]>
Commit | Line | Data |
---|---|---|
30a4f2a8 | 1 | /* |
b510f3a1 | 2 | * DEBUG: section 05 Socket Functions |
30a4f2a8 | 3 | * AUTHOR: Harvest Derived |
4 | * | |
2b6662ba | 5 | * SQUID Web Proxy Cache http://www.squid-cache.org/ |
e25c139f | 6 | * ---------------------------------------------------------- |
30a4f2a8 | 7 | * |
2b6662ba | 8 | * Squid is the result of efforts by numerous individuals from |
9 | * the Internet community; see the CONTRIBUTORS file for full | |
10 | * details. Many organizations have provided support for Squid's | |
11 | * development; see the SPONSORS file for full details. Squid is | |
12 | * Copyrighted (C) 2001 by the Regents of the University of | |
13 | * California; see the COPYRIGHT file for full details. Squid | |
14 | * incorporates software developed and/or copyrighted by other | |
15 | * sources; see the CREDITS file for full details. | |
30a4f2a8 | 16 | * |
17 | * This program is free software; you can redistribute it and/or modify | |
18 | * it under the terms of the GNU General Public License as published by | |
19 | * the Free Software Foundation; either version 2 of the License, or | |
20 | * (at your option) any later version. | |
9e008dda | 21 | * |
30a4f2a8 | 22 | * This program is distributed in the hope that it will be useful, |
23 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
24 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
25 | * GNU General Public License for more details. | |
9e008dda | 26 | * |
30a4f2a8 | 27 | * You should have received a copy of the GNU General Public License |
28 | * along with this program; if not, write to the Free Software | |
cbdec147 | 29 | * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111, USA. |
e25c139f | 30 | * |
2d8c0b1a | 31 | * |
32 | * Copyright (c) 2003, Robert Collins <robertc@squid-cache.org> | |
30a4f2a8 | 33 | */ |
090089c4 | 34 | |
f7f3304a | 35 | #include "squid-old.h" |
cfd66529 | 36 | #include "base/AsyncCall.h" |
c4b7a5a9 | 37 | #include "StoreIOBuffer.h" |
38 | #include "comm.h" | |
a553a5a3 | 39 | #include "event.h" |
528b2c61 | 40 | #include "fde.h" |
04f55905 AJ |
41 | #include "comm/AcceptLimiter.h" |
42 | #include "comm/comm_internal.h" | |
cfd66529 | 43 | #include "comm/Connection.h" |
ec41b64c | 44 | #include "comm/IoCallback.h" |
d841c88d | 45 | #include "comm/Loops.h" |
ec41b64c | 46 | #include "comm/Write.h" |
cbff89ba | 47 | #include "comm/TcpAcceptor.h" |
a553a5a3 | 48 | #include "CommRead.h" |
0eb49b6d | 49 | #include "MemBuf.h" |
781ce8ff | 50 | #include "pconn.h" |
985c86bc | 51 | #include "SquidTime.h" |
b0469965 | 52 | #include "CommCalls.h" |
74257126 | 53 | #include "DescriptorSet.h" |
9b5c4a9a | 54 | #include "icmp/net_db.h" |
96d89ea0 HN |
55 | #include "ip/Address.h" |
56 | #include "ip/Intercept.h" | |
425de4c8 | 57 | #include "ip/QosConfig.h" |
055421ee | 58 | #include "ip/tools.h" |
b4cd430a | 59 | #include "ClientInfo.h" |
e4f1fdae | 60 | #include "StatCounters.h" |
4db984be CT |
61 | #if USE_SSL |
62 | #include "ssl/support.h" | |
63 | #endif | |
090089c4 | 64 | |
b4cd430a | 65 | #include "cbdata.h" |
be266cb2 | 66 | #if _SQUID_CYGWIN_ |
b671cc68 | 67 | #include <sys/ioctl.h> |
68 | #endif | |
30a4f2a8 | 69 | #ifdef HAVE_NETINET_TCP_H |
70 | #include <netinet/tcp.h> | |
71 | #endif | |
090089c4 | 72 | |
2b663917 | 73 | /* |
74 | * New C-like simple comm code. This stuff is a mess and doesn't really buy us anything. | |
75 | */ | |
76 | ||
82ec8dfc AR |
77 | static void commStopHalfClosedMonitor(int fd); |
78 | static IOCB commHalfClosedReader; | |
9c8a6c3b | 79 | static void comm_init_opened(const Comm::ConnectionPointer &conn, tos_t tos, nfmark_t nfmark, const char *note, struct addrinfo *AI); |
a67d2b2e | 80 | static int comm_apply_flags(int new_socket, Ip::Address &addr, int flags, struct addrinfo *AI); |
82ec8dfc | 81 | |
9a0a18de | 82 | #if USE_DELAY_POOLS |
b4cd430a CT |
83 | CBDATA_CLASS_INIT(CommQuotaQueue); |
84 | ||
85 | static void commHandleWriteHelper(void * data); | |
86 | #endif | |
87 | ||
090089c4 | 88 | /* STATIC */ |
62e76326 | 89 | |
74257126 AR |
90 | static DescriptorSet *TheHalfClosed = NULL; /// the set of half-closed FDs |
91 | static bool WillCheckHalfClosed = false; /// true if check is scheduled | |
92 | static EVH commHalfClosedCheck; | |
93 | static void commPlanHalfClosedCheck(); | |
94 | ||
cc192b50 | 95 | static comm_err_t commBind(int s, struct addrinfo &); |
f5b8bbc4 | 96 | static void commSetReuseAddr(int); |
97 | static void commSetNoLinger(int); | |
30a4f2a8 | 98 | #ifdef TCP_NODELAY |
f5b8bbc4 | 99 | static void commSetTcpNoDelay(int); |
30a4f2a8 | 100 | #endif |
f5b8bbc4 | 101 | static void commSetTcpRcvbuf(int, int); |
723123a9 | 102 | |
b001e822 | 103 | static MemAllocator *conn_close_pool = NULL; |
c4b7a5a9 | 104 | fd_debug_t *fdd_table = NULL; |
62e76326 | 105 | |
04f55905 | 106 | bool |
b0469965 | 107 | isOpen(const int fd) |
b300c36d | 108 | { |
9a5ac93d | 109 | return fd >= 0 && fd_table && fd_table[fd].flags.open != 0; |
b300c36d | 110 | } |
111 | ||
e1a88700 | 112 | /** |
c4b7a5a9 | 113 | * Attempt a read |
114 | * | |
115 | * If the read attempt succeeds or fails, call the callback. | |
116 | * Else, wait for another IO notification. | |
117 | */ | |
2d8c0b1a | 118 | void |
2b663917 | 119 | commHandleRead(int fd, void *data) |
2d8c0b1a | 120 | { |
ec41b64c | 121 | Comm::IoCallback *ccb = (Comm::IoCallback *) data; |
9e008dda | 122 | |
2b663917 | 123 | assert(data == COMMIO_FD_READCB(fd)); |
ec41b64c | 124 | assert(ccb->active()); |
62e76326 | 125 | /* Attempt a read */ |
95dc7ff4 | 126 | ++ statCounter.syscalls.sock.reads; |
62e76326 | 127 | errno = 0; |
2d8c0b1a | 128 | int retval; |
2b663917 | 129 | retval = FD_READ_METHOD(fd, ccb->buf, ccb->size); |
bf8fe701 | 130 | debugs(5, 3, "comm_read_try: FD " << fd << ", size " << ccb->size << ", retval " << retval << ", errno " << errno); |
62e76326 | 131 | |
132 | if (retval < 0 && !ignoreErrno(errno)) { | |
bf8fe701 | 133 | debugs(5, 3, "comm_read_try: scheduling COMM_ERROR"); |
9e008dda | 134 | ccb->offset = 0; |
ec41b64c | 135 | ccb->finish(COMM_ERROR, errno); |
62e76326 | 136 | return; |
137 | }; | |
138 | ||
139 | /* See if we read anything */ | |
140 | /* Note - read 0 == socket EOF, which is a valid read */ | |
141 | if (retval >= 0) { | |
142 | fd_bytes(fd, retval, FD_READ); | |
9e008dda | 143 | ccb->offset = retval; |
ec41b64c | 144 | ccb->finish(COMM_OK, errno); |
62e76326 | 145 | return; |
146 | } | |
c4b7a5a9 | 147 | |
62e76326 | 148 | /* Nope, register for some more IO */ |
d841c88d | 149 | Comm::SetSelect(fd, COMM_SELECT_READ, commHandleRead, data, 0); |
c4b7a5a9 | 150 | } |
151 | ||
abd8f140 AJ |
152 | /** |
153 | * Queue a read. handler/handler_data are called when the read | |
154 | * completes, on error, or on file descriptor close. | |
155 | */ | |
b0469965 | 156 | void |
ec20038e | 157 | comm_read(const Comm::ConnectionPointer &conn, char *buf, int size, AsyncCall::Pointer &callback) |
c4b7a5a9 | 158 | { |
ec20038e | 159 | debugs(5, 5, "comm_read, queueing read for " << conn << "; asynCall " << callback); |
c4b7a5a9 | 160 | |
82ec8dfc | 161 | /* Make sure we are open and not closing */ |
ec20038e AJ |
162 | assert(Comm::IsConnOpen(conn)); |
163 | assert(!fd_table[conn->fd].closing()); | |
b0388924 | 164 | Comm::IoCallback *ccb = COMMIO_FD_READCB(conn->fd); |
82ec8dfc AR |
165 | |
166 | // Make sure we are either not reading or just passively monitoring. | |
167 | // Active/passive conflicts are OK and simply cancel passive monitoring. | |
168 | if (ccb->active()) { | |
169 | // if the assertion below fails, we have an active comm_read conflict | |
ec20038e AJ |
170 | assert(fd_table[conn->fd].halfClosedReader != NULL); |
171 | commStopHalfClosedMonitor(conn->fd); | |
82ec8dfc AR |
172 | assert(!ccb->active()); |
173 | } | |
b0388924 | 174 | ccb->conn = conn; |
528b2c61 | 175 | |
2b663917 | 176 | /* Queue the read */ |
ec41b64c | 177 | ccb->setCallback(Comm::IOCB_READ, callback, (char *)buf, NULL, size); |
8bbb16e3 | 178 | Comm::SetSelect(conn->fd, COMM_SELECT_READ, commHandleRead, ccb, 0); |
c4b7a5a9 | 179 | } |
180 | ||
e1a88700 | 181 | /** |
c4b7a5a9 | 182 | * Empty the read buffers |
183 | * | |
184 | * This is a magical routine that empties the read buffers. | |
185 | * Under some platforms (Linux) if a buffer has data in it before | |
186 | * you call close(), the socket will hang and take quite a while | |
187 | * to timeout. | |
188 | */ | |
189 | static void | |
190 | comm_empty_os_read_buffers(int fd) | |
191 | { | |
1191b93b | 192 | #if _SQUID_LINUX_ |
c4b7a5a9 | 193 | /* prevent those nasty RST packets */ |
194 | char buf[SQUID_TCP_SO_RCVBUF]; | |
62e76326 | 195 | |
cc192b50 | 196 | if (fd_table[fd].flags.nonblocking == 1) { |
197 | while (FD_READ_METHOD(fd, buf, SQUID_TCP_SO_RCVBUF) > 0) {}; | |
198 | } | |
c4b7a5a9 | 199 | #endif |
200 | } | |
201 | ||
202 | ||
e1a88700 | 203 | /** |
2b663917 | 204 | * Return whether the FD has a pending completed callback. |
e884bbde | 205 | * NP: does not work. |
c4b7a5a9 | 206 | */ |
207 | int | |
208 | comm_has_pending_read_callback(int fd) | |
209 | { | |
b0469965 | 210 | assert(isOpen(fd)); |
211 | // XXX: We do not know whether there is a read callback scheduled. | |
212 | // This is used for pconn management that should probably be more | |
213 | // tightly integrated into comm to minimize the chance that a | |
214 | // closing pconn socket will be used for a new transaction. | |
545d554b | 215 | return false; |
c4b7a5a9 | 216 | } |
217 | ||
b0469965 | 218 | // Does comm check this fd for read readiness? |
219 | // Note that when comm is not monitoring, there can be a pending callback | |
220 | // call, which may resume comm monitoring once fired. | |
528b2c61 | 221 | bool |
b0469965 | 222 | comm_monitors_read(int fd) |
c4b7a5a9 | 223 | { |
b0469965 | 224 | assert(isOpen(fd)); |
225 | // Being active is usually the same as monitoring because we always | |
ec41b64c AJ |
226 | // start monitoring the FD when we configure Comm::IoCallback for I/O |
227 | // and we usually configure Comm::IoCallback for I/O when we starting | |
228 | // monitoring a FD for reading. | |
b0469965 | 229 | return COMMIO_FD_READCB(fd)->active(); |
c4b7a5a9 | 230 | } |
231 | ||
e1a88700 | 232 | /** |
c4b7a5a9 | 233 | * Cancel a pending read. Assert that we have the right parameters, |
234 | * and that there are no pending read events! | |
2b663917 | 235 | * |
b0469965 | 236 | * XXX: We do not assert that there are no pending read events and |
237 | * with async calls it becomes even more difficult. | |
238 | * The whole interface should be reworked to do callback->cancel() | |
239 | * instead of searching for places where the callback may be stored and | |
240 | * updating the state of those places. | |
241 | * | |
2b663917 | 242 | * AHC Don't call the comm handlers? |
c4b7a5a9 | 243 | */ |
244 | void | |
245 | comm_read_cancel(int fd, IOCB *callback, void *data) | |
