Log in to Gmail (IMAP/SMTP) using OAuth in backend

RESOLVED FIXED in Thunderbird 40.0

Status

Thunderbird
General
--
enhancement
RESOLVED FIXED
4 years ago
6 months ago

People

(Reporter: Nicholas Miell, Assigned: jcranmer, Mentored)

Tracking

(Depends on: 1 bug, Blocks: 3 bugs, {feature, user-doc-needed})

Trunk
Thunderbird 40.0
feature, user-doc-needed
Dependency tree / graph
Bug Flags:
in-moztrap +

Thunderbird Tracking Flags

(thunderbird38+ fixed, thunderbird39 fixed, thunderbird40 fixed)

Details

Attachments

(1 attachment, 7 obsolete attachments)

(Reporter)

Description

4 years ago
User Agent: Mozilla/5.0 (X11; Linux x86_64; rv:19.0) Gecko/20100101 Firefox/19.0
Build ID: 20130219173943

Steps to reproduce:

Added a Google Talk or Gmail account to Thunderbird.


Actual results:

I had to generate three different application specific passwords.


Expected results:

Thunderbird should use the SASL XOAUTH2 mechanism when logging in to Gmail using IMAP and SMTP or to Google Talk using XMPP.

https://developers.google.com/talk/jep_extensions/oauth
https://developers.google.com/google-apps/gmail/oauth_overview
This is related to having 2-step authentication enabled on your google account?
(Reporter)

Comment 2

4 years ago
Yes.
BenB: Do you think this is a good idea?
Severity: normal → enhancement
Component: Untriaged → General
Flags: needinfo?(ben.bucksch)
OS: Linux → All
Hardware: x86_64 → All
Version: 17 → Trunk

Comment 4

4 years ago
No. I don't see a reason why we should use a 2-factor auth.
- It's more network calls
- it's more complicated code
- Not many ISPs support it for IMAP

> I had to generate three different application specific passwords.

No, you don't. You can use the same password for IMAP and XMPP.

Suggesting WONTFIX.
Flags: needinfo?(ben.bucksch)
BenB: By 2-factor auth, do you mean OAuth? Are these the same thing?

Comment 6

4 years ago
My comment applies to both.

Comment 7

4 years ago
Google implements the 2-factor auth using OAuth.

Supporting it would involve Thunderbird poping up a browser window. I think that's a very bad idea to start with. Checking email shouldn't require a web browser.
BenB: Thanks for the info. I am not quite as expert on authentication as you are :) I agree with the WONTFIX.

Nicholas Miell: I have noticed many other people having problems recently as a result of enabling 2-step authentication in gmail. Although it does not look like your suggestion is the best fix, I think that we can do something else to help users in this regard.
Status: UNCONFIRMED → RESOLVED
Last Resolved: 4 years ago
Resolution: --- → WONTFIX
(Reporter)

Comment 9

4 years ago
This is literally the only way to log in to GMail without using an application specific password, there's nothing wrong with opening up a browser window to do it (especially when Thunderbird has one built in), and the OAuth authentication token can be cached until Google decides it has expired. And I haven't had any problems with Google's two-factor auth other than your refusal to support it.
(Assignee)

Comment 10

4 years ago
(In reply to Ben Bucksch (:BenB) from comment #7)
> Google implements the 2-factor auth using OAuth.
> 
> Supporting it would involve Thunderbird poping up a browser window. I think
> that's a very bad idea to start with. Checking email shouldn't require a web
> browser.

We can "pop up" the window in Thunderbird itself, and reading the documentation in detail, I think that it wouldn't be too hard to implement OAuth in TB. The IETF Kitten group has a draft for SASL and GSSAPI OAuth mechanisms (we already support GSSAPI, so in principle GMail could implement GSSAPI)--assuming that the platform GSSAPI calls add in support for OAuth, that leaves integration issues to other people (modulo things like bug 524698).

Since the SASL protocol gives no indication of how to find the auth servers, the only thing we could do is heuristically guess based off of either the server name or the AUTH=XOAUTH2 token; the former is extremely unpleasant to me, and the latter strikes me as unpleasant. I don't think it's worth supporting for IMAP, but I could see an argument for XMPP as we already special-case Google Talk.

The biggest reason to support this is to make life easier for people who turn on two-factor auth, which, quite frankly, is likely limited to people who are not going to be scared off by generating app-specific passwords. We're not tied to Google, and this is an authentication mechanism that is inherently tied to Google--hence I concur with the WONTFIX.

Comment 11

4 years ago
There are bigger ramifications here. For me, one of the primary purposes of Thunderbird is to keep email an open and viable communication method, and to preserve open standards that can be implemented by anyone. In other words, one of the purposes of Thunderbird is to allow for other clients as well, on all kinds of platforms, for all kinds of usecases, not all of which are interactive (see e.g. Android app "SMS Backup+").

By supporting OAuth in Thunderbird, we make it more likely that Google will make such obnoxious auth methods mandatory at some point in the future.

While it may be possible for Thunderbird to open a web browser window, it is not possible for other clients.  Any email client would have to have a web browser, which I personally find ridiculous and dangerous.

More generally, right now, ISPs are limited to what the IMAP standard allows, and to the specific purpose of email. If we open a browser window and make auth dependent on that, it means that we hand control entirely over to Google. Google can do in that window whatever they want, and make completely arbitrary demands on IMAP users. Currently, IMAP sets the rules. This would be over.

So, I consider this to be a very dangerous move for the freedom of email.

Comment 12

3 years ago
Reopening in new light of http://googleonlinesecurity.blogspot.co.uk/2014/04/new-security-measures-will-affect-older.html
Status: RESOLVED → REOPENED
Ever confirmed: true
Resolution: WONTFIX → ---

Updated

3 years ago
Status: REOPENED → NEW

Comment 13

3 years ago
I still stand to my comment 11. We cannot let a single provider do whatever they want. Even more so when that single provider has 20% of the user base. Then it's all the more dangerous, because there's nobody to keep them in check.

Email != Web. We cannot allow email to depend on the web, or proprietary auth mechanisms. Playing along here will be highly damaging to email and the Internet. We need to insist on the email standards. This is Thunderbird's mission.

Comment 14

3 years ago
http://en.wikipedia.org/wiki/Embrace,_extend_and_extinguish

Comment 15

3 years ago
Quote:
"Embrace, extend, and extinguish" is a phrase that the U.S. Department of Justice found was used internally by Microsoft to describe its strategy for entering product categories involving widely used standards, extending those standards with proprietary capabilities, and then using those differences to disadvantage its competitors.

That's precisely what's happening here. Google's strategy is to bring everything to the web, and everything on their services and servers. Including email.

They are
1. "entering product categories involving widely used standards" -- email, using gmail
2. "extending those standards with proprietary capabilities" -- OAuth for IMAP, this is happening right now
3. "using those differences to disadvantage its competitors" -- because soon you won't be able to use an email client without a web browser, which is a huge disadvantage for most situations where IMAP/SMTP is used. So why not use GMail web frontend straight? Email dead.

I don't care what rationale they use. Steps 2 and 3 are dangerous.

(And forcing me to enter a working phone number - and *any* number -, just to access my own account, doesn't count as "security" for me, but as privacy violation.)

Google knows they can't just cut off Thunderbird. The blog post didn't say they will, it's vaguely phrased. This is power play, and they are testing waters. But if they go ahead with this and require OAuth or otherwise make using Thunderbird hard, we need to jump up and down in the press and cry "foul". I'm not giving in.
I think the tb-planning discussion covers the meta aspects pretty well, but in general, oauth2 is a win for the user in terms of protecting their account.

I'm not sure the web browser argument is a reasonable one.  While an email client shouldn't be a web browser, it's very common in application platforms now to provide embeddable web browser widgets to allow for flows like this.  (Or even to provide explicit mechanisms to enable oauth2-style flows.)

In cases like Firefox OS we are able to do oauth2 while respecting the underlying origin-based security mechanism of the web.  The Gaia email app is not able to corrupt the oauth2 process (although impersonation issues have not been conclusively addressed by the platform yet.)

More specifically, it's not clear what alternative could allow Google to improve their security situation using an in-channel mechanism other than proposing an IMAP extension.  (Application specific passwords are already an option and presumably will not be impacted by the first phrase of Google's roll-out.)  The IMAP extension would require all the private data to be funneled through the email app.  Since most all logic in Thunderbird runs with chrome privileges, a compromise of Thunderbird is still a compromise of the credentials, but in cases like the Gaia Email app, such an IMAP protocol is much riskier than the oauth2 approach.

Anyways, I would then argue that the IMAP extension, besides being less secure, is also going to be a more proprietary thing than oauth2 which is an already-adopted standard, as these things go.  When people are constrained by IMAP, that's basically saying they have to implement a (potentially broken) subset of RFC 3501 and maybe some other stuff on top; it would be surprising to see them implement such an extension.


I'll probably refrain from discussing further on this bug.  I'm not an active contributor to Thunderbird, but this debate is relevant to the Gaia email app so I do want to keep informed about what Thunderbird does.  I will note that the Gaia email app will be implementing oauth2 support and using it by default, but we wouldn't do anything to stop the use of the manual configuration mode to use usernames/passwords with gmail.
(Assignee)

Comment 17

3 years ago
(In reply to Magnus Melin from comment #12)
> Reopening in new light of
> http://googleonlinesecurity.blogspot.co.uk/2014/04/new-security-measures-
> will-affect-older.html

(In reply to Ben Bucksch (:BenB) from comment #13)
> I still stand to my comment 11. We cannot let a single provider do whatever
> they want. Even more so when that single provider has 20% of the user base.
> Then it's all the more dangerous, because there's nobody to keep them in
> check.

To carry over some of the discussion from tb-planning:

I personally stand by comment 10: until there is a standardized mechanism to look up the details on how to access the OAuth server, I do not wish to see this implemented. Andrew Sutherland has mentioned that Google appears to have some stuff set up to make this possible but it's not fully set up in an end-to-end usable approach.

Should such a mechanism exist and be informally or formally specified, I don't object to this feature. Until that mechanism exists, I object.
(Assignee)

Comment 18

3 years ago
(In reply to Andrew Sutherland (:asuth) from comment #16)
> More specifically, it's not clear what alternative could allow Google to
> improve their security situation using an in-channel mechanism other than
> proposing an IMAP extension.

Mandate SASL SCRAM or other powerful authentication techniques? A well-coded client that refuses to step down from an encrypted SASL mechanism can never be forced to divulge a users' password, even in extreme circumstances like connecting to a server that managed to somehow validate a certificate.

OAuth2 assumes that the data must be authorized in an external mechanism. Even in the case of B2G, it's possible to do the authorization manually and internally and virtually undetectably. HTML widgets are basically built-in to most GUI frameworks, running the HTTP yourself isn't a difficult task, and virtually all standard networking libraries allow you to easily wrap it in SSL (if really insecurely, too!). When the authorization step can be intercepted, you can just as easily get the password. So the use of external authorization is really a myth: the server has to assume that the authorization has to be under the control of the client anyways.

And when you look at how OAuth2 authorization is supposed to work... it is really, really easy to implement OAuth2 completely insecurely and in such a manner for a legitimate client to be inadvertently tricked into divulging the password to an attacking party. In contrast, a SCRAM-like approach only requires the client to refuse to fallback from a secure implementation--that won't necessarily happen, but it's much less (and more obvious) steps to writing a secure implementation than with OAuth2.

> Anyways, I would then argue that the IMAP extension, besides being less
> secure, is also going to be a more proprietary thing than oauth2 which is an
> already-adopted standard, as these things go.

"Standard." A large number of details about authorization parameters are woefully underspecified (== clients must guess them), not least of which is which authorization server to ask in the first place. In contrast, SASL and GSSAPI require protocols and mechanisms to fully specify details such as "what service name should I use?".

Comment 19

3 years ago
> it's not clear what alternative could allow Google to improve their security situation
> using an in-channel mechanism other than proposing an IMAP extension.

Lesson from "West Wing": Don't accept the premise. I don't accept that we have a problem in the first place.

Simple solution: If Google thinks that a particular, concrete login attempt might be fraudulent (e.g. because I usually log in from Germany and this attempt is suddenly from Russia), then they can simply block this login attempt, including a IMAP/POP3/SMTP error message in plain language that explains why they block it and what to do to unblock (e.g. "You are logging in from a new and unusual country. Please log in to http://gmail.com first and approve this attempt. Then try again."). This would be 1) self-explanatory for every use 2) not require any new standards 3) work with all current and well-implemented software.

> Application specific passwords are already an option

So, if that works, I don't see a problem.

Aside from the initial setup, of course. Again, well-crafted error messages (which I think all standard protocols allow) do wonders here for usability.
(In reply to Ben Bucksch (:BenB) from comment #19)
> Simple solution: If Google thinks that a particular, concrete login attempt
> might be fraudulent (e.g. because I usually log in from Germany and this
> attempt is suddenly from Russia), then they can simply block this login
> attempt, including a IMAP/POP3/SMTP error message in plain language that
> explains why they block it and what to do to unblock (e.g. "You are logging
> in from a new and unusual country. Please log in to http://gmail.com first
> and approve this attempt. Then try again."). This would be 1)
> self-explanatory for every use 2) not require any new standards 3) work with
> all current and well-implemented software.

