When starting a daemonized container, only the original parent
thread should return to the caller. The first forked child
immediately exits after forking, but the grandparent child
was in some places returning on error - causing a second instance
of the calling function.
Signed-off-by: Serge Hallyn <serge.hallyn@ubuntu.com>
Acked-by: Tycho Andersen <tycho.andersen@canonical.com>
In various places throughout the code, we want to "nullify" the std fds,
opening them to /dev/null or zero or so. Instead, let's unify this code and do
it in such a way that Coverity (probably) won't complain.
v2: use /dev/null for stdin as well
v3: add a comment about use of C's short circuiting
v4: axe comment, check errors on dup2, s/quiet/need_null_stdfds
Reported-by: Coverity
Signed-off-by: Tycho Andersen <tycho.andersen@canonical.com>
Signed-off-by: Serge Hallyn <serge.hallyn@ubuntu.com>
We're leaking the FILE* here while closing the underlying fd; let's just
close the file and thus close both.
Reported-by: Coverity
Signed-off-by: Tycho Andersen <tycho.andersen@canonical.com>
Signed-off-by: Serge Hallyn <serge.hallyn@ubuntu.com>
Initialize ret to 0 so compiler no longer complains about
monitor.c: In function 'lxc_monitor_open':
monitor.c:212:5: error: 'ret' may be used uninitialized in this function [-Werror=maybe-uninitialized]
https://github.com/openwrt/packages/issues/1356
Signed-off-by: Daniel Golle <daniel@makrotopia.org>
Instead of re-defining MS_ options all over the place, just revert the
last change to bdev.h so we have all the defines in there again.
Signed-off-by: Stéphane Graber <stgraber@ubuntu.com>
Commit 37cf711b added a destroy hook, but when it checks
at destroy time whether that hook exists, it assumes that
c->lxc_conf is good. In fact lxc_conf can be NULL, so check
for that.
Signed-off-by: Serge Hallyn <serge.hallyn@ubuntu.com>
Acked-by: Stéphane Graber <stgraber@ubuntu.com>
We use the right path when actually execing criu to checkpoint and restore, but
when checking versions we didn't. Let's use the right path.
Reported-by: Dietmar Maurer <dietmar@proxmox.com>
Signed-off-by: Tycho Andersen <tycho.andersen@canonical.com>
Acked-by: Stéphane Graber <stgraber@ubuntu.com>
and don't use it if not. This fixes failure to build with older
cgmanager.
Signed-off-by: Serge Hallyn <serge.hallyn@ubuntu.com>
Acked-by: Stéphane Graber <stgraber@ubuntu.com>
commit 507cee3618 moved the close and re-open of fds 0-2 into
do_start. But this means that the lxc monitor itself keeps the
caller's fds 0-2 open, which is wrong for daemonized containers.
Closes#548
Reported-by: Mathieu Le Marec - Pasquet <kiorky@cryptelium.net>
Signed-off-by: Serge Hallyn <serge.hallyn@ubuntu.com>
Acked-by: Stéphane Graber <stgraber@ubuntu.com>
We were using 'all' controller if current was in all the
same cgroup. That doesn't suffice. We'd have to check
the target. At that point we may as well just attach
controller by controller.
An optimization to consider is to check the /proc/initpid/cgroup
for all identical controllers. Let's start by just getting it
right.
Signed-off-by: Serge Hallyn <serge.hallyn@ubuntu.com>
Acked-by: Stéphane Graber <stgraber@ubuntu.com>
When restoring, we didn't clean up the pidfile that criu uses to pass us the
init pid on error or success; let's do that.
Signed-off-by: Tycho Andersen <tycho.andersen@canonical.com>
Acked-by: Serge E. Hallyn <serge.hallyn@ubuntu.com>
Until we bump the SONAME to liblxc2, only symbol additions and struct
member additions are allowed.
Adding struct members in the middle of the struct breaks backward
compatibility.
This commit makes it clear when struct members were added and moves a
few members that were added in the middle of the 1.0 struct to the end
of it.
Note that unfortunately that means we're breaking backward compatibility
between LXC 1.1.0 and the state after this commit, given 1.1 is
reasonably new, this is the least damaging way of fixing the problem.
Signed-off-by: Stéphane Graber <stgraber@ubuntu.com>
Acked-by: Serge E. Hallyn <serge.hallyn@ubuntu.com>
Current aufs supports FS_USERNS_MOUNT by using allow_userns module
parameter. It allows root in userns to mount aufs.
This patch allows an unprivileged container to use aufs. The value of
xino option is changed to /dev/shm/aufs.xino that an unpriv user can
write.
