2 * "Interpreter" that you can put in #! like this
3 * #!/usr/bin/cgi-fcgi-interp [<options>] <interpreter>
4 * #!/usr/bin/cgi-fcgi-interp [<options>],<interpreter>
7 * cgi-fcgi-interp.[ch] - C helpers common to the whole of chiark-utils
9 * Copyright 2016 Ian Jackson
10 * Copyright 1982,1986,1993 The Regents of the University of California
12 * This program is free software; you can redistribute it and/or modify
13 * it under the terms of the GNU General Public License as published by
14 * the Free Software Foundation; either version 3 of the License, or
15 * (at your option) any later version.
17 * This program is distributed in the hope that it will be useful,
18 * but WITHOUT ANY WARRANTY; without even the implied warranty of
19 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
20 * GNU General Public License for more details.
22 * You should have received a copy of the GNU General Public
23 * License along with this file; if not, consult the Free Software
24 * Foundation's website at www.fsf.org, or the GNU Project website at
27 * See below for a BSD 3-clause notice regarding timespeccmp.
30 * The result is a program which looks, when executed via the #!
31 * line, like a CGI program. But the script inside will be executed
32 * via <interpreter> in an fcgi context.
37 * The real interpreter to use. Eg "perl". Need not
38 * be an absolute path; will be fed to execvp.
41 * Use <ident> rather than hex(sha256(<script>))
42 * as the basename of the leafname of the fcgi rendezvous
43 * socket. If <ident> contains only hex digit characters it
44 * ought to be no more than 32 characters. <ident> should
45 * not contain spaces or commas (see below).
48 * Start <numservers> instances of the program. This
49 * determines the maximum concurrency. (Note that unlike
50 * speedy, the specified number of servers is started
51 * right away.) The default is 4.
54 * Stale server check interval, in seconds. The worker
55 * process group will get a SIGTERM when it is no longer
56 * needed to process new requests. Ideally it would continue
57 * to serve any existing requests. The SIGTERM will arrive no
58 * earlier than <interval> after the last request arrived at
59 * the containing webserver. Default is 300.
62 * Debug mode. Do not actually run program. Instead, print
63 * out what we would do.
65 * <options> and <interpreter> can be put into a single argument
66 * to cgi-fcgi-interp, separated by spaces or commas. <interpreter>
69 * cgi-fcgi-interp automatically expires old sockets, including
70 * ones where the named script is out of date.
74 * Uses one of two directories
75 * /var/run/user/<UID>/cgi-fcgi-interp/
76 * ~/.cgi-fcgi-interp/<node>/
77 * and inside there uses these paths
79 * l<ident> used to lock around garbage collection
81 * If -M<ident> is not specified then an initial substricg of the
82 * lowercase hex of the sha256 of the <script> (ie, our argv[1]) is
83 * used. The substring is chosen so that the whole path is 10 bytes
84 * shorter than sizeof(sun_path). But always at least 33 characters.
86 * <node> is truncated at the first `.' and after the first 32
90 * - see if /var/run/user exists
91 * if so, lstat /var/run/user/<UID> and check that
92 * we own it and it's X700; if not, fail
93 * if it's ok then <base> is /var/run/user/<UID>
94 * otherwise, look for and maybe create ~/.cgi-fcgi-interp
95 * (where ~ is HOME or from getpwuid)
96 * and then <base> is ~/.cgi-fcgi-interp/<node>
97 * - calculate pathname (checking <ident> length is OK)
98 * - check for and maybe create <base>
99 * - stat and lstat the <script>
100 * - stat the socket and check its timestamp
101 * if it is too old, unlink it
102 * - dup stderr, mark no cloexec
103 * - run cgi-fcgi -connect SOCKET \
105 * --stage2 <was-stderr> <socket> \
110 * --stage2 does this:
111 * - dup2 <was-stderr> to fd 2
112 * - open /dev/null and expect fd 1 (and if not, close it)