The main problem with this is if the user is using single-factor authentication (which again, I am inferring is Google's primary initial concern), the only way to distinguish the user from an attacker with the user's credentials are:
- by IP address, which assumes the attacker has no way of using a more convincing IP if this is applied at a broad geographic scale
- depending on secondary tokens handed out as part of the normal 1-factor login process (like a login cookie that was minted in the past, before the user started acting sketchy)
- by using Google's recovery procedures and possibly invalidating the existing credentials.  These usually involve other e-mail addresses on file, phone numbers on file, or, for last-ditch recovery purposes, knowing exactly when the account was created and maybe using the exact IP address or class C subnet that was originally used.


In terms of error messages, gmail actually already does this.  They provide a structured specific error code that software can identify, plus they include human-readable text and a link to a support document.  Frequently not the world's best support document, but a good starting point.

A big problem with these error messages is that most servers provide useless, unlocalized, confusing error messages so email clients have an incentive to not show the error message directly to the user.  I think I've come around to the idea, however, that if you see something that looks like a URL in the error message, then it is something useful and should be presented to the user and possibly even automatically browsed to.
 

> > Application specific passwords are already an option
> 
> So, if that works, I don't see a problem.
> 
> Aside from the initial setup, of course. Again, well-crafted error messages
> (which I think all standard protocols allow) do wonders here for usability.

Here's the documentation on this, and I think which is indirectly linked to by the existing error message:  https://support.google.com/mail/answer/1173270?hl=en

Note that this is a less horrible work-flow for Thunderbird than it is on a tiny mobile phone (which in the case of Firefox OS, currently lacks copy and paste)
(In reply to Joshua Cranmer [:jcranmer] from comment #18)
> (In reply to Andrew Sutherland (:asuth) from comment #16)
> > More specifically, it's not clear what alternative could allow Google to
> > improve their security situation using an in-channel mechanism other than
> > proposing an IMAP extension.
> 
> Mandate SASL SCRAM or other powerful authentication techniques? A well-coded
> client that refuses to step down from an encrypted SASL mechanism can never
> be forced to divulge a users' password, even in extreme circumstances like
> connecting to a server that managed to somehow validate a certificate.

SCRAM-SHA-1 is clearly a big improvement over sending the password in the clear, even over an encrypted connection.  But it's also an orthogonal issue.  It would be an improvement if the oauth2 flow used SCRAM-SHA-1 to avoid sending the token in the clear.


> OAuth2 assumes that the data must be authorized in an external mechanism.

I think the thrust of your argument here is that if you can get the user to type in the password, then you can steal the user's password while convincing them that they're being safe.

The greater context here that Google is trying to enable 2-factor authentication across the board, and that's why they want people to adopt OAuth2.  If the user is required to type a password in, plus a code that gets texted to them, then you stealing their password only grants you an access window *once* to be 2-factor authenticated.  You can do a ton of evil stuff during that time window including minting other tokens, but unless you change how the two-factor auth mechanism works, you can't get two-factor authenticated again in the future.  (And I should note that Google is pretty smart about making you use a new text-message/TOTP code to do dangerous things even if you've really recently re-authed.)


Also, in many cases the device in use will already have trusted Google credentials, so it's not a question of the user needing to enter a password again, it's a question of saying "yes" or "no" to the answer of whether Thunderbird should be given read/write access to your mail store.  You can fake a "yes"/"no" dialog, but that obviously won't get you anywhere.  (And users are getting used to this, so they may in fact think twice before typing in their password.)  In a browser, the user is probably already authenticated to Google, and so can say yes/no (depending on Google's paranoia for the scope).  On Android, same deal, but web activities potentially get involved.  I believe we may even be able to do the right thing on Firefox OS so you don't need to actually be typing your password in if you've already authenticated.


> Even in the case of B2G, it's possible to do the authorization manually and
> internally and virtually undetectably. HTML widgets are basically built-in
> to most GUI frameworks, running the HTTP yourself isn't a difficult task,
> and virtually all standard networking libraries allow you to easily wrap it
> in SSL (if really insecurely, too!). When the authorization step can be
> intercepted, you can just as easily get the password. So the use of external
> authorization is really a myth: the server has to assume that the
> authorization has to be under the control of the client anyways.




> 
> And when you look at how OAuth2 authorization is supposed to work... it is
> really, really easy to implement OAuth2 completely insecurely and in such a
> manner for a legitimate client to be inadvertently tricked into divulging
> the password to an attacking party. In contrast, a SCRAM-like approach only
> requires the client to refuse to fallback from a secure implementation--that
> won't necessarily happen, but it's much less (and more obvious) steps to
> writing a secure implementation than with OAuth2.
> 
> > Anyways, I would then argue that the IMAP extension, besides being less
> > secure, is also going to be a more proprietary thing than oauth2 which is an
> > already-adopted standard, as these things go.
> 
> "Standard." A large number of details about authorization parameters are
> woefully underspecified (== clients must guess them), not least of which is
> which authorization server to ask in the first place. In contrast, SASL and
> GSSAPI require protocols and mechanisms to fully specify details such as
> "what service name should I use?".
(Sorry about the un-trimmed quoted stuff, accidental submit.  I had nothing more to say about those quote excerpts.)

Comment 23

3 years ago
> the main problem with this is if the user is using single-factor authentication ...,
> the only way to distinguish the user from an attacker with the user's credentials are:
> - by IP address
> - by using Google's recovery procedures and possibly invalidating the existing credentials.
>   These usually involve other e-mail addresses on file, phone numbers on file ...

Why recovery by third party means? If Google thinks this attempt is suspicious, it can give the error message and require the user to log in on gmail.com manually (if that helps Google someway).

If Google tries to use OAuth2 to identify me by earlier-set cookies in the browser, then we won't have these cookies anyway.

Comment 24

3 years ago
> A big problem with these error messages is that most servers provide useless,
> unlocalized, confusing error messages so email clients have an incentive to not
> show the error message directly to the user.

What I usually do my in apps is to show a simple, dumbed-down error message, then further down in the dialog the raw server message. E.g. alert("Login failed, please check username and password\n\n" + servermessage); A client that doesn't show the server's error message to the user is reckless IMHO. That's much easier fixed than implementing OAuth, too.

> [URLs from browser] even automatically browsed to.

Uh, you don't want to do that automatically. The URL is coming from the server, and you can't know what it does, nor whether it's related to the same business. That could get nasty.

> Note that this is a less horrible work-flow

Google required me to enter a valid phone number, against my explicit will, and type the code sent there. Media break. And the first attempt failed. Google's 2-factor is a horrible work-flow either way, and that's entirely on Google. So, that's beside the point.
(In reply to Ben Bucksch (:BenB) from comment #23)
> Why recovery by third party means? If Google thinks this attempt is
> suspicious, it can give the error message and require the user to log in on
> gmail.com manually (if that helps Google someway).

My point was that if the user logs into google.com with username/password, and Thunderbird logs into email with the same username/password, then an attacker who gets a warning message "log in via google.com" can absolutely log in via google.com and say "oh yes, that was totally me" if there's no other information involved.


(In reply to Ben Bucksch (:BenB) from comment #24)
> > [URLs from browser] even automatically browsed to.
> 
> Uh, you don't want to do that automatically. The URL is coming from the
> server, and you can't know what it does, nor whether it's related to the
> same business. That could get nasty.

Hm, yeah, thinking about it more, we definitely need to frame the situation using at least one UI card, so we might as well show the URI and a prompt to ask if they want to browse there.  Note that I am only talking about doing this in Firefox OS where we require a secure connection and the link would end up getting opened in its own attenuated subprocess.

 
> > Note that this is a less horrible work-flow
> 
> Google required me to enter a valid phone number, against my explicit will,
> and type the code sent there. Media break. And the first attempt failed.
> Google's 2-factor is a horrible work-flow either way, and that's entirely on
> Google. So, that's beside the point.

Google's 2-factor auth allows for either using text messages for primary or the constantly changing TOTP code stuff (http://en.wikipedia.org/wiki/Time-based_One-time_Password_Algorithm), with fallback one-shot codes you can generate.  I use the TOTP stuff.

If you're trying to establish a new account by just going to the webpage without any other context, it's possible Google may try and use text messages to stop spammers from creating new accounts/bulk accounts/etc.  That is a tricky problem, but there may be other avenues available.  Gmail seems to let me send invites non-gmail accounts; that presumably might be sufficient for them to not require use of a text message since there's some trust graph stuff in there.

Comment 26

3 years ago
> My point was that if the user logs into google.com with username/password, and Thunderbird logs
> into email with the same username/password, then an attacker who gets a warning message
> "log in via google.com" can absolutely log in via google.com and say
> "oh yes, that was totally me" if there's no other information involved.

Right, but that problem is inherent. If you assume the attacker has the master password, then Thunderbird's auth method doesn't matter anymore. The account is already lost. If the attacker can log into gmail.com just fine, then he can do any OAuth as well, so nothing is gained. (!)

---

FWIW, as for IMAP without SSL, they easily can and should cut that off, and all big German ISPs did this this month. With IMAP+SSL, I don't see how IMAP clients would be more risky to lose passwords than a web browser logging into gmail.com. I'd argue they are much less risky than a web browser, because phishing is much easier on the web than on email clients. So, I don't buy the whole point. 

But if there is a point, in a specific, suspicious situation, they can do the error message trick.

Comment 27

3 years ago
Just wanted to point out that this is not Google specific, Hotmail (now Outlook) also uses Oauth+IMAP when multifactor authentication is enabled.
(Reporter)

Comment 28

3 years ago
(In reply to Stephen Baker from comment #27)
> Just wanted to point out that this is not Google specific, Hotmail (now
> Outlook) also uses Oauth+IMAP when multifactor authentication is enabled.

Unfortunately, the OAuth endpoint or whatever it's called and the list of permissions needed to interact with IMAP are completely different from GMail, and there's still no way at the protocol level to discover those details. (Once you have them, though, the rest of the protocol is identical.)

Comment 29

3 years ago
Google doesn't allow me to login anymore in Thunderbird using SSL/TLS, sign-in attempts are blocked. Please implement Oauth2.

The only 'work-around' is currently to enable 'less safe' apps via:
https://www.google.com/settings/security/lesssecureapps
(In reply to Melroy from comment #29)
> The only 'work-around' is currently to enable 'less safe' apps via:
> https://www.google.com/settings/security/lesssecureapps

Note that anyone who had been using Thunderbird (or any other IMAP client that uses PLAIN auth) within some recent time period around when the switchover occurred around ~July 15th, 2014 will already have the setting enabled.  Likewise, any new gmail accounts that start using Thunderbird within a month of being created.

Reference: http://mailman13.u.washington.edu/pipermail/imap-protocol/2014-September/002326.html

Also, "less secure apps" is irrelevant if 2-factor auth is enabled.  In that case you need to create an application-specific password for Thunderbird until Thunderbird supports OAuth2.  In general I'd suggest moving to 2-factor auth regardless.
(Assignee)

Comment 31

3 years ago
(In reply to Andrew Sutherland [:asuth] from comment #30)
> (In reply to Melroy from comment #29)
> > The only 'work-around' is currently to enable 'less safe' apps via:
> > https://www.google.com/settings/security/lesssecureapps
> 
> Note that anyone who had been using Thunderbird (or any other IMAP client
> that uses PLAIN auth) within some recent time period around when the
> switchover occurred around ~July 15th, 2014 will already have the setting
> enabled.  Likewise, any new gmail accounts that start using Thunderbird
> within a month of being created.

I pointed out when the issue first came out that I did not wish to implement it so long as I had to hardcode various values. That they rolled out changes without providing a generic mechanism, even after I (and others supported me on the IMAP mailing list) pointed out to them the critical issue of lack of support is to me disreputable and underhanded.

> Also, "less secure apps" is irrelevant if 2-factor auth is enabled.  In that
> case you need to create an application-specific password for Thunderbird
> until Thunderbird supports OAuth2.  In general I'd suggest moving to
> 2-factor auth regardless.

We will have to eventually, much as I don't like it. From what I've talked to Google before, they do badly want us to move as well. But I've communicated that the price of support is essentially them rolling out a dynamic client registration and endpoint discovery mechanism, and I don't want to change that opinion, especially since I fear they won't be motivated to roll those out if we don't push back hard on it.
(In reply to Joshua Cranmer [:jcranmer] from comment #31)
> I pointed out when the issue first came out that I did not wish to implement
> it so long as I had to hardcode various values. That they rolled out changes
> without providing a generic mechanism, even after I (and others supported me
> on the IMAP mailing list) pointed out to them the critical issue of lack of
> support is to me disreputable and underhanded.

I agree that it would have been preferable for their XOAuth2 implementation to avoid requiring any hardcoding/registration before introducing anything that resembles a "stick" (versus a "carrot").  But I think it's appropriate to keep in mind that the context here is people having their email accounts compromised.  Login via https is more amenable to making sure the user is the user, whereas IMAP state is basically IP address and username/password.  The former is not reliably stable for most users, the latter is something that is all-too-easily compromised through phishing or through reusing the same password across multiple accounts or other poor password hygiene.  (Reiterating that 2-factor auth is completely unaffected by the "less secure apps" feature.)

