> > When polling for entropy, the PRNG will start with the first entropy > source added and then go through the list. The last entropy source, > Unix_EntropySource, will include the list of commands as mentioned earlier > in this email thread. The “ls -alni /tmp” has priority 4 out of 5. This > makes it one of the last commands in the last entropy source. > > Each byte gathered by the Unix_EntropySource is counted as 0.005 bits toward the 128 bit polling goal. We can count the number of bytes returned from each command in priority 1 to priority 3. My system did not have all commands, but this was the number of bytes I polled using the commands.
netstat -in | wc -c 269 vmstat -s | wc -c 764 vmstat | wc -c 234 arp -a -n | wc -c 112 ifconfig -a | wc -c 922 ipcs -a | wc -c 1591 netstat -an | wc -c 39167 netstat -s | wc -c 2267 uname -a | wc -c 104 uptime | wc -c 71 ps -A | wc -c 6246 Total number of bytes: 51747. Multiply this with 0.005 and we get 258.735 bits of entropy according to Botan. This is more than the goal of 128 bits. So even if we ignore all of the other entropy sources than Unix_EntropySource, we would still most likely return before even calling the command "ls -alni /tmp". I believe we have the following discussion points: 1. (Original topic) vmstat causes kernel messages. What modifications do we want to Botan? 2. "ls -alni /tmp" is used as a last resort entropy source. It is not good to poll entropy from a directory that users can control. Botan will most likely not use this source. Should Botan remove this command? Even if it is not directly used in the cryptographic operations (the seed is passed through two RNGs)? 3. Should SoftHSM add checks verifying that the RNG is seeded with good sources (the build flags from botan/build.h)? // Rickard
_______________________________________________ Opendnssec-user mailing list [email protected] https://lists.opendnssec.org/mailman/listinfo/opendnssec-user