246 | { | |
b0469965 | 247 | if (!isOpen(fd)) { |
248 | debugs(5, 4, "comm_read_cancel fails: FD " << fd << " closed"); | |
249 | return; | |
9e008dda | 250 | } |
b0469965 | 251 | |
ec41b64c | 252 | Comm::IoCallback *cb = COMMIO_FD_READCB(fd); |
b0469965 | 253 | // TODO: is "active" == "monitors FD"? |
254 | if (!cb->active()) { | |
255 | debugs(5, 4, "comm_read_cancel fails: FD " << fd << " inactive"); | |
256 | return; | |
9e008dda | 257 | } |
b0469965 | 258 | |
259 | typedef CommCbFunPtrCallT<CommIoCbPtrFun> Call; | |
260 | Call *call = dynamic_cast<Call*>(cb->callback.getRaw()); | |
261 | if (!call) { | |
262 | debugs(5, 4, "comm_read_cancel fails: FD " << fd << " lacks callback"); | |
263 | return; | |
9e008dda | 264 | } |
b0469965 | 265 | |
82ec8dfc AR |
266 | call->cancel("old comm_read_cancel"); |
267 | ||
b0469965 | 268 | typedef CommIoCbParams Params; |
269 | const Params ¶ms = GetCommParams<Params>(cb->callback); | |
c4b7a5a9 | 270 | |
c4b7a5a9 | 271 | /* Ok, we can be reasonably sure we won't lose any data here! */ |
b0469965 | 272 | assert(call->dialer.handler == callback); |
273 | assert(params.data == data); | |
c4b7a5a9 | 274 | |
275 | /* Delete the callback */ | |
ec41b64c | 276 | cb->cancel("old comm_read_cancel"); |
420f2ac8 | 277 | |
278 | /* And the IO event */ | |
d841c88d | 279 | Comm::SetSelect(fd, COMM_SELECT_READ, NULL, NULL, 0); |
c4b7a5a9 | 280 | } |
281 | ||
c4b7a5a9 | 282 | void |
b0469965 | 283 | comm_read_cancel(int fd, AsyncCall::Pointer &callback) |
c4b7a5a9 | 284 | { |
b0469965 | 285 | callback->cancel("comm_read_cancel"); |
9e008dda | 286 | |
b0469965 | 287 | if (!isOpen(fd)) { |
288 | debugs(5, 4, "comm_read_cancel fails: FD " << fd << " closed"); | |
289 | return; | |
290 | } | |
291 | ||
ec41b64c | 292 | Comm::IoCallback *cb = COMMIO_FD_READCB(fd); |
b0469965 | 293 | |
294 | if (!cb->active()) { | |
295 | debugs(5, 4, "comm_read_cancel fails: FD " << fd << " inactive"); | |
296 | return; | |
297 | } | |
298 | ||
299 | AsyncCall::Pointer call = cb->callback; | |
82ec8dfc | 300 | assert(call != NULL); // XXX: should never fail (active() checks for callback==NULL) |
9e008dda | 301 | |
b0469965 | 302 | /* Ok, we can be reasonably sure we won't lose any data here! */ |
303 | assert(call == callback); | |
c4b7a5a9 | 304 | |
b0469965 | 305 | /* Delete the callback */ |
ec41b64c | 306 | cb->cancel("comm_read_cancel"); |
b0469965 | 307 | |
308 | /* And the IO event */ | |
d841c88d | 309 | Comm::SetSelect(fd, COMM_SELECT_READ, NULL, NULL, 0); |
c4b7a5a9 | 310 | } |
311 | ||
312 | ||
e1a88700 | 313 | /** |
ce767c23 | 314 | * synchronous wrapper around udp socket functions |
315 | */ | |
ce767c23 | 316 | int |
b7ac5457 | 317 | comm_udp_recvfrom(int fd, void *buf, size_t len, int flags, Ip::Address &from) |
ce767c23 | 318 | { |
95dc7ff4 | 319 | ++ statCounter.syscalls.sock.recvfroms; |
cc192b50 | 320 | int x = 0; |
321 | struct addrinfo *AI = NULL; | |
322 | ||
323 | debugs(5,8, "comm_udp_recvfrom: FD " << fd << " from " << from); | |
324 | ||
325 | assert( NULL == AI ); | |
326 | ||
327 | from.InitAddrInfo(AI); | |
328 | ||
329 | x = recvfrom(fd, buf, len, flags, AI->ai_addr, &AI->ai_addrlen); | |
330 | ||
331 | from = *AI; | |
332 | ||
333 | from.FreeAddrInfo(AI); | |
334 | ||
335 | return x; | |
ce767c23 | 336 | } |
337 | ||
365f12a9 | 338 | int |
7d21986b | 339 | comm_udp_recv(int fd, void *buf, size_t len, int flags) |
365f12a9 | 340 | { |
b7ac5457 | 341 | Ip::Address nul; |
cc192b50 | 342 | return comm_udp_recvfrom(fd, buf, len, flags, nul); |
365f12a9 | 343 | } |
344 | ||
f71da12c | 345 | ssize_t |
7d21986b | 346 | comm_udp_send(int s, const void *buf, size_t len, int flags) |
f71da12c | 347 | { |
62e76326 | 348 | return send(s, buf, len, flags); |
f71da12c | 349 | } |
ce767c23 | 350 | |
351 | ||
545d554b | 352 | bool |
353 | comm_has_incomplete_write(int fd) | |
354 | { | |
b0469965 | 355 | assert(isOpen(fd)); |
356 | return COMMIO_FD_WRITECB(fd)->active(); | |
d4cb310b | 357 | } |
358 | ||
e1a88700 | 359 | /** |
cf3c0ee3 | 360 | * Queue a write. handler/handler_data are called when the write fully |
361 | * completes, on error, or on file descriptor close. | |
362 | */ | |
9864ee44 | 363 | |
090089c4 | 364 | /* Return the local port associated with fd. */ |
f45dd259 | 365 | unsigned short |
b8d8561b | 366 | comm_local_port(int fd) |
090089c4 | 367 | { |
b7ac5457 | 368 | Ip::Address temp; |
cc192b50 | 369 | struct addrinfo *addr = NULL; |
76f87348 | 370 | fde *F = &fd_table[fd]; |
090089c4 | 371 | |
090089c4 | 372 | /* If the fd is closed already, just return */ |
62e76326 | 373 | |
60c0b5a2 | 374 | if (!F->flags.open) { |
bf8fe701 | 375 | debugs(5, 0, "comm_local_port: FD " << fd << " has been closed."); |
62e76326 | 376 | return 0; |
090089c4 | 377 | } |
62e76326 | 378 | |
cc192b50 | 379 | if (F->local_addr.GetPort()) |
380 | return F->local_addr.GetPort(); | |
62e76326 | 381 | |
6084c0b6 AJ |
382 | if (F->sock_family == AF_INET) |
383 | temp.SetIPv4(); | |
6084c0b6 | 384 | |
cc192b50 | 385 | temp.InitAddrInfo(addr); |
62e76326 | 386 | |
cc192b50 | 387 | if (getsockname(fd, addr->ai_addr, &(addr->ai_addrlen)) ) { |
bf8fe701 | 388 | debugs(50, 1, "comm_local_port: Failed to retrieve TCP/UDP port number for socket: FD " << fd << ": " << xstrerror()); |
cc192b50 | 389 | temp.FreeAddrInfo(addr); |
62e76326 | 390 | return 0; |
090089c4 | 391 | } |
cc192b50 | 392 | temp = *addr; |
393 | ||
394 | temp.FreeAddrInfo(addr); | |
395 | ||
3d031c35 AJ |
396 | if (F->local_addr.IsAnyAddr()) { |
397 | /* save the whole local address, not just the port. */ | |
398 | F->local_addr = temp; | |
399 | } else { | |
400 | F->local_addr.SetPort(temp.GetPort()); | |
401 | } | |
62e76326 | 402 | |
6084c0b6 | 403 | debugs(5, 6, "comm_local_port: FD " << fd << ": port " << F->local_addr.GetPort() << "(family=" << F->sock_family << ")"); |
cc192b50 | 404 | return F->local_addr.GetPort(); |
090089c4 | 405 | } |
406 | ||
3d7e9d7c | 407 | static comm_err_t |
cc192b50 | 408 | commBind(int s, struct addrinfo &inaddr) |
090089c4 | 409 | { |
95dc7ff4 | 410 | ++ statCounter.syscalls.sock.binds; |
62e76326 | 411 | |
ac760b5e AJ |
412 | if (bind(s, inaddr.ai_addr, inaddr.ai_addrlen) == 0) { |
413 | debugs(50, 6, "commBind: bind socket FD " << s << " to " << fd_table[s].local_addr); | |
62e76326 | 414 | return COMM_OK; |
ac760b5e | 415 | } |
62e76326 | 416 | |
cc192b50 | 417 | debugs(50, 0, "commBind: Cannot bind socket FD " << s << " to " << fd_table[s].local_addr << ": " << xstrerror()); |
62e76326 | 418 | |
090089c4 | 419 | return COMM_ERROR; |
420 | } | |
421 | ||
e1a88700 | 422 | /** |
423 | * Create a socket. Default is blocking, stream (TCP) socket. IO_TYPE | |
424 | * is OR of flags specified in comm.h. Defaults TOS | |
425 | */ | |
b8d8561b | 426 | int |
16b204c4 | 427 | comm_open(int sock_type, |
62e76326 | 428 | int proto, |
b7ac5457 | 429 | Ip::Address &addr, |
62e76326 | 430 | int flags, |
431 | const char *note) | |
d6827718 | 432 | { |
425de4c8 | 433 | return comm_openex(sock_type, proto, addr, flags, 0, 0, note); |
d6827718 | 434 | } |
435 | ||
e0d28505 AJ |
436 | void |
437 | comm_open_listener(int sock_type, | |
438 | int proto, | |
439 | Comm::ConnectionPointer &conn, | |
440 | const char *note) | |
441 | { | |
442 | /* all listener sockets require bind() */ | |
443 | conn->flags |= COMM_DOBIND; | |
444 | ||
445 | /* attempt native enabled port. */ | |
b5523edc | 446 | conn->fd = comm_openex(sock_type, proto, conn->local, conn->flags, 0, 0, note); |
e0d28505 AJ |
447 | } |
448 | ||
31be869c AJ |
449 | int |
450 | comm_open_listener(int sock_type, | |
04f7fd38 | 451 | int proto, |
b7ac5457 | 452 | Ip::Address &addr, |
04f7fd38 AJ |
453 | int flags, |
454 | const char *note) | |
31be869c AJ |
455 | { |
456 | int sock = -1; | |
457 | ||
ac760b5e AJ |
458 | /* all listener sockets require bind() */ |
459 | flags |= COMM_DOBIND; | |
460 | ||
31be869c | 461 | /* attempt native enabled port. */ |
425de4c8 | 462 | sock = comm_openex(sock_type, proto, addr, flags, 0, 0, note); |
31be869c | 463 | |
31be869c AJ |
464 | return sock; |
465 | } | |
466 | ||
2d8c0b1a | 467 | static bool |
468 | limitError(int const anErrno) | |
469 | { | |
470 | return anErrno == ENFILE || anErrno == EMFILE; | |
471 | } | |
d6827718 | 472 | |
cc192b50 | 473 | void |
474 | comm_set_v6only(int fd, int tos) | |
475 | { | |
476 | #ifdef IPV6_V6ONLY | |
477 | if (setsockopt(fd, IPPROTO_IPV6, IPV6_V6ONLY, (char *) &tos, sizeof(int)) < 0) { | |
7e07ced1 | 478 | debugs(50, 1, "comm_open: setsockopt(IPV6_V6ONLY) " << (tos?"ON":"OFF") << " for FD " << fd << ": " << xstrerror()); |
cc192b50 | 479 | } |
480 | #else | |
481 | debugs(50, 0, "WARNING: comm_open: setsockopt(IPV6_V6ONLY) not supported on this platform"); | |
482 | #endif /* sockopt */ | |
483 | } | |
057f5854 | 484 | |
40d6264d AJ |
485 | /** |
486 | * Set the socket IP_TRANSPARENT option for Linux TPROXY v4 support. | |
487 | */ | |
f1e0717c | 488 | void |
e950e673 | 489 | comm_set_transparent(int fd) |
f1e0717c | 490 | { |
2ad20b4f | 491 | #if defined(IP_TRANSPARENT) |
e950e673 | 492 | int tos = 1; |
ef88b51d AJ |
493 | if (setsockopt(fd, SOL_IP, IP_TRANSPARENT, (char *) &tos, sizeof(int)) < 0) { |
494 | debugs(50, DBG_IMPORTANT, "comm_open: setsockopt(IP_TRANSPARENT) on FD " << fd << ": " << xstrerror()); | |
9e008dda | 495 | } else { |
3949d8b7 AJ |
496 | /* mark the socket as having transparent options */ |
497 | fd_table[fd].flags.transparent = 1; | |
498 | } | |
f1e0717c | 499 | #else |
ef88b51d | 500 | debugs(50, DBG_CRITICAL, "WARNING: comm_open: setsockopt(IP_TRANSPARENT) not supported on this platform"); |
f1e0717c AJ |
501 | #endif /* sockopt */ |
502 | } | |
503 | ||
e1a88700 | 504 | /** |
505 | * Create a socket. Default is blocking, stream (TCP) socket. IO_TYPE | |
506 | * is OR of flags specified in defines.h:COMM_* | |
507 | */ | |
d6827718 | 508 | int |
509 | comm_openex(int sock_type, | |
62e76326 | 510 | int proto, |
b7ac5457 | 511 | Ip::Address &addr, |
62e76326 | 512 | int flags, |
425de4c8 AJ |
513 | tos_t tos, |
514 | nfmark_t nfmark, | |
62e76326 | 515 | const char *note) |
090089c4 | 516 | { |
517 | int new_socket; | |
cc192b50 | 518 | struct addrinfo *AI = NULL; |
090089c4 | 519 | |
88bfe092 | 520 | PROF_start(comm_open); |
090089c4 | 521 | /* Create socket for accepting new connections. */ |
95dc7ff4 | 522 | ++ statCounter.syscalls.sock.sockets; |
62e76326 | 523 | |
cc192b50 | 524 | /* Setup the socket addrinfo details for use */ |
525 | addr.GetAddrInfo(AI); | |
526 | AI->ai_socktype = sock_type; | |
527 | AI->ai_protocol = proto; | |
cc192b50 | 528 | |
e884bbde | 529 | debugs(50, 3, "comm_openex: Attempt open socket for: " << addr ); |
cc192b50 | 530 | |
0eb08770 | 531 | new_socket = socket(AI->ai_family, AI->ai_socktype, AI->ai_protocol); |
055421ee | 532 | |
0eb08770 HN |
533 | /* under IPv6 there is the possibility IPv6 is present but disabled. */ |
