allow running lxc-monitord as a system daemon

lxc-monitord instances are spawned on demand and, if this
happens from a service, the daemon is considered part of
it by systemd, as it is running in the same cgroups. This
can be avoided by leaving it running permanently.

Signed-off-by: Wolfgang Bumiller <w.bumiller@proxmox.com>

Merge pull request #3340 from brauner/2020-03-30/fixes

cgroups: handle older kernels (e.g. v4.9)

cgroups: send two fds to attach to unified cgroup

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

cgroups: send two attach fds

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

start: log error when failing to create cgroup

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

cgroups: handle older kernels (e.g. v4.9)

On olders kernels the restrictions to move processes between cgroups are
different than they are on newer kernels. Specifically, we're running into the
following check:

if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) &&
    !uid_eq(cred->euid, tcred->uid) &&
    !uid_eq(cred->euid, tcred->suid))
        ret = -EACCES;

which dictates that in order to move a process into a cgroup one either needs
to be global root (no restrictions apply) or the effective uid of the process
trying to move the process and the {saved}uid of the process that is supposed
to be moved need to be identical. The new attaching logic we did didn't
fulfill this criterion for because it's not present on new kernels.

Closes https://github.com/lxc/lxd/issues/7104.
Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>

Merge pull request #3339 from Blub/cmd-get-cgroup-string-termination

verify cgroup controller name

verify cgroup controller name

validate that a cgroup controller name is a valid
zero-terminated string before passing it to
`cgroup_ops->get_cgroup()`.

Signed-off-by: Wolfgang Bumiller <w.bumiller@proxmox.com>

Merge pull request #3338 from brauner/2020-03-28/fixes

tree-wide: fixes

tree-wide: s/recursive_destroy/lxc_rm_rf/g

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

cgroups: better helper naming

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

cgroups: move check for valid monitor process up

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

Merge pull request #3337 from bale-cen/master

monitor process exited by signal SIGKILL, clean cgroup resource by th…

Merge pull request #3336 from brauner/2020-03-28/fixes

cgroups: please compilers

monitor process exited by signal SIGKILL, clean cgroup resource by third party

Writing the value 0 to a cgroup.procs file causes the
writing process to be moved to the corresponding cgroup

Signed-off-by: cenxianlong <cenxianlong@huawei.com>

cgroups: please compilers

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

Merge pull request #3335 from brauner/2020-03-27/fixes

cgroups: use hidden directory for attaching cgroup

cgroups: use hidden directory for attaching cgroup

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

Merge pull request #3333 from brauner/2020-03-27/fixes

conf: simplify userns_exec_minimal()

conf: simplify userns_exec_minimal()

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

Merge pull request #3332 from brauner/2020-03-27/fixes

attach: fixes

conf: introduce and use userns_exec_minimal()

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

Revert "cgroups: fix unified cgroup attach"

This reverts commit ba7ca43b0be417275db7865336191681d915e97c.

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

Merge pull request #3331 from brauner/2020-03-27/fixes

tree-wide: fixes

fixup i/o handler return values

Particularly important for lxc_cmd_handler() handles client
input and should not be capable of canceling the main loop,
some syscall return values leaked through overlapping with
LXC_MAINLOOP_ERROR, causing unauthorized clients connecting
to the command socket to shutdown the main loop.

In turn, signal_handler() receiving unexpected
`signalfd_siginfo` struct sizes seems like a reason to bail
(since it's a kernel interface).

Signed-off-by: Wolfgang Bumiller <w.bumiller@proxmox.com>
Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>

cgroups: fix unified cgroup attach

There's a fundamental problem with futexes and setid calls and the go runtime.
POSIX requires that when one thread setids all threas must setids and it uses
futexes and signals to synchronize the state across threads. This causes
deadlocks which means we can't use the pretty solution I first implemented.
Instead we need to chown after we create the directory. I might come up with
something smarter later but for now this will do.

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

Merge pull request #3330 from brauner/2020-03-27/fixes

conf: rework and fix leak in userns_exec_1()

cgroups: remove unused variable

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

attach: use close_prot_errno_disarm()

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

cgroups: rework __cg_unified_attach()

We didn't account for cgroup_attach() succeeding and just tried to attach to
the same cgroup again which doesn't make sense.

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

cgroups: move pointer dereference after check

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

commands: log actual errno when lxc_cmd_get_cgroup2_fd() fails

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

conf: rework and fix leak in userns_exec_1()

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

Merge pull request #3329 from brauner/2020-03-25/fixes

cgroups: fix attaching to the unified cgroup

cgroups: fix attaching to the unified cgroup

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

Merge pull request #3328 from brauner/2020-03-25/fixes

tree-wide: fixes

dir: improve dir backend

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

dir: use cleanup macro in dir_mount()

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

tree-wide: harden mount option parsing

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

Merge pull request #3327 from P-EB/master

lxc.service: Starts after remote-fs.target

[lxc.service] Starts after remote-fs.target to allow containers relying on remote FS to work

