You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@httpd.apache.org by Randy Terbush <ra...@zyzzyva.com> on 1996/12/04 03:15:26 UTC
Re: WWW Form Bug Report: "signal handlers use non-rentrant functions" on OTHER:NetBSD
> All right! This is probably the cause of all the mysterious spins we've been
> seeing. I think I'd prefer a proper fix, though. Could log_error be made safe
> instead?
>
> Cheers,
>
> Ben.
I took a stab at this. The question is how many other places need this
sort of change? sprintf is considered safe on Solaris.
Index: http_log.c
===================================================================
RCS file: /export/home/cvs/apache/src/http_log.c,v
retrieving revision 1.10
diff -c -r1.10 http_log.c
*** http_log.c 1996/12/01 20:28:25 1.10
--- http_log.c 1996/12/04 02:10:20
***************
*** 145,157 ****
fclose(pid_file);
}
! void log_error(char *err, server_rec *s) {
! fprintf(s->error_log, "[%s] %s\n",get_time(),err);
! fflush(s->error_log);
}
! void
! log_unixerr(const char *routine, const char *file, const char *msg,
server_rec *s)
{
const char *p, *q;
--- 145,160 ----
fclose(pid_file);
}
! void log_error(char *err, server_rec *s)
! {
! int len;
! char buf[512];
!
! len = sprintf(buf, "[%s] %s\n", get_time(), err);
! write(fileno(s->error_log), buf, len + 1);
}
! void log_unixerr(const char *routine, const char *file, const char *msg,
server_rec *s)
{
const char *p, *q;
> Rob Hartill wrote:
> >
> >
> > Thanks for the patch. I'll forward it to the developers list.
> >
> > cheers,
> > rob
> >
> > Liz Stokes wrote:
> > >> ilaine@panix.com wrote:
> > >> >
> > >> >Submitter: ilaine@panix.com
> > >> >Operating system: OTHER:NetBSD, version: 1.2_BETA
> > >> >Version of Apache Used: 1.1.1
> > >> >Extra Modules used:
> > >> >URL exhibiting problem:
> > >> >
> > >> >Symptoms:
> > >> >--
> > >> >sig_term, sig_bus, and sig_fault all call log_error, which calls many
> > >> >non re-entrant functions. In the example which brought the problem to my
> > >> >attention, httpd was in vfprintf when it received a SIGSEGV. It
> > >> >called log_error which caused it to re-enter vfprintf with unhappy
> > >> >results - it went into a tight loop inside vfprintf and chewed cycles
> > >> >at a prodigious rate till exterminated.
> > >> >
> > >> >I have patched our version to use a simple write() inside the signal handlers
> > >> >instead of log_error. It's not terribly elegant, and of course you lose the
> > >> >timestamp, but I'd be happy to provide the diff if you wish.
> > >>
> > >> Thanks for the info. Apache 1.2b1 will be released in the next few hours.
> > >> Many bugs have been fixed and problems resolved. Can you send us a patch
> > >> for 1.2b1 if the problem persists.
> > >
> > >The bug is still there - the signal handlers are unchanged. Appended is a
> > >diff for http_main.c
> > >
> > >-Liz
> > >
> > >
> > >*** http_main.c Sun Dec 1 15:28:26 1996
> > >--- http_main.c.new Tue Dec 3 14:58:52 1996
> > >***************
> > >*** 1066,1072 ****
> > > }
> > >
> > > void sig_term() {
> > >! log_error("httpd: caught SIGTERM, shutting down", server_conf);
> > > cleanup_scoreboard();
> > > #ifndef NO_KILLPG
> > > killpg(pgrp,SIGKILL);
> > >--- 1066,1073 ----
> > > }
> > >
> > > void sig_term() {
> > >! write (fileno(server_conf->error_log),
> > >! "httpd: caught SIGTERM, shutting down\n", 38);
> > > cleanup_scoreboard();
> > > #ifndef NO_KILLPG
> > > killpg(pgrp,SIGKILL);
> > >***************
> > >*** 1079,1085 ****
> > > }
> > >
> > > void bus_error() {
> > >! log_error("httpd: caught SIGBUS, dumping core", server_conf);
> > > chdir(server_root);
> > > abort();
> > > exit(1);
> > >--- 1080,1087 ----
> > > }
> > >
> > > void bus_error() {
> > >! write (fileno(server_conf->error_log),
> > >! "httpd: caught SIGBUS, dumping core\n", 36);
> > > chdir(server_root);
> > > abort();
> > > exit(1);
> > >***************
> > >*** 1086,1092 ****
> > > }
> > >
> > > void seg_fault() {
> > >! log_error("httpd: caught SIGSEGV, dumping core", server_conf);
> > > chdir(server_root);
> > > abort();
> > > exit(1);
> > >--- 1088,1095 ----
> > > }
> > >
> > > void seg_fault() {
> > >! write (fileno(server_conf->error_log),
> > >! "httpd: caught SIGSEGV, dumping core\n", 37);
> > > chdir(server_root);
> > > abort();
> > > exit(1);
> > >
> > >
> >
> >
>
> --
> Ben Laurie Phone: +44 (181) 994 6435 Email: ben@algroup.co.uk
> Freelance Consultant and Fax: +44 (181) 994 6472
> Technical Director URL: http://www.algroup.co.uk/Apache-SSL
> A.L. Digital Ltd, Apache Group member (http://www.apache.org)
> London, England. Apache-SSL author
Re: WWW Form Bug Report: "signal handlers use non-rentrant functions"
on OTHER:NetBSD
Posted by Paul Richards <p....@elsevier.co.uk>.
