Issue #29157: Prefer getrandom() over getentropy()

vstinner · vstinner · commit ff558f5aba40 · 2017-01-07T00:07:45.000+01:00
* dev_urandom() now calls py_getentropy(). Prepare the fallback to support
  getentropy() failure and falls back on reading from /dev/urandom.
* Simplify dev_urandom(). pyurandom() is now responsible to call getentropy()
  or getrandom(). Enhance also dev_urandom() and pyurandom() documentation.
* getrandom() is now preferred over getentropy(). The glibc 2.24 now implements
  getentropy() on Linux using the getrandom() syscall.  But getentropy()
  doesn't support non-blocking mode. Since getrandom() is tried first, it's not
  more needed to explicitly exclude getentropy() on Solaris. Replace:
  "if defined(HAVE_GETENTROPY) &amp;&amp; !defined(sun)"
  with "if defined(HAVE_GETENTROPY)"
* Enhance py_getrandom() documentation. py_getentropy() now supports ENOSYS,
  EPERM &amp; EINTR
diff --git a/Python/random.c b/Python/random.c
@@ -77,57 +77,23 @@ win32_urandom(unsigned char *buffer, Py_ssize_t size, int raise)
     return 0;
 }
 
-/* Issue #25003: Don't use getentropy() on Solaris (available since
- * Solaris 11.3), it is blocking whereas os.urandom() should not block. */
-#elif defined(HAVE_GETENTROPY) && !defined(sun)
-#define PY_GETENTROPY 1
-
-/* Fill buffer with size pseudo-random bytes generated by getentropy().
-   Return 0 on success, or raise an exception and return -1 on error.
-
-   If raise is zero, don't raise an exception on error. */
-static int
-py_getentropy(char *buffer, Py_ssize_t size, int raise)
-{
-    while (size > 0) {
-        Py_ssize_t len = Py_MIN(size, 256);
-        int res;
-
-        if (raise) {
-            Py_BEGIN_ALLOW_THREADS
-            res = getentropy(buffer, len);
-            Py_END_ALLOW_THREADS
-        }
-        else {
-            res = getentropy(buffer, len);
-        }
-
-        if (res < 0) {
-            if (raise) {
-                PyErr_SetFromErrno(PyExc_OSError);
-            }
-            return -1;
-        }
-
-        buffer += len;
-        size -= len;
-    }
-    return 0;
-}
-
-#else
+#else /* !MS_WINDOWS */
 
 #if defined(HAVE_GETRANDOM) || defined(HAVE_GETRANDOM_SYSCALL)
 #define PY_GETRANDOM 1
 
-/* Call getrandom()
+/* Call getrandom() to get random bytes:
+
    - Return 1 on success
-   - Return 0 if getrandom() syscall is not available (failed with ENOSYS or
-     EPERM) or if getrandom(GRND_NONBLOCK) failed with EAGAIN (system urandom
-     not initialized yet) and raise=0.
+   - Return 0 if getrandom() is not available (failed with ENOSYS or EPERM),
+     or if getrandom(GRND_NONBLOCK) failed with EAGAIN (system urandom not
+     initialized yet) and raise=0.
    - Raise an exception (if raise is non-zero) and return -1 on error:
-     getrandom() failed with EINTR and the Python signal handler raised an
-     exception, or getrandom() failed with a different error. */
+     if getrandom() failed with EINTR, raise is non-zero and the Python signal
+     handler raised an exception, or if getrandom() failed with a different
+     error.
+
+   getrandom() is retried if it failed with EINTR: interrupted by a signal. */
 static int
 py_getrandom(void *buffer, Py_ssize_t size, int blocking, int raise)
 {
@@ -148,7 +114,8 @@ py_getrandom(void *buffer, Py_ssize_t size, int blocking, int raise)
     while (0 < size) {
 #ifdef sun
         /* Issue #26735: On Solaris, getrandom() is limited to returning up
-           to 1024 bytes */
+           to 1024 bytes. Call it multiple times if more bytes are
+           requested. */
         n = Py_MIN(size, 1024);
 #else
         n = Py_MIN(size, LONG_MAX);
@@ -179,18 +146,19 @@ py_getrandom(void *buffer, Py_ssize_t size, int blocking, int raise)
 #endif
 
         if (n < 0) {
-            /* ENOSYS: getrandom() syscall not supported by the kernel (but
-             * maybe supported by the host which built Python). EPERM:
-             * getrandom() syscall blocked by SECCOMP or something else. */
+            /* ENOSYS: the syscall is not supported by the kernel.
+               EPERM: the syscall is blocked by a security policy (ex: SECCOMP)
+               or something else. */
             if (errno == ENOSYS || errno == EPERM) {
                 getrandom_works = 0;
                 return 0;
             }
 
