 | Ip::Address &addr, |
62e76326 | 245 | int flags, |
246 | const char *note) | |
d6827718 | 247 | { |
74d7d196 | 248 | // assume zero-port callers do not need to know the assigned port right away |
249 | if (sock_type == SOCK_STREAM && addr.port() == 0 && ((flags & COMM_DOBIND) || !addr.isAnyAddr())) | |
250 | flags |= COMM_DOBIND_PORT_LATER; | |
251 | ||
c6f168c1 | 252 | return comm_openex(sock_type, proto, addr, flags, note); |
d6827718 | 253 | } |
254 | ||
e0d28505 AJ |
255 | void |
256 | comm_open_listener(int sock_type, | |
257 | int proto, | |
e02ed2e3 | 258 | Comm::ConnectionPointer &conn, |
e0d28505 AJ |
259 | const char *note) |
260 | { | |
261 | /* all listener sockets require bind() */ | |
262 | conn->flags |= COMM_DOBIND; | |
263 | ||
264 | /* attempt native enabled port. */ | |
c6f168c1 | 265 | conn->fd = comm_openex(sock_type, proto, conn->local, conn->flags, note); |
e0d28505 AJ |
266 | } |
267 | ||
31be869c AJ |
268 | int |
269 | comm_open_listener(int sock_type, | |
04f7fd38 | 270 | int proto, |
b7ac5457 | 271 | Ip::Address &addr, |
04f7fd38 AJ |
272 | int flags, |
273 | const char *note) | |
31be869c | 274 | { |
e02ed2e3 AJ |
275 | int sock = -1; |
276 | ||
ac760b5e AJ |
277 | /* all listener sockets require bind() */ |
278 | flags |= COMM_DOBIND; | |
279 | ||
e02ed2e3 | 280 | /* attempt native enabled port. */ |
c6f168c1 | 281 | sock = comm_openex(sock_type, proto, addr, flags, note); |
e02ed2e3 | 282 | |
31be869c AJ |
283 | return sock; |
284 | } | |
285 | ||
2d8c0b1a | 286 | static bool |
287 | limitError(int const anErrno) | |
288 | { | |
289 | return anErrno == ENFILE || anErrno == EMFILE; | |
290 | } | |
d6827718 | 291 | |
8b082ed9 | 292 | static void |
cc192b50 | 293 | comm_set_v6only(int fd, int tos) |
294 | { | |
295 | #ifdef IPV6_V6ONLY | |
296 | if (setsockopt(fd, IPPROTO_IPV6, IPV6_V6ONLY, (char *) &tos, sizeof(int)) < 0) { | |
b69e9ffa AJ |
297 | int xerrno = errno; |
298 | debugs(50, DBG_IMPORTANT, MYNAME << "setsockopt(IPV6_V6ONLY) " << (tos?"ON":"OFF") << " for FD " << fd << ": " << xstrerr(xerrno)); | |
cc192b50 | 299 | } |
300 | #else | |
b69e9ffa | 301 | debugs(50, DBG_CRITICAL, MYNAME << "WARNING: setsockopt(IPV6_V6ONLY) not supported on this platform"); |
cc192b50 | 302 | #endif /* sockopt */ |
303 | } | |
057f5854 | 304 | |
40d6264d | 305 | /** |
b2192042 AJ |
306 | * Set the socket option required for TPROXY spoofing for: |
307 | * - Linux TPROXY v4 support, | |
308 | * - OpenBSD divert-to support, | |
309 | * - FreeBSD IPFW TPROXY v4 support. | |
40d6264d | 310 | */ |
8b082ed9 | 311 | static void |
e950e673 | 312 | comm_set_transparent(int fd) |
f1e0717c | 313 | { |
b2192042 AJ |
314 | #if _SQUID_LINUX_ && defined(IP_TRANSPARENT) // Linux |
315 | # define soLevel SOL_IP | |
316 | # define soFlag IP_TRANSPARENT | |
d0e6f578 | 317 | bool doneSuid = false; |
b2192042 AJ |
318 | |
319 | #elif defined(SO_BINDANY) // OpenBSD 4.7+ and NetBSD with PF | |
320 | # define soLevel SOL_SOCKET | |
321 | # define soFlag SO_BINDANY | |
322 | enter_suid(); | |
d0e6f578 | 323 | bool doneSuid = true; |
1125ea7b | 324 | |
b2192042 AJ |
325 | #elif defined(IP_BINDANY) // FreeBSD with IPFW |
326 | # define soLevel IPPROTO_IP | |
327 | # define soFlag IP_BINDANY | |
1125ea7b | 328 | enter_suid(); |
d0e6f578 | 329 | bool doneSuid = true; |
b2192042 AJ |
330 | |
331 | #else | |
332 | debugs(50, DBG_CRITICAL, "WARNING: comm_open: setsockopt(TPROXY) not supported on this platform"); | |
c46e7feb | 333 | (void)fd; |
b2192042 AJ |
334 | #endif /* sockopt */ |
335 | ||
336 | #if defined(soLevel) && defined(soFlag) | |
337 | int tos = 1; | |
338 | if (setsockopt(fd, soLevel, soFlag, (char *) &tos, sizeof(int)) < 0) { | |
b69e9ffa AJ |
339 | int xerrno = errno; |
340 | debugs(50, DBG_IMPORTANT, MYNAME << "setsockopt(TPROXY) on FD " << fd << ": " << xstrerr(xerrno)); | |
1125ea7b MM |
341 | } else { |
342 | /* mark the socket as having transparent options */ | |
343 | fd_table[fd].flags.transparent = true; | |
344 | } | |
b2192042 AJ |
345 | if (doneSuid) |
346 | leave_suid(); | |
347 | #endif | |
f1e0717c AJ |
348 | } |
349 | ||
e1a88700 | 350 | /** |
351 | * Create a socket. Default is blocking, stream (TCP) socket. IO_TYPE | |
352 | * is OR of flags specified in defines.h:COMM_* | |
353 | */ | |
74d7d196 | 354 | static int |
d6827718 | 355 | comm_openex(int sock_type, |
62e76326 | 356 | int proto, |
e02ed2e3 | 357 | Ip::Address &addr, |
62e76326 | 358 | int flags, |
62e76326 | 359 | const char *note) |
090089c4 | 360 | { |
e02ed2e3 | 361 | int new_socket; |
aee3523a | 362 | struct addrinfo *AI = nullptr; |
090089c4 | 363 | |
364 | /* Create socket for accepting new connections. */ | |
95dc7ff4 | 365 | ++ statCounter.syscalls.sock.sockets; |
62e76326 | 366 | |
97bbba61 AR |
367 | if (!Ip::EnableIpv6 && addr.isIPv6()) { |
368 | debugs(50, 2, "refusing to open an IPv6 socket when IPv6 support is disabled: " << addr); | |
369 | errno = ENOTSUP; | |
370 | return -1; | |
371 | } | |
372 | ||
cc192b50 | 373 | /* Setup the socket addrinfo details for use */ |
e02ed2e3 | 374 | addr.getAddrInfo(AI); |
cc192b50 | 375 | AI->ai_socktype = sock_type; |
376 | AI->ai_protocol = proto; | |
cc192b50 | 377 | |
e02ed2e3 | 378 | debugs(50, 3, "comm_openex: Attempt open socket for: " << addr ); |
cc192b50 | 379 | |
e02ed2e3 | 380 | new_socket = socket(AI->ai_family, AI->ai_socktype, AI->ai_protocol); |
c3d6a21b | 381 | const auto firstErrNo = errno; |
055421ee | 382 | |
0eb08770 HN |
383 | /* under IPv6 there is the possibility IPv6 is present but disabled. */ |
384 | /* try again as IPv4-native if possible */ | |
e02ed2e3 | 385 | if ( new_socket < 0 && Ip::EnableIpv6 && addr.isIPv6() && addr.setIPv4() ) { |
0eb08770 | 386 | /* attempt to open this IPv4-only. */ |
851614a8 | 387 | Ip::Address::FreeAddr(AI); |
0eb08770 | 388 | /* Setup the socket addrinfo details for use */ |
e02ed2e3 | 389 | addr.getAddrInfo(AI); |
0eb08770 HN |
390 | AI->ai_socktype = sock_type; |
391 | AI->ai_protocol = proto; | |
b69e9ffa | 392 | debugs(50, 3, "Attempt fallback open socket for: " << addr ); |
e02ed2e3 | 393 | new_socket = socket(AI->ai_family, AI->ai_socktype, AI->ai_protocol); |
c3d6a21b AR |
394 | // TODO: Report failures of this second socket() call. |
395 | // if both socket() calls fail, we use firstErrNo | |
b69e9ffa | 396 | debugs(50, 2, "attempt open " << note << " socket on: " << addr); |
0eb08770 | 397 | } |
0eb08770 | 398 | |
e02ed2e3 | 399 | if (new_socket < 0) { |
62e76326 | 400 | /* Increase the number of reserved fd's if calls to socket() |
401 | * are failing because the open file table is full. This | |
402 | * limits the number of simultaneous clients */ | |
403 | ||
c3d6a21b AR |
404 | if (limitError(firstErrNo)) { |
405 | debugs(50, DBG_IMPORTANT, MYNAME << "socket failure: " << xstrerr(firstErrNo)); | |
62e76326 | 406 | fdAdjustReserved(); |
2d8c0b1a | 407 | } else { |
c3d6a21b | 408 | debugs(50, DBG_CRITICAL, MYNAME << "socket failure: " << xstrerr(firstErrNo)); |
62e76326 | 409 | } |
410 | ||
851614a8 | 411 | Ip::Address::FreeAddr(AI); |
cc192b50 | 412 | |
c3d6a21b | 413 | errno = firstErrNo; // restore for caller |
62e76326 | 414 | return -1; |
090089c4 | 415 | } |
62e76326 | 416 | |
e02ed2e3 AJ |
417 | // XXX: temporary for the transition. comm_openex will eventually have a conn to play with. |
418 | Comm::ConnectionPointer conn = new Comm::Connection; | |
419 | conn->local = addr; | |
420 | conn->fd = new_socket; | |
421 | ||
422 | debugs(50, 3, "comm_openex: Opened socket " << conn << " : family=" << AI->ai_family << ", type=" << AI->ai_socktype << ", protocol=" << AI->ai_protocol ); | |
cc192b50 | 423 | |
e02ed2e3 | 424 | if ( Ip::EnableIpv6&IPV6_SPECIAL_SPLITSTACK && addr.isIPv6() ) |
b5523edc | 425 | comm_set_v6only(conn->fd, 1); |
62e76326 | 426 | |
cc192b50 | 427 | /* Windows Vista supports Dual-Sockets. BUT defaults them to V6ONLY. Turn it OFF. */ |
