Appologise for sending my previous email in HTML, this email address was never meant to be used with lists. I resend in plain text. On Wed, Jul 29, 2015 at 5:25 PM, Serge E. Hallyn <serge@xxxxxxxxxx> wrote: >> +Enabling Smack related capabilities (CAP_MAC_ADMIN and >> +CAP_MAC_OVERRIDE) is main goal of Smack namespace, so it can work >> +properly in the container. And those capabilities do work to some >> +extent. In several places where capabilities are checked compatibility >> +with Smack namespace has been introduced. Capabilities are of course >> +limited to operate only on mapped labels. >> + >> +CAP_MAC_OVERRIDE works fully, will allow you to ignore Smack access >> +rules, but only between objects that have labels mapped. So in the >> +example above having this CAP will allow e.g. label2 to write to >> +label1, but will not allow any access to label3. > > (Sorry it took me this long to get to reading, and thanks for working > on this) > > Oh my. All this is not at all what I'd expected :) > > Is there rationale for these decisions? Hm, I guess it really is > following the user_ns design, but the huge difference is that the > user_ns is partitioning an already-enumerated set of kuids. The > smack labels are inherently different. There is a big rationale for this. This is not to make Smack limit how namespace can be created (this can be done separately, no conflict here). This is to make Smack work at all inside a namespace. Please note that contrary to SELinux and AppArmor Smack needs CAP_MAC_ADMIN for it to operate on a normal basis. There is no rule for changing labels. CAP_MAC_ADMIN is always required for this. And you cannot unlock and give this capability just like that to anyone. Like in every namespace you need a level of abstraction to allow an unprivileged namespace to administer something. > In containers, something we'd really like to be able to do is: > > Create a new container. Just run it as label 'c1'. Inside the > container, let the admin install mysql from a package which assigns > type 'mysqld', protecting the rest of the container from mysql. > Without the host admin doing anything. > > Normally the way I think of implementing something like this would be > to allow the host to say "c1 is to be namespaceable." Then on a userns > unshare, if the task is in c1, it gets transitioned into the ns. Then the > container sees c1 as _ (or whatever). It can create 'mysql' which is > actually 'c1.mysql' on the host, and it can create and override rules > to c1.*. Few things here. 1. Such an extension with using prefixes (with two exceptions, see below) could be added to my patches. I even planned to do so (make a prefix for a container and assign it a group of labels, this doesn't conflict with arbitrary mapping). But this was refused by Casey on a basis that by Smack defintion labels have no meaning. So no prefixes. 2. (expcetion #1) Changing any rules in a container has been deemed too insecure at this point. 3. (expcetion #2) About the: "Without the host admin doing anything.". With this namespace you delegate part of CAP_MAC_ADMIN privilege to an unprivileged user (as with any other namespace). There is now way that this will not involve host admin. The way you described it you allow an unprivileged process to change its own label and change labels on a filesystem. This is simply against Smack rules and completely insecure. Even with user_ns if you map several UID you need admin intervention. > > Also, allowing CAP_MAC_OVERRIDE in this way seems overly dangerous. > If there were rules defined by the container, then I'd expect those to > be overrideable - but not all rules pertaining to all labels mapped > into the container. But I guess based on your envisioned usage (where > I assume 'label1' is meant to *only* be used for that container) it > might be ok. CAP_MAC_OVERRIDE is only possible for labels that the admin explicitly mapped. So it's up to him to decide what is dangerous or not. It can only map labels that are not used outside of the container if it wishes to. But the user himself will not be able to explot that without permission from the admin. So sorry this is not what you expected, but it seems that what you expected is simply not feasible. > thanks, > -serge Thanks, Lukasz -- To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html