Bug #71482 [Asn]: Lost zend_string after "reached pm.max_children setting" with segfault

Edit report at https://bugs.php.net/bug.php?id=71482&edit=1 ID: 71482 Updated by: laruence@php.net Reported by: alex dot schneider at sevenval dot com Summary: Lost zend_string after "reached pm.max_children setting" with segfault Status: Assigned Type: Bug Package: PCRE related Operating System: Linux Mint 17.3 PHP Version: 7.0.2 Assigned To: laruence Block user comment: N Private report: N New Comment: could you please show me the real backtrace? it may be helpful Previous Comments: ------------------------------------------------------------------------ [2016-01-30 05:34:27] alex dot schneider at sevenval dot com Yes, we use, but i'a almost sure that the error comes from module "pcre" self. Is also interesting that there is no longer segfaults, once we disable the "opcache". Is here a big intern difference between the both functions (original and patched)? ------------------------------------------------------------------------ [2016-01-30 04:25:09] laruence@php.net hmm, do you use any other extensions which is not bundled with php ? ------------------------------------------------------------------------ [2016-01-29 09:12:07] alex dot schneider at sevenval dot com Description: ------------ The destruction of "HashTable module_registry" sometimes "throws" a segfault in module "pcre". I can give you the following infos only: * The segfault is "thrown" while the destruction of "HashTable module_registry" by "zend_hash_graceful_reverse_destroy(&module_registry)", but before "PHP_GSHUTDOWN_FUNCTION(pcre)" * The trigger is the "ZEND_API void ZEND_FASTCALL zend_hash_destroy(HashTable *ht)" function in zend_hash.c (line 1272) "if (EXPECTED(p->key)) {" with "Access to address 0xXXXXXXXXX is not allowed" In the meantime while my test suite runs, I see in the log: WARNING: [pool fit] server reached pm.max_children setting (2), consider raising it I think, the bug is related to #63180. My patch (attached here) works for me. Perhaps your experts can examine the differences here. PS: Sorry for my english :) Test script: --------------- Unfortunately, i cannot isolate the trigger code from my very complex test suite with a lot of test methods and requests. fpmlimits.conf for triggering the segfault: ... pm = dynamic pm.max_children = 2 pm.start_servers = 2 pm.min_spare_servers = 2 pm.max_spare_servers = 2 pm.max_requests = 5 ... Expected result: ---------------- No segfaults Actual result: -------------- segfaults ------------------------------------------------------------------------ -- Edit this bug report at https://bugs.php.net/bug.php?id=71482&edit=1