9b1f7ee8 | 428 | /* Other OS may have this administratively disabled for general use. Same deal. */ |
e02ed2e3 | 429 | if ( Ip::EnableIpv6&IPV6_SPECIAL_V4MAPPING && addr.isIPv6() ) |
b5523edc | 430 | comm_set_v6only(conn->fd, 0); |
cc192b50 | 431 | |
c6f168c1 | 432 | comm_init_opened(conn, note, AI); |
e02ed2e3 | 433 | new_socket = comm_apply_flags(conn->fd, addr, flags, AI); |
38344a8e | 434 | |
851614a8 | 435 | Ip::Address::FreeAddr(AI); |
38344a8e | 436 | |
e02ed2e3 AJ |
437 | // XXX transition only. prevent conn from closing the new FD on function exit. |
438 | conn->fd = -1; | |
c3d6a21b AR |
439 | // XXX: firstErrNo is not applicable here -- socket() calls succeeded above! |
440 | // TODO: Stop reporting error codes via errno. | |
441 | errno = firstErrNo; | |
e02ed2e3 | 442 | return new_socket; |
38344a8e AR |
443 | } |
444 | ||
445 | /// update FD tables after a local or remote (IPC) comm_openex(); | |
446 | void | |
e0d28505 | 447 | comm_init_opened(const Comm::ConnectionPointer &conn, |
5667a628 AR |
448 | const char *note, |
449 | struct addrinfo *AI) | |
38344a8e | 450 | { |
e0d28505 | 451 | assert(Comm::IsConnOpen(conn)); |
38344a8e AR |
452 | assert(AI); |
453 | ||
090089c4 | 454 | /* update fdstat */ |
bf95c10a | 455 | debugs(5, 5, conn << " is a new socket"); |
62e76326 | 456 | |
e0d28505 AJ |
457 | assert(!isOpen(conn->fd)); // NP: global isOpen checks the fde entry for openness not the Comm::Connection |
458 | fd_open(conn->fd, FD_SOCKET, note); | |
62e76326 | 459 | |
e0d28505 AJ |
460 | fde *F = &fd_table[conn->fd]; |
461 | F->local_addr = conn->local; | |
cc192b50 | 462 | |
463 | F->sock_family = AI->ai_family; | |
38344a8e AR |
464 | } |
465 | ||
466 | /// apply flags after a local comm_open*() call; | |
467 | /// returns new_socket or -1 on error | |
468 | static int | |
469 | comm_apply_flags(int new_socket, | |
a67d2b2e | 470 | Ip::Address &addr, |
5667a628 AR |
471 | int flags, |
472 | struct addrinfo *AI) | |
38344a8e AR |
473 | { |
474 | assert(new_socket >= 0); | |
475 | assert(AI); | |
476 | const int sock_type = AI->ai_socktype; | |
62e76326 | 477 | |
79a15e0a | 478 | if (!(flags & COMM_NOCLOEXEC)) |
62e76326 | 479 | commSetCloseOnExec(new_socket); |
480 | ||
cdc33f35 | 481 | if ((flags & COMM_REUSEADDR)) |
62e76326 | 482 | commSetReuseAddr(new_socket); |
483 | ||
4dd643d5 | 484 | if (addr.port() > (unsigned short) 0) { |
7aa9bb3e | 485 | #if _SQUID_WINDOWS_ |
a50bfe93 | 486 | if (sock_type != SOCK_DGRAM) |
487 | #endif | |
a50bfe93 | 488 | commSetNoLinger(new_socket); |
62e76326 | 489 | |
490 | if (opt_reuseaddr) | |
491 | commSetReuseAddr(new_socket); | |
090089c4 | 492 | } |
62e76326 | 493 | |
a35595cd | 494 | /* MUST be done before binding or face OS Error: "(99) Cannot assign requested address"... */ |
9e008dda | 495 | if ((flags & COMM_TRANSPARENT)) { |
a35595cd AJ |
496 | comm_set_transparent(new_socket); |
497 | } | |
a35595cd | 498 | |
4dd643d5 AJ |
499 | if ( (flags & COMM_DOBIND) || addr.port() > 0 || !addr.isAnyAddr() ) { |
500 | if ( !(flags & COMM_DOBIND) && addr.isAnyAddr() ) | |
e0236918 | 501 | debugs(5, DBG_IMPORTANT,"WARNING: Squid is attempting to bind() port " << addr << " without being a listener."); |
4dd643d5 | 502 | if ( addr.isNoAddr() ) |
d816f28d | 503 | debugs(5, DBG_CRITICAL, "ERROR: Squid is attempting to bind() port " << addr << "!!"); |
ac760b5e | 504 | |
7d17a6a6 EB |
505 | #if defined(SO_REUSEPORT) |
506 | if (flags & COMM_REUSEPORT) { | |
507 | int on = 1; | |
508 | if (setsockopt(new_socket, SOL_SOCKET, SO_REUSEPORT, reinterpret_cast<char*>(&on), sizeof(on)) < 0) { | |
509 | const auto savedErrno = errno; | |
510 | const auto errorMessage = ToSBuf("cannot enable SO_REUSEPORT socket option when binding to ", | |
511 | addr, ": ", xstrerr(savedErrno)); | |
512 | if (reconfiguring) | |
513 | debugs(5, DBG_IMPORTANT, "ERROR: " << errorMessage); | |
514 | else | |
515 | throw TexcHere(errorMessage); | |
516 | } | |
517 | } | |
518 | #endif | |
74d7d196 | 519 | |
520 | if ((flags & COMM_DOBIND_PORT_LATER)) | |
521 | commSetBindAddressNoPort(new_socket); | |
522 | ||
c8407295 | 523 | if (commBind(new_socket, *AI) != Comm::OK) { |
62e76326 | 524 | comm_close(new_socket); |
525 | return -1; | |
62e76326 | 526 | } |
23ff6968 | 527 | } |
62e76326 | 528 | |
79a15e0a | 529 | if (flags & COMM_NONBLOCKING) |
4ee57cbe | 530 | if (commSetNonBlocking(new_socket) == Comm::COMM_ERROR) { |
38344a8e | 531 | comm_close(new_socket); |
62e76326 | 532 | return -1; |
62e76326 | 533 | } |
534 | ||
30a4f2a8 | 535 | #ifdef TCP_NODELAY |
536 | if (sock_type == SOCK_STREAM) | |
62e76326 | 537 | commSetTcpNoDelay(new_socket); |
538 | ||
30a4f2a8 | 539 | #endif |
62e76326 | 540 | |
1241e63e | 541 | if (Config.tcpRcvBufsz > 0 && sock_type == SOCK_STREAM) |
62e76326 | 542 | commSetTcpRcvbuf(new_socket, Config.tcpRcvBufsz); |
543 | ||
090089c4 | 544 | return new_socket; |
545 | } | |
546 | ||
a4c0f9c6 | 547 | void |
e0d28505 | 548 | comm_import_opened(const Comm::ConnectionPointer &conn, |
5667a628 AR |
549 | const char *note, |
550 | struct addrinfo *AI) | |
a4c0f9c6 | 551 | { |
bf95c10a | 552 | debugs(5, 2, conn); |
e0d28505 | 553 | assert(Comm::IsConnOpen(conn)); |
a4c0f9c6 AR |
554 | assert(AI); |
555 | ||
c6f168c1 | 556 | comm_init_opened(conn, note, AI); |
a4c0f9c6 | 557 | |
4dd643d5 | 558 | if (conn->local.port() > (unsigned short) 0) { |
7aa9bb3e | 559 | #if _SQUID_WINDOWS_ |
e0d28505 | 560 | if (AI->ai_socktype != SOCK_DGRAM) |
a4c0f9c6 | 561 | #endif |
be4d35dc | 562 | fd_table[conn->fd].flags.nolinger = true; |
a4c0f9c6 AR |
563 | } |
564 | ||
e0d28505 | 565 | if ((conn->flags & COMM_TRANSPARENT)) |
be4d35dc | 566 | fd_table[conn->fd].flags.transparent = true; |
a4c0f9c6 | 567 | |
e0d28505 | 568 | if (conn->flags & COMM_NONBLOCKING) |
be4d35dc | 569 | fd_table[conn->fd].flags.nonblocking = true; |
a4c0f9c6 AR |
570 | |
571 | #ifdef TCP_NODELAY | |
572 | if (AI->ai_socktype == SOCK_STREAM) | |
be4d35dc | 573 | fd_table[conn->fd].flags.nodelay = true; |
a4c0f9c6 AR |
574 | #endif |
575 | ||
576 | /* no fd_table[fd].flags. updates needed for these conditions: | |
577 | * if ((flags & COMM_REUSEADDR)) ... | |
578 | * if ((flags & COMM_DOBIND) ...) ... | |
579 | */ | |
580 | } | |
581 | ||
933dd095 AJ |
582 | // XXX: now that raw-FD timeouts are only unset for pipes and files this SHOULD be a no-op. |
583 | // With handler already unset. Leaving this present until that can be verified for all code paths. | |
584 | void | |
585 | commUnsetFdTimeout(int fd) | |
b0469965 | 586 | { |
bf95c10a | 587 | debugs(5, 3, "Remove timeout for FD " << fd); |
b0469965 | 588 | assert(fd >= 0); |
589 | assert(fd < Squid_MaxFD); | |
590 | fde *F = &fd_table[fd]; | |
591 | assert(F->flags.open); | |
592 | ||
aee3523a | 593 | F->timeoutHandler = nullptr; |
933dd095 | 594 | F->timeout = 0; |
7957e704 AJ |
595 | } |
596 | ||
597 | int | |
4650a4fa | 598 | commSetConnTimeout(const Comm::ConnectionPointer &conn, time_t timeout, AsyncCall::Pointer &callback) |
7957e704 | 599 | { |
bf95c10a | 600 | debugs(5, 3, conn << " timeout " << timeout); |
7957e704 AJ |
601 | assert(Comm::IsConnOpen(conn)); |
602 | assert(conn->fd < Squid_MaxFD); | |
603 | fde *F = &fd_table[conn->fd]; | |
604 | assert(F->flags.open); | |
605 | ||
606 | if (timeout < 0) { | |
aee3523a | 607 | F->timeoutHandler = nullptr; |
7957e704 AJ |
608 | F->timeout = 0; |
609 | } else { | |
aee3523a | 610 | if (callback != nullptr) { |
7957e704 AJ |
611 | typedef CommTimeoutCbParams Params; |
612 | Params ¶ms = GetCommParams<Params>(callback); | |
613 | params.conn = conn; | |
614 | F->timeoutHandler = callback; | |
615 | } | |
b0469965 | 616 | |
7957e704 AJ |
617 | F->timeout = squid_curtime + (time_t) timeout; |
618 | } | |
619 | ||
620 | return F->timeout; | |
b0469965 | 621 | } |
090089c4 | 622 | |
8d77a37c AJ |
623 | int |
624 | commUnsetConnTimeout(const Comm::ConnectionPointer &conn) | |
625 | { | |
