linux-kernel - Re: [PATCH 5/6] coredump: Don't perform any cleanups before dumping core

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:   Fri, 24 Sep 2021 14:41:07 -0700
From:   Kees Cook <keescook@...omium.org>
To:     "Eric W. Biederman" <ebiederm@...ssion.com>
Cc:     linux-kernel@...r.kernel.org,
        Linus Torvalds <torvalds@...ux-foundation.org>,
        Oleg Nesterov <oleg@...hat.com>,
        Al Viro <viro@...iv.linux.org.uk>, linux-api@...r.kernel.org
Subject: Re: [PATCH 5/6] coredump:  Don't perform any cleanups before dumping
 core

On Fri, Sep 24, 2021 at 04:28:58PM -0500, Eric W. Biederman wrote:
> So everything should work from a locking perspective as I am not
> changing the locking I am simply moving the call from exit_mm earlier.
> 
> To explain how the locking works.
> 
> Coredumps are not handled in complete_signal, so when
> a thread dequeues the signal the other threads are all running.
> 
> If two threads dequeue core dumping signals at the same time both will
> contend for mmap_write_lock one will get it and set core_state the
> second will return -EBUSY from coredump_wait and return from do_coredump
> and proceed to do_exit.
> 
> There similar set of races that zap_threads called from coredump_wait
> resolves by testing for signal_group_exit inside of sighand lock.
> 
> The normal case is one thread runs through do_coredump, coredump_wait,
> and zap_threads, counts the threads and waits in coredump_wait for
> all of the other threads to stop.
> 
> The other threads proceed to do_exit and  coredump_task_exit.
> From their the discover that core_state is set.  And holding
> the mmap_read_lock is enough to know ensure that either
> no coredump is in progress or all of the setup is complete in
> coredump_wait.
> 
> If core_state is found it is known that there is a waiter waiting in
> coredump_wait.  So the tasks add themselves to the list.  Decrement
> the count to indicate they don't need to be waited for any longer
> and the last one waits up the waiter in coredump_wait.
> 
> The threads then spin in TASK_UNINTERRUPTLE until the coredump
> completes.
> 
> The dumping thread takes the coredump then calls coredump_finish.
> In coredump_finish the linked list is torn down, and each element
> of the linked list has sets ".task = NULL"  Then the task is woken
> up.
> 
> The task waiting in TASK_UNINTERRUPTIBLE wakes up. Sees that .task =
> NULL and proceeds with the rest of do_exit.
> 
> 
> The only change  this patch makes to that entire process is
> "task->mm = NULL" is replaced by setting PF_POSTCOREDUMP.
> 
> Does that help?

That does! Thanks very much for the run-down on the flow there. I'm
convinced. :)

Reviewed-by: Kees Cook <keescook@...omium.org>

-Kees

-- 
Kees Cook