Commit | Line | Data |
aadcc3c8 |
1 | /* |
0198fd3c |
2 | * os_unix.c -- |
3 | * |
4 | * Description of file. |
5 | * |
6 | * |
7 | * Copyright (c) 1995 Open Market, Inc. |
8 | * All rights reserved. |
9 | * |
10 | * This file contains proprietary and confidential information and |
aadcc3c8 |
11 | * remains the unpublished property of Open Market, Inc. Use, |
12 | * disclosure, or reproduction is prohibited except as permitted by |
13 | * express written license agreement with Open Market, Inc. |
0198fd3c |
14 | * |
15 | * Bill Snapper |
16 | * snapper@openmarket.com |
17 | */ |
18 | |
19 | #ifndef lint |
1dd5d7a8 |
20 | static const char rcsid[] = "$Id: os_unix.c,v 1.14 1999/08/12 23:56:11 roberts Exp $"; |
0198fd3c |
21 | #endif /* not lint */ |
22 | |
6ad90ad2 |
23 | #include "fcgi_config.h" |
0198fd3c |
24 | |
6ad90ad2 |
25 | #include <arpa/inet.h> |
0198fd3c |
26 | #include <assert.h> |
0198fd3c |
27 | #include <errno.h> |
6ad90ad2 |
28 | #include <fcntl.h> /* for fcntl */ |
0198fd3c |
29 | #include <math.h> |
6ad90ad2 |
30 | #include <memory.h> /* for memchr() */ |
31 | #include <netinet/tcp.h> |
32 | #include <stdarg.h> |
33 | #include <stdio.h> |
34 | #include <stdlib.h> |
35 | #include <string.h> |
36 | #include <sys/time.h> |
37 | #include <sys/types.h> |
0198fd3c |
38 | #include <sys/un.h> |
6ad90ad2 |
39 | |
0198fd3c |
40 | #ifdef HAVE_NETDB_H |
41 | #include <netdb.h> |
42 | #endif |
0198fd3c |
43 | |
0198fd3c |
44 | #ifdef HAVE_NETINET_IN_H |
45 | #include <netinet/in.h> |
46 | #endif |
0198fd3c |
47 | |
6ad90ad2 |
48 | #ifdef HAVE_SYS_SOCKET_H |
49 | #include <sys/socket.h> /* for getpeername */ |
50 | #endif |
51 | |
52 | #ifdef HAVE_UNISTD_H |
53 | #include <unistd.h> |
54 | #endif |
55 | |
56 | #include "fastcgi.h" |
57 | #include "fcgiapp.h" |
58 | #include "fcgiappmisc.h" |
59 | #include "fcgimisc.h" |
0198fd3c |
60 | #include "fcgios.h" |
61 | |
6ad90ad2 |
62 | #ifndef FALSE |
63 | #define FALSE 0 |
64 | #endif |
65 | |
66 | #ifndef TRUE |
67 | #define TRUE 1 |
0198fd3c |
68 | #endif |
69 | |
70 | /* |
71 | * This structure holds an entry for each oustanding async I/O operation. |
72 | */ |
73 | typedef struct { |
74 | OS_AsyncProc procPtr; /* callout completion procedure */ |
75 | ClientData clientData; /* caller private data */ |
76 | int fd; |
77 | int len; |
78 | int offset; |
79 | void *buf; |
80 | int inUse; |
81 | } AioInfo; |
82 | |
83 | /* |
84 | * Entries in the async I/O table are allocated 2 per file descriptor. |
85 | * |
86 | * Read Entry Index = fd * 2 |
87 | * Write Entry Index = (fd * 2) + 1 |
88 | */ |
89 | #define AIO_RD_IX(fd) (fd * 2) |
90 | #define AIO_WR_IX(fd) ((fd * 2) + 1) |
91 | |
92 | static int asyncIoTableSize = 16; |
93 | static AioInfo *asyncIoTable = NULL; |
0198fd3c |
94 | |
0198fd3c |
95 | static int libInitialized = FALSE; |
96 | |
97 | static fd_set readFdSet; |
98 | static fd_set writeFdSet; |
99 | |
100 | static fd_set readFdSetPost; |
101 | static int numRdPosted = 0; |
102 | static fd_set writeFdSetPost; |
103 | static int numWrPosted = 0; |
104 | static int volatile maxFd = -1; |
105 | |
0198fd3c |
106 | \f |
107 | /* |
108 | *-------------------------------------------------------------- |
109 | * |
110 | * OS_LibInit -- |
111 | * |
112 | * Set up the OS library for use. |
113 | * |
114 | * NOTE: This function is really only needed for application |
115 | * asynchronous I/O. It will most likely change in the |
116 | * future to setup the multi-threaded environment. |
117 | * |
118 | * Results: |
119 | * Returns 0 if success, -1 if not. |
120 | * |
121 | * Side effects: |
122 | * Async I/O table allocated and initialized. |
123 | * |
124 | *-------------------------------------------------------------- |
125 | */ |
126 | int OS_LibInit(int stdioFds[3]) |
127 | { |
128 | if(libInitialized) |
129 | return 0; |
aadcc3c8 |
130 | |
131 | asyncIoTable = (AioInfo *)malloc(asyncIoTableSize * sizeof(AioInfo)); |
0198fd3c |
132 | if(asyncIoTable == NULL) { |
133 | errno = ENOMEM; |
134 | return -1; |
135 | } |
136 | memset((char *) asyncIoTable, 0, |
137 | asyncIoTableSize * sizeof(AioInfo)); |
138 | |
139 | FD_ZERO(&readFdSet); |
140 | FD_ZERO(&writeFdSet); |
141 | FD_ZERO(&readFdSetPost); |
142 | FD_ZERO(&writeFdSetPost); |
143 | libInitialized = TRUE; |
144 | return 0; |
145 | } |
146 | |
147 | \f |
148 | /* |
149 | *-------------------------------------------------------------- |
150 | * |
151 | * OS_LibShutdown -- |
152 | * |
153 | * Shutdown the OS library. |
154 | * |
155 | * Results: |
156 | * None. |
157 | * |
158 | * Side effects: |
159 | * Memory freed, fds closed. |
160 | * |
161 | *-------------------------------------------------------------- |