bf95c10a | 626 | debugs(5, 3, "Remove timeout for " << conn); |
8d77a37c AJ |
627 | AsyncCall::Pointer nil; |
628 | return commSetConnTimeout(conn, -1, nil); | |
629 | } | |
630 | ||
8630961c AD |
631 | /** |
632 | * Connect socket FD to given remote address. | |
633 | * If return value is an error flag (COMM_ERROR, ERR_CONNECT, ERR_PROTOCOL, etc.), | |
634 | * then error code will also be returned in errno. | |
635 | */ | |
b8d8561b | 636 | int |
b7ac5457 | 637 | comm_connect_addr(int sock, const Ip::Address &address) |
090089c4 | 638 | { |
c8407295 | 639 | Comm::Flag status = Comm::OK; |
76f87348 | 640 | fde *F = &fd_table[sock]; |
cc192b50 | 641 | int x = 0; |
b5568a61 | 642 | int err = 0; |
9689d97c | 643 | socklen_t errlen; |
aee3523a | 644 | struct addrinfo *AI = nullptr; |
cc192b50 | 645 | |
4dd643d5 | 646 | assert(address.port() != 0); |
cc192b50 | 647 | |
bf95c10a | 648 | debugs(5, 9, "connecting socket FD " << sock << " to " << address << " (want family: " << F->sock_family << ")"); |
cc192b50 | 649 | |
3d98ff81 | 650 | /* Handle IPv6 over IPv4-only socket case. |
4dd643d5 | 651 | * this case must presently be handled here since the getAddrInfo asserts on bad mappings. |
3d98ff81 | 652 | * NP: because commResetFD is private to ConnStateData we have to return an error and |
9d92af86 AJ |
653 | * trust its handled properly. |
654 | */ | |
4dd643d5 | 655 | if (F->sock_family == AF_INET && !address.isIPv4()) { |
3d98ff81 | 656 | errno = ENETUNREACH; |
c8407295 | 657 | return Comm::ERR_PROTOCOL; |
3d98ff81 HN |
658 | } |
659 | ||
660 | /* Handle IPv4 over IPv6-only socket case. | |
661 | * This case is presently handled here as it's both a known case and it's | |
662 | * uncertain what error will be returned by the IPv6 stack in such case. It's | |
663 | * possible this will also be handled by the errno checks below after connect() | |
2f8abb64 | 664 | * but needs careful cross-platform verification, and verifying the address |
3d98ff81 HN |
665 | * condition here is simple. |
666 | */ | |
4dd643d5 | 667 | if (!F->local_addr.isIPv4() && address.isIPv4()) { |
3d98ff81 | 668 | errno = ENETUNREACH; |
c8407295 | 669 | return Comm::ERR_PROTOCOL; |
9d92af86 | 670 | } |
9d92af86 | 671 | |
4dd643d5 | 672 | address.getAddrInfo(AI, F->sock_family); |
cc192b50 | 673 | |
090089c4 | 674 | /* Establish connection. */ |
8630961c | 675 | int xerrno = 0; |
62e76326 | 676 | |
9e008dda | 677 | if (!F->flags.called_connect) { |
be4d35dc | 678 | F->flags.called_connect = true; |
95dc7ff4 | 679 | ++ statCounter.syscalls.sock.connects; |
62e76326 | 680 | |
07383e11 CT |
681 | errno = 0; |
682 | if ((x = connect(sock, AI->ai_addr, AI->ai_addrlen)) < 0) { | |
683 | xerrno = errno; | |
684 | debugs(5,5, "sock=" << sock << ", addrinfo(" << | |
9e008dda AJ |
685 | " flags=" << AI->ai_flags << |
686 | ", family=" << AI->ai_family << | |
687 | ", socktype=" << AI->ai_socktype << | |
688 | ", protocol=" << AI->ai_protocol << | |
689 | ", &addr=" << AI->ai_addr << | |
07383e11 | 690 | ", addrlen=" << AI->ai_addrlen << " )"); |
8630961c | 691 | debugs(5, 9, "connect FD " << sock << ": (" << x << ") " << xstrerr(xerrno)); |
07383e11 CT |
692 | debugs(14,9, "connecting to: " << address); |
693 | ||
694 | } else if (x == 0) { | |
695 | // XXX: ICAP code refuses callbacks during a pending comm_ call | |
696 | // Async calls development will fix this. | |
697 | x = -1; | |
698 | xerrno = EINPROGRESS; | |
cc192b50 | 699 | } |
8630961c | 700 | |
9e008dda | 701 | } else { |
8630961c | 702 | errno = 0; |
8a09e810 | 703 | #if _SQUID_NEWSOS6_ |
62e76326 | 704 | /* Makoto MATSUSHITA <matusita@ics.es.osaka-u.ac.jp> */ |
8630961c AD |
705 | if (connect(sock, AI->ai_addr, AI->ai_addrlen) < 0) |
706 | xerrno = errno; | |
62e76326 | 707 | |
8630961c | 708 | if (xerrno == EINVAL) { |
62e76326 | 709 | errlen = sizeof(err); |
710 | x = getsockopt(sock, SOL_SOCKET, SO_ERROR, &err, &errlen); | |
62e76326 | 711 | if (x >= 0) |
8630961c | 712 | xerrno = x; |
62e76326 | 713 | } |
33ac9442 | 714 | #else |
62e76326 | 715 | errlen = sizeof(err); |
62e76326 | 716 | x = getsockopt(sock, SOL_SOCKET, SO_ERROR, &err, &errlen); |
62e76326 | 717 | if (x == 0) |
8630961c | 718 | xerrno = err; |
62e76326 | 719 | |
8a09e810 | 720 | #if _SQUID_SOLARIS_ |
62e76326 | 721 | /* |
722 | * Solaris 2.4's socket emulation doesn't allow you | |
723 | * to determine the error from a failed non-blocking | |
724 | * connect and just returns EPIPE. Create a fake | |
725 | * error message for connect. -- fenner@parc.xerox.com | |
726 | */ | |
8630961c AD |
727 | if (x < 0 && xerrno == EPIPE) |
728 | xerrno = ENOTCONN; | |
729 | else | |
730 | xerrno = errno; | |
33ac9442 | 731 | #endif |
30a4f2a8 | 732 | #endif |
e5f6c5c2 | 733 | } |
62e76326 | 734 | |
851614a8 | 735 | Ip::Address::FreeAddr(AI); |
feca3b9a | 736 | |
8630961c AD |
737 | errno = xerrno; |
738 | if (xerrno == 0 || xerrno == EISCONN) | |
c8407295 | 739 | status = Comm::OK; |
8630961c | 740 | else if (ignoreErrno(xerrno)) |
c8407295 | 741 | status = Comm::INPROGRESS; |
8630961c | 742 | else if (xerrno == EAFNOSUPPORT || xerrno == EINVAL) |
c8407295 | 743 | return Comm::ERR_PROTOCOL; |
b5568a61 | 744 | else |
4ee57cbe | 745 | return Comm::COMM_ERROR; |
62e76326 | 746 | |
4dd643d5 | 747 | address.toStr(F->ipaddr, MAX_IPSTRLEN); |
62e76326 | 748 | |
4dd643d5 | 749 | F->remote_port = address.port(); /* remote_port is HS */ |
62e76326 | 750 | |
c8407295 | 751 | if (status == Comm::OK) { |
e0236918 | 752 | debugs(5, DBG_DATA, "comm_connect_addr: FD " << sock << " connected to " << address); |
c8407295 | 753 | } else if (status == Comm::INPROGRESS) { |
e0236918 | 754 | debugs(5, DBG_DATA, "comm_connect_addr: FD " << sock << " connection pending"); |
090089c4 | 755 | } |
62e76326 | 756 | |
8630961c | 757 | errno = xerrno; |
090089c4 | 758 | return status; |
759 | } | |
760 | ||
cb201b7e | 761 | void |
762 | commCallCloseHandlers(int fd) | |
763 | { | |
76f87348 | 764 | fde *F = &fd_table[fd]; |
bf8fe701 | 765 | debugs(5, 5, "commCallCloseHandlers: FD " << fd); |
62e76326 | 766 | |
aee3523a | 767 | while (F->closeHandler != nullptr) { |
b0469965 | 768 | AsyncCall::Pointer call = F->closeHandler; |
9e008dda | 769 | F->closeHandler = call->Next(); |
aee3523a | 770 | call->setNext(nullptr); |
9e008dda AJ |
771 | // If call is not canceled schedule it for execution else ignore it |
772 | if (!call->canceled()) { | |
773 | debugs(5, 5, "commCallCloseHandlers: ch->handler=" << call); | |
2b6b1bcb AR |
774 | // XXX: Without the following code, callback fd may be -1. |
775 | // typedef CommCloseCbParams Params; | |
776 | // auto ¶ms = GetCommParams<Params>(call); | |
777 | // params.fd = fd; | |
9e008dda AJ |
778 | ScheduleCallHere(call); |
779 | } | |
cb201b7e | 780 | } |
781 | } | |
782 | ||
27774cee | 783 | /** |
98264874 | 784 | * enable linger with time of 0 so that when the socket is |
785 | * closed, TCP generates a RESET | |
786 | */ | |
787 | void | |
8ace824c | 788 | comm_reset_close(const Comm::ConnectionPointer &conn) |
98264874 | 789 | { |
5c336a3b AJ |
790 | struct linger L; |
791 | L.l_onoff = 1; | |
792 | L.l_linger = 0; | |
62e76326 | 793 | |
b69e9ffa AJ |
794 | if (setsockopt(conn->fd, SOL_SOCKET, SO_LINGER, (char *) &L, sizeof(L)) < 0) { |
795 | int xerrno = errno; | |
796 | debugs(50, DBG_CRITICAL, "ERROR: Closing " << conn << " with TCP RST: " << xstrerr(xerrno)); | |
797 | } | |
5c336a3b AJ |
798 | conn->close(); |
799 | } | |
800 | ||
801 | // Legacy close function. | |
802 | void | |
803 | old_comm_reset_close(int fd) | |
804 | { | |
98264874 | 805 | struct linger L; |
806 | L.l_onoff = 1; | |
807 | L.l_linger = 0; | |
62e76326 | 808 | |
b69e9ffa AJ |
809 | if (setsockopt(fd, SOL_SOCKET, SO_LINGER, (char *) &L, sizeof(L)) < 0) { |
810 | int xerrno = errno; | |
811 | debugs(50, DBG_CRITICAL, "ERROR: Closing FD " << fd << " with TCP RST: " << xstrerr(xerrno)); | |
812 | } | |
98264874 | 813 | comm_close(fd); |
814 | } | |
