When namespace_enter_full() is called with namespaces that the current
process is already a part of, skip the setns() call on the user
namespace. This allows the function to succeed without CAP_SYS_ADMIN.
Follow-up for 211c3c9d41
(cherry picked from commit 27b1fd3a1e2e6c28bb09ec79cb3dc949a4ba416d)
namespace_enter() tried to become root after joining a user namespace
but ignored failures, so if root wasn't mapped we'd just carry on with
our original UID/GID. Let's make that an error instead.
Dropping supplementary groups stays best effort, as setgroups() might
be denied (e.g. by the nsresourced BPF LSM) even when root is mapped.
Some callers do need to enter namespaces without root, so add
NAMESPACE_ENTER_KEEP_UID_GID_IF_ROOT_UNMAPPED for those. It's used by
systemd-analyze unit-shell and by the credential refresh in the user
manager. Everything else enters containers or namespaces we allocated
ourselves, where root is always mapped.
Lennart reminded me in [1] that we need to add assert() in functions
that do pointer access. For the simple `*p` pointer dereferences
we even have an automatic coccinelle script that ensures that as
part of the automatic code checks.
However for deref in the `p->` style this is not supported right
now and adding it to coccinelle is hard because its too slow for
this kind of check. So I created a (slightly messy) tree-sitter
python script to see how many asserts we are currently missing.
This commit is the result of running it over the `src/basic`
dir and fixing the flagged issues. I plan to tidy it up and
add it to the checks too but this is orthogonal to this commit.
[1] https://github.com/systemd/systemd/pull/42360#discussion_r3426964562
The BPF kprobe that fires on user namespace destruction is the only thing
that triggers registry cleanup, so any time it doesn't run — ring buffer
overflow, kprobe missing, fdstore entry dropped outside our cleanup path
— a registry entry is left behind forever.
Stamp each registry entry with the kernel's unique namespace identifier
(NS_GET_ID, kernel ≥ 6.13) at allocation time. At manager startup, after
the existing fdstore→registry sweep, walk the registry and ask the kernel
to look each namespace up by id via open_by_handle_at() on nsfs; if the
lookup returns -ESTALE the namespace is gone and we release the entry.
Old entries written before this change carry no identifier and are left
alone.
Add a namespace_open_by_id() helper for the lookup. The kernel restricts
open_by_handle_at() on nsfs to processes in the initial user namespace,
collapsing both permission denials and dead namespaces onto -ESTALE; the
helper refuses early with -EHOSTDOWN outside the initial user namespace
so callers can tell the two apart.
Let's check if we're already in the right namespaces and call connect()
directly if that's the case. This can easily happen when the machine is
specified as .host or so.
These checks may rely on /proc on older kernels which we could lose access
to by joining namespaces so let's do all the checks first and then join
namespaces.
There's no need to pass in a boolean to decide whether we use
namespace_enter_delegated() or not. Instead, we can just check if we
have CAP_SYS_ADMIN in our own user namespace. If we don't, then we have
to insist on a child user namespace being passed in and we have to enter
it first to get CAP_SYS_ADMIN as without CAP_SYS_ADMIN we wouldn't be able
to call setns() in the first place. If we do have CAP_SYS_ADMIN, we can
always enter the other namespaces first before entering the user namespace.
Additionally, we don't fail anymore if we can't reset the UID/GID since a
root user might not always be available in every user namespace we might
enter.
Typically when entering a namespace the userns is handled last,
because we assume our process is more privileged than the userns.
However, that assumption no longer holds for user managers, which
have no privilege over initial userns and all other namespaces
are actually owned by the userns unshared first (in executor).
Hence, let's add another flavor namespace_enter_delegated() to
accommodate that use case.
While blocking dlopen() in child generally makes sense, it does
also lead to misery - in the very case of extension/cred refreshing
it then enforces loading of libcryptsetup/libacl in pid1, which
are otherwise never used. Let's add a flag to opt out hence, but
still prohibit it in namespace_fork/enter().
These source files uses symbols provided by sched.h, e.g.
setns(), unshare(), CLONE_NEWNS, and friends, but they do not explicitly
include sched.h. Currently, it is included indirectly via missing_syscall.h,
which is included by e.g. pidfd-util.h.
Let's explicitly include headers that provides symbols used in the code.
This is similar to 4f18ff2e29, but for sched.h.
We use symbols provided by unistd.h without including it. E.g.
open(), close(), read(), write(), access(), symlink(), unlink(), rmdir(),
fsync(), syncfs(), lseek(), ftruncate(), fchown(), dup2(), pipe2(),
getuid(), getgid(), gettid(), getppid(), pipe2(), execv(), _exit(),
environ, STDIN_FILENO, STDOUT_FILENO, STDERR_FILENO, F_OK, and their
friends and variants, so on.