162 | */ |
163 | void OS_LibShutdown() |
164 | { |
165 | if(!libInitialized) |
166 | return; |
aadcc3c8 |
167 | |
0198fd3c |
168 | free(asyncIoTable); |
169 | asyncIoTable = NULL; |
170 | libInitialized = FALSE; |
171 | return; |
172 | } |
173 | |
174 | \f |
175 | /* |
176 | *---------------------------------------------------------------------- |
177 | * |
178 | * OS_BuildSockAddrUn -- |
179 | * |
180 | * Using the pathname bindPath, fill in the sockaddr_un structure |
181 | * *servAddrPtr and the length of this structure *servAddrLen. |
182 | * |
183 | * The format of the sockaddr_un structure changed incompatibly in |
184 | * 4.3BSD Reno. Digital UNIX supports both formats, other systems |
185 | * support one or the other. |
186 | * |
187 | * Results: |
188 | * 0 for normal return, -1 for failure (bindPath too long). |
189 | * |
190 | *---------------------------------------------------------------------- |
191 | */ |
192 | |
0b7c9662 |
193 | static int OS_BuildSockAddrUn(const char *bindPath, |
0198fd3c |
194 | struct sockaddr_un *servAddrPtr, |
195 | int *servAddrLen) |
196 | { |
197 | int bindPathLen = strlen(bindPath); |
198 | |
199 | #ifdef HAVE_SOCKADDR_UN_SUN_LEN /* 4.3BSD Reno and later: BSDI, DEC */ |
200 | if(bindPathLen >= sizeof(servAddrPtr->sun_path)) { |
201 | return -1; |
202 | } |
203 | #else /* 4.3 BSD Tahoe: Solaris, HPUX, DEC, ... */ |
204 | if(bindPathLen > sizeof(servAddrPtr->sun_path)) { |
205 | return -1; |
206 | } |
207 | #endif |
208 | memset((char *) servAddrPtr, 0, sizeof(*servAddrPtr)); |
209 | servAddrPtr->sun_family = AF_UNIX; |
210 | memcpy(servAddrPtr->sun_path, bindPath, bindPathLen); |
211 | #ifdef HAVE_SOCKADDR_UN_SUN_LEN /* 4.3BSD Reno and later: BSDI, DEC */ |
212 | *servAddrLen = sizeof(servAddrPtr->sun_len) |
213 | + sizeof(servAddrPtr->sun_family) |
214 | + bindPathLen + 1; |
215 | servAddrPtr->sun_len = *servAddrLen; |
216 | #else /* 4.3 BSD Tahoe: Solaris, HPUX, DEC, ... */ |
217 | *servAddrLen = sizeof(servAddrPtr->sun_family) + bindPathLen; |
218 | #endif |
219 | return 0; |
220 | } |
221 | \f |
222 | union SockAddrUnion { |
223 | struct sockaddr_un unixVariant; |
224 | struct sockaddr_in inetVariant; |
225 | }; |
226 | |
227 | \f |
228 | /* |
229 | * OS_CreateLocalIpcFd -- |
230 | * |
231 | * This procedure is responsible for creating the listener socket |
232 | * on Unix for local process communication. It will create a |
233 | * domain socket or a TCP/IP socket bound to "localhost" and return |
234 | * a file descriptor to it to the caller. |
235 | * |
236 | * Results: |
237 | * Listener socket created. This call returns either a valid |
238 | * file descriptor or -1 on error. |
239 | * |
240 | * Side effects: |
241 | * None. |
242 | * |
243 | *---------------------------------------------------------------------- |
244 | */ |
0b7c9662 |
245 | int OS_CreateLocalIpcFd(const char *bindPath, int backlog) |
0198fd3c |
246 | { |
247 | int listenSock, servLen; |
248 | union SockAddrUnion sa; |
249 | int tcp = FALSE; |
250 | char *tp; |
251 | short port; |
252 | char host[MAXPATHLEN]; |
253 | |
254 | strcpy(host, bindPath); |
255 | if((tp = strchr(host, ':')) != 0) { |
256 | *tp++ = 0; |
257 | if((port = atoi(tp)) == 0) { |
258 | *--tp = ':'; |
259 | } else { |
260 | tcp = TRUE; |
261 | } |
262 | } |
263 | if(tcp && (*host && strcmp(host, "localhost") != 0)) { |
264 | fprintf(stderr, "To start a service on a TCP port can not " |
265 | "specify a host name.\n" |
266 | "You should either use \"localhost:<port>\" or " |
267 | " just use \":<port>.\"\n"); |
268 | exit(1); |
269 | } |
270 | |
271 | if(tcp) { |
272 | listenSock = socket(AF_INET, SOCK_STREAM, 0); |
273 | if(listenSock >= 0) { |
274 | int flag = 1; |
275 | if(setsockopt(listenSock, SOL_SOCKET, SO_REUSEADDR, |
276 | (char *) &flag, sizeof(flag)) < 0) { |
277 | fprintf(stderr, "Can't set SO_REUSEADDR.\n"); |
278 | exit(1001); |
279 | } |
280 | } |
281 | } else { |
282 | listenSock = socket(AF_UNIX, SOCK_STREAM, 0); |
283 | } |
284 | if(listenSock < 0) { |
285 | return -1; |
286 | } |
287 | |
288 | /* |
289 | * Bind the listening socket. |
290 | */ |
291 | if(tcp) { |
292 | memset((char *) &sa.inetVariant, 0, sizeof(sa.inetVariant)); |
293 | sa.inetVariant.sin_family = AF_INET; |
294 | sa.inetVariant.sin_addr.s_addr = htonl(INADDR_ANY); |
295 | sa.inetVariant.sin_port = htons(port); |
296 | servLen = sizeof(sa.inetVariant); |
297 | } else { |
298 | unlink(bindPath); |
299 | if(OS_BuildSockAddrUn(bindPath, &sa.unixVariant, &servLen)) { |
300 | fprintf(stderr, "Listening socket's path name is too long.\n"); |
301 | exit(1000); |
302 | } |
303 | } |
