| Age | Commit message (Collapse) | Author |
|---|
|
Add a coverage-report flag that will cause the sandbox to generate a coverage
report (with suffix .cov) in the debug log directory upon exiting. For the
report to be generated, runsc must have been built with the following Bazel
flags: `--collect_code_coverage --instrumentation_filter=...`.
With coverage reports, we should be able to aggregate results across all tests
to surface code coverage statistics for the project as a whole.
The report is simply a text file with each line representing a covered block
as `file:start_line.start_col,end_line.end_col`. Note that this is similar to
the format of coverage reports generated with `go test -coverprofile`,
although we omit the count and number of statements, which are not useful for
us.
Some simple ways of getting coverage reports:
bazel test <some_test> --collect_code_coverage \
--instrumentation_filter=//pkg/...
bazel build //runsc --collect_code_coverage \
--instrumentation_filter=//pkg/...
runsc -coverage-report=dir/ <other_flags> do ...
PiperOrigin-RevId: 368952911
|
|
Allow user mounting a verity fs on an existing mount by specifying mount
flags root_hash and lower_path.
PiperOrigin-RevId: 366843846
|
|
A skeleton implementation of cgroupfs. It supports trivial cpu and
memory controllers with no support for hierarchies.
PiperOrigin-RevId: 366561126
|
|
Implement a new runsc command to set up a sandbox with verityfs and
run the measure tool. This is loosely forked from the do command, and
currently requires the caller to provide the measure tool binary.
PiperOrigin-RevId: 366553769
|
|
VFS1 skips over mounts that overrides files in /dev because the list of
files is hardcoded. This is not needed for VFS2 and a recent change
lifted this restriction. However, parts of the code were still skipping
/dev mounts even in VFS2, causing the loader to panic when it ran short
of FDs to connect to the gofer.
PiperOrigin-RevId: 365858436
|
|
--file-access-mounts flag is similar to --file-access, but controls
non-root mounts that were previously mounted in shared mode only.
This gives more flexibility to control how mounts are shared within
a container.
PiperOrigin-RevId: 364669882
|
|
containerd usually configures both /dev and /dev/shm as tmpfs mounts, e.g.:
```
"mounts": [
...
{
"destination": "/dev",
"type": "tmpfs",
"source": "/run/containerd/io.containerd.runtime.v2.task/moby/10eedbd6a0e7937ddfcab90f2c25bd9a9968b734c4ae361318142165d445e67e/tmpfs",
"options": [
"nosuid",
"strictatime",
"mode=755",
"size=65536k"
]
},
...
{
"destination": "/dev/shm",
"type": "tmpfs",
"source": "/run/containerd/io.containerd.runtime.v2.task/moby/10eedbd6a0e7937ddfcab90f2c25bd9a9968b734c4ae361318142165d445e67e/shm",
"options": [
"nosuid",
"noexec",
"nodev",
"mode=1777",
"size=67108864"
]
},
...
```
(This is mostly consistent with how Linux is usually configured, except that
/dev is conventionally devtmpfs, not regular tmpfs. runc/libcontainer
implements OCI-runtime-spec-undocumented behavior to create
/dev/{ptmx,fd,stdin,stdout,stderr} in non-bind /dev mounts. runsc silently
switches /dev to devtmpfs. In VFS1, this is necessary to get device files like
/dev/null at all, since VFS1 doesn't support real device special files, only
what is hardcoded in devfs. VFS2 does support device special files, but using
devtmpfs is the easiest way to get pre-created files in /dev.)
runsc ignores many /dev submounts in the spec, including /dev/shm. In VFS1,
this appears to be to avoid introducing a submount overlay for /dev, and is
mostly fine since the typical mode for the /dev/shm mount is ~consistent with
the mode of the /dev/shm directory provided by devfs (modulo the sticky bit).
In VFS2, this is vestigial (VFS2 does not use submount overlays), and devtmpfs'
/dev/shm mode is correct for the mount point but not the mount. So turn off
this behavior for VFS2.
After this change:
```
$ docker run --rm -it ubuntu:focal ls -lah /dev/shm
total 0
drwxrwxrwt 2 root root 40 Mar 18 00:16 .
drwxr-xr-x 5 root root 360 Mar 18 00:16 ..
