Re: What DKIM RSA key length to use
That´s weird. I just tested this with simple/simple, relaxed/simple, relaxed/relaxed and simple/relaxed, all with a 2048 bits key, but all my messages got accepted. Can you reproduce this issue and share me the content of the mail (ncluding headers) that had the issue? I did send you something off list.
Re: What DKIM RSA key length to use
On Sun, 2021-04-11 at 04:13 +0200, Thomas Bohl wrote: > Hello, > > > In the filter-dkimsign readme I suggest to use 2048 and I stand by it. > > Thanks for mentioning and coding filter-dkimsign! Somehow I was unaware > of it. I used rspamd just for DKIM. Which is overkill. The daemon racks > up nearly 28000 daily DNS requests to free services (like dnswl.org, > senderscore.com, spamhaus.org etc.) just by running. (I didn't use it as > an inbound filter. I overwrote rbl.conf. I have no clue what it is > doing.) So I switched to filter-dkimsign. Glad you like it. > > I also switched to a 2048 bits key. Which looks good so far. Ironically > only dkimvalidator.com had a problem verifying until I relaxed the > canonicalization algorithms. That´s weird. I just tested this with simple/simple, relaxed/simple, relaxed/relaxed and simple/relaxed, all with a 2048 bits key, but all my messages got accepted. Can you reproduce this issue and share me the content of the mail (ncluding headers) that had the issue? > (Other tests like mail-tester.com or github.com/lieser/dkim_verifier had > no problem with it being simple.) > Cool, it´s always good to hear from more test-cases. martijn@
Re: What DKIM RSA key length to use
Hello, In the filter-dkimsign readme I suggest to use 2048 and I stand by it. Thanks for mentioning and coding filter-dkimsign! Somehow I was unaware of it. I used rspamd just for DKIM. Which is overkill. The daemon racks up nearly 28000 daily DNS requests to free services (like dnswl.org, senderscore.com, spamhaus.org etc.) just by running. (I didn't use it as an inbound filter. I overwrote rbl.conf. I have no clue what it is doing.) So I switched to filter-dkimsign. I also switched to a 2048 bits key. Which looks good so far. Ironically only dkimvalidator.com had a problem verifying until I relaxed the canonicalization algorithms. (Other tests like mail-tester.com or github.com/lieser/dkim_verifier had no problem with it being simple.)
Re: What DKIM RSA key length to use
In the filter-dkimsign readme I suggest to use 2048 and I stand by it. >From RFC1035: is a single length octet followed by that number of characters. is treated as binary information, and can be up to 256 characters in length (including the length octet). Followed by: TXT-DATAOne or more s. An RR has an RDLENGTH of 16 bits, meaning that you can fit about shy of 64k characters in the RR devided over aprox 256 character strings. (not accounting for the maximum DNS packet length). Since you already pointed to RFC6376, also have a look at section 3.6.2.2: Strings in a TXT RR MUST be concatenated together before use with no intervening whitespace. So if an implementation were to truncate a DKIM TXT record at the 255 bytes boundry, it is a violation of the RFC and the validator should be fixed. The same goes for SPF, which specifies in RFC7208 section 3.3: As defined in [RFC1035], Sections 3.3 and 3.3.14, a single text DNS record can be composed of more than one string. If a published record contains multiple character-strings, then the record MUST be treated as if those strings are concatenated together without adding spaces. martijn@ On Sun, 2021-03-28 at 16:46 +0200, Thomas Bohl wrote: > Hello, > > I only recently started to use DKIM and DMARC. (Yesterday to be exact. Now > mails to Gmail go to the inbox and not the spam-folder. Which is nice.) I > started with a 1024 bits RSA key. > > I followed > https://poolp.org/posts/2019-09-14/setting-up-a-mail-server-with-opensmtpd-dovecot-and-rspamd/ > and > https://prefetch.eu/blog/2020/email-server/#dkim > > poolp.org talks in length about why to use a 1024 bits key in order not to > truncate the DNS TXT record. > prefetch.eu uses 2048 bits and talks shortly about why not to use something > bigger. (Which makes sense, since RFC 6376 says that up to 2048 bits MUST be > supported and larger keys only MAY be.) > > Microsoft 365 talks about that 1024 and 2048 bitness is supported, but > defaults to 1024. > https://docs.microsoft.com/en-us/microsoft-365/security/office-365-security/use-dkim-to-validate-outbound-email?view=o365-worldwide#manually-upgrade-your-1024-bit-keys-to-2048-bit-dkim-encryption-keys > > Google Workspace recommends a key with 2048 bits, if your domain host can > manage it. > https://support.google.com/a/answer/174126 > > > I guess my question is: Is the problem with a truncate the DNS TXT record, > as described on poolp.org still a think to worry about, or have think > improved since 2019 and one can unhesitatingly use a > 2048 bits key? > > Thanks for reading > >
What DKIM RSA key length to use
Hello, I only recently started to use DKIM and DMARC. (Yesterday to be exact. Now mails to Gmail go to the inbox and not the spam-folder. Which is nice.) I started with a 1024 bits RSA key. I followed https://poolp.org/posts/2019-09-14/setting-up-a-mail-server-with-opensmtpd-dovecot-and-rspamd/ and https://prefetch.eu/blog/2020/email-server/#dkim poolp.org talks in length about why to use a 1024 bits key in order not to truncate the DNS TXT record. prefetch.eu uses 2048 bits and talks shortly about why not to use something bigger. (Which makes sense, since RFC 6376 says that up to 2048 bits MUST be supported and larger keys only MAY be.) Microsoft 365 talks about that 1024 and 2048 bitness is supported, but defaults to 1024. https://docs.microsoft.com/en-us/microsoft-365/security/office-365-security/use-dkim-to-validate-outbound-email?view=o365-worldwide#manually-upgrade-your-1024-bit-keys-to-2048-bit-dkim-encryption-keys Google Workspace recommends a key with 2048 bits, if your domain host can manage it. https://support.google.com/a/answer/174126 I guess my question is: Is the problem with a truncate the DNS TXT record, as described on poolp.org still a think to worry about, or have think improved since 2019 and one can unhesitatingly use a 2048 bits key? Thanks for reading
