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11 | * remains the unpublished property of Open Market, Inc. Use, |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 |
27064097 |
20 | static const char rcsid[] = "$Id: os_unix.c,v 1.16 2000/02/03 08:52:56 skimo Exp $"; |
0198fd3c |
21 | #endif /* not lint */ |
22 | |
6ad90ad2 |
23 | #include "fcgi_config.h" |
0198fd3c |
24 | |
27064097 |
25 | #include <sys/types.h> |
26 | |
8db9dd8f |
27 | #ifdef HAVE_NETINET_IN_H |
28 | #include <netinet/in.h> |
29 | #endif |
30 | |
6ad90ad2 |
31 | #include <arpa/inet.h> |
0198fd3c |
32 | #include <assert.h> |
0198fd3c |
33 | #include <errno.h> |
6ad90ad2 |
34 | #include <fcntl.h> /* for fcntl */ |
0198fd3c |
35 | #include <math.h> |
6ad90ad2 |
36 | #include <memory.h> /* for memchr() */ |
37 | #include <netinet/tcp.h> |
38 | #include <stdarg.h> |
39 | #include <stdio.h> |
40 | #include <stdlib.h> |
41 | #include <string.h> |
42 | #include <sys/time.h> |
0198fd3c |
43 | #include <sys/un.h> |
6ad90ad2 |
44 | |
0198fd3c |
45 | #ifdef HAVE_NETDB_H |
46 | #include <netdb.h> |
47 | #endif |
0198fd3c |
48 | |
6ad90ad2 |
49 | #ifdef HAVE_SYS_SOCKET_H |
50 | #include <sys/socket.h> /* for getpeername */ |
51 | #endif |
52 | |
53 | #ifdef HAVE_UNISTD_H |
54 | #include <unistd.h> |
55 | #endif |
56 | |
57 | #include "fastcgi.h" |
58 | #include "fcgiapp.h" |
59 | #include "fcgiappmisc.h" |
60 | #include "fcgimisc.h" |
0198fd3c |
61 | #include "fcgios.h" |
62 | |
6ad90ad2 |
63 | #ifndef FALSE |
64 | #define FALSE 0 |
65 | #endif |
66 | |
67 | #ifndef TRUE |
68 | #define TRUE 1 |
0198fd3c |
69 | #endif |
70 | |
71 | /* |
72 | * This structure holds an entry for each oustanding async I/O operation. |
73 | */ |
74 | typedef struct { |
75 | OS_AsyncProc procPtr; /* callout completion procedure */ |
76 | ClientData clientData; /* caller private data */ |
77 | int fd; |
78 | int len; |
79 | int offset; |
80 | void *buf; |
81 | int inUse; |
82 | } AioInfo; |
83 | |
84 | /* |
85 | * Entries in the async I/O table are allocated 2 per file descriptor. |
86 | * |
87 | * Read Entry Index = fd * 2 |
88 | * Write Entry Index = (fd * 2) + 1 |
89 | */ |
90 | #define AIO_RD_IX(fd) (fd * 2) |
91 | #define AIO_WR_IX(fd) ((fd * 2) + 1) |
92 | |
93 | static int asyncIoTableSize = 16; |
94 | static AioInfo *asyncIoTable = NULL; |
0198fd3c |
95 | |
0198fd3c |
96 | static int libInitialized = FALSE; |
97 | |
98 | static fd_set readFdSet; |
99 | static fd_set writeFdSet; |
100 | |
101 | static fd_set readFdSetPost; |
102 | static int numRdPosted = 0; |
103 | static fd_set writeFdSetPost; |
104 | static int numWrPosted = 0; |
105 | static int volatile maxFd = -1; |
106 | |
0198fd3c |
107 | \f |
108 | /* |
109 | *-------------------------------------------------------------- |
110 | * |
111 | * OS_LibInit -- |
112 | * |
113 | * Set up the OS library for use. |
114 | * |
115 | * NOTE: This function is really only needed for application |
116 | * asynchronous I/O. It will most likely change in the |
117 | * future to setup the multi-threaded environment. |
118 | * |
119 | * Results: |
120 | * Returns 0 if success, -1 if not. |
121 | * |
122 | * Side effects: |
123 | * Async I/O table allocated and initialized. |
124 | * |
125 | *-------------------------------------------------------------- |
126 | */ |
127 | int OS_LibInit(int stdioFds[3]) |
128 | { |
129 | if(libInitialized) |
130 | return 0; |
aadcc3c8 |
131 | |
132 | asyncIoTable = (AioInfo *)malloc(asyncIoTableSize * sizeof(AioInfo)); |
0198fd3c |
133 | if(asyncIoTable == NULL) { |
134 | errno = ENOMEM; |
135 | return -1; |
136 | } |
137 | memset((char *) asyncIoTable, 0, |
138 | asyncIoTableSize * sizeof(AioInfo)); |
139 | |
140 | FD_ZERO(&readFdSet); |
141 | FD_ZERO(&writeFdSet); |
142 | FD_ZERO(&readFdSetPost); |
143 | FD_ZERO(&writeFdSetPost); |
144 | libInitialized = TRUE; |
145 | return 0; |
146 | } |
147 | |
148 | \f |
149 | /* |
150 | *-------------------------------------------------------------- |
151 | * |
152 | * OS_LibShutdown -- |
153 | * |
154 | * Shutdown the OS library. |
155 | * |
156 | * Results: |
157 | * None. |
158 | * |
159 | * Side effects: |
160 | * Memory freed, fds closed. |
161 | * |
162 | *-------------------------------------------------------------- |
163 | */ |
164 | void OS_LibShutdown() |
165 | { |
166 | if(!libInitialized) |
167 | return; |
aadcc3c8 |
168 | |
0198fd3c |
169 | free(asyncIoTable); |
170 | asyncIoTable = NULL; |
171 | libInitialized = FALSE; |
172 | return; |
173 | } |
174 | |
175 | \f |
176 | /* |
177 | *---------------------------------------------------------------------- |
178 | * |
