Hi,
first off, allow me to express that this is my first time ever writing
on such a mailing list, and that if something is unclear or you would
need more information, just let me know.
I write to this list in hoping to see this change reverted. The linux
kernel always said it would avoid breaking user namespace as much as
possible, and yet this is what happens. I was hence very much surprised
when my perfectly working containers on systemd-nspawn which makes use
of userns by default, stopped working from one day to the next, till I
identified the problem as being kernel >= 4.18. This container is in
production, hence the annoyance it was. From one day to the next the
container started failing with stranges problems:
* nginx, dovecot, postgresql, and postfix complained about getting
permission denied on /dev/null even though it appeared perfectly normal
to me, the correct permissions, all that
* /var was also acting very strangely, getting a lot of permission
denied or operation not supported messages.
* I could not delete a file that my user had the right to create, write
to and read in /var, I needed root
Here is the pull request that was made to systemd, along with a small
amount of talk around the issue:
https://github.com/systemd/systemd/pull/9483
It was ultimately decided among the systemd folks to bail out of the
issue, as shown in the news entry for systemd 240:
* KERNEL API BREAKAGE: Linux kernel 4.18 changed behaviour
regarding
mknod() handling in user namespaces. Previously mknod() would
always
fail with EPERM in user namespaces. Since 4.18 mknod() will
succeed
but device nodes generated that way cannot be opened, and
attempts to
open them result in EPERM. This breaks the "graceful
fallback" logic
in systemd's PrivateDevices= sand-boxing option. This option is
implemented defensively, so that when systemd detects it runs
in a
restricted environment (such as a user namespace, or an
environment
where mknod() is blocked through seccomp or absence of
CAP_SYS_MKNOD)
where device nodes cannot be created the effect of
PrivateDevices= is
bypassed (following the logic that 2nd-level sand-boxing is not
essential if the system systemd runs in is itself already
sand-boxed
as a whole). This logic breaks with 4.18 in container
managers where
user namespacing is used: suddenly PrivateDevices= succeeds
setting
up a private /dev/ file system containing devices nodes — but
when
these are opened they don't work.
At this point is is recommended that container managers utilizing
user namespaces that intend to run systemd in the payload
explicitly
block mknod() with seccomp or similar, so that the graceful
fallback
logic works again.
We are very sorry for the breakage and the requirement to change
container configurations for newer kernels. It's purely
caused by an
incompatible kernel change. The relevant kernel developers
have been
notified about this userspace breakage quickly, but they chose to
ignore it.
Here's an email that was sent to lkml about the subject:
https://lkml.org/lkml/2018/7/5/742
I link also this, quoting the last of it:
https://lkml.org/lkml/2018/7/5/701
It has never been the case that mknod on a device node will guarantee
that you even can open the device node. The applications that regress
are broken. It doesn't mean we shouldn't be bug compatible, but we darn
well should document very clearly the bugs we are being bug compatible with.
I'm in the opinion that it is a kernel bug, and I quote someone from the
systemd irc channel:
ewb said applications were broken. But the rule is, if userspace breaks,
its a bug. The kernel *has* to revert it. And honestly, this change
doesn't make much sense. You can set nodev yourself but then you know
mknod will not allow you to open the object. Here, the kernel does it
without your knowledge
Also, it seems that if this change is reverted, things that were fixed
to work around the issue this breakage caused will not be broken again,
they should simply go back to their previous way of working. I
understand there may be security reason why this change was made in the
first place, but it is not so big a problem is it ? I can mknode
arbitrary devices in userns and open them as userns root. But my point
is, several things broke. My *working* stuff was broken from one day to
the next.
I am not trying to pick a fight. I want to understand the reasoning
behind this change in the first place, and I'm simply making an attempt
at getting it reverted, because it is true that I don't much fancy
blocking the mknode() syscall in every template unit on every machine we
administer here, and that staying on kernel < 4.18 is not a good
sollution either.
I would also like to be personally CC'ed the comments or answers posted
to this mailing list in response to this message.