534 | /* try again as IPv4-native if possible */ | |
055421ee | 535 | if ( new_socket < 0 && Ip::EnableIpv6 && addr.IsIPv6() && addr.SetIPv4() ) { |
0eb08770 HN |
536 | /* attempt to open this IPv4-only. */ |
537 | addr.FreeAddrInfo(AI); | |
538 | /* Setup the socket addrinfo details for use */ | |
539 | addr.GetAddrInfo(AI); | |
540 | AI->ai_socktype = sock_type; | |
541 | AI->ai_protocol = proto; | |
542 | debugs(50, 3, "comm_openex: Attempt fallback open socket for: " << addr ); | |
ec505200 | 543 | new_socket = socket(AI->ai_family, AI->ai_socktype, AI->ai_protocol); |
0eb08770 HN |
544 | debugs(50, 2, HERE << "attempt open " << note << " socket on: " << addr); |
545 | } | |
0eb08770 HN |
546 | |
547 | if (new_socket < 0) { | |
62e76326 | 548 | /* Increase the number of reserved fd's if calls to socket() |
549 | * are failing because the open file table is full. This | |
550 | * limits the number of simultaneous clients */ | |
551 | ||
2d8c0b1a | 552 | if (limitError(errno)) { |
dd4e714f | 553 | debugs(50, DBG_IMPORTANT, "comm_open: socket failure: " << xstrerror()); |
62e76326 | 554 | fdAdjustReserved(); |
2d8c0b1a | 555 | } else { |
8ca8bd92 | 556 | debugs(50, DBG_CRITICAL, "comm_open: socket failure: " << xstrerror()); |
62e76326 | 557 | } |
558 | ||
cc192b50 | 559 | addr.FreeAddrInfo(AI); |
560 | ||
62e76326 | 561 | PROF_stop(comm_open); |
562 | return -1; | |
090089c4 | 563 | } |
62e76326 | 564 | |
1b76e6c1 | 565 | // XXX: temporary for the transition. comm_openex will eventually have a conn to play with. |
e884bbde AJ |
566 | Comm::ConnectionPointer conn = new Comm::Connection; |
567 | conn->local = addr; | |
b5523edc AJ |
568 | conn->fd = new_socket; |
569 | ||
570 | debugs(50, 3, "comm_openex: Opened socket " << conn << " : family=" << AI->ai_family << ", type=" << AI->ai_socktype << ", protocol=" << AI->ai_protocol ); | |
cc192b50 | 571 | |
d6827718 | 572 | /* set TOS if needed */ |
425de4c8 | 573 | if (tos) |
b5523edc | 574 | Ip::Qos::setSockTos(conn, tos); |
425de4c8 AJ |
575 | |
576 | /* set netfilter mark if needed */ | |
577 | if (nfmark) | |
b5523edc | 578 | Ip::Qos::setSockNfmark(conn, nfmark); |
62e76326 | 579 | |
055421ee | 580 | if ( Ip::EnableIpv6&IPV6_SPECIAL_SPLITSTACK && addr.IsIPv6() ) |
b5523edc | 581 | comm_set_v6only(conn->fd, 1); |
62e76326 | 582 | |
cc192b50 | 583 | /* Windows Vista supports Dual-Sockets. BUT defaults them to V6ONLY. Turn it OFF. */ |
9b1f7ee8 | 584 | /* Other OS may have this administratively disabled for general use. Same deal. */ |
055421ee | 585 | if ( Ip::EnableIpv6&IPV6_SPECIAL_V4MAPPING && addr.IsIPv6() ) |
b5523edc | 586 | comm_set_v6only(conn->fd, 0); |
cc192b50 | 587 | |
b5523edc AJ |
588 | comm_init_opened(conn, tos, nfmark, note, AI); |
589 | new_socket = comm_apply_flags(conn->fd, addr, flags, AI); | |
38344a8e AR |
590 | |
591 | addr.FreeAddrInfo(AI); | |
592 | ||
593 | PROF_stop(comm_open); | |
594 | ||
1b76e6c1 | 595 | // XXX transition only. prevent conn from closing the new FD on function exit. |
b5523edc | 596 | conn->fd = -1; |
38344a8e AR |
597 | return new_socket; |
598 | } | |
599 | ||
600 | /// update FD tables after a local or remote (IPC) comm_openex(); | |
601 | void | |
e0d28505 | 602 | comm_init_opened(const Comm::ConnectionPointer &conn, |
425de4c8 AJ |
603 | tos_t tos, |
604 | nfmark_t nfmark, | |
5667a628 AR |
605 | const char *note, |
606 | struct addrinfo *AI) | |
38344a8e | 607 | { |
e0d28505 | 608 | assert(Comm::IsConnOpen(conn)); |
38344a8e AR |
609 | assert(AI); |
610 | ||
090089c4 | 611 | /* update fdstat */ |
e0d28505 | 612 | debugs(5, 5, HERE << conn << " is a new socket"); |
62e76326 | 613 | |
e0d28505 AJ |
614 | assert(!isOpen(conn->fd)); // NP: global isOpen checks the fde entry for openness not the Comm::Connection |
615 | fd_open(conn->fd, FD_SOCKET, note); | |
62e76326 | 616 | |
e0d28505 AJ |
617 | fdd_table[conn->fd].close_file = NULL; |
618 | fdd_table[conn->fd].close_line = 0; | |
62e76326 | 619 | |
e0d28505 AJ |
620 | fde *F = &fd_table[conn->fd]; |
621 | F->local_addr = conn->local; | |
425de4c8 AJ |
622 | F->tosToServer = tos; |
623 | ||
624 | F->nfmarkToServer = nfmark; | |
cc192b50 | 625 | |
626 | F->sock_family = AI->ai_family; | |
38344a8e AR |
627 | } |
628 | ||
629 | /// apply flags after a local comm_open*() call; | |
630 | /// returns new_socket or -1 on error | |
631 | static int | |
632 | comm_apply_flags(int new_socket, | |
a67d2b2e | 633 | Ip::Address &addr, |
5667a628 AR |
634 | int flags, |
635 | struct addrinfo *AI) | |
38344a8e AR |
636 | { |
637 | assert(new_socket >= 0); | |
638 | assert(AI); | |
639 | const int sock_type = AI->ai_socktype; | |
62e76326 | 640 | |
79a15e0a | 641 | if (!(flags & COMM_NOCLOEXEC)) |
62e76326 | 642 | commSetCloseOnExec(new_socket); |
643 | ||
cdc33f35 | 644 | if ((flags & COMM_REUSEADDR)) |
62e76326 | 645 | commSetReuseAddr(new_socket); |
646 | ||
f45dd259 | 647 | if (addr.GetPort() > (unsigned short) 0) { |
1191b93b | 648 | #if _SQUID_MSWIN_ |
a50bfe93 | 649 | if (sock_type != SOCK_DGRAM) |
650 | #endif | |
a50bfe93 | 651 | commSetNoLinger(new_socket); |
62e76326 | 652 | |
653 | if (opt_reuseaddr) | |
654 | commSetReuseAddr(new_socket); | |
090089c4 | 655 | } |
62e76326 | 656 | |
a35595cd | 657 | /* MUST be done before binding or face OS Error: "(99) Cannot assign requested address"... */ |
9e008dda | 658 | if ((flags & COMM_TRANSPARENT)) { |
a35595cd AJ |
659 | comm_set_transparent(new_socket); |
660 | } | |
a35595cd | 661 | |
ac760b5e AJ |
662 | if ( (flags & COMM_DOBIND) || addr.GetPort() > 0 || !addr.IsAnyAddr() ) { |
663 | if ( !(flags & COMM_DOBIND) && addr.IsAnyAddr() ) | |
664 | debugs(5,1,"WARNING: Squid is attempting to bind() port " << addr << " without being a listener."); | |
665 | if ( addr.IsNoAddr() ) | |
666 | debugs(5,0,"CRITICAL: Squid is attempting to bind() port " << addr << "!!"); | |
667 | ||
cc192b50 | 668 | if (commBind(new_socket, *AI) != COMM_OK) { |
62e76326 | 669 | comm_close(new_socket); |
670 | return -1; | |
62e76326 | 671 | } |
23ff6968 | 672 | } |
62e76326 | 673 | |
79a15e0a | 674 | if (flags & COMM_NONBLOCKING) |
9e008dda | 675 | if (commSetNonBlocking(new_socket) == COMM_ERROR) { |
38344a8e | 676 | comm_close(new_socket); |
62e76326 | 677 | return -1; |
62e76326 | 678 | } |
679 | ||
30a4f2a8 | 680 | #ifdef TCP_NODELAY |
681 | if (sock_type == SOCK_STREAM) | |
62e76326 | 682 | commSetTcpNoDelay(new_socket); |
683 | ||
30a4f2a8 | 684 | #endif |
62e76326 | 685 | |
1241e63e | 686 | if (Config.tcpRcvBufsz > 0 && sock_type == SOCK_STREAM) |
62e76326 | 687 | commSetTcpRcvbuf(new_socket, Config.tcpRcvBufsz); |
688 | ||
090089c4 | 689 | return new_socket; |
690 | } | |
691 | ||
a4c0f9c6 | 692 | void |
e0d28505 | 693 | comm_import_opened(const Comm::ConnectionPointer &conn, |
5667a628 AR |
694 | const char *note, |
695 | struct addrinfo *AI) | |
a4c0f9c6 | 696 | { |
e0d28505 AJ |
697 | debugs(5, 2, HERE << conn); |
698 | assert(Comm::IsConnOpen(conn)); | |
a4c0f9c6 AR |
699 | assert(AI); |
700 | ||
9c8a6c3b | 701 | comm_init_opened(conn, 0, 0, note, AI); |
a4c0f9c6 | 702 | |
e0d28505 AJ |
703 | if (!(conn->flags & COMM_NOCLOEXEC)) |
704 | fd_table[conn->fd].flags.close_on_exec = 1; | |
a4c0f9c6 | 705 | |
f45dd259 | 706 | if (conn->local.GetPort() > (unsigned short) 0) { |
1191b93b | 707 | #if _SQUID_MSWIN_ |
e0d28505 | 708 | if (AI->ai_socktype != SOCK_DGRAM) |
a4c0f9c6 | 709 | #endif |
e0d28505 | 710 | fd_table[conn->fd].flags.nolinger = 1; |
a4c0f9c6 AR |
711 | } |
712 | ||
e0d28505 AJ |
713 | if ((conn->flags & COMM_TRANSPARENT)) |
714 | fd_table[conn->fd].flags.transparent = 1; | |
a4c0f9c6 | 715 | |
e0d28505 AJ |
716 | if (conn->flags & COMM_NONBLOCKING) |
717 | fd_table[conn->fd].flags.nonblocking = 1; | |
a4c0f9c6 AR |
718 | |
719 | #ifdef TCP_NODELAY | |
720 | if (AI->ai_socktype == SOCK_STREAM) | |
e0d28505 | 721 | fd_table[conn->fd].flags.nodelay = 1; |
a4c0f9c6 AR |
722 | #endif |
723 | ||
724 | /* no fd_table[fd].flags. updates needed for these conditions: | |
725 | * if ((flags & COMM_REUSEADDR)) ... | |
726 | * if ((flags & COMM_DOBIND) ...) ... | |
727 | */ | |
728 | } | |
729 | ||
933dd095 AJ |
730 | // XXX: now that raw-FD timeouts are only unset for pipes and files this SHOULD be a no-op. |
731 | // With handler already unset. Leaving this present until that can be verified for all code paths. | |
732 | void | |
733 | commUnsetFdTimeout(int fd) | |
b0469965 | 734 | { |
933dd095 | 735 | debugs(5, 3, HERE << "Remove timeout for FD " << fd); |
b0469965 | 736 | assert(fd >= 0); |
737 | assert(fd < Squid_MaxFD); | |
738 | fde *F = &fd_table[fd]; | |
739 | assert(F->flags.open); | |
740 | ||
933dd095 AJ |
741 | F->timeoutHandler = NULL; |
742 | F->timeout = 0; | |
7957e704 AJ |
743 | } |
744 | ||
745 | int | |
746 | commSetConnTimeout(const Comm::ConnectionPointer &conn, int timeout, AsyncCall::Pointer &callback) | |
747 | { | |
748 | debugs(5, 3, HERE << conn << " timeout " << timeout); | |
749 | assert(Comm::IsConnOpen(conn)); | |
750 | assert(conn->fd < Squid_MaxFD); | |
751 | fde *F = &fd_table[conn->fd]; | |
752 | assert(F->flags.open); | |
753 | ||
754 | if (timeout < 0) { | |
755 | F->timeoutHandler = NULL; | |
756 | F->timeout = 0; | |
757 | } else { | |
758 | if (callback != NULL) { | |
759 | typedef CommTimeoutCbParams Params; | |
760 | Params ¶ms = GetCommParams<Params>(callback); | |
761 | params.conn = conn; | |
762 | F->timeoutHandler = callback; | |
763 | } | |
b0469965 | 764 | |
7957e704 AJ |
765 | F->timeout = squid_curtime + (time_t) timeout; |
766 | } | |
767 | ||
768 | return F->timeout; | |
b0469965 | 769 | } |
090089c4 | 770 | |
8d77a37c AJ |
771 | int |
772 | commUnsetConnTimeout(const Comm::ConnectionPointer &conn) | |
773 | { | |
774 | debugs(5, 3, HERE << "Remove timeout for " << conn); | |
775 | AsyncCall::Pointer nil; | |
776 | return commSetConnTimeout(conn, -1, nil); | |
777 | } | |
778 | ||
b8d8561b | 779 | int |
b7ac5457 | 780 | comm_connect_addr(int sock, const Ip::Address &address) |
090089c4 | 781 | { |
3d7e9d7c | 782 | comm_err_t status = COMM_OK; |
76f87348 | 783 | fde *F = &fd_table[sock]; |
cc192b50 | 784 | int x = 0; |
b5568a61 | 785 | int err = 0; |
9689d97c | 786 | socklen_t errlen; |
feca3b9a | 787 | struct addrinfo *AI = NULL; |
88bfe092 | 788 | PROF_start(comm_connect_addr); |
cc192b50 | 789 | |
790 | assert(address.GetPort() != 0); | |
791 | ||
fa6a39d7 | 792 | debugs(5, 9, HERE << "connecting socket FD " << sock << " to " << address << " (want family: " << F->sock_family << ")"); |
cc192b50 | 793 | |
3d98ff81 | 794 | /* Handle IPv6 over IPv4-only socket case. |
9d92af86 | 795 | * this case must presently be handled here since the GetAddrInfo asserts on bad mappings. |
3d98ff81 | 796 | * NP: because commResetFD is private to ConnStateData we have to return an error and |
9d92af86 AJ |
797 | * trust its handled properly. |
798 | */ | |
9e008dda | 799 | if (F->sock_family == AF_INET && !address.IsIPv4()) { |
3d98ff81 HN |