179 | * OS_BuildSockAddrUn -- |
180 | * |
181 | * Using the pathname bindPath, fill in the sockaddr_un structure |
182 | * *servAddrPtr and the length of this structure *servAddrLen. |
183 | * |
184 | * The format of the sockaddr_un structure changed incompatibly in |
185 | * 4.3BSD Reno. Digital UNIX supports both formats, other systems |
186 | * support one or the other. |
187 | * |
188 | * Results: |
189 | * 0 for normal return, -1 for failure (bindPath too long). |
190 | * |
191 | *---------------------------------------------------------------------- |
192 | */ |
193 | |
0b7c9662 |
194 | static int OS_BuildSockAddrUn(const char *bindPath, |
0198fd3c |
195 | struct sockaddr_un *servAddrPtr, |
196 | int *servAddrLen) |
197 | { |
198 | int bindPathLen = strlen(bindPath); |
199 | |
200 | #ifdef HAVE_SOCKADDR_UN_SUN_LEN /* 4.3BSD Reno and later: BSDI, DEC */ |
201 | if(bindPathLen >= sizeof(servAddrPtr->sun_path)) { |
202 | return -1; |
203 | } |
204 | #else /* 4.3 BSD Tahoe: Solaris, HPUX, DEC, ... */ |
205 | if(bindPathLen > sizeof(servAddrPtr->sun_path)) { |
206 | return -1; |
207 | } |
208 | #endif |
209 | memset((char *) servAddrPtr, 0, sizeof(*servAddrPtr)); |
210 | servAddrPtr->sun_family = AF_UNIX; |
211 | memcpy(servAddrPtr->sun_path, bindPath, bindPathLen); |
212 | #ifdef HAVE_SOCKADDR_UN_SUN_LEN /* 4.3BSD Reno and later: BSDI, DEC */ |
213 | *servAddrLen = sizeof(servAddrPtr->sun_len) |
214 | + sizeof(servAddrPtr->sun_family) |
215 | + bindPathLen + 1; |
216 | servAddrPtr->sun_len = *servAddrLen; |
217 | #else /* 4.3 BSD Tahoe: Solaris, HPUX, DEC, ... */ |
218 | *servAddrLen = sizeof(servAddrPtr->sun_family) + bindPathLen; |
219 | #endif |
220 | return 0; |
221 | } |
222 | \f |
223 | union SockAddrUnion { |
224 | struct sockaddr_un unixVariant; |
225 | struct sockaddr_in inetVariant; |
226 | }; |
227 | |
228 | \f |
229 | /* |
230 | * OS_CreateLocalIpcFd -- |
231 | * |
232 | * This procedure is responsible for creating the listener socket |
233 | * on Unix for local process communication. It will create a |
234 | * domain socket or a TCP/IP socket bound to "localhost" and return |
235 | * a file descriptor to it to the caller. |
236 | * |
237 | * Results: |
238 | * Listener socket created. This call returns either a valid |
239 | * file descriptor or -1 on error. |
240 | * |
241 | * Side effects: |
242 | * None. |
243 | * |
244 | *---------------------------------------------------------------------- |
245 | */ |
0b7c9662 |
246 | int OS_CreateLocalIpcFd(const char *bindPath, int backlog) |
0198fd3c |
247 | { |
248 | int listenSock, servLen; |
249 | union SockAddrUnion sa; |
250 | int tcp = FALSE; |
251 | char *tp; |
252 | short port; |
253 | char host[MAXPATHLEN]; |
254 | |
255 | strcpy(host, bindPath); |
256 | if((tp = strchr(host, ':')) != 0) { |
257 | *tp++ = 0; |
258 | if((port = atoi(tp)) == 0) { |
259 | *--tp = ':'; |
260 | } else { |
261 | tcp = TRUE; |
262 | } |
263 | } |
264 | if(tcp && (*host && strcmp(host, "localhost") != 0)) { |
265 | fprintf(stderr, "To start a service on a TCP port can not " |
266 | "specify a host name.\n" |
267 | "You should either use \"localhost:<port>\" or " |
268 | " just use \":<port>.\"\n"); |
269 | exit(1); |
270 | } |
271 | |
272 | if(tcp) { |
273 | listenSock = socket(AF_INET, SOCK_STREAM, 0); |
274 | if(listenSock >= 0) { |
275 | int flag = 1; |
276 | if(setsockopt(listenSock, SOL_SOCKET, SO_REUSEADDR, |
277 | (char *) &flag, sizeof(flag)) < 0) { |
278 | fprintf(stderr, "Can't set SO_REUSEADDR.\n"); |
279 | exit(1001); |
280 | } |
281 | } |
282 | } else { |
283 | listenSock = socket(AF_UNIX, SOCK_STREAM, 0); |
284 | } |
285 | if(listenSock < 0) { |
286 | return -1; |
287 | } |
288 | |
289 | /* |
290 | * Bind the listening socket. |
291 | */ |
292 | if(tcp) { |
293 | memset((char *) &sa.inetVariant, 0, sizeof(sa.inetVariant)); |
294 | sa.inetVariant.sin_family = AF_INET; |
295 | sa.inetVariant.sin_addr.s_addr = htonl(INADDR_ANY); |
296 | sa.inetVariant.sin_port = htons(port); |
297 | servLen = sizeof(sa.inetVariant); |
298 | } else { |
299 | unlink(bindPath); |
300 | if(OS_BuildSockAddrUn(bindPath, &sa.unixVariant, &servLen)) { |
301 | fprintf(stderr, "Listening socket's path name is too long.\n"); |
302 | exit(1000); |
303 | } |
304 | } |
305 | if(bind(listenSock, (struct sockaddr *) &sa.unixVariant, servLen) < 0 |
0b7c9662 |
306 | || listen(listenSock, backlog) < 0) { |
0198fd3c |
307 | perror("bind/listen"); |
308 | exit(errno); |
309 | } |
310 | |
311 | return listenSock; |
312 | } |
313 | |
314 | \f |
315 | /* |
316 | *---------------------------------------------------------------------- |
