Fix some cornercase locking issues; make AIO lock re-entrant

fs/aio/aio.h

@@ -122,7 +122,9 @@ void aio_initialize(void);
  * Name: aio_lock/aio_unlock
  *
  * Description:
- *   Take/give the lock on the pending asynchronous I/O list
+ *   Take/give the lock on the pending asynchronous I/O list.  These locks
+ *   are implemented with re-entrant semaphores -- deadlocks will not occur
+ *   if the lock is taken multiple times on the same thread.
  *
  * Input Parameters:
  *   None

fs/aio/aio_cancel.c

@@ -130,6 +130,7 @@ int aio_cancel(int fildes, FAR struct aiocb *aiocbp)
 
   ret = AIO_ALLDONE;
   sched_lock();
+  aio_lock();
 
   if (aiocbp)
     {
@@ -139,11 +140,9 @@ int aio_cancel(int fildes, FAR struct aiocb *aiocbp)
         {
           /* No.. Find the container for this AIO control block */
 
-          aio_lock();
           for (aioc = (FAR struct aio_container_s *)g_aio_pending.head;
                aioc && aioc->aioc_aiocbp != aiocbp;
                aioc = (FAR struct aio_container_s *)aioc->aioc_link.flink);
-          aio_unlock();
 
           /* Did we find a container for this fildes?  We should; the aio_result says
            * that the transfer is pending.  If not we return AIO_ALLDONE.
@@ -185,11 +184,9 @@ int aio_cancel(int fildes, FAR struct aiocb *aiocbp)
         {
           /* Find the next container with this AIO control block */
 
-          aio_lock();
           for (aioc = next;
                aioc && aioc->aioc_aiocbp->aio_fildes != fildes;
                aioc = (FAR struct aio_container_s *)aioc->aioc_link.flink);
-          aio_unlock();
 
           /* Did we find the container?  We should; the aio_result says
            * that the transfer is pending.  If not we return AIO_ALLDONE.
@@ -230,6 +227,7 @@ int aio_cancel(int fildes, FAR struct aiocb *aiocbp)
       while (aioc);
     }
 
+  aio_unlock();
   sched_unlock();
   return ret;
 }

fs/aio/aio_initialize.c

@@ -44,6 +44,8 @@
 #include <errno.h>
 #include <queue.h>
 
+#include <nuttx/sched.h>
+
 #include "aio/aio.h"
 
 #ifdef CONFIG_FS_AIO
@@ -73,10 +75,13 @@ static dq_queue_t g_aioc_free;
 static sem_t g_aioc_freesem;
 
 /* This binary semaphore supports exclusive access to the list of pending
- * asynchronous I/O.
+ * asynchronous I/O.  g_aio_holder and a_aio_count support the reentrant
+ * lock.
  */
 
 static sem_t g_aio_exclsem;
+static pid_t g_aio_holder;
+static uint16_t g_aio_count;
 
 /****************************************************************************
  * Public Data
@@ -114,9 +119,11 @@ void aio_initialize(void)
 
   /* Initialize counting semaphores */
 
-  (void)sem_init(&g_aioc_freesem, 0, 0);
+  (void)sem_init(&g_aioc_freesem, 0, CONFIG_FS_NAIOC);
   (void)sem_init(&g_aio_exclsem, 0, 1);
 
+  g_aio_holder = INVALID_PROCESS_ID;
+
   /* Initialize the container queues */
 
   dq_init(&g_aioc_free);
@@ -126,11 +133,9 @@ void aio_initialize(void)
 
   for (i = 0; i < CONFIG_FS_NAIOC; i++)
     {
-      /* "Free" the entry, adding it to the free list and bumping up the
-       * semaphore count.
-       */
+      /* Add the container to the free list */
 
-      aioc_free(&g_aioc_alloc[i]);
+      dq_addlast(&g_aioc_alloc[i], &g_aioc_free);
     }
 }
 
@@ -150,15 +155,53 @@ void aio_initialize(void)
 
 void aio_lock(void)
 {
-  while (sem_wait(&g_aio_exclsem) < 0)
+  pid_t me = getpid();
+
+  /* Does this thread already hold the semaphore? */
+
+  if (g_aio_holder == me)
     {
-      DEBUGASSERT(get_errno() == EINTR);
+      /* Yes, just increment the counts held */
+
+      DEBUGASSERT(g_aio_count > 0 && g_aio_count < UINT16_MAX);
+      g_aio_count++;
+    }
+  else
+    {
+      /* No.. take the semaphore */
+
+      while (sem_wait(&g_aio_exclsem) < 0)
+        {
+          DEBUGASSERT(get_errno() == EINTR);
+        }
+
+      /* And mark it as ours */
+
+      g_aio_holder = me;
+      g_aio_count  = 1;
     }
 }
 
 void aio_unlock(void)
 {
-  sem_post(&g_aio_exclsem);
+  DEBUGASSERT(g_aio_holder == getpid() && g_aio_count > 0);
+
+  /* Would decrementing the count release the lock? */
+
+  if (g_aio_count <= 1)
+    {
+      /* Yes.. that we will no longer be the holder */
+
+      g_aio_holder = INVALID_PROCESS_ID;
+      g_aio_count  = 0;
+      sem_post(&g_aio_exclsem);
+    }
+  else
+    {
+      /* Otherwise, just decrement the count.  We still hold the lock. */
+
+      g_aio_count--;
+    }
 }
 
 /****************************************************************************
@@ -184,8 +227,8 @@ FAR struct aio_container_s *aioc_alloc(void)
 {
   FAR struct aio_container_s *aioc;
 
-  /* Take a semaphore, thus guaranteeing that we have an AIO container
-   * set aside for us.
+  /* Take a count from semaphore, thus guaranteeing that we have an AIO
+   * container set aside for us.
    */
 
   while (sem_wait(&g_aioc_freesem) < 0)

fs/aio/aioc_contain.c

@@ -153,12 +153,13 @@ FAR struct aiocb *aioc_decant(FAR struct aio_container_s *aioc)
 
   aio_lock();
   dq_rem(&aioc->aioc_link, &g_aio_pending);
-  aio_unlock();
 
   /* De-cant the AIO control block and return the container to the free list */
 
   aiocbp = aioc->aioc_aiocbp;
   aioc_free(aioc);
+
+  aio_unlock();
   return aiocbp;
 }