We need to make sure containers cannot modify host path unless it is explicitly shared to it. Right now we expose an additional top level shared directory to the guest and allow it to be modified. This is less ideal and can be enhanced by following method:
1. create two directories for each sandbox:
-. /run/kata-containers/shared/sandboxes/$sbx_id/mounts/, a directory to hold all host/guest shared mounts
-. /run/kata-containers/shared/sandboxes/$sbx_id/shared/, a host/guest shared directory (9pfs/virtiofs source dir)
2. /run/kata-containers/shared/sandboxes/$sbx_id/mounts/ is bind mounted readonly to /run/kata-containers/shared/sandboxes/$sbx_id/shared/, so guest cannot modify it
3. host-guest shared files/directories are mounted one-level under /run/kata-containers/shared/sandboxes/$sbx_id/mounts/ and thus present to guest at one level under /run/kata-containers/shared/sandboxes/$sbx_id/shared/
Signed-off-by: Peng Tao <bergwolf@hyper.sh>
When an x86 sandbox has a vIOMMU (needed for VFIO), it needs the
'kernel_irqchip=split' option or it can't start. fdcd1f3a2 attempts to set
that, but ends up just writing it to a temporary (looks like Go for range
loops pass by value).
Fixes: #2694
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Fabiano Fidêncio <fidencio@redhat.com>
Add a configuration option and a Pod Annotation
If activated:
- Add kernel parameters to load iommu
- Add irqchip=split in the kvm options
- Add a vIOMMU to the VM
Fixes#2694
Signed-off-by: Adrian Moreno <amorenoz@redhat.com>
Signed-off-by: Fabiano Fidêncio <fidencio@redhat.com>
Add a new function appendIOMMU() to the qemuArch interface
and provide an implementation on amd64 architecture.
Signed-off-by: Adrian Moreno <amorenoz@redhat.com>
Signed-off-by: Fabiano Fidêncio <fidencio@redhat.com>
The ppc64 specific qemu setup code adds a "pmu=off" parameter to the cpu
model if the nestedRun option is set. But, not only does availability of
the pmu have nothing to do with nesting on POWER, there is no "pmu=" cpu
opton for ppc64 at all.
So, simply remove it.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Hard-coded Qemu machine options create challenges when running Kata
with latest Qemu (v5.0) or with latest processor version.
This patch makes it configurable by leveraging the existing machine_accelerators
option in configuration.toml.
This patch fixes#2657 for ppc64le
Signed-off-by: bpradipt@in.ibm.com
The default ppc64le Qemu binary path was specific for Ubuntu.
This patch fixes the default binary path for both Fedora and Ubuntu
Fixes: #2738
Signed-off-by: bpradipt@in.ibm.com
qemu_ppc64le.go applies the "tsc=reliable", "no_timer_check" and
"noreplace-smp" kernel parameters, despite those being x86 specific. So,
just remove them.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Dup a new file descriptor for temporary logger writer,
since this logger would be dropped and it's writer would
be closed out of if definition scope, which would cause
the logger process thread terminated if it used the original
pipe write fd.
Fixes: #318
Signed-off-by: fupan.lfp <fupan.lfp@antfin.com>
With this change, a container is not longer given access to
the underlying root partition.
This is done by explicitly adding the root partition
to the device cgroup of the container.
Fixes: #317
Signed-off-by: fupan.lfp <fupan.lfp@antfin.com>
The Qemu version check in unit test case is no longer needed for
Power since we don't support Kata with Qemu version < 4.x.
Fixes: #315
Signed-off-by: bpradipt@in.ibm.com
Improve the output of the data collection script to use lots more folds.
This makes it easier to review the information when viewing the pasted
output in a GitHub issue.
Fixes: #313.
Signed-off-by: James O. D. Hunt <james.o.hunt@intel.com>
Add `start_section()` and `end_section()` functions to the data
collection script to allow new unfoldable sections to be created.
