Hello,
I observed that on few platforms, the random device is missing. Due to this,
the SSL_connect fails. I would have to use -rand option of s_client for
successful connection.
On such platforms, even my client application fails due to the missing random
number generator device.
For eg. On my solaris box, my client fails and truss gives the following output:
=============
0.0999 open("/dev/urandom", O_RDONLY|O_NONBLOCK|O_NOCTTY) Err#2 ENOENT
0.1003 open("/dev/random", O_RDONLY|O_NONBLOCK|O_NOCTTY) Err#2 ENOENT
0.1007 open("/dev/srandom", O_RDONLY|O_NONBLOCK|O_NOCTTY) Err#2 ENOENT
0.1013 so_socket(1, 2, 0, "", 1) = 5
0.1018 connect(5, 0xFFBEE3F8, 19, 1) Err#2 ENOENT
0.1021 close(5) = 0
0.1025 so_socket(1, 2, 0, "", 1) = 5
0.1029 connect(5, 0xFFBEE3F8, 15, 1) Err#2 ENOENT
0.1032 close(5) = 0
0.1035 so_socket(1, 2, 0, "", 1) = 5
0.1039 connect(5, 0xFFBEE3F8, 15, 1) Err#2 ENOENT
0.1041 close(5) = 0
0.1046 so_socket(1, 2, 0, "", 1) = 5
0.1050 connect(5, 0xFFBEE3F8, 14, 1) Err#2 ENOENT
=============
And then the SSL_connect fails with SSL_ERROR_SYSCALL and errno set to 2.
In such cases, I had thought of using RAND_write_file() followed by
RAND_load_file() just after loading required libraries (SSL_load_error_strings,
SSL_library_init).
Do you think this is sensible approach? If not, could you suggest ways to deal
with situations when the random device is missing?
If yes, I have another question. How to detect, programatically if the random
device is missing? The RAND_write_file() and RAND_load_file() should be used
only if the random device is missing, right?
Any help in understanding this is highly appriciated.
Thank you,
~ Urjit