Randy Terbush <ra...@zyzzyva.com> writes:
> I took a stab at this. The question is how many other places need this
> sort of change? sprintf is considered safe on Solaris.
>
After actually looking at the patch, this is missing the real
point. You're still making a function call after a trap and that's not
safe to start with, the stack may not be in any sound state, the
data structures may be hosed. *Something* caused the trap in the first
place, how do you know it isn't what you're subsequently going to try
and use?
Take a look at the new Perl5 signal documentation, it makes the point
that about all you should do in such a case is die and even that isn't
guaranteed to be safe.
> Index: http_log.c
> ===================================================================
> RCS file: /export/home/cvs/apache/src/http_log.c,v
> retrieving revision 1.10
> diff -c -r1.10 http_log.c
> *** http_log.c 1996/12/01 20:28:25 1.10
> --- http_log.c 1996/12/04 02:10:20
> ***************
> *** 145,157 ****
> fclose(pid_file);
> }
>
> ! void log_error(char *err, server_rec *s) {
> ! fprintf(s->error_log, "[%s] %s\n",get_time(),err);
> ! fflush(s->error_log);
> }
>
> ! void
> ! log_unixerr(const char *routine, const char *file, const char *msg,
> server_rec *s)
> {
> const char *p, *q;
> --- 145,160 ----
> fclose(pid_file);
> }
>
> ! void log_error(char *err, server_rec *s)
> ! {
> ! int len;
> ! char buf[512];
> !
> ! len = sprintf(buf, "[%s] %s\n", get_time(), err);
> ! write(fileno(s->error_log), buf, len + 1);
> }
>
> ! void log_unixerr(const char *routine, const char *file, const char *msg,
> server_rec *s)
> {
> const char *p, *q;
>
>
>
>
> > Rob Hartill wrote:
> > >
> > >
> > > Thanks for the patch. I'll forward it to the developers list.
> > >
> > > cheers,
> > > rob
> > >
> > > Liz Stokes wrote:
> > > >> ilaine@panix.com wrote:
> > > >> >
> > > >> >Submitter: ilaine@panix.com
> > > >> >Operating system: OTHER:NetBSD, version: 1.2_BETA
> > > >> >Version of Apache Used: 1.1.1
> > > >> >Extra Modules used:
> > > >> >URL exhibiting problem:
> > > >> >
> > > >> >Symptoms:
> > > >> >--
> > > >> >sig_term, sig_bus, and sig_fault all call log_error, which calls many
> > > >> >non re-entrant functions. In the example which brought the problem to my
> > > >> >attention, httpd was in vfprintf when it received a SIGSEGV. It
> > > >> >called log_error which caused it to re-enter vfprintf with unhappy
> > > >> >results - it went into a tight loop inside vfprintf and chewed cycles
> > > >> >at a prodigious rate till exterminated.
> > > >> >
> > > >> >I have patched our version to use a simple write() inside the signal handlers
> > > >> >instead of log_error. It's not terribly elegant, and of course you lose the
> > > >> >timestamp, but I'd be happy to provide the diff if you wish.
> > > >>
> > > >> Thanks for the info. Apache 1.2b1 will be released in the next few hours.
