libretro-common/nbio_linux: fix realloc-to-self leak on OOM in nbio_linux_resize

nbio_linux_resize committed the new length before realloc could be
validated:

    handle->ptr = realloc(handle->ptr, len);
    handle->len = len;

If realloc() returns NULL, handle->ptr is overwritten with NULL
and the original buffer is leaked.  handle->len is then updated to
the new (larger) size, so the handle reports that it owns 'len'
bytes of data that actually live at a NULL pointer.  Every
subsequent nbio_linux_get_ptr / read / write call that iterates
up to handle->len walks off NULL.

The sibling nbio_stdio_resize (libretro-common/file/nbio/nbio_stdio.c
line 261) already uses the correct tmp-pointer-then-commit pattern
and explicitly calls out the same pre-patch bug in a code comment.
The linux variant was the lone holdout.

Fix: realloc into a new_ptr local, only commit on success.  On OOM
return early without updating handle->ptr or handle->len, leaving
the handle in its pre-resize state.  Callers that currently crash
on NULL-deref now see the old len and the old pointer - which is
the same 'no-op on OOM' behaviour that nbio_stdio_resize already
promises.

Thread-safety: unchanged.  nbio handles are owned by whichever
thread is servicing them (typically a task queue thread); no
shared state touched.

Author: LibretroAdmin

libretro-common/file/nbio/nbio_linux.c

@@ -153,6 +153,7 @@ static bool nbio_linux_iterate(void *data)
 static void nbio_linux_resize(void *data, size_t len)
 {
    struct nbio_linux_t* handle = (struct nbio_linux_t*)data;
+   void *new_ptr;
    if (!handle)
       return;
 
@@ -166,7 +167,21 @@ static void nbio_linux_resize(void *data, size_t len)
       abort(); /* this one returns void and I can't find any other way
                   for it to report failure */
 
-   handle->ptr = realloc(handle->ptr, len);
+   /* Attempt the realloc BEFORE committing the new length.  If it
+    * fails, the old pointer and its old size are still valid; the
+    * caller can retry or abandon.  Matches the sibling pattern in
+    * nbio_stdio_resize - the pre-patch form
+    *
+    *    handle->ptr = realloc(handle->ptr, len);
+    *    handle->len = len;
+    *
+    * set handle->ptr to NULL on OOM (leaking the old buffer) and
+    * then claimed the new length, so subsequent get_ptr()/read/write
+    * walked a NULL pointer assuming len bytes. */
+   if (!(new_ptr = realloc(handle->ptr, len)))
+      return;
+
+   handle->ptr = new_ptr;
    handle->len = len;
 }