]>
Commit | Line | Data |
---|---|---|
da2b3a17 | 1 | |
30a4f2a8 | 2 | /* |
d88f85f2 | 3 | * $Id: comm.cc,v 1.363 2003/02/09 17:24:02 hno Exp $ |
30a4f2a8 | 4 | * |
5 | * DEBUG: section 5 Socket Functions | |
6 | * AUTHOR: Harvest Derived | |
7 | * | |
2b6662ba | 8 | * SQUID Web Proxy Cache http://www.squid-cache.org/ |
e25c139f | 9 | * ---------------------------------------------------------- |
30a4f2a8 | 10 | * |
2b6662ba | 11 | * Squid is the result of efforts by numerous individuals from |
12 | * the Internet community; see the CONTRIBUTORS file for full | |
13 | * details. Many organizations have provided support for Squid's | |
14 | * development; see the SPONSORS file for full details. Squid is | |
15 | * Copyrighted (C) 2001 by the Regents of the University of | |
16 | * California; see the COPYRIGHT file for full details. Squid | |
17 | * incorporates software developed and/or copyrighted by other | |
18 | * sources; see the CREDITS file for full details. | |
30a4f2a8 | 19 | * |
20 | * This program is free software; you can redistribute it and/or modify | |
21 | * it under the terms of the GNU General Public License as published by | |
22 | * the Free Software Foundation; either version 2 of the License, or | |
23 | * (at your option) any later version. | |
24 | * | |
25 | * This program is distributed in the hope that it will be useful, | |
26 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
27 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
28 | * GNU General Public License for more details. | |
29 | * | |
30 | * You should have received a copy of the GNU General Public License | |
31 | * along with this program; if not, write to the Free Software | |
cbdec147 | 32 | * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111, USA. |
e25c139f | 33 | * |
30a4f2a8 | 34 | */ |
090089c4 | 35 | |
44a47c6e | 36 | #include "squid.h" |
c4b7a5a9 | 37 | #include "StoreIOBuffer.h" |
38 | #include "comm.h" | |
528b2c61 | 39 | #include "fde.h" |
090089c4 | 40 | |
b671cc68 | 41 | #if defined(_SQUID_CYGWIN_) |
42 | #include <sys/ioctl.h> | |
43 | #endif | |
30a4f2a8 | 44 | #ifdef HAVE_NETINET_TCP_H |
45 | #include <netinet/tcp.h> | |
46 | #endif | |
090089c4 | 47 | |
f49a8979 | 48 | /* |
49 | * This magic determines how many times to call accept() | |
50 | * at a go. | |
51 | */ | |
52 | #define MAX_ACCEPT_PER_LOOP 10 | |
53 | ||
f88211e8 | 54 | typedef struct { |
55 | char *host; | |
56 | u_short port; | |
57 | struct sockaddr_in S; | |
58 | CNCB *callback; | |
59 | void *data; | |
f88211e8 | 60 | struct in_addr in_addr; |
61 | int locks; | |
03a1ee42 | 62 | int fd; |
22c653cd | 63 | int tries; |
64 | int addrcount; | |
65 | int connstart; | |
f88211e8 | 66 | } ConnectStateData; |
67 | ||
090089c4 | 68 | /* STATIC */ |
e6ccf245 | 69 | static comm_err_t commBind(int s, struct in_addr, u_short port); |
f5b8bbc4 | 70 | static void commSetReuseAddr(int); |
71 | static void commSetNoLinger(int); | |
3d7e9d7c | 72 | static void CommWriteStateCallbackAndFree(int fd, comm_err_t code); |
30a4f2a8 | 73 | #ifdef TCP_NODELAY |
f5b8bbc4 | 74 | static void commSetTcpNoDelay(int); |
30a4f2a8 | 75 | #endif |
f5b8bbc4 | 76 | static void commSetTcpRcvbuf(int, int); |
f88211e8 | 77 | static PF commConnectFree; |
03a1ee42 | 78 | static PF commConnectHandle; |
79 | static PF commHandleWrite; | |
edeb28fd | 80 | static IPH commConnectDnsHandle; |
3d7e9d7c | 81 | static void commConnectCallback(ConnectStateData * cs, comm_err_t status); |
22c653cd | 82 | static int commResetFD(ConnectStateData * cs); |
83 | static int commRetryConnect(ConnectStateData * cs); | |
28c60158 | 84 | CBDATA_TYPE(ConnectStateData); |
723123a9 | 85 | |
3c1a197f | 86 | static PF comm_accept_try; |
c4b7a5a9 | 87 | |
88 | struct _fdc_t { | |
89 | int active; | |
2fb9d240 | 90 | int fd; |
c4b7a5a9 | 91 | dlink_list CommCallbackList; |
92 | struct { | |
93 | char *buf; | |
94 | int size; | |
95 | IOCB *handler; | |
96 | void *handler_data; | |
97 | } read; | |
d4cb310b | 98 | struct { |
dba1a67d | 99 | const char *buf; |
d4cb310b | 100 | int size; |
101 | int curofs; | |
102 | IOCB *handler; | |
103 | void *handler_data; | |
104 | } write; | |
c4b7a5a9 | 105 | struct { |
2fb9d240 | 106 | /* how often (in msec) to re-check if we're out of fds on an accept() */ |
107 | int check_delay; | |
c4b7a5a9 | 108 | struct sockaddr_in me; |
109 | struct sockaddr_in pn; | |
110 | IOACB *handler; | |
111 | void *handler_data; | |
112 | } accept; | |
113 | struct CommFiller { | |
114 | StoreIOBuffer requestedData; | |
115 | size_t amountDone; | |
116 | IOFCB *handler; | |
117 | void *handler_data; | |
118 | } fill; | |
119 | ||
120 | }; | |
121 | typedef struct _fdc_t fdc_t; | |
122 | ||
123 | typedef enum { | |
124 | COMM_CB_READ = 1, | |
125 | COMM_CB_WRITE, | |
126 | COMM_CB_ACCEPT, | |
127 | COMM_CB_FILL | |
128 | } comm_callback_t; | |
129 | ||
130 | struct _CommCallbackData { | |
131 | comm_callback_t type; | |
132 | dlink_node fd_node; | |
133 | dlink_node h_node; | |
134 | int fd; | |
135 | int newfd; /* for accept() */ | |
136 | char *buf; | |
137 | int retval; | |
138 | union { | |
139 | IOCB *r_callback; | |
140 | IOACB *a_callback; | |
141 | IOFCB *f_callback; | |
d4cb310b | 142 | IOWCB *w_callback; |
c4b7a5a9 | 143 | } c; |
144 | void *callback_data; | |
145 | comm_err_t errcode; | |
146 | int xerrno; | |
147 | int seqnum; | |
148 | struct sockaddr_in me; | |
149 | struct sockaddr_in pn; | |
150 | StoreIOBuffer sb; | |
151 | }; | |
152 | typedef struct _CommCallbackData CommCallbackData; | |
153 | ||
154 | struct _fd_debug_t { | |
155 | char *close_file; | |
156 | int close_line; | |
157 | }; | |
158 | typedef struct _fd_debug_t fd_debug_t; | |
159 | ||
723123a9 | 160 | static MemPool *comm_write_pool = NULL; |
58cd5bbd | 161 | static MemPool *conn_close_pool = NULL; |
c4b7a5a9 | 162 | static MemPool *comm_callback_pool = NULL; |
163 | fdc_t *fdc_table = NULL; | |
164 | fd_debug_t *fdd_table = NULL; | |
165 | dlink_list CommCallbackList; | |
166 | static int CommCallbackSeqnum = 1; | |
167 | ||
168 | ||
169 | /* New and improved stuff */ | |
170 | ||
171 | /* | |
172 | * return whether there are entries in the callback queue | |
173 | */ | |
174 | int | |
175 | comm_existsiocallback(void) | |
176 | { | |
177 | return CommCallbackList.head == NULL; | |
178 | } | |
179 | ||
180 | /* | |
181 | * add an IO callback | |
182 | * | |
183 | * IO callbacks are added when we want to notify someone that some IO | |
184 | * has finished but we don't want to risk re-entering a non-reentrant | |
185 | * code block. | |
186 | */ | |
187 | static void | |
188 | comm_addreadcallback(int fd, IOCB *callback, char *buf, size_t retval, comm_err_t errcode, | |
189 | int xerrno, void *callback_data) | |
190 | { | |
191 | CommCallbackData *cio; | |
192 | ||
193 | assert(fdc_table[fd].active == 1); | |
194 | ||
195 | /* Allocate a new struct */ | |
196 | cio = (CommCallbackData *)memPoolAlloc(comm_callback_pool); | |
197 | ||
198 | /* Throw our data into it */ | |
199 | cio->fd = fd; | |
200 | cio->retval = retval; | |
201 | cio->xerrno = xerrno; | |
202 | cio->errcode = errcode; | |
203 | cio->c.r_callback = callback; | |
204 | cio->callback_data = callback_data; | |
205 | cio->seqnum = CommCallbackSeqnum; | |
206 | cio->buf = buf; | |
207 | cio->type = COMM_CB_READ; | |
208 | ||
209 | /* Add it to the end of the list */ | |
210 | dlinkAddTail(cio, &(cio->h_node), &CommCallbackList); | |
211 | ||
212 | /* and add it to the end of the fd list */ | |
213 | dlinkAddTail(cio, &(cio->fd_node), &(fdc_table[fd].CommCallbackList)); | |
214 | ||
215 | } | |
216 | ||
217 | ||
218 | static void | |
219 | comm_addacceptcallback(int fd, int newfd, IOACB *callback, struct sockaddr_in *pn, | |
220 | struct sockaddr_in *me, comm_err_t errcode, int xerrno, void *callback_data) | |
221 | { | |
222 | CommCallbackData *cio; | |
223 | ||
224 | assert(fdc_table[fd].active == 1); | |
225 | ||
226 | /* Allocate a new struct */ | |
227 | cio = (CommCallbackData *)memPoolAlloc(comm_callback_pool); | |
228 | ||
229 | /* Throw our data into it */ | |
230 | cio->fd = fd; | |
231 | cio->xerrno = xerrno; | |
232 | cio->errcode = errcode; | |
233 | cio->c.a_callback = callback; | |
234 | cio->callback_data = callback_data; | |
235 | cio->seqnum = CommCallbackSeqnum; | |
236 | cio->type = COMM_CB_ACCEPT; | |
237 | cio->newfd = newfd; | |
238 | cio->pn = *pn; | |
239 | cio->me = *me; | |
240 | ||
241 | /* Add it to the end of the list */ | |
242 | dlinkAddTail(cio, &(cio->h_node), &CommCallbackList); | |
243 | ||
244 | /* and add it to the end of the fd list */ | |
245 | dlinkAddTail(cio, &(cio->fd_node), &(fdc_table[fd].CommCallbackList)); | |
246 | ||
247 | } | |
248 | ||
249 | static void | |
250 | comm_add_fill_callback(int fd, size_t retval, comm_err_t errcode, int xerrno) | |
251 | { | |
252 | CommCallbackData *cio; | |
253 | ||
254 | assert(fdc_table[fd].active == 1); | |
255 | ||
256 | /* Allocate a new struct */ | |
257 | cio = (CommCallbackData *)memPoolAlloc(comm_callback_pool); | |
258 | ||
259 | /* Throw our data into it */ | |
260 | cio->fd = fd; | |
261 | cio->xerrno = xerrno; | |
262 | cio->errcode = errcode; | |
263 | cio->c.f_callback = fdc_table[fd].fill.handler; | |