317 | * |
318 | * OS_FcgiConnect -- |
319 | * |
320 | * Create the socket and connect to the remote application if |
321 | * possible. |
322 | * |
323 | * This was lifted from the cgi-fcgi application and was abstracted |
324 | * out because Windows NT does not have a domain socket and must |
325 | * use a named pipe which has a different API altogether. |
326 | * |
327 | * Results: |
328 | * -1 if fail or a valid file descriptor if connection succeeds. |
329 | * |
330 | * Side effects: |
331 | * Remote connection established. |
332 | * |
333 | *---------------------------------------------------------------------- |
334 | */ |
335 | int OS_FcgiConnect(char *bindPath) |
336 | { |
337 | union SockAddrUnion sa; |
338 | int servLen, resultSock; |
339 | int connectStatus; |
340 | char *tp; |
341 | char host[MAXPATHLEN]; |
342 | short port; |
343 | int tcp = FALSE; |
344 | |
345 | strcpy(host, bindPath); |
346 | if((tp = strchr(host, ':')) != 0) { |
347 | *tp++ = 0; |
348 | if((port = atoi(tp)) == 0) { |
349 | *--tp = ':'; |
350 | } else { |
351 | tcp = TRUE; |
352 | } |
353 | } |
354 | if(tcp == TRUE) { |
355 | struct hostent *hp; |
356 | if((hp = gethostbyname((*host ? host : "localhost"))) == NULL) { |
357 | fprintf(stderr, "Unknown host: %s\n", bindPath); |
358 | exit(1000); |
359 | } |
360 | sa.inetVariant.sin_family = AF_INET; |
203fd55e |
361 | memcpy(&sa.inetVariant.sin_addr, hp->h_addr, hp->h_length); |
0198fd3c |
362 | sa.inetVariant.sin_port = htons(port); |
363 | servLen = sizeof(sa.inetVariant); |
364 | resultSock = socket(AF_INET, SOCK_STREAM, 0); |
365 | } else { |
366 | if(OS_BuildSockAddrUn(bindPath, &sa.unixVariant, &servLen)) { |
367 | fprintf(stderr, "Listening socket's path name is too long.\n"); |
368 | exit(1000); |
369 | } |
370 | resultSock = socket(AF_UNIX, SOCK_STREAM, 0); |
371 | } |
372 | |
373 | assert(resultSock >= 0); |
374 | connectStatus = connect(resultSock, (struct sockaddr *) &sa.unixVariant, |
375 | servLen); |
376 | if(connectStatus >= 0) { |
377 | return resultSock; |
378 | } else { |
379 | /* |
380 | * Most likely (errno == ENOENT || errno == ECONNREFUSED) |
381 | * and no FCGI application server is running. |
382 | */ |
383 | close(resultSock); |
384 | return -1; |
385 | } |
386 | } |
aadcc3c8 |
387 | |
0198fd3c |
388 | \f |
389 | /* |
390 | *-------------------------------------------------------------- |
391 | * |
392 | * OS_Read -- |
393 | * |
394 | * Pass through to the unix read function. |
395 | * |
396 | * Results: |
397 | * Returns number of byes read, 0, or -1 failure: errno |
398 | * contains actual error. |
399 | * |
400 | * Side effects: |
401 | * None. |
402 | * |
403 | *-------------------------------------------------------------- |
404 | */ |
405 | int OS_Read(int fd, char * buf, size_t len) |
406 | { |
407 | return(read(fd, buf, len)); |
408 | } |
409 | \f |
410 | /* |
411 | *-------------------------------------------------------------- |
412 | * |
413 | * OS_Write -- |
414 | * |
415 | * Pass through to unix write function. |
416 | * |
417 | * Results: |
418 | * Returns number of byes read, 0, or -1 failure: errno |
419 | * contains actual error. |
420 | * |
421 | * Side effects: |
422 | * none. |
423 | * |
424 | *-------------------------------------------------------------- |
425 | */ |
426 | int OS_Write(int fd, char * buf, size_t len) |
427 | { |
428 | return(write(fd, buf, len)); |
429 | } |
430 | |
431 | \f |
432 | /* |
433 | *---------------------------------------------------------------------- |
434 | * |
435 | * OS_SpawnChild -- |
436 | * |
437 | * Spawns a new FastCGI listener process. |
438 | * |
439 | * Results: |
440 | * 0 if success, -1 if error. |
441 | * |
442 | * Side effects: |
443 | * Child process spawned. |
444 | * |
445 | *---------------------------------------------------------------------- |
446 | */ |
447 | int OS_SpawnChild(char *appPath, int listenFd) |
448 | { |
449 | int forkResult; |
450 | |
451 | forkResult = fork(); |
452 | if(forkResult < 0) { |
453 | exit(errno); |
454 | } |
455 | |
456 | if(forkResult == 0) { |
457 | /* |
458 | * Close STDIN unconditionally. It's used by the parent |
459 | * process for CGI communication. The FastCGI applciation |
460 | * will be replacing this with the FastCGI listenFd IF |
461 | * STDIN_FILENO is the same as FCGI_LISTENSOCK_FILENO |
462 | * (which it is on Unix). Regardless, STDIN, STDOUT, and |
463 | * STDERR will be closed as the FastCGI process uses a |
464 | * multiplexed socket in their place. |
465 | */ |
466 | close(STDIN_FILENO); |
467 | |
468 | /* |
469 | * If the listenFd is already the value of FCGI_LISTENSOCK_FILENO |