800 | errno = ENETUNREACH; |
801 | return COMM_ERR_PROTOCOL; | |
802 | } | |
803 | ||
804 | /* Handle IPv4 over IPv6-only socket case. | |
805 | * This case is presently handled here as it's both a known case and it's | |
806 | * uncertain what error will be returned by the IPv6 stack in such case. It's | |
807 | * possible this will also be handled by the errno checks below after connect() | |
808 | * but needs carefull cross-platform verification, and verifying the address | |
809 | * condition here is simple. | |
810 | */ | |
811 | if (!F->local_addr.IsIPv4() && address.IsIPv4()) { | |
812 | errno = ENETUNREACH; | |
9d92af86 AJ |
813 | return COMM_ERR_PROTOCOL; |
814 | } | |
9d92af86 | 815 | |
feca3b9a | 816 | address.GetAddrInfo(AI, F->sock_family); |
cc192b50 | 817 | |
090089c4 | 818 | /* Establish connection. */ |
b5568a61 | 819 | errno = 0; |
62e76326 | 820 | |
9e008dda | 821 | if (!F->flags.called_connect) { |
62e76326 | 822 | F->flags.called_connect = 1; |
95dc7ff4 | 823 | ++ statCounter.syscalls.sock.connects; |
62e76326 | 824 | |
feca3b9a | 825 | x = connect(sock, AI->ai_addr, AI->ai_addrlen); |
62e76326 | 826 | |
5a33a66a | 827 | // XXX: ICAP code refuses callbacks during a pending comm_ call |
828 | // Async calls development will fix this. | |
829 | if (x == 0) { | |
830 | x = -1; | |
831 | errno = EINPROGRESS; | |
832 | } | |
833 | ||
9e008dda | 834 | if (x < 0) { |
cc192b50 | 835 | debugs(5,5, "comm_connect_addr: sock=" << sock << ", addrinfo( " << |
9e008dda AJ |
836 | " flags=" << AI->ai_flags << |
837 | ", family=" << AI->ai_family << | |
838 | ", socktype=" << AI->ai_socktype << | |
839 | ", protocol=" << AI->ai_protocol << | |
840 | ", &addr=" << AI->ai_addr << | |
841 | ", addrlen=" << AI->ai_addrlen << | |
842 | " )" ); | |
cc192b50 | 843 | debugs(5, 9, "connect FD " << sock << ": (" << x << ") " << xstrerror()); |
844 | debugs(14,9, "connecting to: " << address ); | |
845 | } | |
9e008dda | 846 | } else { |
8a09e810 | 847 | #if _SQUID_NEWSOS6_ |
62e76326 | 848 | /* Makoto MATSUSHITA <matusita@ics.es.osaka-u.ac.jp> */ |
849 | ||
feca3b9a | 850 | connect(sock, AI->ai_addr, AI->ai_addrlen); |
62e76326 | 851 | |
852 | if (errno == EINVAL) { | |
853 | errlen = sizeof(err); | |
854 | x = getsockopt(sock, SOL_SOCKET, SO_ERROR, &err, &errlen); | |
855 | ||
856 | if (x >= 0) | |
857 | errno = x; | |
858 | } | |
859 | ||
33ac9442 | 860 | #else |
62e76326 | 861 | errlen = sizeof(err); |
862 | ||
863 | x = getsockopt(sock, SOL_SOCKET, SO_ERROR, &err, &errlen); | |
864 | ||
865 | if (x == 0) | |
866 | errno = err; | |
867 | ||
8a09e810 | 868 | #if _SQUID_SOLARIS_ |
62e76326 | 869 | /* |
870 | * Solaris 2.4's socket emulation doesn't allow you | |
871 | * to determine the error from a failed non-blocking | |
872 | * connect and just returns EPIPE. Create a fake | |
873 | * error message for connect. -- fenner@parc.xerox.com | |
874 | */ | |
875 | if (x < 0 && errno == EPIPE) | |
876 | errno = ENOTCONN; | |
877 | ||
33ac9442 | 878 | #endif |
30a4f2a8 | 879 | #endif |
62e76326 | 880 | |
e5f6c5c2 | 881 | } |
62e76326 | 882 | |
9e008dda AJ |
883 | /* Squid seems to be working fine without this code. With this code, |
884 | * we leak memory on many connect requests because of EINPROGRESS. | |
885 | * If you find that this code is needed, please file a bug report. */ | |
82ec8dfc | 886 | #if 0 |
1191b93b | 887 | #if _SQUID_LINUX_ |
feca3b9a | 888 | /* 2007-11-27: |
9e008dda | 889 | * Linux Debian replaces our allocated AI pointer with garbage when |
feca3b9a AJ |
890 | * connect() fails. This leads to segmentation faults deallocating |
891 | * the system-allocated memory when we go to clean up our pointer. | |
892 | * HACK: is to leak the memory returned since we can't deallocate. | |
893 | */ | |
9e008dda | 894 | if (errno != 0) { |
feca3b9a AJ |
895 | AI = NULL; |
896 | } | |
82ec8dfc | 897 | #endif |
feca3b9a AJ |
898 | #endif |
899 | ||
900 | address.FreeAddrInfo(AI); | |
901 | ||
88bfe092 | 902 | PROF_stop(comm_connect_addr); |
62e76326 | 903 | |
b5568a61 | 904 | if (errno == 0 || errno == EISCONN) |
62e76326 | 905 | status = COMM_OK; |
b5568a61 | 906 | else if (ignoreErrno(errno)) |
62e76326 | 907 | status = COMM_INPROGRESS; |
3d98ff81 HN |
908 | else if (errno == EAFNOSUPPORT || errno == EINVAL) |
909 | return COMM_ERR_PROTOCOL; | |
b5568a61 | 910 | else |
3d98ff81 | 911 | return COMM_ERROR; |
62e76326 | 912 | |
cc192b50 | 913 | address.NtoA(F->ipaddr, MAX_IPSTRLEN); |
62e76326 | 914 | |
cc192b50 | 915 | F->remote_port = address.GetPort(); /* remote_port is HS */ |
62e76326 | 916 | |
9e008dda | 917 | if (status == COMM_OK) { |
cc192b50 | 918 | debugs(5, 10, "comm_connect_addr: FD " << sock << " connected to " << address); |
9e008dda | 919 | } else if (status == COMM_INPROGRESS) { |
bf8fe701 | 920 | debugs(5, 10, "comm_connect_addr: FD " << sock << " connection pending"); |
090089c4 | 921 | } |
62e76326 | 922 | |
090089c4 | 923 | return status; |
924 | } | |
925 | ||
cb201b7e | 926 | void |
927 | commCallCloseHandlers(int fd) | |
928 | { | |
76f87348 | 929 | fde *F = &fd_table[fd]; |
bf8fe701 | 930 | debugs(5, 5, "commCallCloseHandlers: FD " << fd); |
62e76326 | 931 | |
8000a965 | 932 | while (F->closeHandler != NULL) { |
b0469965 | 933 | AsyncCall::Pointer call = F->closeHandler; |
9e008dda AJ |
934 | F->closeHandler = call->Next(); |
935 | call->setNext(NULL); | |
936 | // If call is not canceled schedule it for execution else ignore it | |
937 | if (!call->canceled()) { | |
938 | debugs(5, 5, "commCallCloseHandlers: ch->handler=" << call); | |
9e008dda AJ |
939 | ScheduleCallHere(call); |
940 | } | |
cb201b7e | 941 | } |
942 | } | |
943 | ||
5492ad1d | 944 | #if LINGERING_CLOSE |
945 | static void | |
946 | commLingerClose(int fd, void *unused) | |
947 | { | |
948 | LOCAL_ARRAY(char, buf, 1024); | |
949 | int n; | |
1f7c9178 | 950 | n = FD_READ_METHOD(fd, buf, 1024); |
62e76326 | 951 | |
5492ad1d | 952 | if (n < 0) |
bf8fe701 | 953 | debugs(5, 3, "commLingerClose: FD " << fd << " read: " << xstrerror()); |
62e76326 | 954 | |
5492ad1d | 955 | comm_close(fd); |
956 | } | |
957 | ||
958 | static void | |
933dd095 | 959 | commLingerTimeout(const FdeCbParams ¶ms) |
5492ad1d | 960 | { |
933dd095 AJ |
961 | debugs(5, 3, "commLingerTimeout: FD " << params.fd); |
962 | comm_close(params.fd); | |
5492ad1d | 963 | } |
964 | ||
965 | /* | |
966 | * Inspired by apache | |
967 | */ | |
968 | void | |
969 | comm_lingering_close(int fd) | |
970 | { | |
d4c19b39 | 971 | #if USE_SSL |
972 | if (fd_table[fd].ssl) | |
575d05c4 | 973 | ssl_shutdown_method(fd_table[fd].ssl); |
d4c19b39 | 974 | #endif |
62e76326 | 975 | |
5492ad1d | 976 | if (shutdown(fd, 1) < 0) { |
62e76326 | 977 | comm_close(fd); |
978 | return; | |
5492ad1d | 979 | } |
62e76326 | 980 | |
5492ad1d | 981 | fd_note(fd, "lingering close"); |
933dd095 AJ |
982 | AsyncCall::Pointer call = commCbCall(5,4, "commLingerTimeout", FdeCbPtrFun(commLingerTimeout, NULL)); |
983 | ||
984 | debugs(5, 3, HERE << "FD " << fd << " timeout " << timeout); | |
985 | assert(fd_table[fd].flags.open); | |
986 | if (callback != NULL) { | |
987 | typedef FdeCbParams Params; | |
988 | Params ¶ms = GetCommParams<Params>(callback); | |
989 | params.fd = fd; | |
990 | fd_table[fd].timeoutHandler = callback; | |
991 | fd_table[fd].timeout = squid_curtime + static_cast<time_t>(10); | |
992 | } | |
993 | ||
d841c88d | 994 | Comm::SetSelect(fd, COMM_SELECT_READ, commLingerClose, NULL, 0); |
5492ad1d | 995 | } |
62e76326 | 996 | |
5492ad1d | 997 | #endif |
998 | ||
27774cee | 999 | /** |
98264874 | 1000 | * enable linger with time of 0 so that when the socket is |
1001 | * closed, TCP generates a RESET | |
1002 | */ | |
1003 | void | |
8ace824c | 1004 | comm_reset_close(const Comm::ConnectionPointer &conn) |
98264874 | 1005 | { |
5c336a3b AJ |
1006 | struct linger L; |
1007 | L.l_onoff = 1; | |
1008 | L.l_linger = 0; | |
62e76326 | 1009 | |
5c336a3b | 1010 | if (setsockopt(conn->fd, SOL_SOCKET, SO_LINGER, (char *) &L, sizeof(L)) < 0) |
1b76e6c1 | 1011 | debugs(50, DBG_CRITICAL, "ERROR: Closing " << conn << " with TCP RST: " << xstrerror()); |
5c336a3b AJ |
1012 | |
1013 | conn->close(); | |
1014 | } | |
1015 | ||
1016 | // Legacy close function. | |
1017 | void | |
1018 | old_comm_reset_close(int fd) | |
1019 | { | |
98264874 | 1020 | struct linger L; |
1021 | L.l_onoff = 1; | |
1022 | L.l_linger = 0; | |
62e76326 | 1023 | |
98264874 | 1024 | if (setsockopt(fd, SOL_SOCKET, SO_LINGER, (char *) &L, sizeof(L)) < 0) |
468fe1b5 | 1025 | debugs(50, DBG_CRITICAL, "ERROR: Closing FD " << fd << " with TCP RST: " << xstrerror()); |
62e76326 | 1026 | |
98264874 | 1027 | comm_close(fd); |
1028 | } | |
1029 | ||
575d05c4 | 1030 | #if USE_SSL |
9e008dda | 1031 | void |
a17bf806 | 1032 | commStartSslClose(const FdeCbParams ¶ms) |
10b06767 | 1033 | { |
575d05c4 AJ |
1034 | assert(&fd_table[params.fd].ssl); |
1035 | ssl_shutdown_method(fd_table[params.fd].ssl); | |
10b06767 | 1036 | } |
575d05c4 | 1037 | #endif |
10b06767 | 1038 | |
9e008dda | 1039 | void |
a17bf806 | 1040 | comm_close_complete(const FdeCbParams ¶ms) |
2d8c0b1a | 1041 | { |
b0469965 | 1042 | #if USE_SSL |
575d05c4 | 1043 | fde *F = &fd_table[params.fd]; |
2d8c0b1a | 1044 | |
b0469965 | 1045 | if (F->ssl) { |
1046 | SSL_free(F->ssl); | |
1047 | F->ssl = NULL; | |
1048 | } | |
2d8c0b1a | 1049 | |
95d2589c CT |
1050 | if (F->dynamicSslContext) { |
1051 | SSL_CTX_free(F->dynamicSslContext); | |
1052 | F->dynamicSslContext = NULL; | |
1053 | } | |
b0469965 | 1054 | #endif |
575d05c4 AJ |
1055 | fd_close(params.fd); /* update fdstat */ |
1056 | close(params.fd); | |
b0469965 | 1057 | |
95dc7ff4 | 1058 | ++ statCounter.syscalls.sock.closes; |
b0469965 | 1059 | |
575d05c4 | 1060 | /* When one connection closes, give accept() a chance, if need be */ |
04f55905 | 1061 | Comm::AcceptLimiter::Instance().kick(); |
2d8c0b1a | 1062 | } |
c4b7a5a9 | 1063 | |
1064 | /* | |
1065 | * Close the socket fd. | |
1066 | * | |
1067 | * + call write handlers with ERR_CLOSING | |
1068 | * + call read handlers with ERR_CLOSING | |
1069 | * + call closing handlers | |
a46d2c0e | 1070 | * |
9e008dda | 1071 | * NOTE: COMM_ERR_CLOSING will NOT be called for CommReads' sitting in a |
a46d2c0e | 1072 | * DeferredReadManager. |
c4b7a5a9 | 1073 | */ |
b8d8561b | 1074 | void |
43ae1d95 | 1075 | _comm_close(int fd, char const *file, int line) |
090089c4 | 1076 | { |
82ec8dfc | 1077 | debugs(5, 3, "comm_close: start closing FD " << fd); |
03eb2f01 | 1078 | assert(fd >= 0); |
1079 | assert(fd < Squid_MaxFD); | |
82ec8dfc AR |
1080 | |
1081 | fde *F = &fd_table[fd]; | |
c4b7a5a9 | 1082 | fdd_table[fd].close_file = file; |
1083 | fdd_table[fd].close_line = line; | |
1f7c9178 | 1084 | |
82ec8dfc | 1085 | if (F->closing()) |
62e76326 | 1086 | return; |
1087 | ||
36afac26 | 1088 | /* XXX: is this obsolete behind F->closing() ? */ |
b8869bcf | 1089 | if ( (shutting_down || reconfiguring) && (!F->flags.open || F->type == FD_FILE)) |