815 | ||
8b082ed9 | 816 | static void |
2949dd02 | 817 | commStartTlsClose(const int fd) |
10b06767 | 818 | { |
2949dd02 | 819 | Security::SessionSendGoodbye(fd_table[fd].ssl); |
10b06767 AJ |
820 | } |
821 | ||
8b082ed9 | 822 | static void |
2949dd02 | 823 | comm_close_complete(const int fd) |
2d8c0b1a | 824 | { |
2949dd02 | 825 | auto F = &fd_table[fd]; |
1ca8bbfb | 826 | F->ssl.reset(); |
0476ec45 | 827 | F->dynamicTlsContext.reset(); |
2949dd02 EB |
828 | fd_close(fd); /* update fdstat */ |
829 | close(fd); | |
b0469965 | 830 | |
95dc7ff4 | 831 | ++ statCounter.syscalls.sock.closes; |
b0469965 | 832 | |
575d05c4 | 833 | /* When one connection closes, give accept() a chance, if need be */ |
ccfbe8f4 | 834 | CodeContext::Reset(); // exit FD-specific context |
04f55905 | 835 | Comm::AcceptLimiter::Instance().kick(); |
2d8c0b1a | 836 | } |
c4b7a5a9 | 837 | |
838 | /* | |
839 | * Close the socket fd. | |
840 | * | |
841 | * + call write handlers with ERR_CLOSING | |
842 | * + call read handlers with ERR_CLOSING | |
843 | * + call closing handlers | |
a46d2c0e | 844 | * |
a928fdfd EB |
845 | * A deferred reader has no Comm read handler mentioned above. To stay in sync, |
846 | * such a reader must register a Comm closing handler. | |
c4b7a5a9 | 847 | */ |
b8d8561b | 848 | void |
43ae1d95 | 849 | _comm_close(int fd, char const *file, int line) |
090089c4 | 850 | { |
4a5d9ea5 | 851 | debugs(5, 3, "start closing FD " << fd << " by " << file << ":" << line); |
03eb2f01 | 852 | assert(fd >= 0); |
853 | assert(fd < Squid_MaxFD); | |
82ec8dfc AR |
854 | |
855 | fde *F = &fd_table[fd]; | |
1f7c9178 | 856 | |
82ec8dfc | 857 | if (F->closing()) |
62e76326 | 858 | return; |
859 | ||
36afac26 | 860 | /* XXX: is this obsolete behind F->closing() ? */ |
b8869bcf | 861 | if ( (shutting_down || reconfiguring) && (!F->flags.open || F->type == FD_FILE)) |
62e76326 | 862 | return; |
863 | ||
c4b7a5a9 | 864 | /* The following fails because ipc.c is doing calls to pipe() to create sockets! */ |
c7e637d7 | 865 | if (!isOpen(fd)) { |
d816f28d | 866 | debugs(50, DBG_IMPORTANT, "ERROR: Squid BUG #3556: FD " << fd << " is not an open socket."); |
c7e637d7 AJ |
867 | // XXX: do we need to run close(fd) or fd_close(fd) here? |
868 | return; | |
869 | } | |
62e76326 | 870 | |
76f87348 | 871 | assert(F->type != FD_FILE); |
62e76326 | 872 | |
be4d35dc | 873 | F->flags.close_request = true; |
62e76326 | 874 | |
ccfbe8f4 AR |
875 | // We have caller's context and fde::codeContext. In the unlikely event they |
876 | // differ, it is not clear which context is more applicable to this closure. | |
877 | // For simplicity sake, we remain in the caller's context while still | |
878 | // allowing individual advanced callbacks to overwrite it. | |
879 | ||
575d05c4 | 880 | if (F->ssl) { |
2949dd02 EB |
881 | const auto startCall = asyncCall(5, 4, "commStartTlsClose", |
882 | callDialer(commStartTlsClose, fd)); | |
575d05c4 AJ |
883 | ScheduleCallHere(startCall); |
884 | } | |
62e76326 | 885 | |
74257126 AR |
886 | // a half-closed fd may lack a reader, so we stop monitoring explicitly |
887 | if (commHasHalfClosedMonitor(fd)) | |
888 | commStopHalfClosedMonitor(fd); | |
933dd095 | 889 | commUnsetFdTimeout(fd); |
62e76326 | 890 | |
a6351f16 | 891 | // notify read/write handlers after canceling select reservations, if any |
ec41b64c | 892 | if (COMMIO_FD_WRITECB(fd)->active()) { |
aee3523a | 893 | Comm::SetSelect(fd, COMM_SELECT_WRITE, nullptr, nullptr, 0); |
c3d6a21b | 894 | COMMIO_FD_WRITECB(fd)->finish(Comm::ERR_CLOSING, 0); |
2b663917 | 895 | } |
ec41b64c | 896 | if (COMMIO_FD_READCB(fd)->active()) { |
aee3523a | 897 | Comm::SetSelect(fd, COMM_SELECT_READ, nullptr, nullptr, 0); |
c3d6a21b | 898 | COMMIO_FD_READCB(fd)->finish(Comm::ERR_CLOSING, 0); |
2b663917 | 899 | } |
2d8c0b1a | 900 | |
9a0a18de | 901 | #if USE_DELAY_POOLS |
b27668ec EB |
902 | if (BandwidthBucket *bucket = BandwidthBucket::SelectBucket(F)) { |
903 | if (bucket->selectWaiting) | |
904 | bucket->onFdClosed(); | |
f33d34a8 | 905 | } |
b4cd430a CT |
906 | #endif |
907 | ||
cb201b7e | 908 | commCallCloseHandlers(fd); |
62e76326 | 909 | |
a7ad6e4e | 910 | comm_empty_os_read_buffers(fd); |
9e008dda | 911 | |
9e008dda | 912 | // must use async call to wait for all callbacks |
82ec8dfc | 913 | // scheduled before comm_close() to finish |
2949dd02 EB |
914 | const auto completeCall = asyncCall(5, 4, "comm_close_complete", |
915 | callDialer(comm_close_complete, fd)); | |
10b06767 | 916 | ScheduleCallHere(completeCall); |
090089c4 | 917 | } |
918 | ||
090089c4 | 919 | /* Send a udp datagram to specified TO_ADDR. */ |
b8d8561b | 920 | int |
5df61230 | 921 | comm_udp_sendto(int fd, |
b7ac5457 | 922 | const Ip::Address &to_addr, |
62e76326 | 923 | const void *buf, |
924 | int len) | |
090089c4 | 925 | { |
95dc7ff4 | 926 | ++ statCounter.syscalls.sock.sendtos; |
62e76326 | 927 | |
cc192b50 | 928 | debugs(50, 3, "comm_udp_sendto: Attempt to send UDP packet to " << to_addr << |
9e008dda | 929 | " using FD " << fd << " using Port " << comm_local_port(fd) ); |
cc192b50 | 930 | |
aee3523a | 931 | struct addrinfo *AI = nullptr; |
4dd643d5 AJ |
932 | to_addr.getAddrInfo(AI, fd_table[fd].sock_family); |
933 | int x = sendto(fd, buf, len, 0, AI->ai_addr, AI->ai_addrlen); | |
b69e9ffa | 934 | int xerrno = errno; |
851614a8 | 935 | Ip::Address::FreeAddr(AI); |
cc192b50 | 936 | |
b69e9ffa AJ |
937 | if (x >= 0) { |
938 | errno = xerrno; // restore for caller to use | |
2d8c0b1a | 939 | return x; |
b69e9ffa | 940 | } |
2d8c0b1a | 941 | |
1191b93b | 942 | #if _SQUID_LINUX_ |
b69e9ffa | 943 | if (ECONNREFUSED != xerrno) |
17d51783 | 944 | #endif |
b69e9ffa | 945 | debugs(50, DBG_IMPORTANT, MYNAME << "FD " << fd << ", (family=" << fd_table[fd].sock_family << ") " << to_addr << ": " << xstrerr(xerrno)); |
62e76326 | 946 | |
b69e9ffa | 947 | errno = xerrno; // restore for caller to use |
4ee57cbe | 948 | return Comm::COMM_ERROR; |
090089c4 | 949 | } |
950 | ||
398bc066 | 951 | AsyncCall::Pointer |
575d05c4 | 952 | comm_add_close_handler(int fd, CLCB * handler, void *data) |
30a4f2a8 | 953 | { |
bf8fe701 | 954 | debugs(5, 5, "comm_add_close_handler: FD " << fd << ", handler=" << |
955 | handler << ", data=" << data); | |
62e76326 | 956 | |
b0469965 | 957 | AsyncCall::Pointer call=commCbCall(5,4, "SomeCloseHandler", |
9e008dda | 958 | CommCloseCbPtrFun(handler, data)); |
b0469965 | 959 | comm_add_close_handler(fd, call); |
398bc066 | 960 | return call; |
b0469965 | 961 | } |
62e76326 | 962 | |
b0469965 | 963 | void |
964 | comm_add_close_handler(int fd, AsyncCall::Pointer &call) | |
965 | { | |
966 | debugs(5, 5, "comm_add_close_handler: FD " << fd << ", AsyncCall=" << call); | |
62e76326 | 967 | |
b0469965 | 968 | /*TODO:Check for a similar scheduled AsyncCall*/ |
969 | // for (c = fd_table[fd].closeHandler; c; c = c->next) | |
970 | // assert(c->handler != handler || c->data != data); | |
62e76326 | 971 | |
a928fdfd EB |
972 | // TODO: Consider enhancing AsyncCallList to support random-access close |
973 | // handlers, perhaps after upgrading the remaining legacy CLCB handlers. | |
b0469965 | 974 | call->setNext(fd_table[fd].closeHandler); |
62e76326 | 975 | |
b0469965 | 976 | fd_table[fd].closeHandler = call; |
30a4f2a8 | 977 | } |
978 | ||
b0469965 | 979 | // remove function-based close handler |
b8d8561b | 980 | void |
575d05c4 | 981 | comm_remove_close_handler(int fd, CLCB * handler, void *data) |
090089c4 | 982 | { |
8ebef9e0 | 983 | assert(isOpen(fd)); |
30a4f2a8 | 984 | /* Find handler in list */ |
bf8fe701 | 985 | debugs(5, 5, "comm_remove_close_handler: FD " << fd << ", handler=" << |
986 | handler << ", data=" << data); | |
62e76326 | 987 | |
aee3523a AR |
988 | AsyncCall::Pointer p, prev = nullptr; |
989 | for (p = fd_table[fd].closeHandler; p != nullptr; prev = p, p = p->Next()) { | |
b0469965 | 990 | typedef CommCbFunPtrCallT<CommCloseCbPtrFun> Call; |