Given that Google (in the guise of Brandon Long) indicated this was done as part of the abuse team's efforts, I do think it's entirely reasonable to add a single flag that's set based on historical usage heuristics.  Which is to say I think you're attributing malice or some kind of Machiavellian plotting here that does not match up with the reality.  If they had completely turned off non-XOAuth2 (AKA PLAIN) auth, that would be another thing.  Or if they had just set the setting to "disabled" for everyone including those who were obviously already using it.  But they didn't.

Which is to say, it's fair for Thunderbird to hold out for not having to hardcode Google-specific things, but it's not fair to condemn Google for addressing the very real problem of account compromise.

Comment 33

3 years ago
2 issues here:
1) OAUTH in IMAP
Even with the server discovery solved, my biggest problem would still stay:

I don't want email to depend on the web. That's just not acceptable in any case, for any reason.

It's basically getting rid of email as independent protocol and sucking everything into webmail. If GMail wants to drop support for IMAP, they should say so outright. Requiring OAuth is just the hidden way to do it.

2) 2-factor auth

> not fair to condemn Google for addressing the very real problem of account compromise.

The way they address it is just wrong.

Just because there's a lion running free doesn't mean that firing shotguns in all directions, including shooting innocent people, is the right solution.

For example, requiring an SMS just to log in to email is breaking stuff. Case just today: Neil can't log in to the test accounts I gave him, because the Hotmail and GMail are asking stupid security questions or want to send an SMS. If email depends having my mobile phone reachable (which I typically do not), this is breaking email as we know it. I don't care why anybody wants that, it's unacceptable.

FWIW, the "security questions" that they use to "recover passwords" are the very reason why accounts are being hacked - even those with strong passwords. The recent Apple iCloud hack that published nude celeb pics was partially done this way. Likewise, I don't want my strong password being circumvented by the extremely weak GSM protocols - it's well documented that GSM security is near null - base stations aren't even authenticated in any way. The more SMS (or any other method) is being used for auth, the more this vector will be attacked. There are juicy rewards, so there'll be criminal energy.

2-factor auth isn't the solution just because they say it is.

Updated

3 years ago
Keywords: feature

Comment 34

3 years ago
(In reply to Ben Bucksch (:BenB) from comment #33)
> 2 issues here:
> 1) OAUTH in IMAP
> Even with the server discovery solved, my biggest problem would still stay:
> 
> I don't want email to depend on the web. That's just not acceptable in any
> case, for any reason.
> 
> It's basically getting rid of email as independent protocol and sucking
> everything into webmail. If GMail wants to drop support for IMAP, they
> should say so outright. Requiring OAuth is just the hidden way to do it.
> 
> 2) 2-factor auth
> 
> > not fair to condemn Google for addressing the very real problem of account compromise.
> 
> The way they address it is just wrong.
> 
> Just because there's a lion running free doesn't mean that firing shotguns
> in all directions, including shooting innocent people, is the right solution.
> 
> For example, requiring an SMS just to log in to email is breaking stuff.
> Case just today: Neil can't log in to the test accounts I gave him, because
> the Hotmail and GMail are asking stupid security questions or want to send
> an SMS. If email depends having my mobile phone reachable (which I typically
> do not), this is breaking email as we know it. I don't care why anybody
> wants that, it's unacceptable.
> 
> FWIW, the "security questions" that they use to "recover passwords" are the
> very reason why accounts are being hacked - even those with strong
> passwords. The recent Apple iCloud hack that published nude celeb pics was
> partially done this way. Likewise, I don't want my strong password being
> circumvented by the extremely weak GSM protocols - it's well documented that
> GSM security is near null - base stations aren't even authenticated in any
> way. The more SMS (or any other method) is being used for auth, the more
> this vector will be attacked. There are juicy rewards, so there'll be
> criminal energy.
> 
> 2-factor auth isn't the solution just because they say it is.

I think you’re the only one on track her.  If we are talking about security, OAuth 2.0 is a joke. It’s easy to circumvent, as long as you’re thinking like a true black hat ;) The other factor to note is the hidden agenda, were talking about google. There is a huge difference between being forced to disclose your real identity by proxy and being able to authenticate using an anonymous persona, guess what oorth2 enforces, security was never the propriety, with google, it’s all about who you ‘really’, where you are, why and what you’re thinking, what you ate this lunchtime, when your horny, when you’re not, and so on. True security is meant to enforce privacy not obliterate it.

Comment 35

3 years ago
(In reply to John Burges from comment #34)
> (In reply to Ben Bucksch (:BenB) from comment #33)
> > 2 issues here:
> > 1) OAUTH in IMAP
> > Even with the server discovery solved, my biggest problem would still stay:
> > 
> > I don't want email to depend on the web. That's just not acceptable in any
> > case, for any reason.
> > 
> > It's basically getting rid of email as independent protocol and sucking
> > everything into webmail. If GMail wants to drop support for IMAP, they
> > should say so outright. Requiring OAuth is just the hidden way to do it.
> > 
> > 2) 2-factor auth
> > 
> > > not fair to condemn Google for addressing the very real problem of account compromise.
> > 
> > The way they address it is just wrong.
> > 
> > Just because there's a lion running free doesn't mean that firing shotguns
> > in all directions, including shooting innocent people, is the right solution.
> > 
> > For example, requiring an SMS just to log in to email is breaking stuff.
> > Case just today: Neil can't log in to the test accounts I gave him, because
> > the Hotmail and GMail are asking stupid security questions or want to send
> > an SMS. If email depends having my mobile phone reachable (which I typically
> > do not), this is breaking email as we know it. I don't care why anybody
> > wants that, it's unacceptable.
> > 
> > FWIW, the "security questions" that they use to "recover passwords" are the
> > very reason why accounts are being hacked - even those with strong
> > passwords. The recent Apple iCloud hack that published nude celeb pics was
> > partially done this way. Likewise, I don't want my strong password being
> > circumvented by the extremely weak GSM protocols - it's well documented that
> > GSM security is near null - base stations aren't even authenticated in any
> > way. The more SMS (or any other method) is being used for auth, the more
> > this vector will be attacked. There are juicy rewards, so there'll be
> > criminal energy.
> > 
> > 2-factor auth isn't the solution just because they say it is.
> 
> I think you’re the only one on track her.  If we are talking about security,
> OAuth 2.0 is a joke. It’s easy to circumvent, as long as you’re thinking
> like a true black hat ;) The other factor to note is the hidden agenda, were
> talking about google. There is a huge difference between being forced to
> disclose your real identity by proxy and being able to authenticate using an
> anonymous persona, guess what oorth2 enforces, security was never the
> propriety, with google, it’s all about who you ‘really’, where you are, why
> and what you’re thinking, what you ate this lunchtime, when your horny, when
> you’re not, and so on. True security is meant to enforce privacy not
> obliterate it.

Hummm, no ability to correct typos. Dam.

Comment 36

3 years ago
http://www.cnet.com/news/serious-security-flaw-in-oauth-and-openid-discovered/

Comment 37

3 years ago
http://hueniverse.com/2012/07/26/oauth-2-0-and-the-road-to-hell/

Comment 38

3 years ago
Perhaps the best route is to educate the user base as to why google is such a bad idea, why they hope to Embrace, extend and extinguish. How the user is sleepwalking into a hostage situation. Provide an excellent alternative (is there one, then create one). Google is best at maps and email, everything else is superfluous. It’s time we stop pandering to this monster and start creating a real alternative. Why Kim dot com hadn’t extended mega to compete in the user service space other than some file sharing, I’m not sure, given is Orwellian credentials.  A migration tool is all that is needed, make the barrier to change zero effort, and you have a winner. Google are evil, not doubt about it.
(Reporter)

Comment 39

3 years ago
This bug has evidently become a chat thread for the mentally ill instead of a Thunderbird feature request, so I'm closing it because I'm tired of it polluting my inbox.

If you stumbled upon this bug because you also wanted OAuth support in Thunderbird, I invite you to take a look at KMail. Not only can it log in to Gmail using OAuth, it is smart enough to not store redundant copies of messages that have multiple labels (thanks to the X-GM-MSGID attribute) and it IMAP IDLEs on the [All Mail] folder instead of INBOX, so it notices message deliveries that are automatically labeled by the user's configured filters or bypass the Inbox entirely (thanks to the X-GM-LABELS attribute).
Status: NEW → RESOLVED
Last Resolved: 4 years ago3 years ago
Resolution: --- → WONTFIX

Comment 40

3 years ago
Reopening, but can people please refrain form advocacy comments.
We would like this for tb38, currently setting up a new gmail account is (for a user not having used gmail imap) is broken.
Status: RESOLVED → REOPENED
Resolution: WONTFIX → ---

Updated

3 years ago
Status: REOPENED → NEW

Comment 41

3 years ago
> currently setting up a new gmail account is (for a user not having used gmail imap) is broken.

Can we fix this in some other way, e.g. helping them to enable proper IMAP with normal auth?

The ISPDB XML provides a way to say "You have to manually do A, B, C to set up, here's a link where you can do it".

Comment 42

3 years ago
The use of that data is not implemented - bug 586364. Of course it would be better than status quo, but still, that's a big hassle that probably lose a lot of people on the way.

Comment 43

3 years ago
> http://hueniverse.com/2012/07/26/oauth-2-0-and-the-road-to-hell/

This article is from the lead author of OAuth 2.0 and definitely worth a read.
(Thanks for posting that, dear "mentally ill" colleague.)

> The use of that data is not implemented - bug 586364

I had always intended to do that. We also need that for other ISPs with other reasons, e.g. the largest German ISP has 2 different passwords, one for webmail, one for IMAP, and somebody needs to explain that to our users, or they'll be lost. We'll need something like this in any case.
(Assignee)

Comment 44

3 years ago
(In reply to Nicholas Miell from comment #39)
> This bug has evidently become a chat thread for the mentally ill instead of
> a Thunderbird feature request, so I'm closing it because I'm tired of it
> polluting my inbox.

There is an option at the top of the page for "never email me about this bug."

(In reply to Ben Bucksch (:BenB) from comment #41)
> Can we fix this in some other way, e.g. helping them to enable proper IMAP
> with normal auth?

In the short run, yes. In the long run, no. Google is fairly desperate and wants all IMAP clients to use OAuth. At this point, are options boil down to:
1. Do nothing, and watch our userbase suffer from poorer usability as Google (and perhaps other large webmail providers) switch to mandating OAuth.
2. Prepare a better alternative to OAuth that we're willing to implement and that email providers are willing to implement. I don't think there's enough manpower to make this feasible.
3. To try to convince Google and other email providers to implement improvements to OAuth that make it more palatable to improve it. I've been doing feeble attempts at pushing this through, and (if I understand these emails properly) Google has committed to some improvements (in part, I think, due to some of my nagging); I don't know the timescales on these improvements.
4. Suck it up and implement it as-is.

Ultimately, it's a rock and a hard place. We have enough market share that we can make people listen to some of our concerns, but we don't have enough to stop people from rolling out making it mandatory. Personally, I prefer option #3, but we do need to have a compelling story by TB 38 and it's unfortunately looking like the needed changes won't make it in place by then. I'll have to see if I can get further details from Brandon, though.

I am really not a fan of OAuth, but, unfortunately, the reality is we don't have much of a say in the matter.

Comment 45

3 years ago
Given

"I am really not a fan of OAuth, but, unfortunately, the reality is we don't have much of a say in the matter"

I'm for:

"4. Suck it up and implement it as-is."
(Assignee)

Updated

3 years ago
Mentor: Pidgeot18@gmail.com

Comment 46

3 years ago
As I understand it, Lightning has had to implement Oauth2.0 to get webdav connections with Google working,  so not only should we suck it up,  we should leverage Fallens work and copy liberally.

Comment 47

3 years ago
Yeah, even though I don't like it either for the reasons stated throughout this bug and elsewhere, given that it's implemented for Calendar already, it may come at low cost to port it for MailNews too.

Updated

3 years ago
See Also: → bug 1096894

Comment 48

3 years ago
(In reply to Joshua Cranmer [:jcranmer] from comment #44)


> 2. Prepare a better alternative to OAuth that we're willing to implement and
> that email providers are willing to implement. I don't think there's enough
> manpower to make this feasible.
> 3. To try to convince Google and other email providers to implement
> improvements to OAuth that make it more palatable to improve it. I've been
> doing feeble attempts at pushing this through, and (if I understand these
> emails properly) Google has committed to some improvements (in part, I
> think, due to some of my nagging); I don't know the timescales on these
> improvements.

The protocols are already secure, there is nothing to fix there.

I agree with BenB this is not a security issue, this is a power issue of Embrace, extend and extinguish disguised as a security issue.

The issue is that Google wants to make itself from a powerful position into an even more powerful position, via using it's current power to attempt to force everyone into going their way, which will give them even more power.

> that email providers are willing to implement. I don't think there's enough
> manpower to make this feasible.

If you give in and let Google have even more power, than they will push even more and then there will be even less manpower.

The thing is, there are more than just 4 things to do.

