Never ever try to close the same fd double times, even in a unit test.
A file descriptor is a number which will be reused, so when you close
the same number twice you may close another file descriptor in the second
time and then there will be an error 'Bad file descriptor (os error 9)'
while the wrongly closed fd is being used.
Fixes: #6679
Signed-off-by: Tim Zhang <tim@hyper.sh>
In cases where the D-Bus connection fails, add a little additional context about
the origin of the error.
Fixes: 6561
Signed-off-by: Christophe de Dinechin <dinechin@redhat.com>
Suggested-by: Archana Shinde <archana.m.shinde@intel.com>
Spell-checked-by: Greg Kurz <gkurz@redhat.com>
There can be an error while connecting to the cgroups managager, for
example a `ENOENT` if a file is not found. Make sure that this is
reported through the proper channels instead of causing a `panic()`
that does not provide much information.
Fixes: #6561
Signed-off-by: Christophe de Dinechin <dinechin@redhat.com>
Reported-by: Greg Kurz <gkurz@redhat.com>
This PR is a continuing work for (kata-containers#3679).
This generalizes the previous VFIO device handling which only
focuses on PCI to include AP (IBM Z specific).
Fixes: kata-containers#3678
Signed-off-by: Hyounggyu Choi <Hyounggyu.Choi@ibm.com>
Initial VFIO-AP support (#578) was simple, but somewhat hacky; a
different code path would be chosen for performing the hotplug, and
agent-side device handling was bound to knowing the assigned queue
numbers (APQNs) through some other means; plus the code for awaiting
them was written for the Go agent and never released. This code also
artificially increased the hotplug timeout to wait for the (relatively
expensive, thus limited to 5 seconds at the quickest) AP rescan, which
is impractical for e.g. common k8s timeouts.
Since then, the general handling logic was improved (#1190), but it
assumed PCI in several places.
In the runtime, introduce and parse AP devices. Annotate them as such
when passing to the agent, and include information about the associated
APQNs.
The agent awaits the passed APQNs through uevents and triggers a
rescan directly.
Fixes: #3678
Signed-off-by: Jakob Naucke <jakob.naucke@ibm.com>
e.g., split_vfio_option is PCI-specific and should instead be named
split_vfio_pci_option. This mutually affects the runtime, most notably
how the labels are named for the agent.
Signed-off-by: Jakob Naucke <jakob.naucke@ibm.com>
- implement update_ephemeral_mounts rpc
- for each mountpoint passed in, remount it with new options
Signed-off-by: Sidhartha Mani <sidhartha_mani@apple.com>
In this commit, we have done:
* modify the tranfer process from grpc::Hooks to oci::Hooks, so the code
can be more clean
* add more tests for create_runtime, create_container, start_container hooks
Signed-off-by: Yushuo <y-shuo@linux.alibaba.com>
StartContainer will be execute in guest container namespace in Kata.
The Hook Path of this kind of hook is also in guest container namespace.
StartContainer is executed after start operation is called, and it
should be executed before user-specific command is executed.
Fixes: #5787
Signed-off-by: Yushuo <y-shuo@linux.alibaba.com>
CreateContainer hook is one kind of OCI hook. In kata, it will be
executed after VM is started, before container is created, and after
CreateRuntime is executed.
The hook path of CreateContainer hook is in host runtime namespace, but
it will be executed in host vmm namespace.
Fixes: #5787
Signed-off-by: Yushuo <y-shuo@linux.alibaba.com>
According to the runtime OCI Spec, there can be some hook
operations in the lifecycle of the container. In these hook
operations, the runtime can execute some commands. There are different
points in time in the container lifecycle and different hook types
can be executed.
In this commit, we are now supporting 4 types of hooks(same in
runtime-go): Prestart hook, CreateRuntime hook, Poststart hook and
Poststop hook.