991 | const Call *call = dynamic_cast<const Call*>(p.getRaw()); | |
992 | if (!call) // method callbacks have their own comm_remove_close_handler | |
993 | continue; | |
62e76326 | 994 | |
b0469965 | 995 | typedef CommCloseCbParams Params; |
996 | const Params ¶ms = GetCommParams<Params>(p); | |
997 | if (call->dialer.handler == handler && params.data == data) | |
f53969cc | 998 | break; /* This is our handler */ |
b0469965 | 999 | } |
7828df5b CT |
1000 | |
1001 | // comm_close removes all close handlers so our handler may be gone | |
aee3523a | 1002 | if (p != nullptr) { |
37cba319 | 1003 | p->dequeue(fd_table[fd].closeHandler, prev); |
7828df5b | 1004 | p->cancel("comm_remove_close_handler"); |
37cba319 | 1005 | } |
b0469965 | 1006 | } |
62e76326 | 1007 | |
b0469965 | 1008 | // remove method-based close handler |
1009 | void | |
1010 | comm_remove_close_handler(int fd, AsyncCall::Pointer &call) | |
1011 | { | |
8ebef9e0 | 1012 | assert(isOpen(fd)); |
b0469965 | 1013 | debugs(5, 5, "comm_remove_close_handler: FD " << fd << ", AsyncCall=" << call); |
62e76326 | 1014 | |
7828df5b | 1015 | // comm_close removes all close handlers so our handler may be gone |
aee3523a AR |
1016 | AsyncCall::Pointer p, prev = nullptr; |
1017 | for (p = fd_table[fd].closeHandler; p != nullptr && p != call; prev = p, p = p->Next()); | |
62e76326 | 1018 | |
aee3523a | 1019 | if (p != nullptr) |
37cba319 | 1020 | p->dequeue(fd_table[fd].closeHandler, prev); |
b0469965 | 1021 | call->cancel("comm_remove_close_handler"); |
30a4f2a8 | 1022 | } |
090089c4 | 1023 | |
b8d8561b | 1024 | static void |
1025 | commSetNoLinger(int fd) | |
30a4f2a8 | 1026 | { |
62e76326 | 1027 | |
30a4f2a8 | 1028 | struct linger L; |
f53969cc | 1029 | L.l_onoff = 0; /* off */ |
090089c4 | 1030 | L.l_linger = 0; |
62e76326 | 1031 | |
b69e9ffa AJ |
1032 | if (setsockopt(fd, SOL_SOCKET, SO_LINGER, (char *) &L, sizeof(L)) < 0) { |
1033 | int xerrno = errno; | |
1034 | debugs(50, DBG_CRITICAL, MYNAME << "FD " << fd << ": " << xstrerr(xerrno)); | |
1035 | } | |
be4d35dc | 1036 | fd_table[fd].flags.nolinger = true; |
090089c4 | 1037 | } |
1038 | ||
b8d8561b | 1039 | static void |
1040 | commSetReuseAddr(int fd) | |
090089c4 | 1041 | { |
1042 | int on = 1; | |
b69e9ffa AJ |
1043 | if (setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, (char *) &on, sizeof(on)) < 0) { |
1044 | int xerrno = errno; | |
1045 | debugs(50, DBG_IMPORTANT, MYNAME << "FD " << fd << ": " << xstrerr(xerrno)); | |
1046 | } | |
090089c4 | 1047 | } |
1048 | ||
b8d8561b | 1049 | static void |
1050 | commSetTcpRcvbuf(int fd, int size) | |
f868539a | 1051 | { |
b69e9ffa AJ |
1052 | if (setsockopt(fd, SOL_SOCKET, SO_RCVBUF, (char *) &size, sizeof(size)) < 0) { |
1053 | int xerrno = errno; | |
1054 | debugs(50, DBG_IMPORTANT, MYNAME << "FD " << fd << ", SIZE " << size << ": " << xstrerr(xerrno)); | |
1055 | } | |
1056 | if (setsockopt(fd, SOL_SOCKET, SO_SNDBUF, (char *) &size, sizeof(size)) < 0) { | |
1057 | int xerrno = errno; | |
1058 | debugs(50, DBG_IMPORTANT, MYNAME << "FD " << fd << ", SIZE " << size << ": " << xstrerr(xerrno)); | |
1059 | } | |
8f0d53ef | 1060 | #ifdef TCP_WINDOW_CLAMP |
b69e9ffa AJ |
1061 | if (setsockopt(fd, SOL_TCP, TCP_WINDOW_CLAMP, (char *) &size, sizeof(size)) < 0) { |
1062 | int xerrno = errno; | |
1063 | debugs(50, DBG_IMPORTANT, MYNAME << "FD " << fd << ", SIZE " << size << ": " << xstrerr(xerrno)); | |
1064 | } | |
8f0d53ef | 1065 | #endif |
f868539a | 1066 | } |
1067 | ||
b8d8561b | 1068 | int |
1069 | commSetNonBlocking(int fd) | |
30a4f2a8 | 1070 | { |
be266cb2 | 1071 | #if _SQUID_WINDOWS_ |
b05490a8 | 1072 | int nonblocking = TRUE; |
62e76326 | 1073 | |
a396d1f8 | 1074 | if (ioctl(fd, FIONBIO, &nonblocking) < 0) { |
b69e9ffa AJ |
1075 | int xerrno = errno; |
1076 | debugs(50, DBG_CRITICAL, MYNAME << "FD " << fd << ": " << xstrerr(xerrno) << " " << fd_table[fd].type); | |
a396d1f8 EA |
1077 | return Comm::COMM_ERROR; |
1078 | } | |
62e76326 | 1079 | |
a396d1f8 EA |
1080 | #else |
1081 | int flags; | |
1082 | int dummy = 0; | |
62e76326 | 1083 | |
a396d1f8 | 1084 | if ((flags = fcntl(fd, F_GETFL, dummy)) < 0) { |
b69e9ffa AJ |
1085 | int xerrno = errno; |
1086 | debugs(50, DBG_CRITICAL, MYNAME << "FD " << fd << ": fcntl F_GETFL: " << xstrerr(xerrno)); | |
a396d1f8 EA |
1087 | return Comm::COMM_ERROR; |
1088 | } | |
62e76326 | 1089 | |
a396d1f8 | 1090 | if (fcntl(fd, F_SETFL, flags | SQUID_NONBLOCK) < 0) { |
b69e9ffa AJ |
1091 | int xerrno = errno; |
1092 | debugs(50, DBG_CRITICAL, MYNAME << "FD " << fd << ": " << xstrerr(xerrno)); | |
a396d1f8 | 1093 | return Comm::COMM_ERROR; |
090089c4 | 1094 | } |
7f6ffd15 | 1095 | #endif |
62e76326 | 1096 | |
a396d1f8 | 1097 | fd_table[fd].flags.nonblocking = true; |
090089c4 | 1098 | return 0; |
1099 | } | |
1100 | ||
7e3ce7b9 | 1101 | int |
1102 | commUnsetNonBlocking(int fd) | |
1103 | { | |
7aa9bb3e | 1104 | #if _SQUID_WINDOWS_ |
a50bfe93 | 1105 | int nonblocking = FALSE; |
1106 | ||
1107 | if (ioctlsocket(fd, FIONBIO, (unsigned long *) &nonblocking) < 0) { | |
1108 | #else | |
7e3ce7b9 | 1109 | int flags; |
1110 | int dummy = 0; | |
62e76326 | 1111 | |
7e3ce7b9 | 1112 | if ((flags = fcntl(fd, F_GETFL, dummy)) < 0) { |
b69e9ffa AJ |
1113 | int xerrno = errno; |
1114 | debugs(50, DBG_CRITICAL, MYNAME << "FD " << fd << ": fcntl F_GETFL: " << xstrerr(xerrno)); | |
4ee57cbe | 1115 | return Comm::COMM_ERROR; |
7e3ce7b9 | 1116 | } |
62e76326 | 1117 | |
7e3ce7b9 | 1118 | if (fcntl(fd, F_SETFL, flags & (~SQUID_NONBLOCK)) < 0) { |
a50bfe93 | 1119 | #endif |
b69e9ffa AJ |
1120 | int xerrno = errno; |
1121 | debugs(50, DBG_CRITICAL, MYNAME << "FD " << fd << ": " << xstrerr(xerrno)); | |
4ee57cbe | 1122 | return Comm::COMM_ERROR; |
7e3ce7b9 | 1123 | } |
62e76326 | 1124 | |
be4d35dc | 1125 | fd_table[fd].flags.nonblocking = false; |
7e3ce7b9 | 1126 | return 0; |
1127 | } | |
1128 | ||
b8d8561b | 1129 | void |
e1381638 AJ |
1130 | commSetCloseOnExec(int fd) |
1131 | { | |
3ca60c86 | 1132 | #ifdef FD_CLOEXEC |
731e4d49 | 1133 | int flags; |
7a18b487 | 1134 | int dummy = 0; |
62e76326 | 1135 | |
2209fe19 | 1136 | if ((flags = fcntl(fd, F_GETFD, dummy)) < 0) { |
b69e9ffa AJ |
1137 | int xerrno = errno; |
1138 | debugs(50, DBG_CRITICAL, MYNAME << "FD " << fd << ": fcntl F_GETFD: " << xstrerr(xerrno)); | |
62e76326 | 1139 | return; |
3ca60c86 | 1140 | } |
62e76326 | 1141 | |
b69e9ffa AJ |
1142 | if (fcntl(fd, F_SETFD, flags | FD_CLOEXEC) < 0) { |
1143 | int xerrno = errno; | |
d816f28d | 1144 | debugs(50, DBG_CRITICAL, "ERROR: " << MYNAME << "FD " << fd << ": set close-on-exec failed: " << xstrerr(xerrno)); |
b69e9ffa | 1145 | } |
3ca60c86 | 1146 | #endif |
1147 | } | |
1148 | ||
e90100aa | 1149 | #ifdef TCP_NODELAY |
1150 | static void | |
e1381638 AJ |
1151 | commSetTcpNoDelay(int fd) |
1152 | { | |
e90100aa | 1153 | int on = 1; |
62e76326 | 1154 | |
b69e9ffa AJ |
1155 | if (setsockopt(fd, IPPROTO_TCP, TCP_NODELAY, (char *) &on, sizeof(on)) < 0) { |
1156 | int xerrno = errno; | |
1157 | debugs(50, DBG_IMPORTANT, MYNAME << "FD " << fd << ": " << xstrerr(xerrno)); | |
1158 | } | |
62e76326 | 1159 | |
be4d35dc | 1160 | fd_table[fd].flags.nodelay = true; |
e90100aa | 1161 | } |
62e76326 | 1162 | |
e90100aa | 1163 | #endif |
1164 | ||
d86b3703 | 1165 | void |
e1381638 AJ |
1166 | comm_init(void) |
1167 | { | |
d7ca82e6 | 1168 | assert(fd_table); |
2d8c0b1a | 1169 | |
04f55905 AJ |
1170 | /* make sure the accept() socket FIFO delay queue exists */ |
1171 | Comm::AcceptLimiter::Instance(); | |
b0469965 | 1172 | |
ec41b64c AJ |
1173 | // make sure the IO pending callback table exists |
1174 | Comm::CallbackTableInit(); | |
2d8c0b1a | 1175 | |
59c4d35b | 1176 | /* XXX account fd_table */ |
090089c4 | 1177 | /* Keep a few file descriptors free so that we don't run out of FD's |
1178 | * after accepting a client but before it opens a socket or a file. | |
e83892e9 | 1179 | * Since Squid_MaxFD can be as high as several thousand, don't waste them */ |