Thanks
Added some people to CC that might want to see this..
Am Sa., 22. Dez. 2018 um 19:14 Uhr schrieb Ellie Reeves <[email protected]>:
>
> Hi,
> first off, allow me to express that this is my first time ever writing
> on such a mailing list, and that if something is unclear or you would
> need more information, just let me know.
> I write to this list in hoping to see this change reverted. The linux
> kernel always said it would avoid breaking user namespace as much as
> possible, and yet this is what happens. I was hence very much surprised
> when my perfectly working containers on systemd-nspawn which makes use
> of userns by default, stopped working from one day to the next, till I
> identified the problem as being kernel >= 4.18. This container is in
> production, hence the annoyance it was. From one day to the next the
> container started failing with stranges problems:
>
> * nginx, dovecot, postgresql, and postfix complained about getting
> permission denied on /dev/null even though it appeared perfectly normal
> to me, the correct permissions, all that
> * /var was also acting very strangely, getting a lot of permission
> denied or operation not supported messages.
> * I could not delete a file that my user had the right to create, write
> to and read in /var, I needed root
>
> Here is the pull request that was made to systemd, along with a small
> amount of talk around the issue:
>
> https://github.com/systemd/systemd/pull/9483
>
> It was ultimately decided among the systemd folks to bail out of the
> issue, as shown in the news entry for systemd 240:
>
> * KERNEL API BREAKAGE: Linux kernel 4.18 changed behaviour
> regarding
> mknod() handling in user namespaces. Previously mknod() would
> always
> fail with EPERM in user namespaces. Since 4.18 mknod() will
> succeed
> but device nodes generated that way cannot be opened, and
> attempts to
> open them result in EPERM. This breaks the "graceful
> fallback" logic
> in systemd's PrivateDevices= sand-boxing option. This option is
> implemented defensively, so that when systemd detects it runs
> in a
> restricted environment (such as a user namespace, or an
> environment
> where mknod() is blocked through seccomp or absence of
> CAP_SYS_MKNOD)
> where device nodes cannot be created the effect of
> PrivateDevices= is
> bypassed (following the logic that 2nd-level sand-boxing is not
> essential if the system systemd runs in is itself already
> sand-boxed
> as a whole). This logic breaks with 4.18 in container
> managers where
> user namespacing is used: suddenly PrivateDevices= succeeds
> setting
> up a private /dev/ file system containing devices nodes — but
> when
> these are opened they don't work.
>
> At this point is is recommended that container managers utilizing
> user namespaces that intend to run systemd in the payload
> explicitly
> block mknod() with seccomp or similar, so that the graceful
> fallback
> logic works again.
>
> We are very sorry for the breakage and the requirement to change
> container configurations for newer kernels. It's purely
> caused by an
> incompatible kernel change. The relevant kernel developers
> have been
> notified about this userspace breakage quickly, but they chose to
> ignore it.
>
> Here's an email that was sent to lkml about the subject:
>
> https://lkml.org/lkml/2018/7/5/742
>
> I link also this, quoting the last of it:
>
> https://lkml.org/lkml/2018/7/5/701
>
> It has never been the case that mknod on a device node will guarantee
> that you even can open the device node. The applications that regress
> are broken. It doesn't mean we shouldn't be bug compatible, but we darn
> well should document very clearly the bugs we are being bug compatible with.
>
> I'm in the opinion that it is a kernel bug, and I quote someone from the
> systemd irc channel:
>
> ewb said applications were broken. But the rule is, if userspace breaks,
> its a bug. The kernel *has* to revert it. And honestly, this change
> doesn't make much sense. You can set nodev yourself but then you know
> mknod will not allow you to open the object. Here, the kernel does it
> without your knowledge
>
> Also, it seems that if this change is reverted, things that were fixed
> to work around the issue this breakage caused will not be broken again,
> they should simply go back to their previous way of working. I
> understand there may be security reason why this change was made in the
> first place, but it is not so big a problem is it ? I can mknode
> arbitrary devices in userns and open them as userns root. But my point
> is, several things broke. My *working* stuff was broken from one day to
> the next.
