On Wed, Nov 8, 2023 at 3:58 AM Dominique Martinet <asmad...@codewreck.org> wrote: > > From: Dominique Martinet <dominique.marti...@atmark-techno.com> > > On alpine linux, calling busybox time with a nonexisting command would > hang forever: > ``` > time: can't execute 'fdsa': No such file or directory > Command exited with non-zero status 127 > real 0m 0.00s > user 0m 0.00s > sys 0m 0.00s > <hang> > ``` > > This is because time uses vfork then BB_EXECVP_or_die, which ultimately > calls exit() after execvp fails. exit() is explicitly listed as unsafe > to call after vfork as the child process shares its memory with the > parent; in particular on musl exit will take the init_fini_lock() and > calling it twice in the same address space will just hang. > _exit() on the other hand is safe to call, so replace die_func with it.
Doesn't happen to me... maybe I have an older musl. > For this particular problem die_func could just be overriden in > time.c's run_command after vfork, but this problem actually came up > before in shell/hush.c as per the fflush_and__exit comment so it seems > more appropriate to do this globally in the xvfork macro and get rid of > this whole class of bugs forever. Maybe we can always use _exit(), or fflush_all(); _exit() in xfunc_die()? _______________________________________________ busybox mailing list busybox@busybox.net http://lists.busybox.net/mailman/listinfo/busybox
