tests: remove lxc-test-ubuntu

This is really taking a long time for not a lot of benefit.

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

CODING_STYLE: add section for str{n}cmp()

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

cgroups: use correct mask for chmod()

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

CODING_STYLE: add CODING_STYLE.md

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

CONTRIBUTING: update

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

cgfsng: simplifications and fixes

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

lxclock: remove pthread_atfork_handlers

They shouldn't be needed anymore.

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

conf: support mount propagation

Closes #810.

Signed-off-by: Yifeng Tan <tanyifeng1@huawei.com>
Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>

Fix lxc-console hang

The variable 'descr' is mistakenly covered with 'descr_console'.

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

Modify .gitignore

add:
src/lxc/cmd/lxc-checkconfig
src/lxc/cmd/lxc-update-config

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

console: they are really not necessary

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

fix userns helper error handling

In both of these cases if there is actually an error, we won't close the
pipe and the api call will hang. Instead, let's be sure to close the pipe
before waiting, so that it doesn't hang.

Signed-off-by: Tycho Andersen <tycho@tycho.ws>

coverity: #1425971

Dereference after null check

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

coverity: #1426734

Argument cannot be negative

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

coverity: #1429139

Resource leak

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

Restore most cases of am_guest_unpriv

The only cases where we really need to be privileged with respect
to the host is when we are trying to mknod, and in some cases
to do with a physical network device.  This patch leaves the
detection of the network device cases as a TODO.

This should fix the currently broken case of starting a privileged
container with at least one veth nic, nested inside an unprivileged
container.

Cc: Tycho Andersen <tycho@tycho.ws>
Signed-off-by: Serge Hallyn <shallyn@cisco.com>

define am_guest_unpriv

Sometimes we want to know whether we are privileged wrt our
namespaces, and sometimes we want to know whether we are priv
wrt init_user_ns.

Signed-off-by: Serge Hallyn <shallyn@cisco.com>

cgfsng: add required remount flags

https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=9566d6742852c527bf5af38af5cbb878dad75705

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

lsm: fix missing @ in function documentation

Signed-off-by: Marcos Paulo de Souza <marcos.souza.org@gmail.com>

selinux: simplify check for default label

Signed-off-by: Marcos Paulo de Souza <marcos.souza.org@gmail.com>
Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>

cgroups: add check for lxc.cgroup.use

Closes #870.

Signed-off-by: Marcos Paulo de Souza <marcos.souza.org@gmail.com>

templates: CentOS fixes

Issues fixed:
- lxc-centos died about a missing /run directory
- lxc-centos complained about some config files it couldn't modify
- the new container got stuck at startup time for a minute
  (literally), waiting for systemd-remount-fs startup script

Of course it still works for RHEL 6, CentOS 6 and 7 as well. I did not
verify earlier CentOS or RHEL releases.

Signed-off-by: Harald Dunkel <harald.dunkel@aixigo.de>
Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>

cgfsng: non-functional changes

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

cgfsng: do MS_REMOUNT

Perform MS_REMOUNT on mounts with MS_RDONLY.

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

attach: non-functional changes

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

attach: set the container's environment variables

Set the same environment variables that were used when starting the container
when attaching to the container.

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

cgroups: more consistent naming

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

cgroups: handle limits on the unified hierarchy

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

cgroups: cgfsng_set: handle unified hierarchy

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

cgroups: cgfsng_get: handle unified hierarchy

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

cgroups: cgfsng_attach: handle unified hierarchy

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

cgroups: cgfsng_create: handle unified hierarchy

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

cgroups: get controllers on the unified hierarchy

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

lxc.init: ignore SIGHUP

see comment for details.

Signed-off-by: Tycho Andersen <tycho@tycho.ws>

monitor: send SIGTERM to the container when SIGHUP is received

For the ->execute() case, we want to make sure the application dies when
SIGHUP is received. The next patch will ignore SIGHUP in the lxc monitor,
because tasks inside the container send SIGHUP to init to have it reload
its config sometimes, and we don't want to do that with init.lxc, since it
might actually kill the container if it forwards SIGHUP to the child and
the child can't handle it.

Signed-off-by: Tycho Andersen <tycho@tycho.ws>

lxc-init: use SIGKILL after alarm timeout

d76e3e1 inadvertently switched the alarm timeout from sigterm to sigkill.
We really want sigkill here, so let's bring it back.