Fixes: #5787
Signed-off-by: Yushuo <y-shuo@linux.alibaba.com>
The logic to decide which cgroup driver is used is currently based on the
cgroup path that the host provides. This requires host and guest to use the
same cgroup driver. If the guest uses kata-agent as init, then systemd can't be
used as the cgroup driver. If the host requests a systemd cgroup, this
currently results in a rustjail panic:
thread 'tokio-runtime-worker' panicked at 'called `Result::unwrap()` on an `Err` value: I/O error: No such file or directory (os error 2)
Caused by:
No such file or directory (os error 2)', rustjail/src/cgroups/systemd/manager.rs:44:51
stack backtrace:
0: 0x7ff0fe77a793 - std::backtrace_rs::backtrace::libunwind::trace::h8c197fa9a679d134
at /rustc/69f9c33d71c871fc16ac445211281c6e7a340943/library/std/src/../../backtrace/src/backtrace/libunwind.rs:93:5
1: 0x7ff0fe77a793 - std::backtrace_rs::backtrace::trace_unsynchronized::h9ee19d58b6d5934a
at /rustc/69f9c33d71c871fc16ac445211281c6e7a340943/library/std/src/../../backtrace/src/backtrace/mod.rs:66:5
2: 0x7ff0fe77a793 - std::sys_common::backtrace::_print_fmt::h4badc450600fc417
at /rustc/69f9c33d71c871fc16ac445211281c6e7a340943/library/std/src/sys_common/backtrace.rs:65:5
3: 0x7ff0fe77a793 - <std::sys_common::backtrace::_print::DisplayBacktrace as core::fmt::Display>::fmt::had334ddb529a2169
at /rustc/69f9c33d71c871fc16ac445211281c6e7a340943/library/std/src/sys_common/backtrace.rs:44:22
4: 0x7ff0fdce815e - core::fmt::write::h1aa7694f03e44db2
at /rustc/69f9c33d71c871fc16ac445211281c6e7a340943/library/core/src/fmt/mod.rs:1209:17
5: 0x7ff0fe74e0c4 - std::io::Write::write_fmt::h61b2bdc565be41b5
at /rustc/69f9c33d71c871fc16ac445211281c6e7a340943/library/std/src/io/mod.rs:1682:15
6: 0x7ff0fe77cd3f - std::sys_common::backtrace::_print::h4ec69798b72ff254
at /rustc/69f9c33d71c871fc16ac445211281c6e7a340943/library/std/src/sys_common/backtrace.rs:47:5
7: 0x7ff0fe77cd3f - std::sys_common::backtrace::print::h0e6c02048dec3c77
at /rustc/69f9c33d71c871fc16ac445211281c6e7a340943/library/std/src/sys_common/backtrace.rs:34:9
8: 0x7ff0fe77c93f - std::panicking::default_hook::{{closure}}::hcdb7e705dc37ea6e
at /rustc/69f9c33d71c871fc16ac445211281c6e7a340943/library/std/src/panicking.rs:267:22
9: 0x7ff0fe77d9b8 - std::panicking::default_hook::he03a933a0f01790f
at /rustc/69f9c33d71c871fc16ac445211281c6e7a340943/library/std/src/panicking.rs:286:9
10: 0x7ff0fe77d9b8 - std::panicking::rust_panic_with_hook::he26b680bfd953008
at /rustc/69f9c33d71c871fc16ac445211281c6e7a340943/library/std/src/panicking.rs:688:13
11: 0x7ff0fe77d482 - std::panicking::begin_panic_handler::{{closure}}::h559120d2dd1c6180
at /rustc/69f9c33d71c871fc16ac445211281c6e7a340943/library/std/src/panicking.rs:579:13
12: 0x7ff0fe77d3ec - std::sys_common::backtrace::__rust_end_short_backtrace::h36db621fc93b005a
at /rustc/69f9c33d71c871fc16ac445211281c6e7a340943/library/std/src/sys_common/backtrace.rs:137:18
13: 0x7ff0fe77d3c1 - rust_begin_unwind
at /rustc/69f9c33d71c871fc16ac445211281c6e7a340943/library/std/src/panicking.rs:575:5
14: 0x7ff0fda52ee2 - core::panicking::panic_fmt::he7679b415d25c5f4
at /rustc/69f9c33d71c871fc16ac445211281c6e7a340943/library/core/src/panicking.rs:65:14
15: 0x7ff0fda53182 - core::result::unwrap_failed::hb71caff146724b6b
at /rustc/69f9c33d71c871fc16ac445211281c6e7a340943/library/core/src/result.rs:1791:5
16: 0x7ff0fe5bd738 - <rustjail::cgroups::systemd::manager::Manager as rustjail::cgroups::Manager>::apply::hd46958d9d807d2ca
17: 0x7ff0fe606d80 - <rustjail::container::LinuxContainer as rustjail::container::BaseContainer>::start::{{closure}}::h1de806d91fcb878f
18: 0x7ff0fe604a76 - <core::future::from_generator::GenFuture<T> as core::future::future::Future>::poll::h1749c148adcc235f
19: 0x7ff0fdc0c992 - kata_agent::rpc::AgentService::do_create_container::{{closure}}::{{closure}}::hc1b87a15dfdf2f64
20: 0x7ff0fdb80ae4 - <core::future::from_generator::GenFuture<T> as core::future::future::Future>::poll::h846a8c9e4fb67707
21: 0x7ff0fe3bb816 - <core::future::from_generator::GenFuture<T> as core::future::future::Future>::poll::h53de16ff66ed3972