113 * - become a new process group
114 * - lstat <socket> to find its inum, mtime
115 * - fork/exec <interp> <script>
116 * - periodically lstat <interp> and <script> and
117 * if mtime is newer than our start time
118 * kill process group (at second iteration)
131 #include <sys/types.h>
132 #include <sys/stat.h>
133 #include <sys/utsname.h>
134 #include <sys/socket.h>
136 #include <sys/file.h>
144 #include <nettle/sha.h>
148 #define die common_die
149 #define diee common_diee
151 #define MINHEXHASH 33
153 static const char *interp, *ident;
154 static int numservers=4, debugmode, stage2;
155 static int check_interval=300;
157 void diee(const char *m) {
158 err(127, "error: %s failed", m);
161 static void fusagemessage(FILE *f) {
162 fprintf(f, "usage: #!/usr/bin/cgi-fcgi-interp [<options>]\n");
165 void usagemessage(void) { fusagemessage(stderr); }
167 static void of_help(const struct cmdinfo *ci, const char *val) {
168 fusagemessage(stdout);
169 if (ferror(stdout)) diee("write usage message to stdout");
173 static void of_iassign(const struct cmdinfo *ci, const char *val) {
176 errno= 0; v= strtol(val,&ep,10);
177 if (!*val || *ep || errno || v<INT_MIN || v>INT_MAX)
178 badusage("bad integer argument `%s' for --%s",val,ci->olong);
184 static const struct cmdinfo cmdinfos[]= {
185 { "help", 0, .call= of_help },
186 { 0, 'g', 1, .sassignto= &ident },
187 { 0, 'M', 1, .call=of_iassign, .iassignto= &numservers },
188 { 0, 'D', 0, .iassignto= &debugmode, .arg= 1 },
189 { 0, 'c', 1, .call=of_iassign, .iassignto= &check_interval },
190 { "--stage2",0, 0, .iassignto= &stage2, .arg= 1 },
195 static const char *run_base, *script, *socket_path;
196 static int stderr_copy;
198 static bool find_run_base_var_run(void) {
203 try = m_asprintf("%s/%lu", "/var/run/user", us);
204 r = lstat(try, &stab);
206 if (errno == ENOENT ||
210 return 0; /* oh well */
211 diee("stat /var/run/user/UID");
213 if (!S_ISDIR(stab.st_mode)) {
214 warnx("%s not a directory, falling back to ~\n", try);
217 if (stab.st_uid != us) {
218 warnx("%s not owned by uid %lu, falling back to ~\n", try,
222 if (stab.st_mode & 0077) {
223 warnx("%s writeable by group or other, falling back to ~\n", try);
226 run_base = m_asprintf("%s/%s", try, "cgi-fcgi-interp");
230 static bool find_run_base_home(void) {
236 pw = getpwuid(us); if (!pw) diee("getpwent(uid)");
238 r = uname(&ut); if (r) diee("uname(2)");
239 dot = strchr(ut.nodename, '.');
241 if (sizeof(ut.nodename) > 32)
244 try = m_asprintf("%s/%s/%s", pw->pw_dir, ".cgi-fcgi-interp", ut.nodename);
249 static void find_socket_path(void) {
250 struct sockaddr_un sun;
253 us = getuid(); if (us==(uid_t)-1) diee("getuid");
255 find_run_base_var_run() ||
256 find_run_base_home() ||
259 int maxidentlen = sizeof(sun.sun_path) - strlen(run_base) - 10 - 2;
262 if (maxidentlen < MINHEXHASH)
263 errx(127,"base directory `%s'"
264 " leaves only %d characters for id hash"
265 " which is too little (<%d)",
266 run_base, maxidentlen, MINHEXHASH);
268 int identlen = maxidentlen > 64 ? 64 : maxidentlen;
269 char *hexident = xmalloc(identlen + 2);
270 struct sha256_ctx sc;
271 unsigned char bbuf[32];
275 sha256_update(&sc,strlen(interp)+1,interp);
276 sha256_update(&sc,strlen(script)+1,script);
277 sha256_digest(&sc,sizeof(bbuf),bbuf);
279 for (i=0; i<identlen; i += 2)
280 sprintf(hexident+i, "%02x", bbuf[i/2]);
282 hexident[identlen] = 0;
286 if (strlen(ident) > maxidentlen)
287 errx(127, "base directory `%s' plus ident `%s' too long"
288 " (with spare) for socket (max ident %d)\n",
289 run_base, ident, maxidentlen);
291 r = mkdir(run_base, 0700);
293 if (!(errno == EEXIST))
294 err(127,"mkdir %s",run_base);
297 socket_path = m_asprintf("%s/g%s",run_base,ident);
301 * Regarding the macro timespeccmp:
303 * Copyright (c) 1982, 1986, 1993
304 * The Regents of the University of California. All rights reserved.