470 | * we're set. If not, change it so the child knows where to |
471 | * get the listen socket from. |
472 | */ |
473 | if(listenFd != FCGI_LISTENSOCK_FILENO) { |
474 | dup2(listenFd, FCGI_LISTENSOCK_FILENO); |
475 | close(listenFd); |
476 | } |
477 | |
478 | close(STDOUT_FILENO); |
479 | close(STDERR_FILENO); |
480 | |
481 | /* |
482 | * We're a child. Exec the application. |
483 | * |
484 | * XXX: entire environment passes through |
485 | */ |
486 | execl(appPath, appPath, NULL); |
487 | /* |
488 | * XXX: Can't do this as we've already closed STDERR!!! |
489 | * |
490 | * perror("exec"); |
491 | */ |
492 | exit(errno); |
493 | } |
494 | return 0; |
495 | } |
496 | |
497 | \f |
498 | /* |
499 | *-------------------------------------------------------------- |
500 | * |
501 | * OS_AsyncReadStdin -- |
502 | * |
503 | * This initiates an asynchronous read on the standard |
504 | * input handle. |
505 | * |
506 | * The abstraction is necessary because Windows NT does not |
507 | * have a clean way of "select"ing a file descriptor for |
508 | * I/O. |
509 | * |
510 | * Results: |
511 | * -1 if error, 0 otherwise. |
512 | * |
513 | * Side effects: |
514 | * Asynchronous bit is set in the readfd variable and |
515 | * request is enqueued. |
516 | * |
517 | *-------------------------------------------------------------- |
518 | */ |
aadcc3c8 |
519 | int OS_AsyncReadStdin(void *buf, int len, OS_AsyncProc procPtr, |
0198fd3c |
520 | ClientData clientData) |
521 | { |
522 | int index = AIO_RD_IX(STDIN_FILENO); |
523 | |
524 | ASSERT(asyncIoTable[index].inUse == 0); |
525 | asyncIoTable[index].procPtr = procPtr; |
526 | asyncIoTable[index].clientData = clientData; |
527 | asyncIoTable[index].fd = STDIN_FILENO; |
528 | asyncIoTable[index].len = len; |
529 | asyncIoTable[index].offset = 0; |
530 | asyncIoTable[index].buf = buf; |
531 | asyncIoTable[index].inUse = 1; |
532 | FD_SET(STDIN_FILENO, &readFdSet); |
533 | if(STDIN_FILENO > maxFd) |
534 | maxFd = STDIN_FILENO; |
535 | return 0; |
536 | } |
537 | |
538 | static void GrowAsyncTable(void) |
539 | { |
540 | int oldTableSize = asyncIoTableSize; |
aadcc3c8 |
541 | |
0198fd3c |
542 | asyncIoTableSize = asyncIoTableSize * 2; |
aadcc3c8 |
543 | asyncIoTable = (AioInfo *)realloc(asyncIoTable, asyncIoTableSize * sizeof(AioInfo)); |
0198fd3c |
544 | if(asyncIoTable == NULL) { |
545 | errno = ENOMEM; |
546 | exit(errno); |
547 | } |
548 | memset((char *) &asyncIoTable[oldTableSize], 0, |
549 | oldTableSize * sizeof(AioInfo)); |
550 | |
551 | } |
552 | |
553 | \f |
554 | /* |
555 | *-------------------------------------------------------------- |
556 | * |
557 | * OS_AsyncRead -- |
558 | * |
559 | * This initiates an asynchronous read on the file |
560 | * handle which may be a socket or named pipe. |
561 | * |
562 | * We also must save the ProcPtr and ClientData, so later |
563 | * when the io completes, we know who to call. |
564 | * |
565 | * We don't look at any results here (the ReadFile may |
566 | * return data if it is cached) but do all completion |
567 | * processing in OS_Select when we get the io completion |
568 | * port done notifications. Then we call the callback. |
569 | * |
570 | * Results: |
571 | * -1 if error, 0 otherwise. |
572 | * |
573 | * Side effects: |
574 | * Asynchronous I/O operation is queued for completion. |
575 | * |
576 | *-------------------------------------------------------------- |
577 | */ |
578 | int OS_AsyncRead(int fd, int offset, void *buf, int len, |
579 | OS_AsyncProc procPtr, ClientData clientData) |
580 | { |
581 | int index = AIO_RD_IX(fd); |
aadcc3c8 |
582 | |
0198fd3c |
583 | ASSERT(asyncIoTable != NULL); |
584 | |
585 | if(fd > maxFd) |
586 | maxFd = fd; |
587 | |
588 | if(index >= asyncIoTableSize) { |
589 | GrowAsyncTable(); |
590 | } |
591 | |
592 | ASSERT(asyncIoTable[index].inUse == 0); |
593 | asyncIoTable[index].procPtr = procPtr; |
594 | asyncIoTable[index].clientData = clientData; |
595 | asyncIoTable[index].fd = fd; |
596 | asyncIoTable[index].len = len; |
597 | asyncIoTable[index].offset = offset; |
598 | asyncIoTable[index].buf = buf; |
599 | asyncIoTable[index].inUse = 1; |
600 | FD_SET(fd, &readFdSet); |
601 | return 0; |
602 | } |
603 | \f |
604 | /* |
605 | *-------------------------------------------------------------- |
606 | * |
607 | * OS_AsyncWrite -- |
608 | * |
609 | * This initiates an asynchronous write on the "fake" file |
610 | * descriptor (which may be a file, socket, or named pipe). |
611 | * We also must save the ProcPtr and ClientData, so later |
612 | * when the io completes, we know who to call. |
613 | * |
