Aki Tuomi wrote:> The use of salt, today, is to prevent the attacker from directly seeing > who has same passwords. Of course it also will make a rainbow table > attack less useful,Not just less useful, but almost infeasible. Given the use of random salts, you would have to generate (number of possible salts) rainbow tables. This drastically changes the CPU/storage tradeoffs.> but then again, no one uses rainbow tables anymore > since it takes about few minutes to brute force a password in the cloud > or on your home computer GPU. SHA512-CRYPT uses by default 4000 rounds > on dovecot, to make it more computationally expensive, but still it's > not very strong protection anymore.I know it is theoretically possible, but really, how feasible is it? The way people talk about it, they make it sound trivial. As a test, I just generated a hash using "doveadm pw". Any takers? {SHA512-CRYPT}$6$qyQANQFbQGqOSqHh$Z4YVIe.QakXtWs63lWUmzlseKpiFzfPCKhwM0NQk8jP6QeRmk6POkCdSrFBmVO2wh.dgPCPA7Gq3sNUeKOZLM/> MD5 with or without salt can be ATTAcked at passwords per second speed, > so using CRAM-MD5 of DIGEST-MD5 is very very poor choice. Anything with > MD5 should not be used for passwords these days. > > With Dovecot 2.3 we are making BLF-CRYPT available on all platforms, and > also adding support for ARGON2 when you have libsodium available.Straight hashes, without question, are weak. I routinely break NTLM hashes with rainbow tables. I'm *not* arguing against the use of stronger hashes, I'm just skeptical of the FUD surrounding the user of older salted hashes. This is especially true if the hashes are guarded against leakage, as most are. Strong crypt-hashes are an extra layer of protection (defense in depth). This page (if you trust 3rd parties) will report whether a password was ever found in a breached databases. You can also download the SHA1 hashes of all their known cracked passwords. It's big -- 5.3Gb 7-zipped file. https://haveibeenpwned.com/Passwords "j.emerlik" further asks:> You mean that today they using encrypted passwords to seeing > who has same passwords using brute force, hashes dictionary attack ?They precompute hashes for large number of password candidates, then look up hashes in that table. Rainbow tables are a way of folding large tables into a smaller ones, trading off space for extra CPU time. However, for hashes that use non-trivial salts, you would need to precompute an astronomically large number of tables. An attacker might be better off with a straight CPU-bound brute force search. Joseph Tam <jtam.home at gmail.com>
> On October 27, 2017 at 11:27 PM Joseph Tam <jtam.home at gmail.com> wrote: > > > Aki Tuomi wrote: > > > The use of salt, today, is to prevent the attacker from directly seeing > > who has same passwords. Of course it also will make a rainbow table > > attack less useful, > > Not just less useful, but almost infeasible. Given the use of random > salts, you would have to generate (number of possible salts) rainbow > tables. This drastically changes the CPU/storage tradeoffs. > > > but then again, no one uses rainbow tables anymore > > since it takes about few minutes to brute force a password in the cloud > > or on your home computer GPU. SHA512-CRYPT uses by default 4000 rounds > > on dovecot, to make it more computationally expensive, but still it's > > not very strong protection anymore. > > I know it is theoretically possible, but really, how feasible is it? > The way people talk about it, they make it sound trivial. As a test, > I just generated a hash using "doveadm pw". Any takers? > > {SHA512-CRYPT}$6$qyQANQFbQGqOSqHh$Z4YVIe.QakXtWs63lWUmzlseKpiFzfPCKhwM0NQk8jP6QeRmk6POkCdSrFBmVO2wh.dgPCPA7Gq3sNUeKOZLM/ >A random article on the internet says it's rather feasible if you want passwords cracked. Of course if the passwords are longer than, say, 8 characters, it becomes less feasible. My point wasn't to say that SHA512 is fully insecure, and adding rounds does make it less so as the expenses rack up. https://killtacknine.com/test-driving-google-cloud-gpus-with-hashcat/ It's not available on the graphics (puzzling), but if you look at the gist, it'll show that SHA512-CRYPT is cracked at 247.9 