/* * Individual synchronization primitives are represented by * struct ntsync_obj, and each primitive is backed by a file. * * The whole namespace is represented by a struct ntsync_device also * backed by a file. * * Both rely on struct file for reference counting. Individual * ntsync_obj objects take a reference to the device when created. * Wait operations take a reference to each object being waited on for * the duration of the wait.
*/
struct ntsync_obj {
spinlock_t lock; int dev_locked;
enum ntsync_type type;
struct file *file; struct ntsync_device *dev;
/* The following fields are protected by the object lock. */ union { struct {
__u32 count;
__u32 max;
} sem; struct {
__u32 count;
pid_t owner; bool ownerdead;
} mutex; struct { bool manual; bool signaled;
} event;
} u;
/* * any_waiters is protected by the object lock, but all_waiters is * protected by the device wait_all_lock.
*/ struct list_head any_waiters; struct list_head all_waiters;
/* * Hint describing how many tasks are queued on this object in a * wait-all operation. * * Any time we do a wake, we may need to wake "all" waiters as well as * "any" waiters. In order to atomically wake "all" waiters, we must * lock all of the objects, and that means grabbing the wait_all_lock * below (and, due to lock ordering rules, before locking this object). * However, wait-all is a rare operation, and grabbing the wait-all * lock for every wake would create unnecessary contention. * Therefore we first check whether all_hint is zero, and, if it is, * we skip trying to wake "all" waiters. * * Since wait requests must originate from user-space threads, we're * limited here by PID_MAX_LIMIT, so there's no risk of overflow.
*/
atomic_t all_hint;
};
/* * Protected via atomic_try_cmpxchg(). Only the thread that wins the * compare-and-swap may actually change object states and wake this * task.
*/
atomic_t signaled;
struct ntsync_device { /* * Wait-all operations must atomically grab all objects, and be totally * ordered with respect to each other and wait-any operations. * If one thread is trying to acquire several objects, another thread * cannot touch the object at the same time. * * This device-wide lock is used to serialize wait-for-all * operations, and operations on an object that is involved in a * wait-for-all.
*/ struct mutex wait_all_lock;
struct file *file;
};
/* * Single objects are locked using obj->lock. * * Multiple objects are 'locked' while holding dev->wait_all_lock. * In this case however, individual objects are not locked by holding * obj->lock, but by setting obj->dev_locked. * * This means that in order to lock a single object, the sequence is slightly * more complicated than usual. Specifically it needs to check obj->dev_locked * after acquiring obj->lock, if set, it needs to drop the lock and acquire * dev->wait_all_lock in order to serialize against the multi-object operation.
*/
staticvoid dev_lock_obj(struct ntsync_device *dev, struct ntsync_obj *obj)
{
lockdep_assert_held(&dev->wait_all_lock);
lockdep_assert(obj->dev == dev);
spin_lock(&obj->lock); /* * By setting obj->dev_locked inside obj->lock, it is ensured that * anyone holding obj->lock must see the value.
*/
obj->dev_locked = 1;
spin_unlock(&obj->lock);
}
for (;;) {
spin_lock(&obj->lock); if (likely(!obj->dev_locked)) break;
spin_unlock(&obj->lock);
mutex_lock(&dev->wait_all_lock);
spin_lock(&obj->lock); /* * obj->dev_locked should be set and released under the same * wait_all_lock section, since we now own this lock, it should * be clear.
*/
lockdep_assert(!obj->dev_locked);
spin_unlock(&obj->lock);
mutex_unlock(&dev->wait_all_lock);
}
}
switch (obj->type) { case NTSYNC_TYPE_SEM: return !!obj->u.sem.count; case NTSYNC_TYPE_MUTEX: if (obj->u.mutex.owner && obj->u.mutex.owner != owner) returnfalse; return obj->u.mutex.count < UINT_MAX; case NTSYNC_TYPE_EVENT: return obj->u.event.signaled;
}
WARN(1, "bad object type %#x\n", obj->type); returnfalse;
}
/* * "locked_obj" is an optional pointer to an object which is already locked and * should not be locked again. This is necessary so that changing an object's * state and waking it can be a single atomic operation.
*/ staticvoid try_wake_all(struct ntsync_device *dev, struct ntsync_q *q, struct ntsync_obj *locked_obj)
{
__u32 count = q->count; bool can_wake = true; int signaled = -1;
__u32 i;
lockdep_assert_held(&dev->wait_all_lock); if (locked_obj)
lockdep_assert(locked_obj->dev_locked);
for (i = 0; i < count; i++) { if (q->entries[i].obj != locked_obj)
dev_lock_obj(dev, q->entries[i].obj);
}
for (i = 0; i < count; i++) { if (!is_signaled(q->entries[i].obj, q->owner)) {
can_wake = false; break;
}
}
if (can_wake && atomic_try_cmpxchg(&q->signaled, &signaled, 0)) { for (i = 0; i < count; i++) { struct ntsync_obj *obj = q->entries[i].obj;
switch (obj->type) { case NTSYNC_TYPE_SEM:
obj->u.sem.count--; break; case NTSYNC_TYPE_MUTEX: if (obj->u.mutex.ownerdead)
q->ownerdead = true;
obj->u.mutex.ownerdead = false;
obj->u.mutex.count++;
obj->u.mutex.owner = q->owner; break; case NTSYNC_TYPE_EVENT: if (!obj->u.event.manual)
obj->u.event.signaled = false; break;
}
}
wake_up_process(q->task);
}
for (i = 0; i < count; i++) { if (q->entries[i].obj != locked_obj)
dev_unlock_obj(dev, q->entries[i].obj);
}
}
if (atomic_try_cmpxchg(&q->signaled, &signaled, entry->index)) { if (!event->u.event.manual)
event->u.event.signaled = false;
wake_up_process(q->task);
}
}
}
/* * Actually change the semaphore state, returning -EOVERFLOW if it is made * invalid.