5. create a custom handler in Thunderbird for gmail, which informs the user that Google is trying to "embrace, extend and extinguish", why there is nothing wrong with current protocols, why and how gmail endangers freedom on the web, why oauth is inherently insecure anyway, and why and how gmail endangers freedom on the web once again. 

6. go public

7. Gently propose email provider alternatives

8. go public


about html:
http://langsec.org/
http://karateairbus.com/2013/10/no-more-turing-complete-input-languages/

Comment 49

3 years ago
anaxagramma@gmail.com: I am unsure what you are proposing.

I think what you are saying is "suck it up and implement it" but then add language to make it clear that we don't like Google's approach even though we support it. Is that correct?

Comment 50

3 years ago
(In reply to Kent James (:rkent) from comment #49)
> anaxagramma@gmail.com: I am unsure what you are proposing.
> 
> I think what you are saying is "suck it up and implement it" but then add
> language to make it clear that we don't like Google's approach even though
> we support it. Is that correct?

sorry for the mess! incorrect.
I am personally against implementing oauth in thunderbird. I agree with BenB and I think BenB outlined the situation perfectly.

Right now, if a Thunderbird user gets an error due to no oauth support, [s]he won't see the wider picture and may just press for oauth support due to the lack of information.

However should a user be informed that Google is playing Embrace, extend and extinguish, a user could focus her/his nagging effors against Google, more specifically against its power-play.

Thus even if Thunderbird development resists oauth, the options are *not* just "do nothing" or trying to persuade google... A better option to inform the users about the dangers they have been putten into: trying to force oauth into email would make the internet less free and less private, while those who prefer Thunderbird to default webmail include those doing it for getting some choice and control back compared to webmail. 

Thus this is not about Thunderbird, this is about google endangering the freedom of its users and injecting a tecnhology, marketed as "security", but meanwhile implementing anti-user-privacy and anti-user-control infrastructure, into the patterns of email operation.

Overall, what I propose for Thunderbird is to *resist* oauth but fight back by briefly informing the users, alongside the technical error, about google's Embrace, extend and extinguish attempt. The more detailed explanation could optionally be put into a separate resource that is referenced.


On a philosophical note, it occured to me that forcing oauth may be an event that tries to troll and ultimately kill a healthy and natural culture. In my understanding after the natural culture of tinkering with microcomputers followed an attempt to closed-source the business, and the concept of free-open-source software was born as a response. After public inhibition of the internet, commercial-driven attempts to monopolize access have been made, and the concept of net-neutrality was born as a response. The current situation a bit reminds me of these events, where a power player attempts to raise the level of entry to a level which strengthens its status quo and monopoly both directly and indirectly.
Comment hidden (advocacy)
Comment hidden (advocacy)

Updated

3 years ago
Flags: needinfo?(Rochsilverado)
Attachment #8541497 - Flags: ui-review+
Attachment #8541497 - Flags: superreview?
Attachment #8541497 - Flags: sec-approval?
Attachment #8541497 - Flags: review+
Attachment #8541497 - Flags: feedback+
Attachment #8541497 - Flags: approval-mozilla-beta?
Attachment #8541497 - Flags: approval-mozilla-b2g34?
Attachment #8541497 - Flags: approval-mozilla-aurora?

Updated

3 years ago
Attachment #8541497 - Attachment is obsolete: true
Attachment #8541497 - Flags: ui-review+
Attachment #8541497 - Flags: superreview?
Attachment #8541497 - Flags: sec-approval?
Attachment #8541497 - Flags: review+
Attachment #8541497 - Flags: feedback+
Attachment #8541497 - Flags: approval-mozilla-beta?
Attachment #8541497 - Flags: approval-mozilla-b2g34?
Attachment #8541497 - Flags: approval-mozilla-aurora?

Updated

3 years ago
Flags: needinfo?(Rochsilverado)

Updated

3 years ago
Flags: needinfo?(nobody)
(Assignee)

Updated

3 years ago
Flags: needinfo?(nobody)

Comment 54

2 years ago
just an errata message, tiny but somehow important:

> like non-https sites are clearly visually different all the time from http ones 

should have been:

like non-https sites are clearly visually different all the time from httpS ones

Comment 55

2 years ago
So what I understand is, authorizing via OAuth and authenticating via an application-specific password offer the same level of security. Google can ask for two-factor authentication during OAuth token creation, or it can ask for two-factor authentication during application-specific password creation, and both are always encrypted by TLS.

The only advantage to OAuth seems to be that it is less of a hassle for the user because an application like Thunderbird would know that it has to prompt the user to create an application-specific authorization token, and it would know how to obtain that token automatically after it is created. (As opposed to the user having to search to figure out how to set up an application-specific password, then manually entering that new password into Thunderbird.)

So my concerns revolve around comment 10:

1. Is there a standard way for Thunderbird to know how the mail provider expects authorization (by web page, phone, text, carrier pidgin, etc.)? How do we communicate this to the user?

2. Is there a standard way for Thunderbird to obtain the token after authorization?

If the answers to these two questions are "yes", then I support implementing OAuth. Otherwise, I'd rather add a "show details" button to view the IMAP error message which will hopefully tell the user what to do next (see comment 20).

(By the way, I found this bug report after Thunderbird said that I had entered my Gmail password incorrectly.)
Comment hidden (advocacy)
Comment hidden (advocacy)
(Assignee)

Updated

2 years ago
tracking-thunderbird38: --- → ?

Updated

2 years ago
tracking-thunderbird38: ? → +
(Assignee)

Comment 58

2 years ago
From a standardization point of view, here is some information that may help guide reviewers to think about the overall picture and path forward.

Google's XOAUTH2 implementation is copied from an old version of <https://tools.ietf.org/html/draft-ietf-kitten-sasl-oauth>, I want to say around -03, but I'm not going to hunt down diffs to find out. In the intervening time, the spec has changed in ways that affect the long-term behavior for supporting OAUTHBEARER; the related mailing list archives actually directly reference this bug. If I understand the IETF process here correctly, this draft and related drafts are effectively in the "ready to publish" phase (~= CR/PR for W3C), and I suspect they will be RFCs within the year, and possibly with implementations as well. Which means Google's timing here sucks for us.

To spare you the reading of draft RFCs, here's a summary:
On authentication failure, the server sends a JSON object indicating the scope to use and a link to a JSON document that includes the list of endpoints to use--notably the intent is to also include dynamic client registration of that draft. Dynamic client registration is included as a "SHOULD"-level requirement, which in 2119 terms means "unless you have a compelling reason not to."

Thus, I think the long-term path is to via oauth login as requiring a combination of issuer (which is what the client id/secret are tied to) and requisite endpoints. This information would be collected at account configuration time, and need to be stored somewhere [my implementation uses prefs right now; a JSON file in the profile is probably a better solution when client id/secret need to be stored]. Autoconfiguration would need to know the SASL steps to lookup--and Google already provides the OpenID.Connect page for us: <https://accounts.google.com/.well-known/openid-configuration>, so I intend to modify the autoconfig XML fileformat to add the OpenID-configuration page and scope parameter as indication of the OAuth2 support.

Something I'm not planning on doing is providing UI to change oauth2 configuration details post-configuration--nothing more than "use OAuth2" versus encrypted password versus normal password.

Comment 59

2 years ago
Just signed in to add some additional experience as an end user, as well as a suggestion for consideration.

I'm now successfully using Thunderbird 31.4 with Google two-factor auth enabled. However initially accessing my Gmail account with a one time password lead to repeated demands for credentials (as comment 57). What I eventually realised was that the "one-time password" needed to be used in two places: once when reading mail (via imap.googlemail.com) and again for sending (via smtp.googlemail.com).

Until the one-time password is used in this way then it will flip-flop on the password challenge as you move from sending to reading the mail account. I've been using Thunderbird seamlessly since realising this, though it wasn't obvious in any way that this was needed (goes to usability and customer experience).

I had contemplated looking at another mail client as I like to have a unified mail folder across all my mail accounts (on multiple domains). Not long embarked down that path though, and I'm back in the Thunderbird fold again now it is solved.


On to the suggestion. With the concern that the Google implementation of Oauth 2.0 being tailored to Google (see comment 10, comment 45 and comment 51, for example) then the best place for this would be a plugin?

"Core" Thunderbird would not pick up the Google specific implementation details, and the relationship with a plugin would establish any shortcomings from standardizations maintained at arms length. The end-user usage can be monitored by download and trends. The users get a simple and convenient way to access their Google mail as well by having a choice to download.

Apologies, I'm not familiar with Thunderbird API's or extensibility (though I have seen an Outlook plugin) so please colour me surprised if the mail application can't extend on sending/receiving email in this way.

The long term view would then be to integrate once an acceptable level of standardisation and implementation is available.

In the meantime any advocacy that needs to take place can happen through targeted channels with formal and informal positions adopted by the Mozilla Foundation. This specifically prevents making casualties of Thunderbird's users to make a point, and avoids invoking the first rule of the fanatic...

Thanks for reading, and hope it helps in some small way.
(Assignee)

Comment 60

2 years ago
Created attachment 8565317 [details]
Prototype of OAuth2 autoconfig file

Here is a prototype of what I'm planning on using for the autoconfiguration. It's inspired by what Gaia uses, but some of the details are different:

1. Auth method is OAuth2, not xoauth2. The intent is to implement what will eventually be SASL OAUTHBEARER--it's easy enough to choose between OAUTHBEARER or XOAUTH2 at login time via regular SASL fallback.
2. I've replaced the endpoint URLs with the OpenID document link. This is roughly equivalent to what we would get back via OAUTHBEARER message, so it feels natural to include that in the list.
3. I dropped the oauth2Settings out of emailProvider, so that it's at the same level as webMail or something like that. Having coded it up, it's slightly easier to put it in emailProvider, so I'm flexible to moving it.

The implication here is that the oauth2 tokens will be the same for both IMAP and SMTP--which, given that we don't have much implementation experience, isn't necessarily something that can be relied on. I doubt authentication would differ, but since scope is vaguely alluded to what a client can do, it's feasible to imagine separating "client can read and trash your emails" from "client can send emails on your behalf." On the other hand, it's much easier to handle the case where they use the same scope, particularly when saving to a password manager.
Assignee: nobody → Pidgeot18
Status: NEW → ASSIGNED
Attachment #8565317 - Flags: feedback?(bugmail)
Attachment #8565317 - Flags: feedback?(ben.bucksch)
Comment on attachment 8565317 [details]
Prototype of OAuth2 autoconfig file

(In reply to Joshua Cranmer [:jcranmer] from comment #60)
> 1. Auth method is OAuth2, not xoauth2. The intent is to implement what will
> eventually be SASL OAUTHBEARER--it's easy enough to choose between
> OAUTHBEARER or XOAUTH2 at login time via regular SASL fallback.

Very reasonable, and handy for gaia email since we can distinguish between our old hack and a newer more standardsy solution.

> 2. I've replaced the endpoint URLs with the OpenID document link. This is
> roughly equivalent to what we would get back via OAUTHBEARER message, so it
> feels natural to include that in the list.

Definitely the right way to go.

> 3. I dropped the oauth2Settings out of emailProvider, so that it's at the
> same level as webMail or something like that. Having coded it up, it's
> slightly easier to put it in emailProvider, so I'm flexible to moving it.

I don't have a strong opinion.  Since emailProvider does have some semantics related to ordering, I could see a stronger argument being made for keeping it outside emailProvider so there's less confusion about whether multiple instances are allowed, etc.  :BenB probably has the most vision for the XML format though.

> The implication here is that the oauth2 tokens will be the same for both
> IMAP and SMTP--which, given that we don't have much implementation
> experience, isn't necessarily something that can be relied on.

It seems like a reasonable simplification to just include all the same scopes required for IMAP/SMTP usage in a single (space-delimited) list and associated with a single token.  In the event that more complicated scope stuff needs to happen in the future, we'd probably want to add elements or attributes that captured a mapping between our semantics and the appropriate server scope.  And we can address that then.
Attachment #8565317 - Flags: feedback?(bugmail) → feedback+
(Assignee)

Comment 62

2 years ago
Created attachment 8565826 [details]
MozReview Request: bz://849540/jcranmer

/r/3967 - Bug 849540, part 1: Add an OAuth2 authentication type
/r/3969 - Bug 849540, part 2: Implement an OAuth2 SASL module helper.
/r/3971 - Bug 849540, part 3: Implement XOAUTH2 support for IMAP.
/r/3973 - Bug 849540, part 4: Implement XOAUTH2 for SMTP.
/r/3975 - Bug 849540, part 5: Implement support for OAuth2 in the account config section.

