Add initial version of renumber backend

Renumber is a way of renumbering container locks after the number
of locks available has changed.

For now, renumber only works with containers.

Signed-off-by: Matthew Heon <matthew.heon@pm.me>

9 files changed, 150 insertions, 7 deletions

diff --git a/libpod/lock/in_memory_locks.go b/libpod/lock/in_memory_locks.go
index db8f20e95..7eb22328f 100644
--- a/libpod/lock/in_memory_locks.go
+++ b/libpod/lock/in_memory_locks.go
@@ -89,3 +89,12 @@ func (m *InMemoryManager) RetrieveLock(id uint32) (Locker, error) {
 
 	return m.locks[id], nil
 }
+
+// FreeAllLocks frees all locks
+func (m *InMemoryManager) FreeAllLocks() error {
+	for _, lock := range m.locks {
+		lock.allocated = false
+	}
+
+	return nil
+}
diff --git a/libpod/lock/lock.go b/libpod/lock/lock.go
index 1f94171fe..b96393574 100644
--- a/libpod/lock/lock.go
+++ b/libpod/lock/lock.go
@@ -24,6 +24,9 @@ type Manager interface {
 	// The underlying lock MUST be the same as another other lock with the
 	// same UUID.
 	RetrieveLock(id uint32) (Locker, error)
+	// FreeAllLocks frees all allocated locks, in preparation for lock
+	// reallocation.
+	FreeAllLocks() error
 }
 
 // Locker is similar to sync.Locker, but provides a method for freeing the lock
diff --git a/libpod/lock/shm/shm_lock.c b/libpod/lock/shm/shm_lock.c
index 4af58d857..367055823 100644
--- a/libpod/lock/shm/shm_lock.c
+++ b/libpod/lock/shm/shm_lock.c
@@ -407,6 +407,36 @@ int32_t deallocate_semaphore(shm_struct_t *shm, uint32_t sem_index) {
   return 0;
 }
 
+// Deallocate all semaphores unconditionally.
+// Returns negative ERRNO values.
+int32_t deallocate_all_semaphores(shm_struct_t *shm) {
+  int ret_code;
+  uint i;
+
+  if (shm == NULL) {
+    return -1 * EINVAL;
+  }
+
+  // Lock the mutex controlling access to our shared memory
+  ret_code = take_mutex(&(shm->segment_lock));
+  if (ret_code != 0) {
+    return -1 * ret_code;
+  }
+
+  // Iterate through all bitmaps and reset to unused
+  for (i = 0; i < shm->num_bitmaps; i++) {
+    shm->locks[i].bitmap = 0;
+  }
+
+  // Unlock the allocation control mutex
+  ret_code = release_mutex(&(shm->segment_lock));
+  if (ret_code != 0) {
+    return -1 * ret_code;
+  }
+
+  return 0;
+}
+
 // Lock a given semaphore
 // Does not check if the semaphore is allocated - this ensures that, even for
 // removed containers, we can still successfully lock to check status (and
diff --git a/libpod/lock/shm/shm_lock.go b/libpod/lock/shm/shm_lock.go
index 87d28e5c1..e70ea8743 100644
--- a/libpod/lock/shm/shm_lock.go
+++ b/libpod/lock/shm/shm_lock.go
@@ -155,6 +155,22 @@ func (locks *SHMLocks) DeallocateSemaphore(sem uint32) error {
 	return nil
 }
 
+// DeallocateAllSemaphores frees all semaphores so they can be reallocated to
+// other containers and pods.
+func (locks *SHMLocks) DeallocateAllSemaphores() error {
+	if !locks.valid {
+		return errors.Wrapf(syscall.EINVAL, "locks have already been closed")
+	}
+
+	retCode := C.deallocate_all_semaphores(locks.lockStruct)
+	if retCode < 0 {
+		// Negative errno return from C
+		return syscall.Errno(-1 * retCode)
+	}
+
+	return nil
+}
+
 // LockSemaphore locks the given semaphore.
 // If the semaphore is already locked, LockSemaphore will block until the lock
 // can be acquired.
diff --git a/libpod/lock/shm/shm_lock.h b/libpod/lock/shm/shm_lock.h
index 8e7e23fb7..58e4297e2 100644
--- a/libpod/lock/shm/shm_lock.h
+++ b/libpod/lock/shm/shm_lock.h
@@ -40,6 +40,7 @@ shm_struct_t *open_lock_shm(char *path, uint32_t num_locks, int *error_code);
 int32_t close_lock_shm(shm_struct_t *shm);
 int64_t allocate_semaphore(shm_struct_t *shm);
 int32_t deallocate_semaphore(shm_struct_t *shm, uint32_t sem_index);
+int32_t deallocate_all_semaphores(shm_struct_t *shm);
 int32_t lock_semaphore(shm_struct_t *shm, uint32_t sem_index);
 int32_t unlock_semaphore(shm_struct_t *shm, uint32_t sem_index);
 
diff --git a/libpod/lock/shm/shm_lock_test.go b/libpod/lock/shm/shm_lock_test.go
index 594eb5d8e..830035881 100644
--- a/libpod/lock/shm/shm_lock_test.go
+++ b/libpod/lock/shm/shm_lock_test.go
@@ -4,7 +4,6 @@ import (
 	"fmt"
 	"os"
 	"runtime"
-	"syscall"
 	"testing"
 	"time"
 
