For review: user_namespace(7) man page
luto at amacapital.net
Mon Sep 1 20:57:33 UTC 2014
On Wed, Aug 20, 2014 at 4:36 PM, Michael Kerrisk (man-pages)
<mtk.manpages at gmail.com> wrote:
> Hello Eric et al.,
> For various reasons, my work on the namespaces man pages
> fell off the table a while back. Nevertheless, the pages have
> been close to completion for a while now, and I recently restarted,
> in an effort to finish them. As you also noted to me f2f, there have
> been recently been some small namespace changes that you may affect
> the content of the pages. Therefore, I'll take the opportunity to
> send the namespace-related pages out for further (final?) review.
> So, here, I start with the user_namespaces(7) page, which is shown
> in rendered form below, with source attached to this mail. I'll
> send various other pages in follow-on mails.
> Review comments/suggestions for improvements / bug fixes welcome.
> user_namespaces - overview of Linux user_namespaces
> For an overview of namespaces, see namespaces(7).
> User namespaces isolate security-related identifiers and
> attributes, in particular, user IDs and group IDs (see creden‐
> tials(7), the root directory, keys (see keyctl(2)), and capabili‐
Putting "root directory" here is odd -- that's really part of a
different namespace. But user namespaces sort of isolate the other
namespaces from each other.
Also, ugh, keys. How did keyctl(2) ever make it through any kind of review?
> ties (see capabilities(7)). A process's user and group IDs can
> be different inside and outside a user namespace. In particular,
> a process can have a normal unprivileged user ID outside a user
> namespace while at the same time having a user ID of 0 inside the
> namespace; in other words, the process has full privileges for
> operations inside the user namespace, but is unprivileged for
> operations outside the namespace.
> Nested namespaces, namespace membership
> User namespaces can be nested; that is, each user namespace—
> except the initial ("root") namespace—has a parent user names‐
> pace, and can have zero or more child user namespaces. The par‐
> ent user namespace is the user namespace of the process that cre‐
> ates the user namespace via a call to unshare(2) or clone(2) with
> the CLONE_NEWUSER flag.
> The kernel imposes (since version 3.11) a limit of 32 nested lev‐
> els of user namespaces. Calls to unshare(2) or clone(2) that
> would cause this limit to be exceeded fail with the error EUSERS.
> Each process is a member of exactly one user namespace. A
> process created via fork(2) or clone(2) without the CLONE_NEWUSER
> flag is a member of the same user namespace as its parent. A
> process can join another user namespace with setns(2) if it has
> the CAP_SYS_ADMIN in that namespace; upon doing so, it gains a
> full set of capabilities in that namespace.
> A call to clone(2) or unshare(2) with the CLONE_NEWUSER flag
> makes the new child process (for clone(2)) or the caller (for
> unshare(2)) a member of the new user namespace created by the
> The child process created by clone(2) with the CLONE_NEWUSER flag
> starts out with a complete set of capabilities in the new user
> namespace. Likewise, a process that creates a new user namespace
> using unshare(2) or joins an existing user namespace using
> setns(2) gains a full set of capabilities in that namespace. On
> the other hand, that process has no capabilities in the parent
> (in the case of clone(2)) or previous (in the case of unshare(2)
> and setns(2)) user namespace, even if the new namespace is cre‐
> ated or joined by the root user (i.e., a process with user ID 0
> in the root namespace).
> Note that a call to execve(2) will cause a process to lose any
> capabilities that it has, unless it has a user ID of 0 within the
Or unless file capabilities have a non-empty inheritable mask.
It may be worth mentioning that execve in a user namespace works
exactly like execve outside a userns.
> $ cat /proc/$$/uid_map
> 0 0 4294967295
> This mapping tells us that the range starting at user ID 0 in
> this namespace maps to a range starting at 0 in the (nonexistent)
> parent namespace, and the length of the range is the largest
> 32-bit unsigned integer.
> Defining user and group ID mappings: writing to uid_map and gid_map
> After the creation of a new user namespace, the uid_map file of
> one of the processes in the namespace may be written to once to
> define the mapping of user IDs in the new user namespace. An
> attempt to write more than once to a uid_map file in a user
> namespace fails with the error EPERM. Similar rules apply for
> gid_map files.
> The lines written to uid_map (gid_map) must conform to the fol‐
> lowing rules:
> * The three fields must be valid numbers, and the last field
> must be greater than 0.
> * Lines are terminated by newline characters.
> * There is an (arbitrary) limit on the number of lines in the
> file. As at Linux 3.8, the limit is five lines. In addition,
> the number of bytes written to the file must be less than the
> system page size, and the write must be performed at the start
> of the file (i.e., lseek(2) and pwrite(2) can't be used to
> write to nonzero offsets in the file).
> * The range of user IDs (group IDs) specified in each line can‐
> not overlap with the ranges in any other lines. In the ini‐
> tial implementation (Linux 3.8), this requirement was satis‐
> fied by a simplistic implementation that imposed the further
> requirement that the values in both field 1 and field 2 of
> successive lines must be in ascending numerical order, which
> prevented some otherwise valid maps from being created. Linux
> 3.9 and later fix this limitation, allowing any valid set of
> nonoverlapping maps.
> * At least one line must be written to the file.
> Writes that violate the above rules fail with the error EINVAL.
> In order for a process to write to the /proc/[pid]/uid_map
> (/proc/[pid]/gid_map) file, all of the following requirements
> must be met:
> 1. The writing process must have the CAP_SETUID (CAP_SETGID)
> capability in the user namespace of the process pid.
This checked for the opening process (and I don't actually remember
whether it's checked for the writing process).
> 2. The writing process must be in either the user namespace of
> the process pid or inside the parent user namespace of the
> process pid.
> 3. The mapped user IDs (group IDs) must in turn have a mapping in
> the parent user namespace.
> 4. One of the following is true:
> * The data written to uid_map (gid_map) consists of a single
> line that maps the writing process's filesystem user ID
> (group ID) in the parent user namespace to a user ID (group
> ID) in the user namespace. The usual case here is that
> this single line provides a mapping for user ID of the
> process that created the namespace.
> * The process has the CAP_SETUID (CAP_SETGID) capability in
> the parent user namespace. Thus, a privileged process can
> make mappings to arbitrary user IDs (group IDs) in the par‐
> ent user namespace.
The opening process.
One other thing that could be worth mentioning it: any non-user
namespace that's created is owned by the user namespace of the process
that created it at the time of creation. Actions on those namespaces
require capabilities in the corresponding user namespace.
Thanks for doing this!
More information about the Containers