306 * Redistribution and use in source and binary forms, with or without
307 * modification, are permitted provided that the following conditions
309 * 1. Redistributions of source code must retain the above copyright
310 * notice, this list of conditions and the following disclaimer.
311 * 2. Redistributions in binary form must reproduce the above copyright
312 * notice, this list of conditions and the following disclaimer in the
313 * documentation and/or other materials provided with the distribution.
314 * 4. Neither the name of the University nor the names of its contributors
315 * may be used to endorse or promote products derived from this software
316 * without specific prior written permission.
318 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
319 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
320 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
321 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
322 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
323 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
324 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
325 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
326 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
327 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
330 * @(#)time.h 8.5 (Berkeley) 5/4/95
331 * $FreeBSD: head/sys/sys/time.h 275985 2014-12-21 05:07:11Z imp $
334 #define timespeccmp(tvp, uvp, cmp) \
335 (((tvp)->tv_sec == (uvp)->tv_sec) ? \
336 ((tvp)->tv_nsec cmp (uvp)->tv_nsec) : \
337 ((tvp)->tv_sec cmp (uvp)->tv_sec))
338 #endif /*timespeccmp*/
342 static bool stab_isnewer(const struct stat *a, const struct stat *b) {
344 return timespeccmp(&a->st_mtim, &b->st_mtim, >);
346 return a->st_mtime > &b->st_mtime;
350 static bool check_garbage_vs(const struct stat *started) {
351 struct stat script_stab;
352 struct stat sock_stab;
355 r = lstat(script, &script_stab);
356 if (r) err(127,"lstat script (%s)",script);
358 if (stab_isnewer(&script_stab, &sock_stab))
361 if (S_ISLNK(script_stab.st_mode)) {
362 r = stat(script, &script_stab);
363 if (r) err(127,"stat script (%s0",script);
365 if (stab_isnewer(&script_stab, &sock_stab))
372 static bool check_garbage(void) {
373 struct stat sock_stab;
376 r = lstat(socket_path, &sock_stab);
378 if ((errno == ENOENT))
379 return 0; /* well, no garbage then */
380 err(127,"stat socket (%s)",socket_path);
383 return check_garbage_vs(&sock_stab);
386 static void tidy_garbage(void) {
387 /* We lock l<ident> and re-check. The effect of this is that each
388 * stale socket is removed only once. So unless multiple updates to
389 * the script happen rapidly, we can't be racing with the cgi-fcgi
390 * (which is recreating the socket */
394 const char *lock_path = m_asprintf("%s/l%s",run_base,ident);
396 lockfd = open(lock_path, O_CREAT|O_RDWR, 0600);
397 if (lockfd<0) err(127,"create lock (%s)", lock_path);
399 r = flock(lockfd, LOCK_EX);
400 if (r) err(127,"lock lock (%s)", lock_path);
402 if (check_garbage()) {
403 r = unlink(socket_path);
405 if (!(errno == ENOENT))
406 err(127,"remove out-of-date socket (%s)", socket_path);
411 if (r) errx(127,"close lock (%s)", lock_path);
414 static void make_stderr_copy(void) {
415 stderr_copy = dup(2);
416 if (stderr_copy < 0) err(127,"dup stderr (for copy for stage2)");
419 static void shbang_opts(const char *const **argv_io,
420 const struct cmdinfo *cmdinfos) {
421 myopt(argv_io, cmdinfos);
423 interp = *(*argv_io)++;
424 if (!interp) errx(127,"need interpreter argument");
427 /* stage2 predeclarations */
428 static void record_baseline_time(void);
429 static void become_pgrp(void);
430 static void setup_handlers(void);
431 static void spawn_script(void);
432 static void queue_alarm(void);
433 static void await_something(void);
435 int main(int argc, const char *const *argv) {
436 const char *smashedopt, *us;