Signed-off-by: KATOH Yasufumi <karma@jazz.email.ne.jp>
Acked-by: Serge E. Hallyn <serge.hallyn@ubuntu.com>
(I erred in the first patch, causing every lxc-attach to unmount the
container-'s /proc)
Since we now use mount_proc_if_needed() from attach, as opposed to only
from start, we cannot assume we are pid 1. So fix the check for whether
to mount a new proc.
Signed-off-by: Serge Hallyn <serge.hallyn@ubuntu.com>
Acked-by: Stéphane Graber <stgraber@ubuntu.com>
To set lsm labels, a namespace-local proc mount is needed.
If a container does not have a lxc.mount.auto = proc set, then
tasks in the container do not have a correct /proc mount until
init feels like doing the mount. At startup we handlie this
by mounting a temporary /proc if needed. We weren't doing this
at attach, though, so that
lxc-start -n $container
lxc-wait -t 5 -s RUNNING -n $container
lxc-attach -n $container -- uname -a
could in a racy way fail with something like
lxc-attach: lsm/apparmor.c: apparmor_process_label_set: 183 No such file or directory - failed to change apparmor profile to lxc-container-default
Thanks to Chris Townsend for finding this bug at
https://bugs.launchpad.net/ubuntu/+source/lxc/+bug/1452451
Signed-off-by: Serge Hallyn <serge.hallyn@ubuntu.com>
Acked-by: Stéphane Graber <stgraber@ubuntu.com>
to populate the list of subsystems to use.
Cgmanager can be started with some subsystems disabled (i.e.
cgmanager -M cpuset). If lxc using cgmanager then uses the
/proc/self/cgroup output to determine which controllers to use,
it will fail when trying to do things to cpuset. Instead, ask
cgmanager which controllers to use.
This still defers (per patch 1/1) to the lxc.cgroup.use values.
Signed-off-by: Serge Hallyn <serge.hallyn@ubuntu.com>
Acked-by: Stéphane Graber <stgraber@ubuntu.com>
Doing this requires some btrfs functions from bdev to be used in
utils.c Because utils.h is imported by lxc_init.c, I had to create
a new initutils.[ch] which are used by both lxc_init.c and utils.c
We could instead put the btrfs functions into utils.c, which would
be a shorter patch, but it really doesn't belong there. So I went
the other way figuring there may be more such cases coming up of
fns in utils.c needing code from bdev.c which can't go into lxc_init.
Currently, if we detect a btrfs subvolume we just remove it. The
st_dev on that dir is different, so we cannot detect if this is
bound in from another fs easily. If we care, we should check
whether this is a mountpoint, this patch doesn't do that.
Signed-off-by: Serge Hallyn <serge.hallyn@ubuntu.com>
Acked-by: Stéphane Graber <stgraber@ubuntu.com>
In setup where we want to sniff with an IDS from inside a container
we can use the 'passthru' mode of macvlan. This was not accessible
from the config and this patch fixes the issue.
Signed-off-by: Eric Leblond <eric@regit.org>
We need to use lxc_list_for_each_safe, otherwise de-allocation
will fail with a list size bigger than 2. The pointer to the head
of the list also need freeing after we've freed all other elements
of the list.
Signed-off-by: Kien Truong <duckientruong@gmail.com>
Add a function to sort the cgroup settings before applying.
Currently, the function will put memory.memsw.limit_in_bytes after
memory.limit_in_bytes setting so the container will start
regardless of the order specified in the input. Fix#453
Signed-off-by: Kien Truong <duckientruong@gmail.com>
Otherwise a container created before we needed workdir will fail
to start after a kernel+lxc update.
Signed-off-by: Serge Hallyn <serge.hallyn@ubuntu.com>
Acked-by: KATOH Yasufumi <karma@jazz.email.ne.jp>
CRIU can get confused if there are two dumps that are written to the same
directory, so we make some minimal effort to prevent people from doing this.
This is a better alternative than forcing liblxc to create the directory, since
it is mostly race free (and neither solution is bullet proof anyway if someone
rsyncs some bad images over the top of the good ones).
Signed-off-by: Tycho Andersen <tycho.andersen@canonical.com>
Acked-by: Serge E. Hallyn <serge.hallyn@ubuntu.com>
If Lua 5.3 is compiled with LUA_COMPAT_5_2 defined, the
luaL_checkunsigned compatibility macro is already defined
in lauxlib.h.
Signed-off-by: Thomas Moschny <thomas.moschny@gmx.de>
(1) Add test for kernel version greater 3.
(2) Use && and || instead of -a and -o as suggested in
http://www.unix.com/man-page/posix/1p/test/.
lxc-checkconfig will currently report "missing" on "Cgroup memory controller"
for kernel versions greater 3. This happens because the script, before checking
for the corresponding memory variable in the kernel config, currently will test
whether we have a major kernel version greater- or equal to 3 and a minor kernel
version greater- or equal to 6. This adds an additional test whether we have a
major kernel version greater than 3.