$ docker run --runtime=runsc --rm -it ubuntu:focal ls -lah /dev/shm
total 0
drwxrwxrwx 1 root root 0 Mar 18 00:16 .
dr-xr-xr-x 1 root root 0 Mar 18 00:16 ..
$ docker run --runtime=runsc-vfs2 --rm -it ubuntu:focal ls -lah /dev/shm
total 0
drwxrwxrwt 2 root root 40 Mar 18 00:16 .
drwxr-xr-x 5 root root 320 Mar 18 00:16 ..
```
Fixes #5687
PiperOrigin-RevId: 363699385
|
|
The syscall package has been deprecated in favor of golang.org/x/sys.
Note that syscall is still used in some places because the following don't seem
to have an equivalent in unix package:
- syscall.SysProcIDMap
- syscall.Credential
Updates #214
PiperOrigin-RevId: 361381490
|
|
Syzkaller hosts contains many audit messages that runsc tries
to call the clock_nanosleep syscall.
PiperOrigin-RevId: 359331413
|
|
Previously, loader.signalProcess was inconsitently using both root and
container's PID namespace to find the process. It used root namespace
for the exec'd process and container's PID namespace for other processes.
This fixes the code to use the root PID namespace across the board, which
is the same PID reported in `runsc ps` (or soon will after
https://github.com/google/gvisor/pull/5519).
PiperOrigin-RevId: 358836297
|
|
Operations are now shut down automatically by the main Stop
command, and it is not necessary to call Stop during Destroy.
Fixes #5454
PiperOrigin-RevId: 357295930
|
|
Some versions of the Go runtime call getcpu(), so add it for compatibility. The
hostcpu package already uses getcpu() on arm64.
PiperOrigin-RevId: 355717757
|
|
Because we lack gVisor-internal cgroups, we take the CPU usage of the entire pod
and divide it proportionally according to sentry-internal usage stats.
This fixes `kubectl top pods`, which gets a pod's CPU usage by summing the usage
of its containers.
Addresses #172.
PiperOrigin-RevId: 355229833
|
|
These are primarily simplification and lint mistakes. However, minor
fixes are also included and tests added where appropriate.
PiperOrigin-RevId: 351425971
|
|
Closes #5226
PiperOrigin-RevId: 351259576
|
|
This includes minor fix-ups:
* Handle SIGTERM in runsc debug, to exit gracefully.
* Fix cmd.debug.go opening all profiles as RDONLY.
* Fix the test name in fio_test.go, and encode the block size in the test.
PiperOrigin-RevId: 350205718
|
|
PiperOrigin-RevId: 350159657
|
|
This allows for a model of profiling when you can start collection, and
it will terminate when the sandbox terminates. Without this synchronous
call, it is effectively impossible to collect length blocking and mutex
profiles.
PiperOrigin-RevId: 349483418
|
|
PiperOrigin-RevId: 348055514
|
|
Closes #5048
PiperOrigin-RevId: 348050472
|
|
There are surprisingly few syscall tests that run with hostinet. For example
running the following command only returns two results:
`bazel query test/syscalls:all | grep hostnet`
I think as a result, as our control messages evolved, hostinet was left
behind. Update it to support all control messages netstack supports.
This change also updates sentry's control message parsing logic to make it up to
date with all the control messages we support.
PiperOrigin-RevId: 347508892
|
|
PiperOrigin-RevId: 347047550
|
|
PiperOrigin-RevId: 346101076
|
|
Fixes #4991
PiperOrigin-RevId: 345800333
|
|
Closes #4022
PiperOrigin-RevId: 343378647
|
|
Fixes #2714
PiperOrigin-RevId: 342950412
|
|
- Make AddressableEndpoint optional for NetworkEndpoint.
Not all NetworkEndpoints need to support addressing (e.g. ARP), so
AddressableEndpoint should only be implemented for protocols that
support addressing such as IPv4 and IPv6.
With this change, tcpip.ErrNotSupported will be returned by the stack
when attempting to modify addresses on a network endpoint that does
not support addressing.
Now that packets are fully handled at the network layer, and (with this
change) addresses are optional for network endpoints, we no longer need
the workaround for ARP where a fake ARP address was added to each NIC
that performs ARP so that packets would be delivered to the ARP layer.
PiperOrigin-RevId: 342722547
|
|
When OOM score adjustment needs to be set, all the containers need to be
loaded to find all containers that belong to the sandbox. However, each
load signals the container to ensure it is still alive. OOM score
adjustment is set during creation and deletion of every container, generating
a flood of signals to all containers. The fix removes the signal check
when it's not needed.