> > > >> Many bugs have been fixed and problems resolved. Can you send us a patch
> > > >> for 1.2b1 if the problem persists.
> > > >
> > > >The bug is still there - the signal handlers are unchanged. Appended is a
> > > >diff for http_main.c
> > > >
> > > >-Liz
> > > >
> > > >
> > > >*** http_main.c Sun Dec 1 15:28:26 1996
> > > >--- http_main.c.new Tue Dec 3 14:58:52 1996
> > > >***************
> > > >*** 1066,1072 ****
> > > > }
> > > >
> > > > void sig_term() {
> > > >! log_error("httpd: caught SIGTERM, shutting down", server_conf);
> > > > cleanup_scoreboard();
> > > > #ifndef NO_KILLPG
> > > > killpg(pgrp,SIGKILL);
> > > >--- 1066,1073 ----
> > > > }
> > > >
> > > > void sig_term() {
> > > >! write (fileno(server_conf->error_log),
> > > >! "httpd: caught SIGTERM, shutting down\n", 38);
> > > > cleanup_scoreboard();
> > > > #ifndef NO_KILLPG
> > > > killpg(pgrp,SIGKILL);
> > > >***************
> > > >*** 1079,1085 ****
> > > > }
> > > >
> > > > void bus_error() {
> > > >! log_error("httpd: caught SIGBUS, dumping core", server_conf);
> > > > chdir(server_root);
> > > > abort();
> > > > exit(1);
> > > >--- 1080,1087 ----
> > > > }
> > > >
> > > > void bus_error() {
> > > >! write (fileno(server_conf->error_log),
> > > >! "httpd: caught SIGBUS, dumping core\n", 36);
> > > > chdir(server_root);
> > > > abort();
> > > > exit(1);
> > > >***************
> > > >*** 1086,1092 ****
> > > > }
> > > >
> > > > void seg_fault() {
> > > >! log_error("httpd: caught SIGSEGV, dumping core", server_conf);
> > > > chdir(server_root);
> > > > abort();
> > > > exit(1);
> > > >--- 1088,1095 ----
> > > > }
> > > >
> > > > void seg_fault() {
> > > >! write (fileno(server_conf->error_log),
> > > >! "httpd: caught SIGSEGV, dumping core\n", 37);
> > > > chdir(server_root);
> > > > abort();
> > > > exit(1);
> > > >
> > > >
> > >
> > >
> >
> > --
> > Ben Laurie Phone: +44 (181) 994 6435 Email: ben@algroup.co.uk
> > Freelance Consultant and Fax: +44 (181) 994 6472
> > Technical Director URL: http://www.algroup.co.uk/Apache-SSL
> > A.L. Digital Ltd, Apache Group member (http://www.apache.org)
> > London, England. Apache-SSL author
>
>
>
--
Paul Richards. Originative Solutions Ltd. (Netcraft Ltd. contractor)
Elsevier Science TIS online journal project.
Email: p.richards@elsevier.co.uk
Phone: 0370 462071 (Mobile), +44 (0)1865 843155
Re: WWW Form Bug Report: "signal handlers use non-rentrant functions"
on OTHER:NetBSD
Posted by Paul Richards <p....@elsevier.co.uk>.
Randy Terbush <ra...@zyzzyva.com> writes:
> > All right! This is probably the cause of all the mysterious spins we've been
> > seeing. I think I'd prefer a proper fix, though. Could log_error be made safe
> > instead?
> >
> > Cheers,
> >
> > Ben.
>
> I took a stab at this. The question is how many other places need this
> sort of change? sprintf is considered safe on Solaris.
>
Geeze, I raised this some 6 months ago when I pointed out that Apache
tries to dump core into a specific directory but either people didn't
understand the problem or they were just ignoring me :-).
The signal handling is all to hell, we trap SIGSEGV, call log_error,
do a chdir and then call sigabrt!
Guys, that's total nonsense, you get a SIGSEGV and there's no way to
know why, there's no guarantee anything will work from that point on.
I don't see any reason to trap SIGSEGV at all (nor the other fatal
signals). If the only purpose is to have a nice message in the logs
then it's not worth the problems it may cause.
--
Paul Richards. Originative Solutions Ltd. (Netcraft Ltd. contractor)
Elsevier Science TIS online journal project.
Email: p.richards@elsevier.co.uk
Phone: 0370 462071 (Mobile), +44 (0)1865 843155