- No real notable API changes. The largest change code-wise is in
Vminitd. The connection loop is set up immediately in the new API by
calling runConnections() on the client, and this blocks until either the
task it's running on is cancelled or beginGracefulShutdown is called.
Because of this, either we'd have to modify our API to have
VirtualMachineAgent implementations call some run() method, or we just
put runConnections in an internal task on Vminitd. This change uses the
latter approach as it's simpler.
- Updated dependencies to pick up latest protobuf and grpc-swift-2
fixes.
- ClientBootstrap.withConnectedSocket(fd).wait() registers the vsock fd
with epoll/kqueue immediately. vminitd sends its HTTP/2 SETTINGS frame
right away upon accepting the connection, so those bytes can arrive on
the fd — and be read by NIO — before the gRPC pipeline is installed.
With a bare pipeline, NIO has nowhere to send the bytes and discards
them.
- Buffer until ClientConnectionHandler is in the pipeline so it can
process the SETTINGS frame and fire .ready, then schedule delivery of
the buffered data with assumeIsolatedUnsafeUnchecked().execute.
- To help produce more accurate merged logs, defines a
`StderrLogHandler` in the vminitd Application.swift that writes
timestamps to the boot log with millisecond precision and format
identical to that of the log files created using the `--log-root` arg on
`container`.
- Updates hawkeye to latest version (v6.5.1).
---------
Co-authored-by: Danny Canter <danny_canter@apple.com>
Now that the static linux SDK we use has libarchive linked against it,
we can finally live our dream (/s) of supporting copying directories
in/out a little easier. This was.. kind of annoying. I think the very
simple route of writing archive to guest/host -> streaming over grpc is
simplest, but on the guest end we have a couple problems:
1. The VMs rootfs is read only today always, which is a good thing to me
and I don't want to change if we don't have to.
2. Writing the archive to the containers rootfs temporarily could work,
but it's a bit weird, and the user can make the containers rootfs
readonly which would screw that plan.
3. We could write it to /run or /tmp, but they're tmpfs and dealing with
the headache of the user possibly tarring an enormous dir is one I don't
want to care about.
So, that leaves us with the (truthfully better to me) approach of trying
to write the tar data directly to the host and skipping grpc which kinda
forces us to have a temp spot. Because of that, I made it such that we
pass a port number from host<->guest, and transfer the actual data
(either the single file or tarred dir contents) over the vsock port
instead. The stream is kinda clunky, but it just serves as a means to
exchange metadata and a "we're done" signifier.
Closes#221
This contains (in the static linux sdk but the swift version and the sdk
must match) libarchive support and mimalloc! This also (unrelated to the
sdk bump it seems) raises the amount of retries we do when dialing
vminitd. It seems CI on rare occasions recently has timed out on some
tests. We could also explore just lowering the concurrency on the
integration suite.
Closes#256
If the target of the uds proxy goes through a symlink (/var/run) we had
the same bug we had with mounts previously where we'd follow this and if
the resolved path existed in the root of the VM we'd end up mounting
there. Fix this by plopping the sockets in a holding spot and then bind
mounting them in (which has the correct resolution logic).
This change fixes an issue in linux pod where the exec process
does not inherit the container process config which means any
environment variables, working directory, user, rlimits,
capabilities would be different for the exec process.
This change ensures that exec process would inherit
environment variables, working directory, user, rlimits,
and capabilities from the container process config but
resets arguments, terminal, and stdio as those should not
be inherited.
This change also adds an integration test to exercise this.
Just like we have for hosts and dns configurations, we can add a
hostname field to make it easier to setup if every container is going to
be using the same hostname.
This PR adds a `networking: Bool = true` parameter to
`ContainerManager.create()` so callers can opt out of network interface
creation on a per-container basis.
## Motivation
Currently, `ContainerManager.create()` unconditionally allocates a vmnet
network interface for every container (when the manager has a network
configured). Some use cases don't need network access and benefit from
having it disabled to reduce attack surface.
There's no way to achieve this today without either:
- Initializing the `ContainerManager` without a network (which disables
networking for *all* containers)
- Clearing `config.interfaces` in the configuration closure (which
wastes an IP allocation from the vmnet pool since `network.create(id)`
has already been called)
## Changes
- Add `networking: Bool = true` to all three `create()` overloads on
`ContainerManager`
- When `false`, `self.network?.create(id)` is skipped entirely. No
interface is allocated, and no DNS is configured.
- `releaseNetwork`/`delete` remain safe to call regardless, since
`Allocator.release` silently ignores unknown IDs.