264 | cio->callback_data = fdc_table[fd].fill.handler_data; | |
265 | cio->seqnum = CommCallbackSeqnum; | |
266 | cio->type = COMM_CB_FILL; | |
267 | /* retval not used */ | |
268 | cio->retval = -1; | |
269 | cio->sb = fdc_table[fd].fill.requestedData; | |
270 | cio->sb.length = retval; | |
271 | /* Clear out fd state */ | |
272 | fdc_table[fd].fill.handler = NULL; | |
273 | fdc_table[fd].fill.handler_data = NULL; | |
274 | ||
275 | /* Add it to the end of the list */ | |
276 | dlinkAddTail(cio, &(cio->h_node), &CommCallbackList); | |
277 | ||
278 | /* and add it to the end of the fd list */ | |
279 | dlinkAddTail(cio, &(cio->fd_node), &(fdc_table[fd].CommCallbackList)); | |
280 | } | |
281 | ||
d4cb310b | 282 | static void |
283 | comm_add_write_callback(int fd, size_t retval, comm_err_t errcode, int xerrno) | |
284 | { | |
285 | CommCallbackData *cio; | |
286 | ||
287 | assert(fdc_table[fd].active == 1); | |
288 | ||
289 | /* Allocate a new struct */ | |
290 | cio = (CommCallbackData *)memPoolAlloc(comm_callback_pool); | |
291 | ||
292 | /* Throw our data into it */ | |
293 | cio->fd = fd; | |
294 | cio->xerrno = xerrno; | |
295 | cio->errcode = errcode; | |
296 | cio->c.w_callback = fdc_table[fd].write.handler; | |
91360be7 | 297 | cio->callback_data = fdc_table[fd].write.handler_data; |
d4cb310b | 298 | cio->seqnum = CommCallbackSeqnum; |
299 | cio->type = COMM_CB_WRITE; | |
300 | cio->retval = retval; | |
301 | ||
302 | /* Clear out fd state */ | |
303 | fdc_table[fd].write.handler = NULL; | |
304 | fdc_table[fd].write.handler_data = NULL; | |
305 | ||
306 | /* Add it to the end of the list */ | |
307 | dlinkAddTail(cio, &(cio->h_node), &CommCallbackList); | |
308 | ||
309 | /* and add it to the end of the fd list */ | |
310 | dlinkAddTail(cio, &(cio->fd_node), &(fdc_table[fd].CommCallbackList)); | |
311 | } | |
c4b7a5a9 | 312 | |
313 | ||
314 | ||
315 | static void | |
316 | comm_call_io_callback(CommCallbackData *cio) | |
317 | { | |
318 | switch(cio->type) { | |
319 | case COMM_CB_READ: | |
320 | cio->c.r_callback(cio->fd, cio->buf, cio->retval, cio->errcode, cio->xerrno, | |
321 | cio->callback_data); | |
322 | break; | |
323 | case COMM_CB_WRITE: | |
d4cb310b | 324 | cio->c.w_callback(cio->fd, cio->buf, cio->retval, cio->errcode, cio->xerrno, |
325 | cio->callback_data); | |
c4b7a5a9 | 326 | break; |
327 | case COMM_CB_ACCEPT: | |
328 | cio->c.a_callback(cio->fd, cio->newfd, &cio->me, &cio->pn, cio->errcode, | |
329 | cio->xerrno, cio->callback_data); | |
330 | break; | |
331 | case COMM_CB_FILL: | |
332 | cio->c.f_callback(cio->fd, cio->sb, cio->errcode, | |
333 | cio->xerrno, cio->callback_data); | |
334 | break; | |
335 | default: | |
336 | fatal("unknown comm io callback type!"); | |
337 | break; | |
338 | }; | |
339 | } | |
340 | ||
341 | ||
342 | /* | |
343 | * call the IO callbacks | |
344 | * | |
345 | * This should be called before comm_select() so code can attempt to | |
346 | * initiate some IO. | |
347 | * | |
348 | * When io callbacks are added, they are added with the current | |
349 | * sequence number. The sequence number is incremented in this routine - | |
350 | * since callbacks are added to the _tail_ of the list, when we hit a | |
351 | * callback with a seqnum _not_ what it was when we entered this routine, | |
352 | * we can stop. | |
353 | */ | |
354 | void | |
355 | comm_calliocallback(void) | |
356 | { | |
357 | CommCallbackData *cio; | |
358 | dlink_node *node; | |
359 | int oldseqnum = CommCallbackSeqnum; | |
360 | ||
361 | /* Call our callbacks until we hit NULL or the seqnum changes */ | |
362 | while (CommCallbackList.head != NULL) { | |
363 | node = (dlink_node *)CommCallbackList.head; | |
364 | cio = (CommCallbackData *)node->data; | |
365 | ||
366 | /* If seqnum isn't the same, its time to die */ | |
367 | if (cio->seqnum != oldseqnum) | |
368 | break; /* we've hit newly-added events */ | |
369 | ||
370 | assert(fdc_table[cio->fd].active == 1); | |
371 | ||
372 | dlinkDelete(&cio->h_node, &CommCallbackList); | |
373 | dlinkDelete(&cio->fd_node, &(fdc_table[cio->fd].CommCallbackList)); | |
374 | comm_call_io_callback(cio); | |
375 | memPoolFree(comm_callback_pool, cio); | |
376 | } | |
377 | } | |
378 | ||
379 | ||
380 | /* | |
381 | * Queue a callback | |
382 | */ | |
383 | static void | |
384 | comm_read_callback(int fd, int retval, comm_err_t errcode, int xerrno) | |
385 | { | |
386 | fdc_t *Fc = &fdc_table[fd]; | |
387 | ||
388 | assert(Fc->read.handler != NULL); | |
389 | ||
390 | comm_addreadcallback(fd, Fc->read.handler, Fc->read.buf, retval, errcode, xerrno, | |
391 | Fc->read.handler_data); | |
392 | Fc->read.handler = NULL; | |
393 | Fc->read.handler_data = NULL; | |
394 | } | |
395 | ||
396 | /* | |
397 | * Attempt a read | |
398 | * | |
399 | * If the read attempt succeeds or fails, call the callback. | |
400 | * Else, wait for another IO notification. | |
401 | */ | |
402 | static void | |
403 | comm_read_try(int fd, void *data) | |
404 | { | |
405 | fdc_t *Fc = &fdc_table[fd]; | |
406 | int retval; | |
407 | ||
408 | /* make sure we actually have a callback */ | |
409 | assert(Fc->read.handler != NULL); | |
410 | ||
411 | /* Attempt a read */ | |
412 | statCounter.syscalls.sock.reads++; | |
2fb9d240 | 413 | errno = 0; |
c4b7a5a9 | 414 | retval = FD_READ_METHOD(fd, Fc->read.buf, Fc->read.size); |
2fb9d240 | 415 | debug(5, 3) ("comm_read_try: fd %d, size %d, retval %d, errno %d\n", |
416 | fd, Fc->read.size, retval, errno); | |
c4b7a5a9 | 417 | if (retval < 0 && !ignoreErrno(errno)) { |
2fb9d240 | 418 | debug(5, 3) ("comm_read_try: scheduling COMM_ERROR\n"); |
c4b7a5a9 | 419 | comm_read_callback(fd, -1, COMM_ERROR, errno); |
420 | return; | |
421 | }; | |
422 | ||
423 | /* See if we read anything */ | |
424 | /* Note - read 0 == socket EOF, which is a valid read */ | |
425 | if (retval >= 0) { | |
426 | fd_bytes(fd, retval, FD_READ); | |
427 | comm_read_callback(fd, retval, COMM_OK, 0); | |
428 | return; | |
429 | } | |
430 | ||
431 | /* Nope, register for some more IO */ | |
432 | commSetSelect(fd, COMM_SELECT_READ, comm_read_try, NULL, 0); | |
433 | } | |
434 | ||
435 | /* | |
436 | * Queue a read. handler/handler_data are called when the read | |
437 | * completes, on error, or on file descriptor close. | |
438 | */ | |
439 | void | |
440 | comm_read(int fd, char *buf, int size, IOCB *handler, void *handler_data) | |
441 | { | |
442 | /* Make sure we're not reading anything and we're not closing */ | |
443 | assert(fdc_table[fd].active == 1); | |
444 | assert(fdc_table[fd].read.handler == NULL); | |
445 | assert(!fd_table[fd].flags.closing); | |
446 | ||
528b2c61 | 447 | debug(5,4)("comm_read, queueing read for FD %d\n",fd); |
448 | ||
c4b7a5a9 | 449 | /* Queue a read */ |
450 | fdc_table[fd].read.buf = buf; | |
451 | fdc_table[fd].read.size = size; | |
452 | fdc_table[fd].read.handler = handler; | |
453 | fdc_table[fd].read.handler_data = handler_data; | |
454 | ||
455 | #if OPTIMISTIC_IO | |
456 | comm_read_try(fd, NULL); | |
457 | #else | |
458 | /* Register intrest in a FD read */ | |
459 | commSetSelect(fd, COMM_SELECT_READ, comm_read_try, NULL, 0); | |
460 | #endif | |
461 | } | |
462 | ||
463 | static void | |
464 | comm_fill_read(int fd, char *buf, size_t len, comm_err_t flag, int xerrno, void *data) | |
465 | { | |
466 | /* TODO use a reference to the table entry, or use C++ :] */ | |
467 | StoreIOBuffer *sb; | |
468 | _fdc_t::CommFiller *fill; | |
469 | assert(fdc_table[fd].active == 1); | |
470 | ||
471 | if (flag != COMM_OK) { | |
472 | /* Error! */ | |
473 | /* XXX This was -1 below, but -1 can't be used for size_t parameters. | |
474 | * The callback should set -1 to the client if needed based on the flags | |
475 | */ | |
476 | comm_add_fill_callback(fd, 0, flag, xerrno); | |
477 | return; | |
478 | } | |
479 | /* flag is COMM_OK */ | |
480 | /* We handle EOFs as read lengths of 0! Its eww, but its consistent */ | |
481 | fill = &fdc_table[fd].fill; | |
482 | fill->amountDone += len; | |
483 | sb = &fdc_table[fd].fill.requestedData; | |
484 | assert(fill->amountDone <= sb->length); | |
485 | comm_add_fill_callback(fd, fill->amountDone, COMM_OK, 0); | |
486 | } | |
487 | ||
488 | /* | |
489 | * Try filling a StoreIOBuffer with some data, and call a callback when successful | |
490 | */ | |
491 | void | |
492 | comm_fill_immediate(int fd, StoreIOBuffer sb, IOFCB *callback, void *data) | |
493 | { | |
494 | assert(fdc_table[fd].fill.handler == NULL); | |
495 | /* prevent confusion */ | |
496 | assert (sb.offset == 0); | |
497 | ||
498 | /* If we don't have any data, record details and schedule a read */ | |
499 | fdc_table[fd].fill.handler = callback; | |
500 | fdc_table[fd].fill.handler_data = data; | |
501 | fdc_table[fd].fill.requestedData = sb; | |
502 | fdc_table[fd].fill.amountDone = 0; | |
503 | ||
504 | comm_read(fd, sb.data, sb.length, comm_fill_read, NULL); | |
505 | } | |
506 | ||
507 | ||
508 | /* | |
509 | * Empty the read buffers | |
510 | * | |
511 | * This is a magical routine that empties the read buffers. | |
512 | * Under some platforms (Linux) if a buffer has data in it before | |
513 | * you call close(), the socket will hang and take quite a while | |
514 | * to timeout. | |
515 | */ | |
