From: ebiederm@xmission.com (Eric W. Biederman)
To: Oleg Nesterov <oleg@redhat.com>
Cc: "Kirill A. Shutemov" <kirill@shutemov.name>,
Andrew Morton <akpm@linux-foundation.org>,
Kees Cook <keescook@chromium.org>,
David Howells <dhowells@redhat.com>,
linux-kernel@vger.kernel.org,
Peter Zijlstra <peterz@infradead.org>,
Ingo Molnar <mingo@kernel.org>,
"Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>,
Rik van Riel <riel@redhat.com>,
Vladimir Davydov <vdavydov@parallels.com>,
Ricky Zhou <rickyz@chromium.org>, Julien Tinnes <jln@google.com>
Subject: Re: [PATCH 1/2] unshare: Unsharing a thread does not require unsharing a vm
Date: Thu, 13 Aug 2015 10:38:03 -0500 [thread overview]
Message-ID: <87vbcjyzac.fsf@x220.int.ebiederm.org> (raw)
In-Reply-To: <20150813125550.GA13984@redhat.com> (Oleg Nesterov's message of "Thu, 13 Aug 2015 14:55:50 +0200")
Oleg Nesterov <oleg@redhat.com> writes:
> Let me first say that CLONE_SIGHAND must die, I think ;) and perhaps
> even sighand_struct... I am wondering if we can add something like
>
> if ((clone_flags & (CLONE_THREAD | CLONE_SIGHAND)) == CLONE_SIGHAND)
> pr_info("You are crazy, please report this to lkml\n");
>
> into copy_process().
The only way killing CLONE_SIGHAND would be viable would be with a
config option. There are entire generations of linux where libpthreads
used this before CLONE_THREAD was implemented. Now perhaps no one cares
anymore, but there are a lot of historic binairies that used it, even to
the point where I know of at least one user outside of glibc's pthread
implementation.
> On 08/12, Eric W. Biederman wrote:
>>
>> Oleg Nesterov <oleg@redhat.com> writes:
>>
>> > On 08/11, Eric W. Biederman wrote:
>> >> if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
>> >> - /* FIXME: get_task_mm() increments ->mm_users */
>> >> - if (atomic_read(¤t->mm->mm_users) > 1)
>> >> + if (!thread_group_empty(current))
>> >> + return -EINVAL;
>> >> + }
>> >> + if (unshare_flags & CLONE_VM) {
>> >> + if (!current_is_single_threaded())
>> >> return -EINVAL;
>> >> }
>> >
>> > OK, but then you can remove "| CLONE_VM" from the previous check...
>>
>> As an optimization, but I don't think anything cares enough for the
>> optimization to be worth the confusion.
>
> current_is_single_threaded() checks task->signal->live at the start,
> so there is no optimization. But I won't argue, this doesn't hurt.
>> >> /*
>> >> + * If unsharing a signal handlers, must also unshare the signal queues.
>> >> + */
>> >> + if (unshare_flags & CLONE_SIGHAND)
>> >> + unshare_flags |= CLONE_THREAD;
>> >
>> > This looks unnecessary, check_unshare_flags() checks "THREAD | SIGHAND".
>> > And to me the comment looks misleading although I won't argue.
>>
>> I absolutely can not understand this code if we jump 5 steps ahead
>> and optimize out the individual dependencies, and try for a flattened
>> dependency tree instead. I can validate the individual dependencies
>> from first principles.
>>
>> If we jump several steps ahead I can not validate the individual
>> dependencies.
>
> OK,
>
>> > And in fact this doesn't look exactly right, or I am totally confused.
>> > Shouldn't we do
>> >
>> > if (unshare_flags & CLONE_SIGHAND)
>> > unshare_flags |= CLONE_VM;
>>
>> Nope. The backward definitions of the flags in unshare has gotten you.
>
> See below,
>
>> CLONE_SIGHAND means that you want a struct sighand_struct with a count
>> of 1.
>
> This is (almost) true,
>
>> Nothing about a sighand_struct with a count of 1 implies or
>> requires mm_users == 1. clone can quite happily create those.
>
> See
>
> if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
>
> in copy_process(). So if you have a shared sighand_struct, your ->mm
> is also shared, current_is_single_threaded() will notice this.
Yes. A shared sighand_struct will have a shared ->mm. But a private
sighand_struct with count == 1 may also have a shared ->mm.
>> > Otherwise suppose that a single threaded process does clone(VM | SIGHAND)
>> > and (say) child does sys_unshare(SIGHAND). This will wrongly succeed
>> > afaics.
>>
>> Why would it be wrong to succeed in that case? struct sighand_struct
>> has a count of 1.