Pull down these commits:

hg pull review -r 59daf3bdb3649323d31e3294b3a1fac020697720
Attachment #8565826 - Flags: review?(rkent)
Attachment #8565826 - Flags: review?(ben.bucksch)
(Assignee)

Comment 63

2 years ago
Note: to test autoconfiguration for gmail, save the file in attachment 8565317 [details] to $objdir/dist/bin/isp/gmail.com.xml

Comment 64

2 years ago
When I revoked access to a test account in Google, and restarted, the spinner runs with "testrkent@gmail.com: Sending login information.." and I see in the error console:

Timestamp: 2/18/2015 2:04:02 PM
Error: NS_ERROR_XPC_JAVASCRIPT_ERROR_WITH_DETAILS: [JavaScript Error: "propEnum is undefined" {file: "resource://gre/modules/LoginHelper.jsm" line: 165}]'[JavaScript Error: "propEnum is undefined" {file: "resource://gre/modules/LoginHelper.jsm" line: 165}]' when calling method: [nsILoginManagerStorage::modifyLogin]
Source File: file:///C:/tb/3-central/tb-debug/dist/bin/components/nsLoginManager.js
Line: 355

Updated

2 years ago
Attachment #8565826 - Flags: review?(rkent)

Comment 65

2 years ago
Comment on attachment 8565826 [details]
MozReview Request: bz://849540/jcranmer

https://reviewboard.mozilla.org/r/3965/#review3173

I tested it all and it works initially. But I never managed to recover from removing authorization for the login from the gmail account, so now it always fails.

Comment 66

2 years ago
https://reviewboard.mozilla.org/r/3967/#review3157

::: mailnews/base/public/MailNewsTypes2.idl
(Diff revision 1)
> +    /// Use OAuth2 to authenticate

Nit: period at end of comment.

Comment 67

2 years ago
https://reviewboard.mozilla.org/r/3969/#review3159

::: mailnews/base/public/msgIOAuth2Module.idl
(Diff revision 1)
> +  bool initFromMail(in nsIMsgIncomingServer aSmtpServer);

Nit: aServer not aSmtpServer

::: mailnews/base/src/msgOAuth2Module.js
(Diff revision 1)
> +      this._appSecret = 'YpeuM0eYPQe_r98HZ7p16zUm';

I think that you need some editorial comments here, similar to what Lightning uses, that mentions that you know this is sensitive, and specifically requests that people do not abuse it. Otherwise you are going to get reports from people of their discovery of a security disclosure in Thunderbird that you will have to argue about after the fact.

See http://mxr.mozilla.org/comm-central/source/calendar/providers/gdata/modules/gdataSession.jsm#500

::: mailnews/base/src/msgOAuth2Module.js
(Diff revision 1)
> +    this._loginUrl = "oauth:" + aUsername + "@" + issuer + ";" + scope;

The format to use for the URL string in the password manager should be reasonable, and also consistent with other items that are there. Yours is quite inconsistent with what else is there.

Here's what I would suggest. Simply use the actual https: URL of the google API used to get the token, use the actual username, and add a realm like "Google Email OAuth2 Token".

::: mailnews/base/src/msgOAuth2Module.js
(Diff revision 1)
> +    let logins = loginMgr.findLogins({}, this._loginUrl, null, this._loginUrl);

That you need here is to get a list of logins for a particular host and realm, and then search that list for a login for the username that you are looking for. I really think that we should use a consistent URL rather than the adhoc pattern that you used.

::: mail/locales/en-US/chrome/messenger/messenger.properties
(Diff revision 1)
> +oauth2WindowTitle=Enter credentials for %1$S

Please see my bug 1110881  That is, you should not be putting up a request for credentials without some indication to the user of why this is being asked for.

Comment 68

2 years ago
https://reviewboard.mozilla.org/r/3971/#review3163

::: mailnews/imap/src/nsImapProtocol.cpp
(Diff revision 1)
> +    mOAuth2Support = new mozilla::mailnews::OAuth2ThreadHelper(server);

Can't you just instantiate this when needed, like after InitPrefAuthMethods (or even as part of that method, in case nsMsgAuthMethod::OAuth2:? You should not have to add the extra memory and time for all of the other protocols that don't need this.

::: mailnews/imap/src/nsImapProtocol.cpp
(Diff revision 1)
> +    

Nit: whitespace

::: mailnews/imap/src/nsImapProtocol.cpp
(Diff revision 1)
> +      return NS_ERROR_UNEXPECTED;

Although you can use these four lines for error checking, if it was me I would just replace this with a single line:

NS_ENSURE_STATE(mOAuth2Support);

::: mailnews/imap/src/nsImapProtocol.cpp
(Diff revision 1)
> +    IncrementCommandTagNumber();

I don't believe you need this. This is called at the top of this method, and you really have not sent anything yet.

::: mailnews/imap/src/nsImapProtocol.cpp
(Diff revision 1)
> +    command += " AUTHENTICATE XOAUTH2 ";

Is there any reason that you capitalized AUTHENTICATE? All of the other commands had it in lower case. I suspect it does not matter, so you should probably be consistent with the other code.

Comment 69

2 years ago
https://reviewboard.mozilla.org/r/3973/#review3165

::: mailnews/compose/src/nsSmtpProtocol.h
(Diff revision 1)
> +#define SMTP_AUTH_OAUTH2_ENABLED        0x00040000

These are not persisted, right, so you are free to choose them as you wish? Then move SMTP_AUTH and SMTP_AUTH_NONE_ENABLED up one hex position, so that SMTP_AUTH_OAUTH2_ENABLED is 0x0010000 and naturally grouped with the prior flags.

::: mailnews/compose/src/nsSmtpProtocol.h
(Diff revision 1)
> +    nsCOMPtr<msgIOAuth2Module> mOAuth2Support;

You are overloading the meaning of this to be both a method used to detect oauth2 support, as well as a flag as to whether it is supported. You need to document here that you are also using its presence as a flag that oauth2 is supported.

::: mailnews/compose/src/nsSmtpProtocol.cpp
(Diff revision 1)
> +            mOAuth2Support = nullptr;

This is where you overload the meaning.

::: mailnews/compose/src/nsSmtpProtocol.cpp
(Diff revision 1)
> +  if (NS_FAILED(rv))

It's a little confusing having this if (NS_FAILED(rv)) several lines after the SendData. Can't you put it right after, and if ... else to deal with the state?

Comment 70

2 years ago
https://reviewboard.mozilla.org/r/3975/#review3171

I would expect that in Manual setup, you would have the choice of selecting OAuth at least for cases where we know it exists from the isp file. But there is no way to enable it in manual setup. If I wanted OAuth I would probably go to the Manual Config screen. When I do, and do Re-test, it shows "Normal Password" for authentication. I think that the Mail Account Setup should have some way of showing the user that OAuth2 is going to be used other than just accepting "Done" and hoping for the best.

Also, doesn't OAuth2 assume that you will be using SSL? Should you have some checks for that?
(Assignee)

Comment 71

2 years ago
https://reviewboard.mozilla.org/r/3971/#review3175

> Is there any reason that you capitalized AUTHENTICATE? All of the other commands had it in lower case. I suspect it does not matter, so you should probably be consistent with the other code.

The protocol command names are case-insensitive; usual convention is to describe them in upper-case.
(Assignee)

Comment 72

2 years ago
https://reviewboard.mozilla.org/r/3969/#review3177

> The format to use for the URL string in the password manager should be reasonable, and also consistent with other items that are there. Yours is quite inconsistent with what else is there.
> 
> Here's what I would suggest. Simply use the actual https: URL of the google API used to get the token, use the actual username, and add a realm like "Google Email OAuth2 Token".

A refresh token requires the scope--the same issuer but different scopes require different tokens, and hence different entries in the password manager. I used the issuer rather than the auth or token URI because that is less likely to change (indeed, the token URI generated in this patch is actually a different one from the one I originally used when doing initial testing).

Most of this was half-cargo-culted from the NNTP login process, since the nsILoginManager documentation actually rather sucks at describing how to use it for non-HTTP authentication.

Comment 73

2 years ago
(In reply to Joshua Cranmer [:jcranmer] from comment #72)
> 
> A refresh token requires the scope--the same issuer but different scopes
> require different tokens, and hence different entries in the password
> manager.

Right. What I am proposing is that the uri contain the actual uri, the username the actual username, and anything else required to make it unique (such as scope, or the access source (like Thunderbird versus Lightning)) be put in the realm.

> I used the issuer rather than the auth or token URI because that is
> less likely to change (indeed, the token URI generated in this patch is
> actually a different one from the one I originally used when doing initia
> testing).

Isn't the uri a well-defined location like https://accounts.google.com/o/oauth2/auth?
(In reply to Kent James (:rkent) from comment #67)
> See
> http://mxr.mozilla.org/comm-central/source/calendar/providers/gdata/modules/
> gdataSession.jsm#500

On the last line of that comment there's a bad tpyo... it says "its" instead of "it's"
(Assignee)

Comment 75

2 years ago
https://reviewboard.mozilla.org/r/3975/#review3207

The way the manual setup works compared to the ISP stuff is, as I recall, squirrelly. The XOAUTH2 response from google right now is insufficient to provide the necessary details, so I can't rely on a "proper" verifyConfig/verifyLogon path to guesstimate those details (nor do I want to try to do all the glue work to make it available from the IMAP/SMTP login code).

The requirements for OAuth2 SSL (or lack thereof, for the most part) are in the https:// token endpoint negotiation, not the SASL mechanism.
(Assignee)

Comment 76

2 years ago
https://reviewboard.mozilla.org/r/3971/#review3299

> Although you can use these four lines for error checking, if it was me I would just replace this with a single line:
> 
> NS_ENSURE_STATE(mOAuth2Support);

I prefer the assertion since it creates a situation that would be fatal in tests.
(Assignee)

Comment 77

2 years ago
Comment on attachment 8565826 [details]
MozReview Request: bz://849540/jcranmer

/r/3967 - Bug 849540, part 1: Add an OAuth2 authentication type
/r/3969 - Bug 849540, part 2: Implement an OAuth2 SASL module helper.
/r/3971 - Bug 849540, part 3: Implement XOAUTH2 support for IMAP.
/r/3973 - Bug 849540, part 4: Implement XOAUTH2 for SMTP.
/r/3975 - Bug 849540, part 5: Implement support for OAuth2 in the account config section.

Pull down these commits:

hg pull review -r eda47e4681e8f83cec3152a03f83c97453eb0771
Attachment #8565826 - Flags: review?(rkent)

Comment 78

2 years ago
https://reviewboard.mozilla.org/r/3967/#review3301

Ship It!

Comment 79

2 years ago
https://reviewboard.mozilla.org/r/3967/#review3303

Looks good to me now. Missed the "Ship it" checkmark before.

Comment 80

2 years ago
https://reviewboard.mozilla.org/r/3969/#review3305

LGTM
(Assignee)

Comment 81

2 years ago
Pushing parts 1 and 2 for the string freeze:
https://hg.mozilla.org/comm-central/rev/47051fbe6edd
https://hg.mozilla.org/comm-central/rev/bdc11668c7ab

Comment 82

2 years ago
Comment on attachment 8565826 [details]
MozReview Request: bz://849540/jcranmer

https://reviewboard.mozilla.org/r/3965/#review4553

::: mailnews/base/prefs/content/accountcreation/readFromXML.js
(Diff revision 2)
> +          if (oO.auth == Ci.nsMsgAuthMethod.OAuth2 && !oauthSettings)

Maybe we should consider this an error for which we should consider the config faulty (set "exception = new ..."), rather than something where we should ignore the section (just continue with next server).

::: mailnews/base/prefs/content/accountcreation/verifyConfig.js
(Diff revision 2)
> -    errorCallback(e);
> +      errorCallback(e);

Thr throw must stay. It's important for how the function works, so that execution stops at that point.

::: mailnews/base/prefs/content/accountcreation/verifyConfig.js
(Diff revision 2)
> +  let result = configLoaded.promise.then(function (result) {

Code style:
* Remove |configUrl| local var, it's not useful
* When breaking long lines, indent the following lines at least 4 spaces, not just 2.

::: mailnews/base/prefs/content/accountcreation/readFromXML.js
(Diff revision 2)
> +    oauthSettings.scope = clientConfigXML.clientConfig.oauth2Settings.scope;

Any values you take from the network and copy into internal variables MUST be security-checked to have a format and values that you expect, using sanitize.foo(). Imagine e.g. a javascript: URL here (just as an example).

Instead, do e.g.:
oauthSettings.configUrl = sanitize.url(clientConfigXML.clientConfig.oauth2Settings.openidConfig);
oauthSettings.scope = sanitize.hostname(clientConfigXML.clientConfig.oauth2Settings.openidScope);


Also, the XML format uses camelCase instead of dashes. As a side-benefit, that allows you to use dots here.

Please note that the XML MUST be defined on
https://wiki.mozilla.org/Thunderbird:Autoconfiguration:ConfigFileFormat
in the style of the existing document, with a realistic example, and approved by me and preferably also documented on
https://developer.mozilla.org/en-US/docs/Mozilla/Thunderbird/Autoconfiguration

For the format, I'm a little confused why we have OpenID config *inside* the OAuth XML section here. Logically, they would be siblings. I'd need to understand more about the protocol to see whether that makes sense or not.

::: mailnews/base/prefs/content/accountcreation/verifyConfig.js
(Diff revision 2)
> +    oauthConfig.tokenEndpoint = config.token_endpoint;

Same as above: Any values you take from the network MUST be security-checked to have values that you expect, using sanitize.foo(). Check sanitize*.js which functions you have and are most applicable. Here, it's probably sanitize.hostname() or sanitize.url() again.