614 | * We don't look at any results here (the WriteFile generally |
615 | * completes immediately) but do all completion processing |
616 | * in OS_DoIo when we get the io completion port done |
617 | * notifications. Then we call the callback. |
618 | * |
619 | * Results: |
620 | * -1 if error, 0 otherwise. |
621 | * |
622 | * Side effects: |
623 | * Asynchronous I/O operation is queued for completion. |
624 | * |
625 | *-------------------------------------------------------------- |
626 | */ |
aadcc3c8 |
627 | int OS_AsyncWrite(int fd, int offset, void *buf, int len, |
0198fd3c |
628 | OS_AsyncProc procPtr, ClientData clientData) |
629 | { |
630 | int index = AIO_WR_IX(fd); |
631 | |
632 | if(fd > maxFd) |
633 | maxFd = fd; |
634 | |
635 | if(index >= asyncIoTableSize) { |
636 | GrowAsyncTable(); |
637 | } |
638 | |
639 | ASSERT(asyncIoTable[index].inUse == 0); |
640 | asyncIoTable[index].procPtr = procPtr; |
641 | asyncIoTable[index].clientData = clientData; |
642 | asyncIoTable[index].fd = fd; |
643 | asyncIoTable[index].len = len; |
644 | asyncIoTable[index].offset = offset; |
645 | asyncIoTable[index].buf = buf; |
646 | asyncIoTable[index].inUse = 1; |
647 | FD_SET(fd, &writeFdSet); |
648 | return 0; |
649 | } |
650 | \f |
651 | /* |
652 | *-------------------------------------------------------------- |
653 | * |
654 | * OS_Close -- |
655 | * |
656 | * Closes the descriptor. This is a pass through to the |
657 | * Unix close. |
658 | * |
659 | * Results: |
660 | * 0 for success, -1 on failure |
661 | * |
662 | * Side effects: |
663 | * None. |
664 | * |
665 | *-------------------------------------------------------------- |
666 | */ |
667 | int OS_Close(int fd) |
668 | { |
669 | int index = AIO_RD_IX(fd); |
aadcc3c8 |
670 | |
0198fd3c |
671 | FD_CLR(fd, &readFdSet); |
672 | FD_CLR(fd, &readFdSetPost); |
673 | if(asyncIoTable[index].inUse != 0) { |
674 | asyncIoTable[index].inUse = 0; |
675 | } |
aadcc3c8 |
676 | |
0198fd3c |
677 | FD_CLR(fd, &writeFdSet); |
678 | FD_CLR(fd, &writeFdSetPost); |
679 | index = AIO_WR_IX(fd); |
680 | if(asyncIoTable[index].inUse != 0) { |
681 | asyncIoTable[index].inUse = 0; |
682 | } |
683 | if(maxFd == fd) |
684 | maxFd--; |
685 | return close(fd); |
686 | } |
687 | \f |
688 | /* |
689 | *-------------------------------------------------------------- |
690 | * |
691 | * OS_CloseRead -- |
692 | * |
693 | * Cancel outstanding asynchronous reads and prevent subsequent |
694 | * reads from completing. |
695 | * |
696 | * Results: |
697 | * Socket or file is shutdown. Return values mimic Unix shutdown: |
698 | * 0 success, -1 failure |
699 | * |
700 | *-------------------------------------------------------------- |
701 | */ |
702 | int OS_CloseRead(int fd) |
703 | { |
704 | if(asyncIoTable[AIO_RD_IX(fd)].inUse != 0) { |
705 | asyncIoTable[AIO_RD_IX(fd)].inUse = 0; |
706 | FD_CLR(fd, &readFdSet); |
707 | } |
aadcc3c8 |
708 | |
0198fd3c |
709 | return shutdown(fd, 0); |
710 | } |
711 | |
712 | \f |
713 | /* |
714 | *-------------------------------------------------------------- |
715 | * |
716 | * OS_DoIo -- |
717 | * |
718 | * This function was formerly OS_Select. It's purpose is |
719 | * to pull I/O completion events off the queue and dispatch |
720 | * them to the appropriate place. |
721 | * |
722 | * Results: |
723 | * Returns 0. |
724 | * |
725 | * Side effects: |
726 | * Handlers are called. |
727 | * |
728 | *-------------------------------------------------------------- |
729 | */ |
730 | int OS_DoIo(struct timeval *tmo) |
731 | { |
732 | int fd, len, selectStatus; |
733 | OS_AsyncProc procPtr; |
734 | ClientData clientData; |
735 | AioInfo *aioPtr; |
736 | fd_set readFdSetCpy; |
737 | fd_set writeFdSetCpy; |
738 | |
739 | FD_ZERO(&readFdSetCpy); |
740 | FD_ZERO(&writeFdSetCpy); |
741 | |
742 | for(fd = 0; fd <= maxFd; fd++) { |
743 | if(FD_ISSET(fd, &readFdSet)) { |
744 | FD_SET(fd, &readFdSetCpy); |
745 | } |
746 | if(FD_ISSET(fd, &writeFdSet)) { |
747 | FD_SET(fd, &writeFdSetCpy); |
748 | } |
749 | } |
aadcc3c8 |
750 | |
0198fd3c |
751 | /* |
752 | * If there were no completed events from a prior call, see if there's |
753 | * any work to do. |
754 | */ |
755 | if(numRdPosted == 0 && numWrPosted == 0) { |
756 | selectStatus = select((maxFd+1), &readFdSetCpy, &writeFdSetCpy, |
757 | NULL, tmo); |
758 | if(selectStatus < 0) { |
759 | exit(errno); |
760 | } |
761 | |
762 | for(fd = 0; fd <= maxFd; fd++) { |
763 | /* |
764 | * Build up a list of completed events. We'll work off of |
765 | * this list as opposed to looping through the read and write |
766 | * fd sets since they can be affected by a callbacl routine. |
