lxccontainer: do_lxcapi_detach_interface to support detaching wlan devices

Signed-off-by: Thomas Parrott <thomas.parrott@canonical.com>

Merge pull request #3109 from brauner/2019-07-28/bugfixes

cgroups: initialize cpuset properly

cgroups: initialize cpuset properly

Closes #3108.
Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>

Merge pull request #3102 from brauner/2019-07-22/bugfixes

tree-wide: initialize all auto-cleanup variables

Merge pull request #3106 from brauner/2019-07-25/bugfixes

network: restore ability to move nl80211 devices

network: restore ability to move nl80211 devices

Closes #3105.
Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>

Merge pull request #3103 from tych0/info-on-enosys

pidfds: don't print a scary warning on ENOSYS

pidfds: don't print a scary warning on ENOSYS

Most kernels don't have this functionality yet, and so the warning is
printed a lot. Our people are scared of warnings, so let's make it INFO
instead in this case.

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

tree-wide: initialize all auto-cleanup variables

Closes: #3101.
Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>

Re-enable devel flag

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

Release LXC 3.2.1

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

Release LXC 3.2.0

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

lxc-download: Pre-release bump of compat

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

Merge pull request #3092 from Blub/seccomp-mem-rdwr

seccomp: open memfd read-write

seccomp: open memfd read-write

Makes it easier to implement syscalls which need to write to
a buffer passed by user space as a pointer.

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

Merge pull request #2921 from tomponline/tp-2019-03-26/routedveth

Adds veth router mode

doc: Documents the lxc.net.[i].veth.mode option

Signed-off-by: Thomas Parrott <thomas.parrott@canonical.com>

network: Adds veth router mode static routes and proxy entries

Signed-off-by: Thomas Parrott <thomas.parrott@canonical.com>

network: Adds mode param (bridge, router) to veth network setting

Defaulting to bridge mode.

Signed-off-by: Thomas Parrott <thomas.parrott@canonical.com>

lxc/log: Adds error_log_errno macro

Suggested usage:

return error_log_errno(err, "Failed: %s", "some error");

It sets errno to the value of err, then calls SYSERROR with the format and remaining args.

It always returns -1.

Suggested-by: Christian Brauner <christian.brauner@ubuntu.com>
Signed-off-by: Thomas Parrott <thomas.parrott@canonical.com>

Merge pull request #3090 from Rachid-Koucha/patch-3

Suppress hardcoded table sizes

Suppress hardcoded table sizes

. Use sizeof() instead of hardcoded values
. snprintf(..., size, ""...) is in error if the return code is >= size (not sufficient to set only ">")

Signed-off-by: Rachid Koucha <rachid.koucha@gmail.com>

Merge pull request #3089 from Rachid-Koucha/patch-2

Typo fix

Typo fix

Fixed a typo in error message

Signed-off-by: Rachid Koucha <rachid.koucha@gmail.com>

Merge pull request #3088 from tenforward/japanese

doc: Add lxc.comp.notify.cookie to Japanese lxc.container.conf(5)

doc: Add lxc.comp.notify.cookie to Japanese lxc.container.conf(5)

update for commit 214008e

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

Merge pull request #3087 from brauner/master

cgroup: check for non-empty conf

cgroup: check for non-empty conf

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

Merge pull request #3086 from brauner/2019-07-09/seccomp_fixes

seccomp: coding style

seccomp: coding style

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

Merge pull request #3080 from Blub/seccomp-notify-api

Seccomp notify api update

af_unix: remove unused variable

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

seccomp: send caller pidfd along with proxied requests

On the one hand this should close the race between the
process exiting until the proxy reads the request.
On the other hand it'll help the proxy quickly access info
from /proc (such as ./cwd, ./ns/mnt, ...)

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

seccomp: recvmsg with MSG_TRUNC

We only read the message without the cookie. For now assert
that the sender also didn't try to send more by letting
`recvmsg()` return the original size of the packet if it was
longer.

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

doc: document lxc.seccomp.notify.cookie

and fix a minor typo

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

seccomp: defer reconnecting to the proxy

With the previous commit we now attempt to reconnect to the
proxy in the beginning of the notify handler if we had no
connection.
If the connection fails later on, we now don't really need
to immediately try to reconnect if we send a default
response anyway (particularly if the recv() fails). (This
also gives the proxy more time, for instance if it was just
restarted.)

