userns.conf: remove obsolete bind-mounts

Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>

conf: NOTICE() on mounts on container's /dev

Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>

utils: rework lxc_deslashify()

Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>

conf: lxc_fill_autodev()

non-functional changes

Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>

utils: switch to has_fs_type()

Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>

utils: add has_fs_type() + is_fs_type()

Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>

conf: mount_autodev()

non-functional changes

Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>

conf: lxchook_names

non-functional changes

Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>

conf: mount_entry()

non-functional changes

Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>

conf: cull_mntent_opt()

non-functional changes

Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>

conf: mount_entry_create_dir_file()

bugfixes

Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>

conf: mount_entry_on_generic()

non-functional changes

Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>

conf: mount_entry_on_systemfs()

non-functional changes

Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>

conf: mount_entry_on_absolute_rootfs()

non-functional changes

Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>

conf: setup_mount_entries()

non-functional changes

Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>

conf: make_anonymous_mount_file()

non-functional changes

Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>

conf: setup_mount()

non-functional changes

Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>

conf: mount_file_entries()

non-functional changes

Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>

Merge pull request #1723 from brauner/2017-07-31/remove_utmp_watch

start: remove utmp watch

Merge pull request #1722 from brauner/2017-07-31/devpts_use_max_mount_option

devpts: use max=<count> option on mount

Merge pull request #1724 from GamerSource/opensuse-template-improvements-v2

Opensuse template improvements

devpts: use max=<count> option on mount

This will only work with kernels >= 3.4

Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>

start: remove utmp watch

Closes #1616.

Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>

templates/opensuse: support leap 42.3

Signed-off-by: Thomas Lamprecht <t.lamprecht@proxmox.com>

templates/opensuse: getty.target.wants does not always exists

Signed-off-by: Thomas Lamprecht <t.lamprecht@proxmox.com>

templates/opensuse: fix tumbleweed software selection

tumbleweed needs a newer version for a few packages, namely:
* liblua5_3
* bncurses6
* breadline7

As Leap it also has no dhcpd, but udhcp which gets pulled in already.
Further iproute2 and net-tools need manual instalation.

Signed-off-by: Thomas Lamprecht <t.lamprecht@proxmox.com>

templates/opensuse: tumbleweed has no update repo

As tumbleweed is a rolling release it has not update repository.

Signed-off-by: Thomas Lamprecht <t.lamprecht@proxmox.com>

Merge pull request #1721 from brauner/2017-07-30/fix_saved_config_handling

network parser fixes

test: test for invalid network key

Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>

Merge pull request #1720 from hallyn/2017-07-29/cg

Fix some bugs in #1719

lxccontainer: clear whole indexed networks

Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>

confile: use deindexed network keys

When we are passed a network key like "lxc.net.[i].ipv4.address" we need to
make sure that we pass the deindexed key "lxc.net.ipv4.address" to the
{get,clr,set} methods otherwise we'll end up in an endless loop.

Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>

Fix some bugs in #1719

In particular, any text (like a #comment) following the mode
would end up in our mode line.

Signed-off-by: Serge Hallyn <serge@hallyn.com>

Merge pull request #1719 from hallyn/2017-07-28/cg

Fix up c8bf519d to fit our coding style

Fix up c8bf519d to fit our coding style

Also reduce the number of mallocs.

Signed-off-by: Serge Hallyn <serge@hallyn.com>

Merge pull request #1708 from aeris/master

Allow full path in lxc.devices.allow

Merge pull request #1717 from hallyn/harri1

Use "rsync -SHaAX" to copy the cached rootfs into place

Use "rsync -SHaAX" to copy the cached rootfs into place

(updated by Serge to also handle hte new lxc-fedora{-legacy{.in
templates)

Signed-off-by: Harald Dunkel <harri@afaics.de>
Signed-off-by: Serge Hallyn <serge@hallyn.com>
Acked-by: Serge Hallyn <serge@hallyn.com>

Merge pull request #1716 from brauner/2017-07-28/bugfix

lvm: fix check

lvm: fix check

Signed-off-by: Thomas Hipp <thipp@suse.de>
Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>

Merge pull request #1715 from brauner/2017-07-27/fix_gcc7_bug

lvm: check whether lxc.bdev.lvm.vg is empty

cgroups: use tight scoping

Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>

lvm: check whether lxc.bdev.lvm.vg is set

Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>

Merge pull request #1714 from brauner/2017-07-27/fix_gcc7_bug

cgroups: workaround gcc-7 bug

cgroups: workaround gcc-7 bug

Also: LOL

Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>

Merge pull request #1713 from brauner/2017-07-26/hybrid_cgroup_support

cgroups: handle hybrid cgroup layouts

cgroups: handle hybrid cgroup layouts

Closes #1669.
Closes #1678.
Relates to https://github.com/systemd/systemd/issues/6408.

Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>

utils: move helpers from cgfsng.c to utils.{c,h}

Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>

Allow full path in lxc.devices.allow

Some devices like LVM or cryptsetup entries have no stable major/minor, changing between host reboots.
In this case, hardcoded numbers are not usable in config file and there is currently no way to use hook with lxc-device to do the link at guest startup :

    * `pre-start`/`autodev` hook runs in host context but has the guest in stopped state and so lxc-device not usable
    * `start` hook is in running state but runs in guest context and so lxc-device not available

This patch converts fullpath in lxc.devices.allow to current major/minor numbers to address those changing numbers.

Signed-off-by: aeris <aeris@imirhil.fr>

Merge pull request #1606 from brauner/2017-06-01/lxc_setup_after_cgroup_unshare

call lxc_setup() after unshare(CLONE_NEWCGROUP)

start: lxc_setup() after unshare(CLONE_NEWCGROUP)

When the running kernel supports cgroup namespaces and users want to manually
set up cgroups via lxc.hook.mount before the init binary starts the cgroup
namespace needs to be already unshared. Otherwise the view on the cgroup mounts
is wrong. This commit places the call to lxc_setup() after the
LXC_SYNC_POST_CGROUP barrier.

Before this commit, the tty fds we allocate from a fresh devpts instance in the
container's namespaces before the init binary starts were referring to the
host's cgroup namespace since lxc_setup() was called before
unshare(CLONE_NEWCGROUP). Although not a security risk at this point since
setns() restricts its calls to /proc/<self>/ns files it's still better to do it
*after* the cgroup namespace has been unshared.

Adding a Suggested-by line for the lxc.mount.hook fix for Quentin.

Closes #1597.

Suggested-by: Quentin Dufour <quentin@dufour.tk>
Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>

Merge pull request #1706 from lifeng68/do_remount

Fix issue #1702, do remount with the MS_REMOUNT flag when mounts with MS_RDONLY

 Using 'add-required_remount_flags' function to add required flags

Signed-off-by: Li Feng <lifeng68@huawei.com>

Add test script to test the ro option of lxc.rootfs.options

Signed-off-by: Li Feng <lifeng68@huawei.com>

Fix issue #1702, do remount with the MS_REMOUNT flag when mounts with MS_RDONLY

Signed-off-by: Li Feng <lifeng68@huawei.com>

Merge pull request #1696 from 0x0916/2017-07-12/remove-old-workaround-case

cgfsng: remove the old workdaround code

Merge pull request #1695 from 0x0916/2017-07-12/update-doc-and-test

update doc and test

cgfsng: remove the old workdaround code

Signed-off-by: Long Wang <w@laoqinren.net>

tests: fix typo

Signed-off-by: Long Wang <w@laoqinren.net>

doc: s/lxc.net.ipv6/lxc.net.ipv6.address/g

Signed-off-by: Long Wang <w@laoqinren.net>

doc: s/lxc.net.ipv4/lxc.net.ipv4.address/g

Signed-off-by: Long Wang <w@laoqinren.net>

doc: s/lxc.mount/lxc.mount.fstab/g

Signed-off-by: Long Wang <w@laoqinren.net>

doc: s/lxc.pts/lxc.pty.max/g

Signed-off-by: Long Wang <w@laoqinren.net>

doc: s/lxc.tty/lxc.tty.max/g

Signed-off-by: Long Wang <w@laoqinren.net>

doc: s/lxc.console/lxc.console.path/g

Signed-off-by: Long Wang <w@laoqinren.net>

doc: s/lxc.rootfs/lxc.rootfs.path/g

Signed-off-by: Long Wang <w@laoqinren.net>

doc: s/seccomp/seccomp.profile/g

Signed-off-by: Long Wang <w@laoqinren.net>

doc: s/se_context/selinux.context/g

Signed-off-by: Long Wang <w@laoqinren.net>

doc: s/aa_profile/apparmor.profile/g

Signed-off-by: Long Wang <w@laoqinren.net>

doc: s/aa_allow_incomplete/apparmor.allow_incomplete/g

Signed-off-by: Long Wang <w@laoqinren.net>

Merge pull request #1692 from 0x0916/2017-07-11/cgfsng

cgfsng: only output debug info when we set cgroup data

Merge pull request #1693 from tenforward/japanese

Update Japanese man pages

doc: Update Japanese lxc-ls(1)

add "--defined" option

Signed-off-by: KATOH Yasufumi <karma@jazz.email.ne.jp>

doc: Update Japanese lxc.containers.conf(5)

Follow updates of configuration keys:
* Delete lxc.kmsg (commit 26a38fb)
* lxc.console -> lxc.console.path (commit 3aed493)
* lxc.mount -> lxc.mount.fstab (commit 47148e9)
* lxc.net.ipv{4,6} -> lxc.net.ipv{4,6}.address (commit 9ff60df, 2e44ae2)
* lxc.tty -> lxc.tty.max, lxc.devttydir -> lxc.tty.dir (commit fe1c588)
* lxc.pts -> lxc.pty.max (commit 232763)

Signed-off-by: KATOH Yasufumi <karma@jazz.email.ne.jp>

cgfsng: only output debug info when we set cgroup data

Only output debug info `cgroup 'xxxx' set to 'yyyy'` when we set
cgroup data.