d85c3078 | 1180 | RESERVED_FD = min(100, Squid_MaxFD / 4); |
2d8c0b1a | 1181 | |
74257126 | 1182 | TheHalfClosed = new DescriptorSet; |
d841c88d AJ |
1183 | |
1184 | /* setup the select loop module */ | |
1185 | Comm::SelectLoopInit(); | |
090089c4 | 1186 | } |
1187 | ||
236d1779 | 1188 | void |
e1381638 AJ |
1189 | comm_exit(void) |
1190 | { | |
74257126 | 1191 | delete TheHalfClosed; |
aee3523a | 1192 | TheHalfClosed = nullptr; |
74257126 | 1193 | |
ec41b64c | 1194 | Comm::CallbackTableDestruct(); |
236d1779 | 1195 | } |
1196 | ||
9a0a18de | 1197 | #if USE_DELAY_POOLS |
b4cd430a | 1198 | // called when the queue is done waiting for the client bucket to fill |
ec41b64c | 1199 | void |
b4cd430a CT |
1200 | commHandleWriteHelper(void * data) |
1201 | { | |
1202 | CommQuotaQueue *queue = static_cast<CommQuotaQueue*>(data); | |
1203 | assert(queue); | |
1204 | ||
1205 | ClientInfo *clientInfo = queue->clientInfo; | |
1206 | // ClientInfo invalidates queue if freed, so if we got here through, | |
1207 | // evenAdd cbdata protections, everything should be valid and consistent | |
f33d34a8 | 1208 | assert(clientInfo); |
b4cd430a CT |
1209 | assert(clientInfo->hasQueue()); |
1210 | assert(clientInfo->hasQueue(queue)); | |
b4cd430a CT |
1211 | assert(clientInfo->eventWaiting); |
1212 | clientInfo->eventWaiting = false; | |
1213 | ||
1214 | do { | |
b90a3414 EB |
1215 | clientInfo->writeOrDequeue(); |
1216 | if (clientInfo->selectWaiting) | |
1217 | return; | |
1218 | } while (clientInfo->hasQueue()); | |
b4cd430a | 1219 | |
b90a3414 EB |
1220 | debugs(77, 3, "emptied queue"); |
1221 | } | |
1222 | ||
1223 | void | |
1224 | ClientInfo::writeOrDequeue() | |
1225 | { | |
1226 | assert(!selectWaiting); | |
1227 | const auto head = quotaPeekFd(); | |
1228 | const auto &headFde = fd_table[head]; | |
1229 | CallBack(headFde.codeContext, [&] { | |
1230 | const auto ccb = COMMIO_FD_WRITECB(head); | |
1231 | // check that the head descriptor is still relevant | |
1232 | if (headFde.clientInfo == this && | |
b6388dfd AJ |
1233 | quotaPeekReserv() == ccb->quotaQueueReserv && |
1234 | !headFde.closing()) { | |
b4cd430a CT |
1235 | |
1236 | // wait for the head descriptor to become ready for writing | |
d841c88d | 1237 | Comm::SetSelect(head, COMM_SELECT_WRITE, Comm::HandleWrite, ccb, 0); |
b90a3414 EB |
1238 | selectWaiting = true; |
1239 | } else { | |
1240 | quotaDequeue(); // remove the no longer relevant descriptor | |
f33d34a8 | 1241 | } |
b90a3414 | 1242 | }); |
b4cd430a CT |
1243 | } |
1244 | ||
1245 | bool | |
1246 | ClientInfo::hasQueue() const | |
1247 | { | |
1248 | assert(quotaQueue); | |
1249 | return !quotaQueue->empty(); | |
1250 | } | |
1251 | ||
1252 | bool | |
1253 | ClientInfo::hasQueue(const CommQuotaQueue *q) const | |
1254 | { | |
1255 | assert(quotaQueue); | |
1256 | return quotaQueue == q; | |
1257 | } | |
1258 | ||
1259 | /// returns the first descriptor to be dequeued | |
1260 | int | |
1261 | ClientInfo::quotaPeekFd() const | |
1262 | { | |
1263 | assert(quotaQueue); | |
1264 | return quotaQueue->front(); | |
1265 | } | |
1266 | ||
1267 | /// returns the reservation ID of the first descriptor to be dequeued | |
1268 | unsigned int | |
1269 | ClientInfo::quotaPeekReserv() const | |
1270 | { | |
1271 | assert(quotaQueue); | |
1272 | return quotaQueue->outs + 1; | |
1273 | } | |
1274 | ||
1275 | /// queues a given fd, creating the queue if necessary; returns reservation ID | |
1276 | unsigned int | |
1277 | ClientInfo::quotaEnqueue(int fd) | |
1278 | { | |
1279 | assert(quotaQueue); | |
1280 | return quotaQueue->enqueue(fd); | |
1281 | } | |
1282 | ||
1283 | /// removes queue head | |
1284 | void | |
1285 | ClientInfo::quotaDequeue() | |
1286 | { | |
1287 | assert(quotaQueue); | |
1288 | quotaQueue->dequeue(); | |
1289 | } | |
1290 | ||
1291 | void | |
1292 | ClientInfo::kickQuotaQueue() | |
1293 | { | |
1294 | if (!eventWaiting && !selectWaiting && hasQueue()) { | |
1295 | // wait at least a second if the bucket is empty | |
b27668ec | 1296 | const double delay = (bucketLevel < 1.0) ? 1.0 : 0.0; |
b4cd430a | 1297 | eventAdd("commHandleWriteHelper", &commHandleWriteHelper, |
f33d34a8 | 1298 | quotaQueue, delay, 0, true); |
b4cd430a CT |
1299 | eventWaiting = true; |
1300 | } | |
1301 | } | |
1302 | ||
1303 | /// calculates how much to write for a single dequeued client | |
1304 | int | |
b27668ec | 1305 | ClientInfo::quota() |
b4cd430a CT |
1306 | { |
1307 | /* If we have multiple clients and give full bucketSize to each client then | |
1308 | * clt1 may often get a lot more because clt1->clt2 time distance in the | |
1309 | * select(2) callback order may be a lot smaller than cltN->clt1 distance. | |
1310 | * We divide quota evenly to be more fair. */ | |
1311 | ||
1312 | if (!rationedCount) { | |
1313 | rationedCount = quotaQueue->size() + 1; | |
1314 | ||
1315 | // The delay in ration recalculation _temporary_ deprives clients from | |
1316 | // bytes that should have trickled in while rationedCount was positive. | |
1317 | refillBucket(); | |
1318 | ||
1319 | // Rounding errors do not accumulate here, but we round down to avoid | |
1320 | // negative bucket sizes after write with rationedCount=1. | |
b27668ec | 1321 | rationedQuota = static_cast<int>(floor(bucketLevel/rationedCount)); |
bf95c10a | 1322 | debugs(77,5, "new rationedQuota: " << rationedQuota << |
f33d34a8 | 1323 | '*' << rationedCount); |
b4cd430a CT |
1324 | } |
1325 | ||
1326 | --rationedCount; | |
bf95c10a | 1327 | debugs(77,7, "rationedQuota: " << rationedQuota << |
f33d34a8 | 1328 | " rations remaining: " << rationedCount); |
b4cd430a CT |
1329 | |
1330 | // update 'last seen' time to prevent clientdb GC from dropping us | |
1331 | last_seen = squid_curtime; | |
1332 | return rationedQuota; | |
1333 | } | |
1334 | ||
b27668ec EB |
1335 | bool |
1336 | ClientInfo::applyQuota(int &nleft, Comm::IoCallback *state) | |
1337 | { | |
1338 | assert(hasQueue()); | |
1339 | assert(quotaPeekFd() == state->conn->fd); | |
1340 | quotaDequeue(); // we will write or requeue below | |
1341 | if (nleft > 0 && !BandwidthBucket::applyQuota(nleft, state)) { | |
1342 | state->quotaQueueReserv = quotaEnqueue(state->conn->fd); | |
1343 | kickQuotaQueue(); | |
1344 | return false; | |
1345 | } | |
1346 | return true; | |
1347 | } | |
1348 | ||
b4cd430a | 1349 | void |
b27668ec | 1350 | ClientInfo::scheduleWrite(Comm::IoCallback *state) |
b4cd430a | 1351 | { |
b27668ec EB |
1352 | if (writeLimitingActive) { |
1353 | state->quotaQueueReserv = quotaEnqueue(state->conn->fd); | |
1354 | kickQuotaQueue(); | |
b4cd430a | 1355 | } |
b27668ec | 1356 | } |
b4cd430a | 1357 | |
b27668ec EB |
1358 | void |
1359 | ClientInfo::onFdClosed() | |
1360 | { | |
1361 | BandwidthBucket::onFdClosed(); | |
1362 | // kick queue or it will get stuck as commWriteHandle is not called | |
1363 | kickQuotaQueue(); | |
1364 | } | |
b4cd430a | 1365 | |
b27668ec EB |
1366 | void |
1367 | ClientInfo::reduceBucket(const int len) | |
1368 | { | |
1369 | if (len > 0) | |
1370 | BandwidthBucket::reduceBucket(len); | |
1371 | // even if we wrote nothing, we were served; give others a chance | |
1372 | kickQuotaQueue(); | |
b4cd430a CT |
1373 | } |
1374 | ||
f33d34a8 | 1375 | void |
b4cd430a CT |
1376 | ClientInfo::setWriteLimiter(const int aWriteSpeedLimit, const double anInitialBurst, const double aHighWatermark) |
1377 | { | |
b27668ec | 1378 | debugs(77,5, "Write limits for " << (const char*)key << |
f33d34a8 A |
1379 | " speed=" << aWriteSpeedLimit << " burst=" << anInitialBurst << |
1380 | " highwatermark=" << aHighWatermark); | |
b4cd430a CT |
1381 | |
1382 | // set or possibly update traffic shaping parameters | |
1383 | writeLimitingActive = true; | |
1384 | writeSpeedLimit = aWriteSpeedLimit; | |
1385 | bucketSizeLimit = aHighWatermark; | |
1386 | ||
1387 | // but some members should only be set once for a newly activated bucket | |
