Fix niova_io_iovs_advance() bug when iovs are completely used

Calling niova_io_iovs_advance() with parameters that completely 'use'
the entire iov set caused the function to assert on an internal check
that verified the ensuing index was < the number of iovs.

The approach taken here is to detect the condition and return the error,
-EXFULL, rather than modify the last iov such that its base pointer
would be out of bounds and its length would be 0.

src/io.c

@@ -361,11 +361,23 @@ niova_io_iovs_advance(struct iovec *iovs, size_t niovs,
             break;
     }
 
-    if (bytes_already_consumed > 0)
+    if (idx == niovs)
     {
-        NIOVA_ASSERT(idx == niovs);
-        return -ERANGE;
+        if (bytes_already_consumed > 0)
+        {
+            NIOVA_ASSERT(idx == niovs);
+            return -ERANGE;
+        }
+        else
+        {
+            /* If this advance consumes all iov space, simply return w/out
+             * modifying any iov.
+             */
+            NIOVA_ASSERT(bytes_already_consumed == 0);
+            return -EXFULL;
+        }
     }
+
     NIOVA_ASSERT(idx < niovs);
 
     if (save_iov)

test/io-iov-test.c

@@ -142,6 +142,10 @@ iov_test(void)
     niova_io_iov_restore(iovs, n, idx, &save);
     NIOVA_ASSERT(niova_io_iovs_total_size_get(iovs, n) == (n * n));
 
+    // Consume everything in one shot:  niova-block issue #141
+    idx = niova_io_iovs_advance(iovs, n, n * n, NULL);
+    NIOVA_ASSERT(idx == -EXFULL);
+
     off_t off = 0;
     size_t niovs = n;