For issues, if it's something like "Google. Inc", use sanitize.label() (not sanitize.string()).

::: mailnews/base/prefs/content/accountcreation/verifyConfig.js
(Diff revision 2)
> +        "chrome://messenger/locale/accountCreationModel.properties")

Please use a specific error message here.

::: mailnews/base/src/msgOAuth2Module.js
(Diff revision 2)
> +      // provide dynamic client registration. Don't copy these values for your

We're a grown-up client application here, not a website. This is simply not applicable. Tell Google to come up with something sane.

In fact, they don't even need to: We already have standing standards (called SASL, working in IMAP and SMTP) with challenge-response authentication mechnisms that can do SMS code login. They just need to use them. No need for OAuth or "secret app keys" at all. Greetings from me :).

The fact that we have nonsense like that is for me a complete non-starter for OAuth in Thunderbird. If it was my decision, I would not allow this into Thunderbird.

::: mailnews/compose/src/nsSmtpProtocol.cpp
(Diff revision 2)
> +        // Query for OAuth2 support. If the SMTP server preferences don't allow

s/allow/use/

::: mailnews/base/public/msgIOAuth2Module.idl
(Diff revision 2)
> +   * Initialize the OAuth2 parameters from an SMTP server, and return whether or

Please describe the interaction a bit more here. It's not clear that "init*" refers to the init of the module, not of the server or connection. Please also write that you expect exactly one of the two functions being called. Also, describe briefly what the functions do, e.g. "Reads the OAuth settings for this server"

::: mailnews/base/public/msgIOAuth2Module.idl
(Diff revision 2)
> +  void connect(in boolean aWithUI, in msgIOAuth2ModuleListener aCallback);

it's unclear at which steps you would call connect() vs. build*()

::: mailnews/base/public/msgIOAuth2Module.idl
(Diff revision 2)
> +  ACString buildXOAuth2String();

maybe call this buildSASLResponse() ? "OAuth" is implied by the module, and "string" is implied by the return type.

::: mailnews/base/public/msgIOAuth2Module.idl
(Diff revision 2)
> + * A listener callback for OAuth2 SASL authentication. This would be represented

Write that the implementation is based on the OAuth module in [where] and just adapts this for email and SASL.

::: mailnews/base/src/msgOAuth2Module.js
(Diff revision 2)
> +

Rest of file not reviewed

::: mailnews/compose/src/nsSmtpProtocol.cpp
(Diff revision 2)
> +    case nsMsgAuthMethod::OAuth2:

Move this up, after NTLM

::: mailnews/compose/src/nsSmtpProtocol.cpp
(Diff revision 2)
> +          SMTP_AUTH_OAUTH2_ENABLED |

Don't add this to "secure". It's a legacy setting.

::: mailnews/compose/src/nsSmtpProtocol.cpp
(Diff revision 2)
> +  else if (SMTP_AUTH_OAUTH2_ENABLED & availCaps)

I don't we should prefer OAuth over passwords. If we are to use OAuth, it should be explicitly configured.

Also, you need to have OAuth configured to work. Imagine what happens for a server that reports XOAUTH, but we don't have an OAuth config for it.

::: mailnews/compose/src/nsSmtpProtocol.cpp
(Diff revision 2)
> +nsresult nsSmtpProtocol::OnSuccess(const nsACString &aAccessToken)

Rename this function (and OnFailure()). It's specific to OAuth, and should have OAuth in its name

::: mailnews/compose/src/nsSmtpProtocol.cpp
(Diff revision 2)
> +nsresult nsSmtpProtocol::OnSuccess(const nsACString &aAccessToken)

Rename this function (and OnFailure()). It's specific to OAuth, and should have OAuth in its name

Most likely, OnSuccess() should be called AuthOAuth2Step2()

::: mailnews/compose/src/nsSmtpProtocol.h
(Diff revision 2)
> +SMTP_SUSPENDED,                                     // 25

SUSPENDED should be called OAUTH_STEP1 or similar. It cannot have a generic name.

::: mailnews/compose/src/nsSmtpProtocol.cpp
(Diff revision 2)
> +  else if (SMTP_AUTH_OAUTH2_ENABLED & availCaps)

Not sure whether this should be before passwords. Consider a server that we don't support, but it reports XOAUTH2 .

::: mailnews/compose/src/nsSmtpProtocol.cpp
(Diff revision 2)
> +      // This state means we're going into an async loop and waiting for

As said above, don't make this generic. It makes the code unreadable, esp. given that this is a state machine. Make a specific state for OAUTH_STEP_1 and OAUTH_STEP_2.

::: mailnews/imap/src/nsImapProtocol.cpp
(Diff revision 2)
> +    // Disable OAuth2 support if we don't have the prefs installed.

Good idea. Do the same for SMTP, please, including removing the server capability.

::: mailnews/imap/src/nsImapProtocol.cpp
(Diff revision 2)
> +    NS_ASSERTION(mOAuth2Support,

Funny message, but better would be: "Lack of oauthmodule should have been caught in [code place]"

::: mailnews/imap/src/nsImapProtocol.cpp
(Diff revision 2)
> +    mOAuth2Support = nullptr; // Its purpose has been served.

IMAP makes several connections with several threads. Are you sure that it will be a new object of this type? If so, write that or better yet ensure it in the right place.

::: mailnews/imap/src/nsSyncRunnableHelpers.cpp
(Diff revision 2)
> +

I didn't review this part
Attachment #8565826 - Flags: review?(ben.bucksch)

Comment 83

2 years ago
Comment on attachment 8565826 [details]
MozReview Request: bz://849540/jcranmer

Sorry for the delay in the review. Also sorry for the bad code quotes. It was my first review using reviewboard. Typically, the quotes lines are above the code I comment on.
Attachment #8565826 - Flags: review?(rkent) → review-
(Assignee)

Comment 84

2 years ago
https://reviewboard.mozilla.org/r/3965/#review4563

> Thr throw must stay. It's important for how the function works, so that execution stops at that point.

Um. This is an asynchronous function body in a promise. Throwing won't stop the function.

> Same as above: Any values you take from the network MUST be security-checked to have values that you expect, using sanitize.foo(). Check sanitize*.js which functions you have and are most applicable. Here, it's probably sanitize.hostname() or sanitize.url() again.
> 
> For issues, if it's something like "Google. Inc", use sanitize.label() (not sanitize.string()).

Issuer is a purely internal string and is never displayed to the user.

> Any values you take from the network and copy into internal variables MUST be security-checked to have a format and values that you expect, using sanitize.foo(). Imagine e.g. a javascript: URL here (just as an example).
> 
> Instead, do e.g.:
> oauthSettings.configUrl = sanitize.url(clientConfigXML.clientConfig.oauth2Settings.openidConfig);
> oauthSettings.scope = sanitize.hostname(clientConfigXML.clientConfig.oauth2Settings.openidScope);
> 
> 
> Also, the XML format uses camelCase instead of dashes. As a side-benefit, that allows you to use dots here.
> 
> Please note that the XML MUST be defined on
> https://wiki.mozilla.org/Thunderbird:Autoconfiguration:ConfigFileFormat
> in the style of the existing document, with a realistic example, and approved by me and preferably also documented on
> https://developer.mozilla.org/en-US/docs/Mozilla/Thunderbird/Autoconfiguration
> 
> For the format, I'm a little confused why we have OpenID config *inside* the OAuth XML section here. Logically, they would be siblings. I'd need to understand more about the protocol to see whether that makes sense or not.

Yes, I intend to change the documentation when the config file format is finalized (doing it before would be putting the cart before the horse).

The OAuth2 SASL stuff is meant to be keyed off of a {configuration issuer, scope} tuple, where the configuration issuer in this case is defined by the openid configuration URL. (In theory, this is what you receive in an error message on a SASL OAUTHBEARER failure. The current XOAUTH2 implementation is based on an old draft which did not include the openid config URL in the error response.)

> We're a grown-up client application here, not a website. This is simply not applicable. Tell Google to come up with something sane.
> 
> In fact, they don't even need to: We already have standing standards (called SASL, working in IMAP and SMTP) with challenge-response authentication mechnisms that can do SMS code login. They just need to use them. No need for OAuth or "secret app keys" at all. Greetings from me :).
> 
> The fact that we have nonsense like that is for me a complete non-starter for OAuth in Thunderbird. If it was my decision, I would not allow this into Thunderbird.

This app secret is not secret, but it is mandated by the protocol (even the Google OAuth information pages admit that it's sometimes not secret). I refuse to do any obfuscation here.
(Assignee)

Comment 85

2 years ago
https://reviewboard.mozilla.org/r/3965/#review4613

> maybe call this buildSASLResponse() ? "OAuth" is implied by the module, and "string" is implied by the return type.

This is meant to suggest that we're building the response to SASL XOAUTH2, not SASL OAUTHBEARER.

Comment 86

2 years ago
> Um. This is an asynchronous function body in a promise. Throwing won't stop the function.

Right. But it should, that was my point. It did before (I think), and the logic of the function shouldn't change.

> Issuer is a purely internal string and is never displayed to the user.

What's it used for? (This should probably be part of the file format documentation)

sanitize.label() still seems like the right check function.

> app secret ... I refuse to do any obfuscation here.

I'm 100% with you on that.
(Assignee)

Comment 87

2 years ago
https://reviewboard.mozilla.org/r/3965/#review4663

> SUSPENDED should be called OAUTH_STEP1 or similar. It cannot have a generic name.

It is generic--it can be used whenever the SMTP connection needs to be "paused" to allow some asynchronous event to fire back (cf. NNTP's NNTP_SUSPENDED state). Since SMTP doesn't have an async authentication driver yet, the only use is oauth at the moment.

> I don't we should prefer OAuth over passwords. If we are to use OAuth, it should be explicitly configured.
> 
> Also, you need to have OAuth configured to work. Imagine what happens for a server that reports XOAUTH, but we don't have an OAuth config for it.

OAuth2 is only enabled if we have mOAuth2Support set to non-null (we clear the flag out otherwise). So, no, that isn't a problem.

> As said above, don't make this generic. It makes the code unreadable, esp. given that this is a state machine. Make a specific state for OAUTH_STEP_1 and OAUTH_STEP_2.

This code is necessarily async, and so we have to drop out of the state machine. I'm actually surprised that we've gotten this far having absolutely no other async callbacks in the SMTP state machine!
(Assignee)

Comment 88

2 years ago
Comment on attachment 8565826 [details]
MozReview Request: bz://849540/jcranmer

/r/3967 - Bug 849540, part 3: Implement XOAUTH2 support for IMAP.
/r/3969 - Bug 849540, part 4: Implement XOAUTH2 for SMTP.
/r/3971 - Bug 849540, part 5: Implement support for OAuth2 in the account config section.

Pull down these commits:

hg pull review -r b81143c80440ef554d473cb9902b1534fc15b6b2
Attachment #8565826 - Flags: review?(rkent)
Attachment #8565826 - Flags: review?(ben.bucksch)
Attachment #8565826 - Flags: review-

Comment 89

2 years ago
> > SUSPENDED
> It is generic

I don't think it should be. It's a state machine and the execution flow is already too difficult to follow. If you drop into generic states, it gets even more difficult.

Plus, it's logically an OAUTH step: You start with OAUTH2, you continue with OAUTH2. You cannot start with OAUTH2 step 1 and continue with SSL client cert STEP 2. Thus, it's inherently a step for OAUTH, not generic.

Please make it an explicit OAUTH2_STEP1 and OAUTH2_STEP2. We have similar step 1 and step 2 in other auth schemes.

That it's async waiting doesn't change the fact that we need to continue with OAUTH after it.
Implementation suggestion (just an idea): Given that it's a case/if switch, you can easily have some "async" code that's executed for several steps, e.g. state = OAUTH2_STEP2, SSL_CLIENT_CERT_STEP2 and CHALLENGE_RESPONSE_STEP2 all are conditions to run the async waiting code. Once the waiting is over, you drop into OAUTH2_STEP3, or CHALLENGE_RESPONSE_STEP3, depending on what the previous state was.

Comment 90

2 years ago
Comment on attachment 8565826 [details]
MozReview Request: bz://849540/jcranmer

Please ask for review again once all my review comments in comment 82 are fixed. Thanks.
Attachment #8565826 - Flags: review?(ben.bucksch) → review-

Comment 91

2 years ago
> The throw must stay. It's important for how the function works, so that execution stops at that point

Given that the throw was at the end of the function anyway, and the error callback is being called, your change was correct. I retract this comment.

[General rule for all async functions here is:
Throw when possible. Otherwise call errorCallback. In both cases, stop execution after the error occurred. Under no circumstance call successCallback after a throw or errorCallback.
The function either throws, calls errorCallback or successCallback, and *exactly one* of them. (As such, the old function was wrong, actually.) Only exception: In rare circumstances, errorCallback might be called several times. But successCallback will never be called after another successCallback or errorCallback.]

Comment 92

2 years ago
I compiled with the latest patch, but it is not working for me. That is, when I attempt to create a gmail account I see no evidence that OAuth is available, and authentication never succeeds.