>
> I am not trying to pick a fight. I want to understand the reasoning
> behind this change in the first place, and I'm simply making an attempt
> at getting it reverted, because it is true that I don't much fancy
> blocking the mknode() syscall in every template unit on every machine we
> administer here, and that staying on kernel < 4.18 is not a good
> sollution either.
>
> I would also like to be personally CC'ed the comments or answers posted
> to this mailing list in response to this message.
>
> Thanks
Eric, this is entirely unacceptable.
On Sat, Dec 22, 2018 at 12:58 PM Gabriel C <[email protected]> wrote:
>
> Added some people to CC that might want to see this..
Thanks.
> > Here's an email that was sent to lkml about the subject:
> >
> > https://lkml.org/lkml/2018/7/5/742
> >
> > I link also this, quoting the last of it:
> >
> > https://lkml.org/lkml/2018/7/5/701
> >
> > It has never been the case that mknod on a device node will guarantee
> > that you even can open the device node. The applications that regress
> > are broken. It doesn't mean we shouldn't be bug compatible, but we darn
> > well should document very clearly the bugs we are being bug compatible with.
Yeah, this is complete garbage.
We have very clear rules in the kernel: if some change breaks existing
setups, it is ABSOLUTELY NEVER the application that is broken.
It is the kernel.
There is absolutely zero gray areas here. Eric, your behavior is
entirely out of line, and now we apparently have a regression that
goes back to June that I was not told about because of your incorrect
stance.
Eric, I want to make this 1000% clear: there are no user space bugs.
If it used to work, then user space was clearly doing the right thing.
The fact that you tried to several times claim it was buggy user space
is a serious breach of trust. You KNOW this is the case.
Seriously. There are no excuses.
That commit is now reverted in my tree, and furthermore I will not
take any pull requests from you until you have made it clear that you
comprehend this very fundamental issue.
Why did it take so long for this issue to be elevated to me?
Linus
On Sun, Dec 23, 2018 at 12:02 AM Linus Torvalds
<[email protected]> wrote:
>
> On Sat, Dec 22, 2018 at 2:49 PM Christian Brauner
> <[email protected]> wrote:
> >
> > To be fair, no one apart from me was pointing out that it actually
> > breaks people including systemd folks
> > even though I was bringing it up with them. I even tried to fix all of
> > userspace after this got NACKED
>
> Seriously, the "we don't break user space" is the #1 rule in the
> kernel, and people should _know_ it's the #1 rule.
>
> If somebody ignores that rule, it needs to be escalated to me.
> Immediately. Because I need to know.
Fair enough. I usually try to be very conservative when sending patches
directly your way and Eric is otherwise very much on top of not regressing
userspace and I trust him.
However, for this case should I resend the revert?
Christian
>
> I need to know so that I can override the bogus NAK, and so that we
> can fix the breakage ASAP. The absolute last thing we need is some
> other user space then starting to rely on the new behavior, which just
> compounds the problem and makes it a *much* bigger problem.
>
> But I also need to know so that I can then make sure I know not to
> trust the person who broke rule #1.
>
> This is not some odd corner case for the kernel. This is literally the
> rule we have lived with for *decades*.
>
> So please escalate to me whenever you feel a kernel developer doesn't
> follow the first rule. Because the code that broke things *will* be
> reverted (*).
>
> Linus
>
> (*) Yes, there are exceptions. We have had situations where some
> interface was simply just a huge security issue or had some other
> fundamental issue. And we've had cases where the breakage was just so
> old that it was no longer fixable. So even rule #1 can sometimes have
> things that hold it back. But it is *very* rare. Certainly nothing
> like this.
On Sat, Dec 22, 2018 at 11:20 PM Linus Torvalds
<[email protected]> wrote:
>
> Eric, this is entirely unacceptable.
i would like to point out that I send a revert for this in *July*
before any kernel with this change
was released for the exact same reason. But I was ignored and no one
came to argumentative aid:
- https://lists.linuxfoundation.org/pipermail/containers/2018-July/039182.html
- https://lists.linuxfoundation.org/pipermail/containers/2018-July/039183.html
To be fair, no one apart from me was pointing out that it actually
breaks people including systemd folks
even though I was bringing it up with them. I even tried to fix all of
userspace after this got NACKED
( https://github.com/systemd/systemd/pull/9483 ).