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

seccomp: keep retrying to reconnect to proxy

If a syscall happens after we already failed to communicate
with the proxy, proxy_fd was -1.
Before the previous commit we'd then be stuck in the state
where there was no proxy registered. With the previous
commit we'd send a default reply and only then try to
reconnect.
Improve this even further by trying to reconnect right at
the start.

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

seccomp: send default response when there's no proxy

Particularly, when there's no proxy registered (iow. none
configured but the seccomp profile still had a 'notify'
rule), we don't want to leave them hanging.

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

seccomp: retry connecting to the proxy once

If the first sendmsg() fails, try to reconnect once before
failing. Otherwise if a proxy restarts while no syscall
happens, the next syscall always fails with ENOSYS.

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

seccomp: don't ignore syscalls when there's no proxy

The container process would just hang.

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

seccomp: remove reconnect-loop

When we fail to send a message, we send a default seccomp
response and try to reconnect to the proxy. It doesn't
really make much sense to retry to send the request over the
new connection as the syscall has already been answered. The
same goes for receiving the response - after reconnecting to
the proxy, we're a new client to a potentially new proxy
process, so awaiting a response without having sent a
request doesn't make all too much sense either.

In the future we should probably have a timeout or retry
count for the entire proxy _transaction_ before sending a
response to seccomp at all (and probably handle requests
asynchronously).

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

seccomp: use SOCK_SEQPACKET for the notify proxy

The seccomp notify API has a few variables: The struct sizes
are queried at runtime, and we now also have a user
configured cookie.
This means that with a SOCK_STREAM connection the proxy
needs to carefully read() the right amount of data based on
the contents of our proxy message struct to avoid ending up
in the middle of a packet.
While for now this may not be too tragic, since we currently
only ever send a single packet and then wait for the
response, we may at some point want to be able to handle
multiple processes simultaneously, hence it makes sense to
switch to a packet based connection.

So switch to using SOCK_SEQPACKET which is packet based,
(and also guarantees ordering). The `MSG_PEEK` flag can be
used with `recvmsg()` to figure out a packet's size on the
other end, and usually the size *should* not change after
that for an existing connection from a running container.

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

seccomp: assert that __reserved is 0 in notify responses

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

seccomp: update notify api

The previous API doesn't reflect the fact that
`seccomp_notif` and `seccomp_notif_resp` are allocatd
dynamically with sizes figured out at runtime.

We now query the sizes via the seccomp(2) syscall and change
`struct seccomp_notify_proxy_msg` to contain the sizes
instead of the data, with the data following afterwards.

Additionally it did not provide a convenient way to identify
the container the message originated from, for which we now
include a cookie configured via `lxc.seccomp.notify.cookie`.

Since we currently always send exactly one request and await
the response immediately, verify the `id` in the client's
response.

Finally, the proxy message's "version" field is removed, and
we reserve 64 bits in its place.

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

conf: add lxc.seccomp.notify.cookie

This is an arbitrary string to to be included in proxied
seccomp notification messages.

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

file_utils: add lxc_recvmsg_nointr_iov

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

af_unix: add lxc_unix_connect_type

we want to use SOCK_SEQPACKET and in the future perhaps
SOCK_DATAGRAM as well

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

af_unix: add lxc_abstract_unix_recv_fds_iov()

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

af_unix: add lxc_abstract_unix_send_fds_iov

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

Merge pull request #3085 from Blub/pidfd_send_signal-fixup

pidf_send_signal: fix return value

pidf_send_signal: fix return value

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

Merge pull request #3083 from brauner/2019-07-07/mount_api

lxccontainer: properly cleanup on mount injection failure

lxccontainer: properly cleanup on mount injection failure

Closes: #3082
Reported-by: Stéphane Graber <stgraber@ubuntu.com>
Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>

Merge pull request #3081 from brauner/2019-07-05/network_unification_fixes

start: call lxc_find_gateway_addresses early

start: call lxc_find_gateway_addresses early

This restores the lxc.net.x.ipv4.gateway = auto and
lxc.net.x.ipv6.gateway = auto functionality.