Comment 93

2 years ago
I tried again to get this to work, got the compile working (I had not successfully applied the ReviewBoard stuff before) but I still can't make it work. Please ping me when you are free Joshua so we can work through this.
Flags: needinfo?(Pidgeot18)
This has just been about logging into IMAP/SMTP over OAuth, stripping the Google Talk XMPP part out of the title.
Summary: Log in to Google Talk (XMPP) and Gmail (IMAP/SMTP) using OAuth → Log in to Gmail (IMAP/SMTP) using OAuth

Comment 95

2 years ago
I've slogged my way through the autoconfig stuff, and I have found and fixed two errors, but still it does not work. Could you please provide me with a working patch and gmail config file? Am I missing something here?

The errors so far:

1)

+    oauthSettings.configUrl = sanitize.url(
+      clientConfigXML.clientConfig.oauth2Settings.openidConfig);

openidConfig does not exist in the config file, there you have openid-config  One of these needs to be changed.

2) In sanitize.url:

120     try {
121       uri = ioService().newURI(str, null, null);
122       uri = uri.QueryInterface(Ci.nsIURL);
123     } catch (e) {
124       throw new MalformedException("url_parsing.error", unchecked);

ioService() was removed in Bug 824150 so this now silently fails. That prevents the local config file from being loaded.

Updated

2 years ago
Blocks: 824150

Comment 96

2 years ago
I will review this URL:
http://hg.mozilla.org/users/Pidgeot18_gmail.com/patch-queues/raw-file/48a1de5f0782/patches-c-c-new-build/oauth2-autoconfig

+          // If we're using OAuth2, but don't have working settings, bail.
+          if (iO.auth == Ci.nsMsgAuthMethod.OAuth2 && !oauthSettings)
+            continue;
(2x)

consider this an error for which we should consider the config faulty (set "exception = new ..."), rather than something where we should ignore the section (just continue with next server).


Please note that the XML MUST be defined on
https://wiki.mozilla.org/Thunderbird:Autoconfiguration:ConfigFileFormat
in the style of the existing document, with a realistic example, and approved by me and preferably also documented on
https://developer.mozilla.org/en-US/docs/Mozilla/Thunderbird/Autoconfiguration

Please make the changes to these documents and let me review these changes.

+    oauthConfig.issuer = sanitizer.nonemptystring(config.issuer);

For issuer, if it's something like "Google. Inc", use sanitize.label() (not sanitize.string()).

------

The following comments reply to changes that are not in this diff, but should be possible to fix now, even for beta:

::: mailnews/compose/src/nsSmtpProtocol.cpp
(Diff revision 2)
> +        // Query for OAuth2 support. If the SMTP server preferences don't allow

s/allow/use/


::: mailnews/base/public/msgIOAuth2Module.idl
(Diff revision 2)
> +   * Initialize the OAuth2 parameters from an SMTP server, and return whether or

Please describe the interaction a bit more here. It's not clear that "init*" refers to the init of the module, not of the server or connection. Please also write that you expect exactly one of the two functions being called. Also, describe briefly what the functions do, e.g. "Reads the OAuth settings for this server"

::: mailnews/base/public/msgIOAuth2Module.idl
(Diff revision 2)
> + * A listener callback for OAuth2 SASL authentication. This would be represented

Write that the implementation is based on the OAuth module in [where] and just adapts this for email and SASL.

::: mailnews/compose/src/nsSmtpProtocol.cpp
(Diff revision 2)
> +    case nsMsgAuthMethod::OAuth2:

Move this up, after NTLM

::: mailnews/compose/src/nsSmtpProtocol.cpp
(Diff revision 2)
> +          SMTP_AUTH_OAUTH2_ENABLED |

Don't add this to "secure". It's a legacy setting.

::: mailnews/compose/src/nsSmtpProtocol.cpp
(Diff revision 2)
> +  else if (SMTP_AUTH_OAUTH2_ENABLED & availCaps)

I don't we should prefer OAuth over passwords. If we are to use OAuth, it should be explicitly configured.

Also, you need to have OAuth configured to work. Imagine what happens for a server that reports XOAUTH, but we don't have an OAuth config for it.

::: mailnews/compose/src/nsSmtpProtocol.cpp/h

> +nsresult nsSmtpProtocol::OnSuccess(const nsACString &aAccessToken)
> +SMTP_SUSPENDED,                                     // 25
> +      // This state means we're going into an async loop and waiting for

Rename this function (and OnFailure()). It's specific to OAuth, and should have OAuth in its name

Most likely, OnSuccess() should be called AuthOAuth2Step2()

SUSPENDED should be called OAUTH_STEP1 or similar. It cannot have a generic name.

Do NOT make this generic. It makes the code unreadable, esp. given that this is a state machine. Make a specific state for OAUTH_STEP_1 and OAUTH_STEP_2.

I know you disagree, but the whole point of the states of a state machine is to keep track of which state we are in and what the next step will be. The next step after OAUTH_STEP_1 is always OAUTH_STEP_2, and never OAUTH_STEP1, SUSPEND and SSL_CLIENT_CERT_STEP2, even if both methods need to "suspend" and wait for some external action. You may share the code that's executed for the 2 steps, but the *state* is always OAUTH before, OAUTH after. That's the whole idea of states in a state machine, to know where you are and what to do next.

See comment 89, too.

This MUST be fixed.

::: mailnews/compose/src/nsSmtpProtocol.cpp
(Diff revision 2)
> +  else if (SMTP_AUTH_OAUTH2_ENABLED & availCaps)

Not sure whether this should be before passwords. Consider a server that we don't support, but it reports XOAUTH2 .

::: mailnews/imap/src/nsImapProtocol.cpp
(Diff revision 2)
> +    // Disable OAuth2 support if we don't have the prefs installed.

Good idea. Do the same for SMTP, please, including removing the server capability.

::: mailnews/imap/src/nsImapProtocol.cpp
(Diff revision 2)
> +    NS_ASSERTION(mOAuth2Support,

Funny message, but better would be: "Lack of oauthmodule should have been caught in [code place]"

::: mailnews/imap/src/nsImapProtocol.cpp
(Diff revision 2)
> +    mOAuth2Support = nullptr; // Its purpose has been served.

IMAP makes several connections with several threads. Are you sure that it will be a new object of this type? If so, write that or better yet ensure it in the right place.

-------

The following comments reply to changes were already commited and are not in this diff. You say we can't change this now, due to string and API freeze. That's fine, so please make a new patch (now), but to be applied after trunk leaves beta.

::: mailnews/base/prefs/content/accountcreation/verifyConfig.js
(Diff revision 2)
> +        "chrome://messenger/locale/accountCreationModel.properties")

Please use a specific error message here.

::: mailnews/base/public/msgIOAuth2Module.idl
(Diff revision 2)
> +  void connect(in boolean aWithUI, in msgIOAuth2ModuleListener aCallback);

it's unclear at which steps you would call connect() vs. build*()

::: mailnews/base/public/msgIOAuth2Module.idl
(Diff revision 2)
> +  ACString buildXOAuth2String();

maybe call this buildSASLResponse() ? "OAuth" is implied by the module, and "string" is implied by the return type.

Comment 97

2 years ago
Created attachment 8589246 [details] [diff] [review]
Last patch from jcramner

jcranmer flat out refuses to attach a diff, so I'll do it. I hope this is the right one. It should match the last patch he attached.
(reviewboard is useless, cumbersome and unusable for me.)

Comment 98

2 years ago
Comment on attachment 8589246 [details] [diff] [review]
Last patch from jcramner

Doesn't contain any of the fixes to the SMTP and IMAP implementations, mentioned above.
Attachment #8589246 - Flags: review-

Comment 99

2 years ago
More bugs that needed fixing before this would work:
verifyConfig.js:
-    oauthConfig.issuer = sanitizer.nonemptystring(config.issuer);
+    oauthConfig.issuer = sanitize.nonemptystring(config.issuer);

msgOAuth2Module.js:
     for (let login of logins) {
-      if (login.username == this._username)
+      if (login.username == this._username) {
         loginMgr.modifyLogin(login, {password: token});
-      return token;
+        return token;
+      }

After working with this for a day, here's my thoughts on direction.

I don't believe that we are going to be ready for TB 38 to suddenly switch all new gmail accounts to OAUTH, though clearly that is what Google is proposing. What I propose is a compromise. I've played around with some changes that will allow us to specify for GMail that we first try passwordCleartext (which is our current default), and only if that fails do we then try OAuth2. I'll propose patch modifications with those changes hopefully on Thursday, as well as any other review comments.

Updated

2 years ago
Depends on: 1153383
Created attachment 8591190 [details] [diff] [review]
rkent changes to jcranmer's patch

I've spent several days now working on this. I had to change the patch to make it work, plus it was easier to test my suggested changes by just doing them. So this is a patch with my changes, applied on top of jcranmer's last patch. I'll use the review system to make comments on my code, and the issues addressed relative to jcranmer's patch.
Comment on attachment 8591190 [details] [diff] [review]
rkent changes to jcranmer's patch

Review of attachment 8591190 [details] [diff] [review]:
-----------------------------------------------------------------

I'm giving my comments here. I'll post the full patch with both sets of changes as well. I think that the best way forward is to have you look at that patch, and do the equivalent of a review on my changes. I'm not taking over the bug, I'm just asking that both jcranmer and rkent agree jointly.

::: mailnews/base/prefs/content/accountcreation/createInBackend.js
@@ +33,5 @@
>      inServer.setCharValue("oauth2.issuer", config.oauthSettings.issuer);
> +    inServer.setCharValue("oauth2.authEndpoint",
> +                          config.oauthSettings.authEndpoint);
> +    inServer.setCharValue("oauth2.tokenEndpoint",
> +                          config.oauthSettings.tokenEndpoint);

If you are going to use openID configuration, the endpoints are key results that need storing with the other oauth parameters.

It turns our that you are not really using discoverOpenIDConfig to do any real work, but in the future if you do these need storing, so we might as well do it now.

::: mailnews/base/prefs/content/accountcreation/emailWizard.js
@@ +36,5 @@
>  
> +if (typeof gEmailWizardLogger == "undefined") {
> +  Cu.import("resource:///modules/gloda/log4moz.js");
> +  var gEmailWizardLogger = Log4Moz.getConfiguredLogger("mail.wizard");
> +}

I did a lot of logging changes to help with debugging, might as well leave these. I won't comment on them further.

::: mailnews/base/prefs/content/accountcreation/sanitizeDatatypes.js
@@ +117,5 @@
>        throw new MalformedException("url_scheme.error", unchecked);
>  
>      var uri;
>      try {
> +      uri = Services.io.newURI(str, null, null);

ioService() was removed many versions ago.

::: mailnews/base/prefs/content/accountcreation/verifyConfig.js
@@ +99,1 @@
>          verifyLogon(config, inServer, alter, msgWindow,

Other authentications are verified at this point so that you do not create an unusable account. I changed OAuth2 so that it behaves the same way.

@@ +361,5 @@
>    let result = configLoaded.promise.then(function (result) {
>      let config = JSON.parse(result);
>      oauthConfig.authEndpoint = sanitize.url(config.authorization_endpoint);
>      oauthConfig.tokenEndpoint = sanitize.url(config.token_endpoint);
> +    oauthConfig.issuer = sanitize.nonemptystring(config.issuer);

This was a another key bug fix, getting rid of this typo.

::: mailnews/base/src/msgOAuth2Module.js
@@ +28,5 @@
>        // own application--register it yourself. This code (and possibly even the
>        // registration itself) will disappear when this is switched to dynamic
>        // client registration.
> +      this._appKey = '406964657835-aq8lmia8j95dhl1a2bvharmfk3t1hgqj.apps.googleusercontent.com';
> +      this._appSecret = 'kSmqreRr0qwBWJgbf5Y-PjSU';

The app secret has been changed to what should be the official appKey rather than your test one.

@@ +65,5 @@
> +                        "https://www.googleapis.com/oauth2/v3/token");
> +      }
> +      else
> +        return false;
> +    }

With the change, you can take an existing GMail account, change the authentication to OAuth2, and it will work. Yes it is tacky to hard-wire for a single provider, but we have done that anyway, so this is not any more tacky, and it adds an important capability.

@@ +138,1 @@
>      }

I don't believe your login manager calls were correct (and they failed when I changed the token from the test account to the final account).

::: mailnews/base/util/OAuth2.jsm
@@ +23,5 @@
>    return result;
>  }
>  
> +// Only allow one connecting window per endpoint.
> +var gConnecting = {};

Since IMAP typically opens up several connections, each of these was putting up a separate authentication window. This change only allows a single authentication window to be open. (Normal password needed window is a modal window which also solves the problem, but this seems preferable).

