diff --git a/mcpp.toml b/mcpp.toml
index 927690d..f4583c7 100644
--- a/mcpp.toml
+++ b/mcpp.toml
@@ -1,7 +1,7 @@
 [package]
 namespace   = "mcpplibs"
 name        = "openkal-linux"
-version     = "0.5.3"
+version     = "0.5.4"
 description = "The reference implementation of openkal for Linux, written on the kernel's own system-call interface so that it can be placed beneath a C library as well as above one."
 license     = "Apache-2.0"
 
@@ -18,7 +18,7 @@ authors     = ["mcpplibs"]
 repo        = "https://github.com/mcpplibs/openkal-linux"
 
 [dependencies]
-openkal = "0.6.0"
+openkal = "0.7.0"
 
 # The package contributes definitions and no modules. The interface it
 # implements is declared by the specification package, which this package
diff --git a/src/random.cpp b/src/random.cpp
new file mode 100644
index 0000000..70fc828
--- /dev/null
+++ b/src/random.cpp
@@ -0,0 +1,55 @@
+// openkal.random on Linux --- getrandom(2).
+//
+// ⭐ THE KERNEL'S OWN CALL AND NOT `/dev/urandom`. The device would need a
+// descriptor, which needs a path, which a capability-oriented filesystem
+// deliberately does not hand out; and a program early enough in its life not to
+// have a filesystem yet still has this call. `getrandom` is the interface the
+// kernel offers for exactly this question.
+#include "sys.h"
+#include <openkal/random.h>
+
+namespace {
+
+// ⚠️ `GRND_NONBLOCK` IS NOT SET, AND THAT IS WHAT `BLOCKING` REPORTS.
+//
+// Without it the call waits until the pool has been initialised, which on a
+// machine seconds into its first boot can be a real wait. Setting it instead
+// would turn that wait into a short read — a partial success this interface
+// does not have — so the wait is kept and named in the capability word.
+constexpr okl_long flags_blocking = 0;
+
+}  // namespace
+
+extern "C" int kal_random_fill(void* out, kal_uintptr len) {
+    if (len == 0) return kal_ok;
+    if (out == nullptr) return kal_err_invalid;
+
+    auto* p = static_cast<unsigned char*>(out);
+    kal_uintptr filled = 0;
+    while (filled < len) {
+        const okl_long r = okl::sys(okl::nr_getrandom,
+                                    reinterpret_cast<okl_long>(p + filled),
+                                    static_cast<okl_long>(len - filled),
+                                    flags_blocking);
+        if (r < 0) {
+            // ⚠️ THE BUFFER IS NOT RESTORED, AND THE CONTRACT SAYS IT NEED NOT
+            // BE: a failed fill leaves the buffer unspecified rather than
+            // unchanged. Restoring it would oblige this function to keep a copy
+            // of what it was handed, which is a cost every successful call
+            // would pay for the benefit of the failing one.
+            if (r == -4 /* EINTR */) continue;
+            if (r == -11 /* EAGAIN */) return kal_err_again;
+            return kal_err_io;
+        }
+        // A short return is the kernel's, not this interface's: `getrandom`
+        // caps a single call at 32 MiB. Looping is what turns it into the
+        // all-or-nothing this interface promises.
+        filled += static_cast<kal_uintptr>(r);
+    }
+    return kal_ok;
+}
+
+// Blocking, because GRND_NONBLOCK is not set above. Not hardware: the kernel's
+// pool is what this reads, and whether the pool was seeded from a hardware
+// source is not something this backend can observe.
+extern "C" const kal_uintptr kal_random_props = KAL_RANDOM_PROP_BLOCKING;
diff --git a/src/sys.h b/src/sys.h
index 6352116..82975e3 100644
--- a/src/sys.h
+++ b/src/sys.h
@@ -97,6 +97,7 @@ enum : okl_long {
     nr_openat = 257, nr_mkdirat = 258, nr_newfstatat = 262, nr_unlinkat = 263,
     nr_renameat = 264, nr_readlinkat = 267, nr_dup3 = 292, nr_execveat = 322,
     nr_dup2 = 33, nr_utimensat = 280,
+    nr_getrandom = 318,
 };
 
 #elif defined(__aarch64__)
@@ -170,6 +171,7 @@ enum : okl_long {
     nr_clone = 220, nr_execve = 221, nr_wait4 = 260, nr_renameat = 38,
     nr_dup3 = 24, nr_execveat = 281, nr_dup2 = -1,
     nr_arch_prctl = -1, nr_utimensat = 88,
+    nr_getrandom = 278,
 };
 
 #else