- Add unit test `testNetworkingFalseSkipsInterfaceCreation` using the
existing `NilGatewayNetwork` fixture
- Add integration tests `testNetworkingDisabled` and
`testNetworkingEnabled` that create containers through a network-enabled
`ContainerManager` and verify the presence/absence of `eth0` via
`/sys/class/net/`
There are some workloads that benefit from having a "true" init process
that reaps zombies. This change implements a minimal init as part of
vminitd and exposes an API to be able to use this init process in our
containers. signals will be forwarded, any child procs will be reaped
and the exit code of the actual workload will be propagated. The
LinuxContainer/LinuxPod API is a very simple bool to ask to use this as
our init process, and the init simply spawns whatever command you want
as a child underneath this.
```
➜ ./bin/cctl run --kernel bin/kernel.arm64 --init
/ # ps aux
PID USER TIME COMMAND
1 root 0:00 /.cz-init -- /bin/sh
2 root 0:00 /bin/sh
3 root 0:00 ps aux
```
This adds support to LinuxContainer to be able to provide an optional
writable layer. This is useful as today the rootfs size is determined by
the size of the image block itself which is set in stone at unpack time
of the image. This leaves a lot to be desired because a user might want
a larger or smaller rootfs size, but it's not configurable on a per
container basis because of this. To support this, we can pass an
additional writable block device that we can overlayfs with the image
contents in the guest. All writes will go to this writable layer, and
the rootfs size in the container is now whatever size this writable
layer is.
The block devices (via our ext4 package, but you could use whatever) are
cheap enough to generate on the fly that you can choose whatever size
you want easily.
These two stats are import to calculate the following
- wss bytes (memory usage - inactive files)
- rss bytes (anonymous memory)
- available bytes (limit - wss bytes)
Just exposing these two in `SandboxContext.proto` as we already collect
them in `readMemoryStats()` in `Cgroup2Manager`
- Downgrades a warning to the debug level in `BidirectionalRelay`
because the shutdown syscall isn't supported for certain socket types
- Rethrows an error if we can't start a socket relay
I've extracted the network cleanup step from `delete` into a new public
`releaseNetwork` method. This mirrors the existing `release` on
`VmnetNetwork` and enables stop-and-restart workflows where the network
allocation needs to be freed without removing the container's rootfs.
## Changes
- Add `ContainerManager.release(_:)` — frees network resources for a
container after it has stopped
- Refactor `delete(_:)` to call `releaseNetwork(_:)` before removing
files, preserving existing behavior
## Motivation
Currently the only way to free a container's network allocation is
`delete`, which also removes its files. This makes it impossible to stop
a container (freeing the network so the ID can be re-allocated) and
later restart it with its filesystem intact. The new `releaseNetwork `
method fills this gap.
- Adds a public method to wait for the relay to complete when both
directions close
- Improves error handling
- Moves `UnixSocketRelayManager` into a separate file
- Renames `SocketRelay` to `UnixSocketRelay`
- Adds logs at the info level after converting other logs to the debug
level previously
- Guarded against ipv4Gateway if nil and throw
`ContainerizationError(.invalidState, ...)` instead of force unwraping
- Added a unit test
`ContainerManagerTests.testCreateThrowsWhenGatewayMissing`
If we have the same source we should dedupe these. Previously
virt.framework would yell at us for a duplicate tag because we derive
the tag from hashing the mount source.
Changes:
- Refactors a bidirectional relay into a new class
- Makes the unique identifier for a socket configuration public
Bug fixes:
- Fixes a bug where partial writes didn't advance the pointer
- Fixes a buffer memory leak
- Fixes error messages
- Fixes log messages
- Fixes log levels
"clean up" is a verb, and "cleanup" is a noun. There are several
examples of the incorrect usage of "cleanup" as a verb in this
repository, including one in the public API (`cleanupOrphanedBlobs`).
For this public method, I've added a deprecation notice to the name with
the incorrect usage.
- Closesapple/container#988.
- macOS libarchive is not built with zstd support, so the workaround in
ArchiveReader is to attempt to decompress every archive as zstd. If
decompression fails, we pass the original archive to libarchive. If it
succeeds, we pass the uncompressed archive.
- Adds blob media type recognition for zstd to EXT4Unpacker. Tested zstd
blob unpack using `image pull tonistiigi/hello-world:zstd-docker`.
Co-authored-by: Aditya Ramani <a_ramani@apple.com>
I was not a huge fan of how rlimits were specified using the OCI type as
it just takes in strings for the type which isn't super ergonomic. This
adds a new type for rlimits that we can convert to oci type like we
already do for capabilities.