@@ -53,12 +52,8 @@ func runLockTest(t *testing.T, testFunc func(*testing.T, *SHMLocks)) {
 	}
 	defer func() {
 		// Deallocate all locks
-		// Ignore ENOENT (lock is not allocated)
-		var i uint32
-		for i = 0; i < numLocks; i++ {
-			if err := locks.DeallocateSemaphore(i); err != nil && err != syscall.ENOENT {
-				t.Fatalf("Error deallocating semaphore %d: %v", i, err)
-			}
+		if err := locks.DeallocateAllSemaphores(); err != nil {
+			t.Fatalf("Error deallocating semaphores: %v", err)
 		}
 
 		if err := locks.Close(); err != nil {
@@ -212,6 +207,25 @@ func TestAllocateDeallocateCycle(t *testing.T) {
 	})
 }
 
+// Test that DeallocateAllSemaphores deallocates all semaphores
+func TestDeallocateAllSemaphoresDeallocatesAll(t *testing.T) {
+	runLockTest(t, func(t *testing.T, locks *SHMLocks) {
+		// Allocate a lock
+		locks1, err := locks.AllocateSemaphore()
+		assert.NoError(t, err)
+
+		// Free all locks
+		err = locks.DeallocateAllSemaphores()
+		assert.NoError(t, err)
+
+		// Allocate another lock
+		locks2, err := locks.AllocateSemaphore()
+		assert.NoError(t, err)
+
+		assert.Equal(t, locks1, locks2)
+	})
+}
+
 // Test that locks actually lock
 func TestLockSemaphoreActuallyLocks(t *testing.T) {
 	runLockTest(t, func(t *testing.T, locks *SHMLocks) {
diff --git a/libpod/lock/shm_lock_manager_linux.go b/libpod/lock/shm_lock_manager_linux.go
index 94dfd7dd7..187661c8b 100644
--- a/libpod/lock/shm_lock_manager_linux.go
+++ b/libpod/lock/shm_lock_manager_linux.go
@@ -71,6 +71,11 @@ func (m *SHMLockManager) RetrieveLock(id uint32) (Locker, error) {
 	return lock, nil
 }
 
+// FreeAllLocks frees all locks in the manager
+func (m *SHMLockManager) FreeAllLocks() error {
+	return m.locks.DeallocateAllSemaphores()
+}
+
 // SHMLock is an individual shared memory lock.
 type SHMLock struct {
 	lockID  uint32
diff --git a/libpod/lock/shm_lock_manager_unsupported.go b/libpod/lock/shm_lock_manager_unsupported.go
index cbdb2f7bc..1d6e3fcbd 100644
--- a/libpod/lock/shm_lock_manager_unsupported.go
+++ b/libpod/lock/shm_lock_manager_unsupported.go
@@ -27,3 +27,8 @@ func (m *SHMLockManager) AllocateLock() (Locker, error) {
 func (m *SHMLockManager) RetrieveLock(id string) (Locker, error) {
 	return nil, fmt.Errorf("not supported")
 }
+
+// FreeAllLocks is not supported on this platform
+func (m *SHMLockManager) FreeAllLocks() error {
+	return fmt.Errorf("not supported")
+}
diff --git a/libpod/runtime_renumber.go b/libpod/runtime_renumber.go
new file mode 100644
index 000000000..bc291156e
--- /dev/null
+++ b/libpod/runtime_renumber.go
@@ -0,0 +1,60 @@
+package libpod
+
+import (
+	"path/filepath"
+
+	"github.com/containers/storage"
+	"github.com/pkg/errors"
+)
+
+// RenumberLocks reassigns lock numbers for all containers, pods, and volumes in
+// the state.
+// It renders the runtime it is called on, and all container/pod/volume structs
+// from that runtime, unusable, and requires that a new runtime be initialized
+// after it is called.
+func (r *Runtime) RenumberLocks() error {
+	r.lock.Lock()
+	locked := true
+	defer func() {
+		if locked {
+			r.lock.Unlock()
+		}
+	}()
+
+	runtimeAliveLock := filepath.Join(r.config.TmpDir, "alive.lck")
+	aliveLock, err := storage.GetLockfile(runtimeAliveLock)
+	if err != nil {
+		return errors.Wrapf(err, "error acquiring runtime init lock")
+	}
+	aliveLock.Lock()
+	// It's OK to defer until Shutdown() has run, so no need to check locked
+	defer aliveLock.Unlock()
+
+	// Start off by deallocating all locks
+	if err := r.lockManager.FreeAllLocks(); err != nil {
+		return err
+	}
+
+	allCtrs, err := r.state.AllContainers()
+	if err != nil {
+		return err
+	}
+	for _, ctr := range allCtrs {
+		lock, err := r.lockManager.AllocateLock()
+		if err != nil {
+			return errors.Wrapf(err, "error allocating lock for container %s", ctr.ID())
+		}
+
+		ctr.config.LockID = lock.ID()
+
+		// Write the new lock ID
+		if err := r.state.RewriteContainerConfig(ctr, ctr.config); err != nil {
+			return err
+		}
+	}
+
+	r.lock.Unlock()
+	locked = false
+
+	return r.Shutdown(false)
+}