jitter option still has problems with systemd reboot about rng-tools HOT 14 CLOSED

run rngd -f -d instead of running it via systemd while the system is up, then kill it with a SIGTERM from another terminal and send me the output of that

from rng-tools.

screenlog.txt

from rng-tools.

Ok description:

if I kill rngd before

jent_read_entropy time on cpu 0 is 1.265280712500e+01 sec

-> you have the first case

Initializing available sources
Initializing entropy source hwrng
Limiting thread count to 2 active cpus
JITTER starts 2 threads
JITTER thread on cpu 0 wakes up for refill
JITTER thread on cpu 1 wakes up for refill
Terminated

Instead if I wait after

Pool full at 4096, sleeping!

I got

Checking on done for thread 0

jent_read_entropy time on cpu 1 is 1.045238600000e+01 sec

Writing to pipe

DONE Writing to pipe with return -1

Error on pipe write: Bad file descriptor

jent_read_entropy time on cpu 0 is 3.505968987500e+01 sec

Writing to pipe

DONE Writing to pipe with return -1

Error on pipe write: Bad file descriptor

Closing thread 0

Checking on done for thread 1

Closing thread 1

but before the closing thread message is very slow

from rng-tools.

Also If I execute

rngd -d -l
Entropy sources that are available but disabled

1: TPM RNG Device (tpm)

and get stuck for a while
then it prints

Available and enabled entropy sources:

0: Hardware RNG Device (hwrng)

5: JITTER Entropy generator (jitter)

Checking on done for thread 0

but if I type

rngd -d -x jitter -l

is very fast

Disabling 5: JITTER Entropy generator (jitter)

Entropy sources that are available but disabled

1: TPM RNG Device (tpm)

5: JITTER Entropy generator (jitter)

Available and enabled entropy sources:

0: Hardware RNG Device (hwrng)

from rng-tools.

you don't have to disable jittier, thats where the problem is that we want to debug.

That said, I'm not sure how this is happening, on my system as soon as close the read side of the pipe, all the writers return immediately with the EBADF error, causing them to break out of their loop and exit.

Let me ask you, when you say "very slow" do you remember about how long it took to exit? Was it approximately 35 seconds?

from rng-tools.

time rngd -l
Entropy sources that are available but disabled

1: TPM RNG Device (tpm)

Available and enabled entropy sources:

0: Hardware RNG Device (hwrng)

5: JITTER Entropy generator (jitter)

Command terminated by signal 2
real    0m 12.59s
user    0m 21.41s
sys     0m 3.28s

time rngd -l -x jitter
Disabling 5: JITTER Entropy generator (jitter)

Entropy sources that are available but disabled

1: TPM RNG Device (tpm)

5: JITTER Entropy generator (jitter)

Available and enabled entropy sources:

0: Hardware RNG Device (hwrng)

Command exited with non-zero status 1
real    0m 0.00s
user    0m 0.00s
sys     0m 0.00s

from rng-tools.

ok, I think I know whats happening. if you look at
#67 (comment)

You'll see that the time spend generating entropy in the jitterentropy daemon is :
jent_read_entropy time on cpu 0 is 3.505968987500e+01 sec

Thats about 35 seconds, While sending pthread_kill(SIGINT) to the writer thread will make it break the loop if its blocked on a write, it won't cause it to return early from that generation function, meaning that you have to wait for some part of that 35 seconds to break out of the loop and exit. Thats not strictly speaking wrong, but it is a bit of an irritant. I might could try to square that up with some setjmp/longjump magic, but I'm not sure.

from rng-tools.

Ok my issue is that rngd get stuck on reboot
Now I have to look at another issue for now I disabled jitter and is working fine

I will try to add more printf and debug again (sadly is quite difficult to add gdb to this system :-( )

from rng-tools.

https://github.com/nhorman/rng-tools/tree/early-jitter-exit

I've created the new branch above, which uses sigsetjmp/siglongjump to effect an early return from jent_read_entropy, which should allow for earlier exiting from threads that are delayed by the gathering of entropy from the jitter library. Please build that branch and give it a try with jitterentropy enabled. If it works for you I'll merge it.

from rng-tools.

ping, any update?

from rng-tools.

Testing it now

from rng-tools.

no differences.

from rng-tools.

I will try to look at this next week (now I'm stuck with another issue).
if you want you can close it and I will reopen it with more debug information.

from rng-tools.

ok, saying no differences without additional information doesn't help. I'm going to add some debug information to this branch. I'll leave this open and when you have time, please run rngd with -d so debug information is recorded in the logs, and send that in.

from rng-tools.