Signed-off-by: Tycho Andersen <tycho@tycho.ws>

debian: We must use iproute on wheezy

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

lsm: do not #ifdefine

Since we stopped linking against AppArmor we can now remove the conditional
compilation.

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

tools: make "-n" optional

This lets users use the tools with "lxc-* -n <container-name>" or
"lxc-* <container-name>".

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

debian: Use iproute2 instead of iproute

The package has pretty much always been iproute2 with iproute being an
alias for it, the alias is now gone so we need to use iproute2.

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

better unprivileged detection

In particular, if we are already in a user namespace we are unprivileged,
and doing things like moving the physical nics back to the host netns won't
work. Let's do the same thing LXD does if euid == 0: inspect
/proc/self/uid_map and see what that says.

Signed-off-by: Tycho Andersen <tycho@tycho.ws>

better check for lock dir

Consider the case where we're running in a user namespace but in the host's
mount ns with the host's filesystem (something like
lxc-usernsexec ... lxc-execute ...), in this case, we'll be euid 0, but we
can't actually write to /run. Let's improve this locking check to make sure
we can actually write to /run before we decide to actually use it as our
locking dir.

Signed-off-by: Tycho Andersen <tycho@tycho.ws>

[cgfsng] show wrong errno

lxc_cgroup_set_data: h = get_hierarchy(controller);
if h is NULL, now errno is old, it donot set new one.
And then,
cgfsng_setup_limits:

    if (lxc_cgroup_set_data(cg->subsystem, cg->value, d)) {
if (do_devices && (errno == EACCES ||
errno == EPERM)) {
WARN("Error setting %s to %s for %s",
cg->subsystem, cg->value,
d->name);
continue;
}
SYSERROR("Error setting %s to %s for
 %s",
 cg->subsystem, cg->value,
 d->name);
goto out;
}

SYSERROR will show old errno, make me confused.

Signed-off-by: duguhaotian <duguhaotian@gmail.com>

network.c: Remove ip_forward_set and callers

The last user of ip_forward_set, lxc_ip_forward_on and
lxc_ip_forward_off was in 2009:

commit 92d385229be040cf034ac6efd9405b7bdfe06a3e
Author: Daniel Lezcano <dlezcano@fr.ibm.com>
Date:   Thu Oct 22 15:33:40 2009 +0200

    remove test directory

These functions are not called anymore.

Signed-off-by: Marcos Paulo de Souza <marcos.souza.org@gmail.com>

autotools: do not link against libapparmor

Since we write the label directly without going through the AppArmor API it
doesn't make sense to link against it.

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

apparmor: do not call aa_change_profile()

We can simply write the label ourselves. There's no magic happening.

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

lsm: add lsm_process_label_set_at()

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

lsm: add lsm_process_label_fd_get()

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

lsm: non-functional changes

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

Fix comments and add check in lxc_poll.

Add check whether handler->conf->console.path is 'none'

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

Modify .gitignore

Add:
src/tests/lxc-test-api-reboot
src/tests/lxc-test-criu-check-feature
src/tests/lxc-test-raw-clone
src/tests/lxc-test-share-ns
src/tests/lxc-test-state-server

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

start: fix mainloop cleanup goto statements

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

coverity: #1426694

remove logically dead code

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

coverity: #1426734

do not call close on bad fd

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

coverity: #1427190

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

coverity: #1427191

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

coverity: #1427638

avoid (however unlikely) double free

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

coverity: #1427639

remove logically dead code

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

coverity: #1427668

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

unlink lxc-init

It's sort of an implementation detail that this exists at all, and we
should probably not pollute the container's mount tables or FS with this.

Signed-off-by: Tycho Andersen <tycho@tycho.ws>

Create console when the rootfs is NULL

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

tools: fix android

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

cgfsng: reduce delta

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

Revert commit "bla" with bad commit message

This reverts commit 04c141a4ed93e5daf20b90b3da80314cb501b1a9.

bla

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

cgroups: reduce delta

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

attach: reduce delta

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

attach_options: reduce delta

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

test: fix console tests

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

console: cleanup

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

set exit status to 1 in the unknown si_code case

Signed-off-by: Tycho Andersen <tycho@tycho.ws>

lxc-execute: actually exit with the status of the spawned task

Now that we have things propagated through init and liblxc correctly, at
least in non-daemon mode, we can exit with the actual exit status of the
task, instead of always succeeding, which is not so helpful.