516 | static void | |
517 | comm_empty_os_read_buffers(int fd) | |
518 | { | |
a42d5c25 | 519 | #ifdef _SQUID_LINUX_ |
c4b7a5a9 | 520 | /* prevent those nasty RST packets */ |
521 | char buf[SQUID_TCP_SO_RCVBUF]; | |
522 | if (fd_table[fd].flags.nonblocking == 1) | |
523 | while (FD_READ_METHOD(fd, buf, SQUID_TCP_SO_RCVBUF) > 0); | |
524 | #endif | |
525 | } | |
526 | ||
527 | ||
528 | /* | |
529 | * Return whether a file descriptor has any pending read request callbacks | |
530 | * | |
531 | * Assumptions: the fd is open (ie, its not closing) | |
532 | */ | |
533 | int | |
534 | comm_has_pending_read_callback(int fd) | |
535 | { | |
536 | dlink_node *node; | |
537 | CommCallbackData *cd; | |
538 | ||
539 | assert(fd_table[fd].flags.open == 1); | |
540 | assert(fdc_table[fd].active == 1); | |
541 | ||
542 | /* | |
543 | * XXX I don't like having to walk the list! | |
544 | * Instead, if this routine is called often enough, we should | |
545 | * also maintain a linked list of _read_ events - we can just | |
546 | * check if the list head a HEAD.. | |
547 | * - adrian | |
548 | */ | |
549 | node = fdc_table[fd].CommCallbackList.head; | |
550 | while (node != NULL) { | |
551 | cd = (CommCallbackData *)node->data; | |
552 | if (cd->type == COMM_CB_READ) | |
553 | return 1; | |
554 | node = node->next; | |
555 | } | |
556 | ||
557 | /* Not found */ | |
558 | return 0; | |
559 | } | |
560 | ||
561 | /* | |
562 | * return whether a file descriptor has a read handler | |
563 | * | |
564 | * Assumptions: the fd is open | |
528b2c61 | 565 | * the fd is a comm fd. |
c4b7a5a9 | 566 | */ |
528b2c61 | 567 | bool |
c4b7a5a9 | 568 | comm_has_pending_read(int fd) |
569 | { | |
570 | assert(fd_table[fd].flags.open == 1); | |
571 | assert(fdc_table[fd].active == 1); | |
572 | ||
573 | return (fdc_table[fd].read.handler != NULL); | |
574 | } | |
575 | ||
576 | /* | |
577 | * Cancel a pending read. Assert that we have the right parameters, | |
578 | * and that there are no pending read events! | |
579 | */ | |
580 | void | |
581 | comm_read_cancel(int fd, IOCB *callback, void *data) | |
582 | { | |
583 | assert(fd_table[fd].flags.open == 1); | |
584 | assert(fdc_table[fd].active == 1); | |
585 | ||
586 | assert(fdc_table[fd].read.handler == callback); | |
587 | assert(fdc_table[fd].read.handler_data == data); | |
588 | ||
589 | assert(!comm_has_pending_read_callback(fd)); | |
590 | ||
591 | /* Ok, we can be reasonably sure we won't lose any data here! */ | |
592 | ||
593 | /* Delete the callback */ | |
594 | fdc_table[fd].read.handler = NULL; | |
595 | fdc_table[fd].read.handler_data = NULL; | |
420f2ac8 | 596 | |
597 | /* And the IO event */ | |
598 | commSetSelect(fd, COMM_SELECT_READ, NULL, NULL, 0); | |
c4b7a5a9 | 599 | } |
600 | ||
601 | ||
d846c273 | 602 | /* |
603 | * Open a filedescriptor, set some sane defaults | |
604 | * + accept() poll time is 250ms | |
605 | */ | |
c4b7a5a9 | 606 | void |
607 | fdc_open(int fd, unsigned int type, char *desc) | |
608 | { | |
609 | assert(fdc_table[fd].active == 0); | |
610 | ||
611 | fdc_table[fd].active = 1; | |
2fb9d240 | 612 | fdc_table[fd].fd = fd; |
d846c273 | 613 | comm_accept_setcheckperiod(fd, 250); |
c4b7a5a9 | 614 | fd_open(fd, type, desc); |
615 | } | |
616 | ||
617 | ||
ce767c23 | 618 | /* |
619 | * synchronous wrapper around udp socket functions | |
620 | */ | |
621 | ||
622 | int | |
7d21986b | 623 | comm_udp_recvfrom(int fd, void *buf, size_t len, int flags, |
ce767c23 | 624 | struct sockaddr *from, socklen_t *fromlen) |
625 | { | |
626 | statCounter.syscalls.sock.recvfroms++; | |
627 | return recvfrom(fd, buf, len, flags, from, fromlen); | |
628 | } | |
629 | ||
365f12a9 | 630 | int |
7d21986b | 631 | comm_udp_recv(int fd, void *buf, size_t len, int flags) |
365f12a9 | 632 | { |
7d21986b | 633 | return comm_udp_recvfrom(fd, buf, len, flags, NULL, 0); |
365f12a9 | 634 | } |
635 | ||
f71da12c | 636 | ssize_t |
7d21986b | 637 | comm_udp_send(int s, const void *buf, size_t len, int flags) |
f71da12c | 638 | { |
639 | return send(s, buf, len, flags); | |
640 | } | |
ce767c23 | 641 | |
642 | ||
d4cb310b | 643 | /* |
644 | * The new-style comm_write magic | |
645 | */ | |
646 | /* | |
647 | * Attempt a write | |
648 | * | |
649 | * If the write attempt succeeds or fails, call the callback. | |
650 | * Else, wait for another IO notification. | |
651 | */ | |
652 | static void | |
653 | comm_write_try(int fd, void *data) | |
654 | { | |
655 | fdc_t *Fc = &fdc_table[fd]; | |
656 | int retval; | |
657 | ||
658 | /* make sure we actually have a callback */ | |
659 | assert(Fc->write.handler != NULL); | |
660 | ||
661 | /* Attempt a write */ | |
662 | statCounter.syscalls.sock.reads++; | |
2fb9d240 | 663 | errno = 0; |
d4cb310b | 664 | retval = FD_WRITE_METHOD(fd, Fc->write.buf + Fc->write.curofs, Fc->write.size - Fc->write.curofs); |
2fb9d240 | 665 | debug(5, 3) ("comm_write_try: fd %d: tried to write %d bytes, retval %d, errno %d\n", |
666 | fd, Fc->write.size - Fc->write.curofs, retval, errno); | |
d88f85f2 | 667 | |
d4cb310b | 668 | if (retval < 0 && !ignoreErrno(errno)) { |
2fb9d240 | 669 | debug(5, 3) ("comm_write_try: can't ignore error: scheduling COMM_ERROR callback\n"); |
d4cb310b | 670 | comm_add_write_callback(fd, 0, COMM_ERROR, errno); |
671 | return; | |
d4cb310b | 672 | } |
d88f85f2 | 673 | |
d4cb310b | 674 | if (retval >= 0) { |
d88f85f2 | 675 | fd_bytes(fd, retval, FD_WRITE); |
d4cb310b | 676 | Fc->write.curofs += retval; |
677 | assert(Fc->write.curofs <= Fc->write.size); | |
678 | /* All? */ | |
679 | if (Fc->write.curofs == Fc->write.size) { | |
d88f85f2 | 680 | comm_add_write_callback(fd, Fc->write.size, COMM_OK, 0); |
d4cb310b | 681 | return; |
682 | } | |
683 | } | |
684 | ||
685 | /* if we get here, we need to write more! */ | |
686 | commSetSelect(fd, COMM_SELECT_WRITE, comm_write_try, NULL, 0); | |
687 | } | |
688 | ||
cf3c0ee3 | 689 | /* |
690 | * Queue a write. handler/handler_data are called when the write fully | |
691 | * completes, on error, or on file descriptor close. | |
692 | */ | |
693 | void | |
dba1a67d | 694 | comm_write(int fd, const char *buf, size_t size, IOWCB *handler, void *handler_data) |
cf3c0ee3 | 695 | { |
696 | /* Make sure we're not writing anything and we're not closing */ | |
697 | assert(fdc_table[fd].active == 1); | |
698 | assert(fdc_table[fd].write.handler == NULL); | |
699 | assert(!fd_table[fd].flags.closing); | |
700 | ||
701 | /* Queue a read */ | |
702 | fdc_table[fd].write.buf = buf; | |
703 | fdc_table[fd].write.size = size; | |
704 | fdc_table[fd].write.handler = handler; | |
705 | fdc_table[fd].write.handler_data = handler_data; | |
706 | fdc_table[fd].write.curofs = 0; | |
707 | ||
708 | #if OPTIMISTIC_IO | |
709 | comm_write_try(fd, NULL); | |
710 | #else | |
711 | /* Register intrest in a FD read */ | |
712 | commSetSelect(fd, COMM_SELECT_WRITE, comm_write_try, NULL, 0); | |
713 | #endif | |
714 | } | |
d4cb310b | 715 | |
3c1a197f | 716 | /* |
717 | * New-style accept stuff | |
718 | */ | |
719 | ||
720 | /* | |
721 | * Set the check delay on accept()ing when we're out of FDs | |
722 | * | |
723 | * The premise behind this is that we can hit a situation where | |
724 | * we've hit our reserved filedescriptor limit and we don't want | |
725 | * to accept any more connections until some others have closed. | |
726 | * | |
727 | * This code will set the period which we register an event to check | |
728 | * to see whether we _have_ enough open FDs to re-register for IO. | |
729 | */ | |
730 | void | |
731 | comm_accept_setcheckperiod(int fd, int mdelay) | |
732 | { | |
733 | assert(fdc_table[fd].active == 1); | |
734 | assert(mdelay != 0); | |
735 | fdc_table[fd].accept.check_delay = mdelay; | |
736 | } | |
737 | ||
738 | /* | |
739 | * Our periodic accept() suitability checker.. | |
740 | */ | |
741 | static void | |
742 | comm_accept_check_event(void *data) | |
743 | { | |
744 | static time_t last_warn = 0; | |
745 | int fd = ((fdc_t *)(data))->fd; | |
746 | ||
a7c5ae5d | 747 | if (fdNFree() < RESERVED_FD) { |
3c1a197f | 748 | commSetSelect(fd, COMM_SELECT_READ, comm_accept_try, NULL, 0); |
749 | return; | |
750 | } | |
751 | if (last_warn + 15 < squid_curtime) { | |
752 | debug(33, 0) ("WARNING! Your cache is running out of filedescriptors\n"); | |
753 | last_warn = squid_curtime; | |
754 | } | |
755 | eventAdd("comm_accept_check_event", comm_accept_check_event, &fdc_table[fd], | |
756 | 1000.0 / (double)(fdc_table[fd].accept.check_delay), 1); | |
757 | } | |
758 | ||
759 | ||
d4cb310b | 760 | |
c4b7a5a9 | 761 | /* Older stuff */ |
309ad3b6 | 762 | |
b8d8561b | 763 | static void |
3d7e9d7c | 764 | CommWriteStateCallbackAndFree(int fd, comm_err_t code) |
9864ee44 | 765 | { |
f17936ab | 766 | CommWriteStateData *CommWriteState = fd_table[fd].rwstate; |
767 | CWCB *callback = NULL; | |
fa80a8ef | 768 | void *cbdata; |
a56a3abe | 769 | fd_table[fd].rwstate = NULL; |
f17936ab | 770 | if (CommWriteState == NULL) |
9864ee44 | 771 | return; |
c0dec081 | 772 | if (CommWriteState->free_func) { |
729dd65c | 773 | FREE *free_func = CommWriteState->free_func; |
774 | void *free_buf = CommWriteState->buf; | |
775 | CommWriteState->free_func = NULL; | |
f17936ab | 776 | CommWriteState->buf = NULL; |