Signed-off-by: Long Wang <w@laoqinren.net>

Fix syntax error in lxc-download

Signed-off-by: Stéphane Graber <stgraber@ubuntu.com>

Sanitize lxc-download script with shellcheck

Fix potential CVE when using spaces in LXC names

Signed-off-by: William Gathoye <william@gathoye.be>

Merge pull request #1690 from brauner/2017-07-10/lifeng68-Modify_fds

start: dup std{in,out,err} to pty slave

utils: set_stdfds()

non-functional changes

Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>

start: dup std{in,out,err} to pty slave

In the case the container has a console with a valid slave pty file descriptor
we duplicate std{in,out,err} to the slave file descriptor so console logging
works correctly. When the container does not have a valid slave pty file
descriptor for its console and is started daemonized we should dup to
/dev/null.

Closes #1646.

Signed-off-by: Li Feng <lifeng68@huawei.com>
Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>

Merge pull request #1687 from 0x0916/2017-07-09/delete-comments

delete comments

Merge pull request #1688 from 0x0916/2017-07-09/use-access

utils: use access instead of stat

utils: use access instead of stat

we shoud test whether the file exists and grants
execute permissions.

Signed-off-by: Long Wang <w@laoqinren.net>

commonds: fix typo

Signed-off-by: Long Wang <w@laoqinren.net>

commands: delete meaningless comments

Signed-off-by: Long Wang <w@laoqinren.net>

Merge pull request #1659 from brauner/2017-06-28/do_not_use_cmd_socket_on_daemonized_start

start: use separate socket on daemonized start

commands: handle EINTR

Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>

commands: abstract cmd socket handling + logging

Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>

commands: add missing translation

Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>

commands: mv lxc_make_abstract_socket_name()

Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>

commands: make state server interface flexible

This adds a little more flexibility to the state server. The idea is to have a
command socket function "lxc_cmd_add_state_client()" whose only task is to add
a new state client to the container's in-memory handler. This function returns
either the state of the container if it is already in the requested state or it
will return the newly registered client's fd in one of its arguments to the
caller. We then provide a separate helper function "lxc_cmd_sock_rcv_state()"
which can be passed the returned client fd and listens on the fd for the
requested state.
This is useful when we want to first register a client, then send a signal to
the container and wait for a state. This ensure that the client fd is
registered before the signal can have any effect and can e.g. be used to catch
something like the "STOPPING" state that is very ephemeral.

Additionally we provide a convenience function "lxc_cmd_sock_get_state()" which
combines both tasks and is used in e.g. "lxc_wait()".

Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>

commands: rename to lxc_cmd_add_state_client()

The new wait commands API is not yet stable so this change is ok.

Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>

lxc static init: report exec*() failure

Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>

test: shortlived daemonized containers

Add a test to see if we can start daemonized containers that have a very
short-lived init process. The point of this is to see whether we can correctly
retrieve the state.

Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>

lxccontainer: non-functional changes

Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>

lxccontainer: make sure memory is free()ed

Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>

start: use separate socket on daemonized start

Since we killed lxc-monitord we rely on the container's command socket to wait
for the container. This doesn't work nicely on daemonized startup since a
container's init process might be something that is so short-lived that we
won't even be able to add a state client before the mainloop closes. But the
container might still have been RUNNING and executed the init binary correctly.
In this case we would erroneously report that the container failed to start
when it actually started just fine.
This commit ensures that we really all cases where the container successfully
ran by switching to a short-lived per-container anonymous unix socket pair that
uses credentials to pass container states around. It is immediately closed once
the container has started successfully.
This should also make daemonized container start way more robust since we don't
rely on the command socket handler to be running.

For the experienced developer: Yes, I did think about utilizing the command
socket directly for this. The problem is that when the mainloop starts it may
end up end accept()ing the connection that we want
do_wait_on_daemonized_start() to accept() so this won't work and might cause us
to hang indefinitely. The same problem arises when the container fails to start
before the mainloop is created. In this case we would hang indefinitely as
well.

Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>

start: generalize lxc_check_inherited()

Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>

utils: lxc_make_abstract_socket_name()

Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>

Merge pull request #1665 from brauner/2017-07-01/deprecate_lxc_rootfs_backend

prepare for LXC 2.1