304 | if(bind(listenSock, (struct sockaddr *) &sa.unixVariant, servLen) < 0 |
0b7c9662 |
305 | || listen(listenSock, backlog) < 0) { |
0198fd3c |
306 | perror("bind/listen"); |
307 | exit(errno); |
308 | } |
309 | |
310 | return listenSock; |
311 | } |
312 | |
313 | \f |
314 | /* |
315 | *---------------------------------------------------------------------- |
316 | * |
317 | * OS_FcgiConnect -- |
318 | * |
319 | * Create the socket and connect to the remote application if |
320 | * possible. |
321 | * |
322 | * This was lifted from the cgi-fcgi application and was abstracted |
323 | * out because Windows NT does not have a domain socket and must |
324 | * use a named pipe which has a different API altogether. |
325 | * |
326 | * Results: |
327 | * -1 if fail or a valid file descriptor if connection succeeds. |
328 | * |
329 | * Side effects: |
330 | * Remote connection established. |
331 | * |
332 | *---------------------------------------------------------------------- |
333 | */ |
334 | int OS_FcgiConnect(char *bindPath) |
335 | { |
336 | union SockAddrUnion sa; |
337 | int servLen, resultSock; |
338 | int connectStatus; |
339 | char *tp; |
340 | char host[MAXPATHLEN]; |
341 | short port; |
342 | int tcp = FALSE; |
343 | |
344 | strcpy(host, bindPath); |
345 | if((tp = strchr(host, ':')) != 0) { |
346 | *tp++ = 0; |
347 | if((port = atoi(tp)) == 0) { |
348 | *--tp = ':'; |
349 | } else { |
350 | tcp = TRUE; |
351 | } |
352 | } |
353 | if(tcp == TRUE) { |
354 | struct hostent *hp; |
355 | if((hp = gethostbyname((*host ? host : "localhost"))) == NULL) { |
356 | fprintf(stderr, "Unknown host: %s\n", bindPath); |
357 | exit(1000); |
358 | } |
359 | sa.inetVariant.sin_family = AF_INET; |
203fd55e |
360 | memcpy(&sa.inetVariant.sin_addr, hp->h_addr, hp->h_length); |
0198fd3c |
361 | sa.inetVariant.sin_port = htons(port); |
362 | servLen = sizeof(sa.inetVariant); |
363 | resultSock = socket(AF_INET, SOCK_STREAM, 0); |
364 | } else { |
365 | if(OS_BuildSockAddrUn(bindPath, &sa.unixVariant, &servLen)) { |
366 | fprintf(stderr, "Listening socket's path name is too long.\n"); |
367 | exit(1000); |
368 | } |
369 | resultSock = socket(AF_UNIX, SOCK_STREAM, 0); |
370 | } |
371 | |
372 | assert(resultSock >= 0); |
373 | connectStatus = connect(resultSock, (struct sockaddr *) &sa.unixVariant, |
374 | servLen); |
375 | if(connectStatus >= 0) { |
376 | return resultSock; |
377 | } else { |
378 | /* |
379 | * Most likely (errno == ENOENT || errno == ECONNREFUSED) |
380 | * and no FCGI application server is running. |
381 | */ |
382 | close(resultSock); |
383 | return -1; |
384 | } |
385 | } |
aadcc3c8 |
386 | |
0198fd3c |
387 | \f |
388 | /* |
389 | *-------------------------------------------------------------- |
390 | * |
391 | * OS_Read -- |
392 | * |
393 | * Pass through to the unix read function. |
394 | * |
395 | * Results: |
396 | * Returns number of byes read, 0, or -1 failure: errno |
397 | * contains actual error. |
398 | * |
399 | * Side effects: |
400 | * None. |
401 | * |
402 | *-------------------------------------------------------------- |
403 | */ |
404 | int OS_Read(int fd, char * buf, size_t len) |
405 | { |
406 | return(read(fd, buf, len)); |
407 | } |
408 | \f |
409 | /* |
410 | *-------------------------------------------------------------- |
411 | * |
412 | * OS_Write -- |
413 | * |
414 | * Pass through to unix write function. |
415 | * |
416 | * Results: |
417 | * Returns number of byes read, 0, or -1 failure: errno |
418 | * contains actual error. |
419 | * |
420 | * Side effects: |
421 | * none. |
422 | * |
423 | *-------------------------------------------------------------- |
424 | */ |
425 | int OS_Write(int fd, char * buf, size_t len) |
426 | { |
427 | return(write(fd, buf, len)); |
428 | } |
429 | |
430 | \f |
431 | /* |
432 | *---------------------------------------------------------------------- |
433 | * |
434 | * OS_SpawnChild -- |
435 | * |
436 | * Spawns a new FastCGI listener process. |
437 | * |
438 | * Results: |
439 | * 0 if success, -1 if error. |
440 | * |
441 | * Side effects: |
442 | * Child process spawned. |
443 | * |
444 | *---------------------------------------------------------------------- |
445 | */ |
446 | int OS_SpawnChild(char *appPath, int listenFd) |
447 | { |
448 | int forkResult; |
449 | |
450 | forkResult = fork(); |
451 | if(forkResult < 0) { |
452 | exit(errno); |
453 | } |
454 | |
455 | if(forkResult == 0) { |
456 | /* |
457 | * Close STDIN unconditionally. It's used by the parent |
458 | * process for CGI communication. The FastCGI applciation |
459 | * will be replacing this with the FastCGI listenFd IF |
460 | * STDIN_FILENO is the same as FCGI_LISTENSOCK_FILENO |
