Fixed allocation of multiple shared memory chunks.

Previously, one shared memory chunk was allocated under mutex and other
chunks (if required) were allocated using atomic operations.  So such
allocation is not guaranteed and the result buffer can be less than
requested.

This commit moves multiple chunks allocation under mutex and guarantees
the result buffer is large enough.

src/nxt_port_memory_int.h

@@ -129,13 +129,20 @@ nxt_port_mmap_chunk_start(nxt_port_mmap_header_t *hdr, nxt_chunk_id_t c)
 static nxt_bool_t
 nxt_port_mmap_get_free_chunk(nxt_free_map_t *m, nxt_chunk_id_t *c)
 {
-    int             ffs;
-    size_t          i;
-    nxt_chunk_id_t  chunk;
-    nxt_free_map_t  bits;
+    const nxt_free_map_t  default_mask = (nxt_free_map_t) -1;
+
+    int             ffs;
+    size_t          i, start;
+    nxt_chunk_id_t  chunk;
+    nxt_free_map_t  bits, mask;
+
+    start = FREE_IDX(*c);
+    mask = default_mask << ((*c) % FREE_BITS);
+
+    for (i = start; i < MAX_FREE_IDX; i++) {
+        bits = m[i] & mask;
+        mask = default_mask;
 
-    for (i = 0; i < MAX_FREE_IDX; i++) {
-        bits = m[i];
         if (bits == 0) {
             continue;
         }