767 | */ |
768 | if(FD_ISSET(fd, &readFdSetCpy)) { |
769 | numRdPosted++; |
770 | FD_SET(fd, &readFdSetPost); |
771 | FD_CLR(fd, &readFdSet); |
772 | } |
773 | |
774 | if(FD_ISSET(fd, &writeFdSetCpy)) { |
775 | numWrPosted++; |
776 | FD_SET(fd, &writeFdSetPost); |
777 | FD_CLR(fd, &writeFdSet); |
778 | } |
779 | } |
780 | } |
781 | |
782 | if(numRdPosted == 0 && numWrPosted == 0) |
783 | return 0; |
aadcc3c8 |
784 | |
0198fd3c |
785 | for(fd = 0; fd <= maxFd; fd++) { |
786 | /* |
787 | * Do reads and dispatch callback. |
788 | */ |
aadcc3c8 |
789 | if(FD_ISSET(fd, &readFdSetPost) |
0198fd3c |
790 | && asyncIoTable[AIO_RD_IX(fd)].inUse) { |
791 | |
792 | numRdPosted--; |
793 | FD_CLR(fd, &readFdSetPost); |
794 | aioPtr = &asyncIoTable[AIO_RD_IX(fd)]; |
aadcc3c8 |
795 | |
0198fd3c |
796 | len = read(aioPtr->fd, aioPtr->buf, aioPtr->len); |
797 | |
798 | procPtr = aioPtr->procPtr; |
799 | aioPtr->procPtr = NULL; |
800 | clientData = aioPtr->clientData; |
801 | aioPtr->inUse = 0; |
802 | |
803 | (*procPtr)(clientData, len); |
804 | } |
805 | |
806 | /* |
807 | * Do writes and dispatch callback. |
808 | */ |
809 | if(FD_ISSET(fd, &writeFdSetPost) && |
810 | asyncIoTable[AIO_WR_IX(fd)].inUse) { |
811 | |
812 | numWrPosted--; |
813 | FD_CLR(fd, &writeFdSetPost); |
814 | aioPtr = &asyncIoTable[AIO_WR_IX(fd)]; |
aadcc3c8 |
815 | |
0198fd3c |
816 | len = write(aioPtr->fd, aioPtr->buf, aioPtr->len); |
817 | |
818 | procPtr = aioPtr->procPtr; |
819 | aioPtr->procPtr = NULL; |
820 | clientData = aioPtr->clientData; |
821 | aioPtr->inUse = 0; |
822 | (*procPtr)(clientData, len); |
823 | } |
824 | } |
825 | return 0; |
826 | } |
827 | |
828 | \f |
829 | /* |
830 | *---------------------------------------------------------------------- |
831 | * |
832 | * ClientAddrOK -- |
833 | * |
834 | * Checks if a client address is in a list of allowed addresses |
835 | * |
836 | * Results: |
837 | * TRUE if address list is empty or client address is present |
838 | * in the list, FALSE otherwise. |
839 | * |
840 | *---------------------------------------------------------------------- |
841 | */ |
0b7c9662 |
842 | static int ClientAddrOK(struct sockaddr_in *saPtr, const char *clientList) |
0198fd3c |
843 | { |
844 | int result = FALSE; |
845 | char *clientListCopy, *cur, *next; |
846 | char *newString = NULL; |
847 | int strLen; |
848 | |
849 | if(clientList == NULL || *clientList == '\0') { |
850 | return TRUE; |
851 | } |
852 | |
853 | strLen = strlen(clientList); |
aadcc3c8 |
854 | clientListCopy = (char *)malloc(strLen + 1); |
0198fd3c |
855 | assert(newString != NULL); |
856 | memcpy(newString, clientList, strLen); |
857 | newString[strLen] = '\000'; |
aadcc3c8 |
858 | |
0198fd3c |
859 | for(cur = clientListCopy; cur != NULL; cur = next) { |
860 | next = strchr(cur, ','); |
861 | if(next != NULL) { |
862 | *next++ = '\0'; |
863 | } |
864 | if(inet_addr(cur) == saPtr->sin_addr.s_addr) { |
865 | result = TRUE; |
866 | break; |
867 | } |
868 | } |
869 | free(clientListCopy); |
870 | return result; |
871 | } |
872 | |
873 | \f |
874 | /* |
875 | *---------------------------------------------------------------------- |
876 | * |
877 | * AcquireLock -- |
878 | * |
879 | * On platforms that implement concurrent calls to accept |
880 | * on a shared listening ipcFd, returns 0. On other platforms, |
881 | * acquires an exclusive lock across all processes sharing a |
882 | * listening ipcFd, blocking until the lock has been acquired. |
883 | * |
884 | * Results: |
885 | * 0 for successful call, -1 in case of system error (fatal). |
886 | * |
887 | * Side effects: |
888 | * This process now has the exclusive lock. |
889 | * |
890 | *---------------------------------------------------------------------- |
891 | */ |
0b7c9662 |
892 | static int AcquireLock(int sock, int fail_on_intr) |
0198fd3c |
893 | { |
894 | #ifdef USE_LOCKING |
0b7c9662 |
895 | do { |
896 | struct flock lock; |
897 | lock.l_type = F_WRLCK; |
898 | lock.l_start = 0; |
899 | lock.l_whence = SEEK_SET; |
900 | lock.l_len = 0; |
901 | |
902 | if (fcntl(sock, F_SETLKW, &lock) != -1) |
903 | return 0; |
904 | } while (errno == EINTR && !fail_on_intr); |
905 | |
906 | return -1; |
907 | |
908 | #else |
0198fd3c |
909 | return 0; |
0b7c9662 |
910 | #endif |
0198fd3c |
911 | } |
912 | \f |
913 | /* |
914 | *---------------------------------------------------------------------- |
915 | * |
916 | * ReleaseLock -- |
917 | * |
918 | * On platforms that implement concurrent calls to accept |
919 | * on a shared listening ipcFd, does nothing. On other platforms, |
920 | * releases an exclusive lock acquired by AcquireLock. |