There is also a race fetching OOM score adjustment value from the parent when
the sandbox exits at the same time (the time it took to signal containers above
made this window quite large). The fix is to store the original value
in the sandbox state file and use it when the value needs to be restored.
Also add more logging and made the existing ones more consistent to help with
debugging.
PiperOrigin-RevId: 340940799
|
|
PiperOrigin-RevId: 339721152
|
|
Also refactor the template and CheckedObject interface to make this cleaner.
Updates #1486.
PiperOrigin-RevId: 339577120
|
|
PiperOrigin-RevId: 339363816
|
|
This is useful to optionally set /dev ro,noexec.
Treat /dev and /dev/pts the same as /proc and /sys.
Make sure the Type is right though. Many config.json snippets
on the Internet suggest /dev is tmpfs, not devtmpfs.
|
|
Inode number consistency checks are now skipped in save/restore tests for
reasons described in greatest detail in StatTest.StateDoesntChangeAfterRename.
They pass in VFS1 due to the bug described in new test case
SimpleStatTest.DifferentFilesHaveDifferentDeviceInodeNumberPairs.
Fixes #1663
PiperOrigin-RevId: 338776148
|
|
Our current reference leak checker uses finalizers to verify whether an object
has reached zero references before it is garbage collected. There are multiple
problems with this mechanism, so a rewrite is in order.
With finalizers, there is no way to guarantee that a finalizer will run before
the program exits. When an unreachable object with a finalizer is garbage
collected, its finalizer will be added to a queue and run asynchronously. The
best we can do is run garbage collection upon sandbox exit to make sure that
all finalizers are enqueued.
Furthermore, if there is a chain of finalized objects, e.g. A points to B
points to C, garbage collection needs to run multiple times before all of the
finalizers are enqueued. The first GC run will register the finalizer for A but
not free it. It takes another GC run to free A, at which point B's finalizer
can be registered. As a result, we need to run GC as many times as the length
of the longest such chain to have a somewhat reliable leak checker.
Finally, a cyclical chain of structs pointing to one another will never be
garbage collected if a finalizer is set. This is a well-known issue with Go
finalizers (https://github.com/golang/go/issues/7358). Using leak checking on
filesystem objects that produce cycles will not work and even result in memory
leaks.
The new leak checker stores reference counted objects in a global map when
leak check is enabled and removes them once they are destroyed. At sandbox
exit, any remaining objects in the map are considered as leaked. This provides
a deterministic way of detecting leaks without relying on the complexities of
finalizers and garbage collection.
This approach has several benefits over the former, including:
- Always detects leaks of objects that should be destroyed very close to
sandbox exit. The old checker very rarely detected these leaks, because it
relied on garbage collection to be run in a short window of time.
- Panics if we forgot to enable leak check on a ref-counted object (we will try
to remove it from the map when it is destroyed, but it will never have been
added).
- Can store extra logging information in the map values without adding to the
size of the ref count struct itself. With the size of just an int64, the ref
count object remains compact, meaning frequent operations like IncRef/DecRef
are more cache-efficient.
- Can aggregate leak results in a single report after the sandbox exits.
Instead of having warnings littered in the log, which were
non-deterministically triggered by garbage collection, we can print all
warning messages at once. Note that this could also be a limitation--the
sandbox must exit properly for leaks to be detected.
Some basic benchmarking indicates that this change does not significantly
affect performance when leak checking is enabled, which is understandable
since registering/unregistering is only done once for each filesystem object.
Updates #1486.
PiperOrigin-RevId: 338685972
|
|
Subcontainers are only configured when the container starts, however because
start doesn't load the spec, flag annotations that may override flags were
not getting applied to the configuration.
Updates #3494
PiperOrigin-RevId: 338610953
|
|
PiperOrigin-RevId: 338072845
|
|
- Check the sticky bit in overlay.filesystem.UnlinkAt(). Fixes
StickyTest.StickyBitPermDenied.
- When configuring a VFS2 overlay in runsc, copy the lower layer's root
owner/group/mode to the upper layer's root (as in the VFS1 equivalent,
boot.addOverlay()). This makes the overlay root owned by UID/GID 65534 with
mode 0755 rather than owned by UID/GID 0 with mode 01777. Fixes
CreateTest.CreateFailsOnUnpermittedDir, which assumes that the test cannot
create files in /.
