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 * Add <ident-info> to the unique identifying information for
42 * this fcgi program. May be repeated; order is significant.
45 * Use <ident> rather than hex(sha256(<interp>\0<script>\0))
46 * as the basename of the leafname of the fcgi rendezvous
47 * socket. If <ident> contains only hex digit characters it
48 * ought to be no more than 32 characters. <ident> should
49 * not contain spaces or commas (see below).
52 * Start <numservers> instances of the program. This
53 * determines the maximum concurrency. (Note that unlike
54 * speedy, the specified number of servers is started
55 * right away.) The default is 4.
58 * Stale server check interval, in seconds. The worker
59 * process group will get a SIGTERM when it is no longer
60 * needed to process new requests. Ideally it would continue
61 * to serve any existing requests. The SIGTERM will arrive no
62 * earlier than <interval> after the last request arrived at
63 * the containing webserver. Default is 300.
66 * Debug mode. Do not actually run program. Instead, print
67 * out what we would do.
69 * <options> and <interpreter> can be put into a single argument
70 * to cgi-fcgi-interp, separated by spaces or commas. <interpreter>
73 * cgi-fcgi-interp automatically expires old sockets, including
74 * ones where the named script is out of date.
77 * Uses one of two directories
78 * /var/run/user/<UID>/cgi-fcgi-interp/
79 * ~/.cgi-fcgi-interp/<node>/
80 * and inside there uses these paths
82 * l<ident> used to lock around garbage collection
84 * If -M<ident> is not specified then an initial substring of the
85 * lowercase hex of the sha256 of <interp>\0<script>\0 is
86 * used. The substring is chosen so that the whole path is 10 bytes
87 * shorter than sizeof(sun_path). But always at least 33 characters.
89 * <node> is truncated at the first `.' and after the first 32
93 * - see if /var/run/user exists
94 * if so, lstat /var/run/user/<UID> and check that
95 * we own it and it's X700; if not, fail
96 * if it's ok then <base> is /var/run/user/<UID>
97 * otherwise, look for and maybe create ~/.cgi-fcgi-interp
98 * (where ~ is HOME or from getpwuid)
99 * and then <base> is ~/.cgi-fcgi-interp/<node>
100 * - calculate pathname (checking <ident> length is OK)
101 * - check for and maybe create <base>
102 * - stat and lstat the <script>
103 * - stat the socket and check its timestamp
104 * if it is too old, unlink it
105 * - dup stderr, mark no cloexec
106 * - set CHIARKUTILS_CGIFCGIINTERP_STAGE2=<stderr-copy-fd>
107 * - run cgi-fcgi -connect SOCKET <script>
109 * When CHIARKUTILS_CGIFCGIINTERP_STAGE2 is set, --stage2 does this:
110 * - dup2 <was-stderr> to fd 2
111 * - open /dev/null and expect fd 1 (and if not, close it)
112 * - become a new process group
113 * - lstat <socket> to find its inum, mtime
114 * - fork/exec <interp> <script>
115 * - periodically lstat <interp> and <script> and
116 * if mtime is newer than our start time
117 * kill process group (at second iteration)
130 #include <sys/types.h>
131 #include <sys/stat.h>
132 #include <sys/utsname.h>
133 #include <sys/socket.h>
135 #include <sys/file.h>
142 #include <sys/wait.h>
144 #include <nettle/sha.h>
148 #define die common_die
149 #define diee common_diee
151 #define MINHEXHASH 33
152 #define STAGE2_VAR "CHIARKUTILS_CGIFCGIINTERP_STAGE2"
154 static const char *interp, *ident;
155 static int numservers=4, debugmode;
156 static int check_interval=300;
158 static struct sha256_ctx identsc;
162 void diee(const char *m) {
163 err(127, "error: %s failed", m);
166 static void fusagemessage(FILE *f) {
167 fprintf(f, "usage: #!/usr/bin/cgi-fcgi-interp [<options>]\n");
170 void usagemessage(void) { fusagemessage(stderr); }
172 static void of_help(const struct cmdinfo *ci, const char *val) {
173 fusagemessage(stdout);
174 if (ferror(stdout)) diee("write usage message to stdout");
178 static void of_iassign(const struct cmdinfo *ci, const char *val) {