Christian
>
> On Sat, Dec 22, 2018 at 12:58 PM Gabriel C <[email protected]> wrote:
> >
> > Added some people to CC that might want to see this..
>
> Thanks.
>
> > > Here's an email that was sent to lkml about the subject:
> > >
> > > https://lkml.org/lkml/2018/7/5/742
> > >
> > > I link also this, quoting the last of it:
> > >
> > > https://lkml.org/lkml/2018/7/5/701
> > >
> > > It has never been the case that mknod on a device node will guarantee
> > > that you even can open the device node. The applications that regress
> > > are broken. It doesn't mean we shouldn't be bug compatible, but we darn
> > > well should document very clearly the bugs we are being bug compatible with.
>
> Yeah, this is complete garbage.
>
> We have very clear rules in the kernel: if some change breaks existing
> setups, it is ABSOLUTELY NEVER the application that is broken.
>
> It is the kernel.
>
> There is absolutely zero gray areas here. Eric, your behavior is
> entirely out of line, and now we apparently have a regression that
> goes back to June that I was not told about because of your incorrect
> stance.
>
> Eric, I want to make this 1000% clear: there are no user space bugs.
> If it used to work, then user space was clearly doing the right thing.
> The fact that you tried to several times claim it was buggy user space
> is a serious breach of trust. You KNOW this is the case.
>
> Seriously. There are no excuses.
>
> That commit is now reverted in my tree, and furthermore I will not
> take any pull requests from you until you have made it clear that you
> comprehend this very fundamental issue.
>
> Why did it take so long for this issue to be elevated to me?
>
> Linus
On Sat, Dec 22, 2018 at 3:07 PM Christian Brauner
<[email protected]> wrote:
>
> However, for this case should I resend the revert?
Since I was pointed at the original email thread, I just picked it up
from there directly. It still applied cleanly, nothing had changed in
that area.
Linus
Am So., 23. Dez. 2018 um 00:02 Uhr schrieb Linus Torvalds
<[email protected]>:
>
> On Sat, Dec 22, 2018 at 2:49 PM Christian Brauner
> <[email protected]> wrote:
> >
> > To be fair, no one apart from me was pointing out that it actually
> > breaks people including systemd folks
> > even though I was bringing it up with them. I even tried to fix all of
> > userspace after this got NACKED
>
> Seriously, the "we don't break user space" is the #1 rule in the
> kernel, and people should _know_ it's the #1 rule.
>
> If somebody ignores that rule, it needs to be escalated to me.
> Immediately. Because I need to know.
>
I do that usually but I didn't saw Christian's revert the time and I
never hit that issue.
Just saw that now because the unusual [BREAKAGE] prefix.
> I need to know so that I can override the bogus NAK, and so that we
> can fix the breakage ASAP. The absolute last thing we need is some
> other user space then starting to rely on the new behavior, which just
> compounds the problem and makes it a *much* bigger problem.
>
Yes and you are right ..
https://github.com/lxc/lxc/pull/2438
I've added an comment there about 4.20.0.
BR,
Gabriel
On Sat, Dec 22, 2018 at 2:49 PM Christian Brauner
<[email protected]> wrote:
>
> To be fair, no one apart from me was pointing out that it actually
> breaks people including systemd folks
> even though I was bringing it up with them. I even tried to fix all of
> userspace after this got NACKED
Seriously, the "we don't break user space" is the #1 rule in the
kernel, and people should _know_ it's the #1 rule.
If somebody ignores that rule, it needs to be escalated to me.
Immediately. Because I need to know.
I need to know so that I can override the bogus NAK, and so that we
can fix the breakage ASAP. The absolute last thing we need is some
other user space then starting to rely on the new behavior, which just
compounds the problem and makes it a *much* bigger problem.