729dd65c | 777 | free_func(free_buf); |
9864ee44 | 778 | } |
f17936ab | 779 | callback = CommWriteState->handler; |
780 | CommWriteState->handler = NULL; | |
fa80a8ef | 781 | if (callback && cbdataReferenceValidDone(CommWriteState->handler_data, &cbdata)) |
782 | callback(fd, CommWriteState->buf, CommWriteState->offset, code, cbdata); | |
723123a9 | 783 | memPoolFree(comm_write_pool, CommWriteState); |
9864ee44 | 784 | } |
785 | ||
090089c4 | 786 | /* Return the local port associated with fd. */ |
b8d8561b | 787 | u_short |
788 | comm_local_port(int fd) | |
090089c4 | 789 | { |
790 | struct sockaddr_in addr; | |
6637e3a5 | 791 | socklen_t addr_len = 0; |
76f87348 | 792 | fde *F = &fd_table[fd]; |
090089c4 | 793 | |
090089c4 | 794 | /* If the fd is closed already, just return */ |
60c0b5a2 | 795 | if (!F->flags.open) { |
a3d5953d | 796 | debug(5, 0) ("comm_local_port: FD %d has been closed.\n", fd); |
30a4f2a8 | 797 | return 0; |
090089c4 | 798 | } |
76f87348 | 799 | if (F->local_port) |
800 | return F->local_port; | |
090089c4 | 801 | addr_len = sizeof(addr); |
802 | if (getsockname(fd, (struct sockaddr *) &addr, &addr_len)) { | |
a3d5953d | 803 | debug(50, 1) ("comm_local_port: Failed to retrieve TCP/UDP port number for socket: FD %d: %s\n", fd, xstrerror()); |
30a4f2a8 | 804 | return 0; |
090089c4 | 805 | } |
76f87348 | 806 | F->local_port = ntohs(addr.sin_port); |
5f6ac48b | 807 | debug(5, 6) ("comm_local_port: FD %d: port %d\n", fd, (int) F->local_port); |
76f87348 | 808 | return F->local_port; |
090089c4 | 809 | } |
810 | ||
3d7e9d7c | 811 | static comm_err_t |
b8d8561b | 812 | commBind(int s, struct in_addr in_addr, u_short port) |
090089c4 | 813 | { |
814 | struct sockaddr_in S; | |
090089c4 | 815 | |
090089c4 | 816 | memset(&S, '\0', sizeof(S)); |
817 | S.sin_family = AF_INET; | |
818 | S.sin_port = htons(port); | |
30a4f2a8 | 819 | S.sin_addr = in_addr; |
83704487 | 820 | statCounter.syscalls.sock.binds++; |
090089c4 | 821 | if (bind(s, (struct sockaddr *) &S, sizeof(S)) == 0) |
822 | return COMM_OK; | |
a3d5953d | 823 | debug(50, 0) ("commBind: Cannot bind socket FD %d to %s:%d: %s\n", |
090089c4 | 824 | s, |
30a4f2a8 | 825 | S.sin_addr.s_addr == INADDR_ANY ? "*" : inet_ntoa(S.sin_addr), |
44a62238 | 826 | (int) port, |
827 | xstrerror()); | |
090089c4 | 828 | return COMM_ERROR; |
829 | } | |
830 | ||
831 | /* Create a socket. Default is blocking, stream (TCP) socket. IO_TYPE | |
d6827718 | 832 | * is OR of flags specified in comm.h. Defaults TOS */ |
b8d8561b | 833 | int |
16b204c4 | 834 | comm_open(int sock_type, |
cc6a9d2e | 835 | int proto, |
836 | struct in_addr addr, | |
837 | u_short port, | |
838 | int flags, | |
0ee4272b | 839 | const char *note) |
d6827718 | 840 | { |
841 | return comm_openex(sock_type, proto, addr, port, flags, 0, note); | |
842 | } | |
843 | ||
844 | ||
845 | /* Create a socket. Default is blocking, stream (TCP) socket. IO_TYPE | |
846 | * is OR of flags specified in defines.h:COMM_* */ | |
847 | int | |
848 | comm_openex(int sock_type, | |
849 | int proto, | |
850 | struct in_addr addr, | |
851 | u_short port, | |
852 | int flags, | |
853 | unsigned char TOS, | |
854 | const char *note) | |
090089c4 | 855 | { |
856 | int new_socket; | |
9056f553 | 857 | int tos = 0; |
76f87348 | 858 | fde *F = NULL; |
090089c4 | 859 | |
88bfe092 | 860 | PROF_start(comm_open); |
090089c4 | 861 | /* Create socket for accepting new connections. */ |
83704487 | 862 | statCounter.syscalls.sock.sockets++; |
16b204c4 | 863 | if ((new_socket = socket(AF_INET, sock_type, proto)) < 0) { |
090089c4 | 864 | /* Increase the number of reserved fd's if calls to socket() |
865 | * are failing because the open file table is full. This | |
866 | * limits the number of simultaneous clients */ | |
867 | switch (errno) { | |
868 | case ENFILE: | |
869 | case EMFILE: | |
a3d5953d | 870 | debug(50, 1) ("comm_open: socket failure: %s\n", xstrerror()); |
9bc73deb | 871 | fdAdjustReserved(); |
090089c4 | 872 | break; |
873 | default: | |
a3d5953d | 874 | debug(50, 0) ("comm_open: socket failure: %s\n", xstrerror()); |
090089c4 | 875 | } |
88bfe092 | 876 | PROF_stop(comm_open); |
603a02fd | 877 | return -1; |
090089c4 | 878 | } |
d6827718 | 879 | /* set TOS if needed */ |
880 | if (TOS) { | |
881 | #ifdef IP_TOS | |
882 | tos = TOS; | |
883 | if (setsockopt(new_socket, IPPROTO_IP, IP_TOS, (char *) &tos, sizeof(int)) < 0) | |
884 | debug(50, 1) ("comm_open: setsockopt(IP_TOS) on FD %d: %s\n", | |
885 | new_socket, xstrerror()); | |
886 | #else | |
887 | debug(50, 0) ("comm_open: setsockopt(IP_TOS) not supported on this platform\n"); | |
888 | #endif | |
889 | } | |
090089c4 | 890 | /* update fdstat */ |
365e5b34 | 891 | debug(5, 5) ("comm_open: FD %d is a new socket\n", new_socket); |
5c5783a2 | 892 | fd_open(new_socket, FD_SOCKET, note); |
c4b7a5a9 | 893 | fdd_table[new_socket].close_file = NULL; |
894 | fdd_table[new_socket].close_line = 0; | |
895 | assert(fdc_table[new_socket].active == 0); | |
896 | fdc_table[new_socket].active = 1; | |
76f87348 | 897 | F = &fd_table[new_socket]; |
d6827718 | 898 | F->local_addr = addr; |
899 | F->tos = tos; | |
79a15e0a | 900 | if (!(flags & COMM_NOCLOEXEC)) |
3ca60c86 | 901 | commSetCloseOnExec(new_socket); |
cdc33f35 | 902 | if ((flags & COMM_REUSEADDR)) |
903 | commSetReuseAddr(new_socket); | |
7690e8eb | 904 | if (port > (u_short) 0) { |
30a4f2a8 | 905 | commSetNoLinger(new_socket); |
3b4be6a6 | 906 | if (opt_reuseaddr) |
090089c4 | 907 | commSetReuseAddr(new_socket); |
090089c4 | 908 | } |
a3724d50 | 909 | if (addr.s_addr != no_addr.s_addr) { |
910 | if (commBind(new_socket, addr, port) != COMM_OK) { | |
911 | comm_close(new_socket); | |
603a02fd | 912 | return -1; |
88bfe092 | 913 | PROF_stop(comm_open); |
a3724d50 | 914 | } |
23ff6968 | 915 | } |
76f87348 | 916 | F->local_port = port; |
090089c4 | 917 | |
79a15e0a | 918 | if (flags & COMM_NONBLOCKING) |
88bfe092 | 919 | if (commSetNonBlocking(new_socket) == COMM_ERROR) { |
603a02fd | 920 | return -1; |
88bfe092 | 921 | PROF_stop(comm_open); |
922 | } | |
30a4f2a8 | 923 | #ifdef TCP_NODELAY |
924 | if (sock_type == SOCK_STREAM) | |
925 | commSetTcpNoDelay(new_socket); | |
926 | #endif | |
1241e63e | 927 | if (Config.tcpRcvBufsz > 0 && sock_type == SOCK_STREAM) |
928 | commSetTcpRcvbuf(new_socket, Config.tcpRcvBufsz); | |
88bfe092 | 929 | PROF_stop(comm_open); |
090089c4 | 930 | return new_socket; |
931 | } | |
932 | ||
e5f6c5c2 | 933 | void |
4f92c80c | 934 | commConnectStart(int fd, const char *host, u_short port, CNCB * callback, void *data) |
e924600d | 935 | { |
28c60158 | 936 | ConnectStateData *cs; |
5117e471 | 937 | debug(5, 3) ("commConnectStart: FD %d, data %p, %s:%d\n", fd, data, host, (int) port); |
72711e31 | 938 | cs = cbdataAlloc(ConnectStateData); |
03a1ee42 | 939 | cs->fd = fd; |
e924600d | 940 | cs->host = xstrdup(host); |
941 | cs->port = port; | |
942 | cs->callback = callback; | |
fa80a8ef | 943 | cs->data = cbdataReference(data); |
e924600d | 944 | comm_add_close_handler(fd, commConnectFree, cs); |
f88211e8 | 945 | cs->locks++; |
8407afee | 946 | ipcache_nbgethostbyname(host, commConnectDnsHandle, cs); |
edeb28fd | 947 | } |
948 | ||
949 | static void | |
03a1ee42 | 950 | commConnectDnsHandle(const ipcache_addrs * ia, void *data) |
edeb28fd | 951 | { |
e6ccf245 | 952 | ConnectStateData *cs = (ConnectStateData *)data; |
f88211e8 | 953 | assert(cs->locks == 1); |
954 | cs->locks--; | |
edeb28fd | 955 | if (ia == NULL) { |
a3d5953d | 956 | debug(5, 3) ("commConnectDnsHandle: Unknown host: %s\n", cs->host); |
6cf028ab | 957 | if (!dns_error_message) { |
958 | dns_error_message = "Unknown DNS error"; | |
0e473d70 | 959 | debug(5, 1) ("commConnectDnsHandle: Bad dns_error_message\n"); |
6cf028ab | 960 | } |
a64c2869 | 961 | assert(dns_error_message != NULL); |
03a1ee42 | 962 | commConnectCallback(cs, COMM_ERR_DNS); |
edeb28fd | 963 | return; |
964 | } | |
f076b37b | 965 | assert(ia->cur < ia->count); |
edeb28fd | 966 | cs->in_addr = ia->in_addrs[ia->cur]; |
52926044 | 967 | ipcacheCycleAddr(cs->host, NULL); |
22c653cd | 968 | cs->addrcount = ia->count; |
969 | cs->connstart = squid_curtime; | |
03a1ee42 | 970 | commConnectHandle(cs->fd, cs); |
e924600d | 971 | } |
972 | ||
f88211e8 | 973 | static void |
3d7e9d7c | 974 | commConnectCallback(ConnectStateData * cs, comm_err_t status) |
f88211e8 | 975 | { |
a3d5953d | 976 | CNCB *callback = cs->callback; |
fa80a8ef | 977 | void *cbdata = cs->data; |
03a1ee42 | 978 | int fd = cs->fd; |
5117e471 | 979 | debug(5, 3) ("commConnectCallback: fd %d, data %p\n", fd, cbdata); |
a3d5953d | 980 | comm_remove_close_handler(fd, commConnectFree, cs); |
9daca08e | 981 | cs->callback = NULL; |
982 | cs->data = NULL; | |
e1b16349 | 983 | commSetTimeout(fd, -1, NULL, NULL); |
a3d5953d | 984 | commConnectFree(fd, cs); |
fa80a8ef | 985 | if (cbdataReferenceValid(cbdata)) |
986 | callback(fd, status, cbdata); | |
5117e471 | 987 | cbdataReferenceDone(cbdata); |
f88211e8 | 988 | } |
989 | ||
e924600d | 990 | static void |
9daca08e | 991 | commConnectFree(int fd, void *data) |
e924600d | 992 | { |
e6ccf245 | 993 | ConnectStateData *cs = (ConnectStateData *)data; |
9daca08e | 994 | debug(5, 3) ("commConnectFree: FD %d\n", fd); |
fa80a8ef | 995 | cbdataReferenceDone(cs->data); |