461 | * (which it is on Unix). Regardless, STDIN, STDOUT, and |
462 | * STDERR will be closed as the FastCGI process uses a |
463 | * multiplexed socket in their place. |
464 | */ |
465 | close(STDIN_FILENO); |
466 | |
467 | /* |
468 | * If the listenFd is already the value of FCGI_LISTENSOCK_FILENO |
469 | * we're set. If not, change it so the child knows where to |
470 | * get the listen socket from. |
471 | */ |
472 | if(listenFd != FCGI_LISTENSOCK_FILENO) { |
473 | dup2(listenFd, FCGI_LISTENSOCK_FILENO); |
474 | close(listenFd); |
475 | } |
476 | |
477 | close(STDOUT_FILENO); |
478 | close(STDERR_FILENO); |
479 | |
480 | /* |
481 | * We're a child. Exec the application. |
482 | * |
483 | * XXX: entire environment passes through |
484 | */ |
485 | execl(appPath, appPath, NULL); |
486 | /* |
487 | * XXX: Can't do this as we've already closed STDERR!!! |
488 | * |
489 | * perror("exec"); |
490 | */ |
491 | exit(errno); |
492 | } |
493 | return 0; |
494 | } |
495 | |
496 | \f |
497 | /* |
498 | *-------------------------------------------------------------- |
499 | * |
500 | * OS_AsyncReadStdin -- |
501 | * |
502 | * This initiates an asynchronous read on the standard |
503 | * input handle. |
504 | * |
505 | * The abstraction is necessary because Windows NT does not |
506 | * have a clean way of "select"ing a file descriptor for |
507 | * I/O. |
508 | * |
509 | * Results: |
510 | * -1 if error, 0 otherwise. |
511 | * |
512 | * Side effects: |
513 | * Asynchronous bit is set in the readfd variable and |
514 | * request is enqueued. |
515 | * |
516 | *-------------------------------------------------------------- |
517 | */ |
aadcc3c8 |
518 | int OS_AsyncReadStdin(void *buf, int len, OS_AsyncProc procPtr, |
0198fd3c |
519 | ClientData clientData) |
520 | { |
521 | int index = AIO_RD_IX(STDIN_FILENO); |
522 | |
523 | ASSERT(asyncIoTable[index].inUse == 0); |
524 | asyncIoTable[index].procPtr = procPtr; |
525 | asyncIoTable[index].clientData = clientData; |
526 | asyncIoTable[index].fd = STDIN_FILENO; |
527 | asyncIoTable[index].len = len; |
528 | asyncIoTable[index].offset = 0; |
529 | asyncIoTable[index].buf = buf; |
530 | asyncIoTable[index].inUse = 1; |
531 | FD_SET(STDIN_FILENO, &readFdSet); |
532 | if(STDIN_FILENO > maxFd) |
533 | maxFd = STDIN_FILENO; |
534 | return 0; |
535 | } |
536 | |
537 | static void GrowAsyncTable(void) |
538 | { |
539 | int oldTableSize = asyncIoTableSize; |
aadcc3c8 |
540 | |
0198fd3c |
541 | asyncIoTableSize = asyncIoTableSize * 2; |
aadcc3c8 |
542 | asyncIoTable = (AioInfo *)realloc(asyncIoTable, asyncIoTableSize * sizeof(AioInfo)); |
0198fd3c |
543 | if(asyncIoTable == NULL) { |
544 | errno = ENOMEM; |
545 | exit(errno); |
546 | } |
547 | memset((char *) &asyncIoTable[oldTableSize], 0, |
548 | oldTableSize * sizeof(AioInfo)); |
549 | |
550 | } |
551 | |
552 | \f |
553 | /* |
554 | *-------------------------------------------------------------- |
555 | * |
556 | * OS_AsyncRead -- |
557 | * |
558 | * This initiates an asynchronous read on the file |
559 | * handle which may be a socket or named pipe. |
560 | * |
561 | * We also must save the ProcPtr and ClientData, so later |
562 | * when the io completes, we know who to call. |
563 | * |
564 | * We don't look at any results here (the ReadFile may |
565 | * return data if it is cached) but do all completion |
566 | * processing in OS_Select when we get the io completion |
567 | * port done notifications. Then we call the callback. |
568 | * |
569 | * Results: |
570 | * -1 if error, 0 otherwise. |
571 | * |
572 | * Side effects: |
573 | * Asynchronous I/O operation is queued for completion. |
574 | * |
575 | *-------------------------------------------------------------- |
576 | */ |
577 | int OS_AsyncRead(int fd, int offset, void *buf, int len, |
578 | OS_AsyncProc procPtr, ClientData clientData) |
579 | { |
580 | int index = AIO_RD_IX(fd); |
aadcc3c8 |
581 | |
0198fd3c |
582 | ASSERT(asyncIoTable != NULL); |
583 | |
584 | if(fd > maxFd) |
585 | maxFd = fd; |
586 | |
587 | if(index >= asyncIoTableSize) { |
588 | GrowAsyncTable(); |
589 | } |
590 | |
591 | ASSERT(asyncIoTable[index].inUse == 0); |
592 | asyncIoTable[index].procPtr = procPtr; |
593 | asyncIoTable[index].clientData = clientData; |
594 | asyncIoTable[index].fd = fd; |
595 | asyncIoTable[index].len = len; |
596 | asyncIoTable[index].offset = offset; |
597 | asyncIoTable[index].buf = buf; |
598 | asyncIoTable[index].inUse = 1; |
599 | FD_SET(fd, &readFdSet); |
600 | return 0; |
601 | } |
602 | \f |
603 | /* |
604 | *-------------------------------------------------------------- |
605 | * |
606 | * OS_AsyncWrite -- |
607 | * |
608 | * This initiates an asynchronous write on the "fake" file |