62e76326 | 1090 | return; |
1091 | ||
c4b7a5a9 | 1092 | /* The following fails because ipc.c is doing calls to pipe() to create sockets! */ |
c7e637d7 AJ |
1093 | if (!isOpen(fd)) { |
1094 | debugs(50, DBG_IMPORTANT, HERE << "BUG 3556: FD " << fd << " is not an open socket."); | |
1095 | // XXX: do we need to run close(fd) or fd_close(fd) here? | |
1096 | return; | |
1097 | } | |
62e76326 | 1098 | |
76f87348 | 1099 | assert(F->type != FD_FILE); |
62e76326 | 1100 | |
88bfe092 | 1101 | PROF_start(comm_close); |
62e76326 | 1102 | |
10b06767 | 1103 | F->flags.close_request = 1; |
62e76326 | 1104 | |
575d05c4 AJ |
1105 | #if USE_SSL |
1106 | if (F->ssl) { | |
575d05c4 | 1107 | AsyncCall::Pointer startCall=commCbCall(5,4, "commStartSslClose", |
a17bf806 AJ |
1108 | FdeCbPtrFun(commStartSslClose, NULL)); |
1109 | FdeCbParams &startParams = GetCommParams<FdeCbParams>(startCall); | |
575d05c4 AJ |
1110 | startParams.fd = fd; |
1111 | ScheduleCallHere(startCall); | |
1112 | } | |
1113 | #endif | |
62e76326 | 1114 | |
74257126 AR |
1115 | // a half-closed fd may lack a reader, so we stop monitoring explicitly |
1116 | if (commHasHalfClosedMonitor(fd)) | |
1117 | commStopHalfClosedMonitor(fd); | |
933dd095 | 1118 | commUnsetFdTimeout(fd); |
62e76326 | 1119 | |
a6351f16 | 1120 | // notify read/write handlers after canceling select reservations, if any |
ec41b64c | 1121 | if (COMMIO_FD_WRITECB(fd)->active()) { |
d841c88d | 1122 | Comm::SetSelect(fd, COMM_SELECT_WRITE, NULL, NULL, 0); |
ec41b64c | 1123 | COMMIO_FD_WRITECB(fd)->finish(COMM_ERR_CLOSING, errno); |
2b663917 | 1124 | } |
ec41b64c | 1125 | if (COMMIO_FD_READCB(fd)->active()) { |
d841c88d | 1126 | Comm::SetSelect(fd, COMM_SELECT_READ, NULL, NULL, 0); |
ec41b64c | 1127 | COMMIO_FD_READCB(fd)->finish(COMM_ERR_CLOSING, errno); |
2b663917 | 1128 | } |
2d8c0b1a | 1129 | |
9a0a18de | 1130 | #if USE_DELAY_POOLS |
b4cd430a CT |
1131 | if (ClientInfo *clientInfo = F->clientInfo) { |
1132 | if (clientInfo->selectWaiting) { | |
1133 | clientInfo->selectWaiting = false; | |
1134 | // kick queue or it will get stuck as commWriteHandle is not called | |
1135 | clientInfo->kickQuotaQueue(); | |
1136 | } | |
f33d34a8 | 1137 | } |
b4cd430a CT |
1138 | #endif |
1139 | ||
cb201b7e | 1140 | commCallCloseHandlers(fd); |
62e76326 | 1141 | |
fb505fa1 | 1142 | if (F->pconn.uses && F->pconn.pool) |
2dba5b8e | 1143 | F->pconn.pool->noteUses(F->pconn.uses); |
62e76326 | 1144 | |
a7ad6e4e | 1145 | comm_empty_os_read_buffers(fd); |
9e008dda | 1146 | |
62e76326 | 1147 | |
10b06767 | 1148 | AsyncCall::Pointer completeCall=commCbCall(5,4, "comm_close_complete", |
a17bf806 AJ |
1149 | FdeCbPtrFun(comm_close_complete, NULL)); |
1150 | FdeCbParams &completeParams = GetCommParams<FdeCbParams>(completeCall); | |
10b06767 | 1151 | completeParams.fd = fd; |
9e008dda | 1152 | // must use async call to wait for all callbacks |
82ec8dfc | 1153 | // scheduled before comm_close() to finish |
10b06767 | 1154 | ScheduleCallHere(completeCall); |
62e76326 | 1155 | |
88bfe092 | 1156 | PROF_stop(comm_close); |
090089c4 | 1157 | } |
1158 | ||
090089c4 | 1159 | /* Send a udp datagram to specified TO_ADDR. */ |
b8d8561b | 1160 | int |
5df61230 | 1161 | comm_udp_sendto(int fd, |
b7ac5457 | 1162 | const Ip::Address &to_addr, |
62e76326 | 1163 | const void *buf, |
1164 | int len) | |
090089c4 | 1165 | { |
cc192b50 | 1166 | int x = 0; |
1167 | struct addrinfo *AI = NULL; | |
1168 | ||
88bfe092 | 1169 | PROF_start(comm_udp_sendto); |
95dc7ff4 | 1170 | ++ statCounter.syscalls.sock.sendtos; |
62e76326 | 1171 | |
cc192b50 | 1172 | debugs(50, 3, "comm_udp_sendto: Attempt to send UDP packet to " << to_addr << |
9e008dda | 1173 | " using FD " << fd << " using Port " << comm_local_port(fd) ); |
cc192b50 | 1174 | |
1175 | /* BUG: something in the above macro appears to occasionally be setting AI to garbage. */ | |
1176 | /* AYJ: 2007-08-27 : or was it because I wasn't then setting 'fd_table[fd].sock_family' to fill properly. */ | |
1177 | assert( NULL == AI ); | |
1178 | ||
1179 | to_addr.GetAddrInfo(AI, fd_table[fd].sock_family); | |
1180 | ||
1181 | x = sendto(fd, buf, len, 0, AI->ai_addr, AI->ai_addrlen); | |
1182 | ||
1183 | to_addr.FreeAddrInfo(AI); | |
1184 | ||
88bfe092 | 1185 | PROF_stop(comm_udp_sendto); |
62e76326 | 1186 | |
2d8c0b1a | 1187 | if (x >= 0) |
1188 | return x; | |
1189 | ||
1191b93b | 1190 | #if _SQUID_LINUX_ |
62e76326 | 1191 | |
2d8c0b1a | 1192 | if (ECONNREFUSED != errno) |
17d51783 | 1193 | #endif |
62e76326 | 1194 | |
cc192b50 | 1195 | debugs(50, 1, "comm_udp_sendto: FD " << fd << ", (family=" << fd_table[fd].sock_family << ") " << to_addr << ": " << xstrerror()); |
62e76326 | 1196 | |
2d8c0b1a | 1197 | return COMM_ERROR; |
090089c4 | 1198 | } |
1199 | ||
b8d8561b | 1200 | void |
575d05c4 | 1201 | comm_add_close_handler(int fd, CLCB * handler, void *data) |
30a4f2a8 | 1202 | { |
bf8fe701 | 1203 | debugs(5, 5, "comm_add_close_handler: FD " << fd << ", handler=" << |
1204 | handler << ", data=" << data); | |
62e76326 | 1205 | |
b0469965 | 1206 | AsyncCall::Pointer call=commCbCall(5,4, "SomeCloseHandler", |
9e008dda | 1207 | CommCloseCbPtrFun(handler, data)); |
b0469965 | 1208 | comm_add_close_handler(fd, call); |
1209 | } | |
62e76326 | 1210 | |
b0469965 | 1211 | void |
1212 | comm_add_close_handler(int fd, AsyncCall::Pointer &call) | |
1213 | { | |
1214 | debugs(5, 5, "comm_add_close_handler: FD " << fd << ", AsyncCall=" << call); | |
62e76326 | 1215 | |
b0469965 | 1216 | /*TODO:Check for a similar scheduled AsyncCall*/ |
1217 | // for (c = fd_table[fd].closeHandler; c; c = c->next) | |
1218 | // assert(c->handler != handler || c->data != data); | |
62e76326 | 1219 | |
b0469965 | 1220 | call->setNext(fd_table[fd].closeHandler); |
62e76326 | 1221 | |
b0469965 | 1222 | fd_table[fd].closeHandler = call; |
30a4f2a8 | 1223 | } |
1224 | ||
b0469965 | 1225 | |
1226 | // remove function-based close handler | |
b8d8561b | 1227 | void |
575d05c4 | 1228 | comm_remove_close_handler(int fd, CLCB * handler, void *data) |
090089c4 | 1229 | { |
b0469965 | 1230 | assert (isOpen(fd)); |
30a4f2a8 | 1231 | /* Find handler in list */ |
bf8fe701 | 1232 | debugs(5, 5, "comm_remove_close_handler: FD " << fd << ", handler=" << |
1233 | handler << ", data=" << data); | |
62e76326 | 1234 | |
37cba319 AR |
1235 | AsyncCall::Pointer p, prev = NULL; |
1236 | for (p = fd_table[fd].closeHandler; p != NULL; prev = p, p = p->Next()) { | |
b0469965 | 1237 | typedef CommCbFunPtrCallT<CommCloseCbPtrFun> Call; |
1238 | const Call *call = dynamic_cast<const Call*>(p.getRaw()); | |
1239 | if (!call) // method callbacks have their own comm_remove_close_handler | |
1240 | continue; | |
62e76326 | 1241 | |
b0469965 | 1242 | typedef CommCloseCbParams Params; |
1243 | const Params ¶ms = GetCommParams<Params>(p); | |
1244 | if (call->dialer.handler == handler && params.data == data) | |
1245 | break; /* This is our handler */ | |
1246 | } | |
7828df5b CT |
1247 | |
1248 | // comm_close removes all close handlers so our handler may be gone | |
37cba319 AR |
1249 | if (p != NULL) { |
1250 | p->dequeue(fd_table[fd].closeHandler, prev); | |
7828df5b | 1251 | p->cancel("comm_remove_close_handler"); |
37cba319 | 1252 | } |
b0469965 | 1253 | } |
62e76326 | 1254 | |
b0469965 | 1255 | // remove method-based close handler |
1256 | void | |
1257 | comm_remove_close_handler(int fd, AsyncCall::Pointer &call) | |
1258 | { | |
1259 | assert (isOpen(fd)); | |
b0469965 | 1260 | debugs(5, 5, "comm_remove_close_handler: FD " << fd << ", AsyncCall=" << call); |
62e76326 | 1261 | |
7828df5b | 1262 | // comm_close removes all close handlers so our handler may be gone |
37cba319 AR |
1263 | AsyncCall::Pointer p, prev = NULL; |
1264 | for (p = fd_table[fd].closeHandler; p != NULL && p != call; prev = p, p = p->Next()); | |
62e76326 | 1265 | |
37cba319 AR |
1266 | if (p != NULL) |
1267 | p->dequeue(fd_table[fd].closeHandler, prev); | |
b0469965 | 1268 | call->cancel("comm_remove_close_handler"); |
30a4f2a8 | 1269 | } |
090089c4 | 1270 | |
b8d8561b | 1271 | static void |
1272 | commSetNoLinger(int fd) | |
30a4f2a8 | 1273 | { |
62e76326 | 1274 | |
30a4f2a8 | 1275 | struct linger L; |
090089c4 | 1276 | L.l_onoff = 0; /* off */ |
1277 | L.l_linger = 0; | |
62e76326 | 1278 | |
30a4f2a8 | 1279 | if (setsockopt(fd, SOL_SOCKET, SO_LINGER, (char *) &L, sizeof(L)) < 0) |
bf8fe701 | 1280 | debugs(50, 0, "commSetNoLinger: FD " << fd << ": " << xstrerror()); |
62e76326 | 1281 | |
58a6c186 | 1282 | fd_table[fd].flags.nolinger = 1; |
090089c4 | 1283 | } |
1284 | ||
b8d8561b | 1285 | static void |
1286 | commSetReuseAddr(int fd) | |
090089c4 | 1287 | { |
1288 | int on = 1; | |
62e76326 | 1289 | |
30a4f2a8 | 1290 | if (setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, (char *) &on, sizeof(on)) < 0) |
bf8fe701 | 1291 | debugs(50, 1, "commSetReuseAddr: FD " << fd << ": " << xstrerror()); |
090089c4 | 1292 | } |
1293 | ||
b8d8561b | 1294 | static void |
1295 | commSetTcpRcvbuf(int fd, int size) | |
f868539a | 1296 | { |
1297 | if (setsockopt(fd, SOL_SOCKET, SO_RCVBUF, (char *) &size, sizeof(size)) < 0) | |
bf8fe701 | 1298 | debugs(50, 1, "commSetTcpRcvbuf: FD " << fd << ", SIZE " << size << ": " << xstrerror()); |
8f0d53ef | 1299 | if (setsockopt(fd, SOL_SOCKET, SO_SNDBUF, (char *) &size, sizeof(size)) < 0) |
1300 | debugs(50, 1, "commSetTcpRcvbuf: FD " << fd << ", SIZE " << size << ": " << xstrerror()); | |
1301 | #ifdef TCP_WINDOW_CLAMP | |
1302 | if (setsockopt(fd, SOL_TCP, TCP_WINDOW_CLAMP, (char *) &size, sizeof(size)) < 0) | |
1303 | debugs(50, 1, "commSetTcpRcvbuf: FD " << fd << ", SIZE " << size << ": " << xstrerror()); | |
1304 | #endif | |
f868539a | 1305 | } |
1306 | ||
b8d8561b | 1307 | int |
1308 | commSetNonBlocking(int fd) | |
30a4f2a8 | 1309 | { |
1191b93b | 1310 | #if !_SQUID_MSWIN_ |
731e4d49 | 1311 | int flags; |
9e205701 | 1312 | int dummy = 0; |
a50bfe93 | 1313 | #endif |
be266cb2 | 1314 | #if _SQUID_WINDOWS_ |
b05490a8 | 1315 | int nonblocking = TRUE; |
62e76326 | 1316 | |
be266cb2 | 1317 | #if _SQUID_CYGWIN_ |
7f6ffd15 | 1318 | if (fd_table[fd].type != FD_PIPE) { |
629b5f75 | 1319 | #endif |
1320 | ||
62e76326 | 1321 | if (ioctl(fd, FIONBIO, &nonblocking) < 0) { |
bf8fe701 | 1322 | debugs(50, 0, "commSetNonBlocking: FD " << fd << ": " << xstrerror() << " " << fd_table[fd].type); |
62e76326 | 1323 | return COMM_ERROR; |
1324 | } | |
629b5f75 | 1325 | |
be266cb2 | 1326 | #if _SQUID_CYGWIN_ |
7f6ffd15 | 1327 | } else { |
1328 | #endif | |
629b5f75 | 1329 | #endif |
1191b93b | 1330 | #if !_SQUID_MSWIN_ |
62e76326 | 1331 | |
1332 | if ((flags = fcntl(fd, F_GETFL, dummy)) < 0) { | |
bf8fe701 | 1333 | debugs(50, 0, "FD " << fd << ": fcntl F_GETFL: " << xstrerror()); |
62e76326 | 1334 | return COMM_ERROR; |
1335 | } | |
1336 | ||
1337 | if (fcntl(fd, F_SETFL, flags | SQUID_NONBLOCK) < 0) { | |
bf8fe701 | 1338 | debugs(50, 0, "commSetNonBlocking: FD " << fd << ": " << xstrerror()); |