8407afee | 996 | safe_free(cs->host); |
997 | cbdataFree(cs); | |
e924600d | 998 | } |
999 | ||
22c653cd | 1000 | /* Reset FD so that we can connect() again */ |
edeb28fd | 1001 | static int |
22c653cd | 1002 | commResetFD(ConnectStateData * cs) |
edeb28fd | 1003 | { |
1004 | int fd2; | |
d6827718 | 1005 | fde *F; |
fa80a8ef | 1006 | if (!cbdataReferenceValid(cs->data)) |
7dd44885 | 1007 | return 0; |
83704487 | 1008 | statCounter.syscalls.sock.sockets++; |
edeb28fd | 1009 | fd2 = socket(AF_INET, SOCK_STREAM, 0); |
83704487 | 1010 | statCounter.syscalls.sock.sockets++; |
edeb28fd | 1011 | if (fd2 < 0) { |
22c653cd | 1012 | debug(5, 0) ("commResetFD: socket: %s\n", xstrerror()); |
9bc73deb | 1013 | if (ENFILE == errno || EMFILE == errno) |
1014 | fdAdjustReserved(); | |
edeb28fd | 1015 | return 0; |
1016 | } | |
22c653cd | 1017 | if (dup2(fd2, cs->fd) < 0) { |
1018 | debug(5, 0) ("commResetFD: dup2: %s\n", xstrerror()); | |
9bc73deb | 1019 | if (ENFILE == errno || EMFILE == errno) |
1020 | fdAdjustReserved(); | |
d6827718 | 1021 | close(fd2); |
edeb28fd | 1022 | return 0; |
1023 | } | |
edeb28fd | 1024 | close(fd2); |
d6827718 | 1025 | F = &fd_table[cs->fd]; |
b5568a61 | 1026 | fd_table[cs->fd].flags.called_connect = 0; |
09544acc | 1027 | /* |
1028 | * yuck, this has assumptions about comm_open() arguments for | |
1029 | * the original socket | |
1030 | */ | |
d6827718 | 1031 | if (commBind(cs->fd, F->local_addr, F->local_port) != COMM_OK) { |
1032 | debug(5, 0) ("commResetFD: bind: %s\n", xstrerror()); | |
1033 | return 0; | |
09544acc | 1034 | } |
d6827718 | 1035 | #ifdef IP_TOS |
1036 | if (F->tos) { | |
1037 | int tos = F->tos; | |
1038 | if (setsockopt(cs->fd, IPPROTO_IP, IP_TOS, (char *) &tos, sizeof(int)) < 0) | |
1039 | debug(50, 1) ("commResetFD: setsockopt(IP_TOS) on FD %d: %s\n", cs->fd, xstrerror()); | |
1040 | } | |
1041 | #endif | |
1042 | if (F->flags.close_on_exec) | |
1043 | commSetCloseOnExec(cs->fd); | |
1044 | if (F->flags.nonblocking) | |
1045 | commSetNonBlocking(cs->fd); | |
09544acc | 1046 | #ifdef TCP_NODELAY |
d6827718 | 1047 | if (F->flags.nodelay) |
1048 | commSetTcpNoDelay(cs->fd); | |
09544acc | 1049 | #endif |
1050 | if (Config.tcpRcvBufsz > 0) | |
1051 | commSetTcpRcvbuf(cs->fd, Config.tcpRcvBufsz); | |
edeb28fd | 1052 | return 1; |
1053 | } | |
1054 | ||
22c653cd | 1055 | static int |
1056 | commRetryConnect(ConnectStateData * cs) | |
1057 | { | |
1058 | assert(cs->addrcount > 0); | |
1059 | if (cs->addrcount == 1) { | |
1060 | if (cs->tries >= Config.retry.maxtries) | |
1061 | return 0; | |
1062 | if (squid_curtime - cs->connstart > Config.Timeout.connect) | |
1063 | return 0; | |
22c653cd | 1064 | } else { |
1065 | if (cs->tries > cs->addrcount) | |
1066 | return 0; | |
1067 | } | |
1068 | return commResetFD(cs); | |
1069 | } | |
1070 | ||
e924600d | 1071 | /* Connect SOCK to specified DEST_PORT at DEST_HOST. */ |
1072 | static void | |
1073 | commConnectHandle(int fd, void *data) | |
090089c4 | 1074 | { |
e6ccf245 | 1075 | ConnectStateData *cs = (ConnectStateData *)data; |
f88211e8 | 1076 | if (cs->S.sin_addr.s_addr == 0) { |
1077 | cs->S.sin_family = AF_INET; | |
1078 | cs->S.sin_addr = cs->in_addr; | |
1079 | cs->S.sin_port = htons(cs->port); | |
17a0a4ee | 1080 | if (Config.onoff.log_fqdn) |
f88211e8 | 1081 | fqdncache_gethostbyaddr(cs->S.sin_addr, FQDN_LOOKUP_IF_MISS); |
e5f6c5c2 | 1082 | } |
f88211e8 | 1083 | switch (comm_connect_addr(fd, &cs->S)) { |
e5f6c5c2 | 1084 | case COMM_INPROGRESS: |
11994bb9 | 1085 | debug(5, 5) ("commConnectHandle: FD %d: COMM_INPROGRESS\n", fd); |
f88211e8 | 1086 | commSetSelect(fd, COMM_SELECT_WRITE, commConnectHandle, cs, 0); |
e5f6c5c2 | 1087 | break; |
1088 | case COMM_OK: | |
22c653cd | 1089 | ipcacheMarkGoodAddr(cs->host, cs->S.sin_addr); |
03a1ee42 | 1090 | commConnectCallback(cs, COMM_OK); |
e5f6c5c2 | 1091 | break; |
1092 | default: | |
22c653cd | 1093 | cs->tries++; |
1094 | ipcacheMarkBadAddr(cs->host, cs->S.sin_addr); | |
194dd3b8 | 1095 | if (Config.onoff.test_reachability) |
1096 | netdbDeleteAddrNetwork(cs->S.sin_addr); | |
22c653cd | 1097 | if (commRetryConnect(cs)) { |
f88211e8 | 1098 | cs->locks++; |
8407afee | 1099 | ipcache_nbgethostbyname(cs->host, commConnectDnsHandle, cs); |
edeb28fd | 1100 | } else { |
03a1ee42 | 1101 | commConnectCallback(cs, COMM_ERR_CONNECT); |
edeb28fd | 1102 | } |
e5f6c5c2 | 1103 | break; |
090089c4 | 1104 | } |
090089c4 | 1105 | } |
22c653cd | 1106 | |
b8d8561b | 1107 | int |
4f92c80c | 1108 | commSetTimeout(int fd, int timeout, PF * handler, void *data) |
090089c4 | 1109 | { |
76f87348 | 1110 | fde *F; |
a3d5953d | 1111 | debug(5, 3) ("commSetTimeout: FD %d timeout %d\n", fd, timeout); |
03eb2f01 | 1112 | assert(fd >= 0); |
1113 | assert(fd < Squid_MaxFD); | |
76f87348 | 1114 | F = &fd_table[fd]; |
60c0b5a2 | 1115 | assert(F->flags.open); |
5c5783a2 | 1116 | if (timeout < 0) { |
a3fa14bf | 1117 | cbdataReferenceDone(F->timeout_data); |
76f87348 | 1118 | F->timeout_handler = NULL; |
a3fa14bf | 1119 | F->timeout = 0; |
5849612f | 1120 | } else { |
1121 | assert(handler || F->timeout_handler); | |
1122 | if (handler) { | |
1123 | cbdataReferenceDone(F->timeout_data); | |
1124 | F->timeout_handler = handler; | |
1125 | F->timeout_data = cbdataReference(data); | |
1126 | } | |
1127 | F->timeout = squid_curtime + (time_t) timeout; | |
30a4f2a8 | 1128 | } |
a3fa14bf | 1129 | return F->timeout; |
090089c4 | 1130 | } |
1131 | ||
b8d8561b | 1132 | int |
0ee4272b | 1133 | comm_connect_addr(int sock, const struct sockaddr_in *address) |
090089c4 | 1134 | { |
3d7e9d7c | 1135 | comm_err_t status = COMM_OK; |
76f87348 | 1136 | fde *F = &fd_table[sock]; |
090089c4 | 1137 | int x; |
b5568a61 | 1138 | int err = 0; |
9689d97c | 1139 | socklen_t errlen; |
489b22c1 | 1140 | assert(ntohs(address->sin_port) != 0); |
88bfe092 | 1141 | PROF_start(comm_connect_addr); |
090089c4 | 1142 | /* Establish connection. */ |
b5568a61 | 1143 | errno = 0; |
1144 | if (!F->flags.called_connect) { | |
1145 | F->flags.called_connect = 1; | |
83704487 | 1146 | statCounter.syscalls.sock.connects++; |
b5568a61 | 1147 | x = connect(sock, (struct sockaddr *) address, sizeof(*address)); |
54f742e7 | 1148 | if (x < 0) |
1149 | debug(5, 9) ("connect FD %d: %s\n", sock, xstrerror()); | |
b5568a61 | 1150 | } else { |
140e2c0b | 1151 | #if defined(_SQUID_NEWSOS6_) |
33ac9442 | 1152 | /* Makoto MATSUSHITA <matusita@ics.es.osaka-u.ac.jp> */ |
1153 | connect(sock, (struct sockaddr *) address, sizeof(*address)); | |
1154 | if (errno == EINVAL) { | |
1155 | errlen = sizeof(err); | |
1156 | x = getsockopt(sock, SOL_SOCKET, SO_ERROR, &err, &errlen); | |
1157 | if (x >= 0) | |
1158 | errno = x; | |
1159 | } | |
1160 | #else | |
b5568a61 | 1161 | errlen = sizeof(err); |
1162 | x = getsockopt(sock, SOL_SOCKET, SO_ERROR, &err, &errlen); | |
1163 | if (x == 0) | |
1164 | errno = err; | |
1165 | #if defined(_SQUID_SOLARIS_) | |
1166 | /* | |
1167 | * Solaris 2.4's socket emulation doesn't allow you | |
1168 | * to determine the error from a failed non-blocking | |
1169 | * connect and just returns EPIPE. Create a fake | |
1170 | * error message for connect. -- fenner@parc.xerox.com | |
1171 | */ | |
1172 | if (x < 0 && errno == EPIPE) | |
1173 | errno = ENOTCONN; | |
33ac9442 | 1174 | #endif |
30a4f2a8 | 1175 | #endif |
e5f6c5c2 | 1176 | } |
88bfe092 | 1177 | PROF_stop(comm_connect_addr); |
b5568a61 | 1178 | if (errno == 0 || errno == EISCONN) |
1179 | status = COMM_OK; | |
1180 | else if (ignoreErrno(errno)) | |
1181 | status = COMM_INPROGRESS; | |
1182 | else | |
1183 | return COMM_ERROR; | |
76f87348 | 1184 | xstrncpy(F->ipaddr, inet_ntoa(address->sin_addr), 16); |
1185 | F->remote_port = ntohs(address->sin_port); | |
090089c4 | 1186 | if (status == COMM_OK) { |
a3d5953d | 1187 | debug(5, 10) ("comm_connect_addr: FD %d connected to %s:%d\n", |
76f87348 | 1188 | sock, F->ipaddr, F->remote_port); |
f21cd581 | 1189 | } else if (status == COMM_INPROGRESS) { |
a3d5953d | 1190 | debug(5, 10) ("comm_connect_addr: FD %d connection pending\n", sock); |
090089c4 | 1191 | } |
090089c4 | 1192 | return status; |
1193 | } | |
1194 | ||
1195 | /* Wait for an incoming connection on FD. FD should be a socket returned | |
1196 | * from comm_listen. */ | |
b8d8561b | 1197 | int |
c4b7a5a9 | 1198 | comm_old_accept(int fd, struct sockaddr_in *pn, struct sockaddr_in *me) |
090089c4 | 1199 | { |
1200 | int sock; | |
1f9afe33 | 1201 | struct sockaddr_in P; |
1202 | struct sockaddr_in M; | |
6637e3a5 | 1203 | socklen_t Slen; |
76f87348 | 1204 | fde *F = NULL; |
1f9afe33 | 1205 | Slen = sizeof(P); |
83704487 | 1206 | statCounter.syscalls.sock.accepts++; |
88bfe092 | 1207 | PROF_start(comm_accept); |
603500e7 | 1208 | if ((sock = accept(fd, (struct sockaddr *) &P, &Slen)) < 0) { |
88bfe092 | 1209 | PROF_stop(comm_accept); |
603500e7 | 1210 | if (ignoreErrno(errno)) { |
c4b7a5a9 | 1211 | debug(50, 5) ("comm_old_accept: FD %d: %s\n", fd, xstrerror()); |
0a0bf5db | 1212 | return COMM_NOMESSAGE; |
603500e7 | 1213 | } else if (ENFILE == errno || EMFILE == errno) { |
c4b7a5a9 | 1214 | debug(50, 3) ("comm_old_accept: FD %d: %s\n", fd, xstrerror()); |
090089c4 | 1215 | return COMM_ERROR; |
603500e7 | 1216 | } else { |