609 | * descriptor (which may be a file, socket, or named pipe). |
610 | * We also must save the ProcPtr and ClientData, so later |
611 | * when the io completes, we know who to call. |
612 | * |
613 | * We don't look at any results here (the WriteFile generally |
614 | * completes immediately) but do all completion processing |
615 | * in OS_DoIo when we get the io completion port done |
616 | * notifications. Then we call the callback. |
617 | * |
618 | * Results: |
619 | * -1 if error, 0 otherwise. |
620 | * |
621 | * Side effects: |
622 | * Asynchronous I/O operation is queued for completion. |
623 | * |
624 | *-------------------------------------------------------------- |
625 | */ |
aadcc3c8 |
626 | int OS_AsyncWrite(int fd, int offset, void *buf, int len, |
0198fd3c |
627 | OS_AsyncProc procPtr, ClientData clientData) |
628 | { |
629 | int index = AIO_WR_IX(fd); |
630 | |
631 | if(fd > maxFd) |
632 | maxFd = fd; |
633 | |
634 | if(index >= asyncIoTableSize) { |
635 | GrowAsyncTable(); |
636 | } |
637 | |
638 | ASSERT(asyncIoTable[index].inUse == 0); |
639 | asyncIoTable[index].procPtr = procPtr; |
640 | asyncIoTable[index].clientData = clientData; |
641 | asyncIoTable[index].fd = fd; |
642 | asyncIoTable[index].len = len; |
643 | asyncIoTable[index].offset = offset; |
644 | asyncIoTable[index].buf = buf; |
645 | asyncIoTable[index].inUse = 1; |
646 | FD_SET(fd, &writeFdSet); |
647 | return 0; |
648 | } |
649 | \f |
650 | /* |
651 | *-------------------------------------------------------------- |
652 | * |
653 | * OS_Close -- |
654 | * |
655 | * Closes the descriptor. This is a pass through to the |
656 | * Unix close. |
657 | * |
658 | * Results: |
659 | * 0 for success, -1 on failure |
660 | * |
661 | * Side effects: |
662 | * None. |
663 | * |
664 | *-------------------------------------------------------------- |
665 | */ |
666 | int OS_Close(int fd) |
667 | { |
668 | int index = AIO_RD_IX(fd); |
aadcc3c8 |
669 | |
0198fd3c |
670 | FD_CLR(fd, &readFdSet); |
671 | FD_CLR(fd, &readFdSetPost); |
672 | if(asyncIoTable[index].inUse != 0) { |
673 | asyncIoTable[index].inUse = 0; |
674 | } |
aadcc3c8 |
675 | |
0198fd3c |
676 | FD_CLR(fd, &writeFdSet); |
677 | FD_CLR(fd, &writeFdSetPost); |
678 | index = AIO_WR_IX(fd); |
679 | if(asyncIoTable[index].inUse != 0) { |
680 | asyncIoTable[index].inUse = 0; |
681 | } |
682 | if(maxFd == fd) |
683 | maxFd--; |
684 | return close(fd); |
685 | } |
686 | \f |
687 | /* |
688 | *-------------------------------------------------------------- |
689 | * |
690 | * OS_CloseRead -- |
691 | * |
692 | * Cancel outstanding asynchronous reads and prevent subsequent |
693 | * reads from completing. |
694 | * |
695 | * Results: |
696 | * Socket or file is shutdown. Return values mimic Unix shutdown: |
697 | * 0 success, -1 failure |
698 | * |
699 | *-------------------------------------------------------------- |
700 | */ |
701 | int OS_CloseRead(int fd) |
702 | { |
703 | if(asyncIoTable[AIO_RD_IX(fd)].inUse != 0) { |
704 | asyncIoTable[AIO_RD_IX(fd)].inUse = 0; |
705 | FD_CLR(fd, &readFdSet); |
706 | } |
aadcc3c8 |
707 | |
0198fd3c |
708 | return shutdown(fd, 0); |
709 | } |
710 | |
711 | \f |
712 | /* |
713 | *-------------------------------------------------------------- |
714 | * |
715 | * OS_DoIo -- |
716 | * |
717 | * This function was formerly OS_Select. It's purpose is |
718 | * to pull I/O completion events off the queue and dispatch |
719 | * them to the appropriate place. |
720 | * |
721 | * Results: |
722 | * Returns 0. |
723 | * |
724 | * Side effects: |
725 | * Handlers are called. |
726 | * |
727 | *-------------------------------------------------------------- |
728 | */ |
729 | int OS_DoIo(struct timeval *tmo) |
730 | { |
731 | int fd, len, selectStatus; |
732 | OS_AsyncProc procPtr; |
733 | ClientData clientData; |
734 | AioInfo *aioPtr; |
735 | fd_set readFdSetCpy; |
736 | fd_set writeFdSetCpy; |
737 | |
738 | FD_ZERO(&readFdSetCpy); |
739 | FD_ZERO(&writeFdSetCpy); |
740 | |
741 | for(fd = 0; fd <= maxFd; fd++) { |
742 | if(FD_ISSET(fd, &readFdSet)) { |
743 | FD_SET(fd, &readFdSetCpy); |
744 | } |
745 | if(FD_ISSET(fd, &writeFdSet)) { |
746 | FD_SET(fd, &writeFdSetCpy); |
747 | } |
748 | } |
aadcc3c8 |
749 | |
0198fd3c |
750 | /* |
751 | * If there were no completed events from a prior call, see if there's |
752 | * any work to do. |
753 | */ |
754 | if(numRdPosted == 0 && numWrPosted == 0) { |
755 | selectStatus = select((maxFd+1), &readFdSetCpy, &writeFdSetCpy, |
756 | NULL, tmo); |
757 | if(selectStatus < 0) { |
758 | exit(errno); |
759 | } |
760 | |
761 | for(fd = 0; fd <= maxFd; fd++) { |
762 | /* |