62e76326 | 1339 | return COMM_ERROR; |
1340 | } | |
1341 | ||
a50bfe93 | 1342 | #endif |
be266cb2 | 1343 | #if _SQUID_CYGWIN_ |
090089c4 | 1344 | } |
7f6ffd15 | 1345 | #endif |
58a6c186 | 1346 | fd_table[fd].flags.nonblocking = 1; |
62e76326 | 1347 | |
090089c4 | 1348 | return 0; |
1349 | } | |
1350 | ||
7e3ce7b9 | 1351 | int |
1352 | commUnsetNonBlocking(int fd) | |
1353 | { | |
1191b93b | 1354 | #if _SQUID_MSWIN_ |
a50bfe93 | 1355 | int nonblocking = FALSE; |
1356 | ||
1357 | if (ioctlsocket(fd, FIONBIO, (unsigned long *) &nonblocking) < 0) { | |
1358 | #else | |
7e3ce7b9 | 1359 | int flags; |
1360 | int dummy = 0; | |
62e76326 | 1361 | |
7e3ce7b9 | 1362 | if ((flags = fcntl(fd, F_GETFL, dummy)) < 0) { |
bf8fe701 | 1363 | debugs(50, 0, "FD " << fd << ": fcntl F_GETFL: " << xstrerror()); |
62e76326 | 1364 | return COMM_ERROR; |
7e3ce7b9 | 1365 | } |
62e76326 | 1366 | |
7e3ce7b9 | 1367 | if (fcntl(fd, F_SETFL, flags & (~SQUID_NONBLOCK)) < 0) { |
a50bfe93 | 1368 | #endif |
bf8fe701 | 1369 | debugs(50, 0, "commUnsetNonBlocking: FD " << fd << ": " << xstrerror()); |
62e76326 | 1370 | return COMM_ERROR; |
7e3ce7b9 | 1371 | } |
62e76326 | 1372 | |
7e3ce7b9 | 1373 | fd_table[fd].flags.nonblocking = 0; |
1374 | return 0; | |
1375 | } | |
1376 | ||
b8d8561b | 1377 | void |
e1381638 AJ |
1378 | commSetCloseOnExec(int fd) |
1379 | { | |
3ca60c86 | 1380 | #ifdef FD_CLOEXEC |
731e4d49 | 1381 | int flags; |
7a18b487 | 1382 | int dummy = 0; |
62e76326 | 1383 | |
2209fe19 AJ |
1384 | if ((flags = fcntl(fd, F_GETFD, dummy)) < 0) { |
1385 | debugs(50, 0, "FD " << fd << ": fcntl F_GETFD: " << xstrerror()); | |
62e76326 | 1386 | return; |
3ca60c86 | 1387 | } |
62e76326 | 1388 | |
24382924 | 1389 | if (fcntl(fd, F_SETFD, flags | FD_CLOEXEC) < 0) |
bf8fe701 | 1390 | debugs(50, 0, "FD " << fd << ": set close-on-exec failed: " << xstrerror()); |
62e76326 | 1391 | |
d6827718 | 1392 | fd_table[fd].flags.close_on_exec = 1; |
62e76326 | 1393 | |
3ca60c86 | 1394 | #endif |
1395 | } | |
1396 | ||
e90100aa | 1397 | #ifdef TCP_NODELAY |
1398 | static void | |
e1381638 AJ |
1399 | commSetTcpNoDelay(int fd) |
1400 | { | |
e90100aa | 1401 | int on = 1; |
62e76326 | 1402 | |
e90100aa | 1403 | if (setsockopt(fd, IPPROTO_TCP, TCP_NODELAY, (char *) &on, sizeof(on)) < 0) |
bf8fe701 | 1404 | debugs(50, 1, "commSetTcpNoDelay: FD " << fd << ": " << xstrerror()); |
62e76326 | 1405 | |
d6827718 | 1406 | fd_table[fd].flags.nodelay = 1; |
e90100aa | 1407 | } |
62e76326 | 1408 | |
e90100aa | 1409 | #endif |
1410 | ||
b2130d58 | 1411 | void |
e1381638 AJ |
1412 | commSetTcpKeepalive(int fd, int idle, int interval, int timeout) |
1413 | { | |
b2130d58 | 1414 | int on = 1; |
1415 | #ifdef TCP_KEEPCNT | |
1416 | if (timeout && interval) { | |
9e008dda AJ |
1417 | int count = (timeout + interval - 1) / interval; |
1418 | if (setsockopt(fd, IPPROTO_TCP, TCP_KEEPCNT, &count, sizeof(on)) < 0) | |
1419 | debugs(5, 1, "commSetKeepalive: FD " << fd << ": " << xstrerror()); | |
b2130d58 | 1420 | } |
1421 | #endif | |
1422 | #ifdef TCP_KEEPIDLE | |
1423 | if (idle) { | |
9e008dda AJ |
1424 | if (setsockopt(fd, IPPROTO_TCP, TCP_KEEPIDLE, &idle, sizeof(on)) < 0) |
1425 | debugs(5, 1, "commSetKeepalive: FD " << fd << ": " << xstrerror()); | |
b2130d58 | 1426 | } |
1427 | #endif | |
1428 | #ifdef TCP_KEEPINTVL | |
1429 | if (interval) { | |
9e008dda AJ |
1430 | if (setsockopt(fd, IPPROTO_TCP, TCP_KEEPINTVL, &interval, sizeof(on)) < 0) |
1431 | debugs(5, 1, "commSetKeepalive: FD " << fd << ": " << xstrerror()); | |
b2130d58 | 1432 | } |
1433 | #endif | |
1434 | if (setsockopt(fd, SOL_SOCKET, SO_KEEPALIVE, (char *) &on, sizeof(on)) < 0) | |
9e008dda | 1435 | debugs(5, 1, "commSetKeepalive: FD " << fd << ": " << xstrerror()); |
b2130d58 | 1436 | } |
6a988308 | 1437 | |
d86b3703 | 1438 | void |
e1381638 AJ |
1439 | comm_init(void) |
1440 | { | |
c4b7a5a9 | 1441 | fd_table =(fde *) xcalloc(Squid_MaxFD, sizeof(fde)); |
1442 | fdd_table = (fd_debug_t *)xcalloc(Squid_MaxFD, sizeof(fd_debug_t)); | |
2d8c0b1a | 1443 | |
04f55905 AJ |
1444 | /* make sure the accept() socket FIFO delay queue exists */ |
1445 | Comm::AcceptLimiter::Instance(); | |
b0469965 | 1446 | |
ec41b64c AJ |
1447 | // make sure the IO pending callback table exists |
1448 | Comm::CallbackTableInit(); | |
2d8c0b1a | 1449 | |
59c4d35b | 1450 | /* XXX account fd_table */ |
090089c4 | 1451 | /* Keep a few file descriptors free so that we don't run out of FD's |
1452 | * after accepting a client but before it opens a socket or a file. | |
e83892e9 | 1453 | * Since Squid_MaxFD can be as high as several thousand, don't waste them */ |
d85c3078 | 1454 | RESERVED_FD = min(100, Squid_MaxFD / 4); |
2d8c0b1a | 1455 | |
04eb0689 | 1456 | conn_close_pool = memPoolCreate("close_handler", sizeof(close_handler)); |
74257126 AR |
1457 | |
1458 | TheHalfClosed = new DescriptorSet; | |
d841c88d AJ |
1459 | |
1460 | /* setup the select loop module */ | |
1461 | Comm::SelectLoopInit(); | |
090089c4 | 1462 | } |
1463 | ||
236d1779 | 1464 | void |
e1381638 AJ |
1465 | comm_exit(void) |
1466 | { | |
74257126 AR |
1467 | delete TheHalfClosed; |
1468 | TheHalfClosed = NULL; | |
1469 | ||
236d1779 | 1470 | safe_free(fd_table); |
1471 | safe_free(fdd_table); | |
ec41b64c | 1472 | Comm::CallbackTableDestruct(); |
236d1779 | 1473 | } |
1474 | ||
9a0a18de | 1475 | #if USE_DELAY_POOLS |
b4cd430a | 1476 | // called when the queue is done waiting for the client bucket to fill |
ec41b64c | 1477 | void |
b4cd430a CT |
1478 | commHandleWriteHelper(void * data) |
1479 | { | |
1480 | CommQuotaQueue *queue = static_cast<CommQuotaQueue*>(data); | |
1481 | assert(queue); | |
1482 | ||
1483 | ClientInfo *clientInfo = queue->clientInfo; | |
1484 | // ClientInfo invalidates queue if freed, so if we got here through, | |
1485 | // evenAdd cbdata protections, everything should be valid and consistent | |
f33d34a8 | 1486 | assert(clientInfo); |
b4cd430a CT |
1487 | assert(clientInfo->hasQueue()); |
1488 | assert(clientInfo->hasQueue(queue)); | |
1489 | assert(!clientInfo->selectWaiting); | |
1490 | assert(clientInfo->eventWaiting); | |
1491 | clientInfo->eventWaiting = false; | |
1492 | ||
1493 | do { | |
1494 | // check that the head descriptor is still relevant | |
1495 | const int head = clientInfo->quotaPeekFd(); | |
ec41b64c | 1496 | Comm::IoCallback *ccb = COMMIO_FD_WRITECB(head); |
b4cd430a CT |
1497 | |
1498 | if (fd_table[head].clientInfo == clientInfo && | |
f33d34a8 A |
1499 | clientInfo->quotaPeekReserv() == ccb->quotaQueueReserv && |
1500 | !fd_table[head].closing()) { | |
b4cd430a CT |
1501 | |
1502 | // wait for the head descriptor to become ready for writing | |
d841c88d | 1503 | Comm::SetSelect(head, COMM_SELECT_WRITE, Comm::HandleWrite, ccb, 0); |
b4cd430a CT |
1504 | clientInfo->selectWaiting = true; |
1505 | return; | |
f33d34a8 | 1506 | } |
b4cd430a | 1507 | |
f33d34a8 A |
1508 | clientInfo->quotaDequeue(); // remove the no longer relevant descriptor |
1509 | // and continue looking for a relevant one | |
b4cd430a CT |
1510 | } while (clientInfo->hasQueue()); |
1511 | ||
1512 | debugs(77,3, HERE << "emptied queue"); | |
1513 | } | |
1514 | ||
1515 | bool | |
1516 | ClientInfo::hasQueue() const | |
1517 | { | |
1518 | assert(quotaQueue); | |
1519 | return !quotaQueue->empty(); | |
1520 | } | |
1521 | ||
1522 | bool | |
1523 | ClientInfo::hasQueue(const CommQuotaQueue *q) const | |
1524 | { | |
1525 | assert(quotaQueue); | |
1526 | return quotaQueue == q; | |
1527 | } | |
1528 | ||
1529 | /// returns the first descriptor to be dequeued | |
1530 | int | |
1531 | ClientInfo::quotaPeekFd() const | |
1532 | { | |
1533 | assert(quotaQueue); | |
1534 | return quotaQueue->front(); | |
1535 | } | |
1536 | ||
1537 | /// returns the reservation ID of the first descriptor to be dequeued | |
1538 | unsigned int | |
1539 | ClientInfo::quotaPeekReserv() const | |
1540 | { | |
1541 | assert(quotaQueue); | |
1542 | return quotaQueue->outs + 1; | |
1543 | } | |
1544 | ||
1545 | /// queues a given fd, creating the queue if necessary; returns reservation ID | |
1546 | unsigned int | |
1547 | ClientInfo::quotaEnqueue(int fd) | |
1548 | { | |
1549 | assert(quotaQueue); | |
1550 | return quotaQueue->enqueue(fd); | |
1551 | } | |
1552 | ||
1553 | /// removes queue head | |
1554 | void | |
1555 | ClientInfo::quotaDequeue() | |
1556 | { | |
1557 | assert(quotaQueue); | |
1558 | quotaQueue->dequeue(); | |
1559 | } | |
1560 | ||
1561 | void | |
1562 | ClientInfo::kickQuotaQueue() | |
1563 | { | |
1564 | if (!eventWaiting && !selectWaiting && hasQueue()) { | |
1565 | // wait at least a second if the bucket is empty | |
1566 | const double delay = (bucketSize < 1.0) ? 1.0 : 0.0; | |
1567 | eventAdd("commHandleWriteHelper", &commHandleWriteHelper, | |
f33d34a8 | 1568 | quotaQueue, delay, 0, true); |
b4cd430a CT |
1569 | eventWaiting = true; |
1570 | } | |
1571 | } | |
1572 | ||
1573 | /// calculates how much to write for a single dequeued client | |
1574 | int | |
1575 | ClientInfo::quotaForDequed() | |
1576 | { | |
1577 | /* If we have multiple clients and give full bucketSize to each client then | |
1578 | * clt1 may often get a lot more because clt1->clt2 time distance in the | |
1579 | * select(2) callback order may be a lot smaller than cltN->clt1 distance. | |
1580 | * We divide quota evenly to be more fair. */ | |
1581 | ||
1582 | if (!rationedCount) { | |
1583 | rationedCount = quotaQueue->size() + 1; | |
1584 | ||
1585 | // The delay in ration recalculation _temporary_ deprives clients from | |
1586 | // bytes that should have trickled in while rationedCount was positive. | |
1587 | refillBucket(); | |
1588 | ||
1589 | // Rounding errors do not accumulate here, but we round down to avoid | |
1590 | // negative bucket sizes after write with rationedCount=1. | |
1591 | rationedQuota = static_cast<int>(floor(bucketSize/rationedCount)); | |
1592 | debugs(77,5, HERE << "new rationedQuota: " << rationedQuota << | |
f33d34a8 | 1593 | '*' << rationedCount); |
b4cd430a CT |
1594 | } |
1595 | ||
1596 | --rationedCount; | |
1597 | debugs(77,7, HERE << "rationedQuota: " << rationedQuota << | |
f33d34a8 | 1598 | " rations remaining: " << rationedCount); |
b4cd430a CT |
1599 | |
1600 | // update 'last seen' time to prevent clientdb GC from dropping us | |
1601 | last_seen = squid_curtime; | |
1602 | return rationedQuota; | |
1603 | } | |
1604 | ||
1605 | ///< adds bytes to the quota bucket based on the rate and passed time | |
1606 | void | |
1607 | ClientInfo::refillBucket() | |
1608 | { | |
1609 | // all these times are in seconds, with double precision | |
1610 | const double currTime = current_dtime; | |
1611 | const double timePassed = currTime - prevTime; | |
1612 | ||
f33d34a8 | 1613 | // Calculate allowance for the time passed. Use double to avoid |
b4cd430a CT |
1614 | // accumulating rounding errors for small intervals. For example, always |
1615 | // adding 1 byte instead of 1.4 results in 29% bandwidth allocation error. | |
