These were introduced as part of the effort of sd-executor
worker pool (#29566), which never landed due to unsignificant
performance improvement. Let's just remove the unused
helpers. If that work ever gets resurrected they can be
restored from this commit pretty easily.
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.
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.
If /proc/self/gid_map is empty, the kernel will refuse setgroups(),
so don't attempt it if that's the case on top of the /proc/self/setgroups
check we already have.
string_has_ansi_sequence() currently does this to look for ESC input:
t = memchr(s, 0x1B, ...)
So each iteration re-searches from the original start pointer. But if we
find an ESC byte that does *not* start a valid ANSI sequence (like "\x1B
", or an ESC at the end of the string), then ansi_sequence_length()
returns 0, and if that ESC is still in the search window, we will just
spin consuming 100% CPU forever.
Fix this by always advancing past rejected ESC bytes.
/dev/console might have no backing driver, in which case
/sys/class/tty/console/active is empty. Unlike get_kernel_consoles()
resolve_dev_console() currently proceeds with empty devnode,
resulting in setup_input() -> acquire_terminal() emitting -EISDIR
as we're trying to open /dev/. Let's catch this and report -ENXIO.
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.
This is generally useful, but is particularly useful in context of
https://github.com/mvo5/varlink-proxy-rs which can expose a set of local
Varlink services via a HTTP bridge. The idea is that the sockets linked
into /run/varlink/registry/ are candidates for being exposed like that.
/cc @mvo5
When a cgroup is selected for termination, send varlink messages
to hooks registered in `/run/systemd/oomd.prekill-hooks/`.
oomd waits up to `PreKillHookTimeoutSec=` seconds for response
before proceeding with the kill.
name_to_handle_at_try_unique_mntid_fid() in its current form is
ill-designed for various reasons:
* AT_HANDLE_FID requires file system support, while unique mount id
is a VFS concept hence is always available if supported. Hence
the fallback for AT_HANDLE_MNT_ID_UNIQUE should be independent
of fid.
* The request for AT_HANDLE_MNT_ID_UNIQUE can be identified via
specifying ret_unique_mnt_id, no need for opening up the control
to caller (and currently the function simply doesn't handle
mismatch between ret params and flags).
* The caller cannot realistically differentiate whether the returned
mount id is actually unique.
* The path_get_unique_mnt_id() fallback did not handle AT_SYMLINK_FOLLOW.
Let's instead move the statx() fallback into name_to_handle_at_loop()
directly, and revamp interaction of ret_mnt_id/ret_unique_mnt_id:
if both are set, it indicates that the caller can handle both, hence
set what we have and return 0/1 for whether we managed to acquire
the unique one.
The !ret_handle && ret_mnt_id logic is removed. Let's not rely on
undocumented bizaare behavior and it's unused anyways.
path_get_mnt_id_at() exists for a reason...
When the extensions for the final system are already set up from the
initrd we should avoid disrupting the boot process with the remount
(which currently isn't atomic) and the daemon reload for
systemd-confext and systemd-sysext. Similarly, when sysupdate ran and
updated extensions it's best to avoid the remount and daemon reload if
no changes are found.
To do this, encode the current extension state in more detail than
before where only the names of the extensions where encoded in the
overlay mount. This can also be used to provide more details about the
extension origin in "systemd-sysext status (--json=)". During the
refresh add a check whether the old state matches the new state and in
this case skip the refresh unless the user provides a flag to always
refresh. Besides the extension name and the resolved path the best
method for identification is the verity hash but that is not available
for plain image files or directories. Therefore, also include data to
check for file/directory replacements. The creation/modification times
are not always real on reproducible images or extracted archive content.
The file handle together with the unique mount ID is the next best
identifier we can use when we have no verity hash. Fall back to an inode
when we get no handle. With the creation/modification time and the path
this should be good enough. Using a unique mount ID is important (with
a fallback to the regular non-unique mount ID) instead of st_dev because
st_dev gets reused too easily, e.g., by a loop device mount and the
mount ID helps to catch this. For the mount ID to be valid it has to be
resolved before we enter the new mount namespace. Thus, it gets provided
by the image dissect logic and handed over to the sysext subprocess
which runs in a new mount namespace.
Luckily, we can rule out online modification of directories or image
files because this is anyway not well supported with overlay mounts, so
we don't do a file checksum nor do we recurse into a directory to look
for the most recently touched files. But, as said, with the
always-refresh flag one can force a reload.
This also effectively reverts eb2e91b981,
as now the object is only used as input, and kernel would need to handle
input with only ID no matter whether the struct later gets extended.
The current code assumes that FILEID_KERNFS filehandles will never grow
in size. This is not a safe assumption and userland shouldn't be trying
to guess the size of the filehandle it'll get.
We have a macro for that: MAX_HANDLE_SZ.
Mike Yuan
Yu Watanabe
Daan De Meyer
andre4ik3
Chris Down
Daan De Meyer
Lennart Poettering
Matteo Croce
Kai Lueke
Jeff Layton