::: mailnews/imap/src/nsSyncRunnableHelpers.h
@@ +122,5 @@
>  
>  namespace mozilla {
>  namespace mailnews {
>  
> +class OAuth2ThreadHelper final : public msgIOAuth2ModuleListener

This change was needed since m-c now requires this.
Attachment #8591190 - Flags: review+
Created attachment 8591191 [details] [diff] [review]
Full patch, jcranmer original plus rkent changes
Attachment #8591191 - Flags: review?(Pidgeot18)
(Assignee)

Comment 103

2 years ago
Comment on attachment 8591191 [details] [diff] [review]
Full patch, jcranmer original plus rkent changes

Review of attachment 8591191 [details] [diff] [review]:
-----------------------------------------------------------------

::: mailnews/base/src/msgOAuth2Module.js
@@ +61,5 @@
> +        Preferences.set(root + "oauth2.scope", scope);
> +        Preferences.set(root + "oauth2.authEndpoint",
> +                        "https://accounts.google.com/o/oauth2/auth");
> +        Preferences.set(root + "oauth2.tokenEndpoint",
> +                        "https://www.googleapis.com/oauth2/v3/token");

I heavily disagree with setting endpoints as preferences. For dynamic client registration, we would need to store a persisted map of issuer => [endpoints, client key, client secret]--I envisioned this as a standalone json file (cf. various threads on m.d.platform about "don't use preferences as a generic key-value store").

Additionally, I dislike hardcoding the two gmail hostnames. Gmail has far more hostnames, such as if you're using a Google Apps-hosted account. (Not that I have one of these to test).

@@ +128,5 @@
> +        if (token) {
> +          let propBag = Cc["@mozilla.org/hash-property-bag;1"].
> +                        createInstance(Ci.nsIWritablePropertyBag);
> +          propBag.setProperty("password", token);
> +          loginMgr.modifyLogin(login, propBag);

Any JS object can be QI'd to a nsIPropertyBag (cf. <https://dxr.mozilla.org/comm-central/source/mozilla/js/xpconnect/src/XPCWrappedJSClass.cpp#489>). I'd prefer a JS token.
(Assignee)

Updated

2 years ago
Flags: needinfo?(Pidgeot18)
(In reply to Joshua Cranmer [:jcranmer] from comment #103)
> Comment on attachment 8591191 [details] [diff] [review]
> Full patch, jcranmer original plus rkent changes
> 
> Review of attachment 8591191 [details] [diff] [review]:
> -----------------------------------------------------------------
> 
> ::: mailnews/base/src/msgOAuth2Module.js
> @@ +61,5 @@
> > +        Preferences.set(root + "oauth2.scope", scope);
> > +        Preferences.set(root + "oauth2.authEndpoint",
> > +                        "https://accounts.google.com/o/oauth2/auth");
> > +        Preferences.set(root + "oauth2.tokenEndpoint",
> > +                        "https://www.googleapis.com/oauth2/v3/token");
> 
> I heavily disagree with setting endpoints as preferences. For dynamic client
> registration, we would need to store a persisted map of issuer =>
> [endpoints, client key, client secret]--I envisioned this as a standalone
> json file (cf. various threads on m.d.platform about "don't use preferences
> as a generic key-value store").

I don't feel particularly strongly about this, but then what is the point of the whole discoverOpenIDConfig? It returns authEndpoint and tokenEndpoint (which you object to storing anywhere) and issuer. So although you are reading those values, you never actually use them, instead they are hard-coded elsewhere. The only thing that you are supposed to do with issuer is to ensure that its value is accounts.googlemail.com so it is also not of any value. Can we just get rid of discoverOpenIDConfig?

> 
> Additionally, I dislike hardcoding the two gmail hostnames. Gmail has far
> more hostnames, such as if you're using a Google Apps-hosted account. (Not
> that I have one of these to test).
> 

The hardcoding was in your original code, the only thing that I did was to not force you to go through autoconfig to use the hard-coded values. How do you propose that we recognize GMail hostnames and relate them to our OAuth account?

And although you say that there are many GMail hostnames, the current autoconfig file only supports the one. Supporting the typical hostname is better than supporting nothing, if we don't cause some issue that we have to backport later.

> @@ +128,5 @@
> > +        if (token) {
> > +          let propBag = Cc["@mozilla.org/hash-property-bag;1"].
> > +                        createInstance(Ci.nsIWritablePropertyBag);
> > +          propBag.setProperty("password", token);
> > +          loginMgr.modifyLogin(login, propBag);
> 
> Any JS object can be QI'd to a nsIPropertyBag (cf.
> <https://dxr.mozilla.org/comm-central/source/mozilla/js/xpconnect/src/
> XPCWrappedJSClass.cpp#489>). I'd prefer a JS token.

OK fine, what I showed was from other examples in existing code. The original code as written did not work, this does. If you have another method please propose it.

Comment 105

2 years ago
(In reply to Kent James (:rkent) from comment #104)

> 
> And although you say that there are many GMail hostnames, the current
> autoconfig file only supports the one. Supporting the typical hostname is
> better than supporting nothing, if we don't cause some issue that we have to
> backport later.

Are we going to have a manual config option for the occasions that the auto config does not work?  Some way to tell Thunderbird the account requires Oauth and the provider is?
(Assignee)

Comment 106

2 years ago
Created attachment 8593467 [details] [diff] [review]
Alternative patch without account creation

It's the account autoconfig that's proving to be the sticky part of the changes, so here's a patch without those controversial pieces.
(In reply to Matt from comment #105)
> (In reply to Kent James (:rkent) from comment #104)
> 
> Are we going to have a manual config option for the occasions that the auto
> config does not work?  Some way to tell Thunderbird the account requires
> Oauth and the provider is?

In the patch proposal that I did, users will be able to select "OAuth" in manual account configuration. If the hostname matches the standard GMail server, then Thunderbird would select the GMail OAuth account for Thunderbird, and attempt to login to that using OAuth. If anything is different (another OAuth IMAP provider, non-standard GMail hostname) it would not work.
Created attachment 8593664 [details] [diff] [review]
Backend only, improve support of failed logins

Joshua, technically this is your patch with my review. If you can approve this we can land it. I just made some changes in the handling of login failures.
Attachment #8593664 - Flags: review+
Attachment #8593664 - Flags: feedback?(Pidgeot18)
(Assignee)

Comment 109

2 years ago
Comment on attachment 8593664 [details] [diff] [review]
Backend only, improve support of failed logins

Review of attachment 8593664 [details] [diff] [review]:
-----------------------------------------------------------------

::: mailnews/base/src/msgOAuth2Module.js
@@ +61,5 @@
> +        Preferences.set(root + "oauth2.scope", scope);
> +        Preferences.set(root + "oauth2.authEndpoint",
> +                        "https://accounts.google.com/o/oauth2/auth");
> +        Preferences.set(root + "oauth2.tokenEndpoint",
> +                        "https://www.googleapis.com/oauth2/v3/token");

f/r+, assuming you take out the last two lines here.
Sorry, I thought I uploaded a patch without these, but I apparently uploaded before hg qref. >_<
Attachment #8593664 - Flags: feedback?(Pidgeot18) → feedback+
Comment on attachment 8593664 [details] [diff] [review]
Backend only, improve support of failed logins

https://hg.mozilla.org/comm-central/rev/eae1195fde6d

(keep open)
The patch that just landed allows OAuth to be selected in the Advanced Config dialog. This will only work for gmail accounts, specifically accounts with a hostname of imap.googlemail.com or smtp.googlemail.com I'm going to try to get this in the next beta as well.

Still remaining to be done:

1) Allow OAuth2 during manual configuration of gmail accounts.

2) Allow OAuth2 to be set as the autoconfig account type for gmail accounts

3) Change ispdb so that default autoconfig for gmail accounts is OAuth2
Comment on attachment 8593664 [details] [diff] [review]
Backend only, improve support of failed logins

[Triage Comment]

https://hg.mozilla.org/releases/comm-aurora/rev/e149d5f4bcdb
Attachment #8593664 - Flags: approval-comm-aurora+
It is difficult to track patches that land in different versions of Thunderbird within a single bug, so I am going to clone this bug for the followup work of supporting autoconfig.

Updated

2 years ago
Blocks: 1155491
Comment on attachment 8593664 [details] [diff] [review]
Backend only, improve support of failed logins

[Triage Comment]

https://hg.mozilla.org/releases/comm-beta/rev/80ec65353d8f
Attachment #8593664 - Flags: approval-comm-beta+

Updated

2 years ago
Attachment #8565317 - Attachment is obsolete: true
Attachment #8565317 - Flags: feedback?(ben.bucksch)

Updated

2 years ago
Attachment #8589246 - Attachment is obsolete: true

Updated

2 years ago
Attachment #8591190 - Attachment is obsolete: true

Updated

2 years ago
Attachment #8591191 - Attachment is obsolete: true
Attachment #8591191 - Flags: review?(Pidgeot18)

Updated

2 years ago
Attachment #8593467 - Attachment is obsolete: true

Updated

2 years ago
Attachment #8565826 - Attachment is obsolete: true
Attachment #8565826 - Flags: review?(rkent)

Updated

2 years ago
status-thunderbird38: --- → fixed
status-thunderbird39: --- → fixed
status-thunderbird40: --- → fixed
Summary: Log in to Gmail (IMAP/SMTP) using OAuth → Log in to Gmail (IMAP/SMTP) using OAuth in backend

Updated

2 years ago
Status: ASSIGNED → RESOLVED
Last Resolved: 3 years ago2 years ago
Resolution: --- → FIXED
Target Milestone: --- → Thunderbird 40.0

Updated

2 years ago
Duplicate of this bug: 1138020

Updated

2 years ago
Blocks: 1163345

Comment 116

2 years ago
I have been following this bug, and thought I would try it out.  So I changed the authentication method for my existing gmail account to oAuth.

The message is activity manager is 
The imap server devilsgatedrive@gmail.com does not support the selected authentication.  Likewise I received a similar error when trying to use SMTP on the account with oauth.

Could you please tell me what it is I am missing here.  I thought Google were requiring oauth on new accounts. This account is relatively new.  I created is after the 24th October last year when Roland edited the Google KB article to test what was happening.
Matt, what version of TB are you using?

Most of the testing has been while creating a new account, though I would expect what you did to work.

Comment 118

2 years ago
(In reply to Kent James (:rkent) from comment #117)
> Matt, what version of TB are you using?
> 
> Most of the testing has been while creating a new account, though I would
> expect what you did to work.

Daily build... says
Built from https://hg.mozilla.org/mozilla-central/rev/35918b0441b4

I have not installed the latest daily,  So I think that will be the build I tried.
In Moztrap - https://moztrap.mozilla.org/manage/case/16262/
Flags: in-moztrap+
Keywords: user-doc-needed
Before opening a bug, I would ask, OAuth2 is supported for POP3 too?
because if I try to use OAuth 2 for a Google account configured with POP3 it fails.
When I create it using TB Account Wizard it is created as a normal POP3 account, without using OAuth, I select it in Server settings.
If I disable “less secure app” support” this is the error message:
Web login required: https://support.google.com/mail/answer/78754
If I enable it nothing happens.
I attach the POP3 logs:
https://dl.dropboxusercontent.com/u/4400966/pop3log.txt
If it is not supported, sorry.
For Imap and SMTP all works fine :-D

Comment 121

2 years ago
For POP OAuth2: bug 1174505
Thanks Magnus for the link.

Comment 123

2 years ago
Could this have broken the compatability with non oauth Servers? Since I use thunderbird 39 I cannot login to my own server anymoore(courier-imap-ssl)I get this:
 LOGIN = 0x2, old-style IMAP login = 0x4, auth external IMAP login = 0x20000000, OAUTH2 = 0x800000000)
10044[142c8860]: trying auth method 0x1000
10044[142c8860]: IMAP: password prompt failed or user canceled it
10044[142c8860]: login failed entirely
10044[142c8860]: 1400c000:owa.extranet.1and1.com:NA:ProcessCurrentURL: aborting queued urls
10044[142c8860]: 1400c000:owa.extranet.1and1.com:NA:TellThreadToDie: close socket connection
10044[142c8860]: ImapThreadMainLoop leaving [this=1400c000]
4004[142c85c0]: ReadNextLine [stream=1b5c6330 nb=82 needmore=0]
4004[142c85c0]: 1251a000:owa.extranet.1and1.com:NA:CreateNewLineFromSocket: * CAPABILITY IMAP4 IMAP4rev1 AUTH=PLAIN UIDPLUS CHILDREN IDLE NAMESPACE LITERAL+

4004[142c85c0]: ReadNextLine [stream=1b5c6330 nb=28 needmore=0]
4004[142c85c0]: 1251a000:owa.extranet.1and1.com:NA:CreateNewLineFromSocket: 3 OK CAPABILITY completed.

4

Comment 124

2 years ago
(In reply to Ben Bucksch (:BenB) from comment #15)
> Quote:
> "Embrace, extend, and extinguish" ...

Ben is/was correct, you guys/Mozilla messed up and you might not be able to recover at this point. I'm sorry, but why is a web browser now required for IMAP e-mail? Seriously!

Like I said here, http://support.mozilla.org/en-US/questions/1075788, the major issue is *I* can't manually download the OAuth2 token and add it *myself* to TB and having this capability solves a lot of issues.
Is there any way to see the patch that actually got checked in here?

I see no hg links to the actual checkin (I see the l10n changes and the backend fix).

And all the mozreview links are old.

What actually got checked in to fix this bug?
Comment 110 was the checkin for this bug, further work was checked in using bug 1153383.
Blocks: 1293958

Updated

8 months ago
Blocks: 1310389
Depends on: 1232227
See Also: → bug 1176399
You need to log in before you can comment on or make changes to this bug.