c4b7a5a9 | 1217 | debug(50, 1) ("comm_old_accept: FD %d: %s\n", fd, xstrerror()); |
090089c4 | 1218 | return COMM_ERROR; |
1219 | } | |
1220 | } | |
9ef28b60 | 1221 | if (pn) |
1222 | *pn = P; | |
4053a845 | 1223 | Slen = sizeof(M); |
1224 | memset(&M, '\0', Slen); | |
1225 | getsockname(sock, (struct sockaddr *) &M, &Slen); | |
1226 | if (me) | |
1f9afe33 | 1227 | *me = M; |
3ca60c86 | 1228 | commSetCloseOnExec(sock); |
090089c4 | 1229 | /* fdstat update */ |
5c5783a2 | 1230 | fd_open(sock, FD_SOCKET, "HTTP Request"); |
c4b7a5a9 | 1231 | fdd_table[sock].close_file = NULL; |
1232 | fdd_table[sock].close_line = 0; | |
1233 | fdc_table[sock].active = 1; | |
76f87348 | 1234 | F = &fd_table[sock]; |
c0dec081 | 1235 | xstrncpy(F->ipaddr, inet_ntoa(P.sin_addr), 16); |
76f87348 | 1236 | F->remote_port = htons(P.sin_port); |
1237 | F->local_port = htons(M.sin_port); | |
090089c4 | 1238 | commSetNonBlocking(sock); |
88bfe092 | 1239 | PROF_stop(comm_accept); |
090089c4 | 1240 | return sock; |
1241 | } | |
1242 | ||
cb201b7e | 1243 | void |
1244 | commCallCloseHandlers(int fd) | |
1245 | { | |
76f87348 | 1246 | fde *F = &fd_table[fd]; |
f1dc9b30 | 1247 | close_handler *ch; |
a3d5953d | 1248 | debug(5, 5) ("commCallCloseHandlers: FD %d\n", fd); |
29b8d8d6 | 1249 | while ((ch = F->closeHandler) != NULL) { |
1250 | F->closeHandler = ch->next; | |
9daca08e | 1251 | debug(5, 5) ("commCallCloseHandlers: ch->handler=%p\n", ch->handler); |
fa80a8ef | 1252 | if (cbdataReferenceValid(ch->data)) |
603a02fd | 1253 | ch->handler(fd, ch->data); |
fa80a8ef | 1254 | cbdataReferenceDone(ch->data); |
7f6ffd15 | 1255 | memPoolFree(conn_close_pool, ch); /* AAA */ |
cb201b7e | 1256 | } |
1257 | } | |
1258 | ||
5492ad1d | 1259 | #if LINGERING_CLOSE |
1260 | static void | |
1261 | commLingerClose(int fd, void *unused) | |
1262 | { | |
1263 | LOCAL_ARRAY(char, buf, 1024); | |
1264 | int n; | |
1f7c9178 | 1265 | n = FD_READ_METHOD(fd, buf, 1024); |
5492ad1d | 1266 | if (n < 0) |
1267 | debug(5, 3) ("commLingerClose: FD %d read: %s\n", fd, xstrerror()); | |
1268 | comm_close(fd); | |
1269 | } | |
1270 | ||
1271 | static void | |
1272 | commLingerTimeout(int fd, void *unused) | |
1273 | { | |
1274 | debug(5, 3) ("commLingerTimeout: FD %d\n", fd); | |
1275 | comm_close(fd); | |
1276 | } | |
1277 | ||
1278 | /* | |
1279 | * Inspired by apache | |
1280 | */ | |
1281 | void | |
1282 | comm_lingering_close(int fd) | |
1283 | { | |
d4c19b39 | 1284 | #if USE_SSL |
1285 | if (fd_table[fd].ssl) | |
79d4ccdf | 1286 | ssl_shutdown_method(fd); |
d4c19b39 | 1287 | #endif |
5492ad1d | 1288 | if (shutdown(fd, 1) < 0) { |
1289 | comm_close(fd); | |
1290 | return; | |
1291 | } | |
1292 | fd_note(fd, "lingering close"); | |
1293 | commSetTimeout(fd, 10, commLingerTimeout, NULL); | |
1294 | commSetSelect(fd, COMM_SELECT_READ, commLingerClose, NULL, 0); | |
1295 | } | |
1296 | #endif | |
1297 | ||
98264874 | 1298 | /* |
1299 | * enable linger with time of 0 so that when the socket is | |
1300 | * closed, TCP generates a RESET | |
1301 | */ | |
1302 | void | |
1303 | comm_reset_close(int fd) | |
1304 | { | |
1305 | struct linger L; | |
1306 | L.l_onoff = 1; | |
1307 | L.l_linger = 0; | |
1308 | if (setsockopt(fd, SOL_SOCKET, SO_LINGER, (char *) &L, sizeof(L)) < 0) | |
1309 | debug(50, 0) ("commResetTCPClose: FD %d: %s\n", fd, xstrerror()); | |
1310 | comm_close(fd); | |
1311 | } | |
1312 | ||
c4b7a5a9 | 1313 | |
1314 | /* | |
1315 | * Close the socket fd. | |
1316 | * | |
1317 | * + call write handlers with ERR_CLOSING | |
1318 | * + call read handlers with ERR_CLOSING | |
1319 | * + call closing handlers | |
1320 | */ | |
b8d8561b | 1321 | void |
c4b7a5a9 | 1322 | _comm_close(int fd, char *file, int line) |
090089c4 | 1323 | { |
76f87348 | 1324 | fde *F = NULL; |
c4b7a5a9 | 1325 | dlink_node *node; |
1326 | CommCallbackData *cio; | |
1f7c9178 | 1327 | |
a3d5953d | 1328 | debug(5, 5) ("comm_close: FD %d\n", fd); |
03eb2f01 | 1329 | assert(fd >= 0); |
1330 | assert(fd < Squid_MaxFD); | |
76f87348 | 1331 | F = &fd_table[fd]; |
c4b7a5a9 | 1332 | fdd_table[fd].close_file = file; |
1333 | fdd_table[fd].close_line = line; | |
1f7c9178 | 1334 | |
58a6c186 | 1335 | if (F->flags.closing) |
e102ebda | 1336 | return; |
60c0b5a2 | 1337 | if (shutting_down && (!F->flags.open || F->type == FD_FILE)) |
6cf028ab | 1338 | return; |
60c0b5a2 | 1339 | assert(F->flags.open); |
c4b7a5a9 | 1340 | /* The following fails because ipc.c is doing calls to pipe() to create sockets! */ |
fd47fdf3 | 1341 | assert(fdc_table[fd].active == 1); |
76f87348 | 1342 | assert(F->type != FD_FILE); |
88bfe092 | 1343 | PROF_start(comm_close); |
58a6c186 | 1344 | F->flags.closing = 1; |
d4c19b39 | 1345 | #if USE_SSL |
1346 | if (F->ssl) | |
79d4ccdf | 1347 | ssl_shutdown_method(fd); |
d4c19b39 | 1348 | #endif |
fa80a8ef | 1349 | commSetTimeout(fd, -1, NULL, NULL); |
96f1be5d | 1350 | CommWriteStateCallbackAndFree(fd, COMM_ERR_CLOSING); |
c4b7a5a9 | 1351 | |
3c1a197f | 1352 | /* Delete any accept check */ |
1353 | if (eventFind(comm_accept_check_event, &fdc_table[fd])) { | |
1354 | eventDelete(comm_accept_check_event, &fdc_table[fd]); | |
1355 | } | |
1356 | ||
691476e2 | 1357 | /* Do callbacks for read/accept/fill routines, if any */ |
511e2383 | 1358 | if (fdc_table[fd].read.handler) { |
691476e2 | 1359 | fdc_table[fd].read.handler(fd, fdc_table[fd].read.buf, 0, |
1360 | COMM_ERR_CLOSING, 0, fdc_table[fd].read.handler_data); | |
511e2383 | 1361 | fdc_table[fd].read.handler = NULL; |
1362 | } | |
1363 | if (fdc_table[fd].accept.handler) { | |
691476e2 | 1364 | fdc_table[fd].accept.handler(fd, -1, NULL, NULL, COMM_ERR_CLOSING, |
1365 | 0, fdc_table[fd].accept.handler_data); | |
511e2383 | 1366 | fdc_table[fd].accept.handler = NULL; |
1367 | } | |
1368 | if (fdc_table[fd].fill.handler) { | |
691476e2 | 1369 | fdc_table[fd].fill.handler(fd, fdc_table[fd].fill.requestedData, COMM_ERR_CLOSING, 0, |
1370 | fdc_table[fd].fill.handler_data); | |
511e2383 | 1371 | fdc_table[fd].fill.handler = NULL; |
1372 | } | |
1373 | /* Complete (w/ COMM_ERR_CLOSING!) any pending io callbacks */ | |
c4b7a5a9 | 1374 | while (fdc_table[fd].CommCallbackList.head != NULL) { |
1375 | node = fdc_table[fd].CommCallbackList.head; | |
1376 | cio = (CommCallbackData *)node->data; | |
1377 | assert(fd == cio->fd); /* just paranoid */ | |
1378 | dlinkDelete(&cio->h_node, &CommCallbackList); | |
1379 | dlinkDelete(&cio->fd_node, &(fdc_table[cio->fd].CommCallbackList)); | |
504cd889 | 1380 | /* We're closing! */ |
1381 | cio->errcode = COMM_ERR_CLOSING; | |
c4b7a5a9 | 1382 | comm_call_io_callback(cio); |
1383 | memPoolFree(comm_callback_pool, cio); | |
1384 | } | |
1385 | ||
cb201b7e | 1386 | commCallCloseHandlers(fd); |
b716a8ad | 1387 | if (F->uses) /* assume persistent connect count */ |
1388 | pconnHistCount(1, F->uses); | |
a7ad6e4e | 1389 | comm_empty_os_read_buffers(fd); |
d4c19b39 | 1390 | #if USE_SSL |
1391 | if (F->ssl) { | |
1392 | SSL_free(F->ssl); | |
1393 | F->ssl = NULL; | |
1394 | } | |
1395 | #endif | |
5c5783a2 | 1396 | fd_close(fd); /* update fdstat */ |
5874bf28 | 1397 | close(fd); |
c4b7a5a9 | 1398 | fdc_table[fd].active = 0; |
1399 | bzero(&fdc_table[fd], sizeof(fdc_t)); | |
83704487 | 1400 | statCounter.syscalls.sock.closes++; |
88bfe092 | 1401 | PROF_stop(comm_close); |
090089c4 | 1402 | } |
1403 | ||
090089c4 | 1404 | /* Send a udp datagram to specified TO_ADDR. */ |
b8d8561b | 1405 | int |
5df61230 | 1406 | comm_udp_sendto(int fd, |
1407 | const struct sockaddr_in *to_addr, | |
1408 | int addr_len, | |
17b6e784 | 1409 | const void *buf, |
5df61230 | 1410 | int len) |
090089c4 | 1411 | { |
5df61230 | 1412 | int x; |
88bfe092 | 1413 | PROF_start(comm_udp_sendto); |
83704487 | 1414 | statCounter.syscalls.sock.sendtos++; |
5df61230 | 1415 | x = sendto(fd, buf, len, 0, (struct sockaddr *) to_addr, addr_len); |
88bfe092 | 1416 | PROF_stop(comm_udp_sendto); |
5df61230 | 1417 | if (x < 0) { |
17d51783 | 1418 | #ifdef _SQUID_LINUX_ |
1419 | if (ECONNREFUSED != errno) | |
1420 | #endif | |
1421 | debug(50, 1) ("comm_udp_sendto: FD %d, %s, port %d: %s\n", | |
1422 | fd, | |
1423 | inet_ntoa(to_addr->sin_addr), | |
1424 | (int) htons(to_addr->sin_port), | |
1425 | xstrerror()); | |
090089c4 | 1426 | return COMM_ERROR; |
1427 | } | |
5df61230 | 1428 | return x; |
090089c4 | 1429 | } |
1430 | ||
b8d8561b | 1431 | void |
70a9dab4 | 1432 | commSetDefer(int fd, DEFER * func, void *data) |
4883993a | 1433 | { |
da2b3a17 | 1434 | fde *F = &fd_table[fd]; |
1435 | F->defer_check = func; | |
70a9dab4 | 1436 | F->defer_data = data; |
4883993a | 1437 | } |
1438 | ||
b8d8561b | 1439 | void |
582b6456 | 1440 | comm_add_close_handler(int fd, PF * handler, void *data) |
30a4f2a8 | 1441 | { |
e6ccf245 | 1442 | close_handler *newHandler = (close_handler *)memPoolAlloc(conn_close_pool); /* AAA */ |
cddc721b | 1443 | close_handler *c; |
a3d5953d | 1444 | debug(5, 5) ("comm_add_close_handler: FD %d, handler=%p, data=%p\n", |
e0c42e90 | 1445 | fd, handler, data); |
29b8d8d6 | 1446 | for (c = fd_table[fd].closeHandler; c; c = c->next) |
aeca2a09 | 1447 | assert(c->handler != handler || c->data != data); |
e6ccf245 | 1448 | newHandler->handler = handler; |
1449 | newHandler->data = cbdataReference(data); | |
1450 | newHandler->next = fd_table[fd].closeHandler; | |