Signed-off-by: Tycho Andersen <tycho@tycho.ws>

start: don't return false when the container's init exits nonzero

This seems slightly counter-intuitive, but IMO it's what we want.
Basically, ->start() should succeed if the container is spawned correctly
(similar to how golang's exec.Cmd.Start() returns nil if the thing spawns
correctly), and users can check error_num (i.e. golang's exec.Cmd.Wait())
to see how it exited.

This preserves previous behavior, which basically was that start was always
successful if the thing actually launched. Since we never kept track of
exit codes, this would always succeed too. Now that we do, it doesn't, and
this change is required.

Signed-off-by: Tycho Andersen <tycho@tycho.ws>

remember the exit code from the init process

error_num seems to be trying to remember the exit code of the init process,
except that nothing actually keeps track of it anywhere. So, let's add a
field to the handler, so that we can keep track of the process' exit
status, and the propagate it to error_num in struct lxc_container so that
people can use it.

Note that this is a slight behavior change, essentially instead of making
error_num always == the return code from start, now it contains slightly
more useful information (the actual exit status). But, there is only one
internal user of error_num which I'll fix in later in the series, so IMO
this is ok.

Signed-off-by: Tycho Andersen <tycho@tycho.ws>

lxc.init: correctly exit with the app's error code

Based on the comments in the code (and the have_status flag), the intent
here (and IMO, the desired behavior) should be for init.lxc to propagate
the actual exit code from the real application process up through.
Otherwise, it is swallowed and nobody can access it.

The bug being fixed here is that ret held the correct exit code, but when
it went around the loop again (to wait for other children) ret is
clobbered. Let's save the desired exit status somewhere else, so it can't
get clobbered, and we propagate things correctly.

Signed-off-by: Tycho Andersen <tycho@tycho.ws>

fix lxc_error_set_and_log to match the docs

The documentation for this function says if the task was killed by a
signal, the return code will be 128+n, where n is the signal number. Let's
make that actually true.

(We'll use this behavior in later patches.)

Signed-off-by: Tycho Andersen <tycho@tycho.ws>

start: don't log stop/continue for non-init processes

This non-init forwarding check should really be before all the log messages
about "init continued" or "init stopped", since they will otherwise lie
about some process that wasn't init being stopped or continued.

Signed-off-by: Tycho Andersen <tycho@tycho.ws>

commands: add LXC_CMD_SERVE_STATE_CLIENTS

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

freezer: non-functional changes

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

lxccontainer: restore blocking wait()

Closes #2027.
Closes lxc/go-lxc#98.

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

utils: check suffix length

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

test: log error on failure

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

utils: do not rely on unitialized variable

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

lxc-alpine: allow retaining sys_ptrace per container

Signed-off-by: Kaarle Ritvanen <kaarle.ritvanen@datakunkku.fi>

console: set SFD_CLOEXEC on signal fd

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

start: non-functional changes

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

tools: honor --console and --console-log

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

attach: minor tweaks

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

console: add lxc_pty_map_ids()

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

console: adapt lxc_console_mainloop_add()

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

attach: cleanup attach_child_main()

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

console: add some pty helpers

- int lxc_make_controlling_pty()
- int lxc_login_pty()
- void lxc_pty_conf_free()
- void lxc_pty_info_init()
- void lxc_pty_init()

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

start: non-functional changes

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

console: move pty creation to separate function

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

confile: improve log messages

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

add some idmap parsing error messages

otherwise, we just get a return value of false from setting config failure,
with no indication as to what actually failed in the log.

Signed-off-by: Tycho Andersen <tycho@tycho.ws>

fix up lxc-usernsexec's exit status

* exit(1) when there is an option parsing error
* exit(0) when the user explicitly asks for help
* exit(1) when the user specifies an invalid option

Signed-off-by: Tycho Andersen <tycho@tycho.ws>

Add return check for 'lxc_cmd_get_name'

If 'lxc_cmd_get_name' failed and return with NULL, this would cause a segment fault.

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

Include -devel suffix in version string

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

Fix broken indentation

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