kH/s and MD5-CRYPT (salted MD5) at 17579.7 kH/s. As comparison, straight non-salted SHA512 goes at 1402.7 MH/s and MD5 with no salting at 33677.6 MH/s. For reference, DIGEST-MD5 uses MD5(username:realm:password) as the stored credential, which is pretty much straight MD5 for cracking. https://gist.github.com/anonymous/1fa176d82af79e879a9efa74c9d1c970> > MD5 with or without salt can be ATTAcked at passwords per second speed, > > so using CRAM-MD5 of DIGEST-MD5 is very very poor choice. Anything with > > MD5 should not be used for passwords these days. > > > > With Dovecot 2.3 we are making BLF-CRYPT available on all platforms, and > > also adding support for ARGON2 when you have libsodium available. > > Straight hashes, without question, are weak. I routinely break NTLM hashes > with rainbow tables. I'm *not* arguing against the use of stronger > hashes, I'm just skeptical of the FUD surrounding the user of older > salted hashes. This is especially true if the hashes are guarded > against leakage, as most are. Strong crypt-hashes are an extra layer > of protection (defense in depth). >Yet these leaks, they seem to occur.> This page (if you trust 3rd parties) will report whether a password > was ever found in a breached databases. You can also download the > SHA1 hashes of all their known cracked passwords. It's big -- 5.3Gb > 7-zipped file. > > https://haveibeenpwned.com/Passwords > > "j.emerlik" further asks: > > > You mean that today they using encrypted passwords to seeing > > who has same passwords using brute force, hashes dictionary attack ? > > They precompute hashes for large number of password candidates, then > look up hashes in that table. Rainbow tables are a way of folding large > tables into a smaller ones, trading off space for extra CPU time. > > However, for hashes that use non-trivial salts, you would need to > precompute an astronomically large number of tables. An attacker might > be better off with a straight CPU-bound brute force search. > > Joseph Tam <jtam.home at gmail.com>
On Sat, 28 Oct 2017, Aki Tuomi wrote:> A random article on the internet ...That would be funny to see in a bibliography. "Accoring to [RANDOM] ..."> says it's rather feasible if you want passwords cracked. Of course if > the passwords are longer than, say, 8 characters, it becomes less > feasible. My point wasn't to say that SHA512 is fully insecure, and > adding rounds does make it less so as the expenses rack up. > > https://killtacknine.com/test-driving-google-cloud-gpus-with-hashcat/ > > It's not available on the graphics (puzzling), but if you look at the > gist, it'll show that SHA512-CRYPT is cracked at 247.9 kH/s and > MD5-CRYPT (salted MD5) at 17579.7 kH/s.Well, let's see. The expected time to crack an 8 alphanumeric password (pretty much scraping the bottom of the strength policy barrel) 62^8/247900/2 ~ 14years (SHA512-CRYPT) 62^8/17579700/2 ~ ~72 days (MD5-CRYPT) Not trivial, but not reassuring, especially since processors get faster.>> This is especially true if the hashes are guarded against leakage, as >> most are. Strong crypt-hashes are an extra layer of protection >> (defense in depth). > > Yet these leaks, they seem to occur.True enough. Case in point: https://techcrunch.com/2016/08/30/dropbox-employees-password-reuse-led-to-theft-of-60m-user-credentials/ (Ironic note: even though half the passwords were using a straight SHA1 hash and thus much weaker, they did not leak the associated salt, and thus, were useless. The bcrypt hashes though, included the salt and were vulnerable.) Take away: use a strong password -- the hardest crypt algortihm in the world can't save you if you don't. Joseph Tam <jtam.home at gmail.com>