181 errno= 0; v= strtol(val,&ep,10);
182 if (!*val || *ep || errno || v<INT_MIN || v>INT_MAX)
183 badusage("bad integer argument `%s' for --%s",val,ci->olong);
187 static void ident_addstring(const struct cmdinfo *ci, const char *string) {
188 /* ci may be 0 and is provided so this can be .call */
189 sha256_update(&identsc,strlen(string)+1,string);
194 static const struct cmdinfo cmdinfos[]= {
195 { "help", 0, .call=of_help },
196 { 0, 'g', 1, .sassignto= &ident },
197 { 0, 'G', 1, .call= ident_addstring },
198 { 0, 'M', 1, .call=of_iassign, .iassignto= &numservers },
199 { 0, 'D', 0, .iassignto= &debugmode, .arg= 1 },
200 { 0, 'c', 1, .call=of_iassign, .iassignto= &check_interval },
205 static const char *run_base, *script, *socket_path;
206 static const char *run_base_mkdir_p;
207 static int stderr_copy;
209 static bool find_run_base_var_run(void) {
214 try = m_asprintf("%s/%lu", "/var/run/user", us);
215 r = lstat(try, &stab);
217 if (errno == ENOENT ||
221 return 0; /* oh well */
222 diee("stat /var/run/user/UID");
224 if (!S_ISDIR(stab.st_mode)) {
225 warnx("%s not a directory, falling back to ~\n", try);
228 if (stab.st_uid != us) {
229 warnx("%s not owned by uid %lu, falling back to ~\n", try,
233 if (stab.st_mode & 0077) {
234 warnx("%s writeable by group or other, falling back to ~\n", try);
237 run_base = m_asprintf("%s/%s", try, "cgi-fcgi-interp");
241 static bool find_run_base_home(void) {
247 pw = getpwuid(us); if (!pw) diee("getpwent(uid)");
249 r = uname(&ut); if (r) diee("uname(2)");
250 dot = strchr(ut.nodename, '.');
252 if (sizeof(ut.nodename) > 32)
255 run_base_mkdir_p = m_asprintf("%s/%s", pw->pw_dir, ".cgi-fcgi-interp");
256 try = m_asprintf("%/%s", run_base_mkdir_p, ut.nodename);
261 static void find_socket_path(void) {
262 struct sockaddr_un sun;
265 us = getuid(); if (us==(uid_t)-1) diee("getuid");
267 find_run_base_var_run() ||
268 find_run_base_home() ||
271 int maxidentlen = sizeof(sun.sun_path) - strlen(run_base) - 10 - 2;
274 if (maxidentlen < MINHEXHASH)
275 errx(127,"base directory `%s'"
276 " leaves only %d characters for id hash"
277 " which is too little (<%d)",
278 run_base, maxidentlen, MINHEXHASH);
280 int identlen = maxidentlen > 64 ? 64 : maxidentlen;
281 char *hexident = xmalloc(identlen + 2);
282 unsigned char bbuf[32];
285 ident_addstring(0,interp);
286 ident_addstring(0,script);
287 sha256_digest(&identsc,sizeof(bbuf),bbuf);
289 for (i=0; i<identlen; i += 2)
290 sprintf(hexident+i, "%02x", bbuf[i/2]);
292 hexident[identlen] = 0;
296 if (strlen(ident) > maxidentlen)
297 errx(127, "base directory `%s' plus ident `%s' too long"
298 " (with spare) for socket (max ident %d)\n",
299 run_base, ident, maxidentlen);
301 r = mkdir(run_base, 0700);
302 if (r && errno==ENOENT && run_base_mkdir_p) {
303 r = mkdir(run_base_mkdir_p, 0700);
304 if (r) err(127,"mkdir %s (since %s was ENOENT)",run_base_mkdir_p,run_base);
305 r = mkdir(run_base, 0700);
308 if (!(errno == EEXIST))
309 err(127,"mkdir %s",run_base);
312 socket_path = m_asprintf("%s/s%s",run_base,ident);
316 * Regarding the macro timespeccmp:
318 * Copyright (c) 1982, 1986, 1993
319 * The Regents of the University of California. All rights reserved.
321 * Redistribution and use in source and binary forms, with or without
322 * modification, are permitted provided that the following conditions
324 * 1. Redistributions of source code must retain the above copyright
325 * notice, this list of conditions and the following disclaimer.
326 * 2. Redistributions in binary form must reproduce the above copyright
327 * notice, this list of conditions and the following disclaimer in the
328 * documentation and/or other materials provided with the distribution.
329 * 4. Neither the name of the University nor the names of its contributors
330 * may be used to endorse or promote products derived from this software
331 * without specific prior written permission.