Signed-off-by: Pierre-Elliott Bécue <becue@crans.org>

lxc_init: add missing O_CLOEXEC

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

lxc_init: move main() down

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

configure.ac: Reset devel flag post-release

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

Release LXC 4.0.0

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

Merge pull request #3325 from brauner/2020-03-24/fixes

make dist: add missing files

Merge pull request #3324 from stgraber/master

lxc-download: Pre-release bump of compat

make dist: add missing files

deleted:    CODING_STYLE.md
deleted:    LICENSE.GPL2
deleted:    LICENSE.LGPL2.1
deleted:    README.md
deleted:    coccinelle/exit.cocci
deleted:    coccinelle/run-coccinelle.sh
deleted:    coccinelle/while-true.cocci
deleted:    doc/api-extensions.md
deleted:    src/tests/lxc-test-exit-code
deleted:    src/tests/travis.sh

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

lxc-download: Pre-release bump of compat

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

Merge pull request #3323 from tych0/fix-readonly-bind-mounts

conf: fix read-only bind mounts

conf: fix read-only bind mounts

Here we would always set MS_RDONLY in required_flags if it was set in
mountflags, so the expression:

!(required_flags & ~mountflags)

would always be true, and we would always skip the remount.

Instead, let's treat readonly as special: always do the remount if
MS_RDONLY is present. Unfortunately it doesn't seem to show up in
sb.f_flag, so we can't use the same path as everything else.

This only inadvertently worked before because of a bug fixed in
f75917858023 ("conf: don't accidently double-mount").

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

Merge pull request #3322 from brauner/2020-03-24/fixes

utils: allow removal of immutable files

utils: allow removal of immutable files

Closes #3185.
Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>

Merge pull request #3321 from brauner/2020-03-24/fixes

lxc-update-config: handle lxc.rootfs.backend correctly

lxc-local: remove -l/--list from help

Closes #2877.
Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>

lvm: don't generate uuid for ext4 snapshots

only btrfs and xfs need it.

Closes #2375.
Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>

lxc-update-config: handle lxc.rootfs.backend correctly

Closes #2424.
Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>

Merge pull request #3320 from brauner/2020-03-22/fixes

overlay: rewrite

lxc_copy: only overmount overlay subdirectory with tmpfs

Closes #2959.
Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>

overlay: rewrite and simplify

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

Merge pull request #3319 from brauner/2020-03-21/fixes

network: fixes

lxc-user-nic: enable uid-marked veth devices for uids with 5 digits

With the newly extended lxc_ifname_alnum_case_sensitive() we'll have 62^4
unique network device names per user.

Closes #3176.
Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>

network: introduce lxc_ifname_alnum_case_sensitive()

Right now we have 36^n unique network device names. Let's make it 62^n.

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

Merge pull request #3318 from brauner/2020-03-20/fixes

log: fix cmd logging

log: fix cmd logging

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

Merge pull request #3317 from brauner/2020-03-20/fixes

mainloop, ringbuf: cleanup

cgroups: simplify

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

Merge pull request #3316 from brauner/2020-03-20/fixes

log: fixes

ringbuf: fix cleanup operations

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

mainloop: cleanup

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

log: add missing variable and fix CMD_SYSINFO()

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

log: cleanup

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

Merge pull request #3315 from brauner/2020-03-20/fixes

log: add missing \

log: add missing \

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

Merge pull request #3314 from brauner/2020-03-20/fixes

start: move reading seccomp profile after pre-start hook

start: move reading seccomp profile after pre-start hook

This allows users to generate custom profiles at startup.

Closes #1848.
Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>

Merge pull request #3313 from brauner/2020-03-20/fixes

network: fixes

lxc_user_nic: rework device creation

Closes #3058.
Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>

nl: improve how we surface errors

Closes #3057.
Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>

Merge pull request #3312 from brauner/2020-03-20/fixes

network: use cleanup macros

network: use cleanup macros

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

Merge pull request #3310 from brauner/2020-03-19/fixes

network: fix ovs removal

network: use cleanup attributes

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

network: cleanup galore

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

network: use is_empty_string() everywhere

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

network: fix ovs removal

Closes #3137.
Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>

Merge pull request #3309 from brauner/2020-03-19/fixes

tree-wide: logging fixes and hardening

log: use global variable to catch statements in loggers

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

Merge pull request #3308 from brauner/2020-03-19/fixes

conf: flatten logic in mount_entry()

cgroups: don't call statements from loggers

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

conf: flatten logic in mount_entry()

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

Merge pull request #3307 from brauner/2020-03-19/fixes

network: fixes

conf: don't accidently double-mount

Closes #3073.
Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>

network: fix moving network devices with custom name

Closes #1271.
Closes #3131.
Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>

network: introduce and use is_empty_string()

since some members of struct lxc_netdev are arrays, not pointers.

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

Merge pull request #3306 from brauner/2020-03-18/fixes

Makefile: fix typo

Makefile: fix typo

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

Merge pull request #3305 from brauner/2020-03-18/fixes

lxc-unshare: add syscall_wrappers.h to build requirements

lxc-unshare: add syscall_wrappers.h to build requirements

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

Merge pull request #3304 from brauner/2020-03-18/fixes

tree-wide: introduce and use syscall number header