Re: [libvirt] [PATCH 01/11] Block SIGPIPE around virExec hook functions

On 01/24/2011 08:13 AM, Daniel P. Berrange wrote: Cute. Yikes. We have a potential deadlock problem. See this bug report against GNU sort: http://lists.gnu.org/archive/html/coreutils/2011-01/msg00085.html In sort, any program that mixes pthread_create with fork (and libvirt falls into that category) must obey this section of POSIX: http://pubs.opengroup.org/onlinepubs/9699919799/functions/fork.html "If a multi-threaded process calls fork(), the new process shall contain a replica of the calling thread and its entire address space, possibly including the states of mutexes and other resources. Consequently, to avoid errors, the child process may only execute async-signal-safe operations until such time as one of the exec functions is called." malloc() (which is called by virReportSystemError(), as well as by _()) is NOT async-signal-safe; therefore, it is quite possible that we fork() in one thread while another thread is in the middle of holding the malloc() mutex, and the child process will deadlock because it no longer has a secondary thread available to release the malloc() mutex. Ultimately, we need to refactor and audit the code so that only async-signal-safe functions are allowed between fork() and exec(); which means that virExec needs to be taught how to hand all errors back to the parent over a secondary pipe for the parent to issue (rather than the child attempting to issue any errors on its own). However, that problem is pre-existing; so your patch, while adding another instance of a violation, is not adding a regression, so: Reluctant ACK. -- Eric Blake eblake(a)redhat.com +1-801-349-2682 Libvirt virtualization library http://libvirt.org