Upon further thought, we should probably support this at the pod level
and allow per container overrides. I've changed my mind on how tedious
it is to specify it on a per-container basis.
Fixes#79
This adds support for single file (virtiofs based) mounts to the two
main container types. This is transparent, and doesn't need to be setup
by a user, so there is no visible API change. Virtiofs does not support
single file mounts today, and it would be less than ideal from a
security standpoint to just expose the parent directory of the file to
the vm and then bind mount in the file, so we chose the following:
1. Create a temp directory the container types will manage.
2. Hardlink in the file we want.
3. Share that tempdir into the vm.
4. Finally bind mount in the file into the container.
The main goal I wanted for this is to leave the logic out of the
`VirtualMachineInstance`. I really want this to just be a little dance
the container types do, and leave the vm out of it. Because of that,
most of the logic is shoved in a new `FileMountContext` type that does
the dirty tricks of rewriting what the user asked for to be a bind mount
from a temporary holding spot we'll use in the guest for these temp
directories.
One of the downsides to this is today the tempdirs are solely on the
main volume, so cross volume will need to copy the file to the tempdir,
so writes won't get written back.
Co-authored-by: Dean Coulstock <d_coulstock@apple.com>
Co-authored-by: Jaewon Hur <hurjaewon@snu.ac.kr>
When we set a gateway outside a subnet in Linux, this fails with:
```
ip-route-add-default: netlink response indicates error, rc = -101
```
To resolve the error, we should first add a route to the gateway outside
the subnet, which this PR does.
Fixes#477
Because we start piping stdin before process launch we can fill up the
guest pipe buffer before the process even starts. We'd need some
backpressure mechanism to handle this and slow consumers (register the
other end with epoll and buffer some data etc.), but we should probably
just start piping after the process is up and running. This change does
exactly that, as well as stops holding the process mutex while draining
stdin in the guest for `CloseStdin()`.
This fixes issues where we try and write > pipe_buf bytes through stdin.
Today this hangs.
The guest agent isn't free, and today we were sizing the VMs memory
allotment and the containers cgroup to the same value. This change gives
some overhead memory to the VM (50MB for now) for the guest agent.
It's possible a user doesn't want the full stats list, and only wants
cpu/mem etc. This plumbs through the ability to filter to only what is
requested. This, while we're already here, adds in memory.event output
to the stats list. For that specifically, I think eventually we may want
a streaming variant of this so you can get alerted of changes in the
file immediately instead of polling/one off reads, but this is useful
for now.
- Simplifies processing and validation of MAC addresses for
apple/container#1005, where we need to create MAC addresses and
corresponding link local IPv6 addresses.
- Adds Codable for IPAddress and CIDR for apple/container#1006, so we
can parse and relay IP addresses regardless of address family for port
forwards.
Add a streaming rpc to copy single files in and out. This can easily be
extended to support copying directories, but we need the guest agent to
be able to tar/untar, and we'll need to figure out what that model is.
Closely related to https://github.com/apple/container/issues/305 I would
like to override the used SSL TrustRoots via standard env variables.
This here would add this configuration and would give an entrypoint for
an implementation of 305 to provide CLI flags or similar.
This has no tests yet, as this would require setting up something like a
MITM proxy when testing against a registry.
As I am unfamiliar with the codebase, I would be willing to do this, but
would require a first nudge on where to best implement this.
To actually use this, we would need to add the allowed env variables to
the `container system start` command env filter.
In a prior change I'd added a way for vminitd to double as a simple
pause container. This change wires this up by adding a new bool to the
pod config to ask for pid ns sharing.
Add a small bit of logic to monitor if vminitd goes over a (somewhat
arbitrary) memory threshold. On average, when running one container it
seems to hover around 25MiB, so this is mostly to catch cases where it's
exceeding a limit we don't deem normal.
Adds `calculateOrphanedBlobsSize()` to calculate the size of orphaned
blobs, will need this to include them under size and reclaimable space
for images in the `container system df` command so it matches up with
what `container image prune` frees up on disk.
- Fixes#417.
Rename `_prune()` to `cleanupOrphanedBlobs()` to clarify what it
actually does, and remove `prune()` method as we'll do all that logic in
container directly.
For the constructor we're using where we pass an already connected
socket, I'm not sure this even does anything, and if it does that's a
bug I'd rather just avoid altogether.
Also gets rid of the unused constructor we have to pass in a UDS.
Michael Crosby