But I also need to know so that I can then make sure I know not to
trust the person who broke rule #1.
This is not some odd corner case for the kernel. This is literally the
rule we have lived with for *decades*.
So please escalate to me whenever you feel a kernel developer doesn't
follow the first rule. Because the code that broke things *will* be
reverted (*).
Linus
(*) Yes, there are exceptions. We have had situations where some
interface was simply just a huge security issue or had some other
fundamental issue. And we've had cases where the breakage was just so
old that it was no longer fixable. So even rule #1 can sometimes have
things that hold it back. But it is *very* rare. Certainly nothing
like this.
Hi,
I would like to thank you all for reacting to this issue so quickly, and
I am really sorry for sending the message several time. I thought there
was a problem with the way it was formatted or some such, hence why I
sent it several times, because none of the messages seemed to get through.
So yeah, real sorry about that bit, and thanking you all
Gabriel C a écrit :
> Am So., 23. Dez. 2018 um 00:02 Uhr schrieb Linus Torvalds
> <[email protected]>:
>>
>> On Sat, Dec 22, 2018 at 2:49 PM Christian Brauner
>> <[email protected]> wrote:
>>>
>>> To be fair, no one apart from me was pointing out that it actually
>>> breaks people including systemd folks
>>> even though I was bringing it up with them. I even tried to fix all of
>>> userspace after this got NACKED
>>
>> Seriously, the "we don't break user space" is the #1 rule in the
>> kernel, and people should _know_ it's the #1 rule.
>>
>> If somebody ignores that rule, it needs to be escalated to me.
>> Immediately. Because I need to know.
>>
>
> I do that usually but I didn't saw Christian's revert the time and I
> never hit that issue.
> Just saw that now because the unusual [BREAKAGE] prefix.
>
>> I need to know so that I can override the bogus NAK, and so that we
>> can fix the breakage ASAP. The absolute last thing we need is some
>> other user space then starting to rely on the new behavior, which just
>> compounds the problem and makes it a *much* bigger problem.
>>
>
> Yes and you are right ..
> https://github.com/lxc/lxc/pull/2438
>
> I've added an comment there about 4.20.0.
>
> BR,
>
> Gabriel
>
Den 23-12-2018 kl. 01:28, skrev Linus Torvalds:
> On Sat, Dec 22, 2018 at 3:07 PM Christian Brauner
> <[email protected]> wrote:
>>
>> However, for this case should I resend the revert?
>
> Since I was pointed at the original email thread, I just picked it up
> from there directly. It still applied cleanly, nothing had changed in
> that area.
>
> Linus
>
This should also be picked up for 4.19 lts
Greg, it's now upstream as:
From 94f82008ce30e2624537d240d64ce718255e0b80 Mon Sep 17 00:00:00 2001
From: Christian Brauner <[email protected]>
Date: Thu, 5 Jul 2018 17:51:20 +0200
Subject: Revert "vfs: Allow userns root to call mknod on owned filesystems."
--
Thomas
On Sun, Dec 23, 2018 at 12:54:11PM +0200, Thomas Backlund wrote:
> Den 23-12-2018 kl. 01:28, skrev Linus Torvalds:
> > On Sat, Dec 22, 2018 at 3:07 PM Christian Brauner
> > <[email protected]> wrote:
> > >
> > > However, for this case should I resend the revert?
> >
> > Since I was pointed at the original email thread, I just picked it up
> > from there directly. It still applied cleanly, nothing had changed in
> > that area.
> >
> > Linus
> >
>
> This should also be picked up for 4.19 lts
>
> Greg, it's now upstream as:
>
> From 94f82008ce30e2624537d240d64ce718255e0b80 Mon Sep 17 00:00:00 2001
> From: Christian Brauner <[email protected]>
> Date: Thu, 5 Jul 2018 17:51:20 +0200
> Subject: Revert "vfs: Allow userns root to call mknod on owned filesystems."
Now queued up, thanks.
greg k-h