22: 0x7ff0fdb519cb - <core::future::from_generator::GenFuture<T> as core::future::future::Future>::poll::h1cbece980286c0f4
23: 0x7ff0fdf4019c - <tokio::future::poll_fn::PollFn<F> as core::future::future::Future>::poll::hc8e72d155feb8d1f
24: 0x7ff0fdfa5fd8 - tokio::loom::std::unsafe_cell::UnsafeCell<T>::with_mut::h0a407ffe2559449a
25: 0x7ff0fdf033a1 - tokio::runtime::task::raw::poll::h1045d9f1db9742de
26: 0x7ff0fe7a8ce2 - tokio::runtime::scheduler::multi_thread::worker::Context::run_task::h4924ae3464af7fbd
27: 0x7ff0fe7afb85 - tokio::runtime::task::raw::poll::h5c843be39646b833
28: 0x7ff0fe7a05ee - std::sys_common::backtrace::__rust_begin_short_backtrace::ha7777c55b98a9bd1
29: 0x7ff0fe7a9bdb - core::ops::function::FnOnce::call_once{{vtable.shim}}::h27ec83c953360cdd
30: 0x7ff0fe7801d5 - <alloc::boxed::Box<F,A> as core::ops::function::FnOnce<Args>>::call_once::hed812350c5aef7a8
at /rustc/69f9c33d71c871fc16ac445211281c6e7a340943/library/alloc/src/boxed.rs:1987:9
31: 0x7ff0fe7801d5 - <alloc::boxed::Box<F,A> as core::ops::function::FnOnce<Args>>::call_once::hc7df8e435a658960
at /rustc/69f9c33d71c871fc16ac445211281c6e7a340943/library/alloc/src/boxed.rs:1987:9
32: 0x7ff0fe7801d5 - std::sys::unix:🧵:Thread:🆕:thread_start::h575491a8a17dbb33
at /rustc/69f9c33d71c871fc16ac445211281c6e7a340943/library/std/src/sys/unix/thread.rs:108:17
Forward the value of "init_mode" to AgentService, so that we can force cgroupfs
when systemd is unavailable.
Fixes: #5779
Signed-off-by: Jeremi Piotrowski <jpiotrowski@microsoft.com>
Right now LinuxContainer::new() gets passed a CreateOpts struct, but then
modifies the use_systemd_cgroup field inside that struct. Pull the cgroups path
parsing logic into do_create_container, so that CreateOpts can be immutable in
LinuxContainer::new. This is just moving things around, there should be no
functional changes.
Signed-off-by: Jeremi Piotrowski <jpiotrowski@microsoft.com>
This patch fixes the issue that do_copy_file changes
the directory permission of the parent directory of
a target file, even when the parent directory already
exists.
Fixes#6367
Signed-off-by: Yohei Ueda <yohei@jp.ibm.com>
Since the cgroup manager is wrapped in a dyn now, the print in
LinuxContainer::new has been useless and just says "CgroupManager". Extend the
Debug trait for 'dyn Manager' to print the type of the cgroup manager so that
it's easier to debug issues.
Fixes: #5779
Signed-off-by: Jeremi Piotrowski <jpiotrowski@microsoft.com>
We should make sure the dns's source file's parent
directory exist, otherwise, it would failed to create
the file directly.
Signed-off-by: Fupan Li <fupan.lfp@antgroup.com>
Trait method cause for std::error::Error is deprecated thus need replace
it with source method for cgroups-fs::error::ErrorKind.
Fixes: #6192
Signed-off-by: Jianyong Wu <jianyong.wu@arm.com>
There are lots of unit test cases fails regularly on aarch64, including
TestIOCopy, create_tmpfs. Temporarily skip it for now and enable it
after them get fixed.
Fixes: #6194
Signed-off-by: Jianyong Wu <jianyong.wu@arm.com>
We have to execute some hooks both in host and guest. And in
/libs/kata-sys-util/src/hooks.rs, the coomon operations are implemented.
In this commit, we are going to refactor the code of guest hooks using
code in /libs/kata-sys-util/src/hooks.rs. At the same time, we move
function valid_env to kata-sys-util to make it usable by both agent and
runtime.
Fixes: #5857
Signed-off-by: Yushuo <y-shuo@linux.alibaba.com>
DEFAULT_REGISTRY pre-registers many metrics that we don't need or have duplicated.
This PR uses a custom register for metrics without interference and ensures that
the registration process is executed only once when the program is running.
Fixes: #5255
Signed-off-by: Yuan-Zhuo <yuanzhuo0118@outlook.com>
Cgroup manager for a container will always be created.
Thus, dropping the option for LinuxContainer.cgroup_manager
is feasible and could simplify the code.
Fixes: #5778
Signed-off-by: Yuan-Zhuo <yuanzhuo0118@outlook.com>
The kata-agent supports SELinux for containers inside the guest
to comply with the OCI runtime specification.
Fixes: #4812
Signed-off-by: Manabu Sugimoto <Manabu.Sugimoto@sony.com>
