Al Viro <viro@xxxxxxxxxxxxxxxxxx> writes: > On Thu, Jun 24, 2021 at 01:57:35PM -0500, Eric W. Biederman wrote: > >> So far the code has been lightly tested, and the descriptions of some >> of the patches are a bit light, but I think this shows the direction >> I am aiming to travel for sorting out exit(2) and exit_group(2). > > FWIW, here's the current picture for do_exit(), aside of exit(2) and do_exit_group(): > > 1) stuff that is clearly oops-like - > alpha:die_if_kernel() alpha:do_entUna() alpha:do_page_fault() arm:oops_end() > arm:__do_kernel_fault() arm64:die() arm64:die_kernel_fault() csky:alignment() > csky:die() csky:no_context() h8300:die() h8300:do_page_fault() hexagon:die() > ia64:die() i64:ia64_do_page_fault() m68k:die_if_kernel() m68k:send_fault_sig() > microblaze:die() mips:die() nds32:handle_fpu_exception() nds32:die() > nds32:unhandled_interruption() nds32:unhandled_exceptions() nds32:do_revinsn() > nds32:do_page_fault() nios:die() openrisc:die() openrisc:do_page_fault() > parisc:die_if_kernel() ppc:oops_end() riscv:die() riscv:die_kernel_fault() > s390:die() s390:do_no_context() s390:do_low_address() sh:die() > sparc32:die_if_kernel() sparc32:do_sparc_fault() sparc64:die_if_kernel() > x86:rewind_stack_do_exit() xtensa:die() xtensa:bad_page_fault() > We really do not want ptrace anywhere near any of those and we do not want > any of that to return; this shit would better be handled right there and > there - no "post a fatal signal" would do. Thanks that makes a good start for digging into these. I think the distinction I would make is: - If the kernel is broken use do_task_dead. - Otherwise cleanup the semantics by using start_group_exit, start_task_exit or by just cleaning up the code. Looking at the reboot case it looks like we the code should have become do_group_exit in 2.5. I have a suspicion we have a bunch of similar cases that want to terminate the entire process, but we simply never updated to deal with multi-thread processes. I suspect in the reboot case panic if machine_halt or or machine_power_off fails is more likely the correct handling. But we do have funny semantics sometimes. I will see what I can do to expand my patchset to handle all of these various callers of do_exit. Eric