Aki, I understand that salted passwords saved in my database and stronger hash algorithm course that it will require more processor time/power to crack my passwords. But only when hackers have direct access to my database what means that hackers have access to my passwords hashes (eg. hackers stolen my database). My Dovecot use passwords saved in database as SHA256 and hackers can use only SMTP, IMAP or POP3 services to try crack it using dictionary attack (I understand that they using plain text dictionaty passwords). Stronger hash algorithm and salt is useful when hackers have direct access to my database but when they use services as SMTP, IMAP or POP3 to crack passwords only longer and more complicated password can be more secure. I do not understand this correctly ? Regards, Jacek 2017-10-28 0:39 GMT+02:00 Aki Tuomi <aki.tuomi at dovecot.fi>:> > > On October 27, 2017 at 11:27 PM Joseph Tam <jtam.home at gmail.com> wrote: > > > > > > Aki Tuomi wrote: > > > > > The use of salt, today, is to prevent the attacker from directly seeing > > > who has same passwords. Of course it also will make a rainbow table > > > attack less useful, > > > > Not just less useful, but almost infeasible. Given the use of random > > salts, you would have to generate (number of possible salts) rainbow > > tables. This drastically changes the CPU/storage tradeoffs. > > > > > but then again, no one uses rainbow tables anymore > > > since it takes about few minutes to brute force a password in the cloud > > > or on your home computer GPU. SHA512-CRYPT uses by default 4000 rounds > > > on dovecot, to make it more computationally expensive, but still it's > > > not very strong protection anymore. > > > > I know it is theoretically possible, but really, how feasible is it? > > The way people talk about it, they make it sound trivial. As a test, > > I just generated a hash using "doveadm pw". Any takers? > > > > {SHA512-CRYPT}$6$qyQANQFbQGqOSqHh$Z4YVIe. > QakXtWs63lWUmzlseKpiFzfPCKhwM0NQk8jP6QeRmk6POkCdSrFBmVO2wh. > dgPCPA7Gq3sNUeKOZLM/ > > > > A random article on the internet says it's rather feasible if you want > passwords cracked. Of course if the passwords are longer than, say, 8 > characters, it becomes less feasible. My point wasn't to say that SHA512 is > fully insecure, and adding rounds does make it less so as the expenses rack > up. > > https://killtacknine.com/test-driving-google-cloud-gpus-with-hashcat/ > > It's not available on the graphics (puzzling), but if you look at the > gist, it'll show that SHA512-CRYPT is cracked at 247.9 kH/s and MD5-CRYPT > (salted MD5) at 17579.7 kH/s. As comparison, straight non-salted SHA512 > goes at 1402.7 MH/s and MD5 with no salting at 33677.6 MH/s. For reference, > DIGEST-MD5 uses MD5(username:realm:password) as the stored credential, > which is pretty much straight MD5 for cracking. > > https://gist.github.com/anonymous/1fa176d82af79e879a9efa74c9d1c970 > > > > MD5 with or without salt can be ATTAcked at passwords per second speed, > > > so using CRAM-MD5 of DIGEST-MD5 is very very poor choice. Anything with > > > MD5 should not be used for passwords these days. > > > > > > With Dovecot 2.3 we are making BLF-CRYPT available on all platforms, > and > > > also adding support for ARGON2 when you have libsodium available. > > > > Straight hashes, without question, are weak. I routinely break NTLM > hashes > > with rainbow tables. I'm *not* arguing against the use of stronger > > hashes, I'm just skeptical of the FUD surrounding the user of older > > salted hashes. This is especially true if the hashes are guarded > > against leakage, as most are. Strong crypt-hashes are an extra layer > > of protection (defense in depth). > > > > Yet these leaks, they seem to occur. > > > This page (if you trust 3rd parties) will report whether a password > > was ever found in a breached databases. You can also download the > > SHA1 hashes of all their known cracked passwords. It's big -- 5.3Gb > > 7-zipped file. > > > > https://haveibeenpwned.com/Passwords > > > > "j.emerlik" further asks: > > > > > You mean that today they using encrypted passwords to seeing > > > who has same passwords using brute force, hashes dictionary attack ? > > > > They precompute hashes for large number of password candidates, then > > look up hashes in that table. Rainbow tables are a way of folding large > > tables into a smaller ones, trading off space for extra CPU time. > > > > However, for hashes that use non-trivial salts, you would need to > > precompute an astronomically large number of tables. An attacker might > > be better off with a straight CPU-bound brute force search. > > > > Joseph Tam <jtam.home at gmail.com> >