763 | * Build up a list of completed events. We'll work off of |
764 | * this list as opposed to looping through the read and write |
765 | * fd sets since they can be affected by a callbacl routine. |
766 | */ |
767 | if(FD_ISSET(fd, &readFdSetCpy)) { |
768 | numRdPosted++; |
769 | FD_SET(fd, &readFdSetPost); |
770 | FD_CLR(fd, &readFdSet); |
771 | } |
772 | |
773 | if(FD_ISSET(fd, &writeFdSetCpy)) { |
774 | numWrPosted++; |
775 | FD_SET(fd, &writeFdSetPost); |
776 | FD_CLR(fd, &writeFdSet); |
777 | } |
778 | } |
779 | } |
780 | |
781 | if(numRdPosted == 0 && numWrPosted == 0) |
782 | return 0; |
aadcc3c8 |
783 | |
0198fd3c |
784 | for(fd = 0; fd <= maxFd; fd++) { |
785 | /* |
786 | * Do reads and dispatch callback. |
787 | */ |
aadcc3c8 |
788 | if(FD_ISSET(fd, &readFdSetPost) |
0198fd3c |
789 | && asyncIoTable[AIO_RD_IX(fd)].inUse) { |
790 | |
791 | numRdPosted--; |
792 | FD_CLR(fd, &readFdSetPost); |
793 | aioPtr = &asyncIoTable[AIO_RD_IX(fd)]; |
aadcc3c8 |
794 | |
0198fd3c |
795 | len = read(aioPtr->fd, aioPtr->buf, aioPtr->len); |
796 | |
797 | procPtr = aioPtr->procPtr; |
798 | aioPtr->procPtr = NULL; |
799 | clientData = aioPtr->clientData; |
800 | aioPtr->inUse = 0; |
801 | |
802 | (*procPtr)(clientData, len); |
803 | } |
804 | |
805 | /* |
806 | * Do writes and dispatch callback. |
807 | */ |
808 | if(FD_ISSET(fd, &writeFdSetPost) && |
809 | asyncIoTable[AIO_WR_IX(fd)].inUse) { |
810 | |
811 | numWrPosted--; |
812 | FD_CLR(fd, &writeFdSetPost); |
813 | aioPtr = &asyncIoTable[AIO_WR_IX(fd)]; |
aadcc3c8 |
814 | |
0198fd3c |
815 | len = write(aioPtr->fd, aioPtr->buf, aioPtr->len); |
816 | |
817 | procPtr = aioPtr->procPtr; |
818 | aioPtr->procPtr = NULL; |
819 | clientData = aioPtr->clientData; |
820 | aioPtr->inUse = 0; |
821 | (*procPtr)(clientData, len); |
822 | } |
823 | } |
824 | return 0; |
825 | } |
826 | |
827 | \f |
828 | /* |
829 | *---------------------------------------------------------------------- |
830 | * |
831 | * ClientAddrOK -- |
832 | * |
833 | * Checks if a client address is in a list of allowed addresses |
834 | * |
835 | * Results: |
836 | * TRUE if address list is empty or client address is present |
837 | * in the list, FALSE otherwise. |
838 | * |
839 | *---------------------------------------------------------------------- |
840 | */ |
0b7c9662 |
841 | static int ClientAddrOK(struct sockaddr_in *saPtr, const char *clientList) |
0198fd3c |
842 | { |
843 | int result = FALSE; |
844 | char *clientListCopy, *cur, *next; |
845 | char *newString = NULL; |
846 | int strLen; |
847 | |
848 | if(clientList == NULL || *clientList == '\0') { |
849 | return TRUE; |
850 | } |
851 | |
852 | strLen = strlen(clientList); |
aadcc3c8 |
853 | clientListCopy = (char *)malloc(strLen + 1); |
0198fd3c |
854 | assert(newString != NULL); |
855 | memcpy(newString, clientList, strLen); |
856 | newString[strLen] = '\000'; |
aadcc3c8 |
857 | |
0198fd3c |
858 | for(cur = clientListCopy; cur != NULL; cur = next) { |
859 | next = strchr(cur, ','); |
860 | if(next != NULL) { |
861 | *next++ = '\0'; |
862 | } |
863 | if(inet_addr(cur) == saPtr->sin_addr.s_addr) { |
864 | result = TRUE; |
865 | break; |
866 | } |
867 | } |
868 | free(clientListCopy); |
869 | return result; |
870 | } |
871 | |
872 | \f |
873 | /* |
874 | *---------------------------------------------------------------------- |
875 | * |
876 | * AcquireLock -- |
877 | * |
878 | * On platforms that implement concurrent calls to accept |
879 | * on a shared listening ipcFd, returns 0. On other platforms, |
880 | * acquires an exclusive lock across all processes sharing a |
881 | * listening ipcFd, blocking until the lock has been acquired. |
882 | * |
883 | * Results: |
884 | * 0 for successful call, -1 in case of system error (fatal). |
885 | * |
886 | * Side effects: |
887 | * This process now has the exclusive lock. |
888 | * |
889 | *---------------------------------------------------------------------- |
890 | */ |
0b7c9662 |
891 | static int AcquireLock(int sock, int fail_on_intr) |
0198fd3c |
892 | { |
893 | #ifdef USE_LOCKING |
0b7c9662 |
894 | do { |
895 | struct flock lock; |
896 | lock.l_type = F_WRLCK; |
897 | lock.l_start = 0; |
898 | lock.l_whence = SEEK_SET; |
899 | lock.l_len = 0; |
900 | |
901 | if (fcntl(sock, F_SETLKW, &lock) != -1) |
902 | return 0; |
903 | } while (errno == EINTR && !fail_on_intr); |
904 | |
905 | return -1; |
906 | |
907 | #else |
0198fd3c |
908 | return 0; |
0b7c9662 |
909 | #endif |
0198fd3c |
910 | } |
911 | \f |
912 | /* |
913 | *---------------------------------------------------------------------- |
914 | * |
915 | * ReleaseLock -- |
916 | * |
917 | * On platforms that implement concurrent calls to accept |