1451 | fd_table[fd].closeHandler = newHandler; | |
30a4f2a8 | 1452 | } |
1453 | ||
b8d8561b | 1454 | void |
582b6456 | 1455 | comm_remove_close_handler(int fd, PF * handler, void *data) |
090089c4 | 1456 | { |
f1dc9b30 | 1457 | close_handler *p; |
1458 | close_handler *last = NULL; | |
30a4f2a8 | 1459 | /* Find handler in list */ |
e869f2bd | 1460 | debug(5, 5) ("comm_remove_close_handler: FD %d, handler=%p, data=%p\n", |
1461 | fd, handler, data); | |
29b8d8d6 | 1462 | for (p = fd_table[fd].closeHandler; p != NULL; last = p, p = p->next) |
30a4f2a8 | 1463 | if (p->handler == handler && p->data == data) |
1464 | break; /* This is our handler */ | |
f88211e8 | 1465 | assert(p != NULL); |
30a4f2a8 | 1466 | /* Remove list entry */ |
1467 | if (last) | |
1468 | last->next = p->next; | |
1469 | else | |
29b8d8d6 | 1470 | fd_table[fd].closeHandler = p->next; |
fa80a8ef | 1471 | cbdataReferenceDone(p->data); |
1472 | memPoolFree(conn_close_pool, p); | |
30a4f2a8 | 1473 | } |
090089c4 | 1474 | |
b8d8561b | 1475 | static void |
1476 | commSetNoLinger(int fd) | |
30a4f2a8 | 1477 | { |
1478 | struct linger L; | |
090089c4 | 1479 | L.l_onoff = 0; /* off */ |
1480 | L.l_linger = 0; | |
30a4f2a8 | 1481 | if (setsockopt(fd, SOL_SOCKET, SO_LINGER, (char *) &L, sizeof(L)) < 0) |
a3d5953d | 1482 | debug(50, 0) ("commSetNoLinger: FD %d: %s\n", fd, xstrerror()); |
58a6c186 | 1483 | fd_table[fd].flags.nolinger = 1; |
090089c4 | 1484 | } |
1485 | ||
b8d8561b | 1486 | static void |
1487 | commSetReuseAddr(int fd) | |
090089c4 | 1488 | { |
1489 | int on = 1; | |
30a4f2a8 | 1490 | if (setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, (char *) &on, sizeof(on)) < 0) |
a3d5953d | 1491 | debug(50, 1) ("commSetReuseAddr: FD %d: %s\n", fd, xstrerror()); |
090089c4 | 1492 | } |
1493 | ||
b8d8561b | 1494 | static void |
1495 | commSetTcpRcvbuf(int fd, int size) | |
f868539a | 1496 | { |
1497 | if (setsockopt(fd, SOL_SOCKET, SO_RCVBUF, (char *) &size, sizeof(size)) < 0) | |
a3d5953d | 1498 | debug(50, 1) ("commSetTcpRcvbuf: FD %d, SIZE %d: %s\n", |
b6f794d6 | 1499 | fd, size, xstrerror()); |
f868539a | 1500 | } |
1501 | ||
b8d8561b | 1502 | int |
1503 | commSetNonBlocking(int fd) | |
30a4f2a8 | 1504 | { |
731e4d49 | 1505 | int flags; |
9e205701 | 1506 | int dummy = 0; |
7f6ffd15 | 1507 | #ifdef _SQUID_CYGWIN_ |
b05490a8 | 1508 | int nonblocking = TRUE; |
7f6ffd15 | 1509 | if (fd_table[fd].type != FD_PIPE) { |
1510 | if (ioctl(fd, FIONBIO, &nonblocking) < 0) { | |
c4b7a5a9 | 1511 | debug(50, 0) ("commSetNonBlocking: FD %d: %s %D\n", fd, xstrerror(), fd_table[fd].type); |
7f6ffd15 | 1512 | return COMM_ERROR; |
1513 | } | |
1514 | } else { | |
1515 | #endif | |
1516 | if ((flags = fcntl(fd, F_GETFL, dummy)) < 0) { | |
1517 | debug(50, 0) ("FD %d: fcntl F_GETFL: %s\n", fd, xstrerror()); | |
1518 | return COMM_ERROR; | |
1519 | } | |
1520 | if (fcntl(fd, F_SETFL, flags | SQUID_NONBLOCK) < 0) { | |
1521 | debug(50, 0) ("commSetNonBlocking: FD %d: %s\n", fd, xstrerror()); | |
1522 | return COMM_ERROR; | |
1523 | } | |
1524 | #ifdef _SQUID_CYGWIN_ | |
090089c4 | 1525 | } |
7f6ffd15 | 1526 | #endif |
58a6c186 | 1527 | fd_table[fd].flags.nonblocking = 1; |
090089c4 | 1528 | return 0; |
1529 | } | |
1530 | ||
7e3ce7b9 | 1531 | int |
1532 | commUnsetNonBlocking(int fd) | |
1533 | { | |
1534 | int flags; | |
1535 | int dummy = 0; | |
1536 | if ((flags = fcntl(fd, F_GETFL, dummy)) < 0) { | |
1537 | debug(50, 0) ("FD %d: fcntl F_GETFL: %s\n", fd, xstrerror()); | |
1538 | return COMM_ERROR; | |
1539 | } | |
1540 | if (fcntl(fd, F_SETFL, flags & (~SQUID_NONBLOCK)) < 0) { | |
1541 | debug(50, 0) ("commUnsetNonBlocking: FD %d: %s\n", fd, xstrerror()); | |
1542 | return COMM_ERROR; | |
1543 | } | |
1544 | fd_table[fd].flags.nonblocking = 0; | |
1545 | return 0; | |
1546 | } | |
1547 | ||
b8d8561b | 1548 | void |
1549 | commSetCloseOnExec(int fd) | |
3ca60c86 | 1550 | { |
1551 | #ifdef FD_CLOEXEC | |
731e4d49 | 1552 | int flags; |
7a18b487 | 1553 | int dummy = 0; |
c7989865 | 1554 | if ((flags = fcntl(fd, F_GETFL, dummy)) < 0) { |
a3d5953d | 1555 | debug(50, 0) ("FD %d: fcntl F_GETFL: %s\n", fd, xstrerror()); |
24382924 | 1556 | return; |
3ca60c86 | 1557 | } |
24382924 | 1558 | if (fcntl(fd, F_SETFD, flags | FD_CLOEXEC) < 0) |
a3d5953d | 1559 | debug(50, 0) ("FD %d: set close-on-exec failed: %s\n", fd, xstrerror()); |
d6827718 | 1560 | fd_table[fd].flags.close_on_exec = 1; |
3ca60c86 | 1561 | #endif |
1562 | } | |
1563 | ||
e90100aa | 1564 | #ifdef TCP_NODELAY |
1565 | static void | |
1566 | commSetTcpNoDelay(int fd) | |
1567 | { | |
1568 | int on = 1; | |
1569 | if (setsockopt(fd, IPPROTO_TCP, TCP_NODELAY, (char *) &on, sizeof(on)) < 0) | |
a3d5953d | 1570 | debug(50, 1) ("commSetTcpNoDelay: FD %d: %s\n", fd, xstrerror()); |
d6827718 | 1571 | fd_table[fd].flags.nodelay = 1; |
e90100aa | 1572 | } |
1573 | #endif | |
1574 | ||
6a988308 | 1575 | |
d86b3703 | 1576 | void |
0673c0ba | 1577 | comm_init(void) |
090089c4 | 1578 | { |
c4b7a5a9 | 1579 | fd_table =(fde *) xcalloc(Squid_MaxFD, sizeof(fde)); |
1580 | fdd_table = (fd_debug_t *)xcalloc(Squid_MaxFD, sizeof(fd_debug_t)); | |
1581 | fdc_table = (fdc_t *)xcalloc(Squid_MaxFD, sizeof(fdc_t)); | |
59c4d35b | 1582 | /* XXX account fd_table */ |
090089c4 | 1583 | /* Keep a few file descriptors free so that we don't run out of FD's |
1584 | * after accepting a client but before it opens a socket or a file. | |
e83892e9 | 1585 | * Since Squid_MaxFD can be as high as several thousand, don't waste them */ |
0254ee29 | 1586 | RESERVED_FD = XMIN(100, Squid_MaxFD / 4); |
28c60158 | 1587 | CBDATA_INIT_TYPE(ConnectStateData); |
c4b7a5a9 | 1588 | |
1589 | comm_callback_pool = memPoolCreate("comm callbacks", sizeof(CommCallbackData)); | |
723123a9 | 1590 | comm_write_pool = memPoolCreate("CommWriteStateData", sizeof(CommWriteStateData)); |
58cd5bbd | 1591 | conn_close_pool = memPoolCreate("close_handler", sizeof(close_handler)); |
090089c4 | 1592 | } |
1593 | ||
30a4f2a8 | 1594 | /* Write to FD. */ |
b8d8561b | 1595 | static void |
582b6456 | 1596 | commHandleWrite(int fd, void *data) |
30a4f2a8 | 1597 | { |
e6ccf245 | 1598 | CommWriteStateData *state = (CommWriteStateData *)data; |
30a4f2a8 | 1599 | int len = 0; |
1600 | int nleft; | |
1601 | ||
88bfe092 | 1602 | PROF_start(commHandleWrite); |
32754419 | 1603 | debug(5, 5) ("commHandleWrite: FD %d: off %ld, sz %ld.\n", |
1604 | fd, (long int) state->offset, (long int) state->size); | |
30a4f2a8 | 1605 | |
1606 | nleft = state->size - state->offset; | |
1f7c9178 | 1607 | len = FD_WRITE_METHOD(fd, state->buf + state->offset, nleft); |
6a988308 | 1608 | debug(5, 5) ("commHandleWrite: write() returns %d\n", len); |
b69f7771 | 1609 | fd_bytes(fd, len, FD_WRITE); |
83704487 | 1610 | statCounter.syscalls.sock.writes++; |
30a4f2a8 | 1611 | |
1612 | if (len == 0) { | |
1613 | /* Note we even call write if nleft == 0 */ | |
1614 | /* We're done */ | |
1615 | if (nleft != 0) | |
02be0294 | 1616 | debug(5, 1) ("commHandleWrite: FD %d: write failure: connection closed with %d bytes remaining.\n", fd, nleft); |
f17936ab | 1617 | CommWriteStateCallbackAndFree(fd, nleft ? COMM_ERROR : COMM_OK); |
30a4f2a8 | 1618 | } else if (len < 0) { |
1619 | /* An error */ | |
e8d6569c | 1620 | if (fd_table[fd].flags.socket_eof) { |
1621 | debug(50, 2) ("commHandleWrite: FD %d: write failure: %s.\n", | |
1622 | fd, xstrerror()); | |
1623 | CommWriteStateCallbackAndFree(fd, COMM_ERROR); | |
1624 | } else if (ignoreErrno(errno)) { | |
a3d5953d | 1625 | debug(50, 10) ("commHandleWrite: FD %d: write failure: %s.\n", |
30a4f2a8 | 1626 | fd, xstrerror()); |
b177367b | 1627 | commSetSelect(fd, |
30a4f2a8 | 1628 | COMM_SELECT_WRITE, |
cd1fb0eb | 1629 | commHandleWrite, |
b177367b | 1630 | state, |
85d7ea98 | 1631 | 0); |
9864ee44 | 1632 | } else { |
a3d5953d | 1633 | debug(50, 2) ("commHandleWrite: FD %d: write failure: %s.\n", |
9864ee44 | 1634 | fd, xstrerror()); |
f17936ab | 1635 | CommWriteStateCallbackAndFree(fd, COMM_ERROR); |
30a4f2a8 | 1636 | } |
30a4f2a8 | 1637 | } else { |
1638 | /* A successful write, continue */ | |
1639 | state->offset += len; | |
e6ccf245 | 1640 | if (state->offset < (off_t)state->size) { |
30a4f2a8 | 1641 | /* Not done, reinstall the write handler and write some more */ |
b177367b | 1642 | commSetSelect(fd, |
30a4f2a8 | 1643 | COMM_SELECT_WRITE, |
cd1fb0eb | 1644 | commHandleWrite, |
b177367b | 1645 | state, |
85d7ea98 | 1646 | 0); |
9864ee44 | 1647 | } else { |
f17936ab | 1648 | CommWriteStateCallbackAndFree(fd, COMM_OK); |
30a4f2a8 | 1649 | } |
30a4f2a8 | 1650 | } |
88bfe092 | 1651 | PROF_stop(commHandleWrite); |
30a4f2a8 | 1652 | } |
1653 | ||
1654 | ||
1655 | ||
7cd8c414 | 1656 | /* |
1657 | * Queue a write. handler/handler_data are called when the write | |
1658 | * completes, on error, or on file descriptor close. | |
1659 | * | |
1660 | * free_func is used to free the passed buffer when the write has completed. | |
1661 | */ | |
b8d8561b | 1662 | void |
d4cb310b | 1663 | comm_old_write(int fd, const char *buf, int size, CWCB * handler, void *handler_data, FREE * free_func) |
30a4f2a8 | 1664 | { |
aa9e2cab | 1665 | CommWriteStateData *state = fd_table[fd].rwstate; |
c4b7a5a9 | 1666 | |
1667 | assert(!fd_table[fd].flags.closing); | |