921 | * |
922 | * Results: |
923 | * 0 for successful call, -1 in case of system error (fatal). |
924 | * |
925 | * Side effects: |
926 | * This process no longer holds the lock. |
927 | * |
928 | *---------------------------------------------------------------------- |
929 | */ |
0b7c9662 |
930 | static int ReleaseLock(int sock) |
0198fd3c |
931 | { |
932 | #ifdef USE_LOCKING |
0b7c9662 |
933 | do { |
934 | struct flock lock; |
935 | lock.l_type = F_UNLCK; |
936 | lock.l_start = 0; |
937 | lock.l_whence = SEEK_SET; |
938 | lock.l_len = 0; |
939 | |
940 | if (fcntl(sock, F_SETLK, &lock) != -1) |
941 | return 0; |
942 | } while (errno == EINTR); |
0198fd3c |
943 | |
0b7c9662 |
944 | return -1; |
945 | |
946 | #else |
0198fd3c |
947 | return 0; |
0b7c9662 |
948 | #endif |
0198fd3c |
949 | } |
950 | |
951 | \f |
69f62c0e |
952 | /********************************************************************** |
aadcc3c8 |
953 | * Determine if the errno resulting from a failed accept() warrants a |
69f62c0e |
954 | * retry or exit(). Based on Apache's http_main.c accept() handling |
955 | * and Stevens' Unix Network Programming Vol 1, 2nd Ed, para. 15.6. |
956 | */ |
957 | static int is_reasonable_accept_errno (const int error) |
958 | { |
959 | switch (error) { |
960 | #ifdef EPROTO |
aadcc3c8 |
961 | /* EPROTO on certain older kernels really means ECONNABORTED, so |
962 | * we need to ignore it for them. See discussion in new-httpd |
963 | * archives nh.9701 search for EPROTO. Also see nh.9603, search |
964 | * for EPROTO: There is potentially a bug in Solaris 2.x x<6, and |
965 | * other boxes that implement tcp sockets in userland (i.e. on top of |
966 | * STREAMS). On these systems, EPROTO can actually result in a fatal |
967 | * loop. See PR#981 for example. It's hard to handle both uses of |
69f62c0e |
968 | * EPROTO. */ |
969 | case EPROTO: |
970 | #endif |
971 | #ifdef ECONNABORTED |
972 | case ECONNABORTED: |
973 | #endif |
974 | /* Linux generates the rest of these, other tcp stacks (i.e. |
975 | * bsd) tend to hide them behind getsockopt() interfaces. They |
976 | * occur when the net goes sour or the client disconnects after the |
977 | * three-way handshake has been done in the kernel but before |
978 | * userland has picked up the socket. */ |
979 | #ifdef ECONNRESET |
980 | case ECONNRESET: |
981 | #endif |
982 | #ifdef ETIMEDOUT |
983 | case ETIMEDOUT: |
984 | #endif |
985 | #ifdef EHOSTUNREACH |
986 | case EHOSTUNREACH: |
987 | #endif |
988 | #ifdef ENETUNREACH |
989 | case ENETUNREACH: |
990 | #endif |
991 | return 1; |
992 | |
993 | default: |
994 | return 0; |
995 | } |
996 | } |
997 | |
998 | /********************************************************************** |
aadcc3c8 |
999 | * This works around a problem on Linux 2.0.x and SCO Unixware (maybe |
1000 | * others?). When a connect() is made to a Unix Domain socket, but its |
1001 | * not accept()ed before the web server gets impatient and close()s, an |
1002 | * accept() results in a valid file descriptor, but no data to read. |
1003 | * This causes a block on the first read() - which never returns! |
1004 | * |
1005 | * Another approach to this is to write() to the socket to provoke a |
1006 | * SIGPIPE, but this is a pain because of the FastCGI protocol, the fact |
1007 | * that whatever is written has to be universally ignored by all FastCGI |
1008 | * web servers, and a SIGPIPE handler has to be installed which returns |
1009 | * (or SIGPIPE is ignored). |
1010 | * |
1011 | * READABLE_UNIX_FD_DROP_DEAD_TIMEVAL = 2,0 by default. |
1012 | * |
1013 | * Making it shorter is probably safe, but I'll leave that to you. Making |
1014 | * it 0,0 doesn't work reliably. The shorter you can reliably make it, |
1015 | * the faster your application will be able to recover (waiting 2 seconds |
1016 | * may _cause_ the problem when there is a very high demand). At any rate, |
1017 | * this is better than perma-blocking. |
1018 | */ |
69f62c0e |
1019 | static int is_af_unix_keeper(const int fd) |
1020 | { |
1021 | struct timeval tval = { READABLE_UNIX_FD_DROP_DEAD_TIMEVAL }; |
1022 | fd_set read_fds; |
1023 | |
1024 | FD_ZERO(&read_fds); |
1025 | FD_SET(fd, &read_fds); |
aadcc3c8 |
1026 | |
69f62c0e |
1027 | return select(fd + 1, &read_fds, NULL, NULL, &tval) >= 0 && FD_ISSET(fd, &read_fds); |
1028 | } |
1029 | |
0198fd3c |
1030 | /* |
1031 | *---------------------------------------------------------------------- |
1032 | * |
0b7c9662 |
1033 | * OS_Accept -- |
0198fd3c |
1034 | * |
1035 | * Accepts a new FastCGI connection. This routine knows whether |