             /* getrandom(GRND_NONBLOCK) fails with EAGAIN if the system urandom
-               is not initialiazed yet. For _PyRandom_Init(), we ignore their
+               is not initialiazed yet. For _PyRandom_Init(), we ignore the
                error and fall back on reading /dev/urandom which never blocks,
-               even if the system urandom is not initialized yet. */
+               even if the system urandom is not initialized yet:
+               see the PEP 524. */
             if (errno == EAGAIN && !raise && !blocking) {
                 return 0;
             }
@@ -217,43 +185,119 @@ py_getrandom(void *buffer, Py_ssize_t size, int blocking, int raise)
     }
     return 1;
 }
-#endif
+
+#elif defined(HAVE_GETENTROPY)
+#define PY_GETENTROPY 1
+
+/* Fill buffer with size pseudo-random bytes generated by getentropy():
+
+   - Return 1 on success
+   - Return 0 if getentropy() syscall is not available (failed with ENOSYS or
+     EPERM).
+   - Raise an exception (if raise is non-zero) and return -1 on error:
+     if getentropy() failed with EINTR, raise is non-zero and the Python signal
+     handler raised an exception, or if getentropy() failed with a different
+     error.
+
+   getentropy() is retried if it failed with EINTR: interrupted by a signal. */
+static int
+py_getentropy(char *buffer, Py_ssize_t size, int raise)
+{
+    /* Is getentropy() supported by the running kernel? Set to 0 if
+       getentropy() failed with ENOSYS or EPERM. */
+    static int getentropy_works = 1;
+
+    if (!getentropy_works) {
+        return 0;
+    }
+
+    while (size > 0) {
+        /* getentropy() is limited to returning up to 256 bytes. Call it
+           multiple times if more bytes are requested. */
+        Py_ssize_t len = Py_MIN(size, 256);
+        int res;
+
+        if (raise) {
+            Py_BEGIN_ALLOW_THREADS
+            res = getentropy(buffer, len);
+            Py_END_ALLOW_THREADS
+        }
+        else {
+            res = getentropy(buffer, len);
+        }
+
+        if (res < 0) {
+            /* ENOSYS: the syscall is not supported by the running kernel.
+               EPERM: the syscall is blocked by a security policy (ex: SECCOMP)
+               or something else. */
+            if (errno == ENOSYS || errno == EPERM) {
+                getentropy_works = 0;
+                return 0;
+            }
+
+            if (errno == EINTR) {
+                if (raise) {
+                    if (PyErr_CheckSignals()) {
+                        return -1;
+                    }
+                }
+
+                /* retry getentropy() if it was interrupted by a signal */
+                continue;
+            }
+
+            if (raise) {
+                PyErr_SetFromErrno(PyExc_OSError);
+            }
+            return -1;
+        }
+
+        buffer += len;
+        size -= len;
+    }
+    return 1;
+}
+#endif /* defined(HAVE_GETENTROPY) && !defined(sun) */
+
 
 static struct {
     int fd;
     dev_t st_dev;
     ino_t st_ino;
 } urandom_cache = { -1 };
 
+/* Read random bytes from the /dev/urandom device:
+
+   - Return 0 on success
+   - Raise an exception (if raise is non-zero) and return -1 on error
+
+   Possible causes of errors:
+
+   - open() failed with ENOENT, ENXIO, ENODEV, EACCES: the /dev/urandom device
+     was not found. For example, it was removed manually or not exposed in a
+     chroot or container.
+   - open() failed with a different error
+   - fstat() failed
+   - read() failed or returned 0
 
-/* Read 'size' random bytes from py_getrandom(). Fall back on reading from
-   /dev/urandom if getrandom() is not available.
+   read() is retried if it failed with EINTR: interrupted by a signal.
 
-   Return 0 on success. Raise an exception (if raise is non-zero) and return -1
-   on error. */
+   The file descriptor of the device is kept open between calls to avoid using
+   many file descriptors when run in parallel from multiple threads:
+   see the issue #18756.
+
+   st_dev and st_ino fields of the file descriptor (from fstat()) are cached to
+   check if the file descriptor was replaced by a different file (which is
+   likely a bug in the application): see the issue #21207.
+
+   If the file descriptor was closed or replaced, open a new file descriptor
+   but don't close the old file descriptor: it probably points to something
+   important for some third-party code. */
 static int
-dev_urandom(char *buffer, Py_ssize_t size, int blocking, int raise)
+dev_urandom(char *buffer, Py_ssize_t size, int raise)
 {
     int fd;
     Py_ssize_t n;
-#ifdef PY_GETRANDOM
-    int res;
-#endif
-
-    assert(size > 0);
-
-#ifdef PY_GETRANDOM
-    res = py_getrandom(buffer, size, blocking, raise);
-    if (res < 0) {
-        return -1;
-    }
-    if (res == 1) {
-        return 0;
-    }
-    /* getrandom() failed with ENOSYS or EPERM,
-       fall back on reading /dev/urandom */
-#endif
-
 