Signed-off-by: Christian Brauner <christianvanbrauner@gmail.com>
Particularly when using the go-lxc api with lots of threads, it
happens that if the open files limit is > 1024, we will try to
select on fd > 1024 which breaks on glibc.
So use poll instead of select.
Signed-off-by: Serge Hallyn <serge.hallyn@ubuntu.com>
Acked-by: Stéphane Graber <stgraber@ubuntu.com>
The logging code uses a global log_fd and log_level to direct
logging (ERROR(), etc). While the container configuration file allows
for lxc.loglevel and lxc.logfile, those are only used at configuration
file read time to set the global variables. This works ok in the
lxc front-end programs, but becomes a problem with threaded API users.
The simplest solution would be to not allow per-container configuration
files, but it'd be nice to avoid that.
Passing a logfd or lxc_conf into every ERROR/INFO/etc call is "possible",
but would be a huge complication as there are many functions, including
struct member functions and callbacks, which don't have that info and
would need to get it from somewhere.
So the approach I'm taking here is to say that all real container work
is done inside api calls, and therefore the API calls themselves can
set a thread-local variable indicating which log info to use. If
unset, then use the global values. The lxc-* programs, when called
with a '-o logfile' argument, set a global variable to indicate that
the user-specified value should be used.
In this patch:
If the lxc container configuration specifies a loglevel/logfile, only
set the lxc_config's logfd and loglevel according to those, not the
global values.
Each API call is wrapped to set/unset the current_config. (The few
exceptions are calls which do not result in any log actions)
Update logfile appender to use the logfile specified in lxc_conf if (a)
current_config is set and (b) the lxc-* command did not override it.
Changelog (2015-04-21):
. always re-set current_config to NULL at end of an API
call, rather than storing the previous value. We don't
nest API calls.
. remove the log_lock stuff which wasn't used
. lxc_conf_free: if the config is current_config, set
current_config to NULL. (It can't be another thread's
current_config, or we wouldn't be freeing it)
. lxc_check_inherited: don't close fd if it is the
current_config->logfd. Note this is only called when
starting a container, so we have no other threads at
this point.
Changelog (2015-04-22)
. Unset the per-container logfd on destroy
.
. Do so before we rm the containerdir. Otherwise if the logfile is set
. to $lxcpath/$name/log, the containerdir won't be fully deleted.
Signed-off-by: Serge Hallyn <serge.hallyn@ubuntu.com>
Acked-by: Stéphane Graber <stgraber@ubuntu.com>
If we don't re-open these after clone, the init process has a pointer to the
parent's /dev/{zero,null}. CRIU seese these and wants to dump the parent's
mount namespace, which is unnecessary. Instead, we should just re-open
stdin/out/err after we do the clone and pivot root, to ensure that we have
pointers to the devcies in init's rootfs instead of the host's.
v2: Only close fds if the container was daemonized. This didn't turn out as
nicely as described on the list because lxc_start() doesn't actually have
the struct lxc_container, so it cant see the flag. Instead, we just pass it
down everywhere.
Signed-off-by: Tycho Andersen <tycho.andersen@canonical.com>
Acked-by: Serge E. Hallyn <serge.hallyn@ubuntu.com>
In vivid containers hugetlbfs is mounted, but it is not one of the hardcoded
fses in criu, so we need to tell criu that it is okay to automount it.
Signed-off-by: Tycho Andersen <tycho.andersen@canonical.com>
Acked-by: Serge E. Hallyn <serge.hallyn@ubuntu.com>
Note that we allow both a tagged version or a git build that has sufficient
patches for the features we require.
v2: close criu's stderr too
Signed-off-by: Tycho Andersen <tycho.andersen@canonical.com>
Acked-by: Serge E. Hallyn <serge.hallyn@ubuntu.com>
Trying to cage the beast that is lxccontainer.c.
Signed-off-by: Tycho Andersen <tycho.andersen@canonical.com>
Acked-by: Serge E. Hallyn <serge.hallyn@ubuntu.com>
As of criu 1.5, the --veth-pair argument supports an additional parameter that
is the bridge name to attach to. This enables us to get rid of the goofy
action-script hack that passed bridge names as environment variables.
This patch is on top of the systemd/lxcfs mount rework patch, as we probably
want to wait to use 1.5 options until it has been out for a while and is in
distros.
Signed-off-by: Tycho Andersen <tycho.andersen@canonical.com>
Acked-by: Serge E. Hallyn <serge.hallyn@ubuntu.com>
