Dear Christoph, * Christoph Biedl <mutt.c...@manchmal.in-ulm.de> [2016-07-23 10:29]: > Peter P. wrote ... > > > When sending encrypted mail from mutt using gpg I do get the following > > error message from gpg: > > > > gpg: DBG: syscall(getrandom) not supported; errno = 38 > > > > The mail is sent nevertheless, but I am wondering why the error message > > is there. A quick online search did not bring up something and I was > > womdering if this list could help me, or tell me where to ask next. > > Came across that message somewhere else, investigated and decided to > share as your question shows up when searching it in the net, but no > explanations. > > Executive summary: It's harmless, and it will go away in the future. Thank you! Indeed that was my experience (and I forgot to report back to this list). > Full story: A while ago, the kernel learned a new syscall "getrandom" > that eases access to random data, something always needed when > dealing with cryptography. The gpg programm uses the libgcrypt library > which tries to use that syscall first, which is a good thing. Upon > failure, it emits that warning, and falls back to the old style of > gaining random data. Which is not perfect but it worked the last > years. > > Code: > http://sources.debian.net/src/libgcrypt20/1.7.1-2/random/rndlinux.c/#L239 > > That syscall was introduced around kernel 3.16 but the Debian jessie > kernel appearently does not have it yet. The change in libgcrypt was > after 1.6.3 (as in Debian jessie) and at 1.7.2 (as in Debian stretch) > the latest. I can trigger the message in Debian by running gpg on a > jessie kernel (3.16) inside a stretch chroot. > > As soon as you'll switch to a newer kernel, libgcrypt will be happy > again. As happy as I am today ;) Thanks for your kind explanation!
Peter