Add option to skip stuck tasks waiting for address space

PiperOrigin-RevId: 297192390

8 files changed, 41 insertions, 25 deletions

diff --git a/pkg/sentry/kernel/kernel.go b/pkg/sentry/kernel/kernel.go
index c62fd6eb1..8b76750e9 100644
--- a/pkg/sentry/kernel/kernel.go
+++ b/pkg/sentry/kernel/kernel.go
@@ -247,6 +247,10 @@ type Kernel struct {
 
 	// VFS keeps the filesystem state used across the kernel.
 	vfs vfs.VirtualFilesystem
+
+	// If set to true, report address space activation waits as if the task is in
+	// external wait so that the watchdog doesn't report the task stuck.
+	SleepForAddressSpaceActivation bool
 }
 
 // InitKernelArgs holds arguments to Init.
diff --git a/pkg/sentry/kernel/task_context.go b/pkg/sentry/kernel/task_context.go
index 2be982684..0158b1788 100644
--- a/pkg/sentry/kernel/task_context.go
+++ b/pkg/sentry/kernel/task_context.go
@@ -140,7 +140,7 @@ func (k *Kernel) LoadTaskImage(ctx context.Context, args loader.LoadArgs) (*Task
 	}
 
 	// Prepare a new user address space to load into.
-	m := mm.NewMemoryManager(k, k)
+	m := mm.NewMemoryManager(k, k, k.SleepForAddressSpaceActivation)
 	defer m.DecUsers(ctx)
 	args.MemoryManager = m
 
diff --git a/pkg/sentry/kernel/task_exec.go b/pkg/sentry/kernel/task_exec.go
index 8f57a34a6..00c425cca 100644
--- a/pkg/sentry/kernel/task_exec.go
+++ b/pkg/sentry/kernel/task_exec.go
@@ -220,7 +220,7 @@ func (r *runSyscallAfterExecStop) execute(t *Task) taskRunState {
 	t.mu.Unlock()
 	t.unstopVforkParent()
 	// NOTE(b/30316266): All locks must be dropped prior to calling Activate.
-	t.MemoryManager().Activate()
+	t.MemoryManager().Activate(t)
 
 	t.ptraceExec(oldTID)
 	return (*runSyscallExit)(nil)
diff --git a/pkg/sentry/kernel/task_usermem.go b/pkg/sentry/kernel/task_usermem.go
index 2bf3ce8a8..b02044ad2 100644
--- a/pkg/sentry/kernel/task_usermem.go
+++ b/pkg/sentry/kernel/task_usermem.go
@@ -30,7 +30,7 @@ var MAX_RW_COUNT = int(usermem.Addr(math.MaxInt32).RoundDown())
 // Activate ensures that the task has an active address space.
 func (t *Task) Activate() {
 	if mm := t.MemoryManager(); mm != nil {
-		if err := mm.Activate(); err != nil {
+		if err := mm.Activate(t); err != nil {
 			panic("unable to activate mm: " + err.Error())
 		}
 	}
diff --git a/pkg/sentry/mm/address_space.go b/pkg/sentry/mm/address_space.go
index 94d39af60..0332fc71c 100644
--- a/pkg/sentry/mm/address_space.go
+++ b/pkg/sentry/mm/address_space.go
@@ -18,6 +18,7 @@ import (
 	"fmt"
 	"sync/atomic"
 
+	"gvisor.dev/gvisor/pkg/context"
 	"gvisor.dev/gvisor/pkg/sentry/platform"
 	"gvisor.dev/gvisor/pkg/usermem"
 )
@@ -38,7 +39,7 @@ func (mm *MemoryManager) AddressSpace() platform.AddressSpace {
 //
 // When this MemoryManager is no longer needed by a task, it should call
 // Deactivate to release the reference.
-func (mm *MemoryManager) Activate() error {
+func (mm *MemoryManager) Activate(ctx context.Context) error {
 	// Fast path: the MemoryManager already has an active
 	// platform.AddressSpace, and we just need to indicate that we need it too.
 	for {
@@ -91,16 +92,20 @@ func (mm *MemoryManager) Activate() error {
 		if as == nil {
 			// AddressSpace is unavailable, we must wait.
 			//
-			// activeMu must not be held while waiting, as the user
-			// of the address space we are waiting on may attempt
-			// to take activeMu.
-			//
-			// Don't call UninterruptibleSleepStart to register the
-			// wait to allow the watchdog stuck task to trigger in
-			// case a process is starved waiting for the address
-			// space.
+			// activeMu must not be held while waiting, as the user of the address
+			// space we are waiting on may attempt to take activeMu.
 			mm.activeMu.Unlock()
+
+			sleep := mm.p.CooperativelySchedulesAddressSpace() && mm.sleepForActivation
+			if sleep {
+				// Mark this task sleeping while waiting for the address space to
+				// prevent the watchdog from reporting it as a stuck task.
+				ctx.UninterruptibleSleepStart(false)
+			}
 			<-c
+			if sleep {
+				ctx.UninterruptibleSleepFinish(false)
+			}
 			continue
 		}
 