1388 | if (firstTimeConnection) { | |
1389 | firstTimeConnection = false; | |
1390 | ||
1391 | assert(!selectWaiting); | |
1392 | assert(!quotaQueue); | |
e9dadd7d | 1393 | quotaQueue = new CommQuotaQueue(this); |
b4cd430a | 1394 | |
b27668ec | 1395 | bucketLevel = anInitialBurst; |
b4cd430a CT |
1396 | prevTime = current_dtime; |
1397 | } | |
1398 | } | |
1399 | ||
1400 | CommQuotaQueue::CommQuotaQueue(ClientInfo *info): clientInfo(info), | |
f53969cc | 1401 | ins(0), outs(0) |
b4cd430a CT |
1402 | { |
1403 | assert(clientInfo); | |
1404 | } | |
1405 | ||
1406 | CommQuotaQueue::~CommQuotaQueue() | |
1407 | { | |
1408 | assert(!clientInfo); // ClientInfo should clear this before destroying us | |
1409 | } | |
1410 | ||
1411 | /// places the given fd at the end of the queue; returns reservation ID | |
1412 | unsigned int | |
1413 | CommQuotaQueue::enqueue(int fd) | |
1414 | { | |
b27668ec | 1415 | debugs(77,5, "clt" << (const char*)clientInfo->key << |
f33d34a8 | 1416 | ": FD " << fd << " with qqid" << (ins+1) << ' ' << fds.size()); |
b4cd430a | 1417 | fds.push_back(fd); |
b90a3414 | 1418 | fd_table[fd].codeContext = CodeContext::Current(); |
b4cd430a CT |
1419 | return ++ins; |
1420 | } | |
1421 | ||
1422 | /// removes queue head | |
1423 | void | |
1424 | CommQuotaQueue::dequeue() | |
1425 | { | |
1426 | assert(!fds.empty()); | |
b27668ec | 1427 | debugs(77,5, "clt" << (const char*)clientInfo->key << |
f33d34a8 A |
1428 | ": FD " << fds.front() << " with qqid" << (outs+1) << ' ' << |
1429 | fds.size()); | |
b4cd430a CT |
1430 | fds.pop_front(); |
1431 | ++outs; | |
1432 | } | |
b27668ec | 1433 | #endif /* USE_DELAY_POOLS */ |
b4cd430a | 1434 | |
89924214 | 1435 | /* |
1436 | * hm, this might be too general-purpose for all the places we'd | |
1437 | * like to use it. | |
1438 | */ | |
b224ea98 | 1439 | int |
e1381638 AJ |
1440 | ignoreErrno(int ierrno) |
1441 | { | |
603500e7 | 1442 | switch (ierrno) { |
62e76326 | 1443 | |
89924214 | 1444 | case EINPROGRESS: |
62e76326 | 1445 | |
603500e7 | 1446 | case EWOULDBLOCK: |
26a880e2 | 1447 | #if EAGAIN != EWOULDBLOCK |
62e76326 | 1448 | |
603500e7 | 1449 | case EAGAIN: |
26a880e2 | 1450 | #endif |
62e76326 | 1451 | |
603500e7 | 1452 | case EALREADY: |
62e76326 | 1453 | |
603500e7 | 1454 | case EINTR: |
db494ab8 | 1455 | #ifdef ERESTART |
62e76326 | 1456 | |
db494ab8 | 1457 | case ERESTART: |
1458 | #endif | |
62e76326 | 1459 | |
1460 | return 1; | |
1461 | ||
603500e7 | 1462 | default: |
62e76326 | 1463 | return 0; |
603500e7 | 1464 | } |
62e76326 | 1465 | |
603500e7 | 1466 | /* NOTREACHED */ |
26a880e2 | 1467 | } |
d723bf6b | 1468 | |
1469 | void | |
e1381638 AJ |
1470 | commCloseAllSockets(void) |
1471 | { | |
d723bf6b | 1472 | int fd; |
aee3523a | 1473 | fde *F = nullptr; |
62e76326 | 1474 | |
95dc7ff4 | 1475 | for (fd = 0; fd <= Biggest_FD; ++fd) { |
62e76326 | 1476 | F = &fd_table[fd]; |
1477 | ||
1478 | if (!F->flags.open) | |
1479 | continue; | |
1480 | ||
1481 | if (F->type != FD_SOCKET) | |
1482 | continue; | |
1483 | ||
f53969cc | 1484 | if (F->flags.ipc) /* don't close inter-process sockets */ |
62e76326 | 1485 | continue; |
1486 | ||
aee3523a | 1487 | if (F->timeoutHandler != nullptr) { |
b0469965 | 1488 | AsyncCall::Pointer callback = F->timeoutHandler; |
aee3523a | 1489 | F->timeoutHandler = nullptr; |
bf8fe701 | 1490 | debugs(5, 5, "commCloseAllSockets: FD " << fd << ": Calling timeout handler"); |
9e008dda | 1491 | ScheduleCallHere(callback); |
62e76326 | 1492 | } else { |
468fe1b5 | 1493 | debugs(5, 5, "commCloseAllSockets: FD " << fd << ": calling comm_reset_close()"); |
5c336a3b | 1494 | old_comm_reset_close(fd); |
62e76326 | 1495 | } |
d723bf6b | 1496 | } |
1497 | } | |
1b3db6d9 | 1498 | |
2d8c0b1a | 1499 | static bool |
e1381638 AJ |
1500 | AlreadyTimedOut(fde *F) |
1501 | { | |
2d8c0b1a | 1502 | if (!F->flags.open) |
1503 | return true; | |
1504 | ||
1505 | if (F->timeout == 0) | |
1506 | return true; | |
1507 | ||
1508 | if (F->timeout > squid_curtime) | |
1509 | return true; | |
1510 | ||
1511 | return false; | |
1512 | } | |
1513 | ||
5ef5e5cc AJ |
1514 | static bool |
1515 | writeTimedOut(int fd) | |
1516 | { | |
ec41b64c | 1517 | if (!COMMIO_FD_WRITECB(fd)->active()) |
5ef5e5cc AJ |
1518 | return false; |
1519 | ||
1520 | if ((squid_curtime - fd_table[fd].writeStart) < Config.Timeout.write) | |
1521 | return false; | |
1522 | ||
1523 | return true; | |
1524 | } | |
1525 | ||
1b3db6d9 | 1526 | void |
e1381638 AJ |
1527 | checkTimeouts(void) |
1528 | { | |
1b3db6d9 | 1529 | int fd; |
aee3523a | 1530 | fde *F = nullptr; |
b0469965 | 1531 | AsyncCall::Pointer callback; |
62e76326 | 1532 | |
95dc7ff4 | 1533 | for (fd = 0; fd <= Biggest_FD; ++fd) { |
62e76326 | 1534 | F = &fd_table[fd]; |
1535 | ||
5ef5e5cc AJ |
1536 | if (writeTimedOut(fd)) { |
1537 | // We have an active write callback and we are timed out | |
ccfbe8f4 | 1538 | CodeContext::Reset(F->codeContext); |
52f6ea9e | 1539 | debugs(5, 5, "checkTimeouts: FD " << fd << " auto write timeout"); |
aee3523a | 1540 | Comm::SetSelect(fd, COMM_SELECT_WRITE, nullptr, nullptr, 0); |
4ee57cbe | 1541 | COMMIO_FD_WRITECB(fd)->finish(Comm::COMM_ERROR, ETIMEDOUT); |
ccfbe8f4 | 1542 | CodeContext::Reset(); |
db43f179 | 1543 | continue; |
b27668ec EB |
1544 | #if USE_DELAY_POOLS |
1545 | } else if (F->writeQuotaHandler != nullptr && COMMIO_FD_WRITECB(fd)->conn != nullptr) { | |
ccfbe8f4 | 1546 | // TODO: Move and extract quota() call to place it inside F->codeContext. |
b27668ec | 1547 | if (!F->writeQuotaHandler->selectWaiting && F->writeQuotaHandler->quota() && !F->closing()) { |
ccfbe8f4 | 1548 | CodeContext::Reset(F->codeContext); |
b27668ec EB |
1549 | F->writeQuotaHandler->selectWaiting = true; |
1550 | Comm::SetSelect(fd, COMM_SELECT_WRITE, Comm::HandleWrite, COMMIO_FD_WRITECB(fd), 0); | |
ccfbe8f4 | 1551 | CodeContext::Reset(); |
b27668ec EB |
1552 | } |
1553 | continue; | |
1554 | #endif | |
1555 | } | |
1556 | else if (AlreadyTimedOut(F)) | |
62e76326 | 1557 | continue; |
1558 | ||
ccfbe8f4 | 1559 | CodeContext::Reset(F->codeContext); |
9e008dda | 1560 | debugs(5, 5, "checkTimeouts: FD " << fd << " Expired"); |
62e76326 | 1561 | |
aee3523a | 1562 | if (F->timeoutHandler != nullptr) { |
bf8fe701 | 1563 | debugs(5, 5, "checkTimeouts: FD " << fd << ": Call timeout handler"); |
b0469965 | 1564 | callback = F->timeoutHandler; |
aee3523a | 1565 | F->timeoutHandler = nullptr; |
9e008dda | 1566 | ScheduleCallHere(callback); |
62e76326 | 1567 | } else { |
bf8fe701 | 1568 | debugs(5, 5, "checkTimeouts: FD " << fd << ": Forcing comm_close()"); |
62e76326 | 1569 | comm_close(fd); |
1570 | } | |
ccfbe8f4 AR |
1571 | |
1572 | CodeContext::Reset(); | |
b5443c04 | 1573 | } |
1574 | } | |
1575 | ||
9e008dda AJ |
1576 | /// Start waiting for a possibly half-closed connection to close |
1577 | // by scheduling a read callback to a monitoring handler that | |
82ec8dfc | 1578 | // will close the connection on read errors. |
a46d2c0e | 1579 | void |
e1381638 AJ |
1580 | commStartHalfClosedMonitor(int fd) |
1581 | { | |
bf95c10a | 1582 | debugs(5, 5, "adding FD " << fd << " to " << *TheHalfClosed); |
8ebef9e0 | 1583 | assert(isOpen(fd) && !commHasHalfClosedMonitor(fd)); |
74257126 | 1584 | (void)TheHalfClosed->add(fd); // could also assert the result |
1f9077c2 | 1585 | fd_table[fd].codeContext = CodeContext::Current(); |
74257126 AR |
1586 | commPlanHalfClosedCheck(); // may schedule check if we added the first FD |
1587 | } | |
1588 | ||
1589 | static | |
1590 | void | |
e1381638 AJ |
1591 | commPlanHalfClosedCheck() |
1592 | { | |
74257126 | 1593 | if (!WillCheckHalfClosed && !TheHalfClosed->empty()) { |
aee3523a | 1594 | eventAdd("commHalfClosedCheck", &commHalfClosedCheck, nullptr, 1.0, 1); |
74257126 AR |
1595 | WillCheckHalfClosed = true; |
1596 | } | |
1597 | } | |
1598 | ||
1599 | /// iterates over all descriptors that may need half-closed tests and | |
