summaryrefslogtreecommitdiff
path: root/miniperlmain.c
diff options
context:
space:
mode:
authorGurusamy Sarathy <gsar@cpan.org>2001-07-20 18:38:48 +0000
committerGurusamy Sarathy <gsar@cpan.org>2001-07-20 18:38:48 +0000
commit52e18b1f277416a33dff2c066a83fdab0520a2d7 (patch)
tree0d35a4f042171ce3c05667667ca19d69d060a292 /miniperlmain.c
parent9046a8ae3aad1f7eda5affd38301ac2313201634 (diff)
downloadperl-52e18b1f277416a33dff2c066a83fdab0520a2d7.tar.gz
Make perl fork()-safe (in a slightly limited way) even on
platforms that don't have pthread_atfork() (extension of the fix in change#11151). Note that this will not help extensions that call fork() directly in C, or that link to libraries that call fork() directly. Such cases must be fixed to either call PerlProc_fork(), or call atfork_lock() in parent before the calling the function that forks and call atfork_unlock() in both parent and child immediately after the fork(). (There are no worries if C code calls exec() in the child immediately after a fork(). Only cases where the child calls perl's API functions (including New()) after the fork() are problematic.) This change also eliminates the use of vfork() from perl, since all such uses were violating the severe restrictions on modifying the state of the process between the vfork() and the exec(). This is a modified version of patches suggested by Abhijit Menon-Sen and Richard Soderberg. p4raw-link: @11151 on //depot/perl: 50dd6e574ff39b609595ddb16b2fe9f625a26f8c p4raw-id: //depot/perl@11423
Diffstat (limited to 'miniperlmain.c')
-rw-r--r--miniperlmain.c11
1 files changed, 10 insertions, 1 deletions
diff --git a/miniperlmain.c b/miniperlmain.c
index 620fed78b2..2c924ebe2e 100644
--- a/miniperlmain.c
+++ b/miniperlmain.c
@@ -48,7 +48,16 @@ main(int argc, char **argv, char **env)
PERL_SYS_INIT3(&argc,&argv,&env);
-#ifdef USE_ITHREADS
+#if defined(USE_THREADS) || defined(USE_ITHREADS)
+ /* XXX Ideally, this should really be happening in perl_alloc() or
+ * perl_construct() to keep libperl.a transparently fork()-safe.
+ * It is currently done here only because Apache/mod_perl have
+ * problems due to lack of a call to cancel pthread_atfork()
+ * handlers when shared objects that contain the handlers may
+ * be dlclose()d. This forces applications that embed perl to
+ * call PTHREAD_ATFORK() explicitly, but if and only if it hasn't
+ * been called at least once before in the current process.
+ * --GSAR 2001-07-20 */
PTHREAD_ATFORK(Perl_atfork_lock,
Perl_atfork_unlock,
Perl_atfork_unlock);