- MknodTest.UnimplementedTypesReturnError assumes that the creation of device
special files is not supported. However, while the VFS2 gofer client still
doesn't support device special files, VFS2 tmpfs does, and in the overlay
test dimension mknod() targets a tmpfs upper layer. The test initially has
all capabilities, including CAP_MKNOD, so its creation of these files
succeeds. Constrain these tests to VFS1.
- Rename overlay.nonDirectoryFD to overlay.regularFileFD and only use it for
regular files, using the original FD for pipes and device special files. This
is more consistent with Linux (which gets the original inode_operations, and
therefore file_operations, for these file types from ovl_fill_inode() =>
init_special_inode()) and fixes remaining mknod and pipe tests.
- Read/write 1KB at a time in PipeTest.Streaming, rather than 4 bytes. This
isn't strictly necessary, but it makes the test less obnoxiously slow on
ptrace.
Fixes #4407
PiperOrigin-RevId: 337971042
|
|
Using the newer bazel rules necessitates a transition from proto1 to
proto2. In order to resolve the incompatibility between proto2 and
gogoproto, the cri runtimeoptions proto must be vendored.
Further, some of the semantics of bazel caching changed during the
transition. It is now necessary to:
- Ensure that :gopath depends only on pure library targets, as the
propagation of go_binary build attributes (pure, static) will
affected the generated files (though content remains the same,
there are conflicts with respect to the gopath).
- Update bazel.mk to include the possibility of binaries in the
bazel-out directory, as it will now put runsc and others there.
This required some refinements to the mechanism of extracting
paths, since some the existing regex resulted in false positives.
- Change nogo rules to prevent escape generation on binary targets.
For some reason, the newer version of bazel attempted to run the
nogo analysis on the binary targets, which fails due to the fact
that objdump does not work on the final binary. This must be due
to a change in the semantics of aspects in bazel3.
PiperOrigin-RevId: 337958324
|
|
This fixes reference leaks related to accidentally forgetting to DecRef()
after calling one or the other.
PiperOrigin-RevId: 336918922
|
|
PiperOrigin-RevId: 336694658
|
|
Updates #267
PiperOrigin-RevId: 335713923
|
|
When all container tasks finish, they release the mount which in turn
will close the 9P session to the gofer. The gofer exits when the connection
closes, triggering the gofer monitor. The gofer monitor will _think_ that
the gofer died prematurely and destroy the container. Then when the caller
attempts to wait for the container, e.g. to get the exit code, wait fails
saying the container doesn't exist.
Gofer monitor now just SIGKILLs the container, and let the normal teardown
process to happen, which will evetually destroy the container at the right
time. Also, fixed an issue with exec racing with container's init process
exiting.
Closes #1487
PiperOrigin-RevId: 335537350
|
|
PiperOrigin-RevId: 335516972
|
|
Updates #1487
PiperOrigin-RevId: 335516732
|
|
With cgroups configured NumCPU is correct, however GOMAXPROCS is still derived from total host core count and ignores cgroup restrictions. This can lead to different and undesired behavior across different hosts.
For example, the total number of threads in the guest process will be larger on machines with more cores.
This change configures the go runtime for the sandbox to only use the number of threads consistent with its restrictions.
|
|
Network or transport protocols may want to reach the stack. Support this
by letting the stack create the protocol instances so it can pass a
reference to itself at protocol creation time.
Note, protocols do not yet use the stack in this CL but later CLs will
make use of the stack from protocols.
PiperOrigin-RevId: 334260210
|
|
https://go.googlesource.com/go/+/0941fc3 switches the Go runtime (on amd64)
from using arch_prctl(ARCH_SET_FS) to CLONE_SETTLS to set the TLS.
PiperOrigin-RevId: 333100550
|
|
Neither CLONE_PARENT_SETTID nor CLONE_CHILD_SETTID are used, so these arguments
will always be NULL.
PiperOrigin-RevId: 333085326
|
|
Go does not call arch_prctl(ARCH_GET_FS), nor am I sure it ever did. Drop the
filter.
PiperOrigin-RevId: 332470532
|
|
All tests under runsc are passing with overlay enabled.
Updates #1487, #1199
PiperOrigin-RevId: 332181267
|