Currently, unistd.h is indirectly included mainly in the following two paths:
- through missing_syscall.h, which is planned to covert to .c file.
- through signal.h -> bits/sigstksz.h, which is new since glibc-2.34.
Note, signal.h is included by sd-eevent.h. So, many source files
indirectly include unistd.h if newer glibc is used.
Currently, our baseline on glibc is 2.31. We need to support glibc older
than 2.34, but unfortunately, we do not have any CI environments with
such old glibc. CIFuzz uses glibc-2.31, but it builds only fuzzers, and
many files are even not compiled.
Now that the necessary functions from log.h have been moved to macro.h,
we can stop including log.h in macro.h. This requires modifying source
files all over the tree to include log.h instead.
This is a simple helper for creating a userns that just maps the
callers user to UID 0 in the namespace. This can be acquired unpriv,
which makes it useful for various purposes, for example for the logic in
is_idmapping_supported(), hence port it over.
(is_idmapping_supported() used a different mapping before, with the
nobody users, but there's no real reason for that, and we'll use
userns_acquire_self_root() elsewhere soon, where the root mapping is
important).
Unprivileged namespaces are only allowed if the "setgroups" file is set
to "deny" for processes. And we need to write it before writing the
gidmap. Hence add a parameter for that.
Then, also patch all current users to actually enable this. The usecase
generally don't need it (because they don't care about unprivileged
userns), but it doesn't hurt to enable the concept anyway in all current
users (none of them actually runs complex userspace in them, but they
mostly use userns_acquire() for idmapped mounts and similar).
Let's anyway make this option explicit in the function call, to indicate
that the concept exists and is applied.
To resolve conflict with sys/mount.h and linux/mount.h or linux/fs.h.
The conflict between sys/mount.h and linux/mount.h is resolved in
glibc-2.37 (774058d72942249f71d74e7f2b639f77184160a6), but our baseline
is still glibc-2.31. Also, even with the version or newer, still
sys/mount.h conflicts with linux/fs.h, which is included by
linux/btrfs.h.
This introduces our own implementation of sys/mount.h, that can be
simultaneously included with linux/mount.h and linux/fs.h. This also
imports linux/fs.h, linux/mount.h, and several other dependent headers.
The introduced sys/mount.h header itself may not be enough simple, but
by using the header, we can drop most of workarounds in other source files.
Often we want to fork off a process that just hangs until we kill it,
let's add a simple flag to create one of this type, and use it at
various places.
This makes sure the the codepath that derives an nsfd from a pid works
the same for the pidfd case and the non-pidfd case: if we can verify
that /proc/ is mounted but the /proc/$PID/ns/ files are missing, we can
assume the ns type is not supported by the kernel. Hence return the same
ENOPKG error in this case as we already do in the pidfd ioctl based
codepath.
- Make fd_is_namespace() take NamespaceType
- Drop support for kernel without NS_GET_NSTYPE (< 4.11)
- Port is_our_namespace() to namespace_open_by_type()
(preparation for later commits, where the latter
would go by pidfd if available, avoiding procfs)
[ 10.056930] (journald)[104]: Failed to fork off '(sd-mkuserns)': Invalid argument
[ 10.063727] systemd[1]: systemd-modules-load.service: About to execute: /usr/lib/systemd/systemd-modules-load
[ 10.071148] (journald)[104]: Failed to fork process (sd-mkuserns): Invalid argument
safe_fork_full() already logs at debug level, so the caller shouldn't.
As reported in https://github.com/systemd/systemd/issues/35400,
on riscv64, with Linux version 6.6.51-linux4microchip+fpga-2024.09, we get:
[ 10.063727] systemd[1]: systemd-modules-load.service: About to execute: /usr/lib/systemd/systemd-modules-load
[ 10.071148] (journald)[104]: Failed to fork process (sd-mkuserns): Invalid argument
Fixes https://github.com/systemd/systemd/issues/35400.
'r' is used to make the repeated checks shorter. Without that, the long variable
name is distracting.
This is a more comprehensive fix compared to #35273. Also adds a minimal
test only.
Based on Luca's #35273 but generalizes the code a bit.
In v258 we really should get rid of the old heuristics around userns and
cgroupns detection, but given we are late in the v257 cycle this keeps
them in.
Skye Soss
Daan De Meyer
Yu Watanabe
Michael Vogt
Mike Yuan
Daan De Meyer
Michal Sekletar
Lennart Poettering
Zbigniew Jędrzejewski-Szmek