diff --git a/pkg/sentry/mm/lifecycle.go b/pkg/sentry/mm/lifecycle.go
index 3c263ebaa..d8a5b9d29 100644
--- a/pkg/sentry/mm/lifecycle.go
+++ b/pkg/sentry/mm/lifecycle.go
@@ -28,16 +28,17 @@ import (
 )
 
 // NewMemoryManager returns a new MemoryManager with no mappings and 1 user.
-func NewMemoryManager(p platform.Platform, mfp pgalloc.MemoryFileProvider) *MemoryManager {
+func NewMemoryManager(p platform.Platform, mfp pgalloc.MemoryFileProvider, sleepForActivation bool) *MemoryManager {
 	return &MemoryManager{
-		p:           p,
-		mfp:         mfp,
-		haveASIO:    p.SupportsAddressSpaceIO(),
-		privateRefs: &privateRefs{},
-		users:       1,
-		auxv:        arch.Auxv{},
-		dumpability: UserDumpable,
-		aioManager:  aioManager{contexts: make(map[uint64]*AIOContext)},
+		p:                  p,
+		mfp:                mfp,
+		haveASIO:           p.SupportsAddressSpaceIO(),
+		privateRefs:        &privateRefs{},
+		users:              1,
+		auxv:               arch.Auxv{},
+		dumpability:        UserDumpable,
+		aioManager:         aioManager{contexts: make(map[uint64]*AIOContext)},
+		sleepForActivation: sleepForActivation,
 	}
 }
 
@@ -79,9 +80,10 @@ func (mm *MemoryManager) Fork(ctx context.Context) (*MemoryManager, error) {
 		envv:                 mm.envv,
 		auxv:                 append(arch.Auxv(nil), mm.auxv...),
 		// IncRef'd below, once we know that there isn't an error.
-		executable:  mm.executable,
-		dumpability: mm.dumpability,
-		aioManager:  aioManager{contexts: make(map[uint64]*AIOContext)},
+		executable:         mm.executable,
+		dumpability:        mm.dumpability,
+		aioManager:         aioManager{contexts: make(map[uint64]*AIOContext)},
+		sleepForActivation: mm.sleepForActivation,
 	}
 
 	// Copy vmas.
diff --git a/pkg/sentry/mm/mm.go b/pkg/sentry/mm/mm.go
index 637383c7a..c2195ae11 100644
--- a/pkg/sentry/mm/mm.go
+++ b/pkg/sentry/mm/mm.go
@@ -226,6 +226,11 @@ type MemoryManager struct {
 	// aioManager keeps track of AIOContexts used for async IOs. AIOManager
 	// must be cloned when CLONE_VM is used.
 	aioManager aioManager
+
+	// sleepForActivation indicates whether the task should report to be sleeping
+	// before trying to activate the address space. When set to true, delays in
+	// activation are not reported as stuck tasks by the watchdog.
+	sleepForActivation bool
 }
 
 // vma represents a virtual memory area.
diff --git a/pkg/sentry/mm/mm_test.go b/pkg/sentry/mm/mm_test.go
index edacca741..fdc308542 100644
--- a/pkg/sentry/mm/mm_test.go
+++ b/pkg/sentry/mm/mm_test.go
@@ -31,7 +31,7 @@ import (
 func testMemoryManager(ctx context.Context) *MemoryManager {
 	p := platform.FromContext(ctx)
 	mfp := pgalloc.MemoryFileProviderFromContext(ctx)
-	mm := NewMemoryManager(p, mfp)
+	mm := NewMemoryManager(p, mfp, false)
 	mm.layout = arch.MmapLayout{
 		MinAddr:      p.MinUserAddress(),
 		MaxAddr:      p.MaxUserAddress(),