Redefine `show_header()` and `show_footer()` to use the new functions.
Signed-off-by: James O. D. Hunt <james.o.hunt@intel.com>
Now that the Kata containerd shim v2 can display a version string,
add those details to the data collection script.
Fixes: #309.
Signed-off-by: James O. D. Hunt <james.o.hunt@intel.com>
All components should support a `--version` option to allow clear
identification of the version of the component being used.
Note that the build changes are required to allow the shim binary to
access the golang code generated by the build (such as the `version`
variable).
Fixes: #307.
Signed-off-by: James O. D. Hunt <james.o.hunt@intel.com>
Add a new system component, used only when tracing is enabled. The
component listens to the agent over VSOCK, forwarding trace spans
created by the agent in the virtual machine onwards to an OpenTelemetry
collector (such as Jaeger) running on the host.
Fixes: #224.
Signed-off-by: James O. D. Hunt <james.o.hunt@intel.com>
Fix a long-standing bug where the KSM throttler logs would not be
collected by removing the last (unused) parameter to the
`find_system_journal_problems()` function.
Signed-off-by: James O. D. Hunt <james.o.hunt@intel.com>
The function that checks for legacy packages in the collect script was
missing pipes denoting regex alternation.
Signed-off-by: James O. D. Hunt <james.o.hunt@intel.com>
The agent logger is useful and generic enough that it can be used by
other components, so move the agent logging package to below a top level
`pkg` to encourage re-use.
Signed-off-by: James O. D. Hunt <james.o.hunt@intel.com>
Clean up all clippy warning.
Also fix a bug in dealing with IFLA_IFNAME attribute.
nlh.addattr_var(IFLA_IFNAME, name.as_ptr() as *const u8, name.len() + 1);
The `name` is a rust String, which doesn't including the trailing '\0',
so name.len() + 1 may cause invalid memory access.
Signed-off-by: Liu Jiang <gerry@linux.alibaba.com>
Implment `TryFrom<IPAddress> for RtIPAddr` instead of From<IPAddress>,
so error code could be returned instead of unwrap().
Do the same for `TryFrom<Route> for RtRoute`.
Signed-off-by: Liu Jiang <gerry@linux.alibaba.com>
There are too much unsafe code in the netlink crate, we need to reduce
unsafe code as much as possible. To achieve this, methods are classified
as public interfaces and internal methods.
All public interface of RtnlHandle has been reimplemented as safe code,
only some public helper functions to manipulater Netlink message data
structures are implemented as unsafe code.
The code to parse IPv4/IPv6/MAC addresses has been moved to a dedicated
file named parser.rs.
Signed-off-by: Liu Jiang <gerry@linux.alibaba.com>
The scan_fmt crate has dependency on other four crates, and it's trivial
to use std library to implement the same logic. Get rid of scan_fmt to
reduce the dependency chain.
Signed-off-by: Liu Jiang <gerry@linux.alibaba.com>
Use features to enable/disable slog and agent handler on demand.
This helps to reduce dependency chains if slog/agent handler is unused.
Signed-off-by: Liu Jiang <gerry@linux.alibaba.com>
The netlink crate is a library to communicate with Linux kenrel by using
the netlink socket. It's generic enough to be reused by other clients.
So get rid of dependency on the rustjail crate by:
1) normalize all pub interfaces to return Result<T, nix::Error>,
2) add helpers to reduce duplicated code,
3) move parse_mac() into lib.rs,
Signed-off-by: Liu Jiang <gerry@linux.alibaba.com>
Add grpc API for adding arp neighbours for a network
interface. These are expected to be static arp entries
sent by the runtime.
Signed-off-by: Tim Zhang <tim@hyper.sh>
Got:
make PROTO_FILE=agent.proto generate-protocol
google/protobuf/descriptor.proto: Read access is denied for file: /usr/local/include/google/protobuf/descriptor.proto
This commit will fix this error.
Signed-off-by: Tim Zhang <tim@hyper.sh>