918 | * on a shared listening ipcFd, does nothing. On other platforms, |
919 | * releases an exclusive lock acquired by AcquireLock. |
920 | * |
921 | * Results: |
922 | * 0 for successful call, -1 in case of system error (fatal). |
923 | * |
924 | * Side effects: |
925 | * This process no longer holds the lock. |
926 | * |
927 | *---------------------------------------------------------------------- |
928 | */ |
0b7c9662 |
929 | static int ReleaseLock(int sock) |
0198fd3c |
930 | { |
931 | #ifdef USE_LOCKING |
0b7c9662 |
932 | do { |
933 | struct flock lock; |
934 | lock.l_type = F_UNLCK; |
935 | lock.l_start = 0; |
936 | lock.l_whence = SEEK_SET; |
937 | lock.l_len = 0; |
938 | |
939 | if (fcntl(sock, F_SETLK, &lock) != -1) |
940 | return 0; |
941 | } while (errno == EINTR); |
0198fd3c |
942 | |
0b7c9662 |
943 | return -1; |
944 | |
945 | #else |
0198fd3c |
946 | return 0; |
0b7c9662 |
947 | #endif |
0198fd3c |
948 | } |
949 | |
950 | \f |
69f62c0e |
951 | /********************************************************************** |
aadcc3c8 |
952 | * Determine if the errno resulting from a failed accept() warrants a |
69f62c0e |
953 | * retry or exit(). Based on Apache's http_main.c accept() handling |
954 | * and Stevens' Unix Network Programming Vol 1, 2nd Ed, para. 15.6. |
955 | */ |
956 | static int is_reasonable_accept_errno (const int error) |
957 | { |
958 | switch (error) { |
959 | #ifdef EPROTO |
aadcc3c8 |
960 | /* EPROTO on certain older kernels really means ECONNABORTED, so |
961 | * we need to ignore it for them. See discussion in new-httpd |
962 | * archives nh.9701 search for EPROTO. Also see nh.9603, search |
963 | * for EPROTO: There is potentially a bug in Solaris 2.x x<6, and |
964 | * other boxes that implement tcp sockets in userland (i.e. on top of |
965 | * STREAMS). On these systems, EPROTO can actually result in a fatal |
966 | * loop. See PR#981 for example. It's hard to handle both uses of |
69f62c0e |
967 | * EPROTO. */ |
968 | case EPROTO: |
969 | #endif |
970 | #ifdef ECONNABORTED |
971 | case ECONNABORTED: |
972 | #endif |
973 | /* Linux generates the rest of these, other tcp stacks (i.e. |
974 | * bsd) tend to hide them behind getsockopt() interfaces. They |
975 | * occur when the net goes sour or the client disconnects after the |
976 | * three-way handshake has been done in the kernel but before |
977 | * userland has picked up the socket. */ |
978 | #ifdef ECONNRESET |
979 | case ECONNRESET: |
980 | #endif |
981 | #ifdef ETIMEDOUT |
982 | case ETIMEDOUT: |
983 | #endif |
984 | #ifdef EHOSTUNREACH |
985 | case EHOSTUNREACH: |
986 | #endif |
987 | #ifdef ENETUNREACH |
988 | case ENETUNREACH: |
989 | #endif |
990 | return 1; |
991 | |
992 | default: |
993 | return 0; |
994 | } |
995 | } |
996 | |
997 | /********************************************************************** |
aadcc3c8 |
998 | * This works around a problem on Linux 2.0.x and SCO Unixware (maybe |
999 | * others?). When a connect() is made to a Unix Domain socket, but its |
1000 | * not accept()ed before the web server gets impatient and close()s, an |
1001 | * accept() results in a valid file descriptor, but no data to read. |
1002 | * This causes a block on the first read() - which never returns! |
1003 | * |
1004 | * Another approach to this is to write() to the socket to provoke a |
1005 | * SIGPIPE, but this is a pain because of the FastCGI protocol, the fact |
1006 | * that whatever is written has to be universally ignored by all FastCGI |
1007 | * web servers, and a SIGPIPE handler has to be installed which returns |
1008 | * (or SIGPIPE is ignored). |
1009 | * |
1010 | * READABLE_UNIX_FD_DROP_DEAD_TIMEVAL = 2,0 by default. |
1011 | * |
1012 | * Making it shorter is probably safe, but I'll leave that to you. Making |
1013 | * it 0,0 doesn't work reliably. The shorter you can reliably make it, |
1014 | * the faster your application will be able to recover (waiting 2 seconds |
1015 | * may _cause_ the problem when there is a very high demand). At any rate, |
1016 | * this is better than perma-blocking. |
1017 | */ |
69f62c0e |
1018 | static int is_af_unix_keeper(const int fd) |
1019 | { |
1020 | struct timeval tval = { READABLE_UNIX_FD_DROP_DEAD_TIMEVAL }; |
1021 | fd_set read_fds; |
1022 | |
1023 | FD_ZERO(&read_fds); |
1024 | FD_SET(fd, &read_fds); |
aadcc3c8 |
1025 | |
69f62c0e |
1026 | return select(fd + 1, &read_fds, NULL, NULL, &tval) >= 0 && FD_ISSET(fd, &read_fds); |
1027 | } |
1028 | |
0198fd3c |
1029 | /* |
1030 | *---------------------------------------------------------------------- |
1031 | * |
0b7c9662 |
1032 | * OS_Accept -- |
0198fd3c |
1033 | * |
1034 | * Accepts a new FastCGI connection. This routine knows whether |