333 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
334 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
335 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
336 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
337 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
338 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
339 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
340 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
341 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
342 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
345 * @(#)time.h 8.5 (Berkeley) 5/4/95
346 * $FreeBSD: head/sys/sys/time.h 275985 2014-12-21 05:07:11Z imp $
349 #define timespeccmp(tvp, uvp, cmp) \
350 (((tvp)->tv_sec == (uvp)->tv_sec) ? \
351 ((tvp)->tv_nsec cmp (uvp)->tv_nsec) : \
352 ((tvp)->tv_sec cmp (uvp)->tv_sec))
353 #endif /*timespeccmp*/
359 static bool stab_isnewer(const struct stat *a, const struct stat *b) {
361 fprintf(stderr,"stab_isnewer mtim %lu.%06lu %lu.06%lu\n",
362 (unsigned long)a->st_mtim.tv_sec,
363 (unsigned long)a->st_mtim.tv_nsec,
364 (unsigned long)b->st_mtim.tv_sec,
365 (unsigned long)b->st_mtim.tv_nsec);
366 return timespeccmp(&a->st_mtim, &b->st_mtim, >);
369 static void stab_mtimenow(struct stat *out) {
370 int r = clock_gettime(CLOCK_REALTIME, &out->st_mtim);
371 if (r) err(127,"(stage2) clock_gettime");
373 fprintf(stderr,"stab_mtimenow mtim %lu.%06lu\n",
374 (unsigned long)out->st_mtim.tv_sec,
375 (unsigned long)out->st_mtim.tv_nsec);
378 #else /* !defined(st_mtime) */
380 static bool stab_isnewer(const struct stat *a, const struct stat *b) {
382 fprintf(stderr,"stab_isnewer mtime %lu %lu\n",
383 (unsigned long)a->st_mtime,
384 (unsigned long)b->st_mtime);
385 return a->st_mtime > &b->st_mtime;
388 static void stab_mtimenow(struct stat *out) {
389 out->st_mtime = time(NULL);
390 if (baseline_time.st_mtime == (time_t)-1) err(127,"(stage2) time()");
392 fprintf(stderr,"stab_mtimenow mtime %lu\n",
393 (unsigned long)out->st_mtime);
396 #endif /* !defined(st_mtime) */
398 static bool check_garbage_vs(const struct stat *started) {
399 struct stat script_stab;
402 r = lstat(script, &script_stab);
403 if (r) err(127,"lstat script (%s)",script);
405 if (stab_isnewer(&script_stab, started))
408 if (S_ISLNK(script_stab.st_mode)) {
409 r = stat(script, &script_stab);
410 if (r) err(127,"stat script (%s0",script);
412 if (stab_isnewer(&script_stab, started))
419 static bool check_garbage(void) {
420 struct stat sock_stab;
423 r = lstat(socket_path, &sock_stab);
425 if ((errno == ENOENT))
426 return 0; /* well, no garbage then */
427 err(127,"stat socket (%s)",socket_path);
430 return check_garbage_vs(&sock_stab);
433 static void tidy_garbage(void) {
434 /* We lock l<ident> and re-check. The effect of this is that each
435 * stale socket is removed only once. So unless multiple updates to
436 * the script happen rapidly, we can't be racing with the cgi-fcgi
437 * (which is recreating the socket */
441 const char *lock_path = m_asprintf("%s/l%s",run_base,ident);
443 lockfd = open(lock_path, O_CREAT|O_RDWR, 0600);
444 if (lockfd<0) err(127,"create lock (%s)", lock_path);
446 r = flock(lockfd, LOCK_EX);
447 if (r) err(127,"lock lock (%s)", lock_path);
449 if (check_garbage()) {
450 r = unlink(socket_path);
452 if (!(errno == ENOENT))
453 err(127,"remove out-of-date socket (%s)", socket_path);
458 if (r) errx(127,"close lock (%s)", lock_path);
461 static void make_stderr_copy(void) {
462 stderr_copy = dup(2);
463 if (stderr_copy < 0) err(127,"dup stderr (for copy for stage2)");
466 static void prep_stage2(void) {
469 const char *stage2_val = m_asprintf("%d", stderr_copy);
470 r = setenv(STAGE2_VAR, stage2_val, 1);
471 if (r) err(127,"set %s (to announce to stage2)", STAGE2_VAR);
474 static void shbang_opts(const char *const **argv_io,
475 const struct cmdinfo *cmdinfos) {
476 myopt(argv_io, cmdinfos);
478 interp = *(*argv_io)++;
479 if (!interp) errx(127,"need interpreter argument");
482 /* stage2 predeclarations */
483 static void record_baseline_time(void);
484 static void become_pgrp(void);
485 static void setup_handlers(void);
486 static void spawn_script(void);
487 static void queue_alarm(void);
488 static void await_something(void);
490 int main(int argc, const char *const *argv) {