1616 | const double gain = timePassed * writeSpeedLimit; | |
1617 | ||
1618 | debugs(77,5, HERE << currTime << " clt" << (const char*)hash.key << ": " << | |
f33d34a8 A |
1619 | bucketSize << " + (" << timePassed << " * " << writeSpeedLimit << |
1620 | " = " << gain << ')'); | |
b4cd430a CT |
1621 | |
1622 | // to further combat error accumulation during micro updates, | |
1623 | // quit before updating time if we cannot add at least one byte | |
1624 | if (gain < 1.0) | |
f33d34a8 | 1625 | return; |
b4cd430a CT |
1626 | |
1627 | prevTime = currTime; | |
1628 | ||
1629 | // for "first" connections, drain initial fat before refilling but keep | |
1630 | // updating prevTime to avoid bursts after the fat is gone | |
1631 | if (bucketSize > bucketSizeLimit) { | |
f33d34a8 A |
1632 | debugs(77,4, HERE << "not refilling while draining initial fat"); |
1633 | return; | |
b4cd430a CT |
1634 | } |
1635 | ||
1636 | bucketSize += gain; | |
1637 | ||
1638 | // obey quota limits | |
1639 | if (bucketSize > bucketSizeLimit) | |
1640 | bucketSize = bucketSizeLimit; | |
1641 | } | |
1642 | ||
f33d34a8 | 1643 | void |
b4cd430a CT |
1644 | ClientInfo::setWriteLimiter(const int aWriteSpeedLimit, const double anInitialBurst, const double aHighWatermark) |
1645 | { | |
f33d34a8 A |
1646 | debugs(77,5, HERE << "Write limits for " << (const char*)hash.key << |
1647 | " speed=" << aWriteSpeedLimit << " burst=" << anInitialBurst << | |
1648 | " highwatermark=" << aHighWatermark); | |
b4cd430a CT |
1649 | |
1650 | // set or possibly update traffic shaping parameters | |
1651 | writeLimitingActive = true; | |
1652 | writeSpeedLimit = aWriteSpeedLimit; | |
1653 | bucketSizeLimit = aHighWatermark; | |
1654 | ||
1655 | // but some members should only be set once for a newly activated bucket | |
1656 | if (firstTimeConnection) { | |
1657 | firstTimeConnection = false; | |
1658 | ||
1659 | assert(!selectWaiting); | |
1660 | assert(!quotaQueue); | |
e9dadd7d | 1661 | quotaQueue = new CommQuotaQueue(this); |
b4cd430a CT |
1662 | |
1663 | bucketSize = anInitialBurst; | |
1664 | prevTime = current_dtime; | |
1665 | } | |
1666 | } | |
1667 | ||
1668 | CommQuotaQueue::CommQuotaQueue(ClientInfo *info): clientInfo(info), | |
f33d34a8 | 1669 | ins(0), outs(0) |
b4cd430a CT |
1670 | { |
1671 | assert(clientInfo); | |
1672 | } | |
1673 | ||
1674 | CommQuotaQueue::~CommQuotaQueue() | |
1675 | { | |
1676 | assert(!clientInfo); // ClientInfo should clear this before destroying us | |
1677 | } | |
1678 | ||
1679 | /// places the given fd at the end of the queue; returns reservation ID | |
1680 | unsigned int | |
1681 | CommQuotaQueue::enqueue(int fd) | |
1682 | { | |
1683 | debugs(77,5, HERE << "clt" << (const char*)clientInfo->hash.key << | |
f33d34a8 | 1684 | ": FD " << fd << " with qqid" << (ins+1) << ' ' << fds.size()); |
b4cd430a CT |
1685 | fds.push_back(fd); |
1686 | return ++ins; | |
1687 | } | |
1688 | ||
1689 | /// removes queue head | |
1690 | void | |
1691 | CommQuotaQueue::dequeue() | |
1692 | { | |
1693 | assert(!fds.empty()); | |
1694 | debugs(77,5, HERE << "clt" << (const char*)clientInfo->hash.key << | |
f33d34a8 A |
1695 | ": FD " << fds.front() << " with qqid" << (outs+1) << ' ' << |
1696 | fds.size()); | |
b4cd430a CT |
1697 | fds.pop_front(); |
1698 | ++outs; | |
1699 | } | |
b4cd430a CT |
1700 | #endif |
1701 | ||
89924214 | 1702 | /* |
1703 | * hm, this might be too general-purpose for all the places we'd | |
1704 | * like to use it. | |
1705 | */ | |
b224ea98 | 1706 | int |
e1381638 AJ |
1707 | ignoreErrno(int ierrno) |
1708 | { | |
603500e7 | 1709 | switch (ierrno) { |
62e76326 | 1710 | |
89924214 | 1711 | case EINPROGRESS: |
62e76326 | 1712 | |
603500e7 | 1713 | case EWOULDBLOCK: |
26a880e2 | 1714 | #if EAGAIN != EWOULDBLOCK |
62e76326 | 1715 | |
603500e7 | 1716 | case EAGAIN: |
26a880e2 | 1717 | #endif |
62e76326 | 1718 | |
603500e7 | 1719 | case EALREADY: |
62e76326 | 1720 | |
603500e7 | 1721 | case EINTR: |
db494ab8 | 1722 | #ifdef ERESTART |
62e76326 | 1723 | |
db494ab8 | 1724 | case ERESTART: |
1725 | #endif | |
62e76326 | 1726 | |
1727 | return 1; | |
1728 | ||
603500e7 | 1729 | default: |
62e76326 | 1730 | return 0; |
603500e7 | 1731 | } |
62e76326 | 1732 | |
603500e7 | 1733 | /* NOTREACHED */ |
26a880e2 | 1734 | } |
d723bf6b | 1735 | |
1736 | void | |
e1381638 AJ |
1737 | commCloseAllSockets(void) |
1738 | { | |
d723bf6b | 1739 | int fd; |
1740 | fde *F = NULL; | |
62e76326 | 1741 | |
95dc7ff4 | 1742 | for (fd = 0; fd <= Biggest_FD; ++fd) { |
62e76326 | 1743 | F = &fd_table[fd]; |
1744 | ||
1745 | if (!F->flags.open) | |
1746 | continue; | |
1747 | ||
1748 | if (F->type != FD_SOCKET) | |
1749 | continue; | |
1750 | ||
1751 | if (F->flags.ipc) /* don't close inter-process sockets */ | |
1752 | continue; | |
1753 | ||
b0469965 | 1754 | if (F->timeoutHandler != NULL) { |
1755 | AsyncCall::Pointer callback = F->timeoutHandler; | |
1756 | F->timeoutHandler = NULL; | |
bf8fe701 | 1757 | debugs(5, 5, "commCloseAllSockets: FD " << fd << ": Calling timeout handler"); |
9e008dda | 1758 | ScheduleCallHere(callback); |
62e76326 | 1759 | } else { |
468fe1b5 | 1760 | debugs(5, 5, "commCloseAllSockets: FD " << fd << ": calling comm_reset_close()"); |
5c336a3b | 1761 | old_comm_reset_close(fd); |
62e76326 | 1762 | } |
d723bf6b | 1763 | } |
1764 | } | |
1b3db6d9 | 1765 | |
2d8c0b1a | 1766 | static bool |
e1381638 AJ |
1767 | AlreadyTimedOut(fde *F) |
1768 | { | |
2d8c0b1a | 1769 | if (!F->flags.open) |
1770 | return true; | |
1771 | ||
1772 | if (F->timeout == 0) | |
1773 | return true; | |
1774 | ||
1775 | if (F->timeout > squid_curtime) | |
1776 | return true; | |
1777 | ||
1778 | return false; | |
1779 | } | |
1780 | ||
5ef5e5cc AJ |
1781 | static bool |
1782 | writeTimedOut(int fd) | |
1783 | { | |
ec41b64c | 1784 | if (!COMMIO_FD_WRITECB(fd)->active()) |
5ef5e5cc AJ |
1785 | return false; |
1786 | ||
1787 | if ((squid_curtime - fd_table[fd].writeStart) < Config.Timeout.write) | |
1788 | return false; | |
1789 | ||
1790 | return true; | |
1791 | } | |
1792 | ||
1b3db6d9 | 1793 | void |
e1381638 AJ |
1794 | checkTimeouts(void) |
1795 | { | |
1b3db6d9 | 1796 | int fd; |
1797 | fde *F = NULL; | |
b0469965 | 1798 | AsyncCall::Pointer callback; |
62e76326 | 1799 | |
95dc7ff4 | 1800 | for (fd = 0; fd <= Biggest_FD; ++fd) { |
62e76326 | 1801 | F = &fd_table[fd]; |
1802 | ||
5ef5e5cc AJ |
1803 | if (writeTimedOut(fd)) { |
1804 | // We have an active write callback and we are timed out | |
52f6ea9e | 1805 | debugs(5, 5, "checkTimeouts: FD " << fd << " auto write timeout"); |
d841c88d | 1806 | Comm::SetSelect(fd, COMM_SELECT_WRITE, NULL, NULL, 0); |
ec41b64c | 1807 | COMMIO_FD_WRITECB(fd)->finish(COMM_ERROR, ETIMEDOUT); |
307b696e | 1808 | } else if (AlreadyTimedOut(F)) |
62e76326 | 1809 | continue; |
1810 | ||
9e008dda | 1811 | debugs(5, 5, "checkTimeouts: FD " << fd << " Expired"); |
62e76326 | 1812 | |
b0469965 | 1813 | if (F->timeoutHandler != NULL) { |
bf8fe701 | 1814 | debugs(5, 5, "checkTimeouts: FD " << fd << ": Call timeout handler"); |
b0469965 | 1815 | callback = F->timeoutHandler; |
1816 | F->timeoutHandler = NULL; | |
9e008dda | 1817 | ScheduleCallHere(callback); |
62e76326 | 1818 | } else { |
bf8fe701 | 1819 | debugs(5, 5, "checkTimeouts: FD " << fd << ": Forcing comm_close()"); |
62e76326 | 1820 | comm_close(fd); |
1821 | } | |
b5443c04 | 1822 | } |
1823 | } | |
1824 | ||
9e008dda AJ |
1825 | /// Start waiting for a possibly half-closed connection to close |
1826 | // by scheduling a read callback to a monitoring handler that | |
82ec8dfc | 1827 | // will close the connection on read errors. |
a46d2c0e | 1828 | void |
e1381638 AJ |
1829 | commStartHalfClosedMonitor(int fd) |
1830 | { | |
74257126 | 1831 | debugs(5, 5, HERE << "adding FD " << fd << " to " << *TheHalfClosed); |
82ec8dfc AR |
1832 | assert(isOpen(fd)); |
1833 | assert(!commHasHalfClosedMonitor(fd)); | |
74257126 AR |
1834 | (void)TheHalfClosed->add(fd); // could also assert the result |
1835 | commPlanHalfClosedCheck(); // may schedule check if we added the first FD | |
1836 | } | |
1837 | ||
1838 | static | |
1839 | void | |
e1381638 AJ |
1840 | commPlanHalfClosedCheck() |
1841 | { | |
74257126 AR |
1842 | if (!WillCheckHalfClosed && !TheHalfClosed->empty()) { |
1843 | eventAdd("commHalfClosedCheck", &commHalfClosedCheck, NULL, 1.0, 1); | |
1844 | WillCheckHalfClosed = true; | |
1845 | } | |
1846 | } | |
1847 | ||
1848 | /// iterates over all descriptors that may need half-closed tests and | |
1849 | /// calls comm_read for those that do; re-schedules the check if needed | |
1850 | static | |
1851 | void | |
e1381638 AJ |
1852 | commHalfClosedCheck(void *) |
1853 | { | |
74257126 AR |
1854 | debugs(5, 5, HERE << "checking " << *TheHalfClosed); |
1855 | ||
1856 | typedef DescriptorSet::const_iterator DSCI; | |
1857 | const DSCI end = TheHalfClosed->end(); | |
1858 | for (DSCI i = TheHalfClosed->begin(); i != end; ++i) { | |
ec20038e AJ |
1859 | Comm::ConnectionPointer c = new Comm::Connection; // XXX: temporary. make HalfClosed a list of these. |
1860 | c->fd = *i; | |
1861 | if (!fd_table[c->fd].halfClosedReader) { // not reading already | |
74257126 | 1862 | AsyncCall::Pointer call = commCbCall(5,4, "commHalfClosedReader", |
9e008dda | 1863 | CommIoCbPtrFun(&commHalfClosedReader, NULL)); |
ec20038e AJ |
1864 | comm_read(c, NULL, 0, call); |
1865 | fd_table[c->fd].halfClosedReader = call; | |
1866 | } else | |
1867 | c->fd = -1; // XXX: temporary. prevent c replacement erase closing listed FD | |
74257126 | 1868 | } |
f900210a | 1869 | |
74257126 AR |
1870 | WillCheckHalfClosed = false; // as far as we know |
1871 | commPlanHalfClosedCheck(); // may need to check again | |
f900210a | 1872 | } |
1873 | ||
82ec8dfc AR |
1874 | /// checks whether we are waiting for possibly half-closed connection to close |
1875 | // We are monitoring if the read handler for the fd is the monitoring handler. | |
1876 | bool | |
e1381638 AJ |
1877 | commHasHalfClosedMonitor(int fd) |
1878 | { | |
74257126 | 1879 | return TheHalfClosed->has(fd); |
a46d2c0e | 1880 | } |
1881 | ||
82ec8dfc AR |
1882 | /// stop waiting for possibly half-closed connection to close |
1883 | static void | |
e1381638 AJ |
1884 | commStopHalfClosedMonitor(int const fd) |
1885 | { | |
74257126 AR |
1886 | debugs(5, 5, HERE << "removing FD " << fd << " from " << *TheHalfClosed); |
1887 | ||
1888 | // cancel the read if one was scheduled | |
1889 | AsyncCall::Pointer reader = fd_table[fd].halfClosedReader; | |
1890 | if (reader != NULL) | |
1891 | comm_read_cancel(fd, reader); | |
1892 | fd_table[fd].halfClosedReader = NULL; | |
1893 | ||
1894 | TheHalfClosed->del(fd); | |
a46d2c0e | 1895 | } |
1896 | ||
82ec8dfc AR |
1897 | /// I/O handler for the possibly half-closed connection monitoring code |
1898 | static void | |
e0d28505 | 1899 | commHalfClosedReader(const Comm::ConnectionPointer &conn, char *, size_t size, comm_err_t flag, int, void *) |