*/ staticint release_sem_state(struct ntsync_obj *sem, __u32 count)
{
__u32 sum;
ntsync_assert_held(sem);
if (check_add_overflow(sem->u.sem.count, count, &sum) ||
sum > sem->u.sem.max) return -EOVERFLOW;
if (copy_from_user(&args, argp, sizeof(args))) return -EFAULT; if (!args.owner) return -EINVAL;
if (mutex->type != NTSYNC_TYPE_MUTEX) return -EINVAL;
all = ntsync_lock_obj(dev, mutex);
prev_count = mutex->u.mutex.count;
ret = unlock_mutex_state(mutex, &args); if (!ret) { if (all)
try_wake_all_obj(dev, mutex);
try_wake_any_mutex(mutex);
}
ntsync_unlock_obj(dev, mutex, all);
if (!ret && put_user(prev_count, &user_args->count))
ret = -EFAULT;
return ret;
}
/* * Actually change the mutex state to mark its owner as dead, * returning -EPERM if not the owner.
*/ staticint kill_mutex_state(struct ntsync_obj *mutex, __u32 owner)
{
ntsync_assert_held(mutex);
staticvoid try_wake_any_obj(struct ntsync_obj *obj)
{ switch (obj->type) { case NTSYNC_TYPE_SEM:
try_wake_any_sem(obj); break; case NTSYNC_TYPE_MUTEX:
try_wake_any_mutex(obj); break; case NTSYNC_TYPE_EVENT:
try_wake_any_event(obj); break;
}
}
staticint ntsync_wait_any(struct ntsync_device *dev, void __user *argp)
{ struct ntsync_wait_args args;
__u32 i, total_count; struct ntsync_q *q; int signaled; bool all; int ret;
if (copy_from_user(&args, argp, sizeof(args))) return -EFAULT;
ret = setup_wait(dev, &args, false, &q); if (ret < 0) return ret;
total_count = args.count; if (args.alert)
total_count++;
/* queue ourselves */
for (i = 0; i < total_count; i++) { struct ntsync_q_entry *entry = &q->entries[i]; struct ntsync_obj *obj = entry->obj;
all = ntsync_lock_obj(dev, obj);
list_add_tail(&entry->node, &obj->any_waiters);
ntsync_unlock_obj(dev, obj, all);
}
/* * Check if we are already signaled. * * Note that the API requires that normal objects are checked before * the alert event. Hence we queue the alert event last, and check * objects in order.
*/
for (i = 0; i < total_count; i++) { struct ntsync_obj *obj = q->entries[i].obj;
if (atomic_read(&q->signaled) != -1) break;
all = ntsync_lock_obj(dev, obj);
try_wake_any_obj(obj);
ntsync_unlock_obj(dev, obj, all);
}
/* sleep */
ret = ntsync_schedule(q, &args);
/* and finally, unqueue */
for (i = 0; i < total_count; i++) { struct ntsync_q_entry *entry = &q->entries[i]; struct ntsync_obj *obj = entry->obj;
all = ntsync_lock_obj(dev, obj);
list_del(&entry->node);
ntsync_unlock_obj(dev, obj, all);
/* even if we caught a signal, we need to communicate success */
ret = q->ownerdead ? -EOWNERDEAD : 0;
if (put_user(signaled, &user_args->index))
ret = -EFAULT;
} elseif (!ret) {
ret = -ETIMEDOUT;
}
kfree(q); return ret;
}
staticint ntsync_wait_all(struct ntsync_device *dev, void __user *argp)
{ struct ntsync_wait_args args; struct ntsync_q *q; int signaled;
__u32 i; int ret;
if (copy_from_user(&args, argp, sizeof(args))) return -EFAULT;
ret = setup_wait(dev, &args, true, &q); if (ret < 0) return ret;
/* queue ourselves */
mutex_lock(&dev->wait_all_lock);
for (i = 0; i < args.count; i++) { struct ntsync_q_entry *entry = &q->entries[i]; struct ntsync_obj *obj = entry->obj;
atomic_inc(&obj->all_hint);
/* * obj->all_waiters is protected by dev->wait_all_lock rather * than obj->lock, so there is no need to acquire obj->lock * here.
*/
list_add_tail(&entry->node, &obj->all_waiters);
} if (args.alert) { struct ntsync_q_entry *entry = &q->entries[args.count]; struct ntsync_obj *obj = entry->obj;
Die Informationen auf dieser Webseite wurden
nach bestem Wissen sorgfältig zusammengestellt. Es wird jedoch weder Vollständigkeit, noch Richtigkeit,
noch Qualität der bereit gestellten Informationen zugesichert.
Bemerkung:
Die farbliche Syntaxdarstellung und die Messung sind noch experimentell.