Hide Forgot
Created attachment 1428700 [details] vmcore dmesg Description of problem: systemd (pid 1) crashes after sending two SIGBUS or SIGFPE. Version-Release number of selected component (if applicable): [root@fastvm-rhel-7-2-25 ~]# rpm -q systemd systemd-219-19.el7.x86_64 [root@fastvm-rhel-7-2-25 ~]# /usr/lib/systemd/systemd --version systemd 219 +PAM +AUDIT +SELINUX +IMA -APPARMOR +SMACK +SYSVINIT +UTMP +LIBCRYPTSETUP +GCRYPT +GNUTLS +ACL +XZ -LZ4 -SECCOMP +BLKID +ELFUTILS +KMOD +IDN How reproducible: everytime Steps to Reproduce: 1. boot system 2. kill -7 1 3. kill -7 1 Actual results: systemd crashes which leads to crash of the system Expected results: no crash Additional info: [ 2232.472981] systemd[1]: segfault at 7fff7d20cff0 ip 00007f095e63b1a7 sp 00007fff7d20cee0 error 6 in libc-2.17.so[7f095e5f4000+1b8000] [ 2232.475325] Kernel panic - not syncing: Attempted to kill init! exitcode=0x0000000b
(In reply to Lukas Herbolt from comment #0) > Expected results: > no crash How do you expect no crash after sending SIGBUS? The expected behavior on a linux kernel is to crash with a core dump.
[ 2232.475325] Kernel panic - not syncing: Attempted to kill init! exitcode=0x0000000b Above shouldn't happen. Instead, systemd should invoke crash handling routine in which it forks off a new process which (optionally) stores coredump and PID1 then enters a state in which does nothing except handling zombies (while (true) { wait() }). PID1 must never call exit() or exit_group().
(In reply to Jan Synacek from comment #2) > (In reply to Lukas Herbolt from comment #0) > > Expected results: > > no crash > > How do you expect no crash after sending SIGBUS? The expected behavior on a > linux kernel is to crash with a core dump. When the pid 1 starts it sets a crash handler for signals: SIGSEGV,SIGILL,SIGFPE,SIGBUS,SIGQUIT,SIGABRT After sending one of this it goes to freeze() where if it just waits for zombie process or ges to pause() and from there the original signal handler is no longer takes care about the signals. > [ 2232.475325] Kernel panic - not syncing: Attempted to kill init! exitcode=0x0000000b > Above shouldn't happen. Instead, systemd should invoke crash handling routine > in which it forks off a new process which (optionally) stores coredump and > PID1 then enters a state in which does nothing except handling zombies (while > (true) { wait() }). Yes, this is what is happening now. If I am reading the code correctly during the coredump handling the .sa_handler is set to nop_handler. So after the second SIGBUS or other from above mentioned signals the pid 1 dies. Which ends with crashed kernel. I am not saying it is wrong behavior, but if it is desired, it should be in the documentation.
After evaluating this issue, there are no plans to address it further or fix it in an upcoming release. Therefore, it is being closed. If plans change such that this issue will be fixed in an upcoming release, then the bug can be reopened.