>
> How that? clone(VM | SIGHAND) will share ->sighand and increment its
> count.
>
>> unshare(CLONE_SIGHAND) requests a sighand_struct with
>> a count of 1.
>
> Exactly, that is why it is wrong to succeed.
Now that I am clear about what you are talking about I agree with you.
My apologies I clearly misread what you were saying yesterday.
>> unshare(SIGHAND) needs to guarantee that when it returns sighand->count == 1.
>> So unshare(SIGHAND) needs to test for sighand->count == 1.
>
> Oh, I do not think we should check sighand->count. This can lead to
> the same problem we have with the current current->mm->mm_users check.
>
> Most probably today nobody increments sighand->count (I didn't even
> try to verify). But this is possible, and I saw the code which did
> this to pin ->sighand...
I have verified that copy_sighand is the only place in the kernel where
we increment sighand->count today. de_thread in fs/exec.c even seems to
rely on that.
So while I agree with you that the sighand->count could suffer a similar
fate as mm_users it does not.
Eric
next prev parent reply other threads:[~2015-08-13 15:44 UTC|newest]
Thread overview: 41+ messages / expand[flat|nested] mbox.gz Atom feed top
2015-07-28 17:15 [PATCH] user_ns: use correct check for single-threadedness Kees Cook
2015-07-28 18:02 ` Rik van Riel
2015-07-28 18:17 ` Eric W. Biederman
2015-07-28 20:55 ` Ricky Zhou
2015-07-28 21:01 ` Kees Cook
2015-08-05 18:13 ` Eric W. Biederman
2015-08-05 19:40 ` Kees Cook
2015-07-28 21:35 ` Andrew Morton
2015-07-28 21:50 ` Kees Cook
2015-07-28 22:11 ` Kirill A. Shutemov
2015-08-05 11:38 ` Ingo Molnar
2015-08-05 11:53 ` Kirill A. Shutemov
2015-08-05 13:13 ` Ricky Zhou
2015-08-05 17:23 ` Oleg Nesterov
2015-08-05 18:00 ` Eric W. Biederman
2015-08-05 18:52 ` Eric W. Biederman
2015-08-06 13:06 ` Oleg Nesterov
2015-08-06 13:44 ` Oleg Nesterov
2015-08-12 1:17 ` Eric W. Biederman
2015-08-12 14:40 ` Oleg Nesterov
2015-08-12 15:11 ` Eric W. Biederman
2015-08-12 1:22 ` [PATCH 0/2] userns: Creation logic fixes Eric W. Biederman
2015-08-12 1:24 ` [PATCH 1/2] unshare: Unsharing a thread does not require unsharing a vm Eric W. Biederman
2015-08-12 17:48 ` Oleg Nesterov
2015-08-12 18:39 ` Eric W. Biederman
2015-08-13 12:55 ` Oleg Nesterov
2015-08-13 15:38 ` Eric W. Biederman [this message]
2015-08-13 16:17 ` Oleg Nesterov
2015-08-13 16:27 ` Eric W. Biederman
2015-08-13 16:50 ` Oleg Nesterov
2015-08-14 17:59 ` Oleg Nesterov
2015-08-12 19:59 ` [PATCH v2] " Eric W. Biederman
2015-08-13 12:57 ` Oleg Nesterov
2015-08-13 16:01 ` Eric W. Biederman
2015-08-13 16:30 ` Oleg Nesterov
2015-08-13 16:39 ` Eric W. Biederman
2015-08-12 1:25 ` [PATCH 2/2] userns,pidns: Force thread group sharing, not signal handler sharing Eric W. Biederman
2015-08-12 17:24 ` Oleg Nesterov
2015-08-12 6:29 ` [PATCH 0/2] userns: Creation logic fixes Kees Cook
2015-08-06 14:35 ` [PATCH] user_ns: use correct check for single-threadedness Oleg Nesterov
2015-08-06 21:16 ` Eric W. Biederman
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=87vbcjyzac.fsf@x220.int.ebiederm.org \
--to=ebiederm@xmission.com \
--cc=akpm@linux-foundation.org \
--cc=dhowells@redhat.com \
--cc=jln@google.com \
--cc=keescook@chromium.org \
--cc=kirill.shutemov@linux.intel.com \
--cc=kirill@shutemov.name \
--cc=linux-kernel@vger.kernel.org \
--cc=mingo@kernel.org \
--cc=oleg@redhat.com \
--cc=peterz@infradead.org \
--cc=rickyz@chromium.org \
--cc=riel@redhat.com \
--cc=vdavydov@parallels.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).