e1381638 | 1900 | { |
82ec8dfc | 1901 | // there cannot be more data coming in on half-closed connections |
9e008dda | 1902 | assert(size == 0); |
e0d28505 AJ |
1903 | assert(conn != NULL); |
1904 | assert(commHasHalfClosedMonitor(conn->fd)); // or we would have canceled the read | |
74257126 | 1905 | |
e0d28505 | 1906 | fd_table[conn->fd].halfClosedReader = NULL; // done reading, for now |
a46d2c0e | 1907 | |
82ec8dfc AR |
1908 | // nothing to do if fd is being closed |
1909 | if (flag == COMM_ERR_CLOSING) | |
1910 | return; | |
a46d2c0e | 1911 | |
82ec8dfc AR |
1912 | // if read failed, close the connection |
1913 | if (flag != COMM_OK) { | |
e0d28505 | 1914 | debugs(5, 3, HERE << "closing " << conn); |
80463bb4 | 1915 | conn->close(); |
82ec8dfc AR |
1916 | return; |
1917 | } | |
a46d2c0e | 1918 | |
82ec8dfc | 1919 | // continue waiting for close or error |
74257126 | 1920 | commPlanHalfClosedCheck(); // make sure this fd will be checked again |
a46d2c0e | 1921 | } |
1922 | ||
a46d2c0e | 1923 | |
3e4bebf8 | 1924 | CommRead::CommRead() : conn(NULL), buf(NULL), len(0), callback(NULL) {} |
a46d2c0e | 1925 | |
3e4bebf8 AJ |
1926 | CommRead::CommRead(const Comm::ConnectionPointer &c, char *buf_, int len_, AsyncCall::Pointer &callback_) |
1927 | : conn(c), buf(buf_), len(len_), callback(callback_) {} | |
a46d2c0e | 1928 | |
a50bfe93 | 1929 | DeferredRead::DeferredRead () : theReader(NULL), theContext(NULL), theRead(), cancelled(false) {} |
a46d2c0e | 1930 | |
a50bfe93 | 1931 | DeferredRead::DeferredRead (DeferrableRead *aReader, void *data, CommRead const &aRead) : theReader(aReader), theContext (data), theRead(aRead), cancelled(false) {} |
a46d2c0e | 1932 | |
e1381638 AJ |
1933 | DeferredReadManager::~DeferredReadManager() |
1934 | { | |
a46d2c0e | 1935 | flushReads(); |
1936 | assert (deferredReads.empty()); | |
1937 | } | |
1938 | ||
97427e90 | 1939 | /* explicit instantiation required for some systems */ |
1940 | ||
63be0a78 | 1941 | /// \cond AUTODOCS-IGNORE |
2236466c | 1942 | template cbdata_type CbDataList<DeferredRead>::CBDATA_CbDataList; |
63be0a78 | 1943 | /// \endcond |
97427e90 | 1944 | |
a46d2c0e | 1945 | void |
e1381638 AJ |
1946 | DeferredReadManager::delayRead(DeferredRead const &aRead) |
1947 | { | |
3e4bebf8 | 1948 | debugs(5, 3, "Adding deferred read on " << aRead.theRead.conn); |
2236466c | 1949 | CbDataList<DeferredRead> *temp = deferredReads.push_back(aRead); |
2796c0d7 | 1950 | |
9e008dda | 1951 | // We have to use a global function as a closer and point to temp |
2796c0d7 AR |
1952 | // instead of "this" because DeferredReadManager is not a job and |
1953 | // is not even cbdata protected | |
575d05c4 | 1954 | // XXX: and yet we use cbdata protection functions on it?? |
2796c0d7 | 1955 | AsyncCall::Pointer closer = commCbCall(5,4, |
9e008dda AJ |
1956 | "DeferredReadManager::CloseHandler", |
1957 | CommCloseCbPtrFun(&CloseHandler, temp)); | |
3e4bebf8 | 1958 | comm_add_close_handler(aRead.theRead.conn->fd, closer); |
2796c0d7 | 1959 | temp->element.closer = closer; // remeber so that we can cancel |
a46d2c0e | 1960 | } |
1961 | ||
1962 | void | |
575d05c4 | 1963 | DeferredReadManager::CloseHandler(const CommCloseCbParams ¶ms) |
e1381638 | 1964 | { |
575d05c4 | 1965 | if (!cbdataReferenceValid(params.data)) |
a46d2c0e | 1966 | return; |
1967 | ||
575d05c4 | 1968 | CbDataList<DeferredRead> *temp = (CbDataList<DeferredRead> *)params.data; |
a46d2c0e | 1969 | |
2796c0d7 | 1970 | temp->element.closer = NULL; |
a46d2c0e | 1971 | temp->element.markCancelled(); |
1972 | } | |
1973 | ||
1974 | DeferredRead | |
e1381638 AJ |
1975 | DeferredReadManager::popHead(CbDataListContainer<DeferredRead> &deferredReads) |
1976 | { | |
a46d2c0e | 1977 | assert (!deferredReads.empty()); |
1978 | ||
2796c0d7 | 1979 | DeferredRead &read = deferredReads.head->element; |
e7e8ebdb AJ |
1980 | |
1981 | // NOTE: at this point the connection has been paused/stalled for an unknown | |
1982 | // amount of time. We must re-validate that it is active and usable. | |
1983 | ||
1984 | // If the connection has been closed already. Cancel this read. | |
1985 | if (!Comm::IsConnOpen(read.theRead.conn)) { | |
1986 | if (read.closer != NULL) { | |
1987 | read.closer->cancel("Connection closed before."); | |
1988 | read.closer = NULL; | |
1989 | } | |
1990 | read.markCancelled(); | |
1991 | } | |
1992 | ||
2796c0d7 | 1993 | if (!read.cancelled) { |
3e4bebf8 | 1994 | comm_remove_close_handler(read.theRead.conn->fd, read.closer); |
2796c0d7 AR |
1995 | read.closer = NULL; |
1996 | } | |
a46d2c0e | 1997 | |
1998 | DeferredRead result = deferredReads.pop_front(); | |
1999 | ||
2000 | return result; | |
2001 | } | |
2002 | ||
2003 | void | |
e1381638 AJ |
2004 | DeferredReadManager::kickReads(int const count) |
2005 | { | |
2236466c | 2006 | /* if we had CbDataList::size() we could consolidate this and flushReads */ |
a46d2c0e | 2007 | |
33cea91c | 2008 | if (count < 1) { |
a46d2c0e | 2009 | flushReads(); |
33cea91c | 2010 | return; |
2011 | } | |
a46d2c0e | 2012 | |
2013 | size_t remaining = count; | |
2014 | ||
2015 | while (!deferredReads.empty() && remaining) { | |
2016 | DeferredRead aRead = popHead(deferredReads); | |
2017 | kickARead(aRead); | |
2018 | ||
2019 | if (!aRead.cancelled) | |
2020 | --remaining; | |
2021 | } | |
2022 | } | |
2023 | ||
2024 | void | |
e1381638 AJ |
2025 | DeferredReadManager::flushReads() |
2026 | { | |
2236466c | 2027 | CbDataListContainer<DeferredRead> reads; |
a46d2c0e | 2028 | reads = deferredReads; |
2236466c | 2029 | deferredReads = CbDataListContainer<DeferredRead>(); |
a46d2c0e | 2030 | |
27b24462 | 2031 | // XXX: For fairness this SHOULD randomize the order |
a46d2c0e | 2032 | while (!reads.empty()) { |
2033 | DeferredRead aRead = popHead(reads); | |
2034 | kickARead(aRead); | |
2035 | } | |
2036 | } | |
2037 | ||
2038 | void | |
e1381638 AJ |
2039 | DeferredReadManager::kickARead(DeferredRead const &aRead) |
2040 | { | |
a46d2c0e | 2041 | if (aRead.cancelled) |
2042 | return; | |
2043 | ||
3e4bebf8 | 2044 | if (Comm::IsConnOpen(aRead.theRead.conn) && fd_table[aRead.theRead.conn->fd].closing()) |
af6a12ee | 2045 | return; |
ed2c738b | 2046 | |
3e4bebf8 | 2047 | debugs(5, 3, "Kicking deferred read on " << aRead.theRead.conn); |
a46d2c0e | 2048 | |
2049 | aRead.theReader(aRead.theContext, aRead.theRead); | |
2050 | } | |
2051 | ||
2052 | void | |
e1381638 AJ |
2053 | DeferredRead::markCancelled() |
2054 | { | |
a46d2c0e | 2055 | cancelled = true; |
2056 | } | |
2d8c0b1a | 2057 | |
8ff3fa2e | 2058 | int |
e1381638 AJ |
2059 | CommSelectEngine::checkEvents(int timeout) |
2060 | { | |
fa3f745b | 2061 | static time_t last_timeout = 0; |
2062 | ||
2063 | /* No, this shouldn't be here. But it shouldn't be in each comm handler. -adrian */ | |
2064 | if (squid_curtime > last_timeout) { | |
2065 | last_timeout = squid_curtime; | |
2066 | checkTimeouts(); | |
2067 | } | |
2068 | ||
d841c88d | 2069 | switch (Comm::DoSelect(timeout)) { |
8ff3fa2e | 2070 | |
2071 | case COMM_OK: | |
2072 | ||
2073 | case COMM_TIMEOUT: | |
2074 | return 0; | |
2075 | ||
2076 | case COMM_IDLE: | |
2077 | ||
2078 | case COMM_SHUTDOWN: | |
2079 | return EVENT_IDLE; | |
2080 | ||
2081 | case COMM_ERROR: | |
2082 | return EVENT_ERROR; | |
2083 | ||
2084 | default: | |
2085 | fatal_dump("comm.cc: Internal error -- this should never happen."); | |
2086 | return EVENT_ERROR; | |
2087 | }; | |
2088 | } | |
10cefb7b | 2089 | |
0ffda73c | 2090 | /// Create a unix-domain socket (UDS) that only supports FD_MSGHDR I/O. |
10cefb7b | 2091 | int |
2092 | comm_open_uds(int sock_type, | |
2093 | int proto, | |
2094 | struct sockaddr_un* addr, | |
2095 | int flags) | |
2096 | { | |
a67d2b2e | 2097 | // TODO: merge with comm_openex() when Ip::Address becomes NetAddress |
ba568924 | 2098 | |
10cefb7b | 2099 | int new_socket; |
10cefb7b | 2100 | |
2101 | PROF_start(comm_open); | |
10cefb7b | 2102 | /* Create socket for accepting new connections. */ |
95dc7ff4 | 2103 | ++ statCounter.syscalls.sock.sockets; |
10cefb7b | 2104 | |
2105 | /* Setup the socket addrinfo details for use */ | |
ba568924 | 2106 | struct addrinfo AI; |
10cefb7b | 2107 | AI.ai_flags = 0; |
2108 | AI.ai_family = PF_UNIX; | |
2109 | AI.ai_socktype = sock_type; | |
2110 | AI.ai_protocol = proto; | |
2111 | AI.ai_addrlen = SUN_LEN(addr); | |
2112 | AI.ai_addr = (sockaddr*)addr; | |
2113 | AI.ai_canonname = NULL; | |
2114 | AI.ai_next = NULL; | |
2115 | ||
ba568924 | 2116 | debugs(50, 3, HERE << "Attempt open socket for: " << addr->sun_path); |
10cefb7b | 2117 | |
2118 | if ((new_socket = socket(AI.ai_family, AI.ai_socktype, AI.ai_protocol)) < 0) { | |
2119 | /* Increase the number of reserved fd's if calls to socket() | |
2120 | * are failing because the open file table is full. This | |
2121 | * limits the number of simultaneous clients */ | |
2122 | ||
2123 | if (limitError(errno)) { | |
ba568924 | 2124 | debugs(50, DBG_IMPORTANT, HERE << "socket failure: " << xstrerror()); |
10cefb7b | 2125 | fdAdjustReserved(); |
2126 | } else { | |
ba568924 | 2127 | debugs(50, DBG_CRITICAL, HERE << "socket failure: " << xstrerror()); |
10cefb7b | 2128 | } |
2129 | ||
2130 | PROF_stop(comm_open); | |
2131 | return -1; | |
2132 | } | |
2133 | ||
ba568924 | 2134 | debugs(50, 3, HERE "Opened UDS FD " << new_socket << " : family=" << AI.ai_family << ", type=" << AI.ai_socktype << ", protocol=" << AI.ai_protocol); |
10cefb7b | 2135 | |
2136 | /* update fdstat */ | |
ba568924 | 2137 | debugs(50, 5, HERE << "FD " << new_socket << " is a new socket"); |
10cefb7b | 2138 | |
2139 | assert(!isOpen(new_socket)); | |
1bac0258 | 2140 | fd_open(new_socket, FD_MSGHDR, NULL); |
10cefb7b | 2141 | |
2142 | fdd_table[new_socket].close_file = NULL; | |
ba568924 | 2143 | |
10cefb7b | 2144 | fdd_table[new_socket].close_line = 0; |
2145 | ||
ba568924 AR |
2146 | fd_table[new_socket].sock_family = AI.ai_family; |
2147 | ||
10cefb7b | 2148 | if (!(flags & COMM_NOCLOEXEC)) |
2149 | commSetCloseOnExec(new_socket); | |
2150 | ||
2151 | if (flags & COMM_REUSEADDR) | |
2152 | commSetReuseAddr(new_socket); | |
2153 | ||
2154 | if (flags & COMM_NONBLOCKING) { | |
2155 | if (commSetNonBlocking(new_socket) != COMM_OK) { | |
2156 | comm_close(new_socket); | |
2157 | PROF_stop(comm_open); | |
2158 | return -1; | |
2159 | } | |
2160 | } | |
2161 | ||
2162 | if (flags & COMM_DOBIND) { | |
2163 | if (commBind(new_socket, AI) != COMM_OK) { | |
2164 | comm_close(new_socket); | |
2165 | PROF_stop(comm_open); | |
2166 | return -1; | |
2167 | } | |
2168 | } | |
2169 | ||
2170 | #ifdef TCP_NODELAY | |
2171 | if (sock_type == SOCK_STREAM) | |
2172 | commSetTcpNoDelay(new_socket); | |
2173 | ||
2174 | #endif | |
2175 | ||
2176 | if (Config.tcpRcvBufsz > 0 && sock_type == SOCK_STREAM) | |
2177 | commSetTcpRcvbuf(new_socket, Config.tcpRcvBufsz); | |
2178 | ||
2179 | PROF_stop(comm_open); | |
2180 | ||
2181 | return new_socket; | |
2182 | } |