When the child is created the parent and child have different views of
struct lxc_handler since - obviously - virtual memory is duplicated. So any
changes to done by the parent that the child should see need to be IPCed to it.
For any non-actual device creation stuff this does not make much sense. This
includes finding gateway addresses. Move it back prior to clone().

Fixes #3078

Signed-off-by: Thomas Parrott <thomas.parrott@canonical.com>
[christian.brauner@ubuntu.com: non-functional changes and update commit message]
Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>

Merge pull request #3077 from brauner/2019-07-03/network_fixes

network: simplify lxc_network_move_created_netdev_priv()

network: simplify lxc_network_move_created_netdev_priv()

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

Merge pull request #3076 from brauner/2019-07-03/network_fixes

network: fixes after unifying network creation

network: send names for all non-trivial network types

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

network: record created_name for instantiate_phys()

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

network: simplify instantiate_phys()

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

network: record created_name for instantiate_vlan()

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

network: simplify instantiate_vlan()

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

network: record created_name for instantiate_ipvlan()

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

network: simplify instantiate_ipvlan()

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

network: stash created_name in instantiate_macvlan()

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

network: simplify instantiate_macvlan()

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

network: s/loDev/loop_device/g

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

Merge pull request #3075 from brauner/2019-07-03/cgroups

cgroups: hande cpuset initialization race

cgroups: hande cpuset initialization race

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

Merge pull request #3074 from brauner/2019-07-03/fix_phys_network_creation

network: remove faulty restriction

network: remove faulty restriction

Reported-by: Thomas Parrott <thomas.parrott@canonical.com>
Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>

Merge pull request #3069 from brauner/2019-07-01/network_creation

start: unify network creation

start: expose LXC_PID to network hooks too

Closes #3066.
Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>

start: unify and simplify network creation

Make sure that network creation happens at the same time for containers started
by privileged and unprivileged users. The only reason we didn't do this so far
was to avoid sending network device ifindices around in the privileged case.

Link: https://github.com/lxc/lxc/issues/3066
Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>

Merge pull request #3064 from brauner/2019-06-27/cleanup_macros

bugfixes

Merge pull request #3059 from brauner/2019-06-21/seccomp_notify

lxccontainer: rework seccomp notify api function

Merge pull request #3067 from Rachid-Koucha/patch-1

Move code/variable in smaller scope

Move code/variable in smaller scope

In start.c, do not fill path[] table if not necessary

Signed-off-by: Rachid Koucha <rachid.koucha@gmail.com>

Merge pull request #3065 from lifeng68/fix_memory_leak

fix memory leak in do_storage_create

fix memory leak in do_storage_create

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

cgroups: move variable into tighter scope

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

cgroups: correctly order variables

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

cgroups: move variable into tighter scope

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

cgroups: simplify cgfsng_nrtasks()

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

cgroups: move variable into tighter scope

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

cgroups: move variable into tighter scope

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

cgroups: use __do_free in cgfsng_attach()

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

cgroups: simplify cgfsng_setup_limits()

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

cgroups: move variables into tighter scope

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

cgroups: use __do_free

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

Merge pull request #3063 from lifeng68/fix_memory_leak

cgfsng: fix memory leak in lxc_cpumask_to_cpulist

cgfsng: fix memory leak in lxc_cpumask_to_cpulist

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

lxccontainer: rework seccomp notify api function

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

Merge pull request #3056 from brauner/2019-06-20/cpuset

cgfsng: write cpuset.mems of correct ancestor

cgfsng: write cpuset.mems of correct ancestor

Reported-by: Free Ekanayaka <free.ekanayaka@canonical.com>
Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>

Merge pull request #3052 from tanyifeng/fd_leak

parse.c: fix fd leak from memfd_create

parse.c: fix fd leak from memfd_create

Signed-off-by: t00416110 <tanyifeng1@huawei.com>

Merge pull request #3050 from hallyn/pkgconfig

lxc.pc.in: add libs.private for static linking

lxc.pc.in: add libs.private for static linking

None of them seem to support pkg-config themselves, else we could add
them to Requires.private.

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