1035 | * we're dealing with TCP based sockets or NT Named Pipes for IPC. |
1036 | * |
1037 | * Results: |
1038 | * -1 if the operation fails, otherwise this is a valid IPC fd. |
1039 | * |
1040 | * Side effects: |
1041 | * New IPC connection is accepted. |
1042 | * |
1043 | *---------------------------------------------------------------------- |
1044 | */ |
1dd5d7a8 |
1045 | int OS_Accept(int listen_sock, int fail_on_intr, const char *webServerAddrs) |
0198fd3c |
1046 | { |
1047 | int socket; |
69f62c0e |
1048 | union { |
0198fd3c |
1049 | struct sockaddr_un un; |
a3c43615 |
1050 | struct sockaddr_in in; |
0198fd3c |
1051 | } sa; |
ed728efb |
1052 | |
0b7c9662 |
1053 | for (;;) { |
1054 | if (AcquireLock(listen_sock, fail_on_intr)) |
1055 | return -1; |
a3c43615 |
1056 | |
0b7c9662 |
1057 | for (;;) { |
69f62c0e |
1058 | do { |
0b7c9662 |
1059 | #ifdef HAVE_SOCKLEN |
1060 | socklen_t len = sizeof(sa); |
1061 | #else |
1062 | int len = sizeof(sa); |
1063 | #endif |
1064 | socket = accept(listen_sock, (struct sockaddr *)&sa, &len); |
1065 | } while (socket < 0 && errno == EINTR && !fail_on_intr); |
a3c43615 |
1066 | |
69f62c0e |
1067 | if (socket < 0) { |
1068 | if (!is_reasonable_accept_errno(errno)) { |
a3c43615 |
1069 | int errnoSave = errno; |
0b7c9662 |
1070 | ReleaseLock(listen_sock); |
a3c43615 |
1071 | errno = errnoSave; |
69f62c0e |
1072 | return (-1); |
a3c43615 |
1073 | } |
69f62c0e |
1074 | errno = 0; |
1075 | } |
0b7c9662 |
1076 | else { /* socket >= 0 */ |
69f62c0e |
1077 | int set = 1; |
aadcc3c8 |
1078 | |
69f62c0e |
1079 | if (sa.in.sin_family != AF_INET) |
1080 | break; |
aadcc3c8 |
1081 | |
69f62c0e |
1082 | #ifdef TCP_NODELAY |
1083 | /* No replies to outgoing data, so disable Nagle */ |
1084 | setsockopt(socket, IPPROTO_TCP, TCP_NODELAY, (char *)&set, sizeof(set)); |
aadcc3c8 |
1085 | #endif |
1086 | |
69f62c0e |
1087 | /* Check that the client IP address is approved */ |
1dd5d7a8 |
1088 | if (ClientAddrOK(&sa.in, webServerAddrs)) |
69f62c0e |
1089 | break; |
aadcc3c8 |
1090 | |
69f62c0e |
1091 | close(socket); |
0b7c9662 |
1092 | } /* socket >= 0 */ |
1093 | } /* for(;;) */ |
aadcc3c8 |
1094 | |
0b7c9662 |
1095 | if (ReleaseLock(listen_sock)) |
69f62c0e |
1096 | return (-1); |
aadcc3c8 |
1097 | |
69f62c0e |
1098 | if (sa.in.sin_family != AF_UNIX || is_af_unix_keeper(socket)) |
1099 | break; |
aadcc3c8 |
1100 | |
69f62c0e |
1101 | close(socket); |
1102 | } /* while(1) - lock */ |
a3c43615 |
1103 | |
a3c43615 |
1104 | return (socket); |
0198fd3c |
1105 | } |
1106 | \f |
1107 | /* |
1108 | *---------------------------------------------------------------------- |
1109 | * |
1110 | * OS_IpcClose |
1111 | * |
1112 | * OS IPC routine to close an IPC connection. |
1113 | * |
1114 | * Results: |
1115 | * |
1116 | * |
1117 | * Side effects: |
1118 | * IPC connection is closed. |
1119 | * |
1120 | *---------------------------------------------------------------------- |
1121 | */ |
1122 | int OS_IpcClose(int ipcFd) |
1123 | { |
1124 | return OS_Close(ipcFd); |
1125 | } |
1126 | |
1127 | \f |
1128 | /* |
1129 | *---------------------------------------------------------------------- |
1130 | * |
1131 | * OS_IsFcgi -- |
1132 | * |
1133 | * Determines whether this process is a FastCGI process or not. |
1134 | * |
1135 | * Results: |
1136 | * Returns 1 if FastCGI, 0 if not. |
1137 | * |
1138 | * Side effects: |
1139 | * None. |
1140 | * |
1141 | *---------------------------------------------------------------------- |
1142 | */ |
0b7c9662 |
1143 | int OS_IsFcgi(int sock) |
0198fd3c |
1144 | { |
b22c3782 |
1145 | union { |
1146 | struct sockaddr_in in; |
1147 | struct sockaddr_un un; |
1148 | } sa; |
8eac3e1b |
1149 | #ifdef HAVE_SOCKLEN |
ed728efb |
1150 | socklen_t len = sizeof(sa); |
1151 | #else |
b22c3782 |
1152 | int len = sizeof(sa); |
ed728efb |
1153 | #endif |
0198fd3c |
1154 | |
0b7c9662 |
1155 | if (getpeername(sock, (struct sockaddr *)&sa, &len) != 0 && errno == ENOTCONN) { |
1156 | return TRUE; |
1157 | } |
1158 | else { |
1159 | return FALSE; |
1160 | } |
0198fd3c |
1161 | } |
1162 | \f |
1163 | /* |
1164 | *---------------------------------------------------------------------- |
1165 | * |
1166 | * OS_SetFlags -- |
1167 | * |
1168 | * Sets selected flag bits in an open file descriptor. |
1169 | * |
1170 | *---------------------------------------------------------------------- |
1171 | */ |
1172 | void OS_SetFlags(int fd, int flags) |
1173 | { |
1174 | int val; |
1175 | if((val = fcntl(fd, F_GETFL, 0)) < 0) { |
1176 | exit(errno); |
1177 | } |
1178 | val |= flags; |
1179 | if(fcntl(fd, F_SETFL, val) < 0) { |
1180 | exit(errno); |
1181 | } |
1182 | } |