     if (raise) {
         struct _Py_stat_struct st;
@@ -275,9 +319,10 @@ dev_urandom(char *buffer, Py_ssize_t size, int blocking, int raise)
             fd = _Py_open("/dev/urandom", O_RDONLY);
             if (fd < 0) {
                 if (errno == ENOENT || errno == ENXIO ||
-                    errno == ENODEV || errno == EACCES)
+                    errno == ENODEV || errno == EACCES) {
                     PyErr_SetString(PyExc_NotImplementedError,
                                     "/dev/urandom (or equivalent) not found");
+                }
                 /* otherwise, keep the OSError exception raised by _Py_open() */
                 return -1;
             }
@@ -349,8 +394,8 @@ dev_urandom_close(void)
         urandom_cache.fd = -1;
     }
 }
+#endif /* !MS_WINDOWS */
 
-#endif
 
 /* Fill buffer with pseudo-random bytes generated by a linear congruent
    generator (LCG):
@@ -373,14 +418,56 @@ lcg_urandom(unsigned int x0, unsigned char *buffer, size_t size)
     }
 }
 
-/* If raise is zero:
-   - Don't raise exceptions on error
-   - Don't call PyErr_CheckSignals() on EINTR (retry directly the interrupted
-     syscall)
-   - Don't release the GIL to call syscalls. */
+/* Read random bytes:
+
+   - Return 0 on success
+   - Raise an exception (if raise is non-zero) and return -1 on error
+
+   Used sources of entropy ordered by preference, preferred source first:
+
+   - CryptGenRandom() on Windows
+   - getrandom() function (ex: Linux and Solaris): call py_getrandom()
+   - getentropy() function (ex: OpenBSD): call py_getentropy()
+   - /dev/urandom device
+
+   Read from the /dev/urandom device if getrandom() or getentropy() function
+   is not available or does not work.
+
+   Prefer getrandom() over getentropy() because getrandom() supports blocking
+   and non-blocking mode: see the PEP 524. Python requires non-blocking RNG at
+   startup to initialize its hash secret, but os.urandom() must block until the
+   system urandom is initialized (at least on Linux 3.17 and newer).
+
+   Prefer getrandom() and getentropy() over reading directly /dev/urandom
+   because these functions don't need file descriptors and so avoid ENFILE or
+   EMFILE errors (too many open files): see the issue #18756.
+
+   Only the getrandom() function supports non-blocking mode.
+
+   Only use RNG running in the kernel. They are more secure because it is
+   harder to get the internal state of a RNG running in the kernel land than a
+   RNG running in the user land. The kernel has a direct access to the hardware
+   and has access to hardware RNG, they are used as entropy sources.
+
+   Note: the OpenSSL RAND_pseudo_bytes() function does not automatically reseed
+   its RNG on fork(), two child processes (with the same pid) generate the same
+   random numbers: see issue #18747. Kernel RNGs don't have this issue,
+   they have access to good quality entropy sources.
+
+   If raise is zero:
+
+   - Don't raise an exception on error
+   - Don't call the Python signal handler (don't call PyErr_CheckSignals()) if
+     a function fails with EINTR: retry directly the interrupted function
+   - Don't release the GIL to call functions.
+*/
 static int
 pyurandom(void *buffer, Py_ssize_t size, int blocking, int raise)
 {
+#if defined(PY_GETRANDOM) || defined(PY_GETENTROPY)
+    int res;
+#endif
+
     if (size < 0) {
         if (raise) {
             PyErr_Format(PyExc_ValueError,
@@ -395,10 +482,25 @@ pyurandom(void *buffer, Py_ssize_t size, int blocking, int raise)
 
 #ifdef MS_WINDOWS
     return win32_urandom((unsigned char *)buffer, size, raise);
-#elif defined(PY_GETENTROPY)
-    return py_getentropy(buffer, size, raise);
 #else
-    return dev_urandom(buffer, size, blocking, raise);
+
+#if defined(PY_GETRANDOM) || defined(PY_GETENTROPY)
+#ifdef PY_GETRANDOM
+    res = py_getrandom(buffer, size, blocking, raise);
+#else
+    res = py_getentropy(buffer, size, raise);
+#endif
+    if (res < 0) {
+        return -1;
+    }
+    if (res == 1) {
+        return 0;
+    }
+    /* getrandom() or getentropy() function is not available: failed with
+       ENOSYS or EPERM. Fall back on reading from /dev/urandom. */
+#endif
+
+    return dev_urandom(buffer, size, raise);
 #endif
 }
 
@@ -491,8 +593,6 @@ _PyRandom_Fini(void)
         CryptReleaseContext(hCryptProv, 0);
         hCryptProv = 0;
     }
-#elif defined(PY_GETENTROPY)
-    /* nothing to clean */
 #else
     dev_urandom_close();
 #endif