1036 | * we're dealing with TCP based sockets or NT Named Pipes for IPC. |
1037 | * |
1038 | * Results: |
1039 | * -1 if the operation fails, otherwise this is a valid IPC fd. |
1040 | * |
1041 | * Side effects: |
1042 | * New IPC connection is accepted. |
1043 | * |
1044 | *---------------------------------------------------------------------- |
1045 | */ |
1dd5d7a8 |
1046 | int OS_Accept(int listen_sock, int fail_on_intr, const char *webServerAddrs) |
0198fd3c |
1047 | { |
1048 | int socket; |
69f62c0e |
1049 | union { |
0198fd3c |
1050 | struct sockaddr_un un; |
a3c43615 |
1051 | struct sockaddr_in in; |
0198fd3c |
1052 | } sa; |
ed728efb |
1053 | |
0b7c9662 |
1054 | for (;;) { |
1055 | if (AcquireLock(listen_sock, fail_on_intr)) |
1056 | return -1; |
a3c43615 |
1057 | |
0b7c9662 |
1058 | for (;;) { |
69f62c0e |
1059 | do { |
0b7c9662 |
1060 | #ifdef HAVE_SOCKLEN |
1061 | socklen_t len = sizeof(sa); |
1062 | #else |
1063 | int len = sizeof(sa); |
1064 | #endif |
1065 | socket = accept(listen_sock, (struct sockaddr *)&sa, &len); |
1066 | } while (socket < 0 && errno == EINTR && !fail_on_intr); |
a3c43615 |
1067 | |
69f62c0e |
1068 | if (socket < 0) { |
1069 | if (!is_reasonable_accept_errno(errno)) { |
a3c43615 |
1070 | int errnoSave = errno; |
0b7c9662 |
1071 | ReleaseLock(listen_sock); |
a3c43615 |
1072 | errno = errnoSave; |
69f62c0e |
1073 | return (-1); |
a3c43615 |
1074 | } |
69f62c0e |
1075 | errno = 0; |
1076 | } |
0b7c9662 |
1077 | else { /* socket >= 0 */ |
69f62c0e |
1078 | int set = 1; |
aadcc3c8 |
1079 | |
69f62c0e |
1080 | if (sa.in.sin_family != AF_INET) |
1081 | break; |
aadcc3c8 |
1082 | |
69f62c0e |
1083 | #ifdef TCP_NODELAY |
1084 | /* No replies to outgoing data, so disable Nagle */ |
1085 | setsockopt(socket, IPPROTO_TCP, TCP_NODELAY, (char *)&set, sizeof(set)); |
aadcc3c8 |
1086 | #endif |
1087 | |
69f62c0e |
1088 | /* Check that the client IP address is approved */ |
1dd5d7a8 |
1089 | if (ClientAddrOK(&sa.in, webServerAddrs)) |
69f62c0e |
1090 | break; |
aadcc3c8 |
1091 | |
69f62c0e |
1092 | close(socket); |
0b7c9662 |
1093 | } /* socket >= 0 */ |
1094 | } /* for(;;) */ |
aadcc3c8 |
1095 | |
0b7c9662 |
1096 | if (ReleaseLock(listen_sock)) |
69f62c0e |
1097 | return (-1); |
aadcc3c8 |
1098 | |
69f62c0e |
1099 | if (sa.in.sin_family != AF_UNIX || is_af_unix_keeper(socket)) |
1100 | break; |
aadcc3c8 |
1101 | |
69f62c0e |
1102 | close(socket); |
1103 | } /* while(1) - lock */ |
a3c43615 |
1104 | |
a3c43615 |
1105 | return (socket); |
0198fd3c |
1106 | } |
1107 | \f |
1108 | /* |
1109 | *---------------------------------------------------------------------- |
1110 | * |
1111 | * OS_IpcClose |
1112 | * |
1113 | * OS IPC routine to close an IPC connection. |
1114 | * |
1115 | * Results: |
1116 | * |
1117 | * |
1118 | * Side effects: |
1119 | * IPC connection is closed. |
1120 | * |
1121 | *---------------------------------------------------------------------- |
1122 | */ |
1123 | int OS_IpcClose(int ipcFd) |
1124 | { |
1125 | return OS_Close(ipcFd); |
1126 | } |
1127 | |
1128 | \f |
1129 | /* |
1130 | *---------------------------------------------------------------------- |
1131 | * |
1132 | * OS_IsFcgi -- |
1133 | * |
1134 | * Determines whether this process is a FastCGI process or not. |
1135 | * |
1136 | * Results: |
1137 | * Returns 1 if FastCGI, 0 if not. |
1138 | * |
1139 | * Side effects: |
1140 | * None. |
1141 | * |
1142 | *---------------------------------------------------------------------- |
1143 | */ |
0b7c9662 |
1144 | int OS_IsFcgi(int sock) |
0198fd3c |
1145 | { |
b22c3782 |
1146 | union { |
1147 | struct sockaddr_in in; |
1148 | struct sockaddr_un un; |
1149 | } sa; |
8eac3e1b |
1150 | #ifdef HAVE_SOCKLEN |
ed728efb |
1151 | socklen_t len = sizeof(sa); |
1152 | #else |
b22c3782 |
1153 | int len = sizeof(sa); |
ed728efb |
1154 | #endif |
0198fd3c |
1155 | |
0b7c9662 |
1156 | if (getpeername(sock, (struct sockaddr *)&sa, &len) != 0 && errno == ENOTCONN) { |
1157 | return TRUE; |
1158 | } |
1159 | else { |
1160 | return FALSE; |
1161 | } |
0198fd3c |
1162 | } |
1163 | \f |
1164 | /* |
1165 | *---------------------------------------------------------------------- |
1166 | * |
1167 | * OS_SetFlags -- |
1168 | * |
1169 | * Sets selected flag bits in an open file descriptor. |
1170 | * |
1171 | *---------------------------------------------------------------------- |
1172 | */ |
1173 | void OS_SetFlags(int fd, int flags) |
1174 | { |
1175 | int val; |
1176 | if((val = fcntl(fd, F_GETFL, 0)) < 0) { |
1177 | exit(errno); |
1178 | } |
1179 | val |= flags; |
1180 | if(fcntl(fd, F_SETFL, val) < 0) { |
1181 | exit(errno); |
1182 | } |
1183 | } |