441 if (argc>=4 && !strcmp(argv[1],"--stage2")) {
445 int stderrfd = atoi(*++argv);
446 r = dup2(stderrfd, 2);
449 r = open("/dev/null",O_WRONLY);
450 if (r<0) err(127,"open /dev/null as stdout");
452 else if (r!=1) errx(127,"open /dev/null for stdout gave bad fd %d",r);
454 socket_path = *++argv;
458 (smashedopt = argv[1]) &&
459 smashedopt[0]=='-' &&
460 (strchr(smashedopt,' ') || strchr(smashedopt,','))) {
461 /* single argument containg all the options and <interp> */
462 argv += 2; /* eat argv[0] and smashedopt */
463 const char *split_args[MAX_OPTS+1];
465 split_args[split_argc++] = argv[0];
467 if (split_argc >= MAX_OPTS) errx(127,"too many options in combined arg");
468 split_args[split_argc++] = smashedopt;
469 if (smashedopt[0] != '-') /* never true on first iteration */
471 char *delim = strchr(smashedopt,' ');
472 if (!delim) delim = strchr(smashedopt,',');
474 errx(127,"combined arg lacks <interpreter>");
476 smashedopt = delim+1;
478 assert(split_argc <= MAX_OPTS);
479 split_args[split_argc++] = 0;
481 const char *const *split_argv = split_args;
483 shbang_opts(&split_argv, cmdinfos);
485 if (!split_argv) errx(127,"combined arg too many non-option arguments");
487 shbang_opts(&argv, cmdinfos);
491 if (!script) errx(127,"need script argument");
492 if (*argv) errx(127,"too many arguments");
498 bool isgarbage = check_garbage();
501 printf("socket: %s\n",socket_path);
502 printf("interp: %s\n",interp);
503 printf("script: %s\n",script);
504 printf("garbage: %d\n",isgarbage);
514 "cgi-fcti", "-connect", socket_path,
516 m_asprintf("-c%d", check_interval),
517 m_asprintf("%d", stderr_copy), socket_path,
520 err(127,"exec cgi-fcgi");
524 record_baseline_time();
537 /* It is most convenient to handle the recheck timeout, as well as
538 * child death, in signal handlers. Our signals all block each other,
539 * and the main program has signals blocked except in sigsuspend, so
540 * we don't need to worry about async-signal-safety, or errno. */
542 static struct stat baseline_time;
543 static pid_t script_child, stage2_pgrp;
544 static bool out_of_date;
546 static void record_baseline_time(void) {
548 int r = clock_gettime(CLOCK_REALTIME, &baseline_time.st_mtim);
549 if (r) err(127,"(stage2) clock_gettime");
551 baseline_time.st_mtime = time(NULL);
552 if (baseline_time.st_mtime == (time_t)-1) err(127,"(stage2) time()");
556 static void become_pgrp(void) {
559 stage2_pgrp = getpid();
562 if (r) err(127,"(stage2) setpgid");
565 static void atexit_handler(void) {
568 sighandler_t sigr = signal(SIGTERM,SIG_IGN);
569 if (sigr == SIG_ERR) warn("(stage2) signal(SIGTERM,SIG_IGN)");
571 r = killpg(stage2_pgrp,SIGTERM);
572 if (r) warn("(stage) killpg failed");
575 static void alarm_handler(int dummy) {
578 exit(0); /* transfers control to atexit_handler */
580 out_of_date = check_garbage_vs(&baseline_time);
584 static void setup_handlers(void) {
588 r = atexit(atexit_handler);
589 if (r) err(127,"(stage2) atexit");
591 sigemptyset(&sa.sa_mask);
592 sigaddset(&sa.sa_mask, SIGALRM);
593 sigaddset(&sa.sa_mask, SIGCHLD);
596 r = sigprocmask(SIG_BLOCK, &sa.sa_mask, 0);
597 if (r) err(127,"(stage2) sigprocmask(SIG_BLOCK,)");
599 sa.sa_handler = alarm_handler;
600 r = sigaction(SIGALRM, &sa, 0);
601 if (r) err(127,"(stage2) sigaction SIGALRM");
603 sa.sa_flags |= SA_NOCLDSTOP;
604 sa.sa_handler = child_handler;
605 r = sigaction(SIGCHLD, &sa, 0);
606 if (r) err(127,"(stage2) sigaction SIGCHLD");
609 static void spawn_script(void) {
610 script_child = fork();
611 if (script_child == (pid_t)-1) err(127,"(stage2) fork");
614 interp, script, (char*)0);
615 err(127,"(stage2) exec interpreter (`%s', for `%s')\n",interp,script);
619 static void queue_alarm(void) {
620 alarm(check_interval);
623 static void await_something(void) {
629 r = sigsuspend(&mask);
631 if (r != EINTR) err(127,"(stage2) sigsuspend");