1668 | ||
a3d5953d | 1669 | debug(5, 5) ("comm_write: FD %d: sz %d: hndl %p: data %p.\n", |
787869c5 | 1670 | fd, size, handler, handler_data); |
aa9e2cab | 1671 | if (NULL != state) { |
afde8a9d | 1672 | debug(5, 1) ("comm_write: fd_table[%d].rwstate != NULL\n", fd); |
723123a9 | 1673 | memPoolFree(comm_write_pool, state); |
6cf028ab | 1674 | fd_table[fd].rwstate = NULL; |
1675 | } | |
e6ccf245 | 1676 | fd_table[fd].rwstate = state = (CommWriteStateData *)memPoolAlloc(comm_write_pool); |
a2c963ae | 1677 | state->buf = (char *) buf; |
30a4f2a8 | 1678 | state->size = size; |
1679 | state->offset = 0; | |
1680 | state->handler = handler; | |
fa80a8ef | 1681 | state->handler_data = cbdataReference(handler_data); |
c0dec081 | 1682 | state->free_func = free_func; |
aa9e2cab | 1683 | commSetSelect(fd, COMM_SELECT_WRITE, commHandleWrite, state, 0); |
30a4f2a8 | 1684 | } |
26a880e2 | 1685 | |
137ee196 | 1686 | /* a wrapper around comm_write to allow for MemBuf to be comm_written in a snap */ |
cb69b4c7 | 1687 | void |
d4cb310b | 1688 | comm_old_write_mbuf(int fd, MemBuf mb, CWCB * handler, void *handler_data) |
cb69b4c7 | 1689 | { |
d4cb310b | 1690 | comm_old_write(fd, mb.buf, mb.size, handler, handler_data, memBufFreeFunc(&mb)); |
cb69b4c7 | 1691 | } |
1692 | ||
c4b7a5a9 | 1693 | |
89924214 | 1694 | /* |
1695 | * hm, this might be too general-purpose for all the places we'd | |
1696 | * like to use it. | |
1697 | */ | |
b224ea98 | 1698 | int |
edd2eb63 | 1699 | ignoreErrno(int ierrno) |
26a880e2 | 1700 | { |
603500e7 | 1701 | switch (ierrno) { |
89924214 | 1702 | case EINPROGRESS: |
603500e7 | 1703 | case EWOULDBLOCK: |
26a880e2 | 1704 | #if EAGAIN != EWOULDBLOCK |
603500e7 | 1705 | case EAGAIN: |
26a880e2 | 1706 | #endif |
603500e7 | 1707 | case EALREADY: |
1708 | case EINTR: | |
db494ab8 | 1709 | #ifdef ERESTART |
1710 | case ERESTART: | |
1711 | #endif | |
26a880e2 | 1712 | return 1; |
603500e7 | 1713 | default: |
1714 | return 0; | |
1715 | } | |
1716 | /* NOTREACHED */ | |
26a880e2 | 1717 | } |
d723bf6b | 1718 | |
1719 | void | |
1720 | commCloseAllSockets(void) | |
1721 | { | |
1722 | int fd; | |
1723 | fde *F = NULL; | |
d723bf6b | 1724 | for (fd = 0; fd <= Biggest_FD; fd++) { |
1725 | F = &fd_table[fd]; | |
60c0b5a2 | 1726 | if (!F->flags.open) |
d723bf6b | 1727 | continue; |
1728 | if (F->type != FD_SOCKET) | |
1729 | continue; | |
de718ec4 | 1730 | if (F->flags.ipc) /* don't close inter-process sockets */ |
1731 | continue; | |
d723bf6b | 1732 | if (F->timeout_handler) { |
fa80a8ef | 1733 | PF *callback = F->timeout_handler; |
1734 | void *cbdata = NULL; | |
1735 | F->timeout_handler = NULL; | |
d723bf6b | 1736 | debug(5, 5) ("commCloseAllSockets: FD %d: Calling timeout handler\n", |
1737 | fd); | |
fa80a8ef | 1738 | if (cbdataReferenceValidDone(F->timeout_data, &cbdata)) |
1739 | callback(fd, cbdata); | |
d723bf6b | 1740 | } else { |
1741 | debug(5, 5) ("commCloseAllSockets: FD %d: calling comm_close()\n", fd); | |
1742 | comm_close(fd); | |
1743 | } | |
1744 | } | |
1745 | } | |
1b3db6d9 | 1746 | |
1747 | void | |
1748 | checkTimeouts(void) | |
1749 | { | |
1750 | int fd; | |
1751 | fde *F = NULL; | |
1752 | PF *callback; | |
1753 | for (fd = 0; fd <= Biggest_FD; fd++) { | |
b5443c04 | 1754 | F = &fd_table[fd]; |
1755 | if (!F->flags.open) | |
1756 | continue; | |
1757 | if (F->timeout == 0) | |
1758 | continue; | |
1759 | if (F->timeout > squid_curtime) | |
1760 | continue; | |
1761 | debug(5, 5) ("checkTimeouts: FD %d Expired\n", fd); | |
1762 | if (F->timeout_handler) { | |
1763 | debug(5, 5) ("checkTimeouts: FD %d: Call timeout handler\n", fd); | |
1764 | callback = F->timeout_handler; | |
1765 | F->timeout_handler = NULL; | |
1766 | callback(fd, F->timeout_data); | |
1767 | } else { | |
1768 | debug(5, 5) ("checkTimeouts: FD %d: Forcing comm_close()\n", fd); | |
1769 | comm_close(fd); | |
1770 | } | |
1771 | } | |
1772 | } | |
1773 | ||
1774 | ||
1775 | int | |
1b3db6d9 | 1776 | commDeferRead(int fd) |
1777 | { | |
1778 | fde *F = &fd_table[fd]; | |
1779 | if (F->defer_check == NULL) | |
b5443c04 | 1780 | return 0; |
1b3db6d9 | 1781 | return F->defer_check(fd, F->defer_data); |
1782 | } | |
c4b7a5a9 | 1783 | |
1784 | ||
1785 | /* | |
1786 | * New-style listen and accept routines | |
1787 | * | |
1788 | * Listen simply registers our interest in an FD for listening, | |
1789 | * and accept takes a callback to call when an FD has been | |
1790 | * accept()ed. | |
1791 | */ | |
1792 | int | |
1793 | comm_listen(int sock) | |
1794 | { | |
1795 | int x; | |
1796 | if ((x = listen(sock, Squid_MaxFD >> 2)) < 0) { | |
1797 | debug(50, 0) ("comm_listen: listen(%d, %d): %s\n", | |
1798 | Squid_MaxFD >> 2, | |
1799 | sock, xstrerror()); | |
1800 | return x; | |
1801 | } | |
1802 | return sock; | |
1803 | } | |
1804 | ||
1805 | ||
1806 | /* | |
1807 | * This callback is called whenever a filedescriptor is ready | |
1808 | * to dupe itself and fob off an accept()ed connection | |
1809 | */ | |
1810 | static void | |
1811 | comm_accept_try(int fd, void *data) | |
1812 | { | |
1813 | int newfd; | |
1814 | fdc_t *Fc; | |
f49a8979 | 1815 | int count; |
1816 | IOACB *hdl; | |
c4b7a5a9 | 1817 | |
1818 | assert(fdc_table[fd].active == 1); | |
1819 | ||
1820 | Fc = &(fdc_table[fd]); | |
1821 | ||
f49a8979 | 1822 | for (count = 0; count < MAX_ACCEPT_PER_LOOP; count++) { |
3c1a197f | 1823 | /* If we're out of fds, register an event and return now */ |
a7c5ae5d | 1824 | if (fdNFree() < RESERVED_FD) { |
3c1a197f | 1825 | debug(5, 3) ("comm_accept_try: we're out of fds - deferring io!\n"); |
1826 | eventAdd("comm_accept_check_event", comm_accept_check_event, &fdc_table[fd], | |
1827 | 1000.0 / (double)(fdc_table[fd].accept.check_delay), 1); | |
1828 | return; | |
1829 | } | |
f49a8979 | 1830 | /* Accept a new connection */ |
1831 | newfd = comm_old_accept(fd, &Fc->accept.pn, &Fc->accept.me); | |
2409da33 | 1832 | /* Check for errors */ |
1833 | if (newfd < 0) { | |
f49a8979 | 1834 | if (newfd == COMM_NOMESSAGE) { |
1835 | /* register interest again */ | |
1836 | commSetSelect(fd, COMM_SELECT_READ, comm_accept_try, NULL, 0); | |
1837 | return; | |
1838 | } | |
2409da33 | 1839 | /* A non-recoverable error - register an error callback */ |
f49a8979 | 1840 | comm_addacceptcallback(fd, -1, Fc->accept.handler, &Fc->accept.pn, |
1841 | &Fc->accept.me, COMM_ERROR, errno, Fc->accept.handler_data); | |
1842 | Fc->accept.handler = NULL; | |
1843 | Fc->accept.handler_data = NULL; | |
c4b7a5a9 | 1844 | return; |
1845 | } | |
c4b7a5a9 | 1846 | |
f49a8979 | 1847 | /* Try the callback! */ |
1848 | hdl = Fc->accept.handler; | |
1849 | Fc->accept.handler = NULL; | |
1850 | hdl(fd, newfd, &Fc->accept.pn, &Fc->accept.me, COMM_OK, 0, Fc->accept.handler_data); | |
c4b7a5a9 | 1851 | |
f49a8979 | 1852 | /* If we weren't re-registed, don't bother trying again! */ |
1853 | if (Fc->accept.handler == NULL) | |
1854 | return; | |
1855 | } | |
c4b7a5a9 | 1856 | } |
1857 | ||
1858 | ||
1859 | /* | |
1860 | * Notes: | |
1861 | * + the current interface will queue _one_ accept per io loop. | |
1862 | * this isn't very optimal and should be revisited at a later date. | |
1863 | */ | |
1864 | void | |
1865 | comm_accept(int fd, IOACB *handler, void *handler_data) | |
1866 | { | |
1867 | fdc_t *Fc; | |
1868 | ||
1869 | assert(fd_table[fd].flags.open == 1); | |
1870 | assert(fdc_table[fd].active == 1); | |
1871 | ||
1872 | /* make sure we're not pending! */ | |
1873 | assert(fdc_table[fd].accept.handler == NULL); | |
1874 | ||
1875 | /* Record our details */ | |
1876 | Fc = &fdc_table[fd]; | |
1877 | Fc->accept.handler = handler; | |
1878 | Fc->accept.handler_data = handler_data; | |
1879 | ||
1880 | /* Kick off the accept */ | |
1881 | #if OPTIMISTIC_IO | |
1882 | comm_accept_try(fd, NULL); | |
1883 | #else | |
1884 | commSetSelect(fd, COMM_SELECT_READ, comm_accept_try, NULL, 0); | |
1885 | #endif | |
1886 | } | |
6cce2334 | 1887 | |
1888 | void CommIO::Initialise() | |
1889 | { | |
1890 | /* Initialize done pipe signal */ | |
1891 | int DonePipe[2]; | |
1892 | pipe(DonePipe); | |
1893 | DoneFD = DonePipe[1]; | |
1894 | DoneReadFD = DonePipe[0]; | |
1895 | fd_open(DonePipe[0], FD_PIPE, "async-io completetion event: main"); | |
1896 | fd_open(DonePipe[1], FD_PIPE, "async-io completetion event: threads"); | |
1897 | commSetNonBlocking(DonePipe[0]); | |
1898 | commSetNonBlocking(DonePipe[1]); | |
1899 | commSetSelect(DonePipe[0], COMM_SELECT_READ, NULLFDHandler, NULL, 0); | |
1900 | Initialised = true; | |
1901 | } | |
1902 | ||
1903 | bool CommIO::Initialised = false; | |
1904 | bool CommIO::DoneSignalled = false; | |
1905 | int CommIO::DoneFD = -1; | |
1906 | int CommIO::DoneReadFD = -1; | |
1907 | ||
1908 | void | |
1909 | CommIO::FlushPipe() | |
1910 | { | |
1911 | char buf[256]; | |
1912 | read(DoneReadFD, buf, sizeof(buf)); | |
1913 | } | |
1914 | ||
1915 | void | |
1916 | CommIO::NULLFDHandler(int fd, void *data) | |
1917 | { | |
1918 | FlushPipe(); | |
1919 | commSetSelect(fd, COMM_SELECT_READ, NULLFDHandler, NULL, 0); | |
1920 | } | |
1921 | ||
1922 | void | |
1923 | CommIO::ResetNotifications() | |
1924 | { | |
1925 | if (DoneSignalled) { | |
1926 | FlushPipe(); | |
1927 | DoneSignalled = false; | |
1928 | } | |
1929 | } |