1600 | /// calls comm_read for those that do; re-schedules the check if needed | |
1601 | static | |
1602 | void | |
e1381638 AJ |
1603 | commHalfClosedCheck(void *) |
1604 | { | |
bf95c10a | 1605 | debugs(5, 5, "checking " << *TheHalfClosed); |
74257126 AR |
1606 | |
1607 | typedef DescriptorSet::const_iterator DSCI; | |
1608 | const DSCI end = TheHalfClosed->end(); | |
1609 | for (DSCI i = TheHalfClosed->begin(); i != end; ++i) { | |
ec20038e AJ |
1610 | Comm::ConnectionPointer c = new Comm::Connection; // XXX: temporary. make HalfClosed a list of these. |
1611 | c->fd = *i; | |
1612 | if (!fd_table[c->fd].halfClosedReader) { // not reading already | |
1f9077c2 EB |
1613 | CallBack(fd_table[c->fd].codeContext, [&c] { |
1614 | AsyncCall::Pointer call = commCbCall(5,4, "commHalfClosedReader", | |
b6388dfd | 1615 | CommIoCbPtrFun(&commHalfClosedReader, nullptr)); |
1f9077c2 EB |
1616 | Comm::Read(c, call); |
1617 | fd_table[c->fd].halfClosedReader = call; | |
1618 | }); | |
ec20038e AJ |
1619 | } else |
1620 | c->fd = -1; // XXX: temporary. prevent c replacement erase closing listed FD | |
74257126 | 1621 | } |
f900210a | 1622 | |
74257126 AR |
1623 | WillCheckHalfClosed = false; // as far as we know |
1624 | commPlanHalfClosedCheck(); // may need to check again | |
f900210a | 1625 | } |
1626 | ||
82ec8dfc AR |
1627 | /// checks whether we are waiting for possibly half-closed connection to close |
1628 | // We are monitoring if the read handler for the fd is the monitoring handler. | |
1629 | bool | |
e1381638 AJ |
1630 | commHasHalfClosedMonitor(int fd) |
1631 | { | |
74257126 | 1632 | return TheHalfClosed->has(fd); |
a46d2c0e | 1633 | } |
1634 | ||
82ec8dfc | 1635 | /// stop waiting for possibly half-closed connection to close |
7e66d5e2 | 1636 | void |
e1381638 AJ |
1637 | commStopHalfClosedMonitor(int const fd) |
1638 | { | |
bf95c10a | 1639 | debugs(5, 5, "removing FD " << fd << " from " << *TheHalfClosed); |
74257126 AR |
1640 | |
1641 | // cancel the read if one was scheduled | |
1642 | AsyncCall::Pointer reader = fd_table[fd].halfClosedReader; | |
aee3523a | 1643 | if (reader != nullptr) |
0d4e382b | 1644 | Comm::ReadCancel(fd, reader); |
aee3523a | 1645 | fd_table[fd].halfClosedReader = nullptr; |
74257126 AR |
1646 | |
1647 | TheHalfClosed->del(fd); | |
a46d2c0e | 1648 | } |
1649 | ||
82ec8dfc AR |
1650 | /// I/O handler for the possibly half-closed connection monitoring code |
1651 | static void | |
c8407295 | 1652 | commHalfClosedReader(const Comm::ConnectionPointer &conn, char *, size_t size, Comm::Flag flag, int, void *) |
e1381638 | 1653 | { |
82ec8dfc | 1654 | // there cannot be more data coming in on half-closed connections |
9e008dda | 1655 | assert(size == 0); |
aee3523a | 1656 | assert(conn != nullptr); |
e0d28505 | 1657 | assert(commHasHalfClosedMonitor(conn->fd)); // or we would have canceled the read |
74257126 | 1658 | |
aee3523a | 1659 | fd_table[conn->fd].halfClosedReader = nullptr; // done reading, for now |
a46d2c0e | 1660 | |
82ec8dfc | 1661 | // nothing to do if fd is being closed |
c8407295 | 1662 | if (flag == Comm::ERR_CLOSING) |
82ec8dfc | 1663 | return; |
a46d2c0e | 1664 | |
82ec8dfc | 1665 | // if read failed, close the connection |
c8407295 | 1666 | if (flag != Comm::OK) { |
bf95c10a | 1667 | debugs(5, 3, "closing " << conn); |
80463bb4 | 1668 | conn->close(); |
82ec8dfc AR |
1669 | return; |
1670 | } | |
a46d2c0e | 1671 | |
82ec8dfc | 1672 | // continue waiting for close or error |
74257126 | 1673 | commPlanHalfClosedCheck(); // make sure this fd will be checked again |
a46d2c0e | 1674 | } |
a46d2c0e | 1675 | |
8ff3fa2e | 1676 | int |
e1381638 AJ |
1677 | CommSelectEngine::checkEvents(int timeout) |
1678 | { | |
fa3f745b | 1679 | static time_t last_timeout = 0; |
1680 | ||
1681 | /* No, this shouldn't be here. But it shouldn't be in each comm handler. -adrian */ | |
1682 | if (squid_curtime > last_timeout) { | |
1683 | last_timeout = squid_curtime; | |
1684 | checkTimeouts(); | |
1685 | } | |
1686 | ||
d841c88d | 1687 | switch (Comm::DoSelect(timeout)) { |
8ff3fa2e | 1688 | |
c8407295 | 1689 | case Comm::OK: |
8ff3fa2e | 1690 | |
c8407295 | 1691 | case Comm::TIMEOUT: |
8ff3fa2e | 1692 | return 0; |
1693 | ||
c8407295 | 1694 | case Comm::IDLE: |
8ff3fa2e | 1695 | |
c8407295 | 1696 | case Comm::SHUTDOWN: |
8ff3fa2e | 1697 | return EVENT_IDLE; |
1698 | ||
4ee57cbe | 1699 | case Comm::COMM_ERROR: |
8ff3fa2e | 1700 | return EVENT_ERROR; |
1701 | ||
1702 | default: | |
1703 | fatal_dump("comm.cc: Internal error -- this should never happen."); | |
1704 | return EVENT_ERROR; | |
1705 | }; | |
1706 | } | |
10cefb7b | 1707 | |
0ffda73c | 1708 | /// Create a unix-domain socket (UDS) that only supports FD_MSGHDR I/O. |
10cefb7b | 1709 | int |
1710 | comm_open_uds(int sock_type, | |
1711 | int proto, | |
1712 | struct sockaddr_un* addr, | |
1713 | int flags) | |
1714 | { | |
a67d2b2e | 1715 | // TODO: merge with comm_openex() when Ip::Address becomes NetAddress |
ba568924 | 1716 | |
10cefb7b | 1717 | int new_socket; |
10cefb7b | 1718 | |
10cefb7b | 1719 | /* Create socket for accepting new connections. */ |
95dc7ff4 | 1720 | ++ statCounter.syscalls.sock.sockets; |
10cefb7b | 1721 | |
1722 | /* Setup the socket addrinfo details for use */ | |
ba568924 | 1723 | struct addrinfo AI; |
10cefb7b | 1724 | AI.ai_flags = 0; |
1725 | AI.ai_family = PF_UNIX; | |
1726 | AI.ai_socktype = sock_type; | |
1727 | AI.ai_protocol = proto; | |
1728 | AI.ai_addrlen = SUN_LEN(addr); | |
1729 | AI.ai_addr = (sockaddr*)addr; | |
aee3523a AR |
1730 | AI.ai_canonname = nullptr; |
1731 | AI.ai_next = nullptr; | |
10cefb7b | 1732 | |
bf95c10a | 1733 | debugs(50, 3, "Attempt open socket for: " << addr->sun_path); |
10cefb7b | 1734 | |
1735 | if ((new_socket = socket(AI.ai_family, AI.ai_socktype, AI.ai_protocol)) < 0) { | |
b69e9ffa | 1736 | int xerrno = errno; |
10cefb7b | 1737 | /* Increase the number of reserved fd's if calls to socket() |
1738 | * are failing because the open file table is full. This | |
1739 | * limits the number of simultaneous clients */ | |
1740 | ||
b69e9ffa AJ |
1741 | if (limitError(xerrno)) { |
1742 | debugs(50, DBG_IMPORTANT, MYNAME << "socket failure: " << xstrerr(xerrno)); | |
10cefb7b | 1743 | fdAdjustReserved(); |
1744 | } else { | |
b69e9ffa | 1745 | debugs(50, DBG_CRITICAL, MYNAME << "socket failure: " << xstrerr(xerrno)); |
10cefb7b | 1746 | } |
10cefb7b | 1747 | return -1; |
1748 | } | |
1749 | ||
fc9d2eb0 | 1750 | debugs(50, 3, "Opened UDS FD " << new_socket << " : family=" << AI.ai_family << ", type=" << AI.ai_socktype << ", protocol=" << AI.ai_protocol); |
10cefb7b | 1751 | |
1752 | /* update fdstat */ | |
bf95c10a | 1753 | debugs(50, 5, "FD " << new_socket << " is a new socket"); |
10cefb7b | 1754 | |
1755 | assert(!isOpen(new_socket)); | |
04efe9dd | 1756 | fd_open(new_socket, FD_MSGHDR, addr->sun_path); |
10cefb7b | 1757 | |
ba568924 AR |
1758 | fd_table[new_socket].sock_family = AI.ai_family; |
1759 | ||
10cefb7b | 1760 | if (!(flags & COMM_NOCLOEXEC)) |
1761 | commSetCloseOnExec(new_socket); | |
1762 | ||
1763 | if (flags & COMM_REUSEADDR) | |
1764 | commSetReuseAddr(new_socket); | |
1765 | ||
1766 | if (flags & COMM_NONBLOCKING) { | |
c8407295 | 1767 | if (commSetNonBlocking(new_socket) != Comm::OK) { |
10cefb7b | 1768 | comm_close(new_socket); |
10cefb7b | 1769 | return -1; |
1770 | } | |
1771 | } | |
1772 | ||
1773 | if (flags & COMM_DOBIND) { | |
c8407295 | 1774 | if (commBind(new_socket, AI) != Comm::OK) { |
10cefb7b | 1775 | comm_close(new_socket); |
10cefb7b | 1776 | return -1; |
1777 | } | |
1778 | } | |
1779 | ||
1780 | #ifdef TCP_NODELAY | |
1781 | if (sock_type == SOCK_STREAM) | |
1782 | commSetTcpNoDelay(new_socket); | |
1783 | ||
1784 | #endif | |
1785 | ||
1786 | if (Config.tcpRcvBufsz > 0 && sock_type == SOCK_STREAM) | |
1787 | commSetTcpRcvbuf(new_socket, Config.tcpRcvBufsz); | |
1788 | ||
10cefb7b | 1789 | return new_socket; |
1790 | } | |
f53969cc | 1791 |