build(deps): bump github.com/Microsoft/hcsshim from 0.8.20 to 0.9.6

Bumps [github.com/Microsoft/hcsshim](https://github.com/Microsoft/hcsshim) from 0.8.20 to 0.9.6.
- [Release notes](https://github.com/Microsoft/hcsshim/releases)
- [Commits](https://github.com/Microsoft/hcsshim/compare/v0.8.20...v0.9.6)

---
updated-dependencies:
- dependency-name: github.com/Microsoft/hcsshim
  dependency-type: direct:production
  update-type: version-update:semver-minor
...

Signed-off-by: dependabot[bot] <support@github.com>

Signed-off-by: dependabot[bot] <support@github.com>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>
Signed-off-by: Matthieu MOREL <matthieu.morel35@gmail.com>

vendor/github.com/Microsoft/hcsshim/internal/hcs/process.go

@@ -3,7 +3,9 @@ package hcs
 import (
 	"context"
 	"encoding/json"
+	"errors"
 	"io"
+	"os"
 	"sync"
 	"syscall"
 	"time"
@@ -16,16 +18,17 @@ import (
 
 // ContainerError is an error encountered in HCS
 type Process struct {
-	handleLock     sync.RWMutex
-	handle         vmcompute.HcsProcess
-	processID      int
-	system         *System
-	hasCachedStdio bool
-	stdioLock      sync.Mutex
-	stdin          io.WriteCloser
-	stdout         io.ReadCloser
-	stderr         io.ReadCloser
-	callbackNumber uintptr
+	handleLock          sync.RWMutex
+	handle              vmcompute.HcsProcess
+	processID           int
+	system              *System
+	hasCachedStdio      bool
+	stdioLock           sync.Mutex
+	stdin               io.WriteCloser
+	stdout              io.ReadCloser
+	stderr              io.ReadCloser
+	callbackNumber      uintptr
+	killSignalDelivered bool
 
 	closedWaitOnce sync.Once
 	waitBlock      chan struct{}
@@ -149,12 +152,45 @@ func (process *Process) Kill(ctx context.Context) (bool, error) {
 		return false, makeProcessError(process, operation, ErrAlreadyClosed, nil)
 	}
 
+	if process.killSignalDelivered {
+		// A kill signal has already been sent to this process. Sending a second
+		// one offers no real benefit, as processes cannot stop themselves from
+		// being terminated, once a TerminateProcess has been issued. Sending a
+		// second kill may result in a number of errors (two of which detailed bellow)
+		// and which we can avoid handling.
+		return true, nil
+	}
+
 	resultJSON, err := vmcompute.HcsTerminateProcess(ctx, process.handle)
+	if err != nil {
+		// We still need to check these two cases, as processes may still be killed by an
+		// external actor (human operator, OOM, random script etc).
+		if errors.Is(err, os.ErrPermission) || IsAlreadyStopped(err) {
+			// There are two cases where it should be safe to ignore an error returned
+			// by HcsTerminateProcess. The first one is cause by the fact that
+			// HcsTerminateProcess ends up calling TerminateProcess in the context
+			// of a container. According to the TerminateProcess documentation:
+			// https://docs.microsoft.com/en-us/windows/win32/api/processthreadsapi/nf-processthreadsapi-terminateprocess#remarks
+			// After a process has terminated, call to TerminateProcess with open
+			// handles to the process fails with ERROR_ACCESS_DENIED (5) error code.
+			// It's safe to ignore this error here. HCS should always have permissions
+			// to kill processes inside any container. So an ERROR_ACCESS_DENIED
+			// is unlikely to be anything else than what the ending remarks in the
+			// documentation states.
+			//
+			// The second case is generated by hcs itself, if for any reason HcsTerminateProcess
+			// is called twice in a very short amount of time. In such cases, hcs may return
+			// HCS_E_PROCESS_ALREADY_STOPPED.
+			return true, nil
+		}
+	}
 	events := processHcsResult(ctx, resultJSON)
 	delivered, err := process.processSignalResult(ctx, err)
 	if err != nil {
 		err = makeProcessError(process, operation, err, events)
 	}
+
+	process.killSignalDelivered = delivered
 	return delivered, err
 }