491 const char *smashedopt;
494 stage2 = getenv(STAGE2_VAR);
496 int stderrfd = atoi(stage2);
499 r = dup2(stderrfd, 2);
502 r = open("/dev/null",O_WRONLY);
503 if (r<0) err(127,"open /dev/null as stdout");
505 else if (r!=1) errx(127,"open /dev/null for stdout gave bad fd %d",r);
508 sha256_init(&identsc);
511 (smashedopt = argv[1]) &&
512 smashedopt[0]=='-' &&
513 (strchr(smashedopt,' ') || strchr(smashedopt,','))) {
514 /* single argument containg all the options and <interp> */
515 argv += 2; /* eat argv[0] and smashedopt */
516 const char *split_args[MAX_OPTS+1];
518 split_args[split_argc++] = argv[0];
520 if (split_argc >= MAX_OPTS) errx(127,"too many options in combined arg");
521 split_args[split_argc++] = smashedopt;
522 if (smashedopt[0] != '-') /* never true on first iteration */
524 char *delim = strchr(smashedopt,' ');
525 if (!delim) delim = strchr(smashedopt,',');
527 errx(127,"combined arg lacks <interpreter>");
529 smashedopt = delim+1;
531 assert(split_argc <= MAX_OPTS);
532 split_args[split_argc++] = 0;
534 const char *const *split_argv = split_args;
536 shbang_opts(&split_argv, cmdinfos);
538 if (!split_argv) errx(127,"combined arg too many non-option arguments");
540 shbang_opts(&argv, cmdinfos);
544 if (!script) errx(127,"need script argument");
545 if (*argv) errx(127,"too many arguments");
551 bool isgarbage = check_garbage();
554 printf("socket: %s\n",socket_path);
555 printf("interp: %s\n",interp);
556 printf("script: %s\n",script);
557 printf("garbage: %d\n",isgarbage);
568 "cgi-fcgi", "-connect", socket_path,
570 m_asprintf("%d", numservers),
572 err(127,"exec cgi-fcgi");
576 record_baseline_time();
589 /* It is most convenient to handle the recheck timeout, as well as
590 * child death, in signal handlers. Our signals all block each other,
591 * and the main program has signals blocked except in sigsuspend, so
592 * we don't need to worry about async-signal-safety, or errno. */
594 static struct stat baseline_time;
595 static pid_t script_child, stage2_pgrp;
596 static bool out_of_date;
598 static void record_baseline_time(void) {
599 stab_mtimenow(&baseline_time);
602 static void become_pgrp(void) {
605 stage2_pgrp = getpid();
608 if (r) err(127,"(stage2) setpgid");
611 static void atexit_handler(void) {
614 sighandler_t sigr = signal(SIGTERM,SIG_IGN);
615 if (sigr == SIG_ERR) warn("(stage2) signal(SIGTERM,SIG_IGN)");
617 r = killpg(stage2_pgrp,SIGTERM);
618 if (r) warn("(stage) killpg failed");
621 static void alarm_handler(int dummy) {
624 exit(0); /* transfers control to atexit_handler */
626 out_of_date = check_garbage_vs(&baseline_time);
630 static void child_handler(int dummy) {
633 pid_t got = waitpid(-1, &status, WNOHANG);
634 if (got == (pid_t)-1) err(127,"(stage2) waitpid");
635 if (got != script_child) {
636 warn("(stage2) waitpid got status %d for unknown child [%lu]",
637 status, (unsigned long)got);
640 if (WIFEXITED(status)) {
641 int v = WEXITSTATUS(status);
642 if (v) warn("program failed with error exit status %d", v);
644 } else if (WIFSIGNALED(status)) {
645 int s = WTERMSIG(status);
646 err(status & 0xff, "program died due to fatal signal %s%s",
647 strsignal(s), WCOREDUMP(status) ? " (core dumped" : "");
649 err(127, "program failed with crazy wait status %#x", status);
655 static void setup_handlers(void) {
659 r = atexit(atexit_handler);
660 if (r) err(127,"(stage2) atexit");
662 sigemptyset(&sa.sa_mask);
663 sigaddset(&sa.sa_mask, SIGALRM);
664 sigaddset(&sa.sa_mask, SIGCHLD);
667 r = sigprocmask(SIG_BLOCK, &sa.sa_mask, 0);
668 if (r) err(127,"(stage2) sigprocmask(SIG_BLOCK,)");
670 sa.sa_handler = alarm_handler;
671 r = sigaction(SIGALRM, &sa, 0);
672 if (r) err(127,"(stage2) sigaction SIGALRM");
674 sa.sa_flags |= SA_NOCLDSTOP;
675 sa.sa_handler = child_handler;
676 r = sigaction(SIGCHLD, &sa, 0);
677 if (r) err(127,"(stage2) sigaction SIGCHLD");
680 static void spawn_script(void) {
681 script_child = fork();
682 if (script_child == (pid_t)-1) err(127,"(stage2) fork");
685 interp, script, (char*)0);
686 err(127,"(stage2) exec interpreter (`%s', for `%s')\n",interp,script);
690 static void queue_alarm(void) {
691 alarm(check_interval);
694 static void await_something(void) {
700 r = sigsuspend(&mask);
702 if (errno != EINTR) err(127,"(stage2) sigsuspend");