log☇︎
28 entries in 1.081s
asciilifeform: since the debian incident, enemy stepped up the 'NOBUS' crapola; no noar '32768 possible keys, total', instead things moar in the spirit of http://qntra.net/2016/08/rng-whitening-bug-weakened-all-versions-of-gpg
diana_coman: http://btcbase.org/log/2017-11-14#1737414 <- confirmed; I do NOT use any nextprime or other "rng"-parts from gpg; current rsatron prototype simply grabs nbits from fg, flips the 2 top bits and 1 bottom bit as per previous discussion and then checks if result is prime; if prime then keep, otherwise discard and try again; no "add 2 until prime" or other such thing ☝︎
asciilifeform: ( she is using my sanitized gpg bignum. but i did not preserve koch's faux-rng atrocity ; so anything pertaining to entropy, is new )
mircea_popescu: the reason is that (in a translation of what koch-gpg does into sanity) you take 2045 bits of rng for each possible prime, stick 11 in front and 1 in the tail and THAT is your 2048 bit prime candidate.
asciilifeform: phun phakt, this calculation is taken from the gpg autopsies last summer, when asciilifeform was chasing imaginary rng boojum after somebody found a real one
a111: 21 results for "gpg rng", http://btcbase.org/log-search?q=gpg%20rng
asciilifeform: !#s gpg rng
asciilifeform: Barbarossa_: see also http://qntra.net/2016/08/rng-whitening-bug-weakened-all-versions-of-gpg
asciilifeform: see also http://qntra.net/2016/08/rng-whitening-bug-weakened-all-versions-of-gpg .
asciilifeform: veen: rng in gpg has serious problems , on top of using /dev/random
asciilifeform: i would say 'replace /dev/random in gpg source with /dev/fg and build' but the rng in gpg is monumentally retarded and i disrecommend its use entirely
asciilifeform: BingoBoingo: http://qntra.net/2016/08/rng-whitening-bug-weakened-all-versions-of-gpg/#comment-67939 << answr.
BingoBoingo: Question http://qntra.net/2016/08/rng-whitening-bug-weakened-all-versions-of-gpg/#comment-67915
mircea_popescu: ;;later tell bingoboingo "Following the news of a serious RNG bug affecting all GPG versions a low energy shitgnome campaign of apologetics and "not that bad" followed." << can i get a "The fact that hundreds of GPG keys have been Phuctored in the past year has, of course, nothing to do with all this." added ?
mircea_popescu: asciilifeform same people who check the gpg rng unwhitened.
asciilifeform: incidentally, mr. lolcow just gave away, unwittingly, another gpg laugh - recall what else comes out of rng when you generate key, after the primes ?
asciilifeform: ;;later tell mircea_popescu https://threatpost.com/gpg-patches-18-year-old-libgcrypt-rng-bug/119984 << the hannobockization is ready!!!1111
BingoBoingo: https://www.reddit.com/r/security/comments/4y8w7s/rng_whitening_bug_weakened_all_versions_of_gpg/
asciilifeform: ;;later tell mircea_popescu http://qntra.net/2016/08/rng-whitening-bug-weakened-all-versions-of-gpg/#comment-67543
BingoBoingo: https://www.reddit.com/r/netsec/comments/4y8xo1/rng_whitening_bug_weakens_gpg/ << try to browse to
mod6: <@deedbot> http://qntra.net/2016/08/rng-whitening-bug-weakened-all-versions-of-gpg/ << Qntra - RNG Whitening Bug Weakened All Versions of GPG << f.
deedbot: http://qntra.net/2016/08/rng-whitening-bug-weakened-all-versions-of-gpg/ << Qntra - RNG Whitening Bug Weakened All Versions of GPG
asciilifeform: mircea_popescu: my best hypothesis is a) khadeer generated key with, e.g., 'jihadcrypt' b) winblowz gpg with the memcpy from rng nopped out by ???
mircea_popescu: asciilifeform incidentally, "owned by whitening" is not altogether a bad theory wrt the null-entropy keys. ie, "they replaced rng with null-outputting one, never noticed because whitening". this, of course, doesn't explain why gpg would end up with null-generated keys, but whatevs.
asciilifeform: the design of gpg rng subsystem assumes extreme entropy-starvation. this is plain as daylight from 10 minutes of reading the src.
ascii_field: there are 3 separate rng subsystems in gpg 1.4 - but more on this later)
mircea_popescu: asciilifeform there's possibly more rng customers than gpg customers imo
mike_c: of course, to securely gpg it i would need a high-quality rng.. where could i get one of those i wonder.