45565 - Make UA string customizable

Status: VERIFIED FIXED

(Core :: Networking, defect, P3)

Description

[Ben Bucksch (:BenB)]

"Mozilla/5.0" is hardcoded. Read from prefs. Some users want to change this. 

Reading the whole beast from prefs would be even better.

Patch will follow. Jud, can you review, please?

Comment 1

[Judson Valeski]

this piece of the UA string is intentionally hardcoded. You can't browse without 
that hardcoded string (modifications break the world).

Could the value be changed programatically via JS in a web page if it were a 
pref?

Comment 2

[Ben Bucksch (:BenB)]

> this piece of the UA string is intentionally hardcoded. You can't browse
> without that hardcoded string (modifications break the world).

Sorry, but this is nonsense. People use very freaky UA-strings (sat by various
proxies). Even if some pages broke, this decision should be up to the user.
There is enough demand for this choice.

> Could the value be changed programatically via JS in a web page if it were a 
> pref?

Webpage JS *should* not be able to set prefs - it would be a security problem in
many other cases as well. Note that you can currently set e.g. the security part
via pref.

Comment 3

[Ben Bucksch (:BenB)]

Attachment: Fix

Comment 4

[Ben Bucksch (:BenB)]

3 new prefs: general.useragent.[appname|appversion|override]. With the first
two, you can change "Mozilla" and "5.0" (falling back to the previous hardcoded
defaults, if they are not sat), with the last one you can override the whole
useragent string at once, which allows to break out of the structure and even
override the OS etc. (which can be a security *feature*, if used well).

Comment 5

[Judson Valeski]

1. It ain't non-sense when modifying Mozilla/x.x results in dozens of emails 
from people complaining about how they can't visit site foo.

2. Proxies, spiders, agents and various clients indeed use many random UA 
strings. However, there's a reason 95% of the clients on the web hard code 
"Mozilla/x.x" and that's because there's a lot of content out there that won't 
service any other requests.

3. I've had this "discussion" a zillion times, and the same logic is always 
used. I agree that the entire string *should* be modifiable, but what worries me 
is someone cracking our JS security model somehow, putting up a page that 
changes this setting, then that browser becomes useless until the user figures 
out what went wrong. Maybe this is an unwaranted fear.

actually, on that last note, I'm going to ask that we *not* make this change. If 
you (or someone else playing w/ this) change your agent and you starting getting 
rejections (of many forms) from content, you could very easily file a bug that 
is a result of your UA string being different, and someone (most often in 
networking) will spend a good hunk of time (I've spent days chasing UA 
string problems) chasing a phantom problem.

Comment 6

[Brendan Eich [:brendan]]

Cc'ing dbaron, who is user-agent owner for Mozilla, and waterson.  Jud, I don't 
think your objection #3 is valid: if someone cracks our security code, many bad 
things can happen, but that doesn't mean we don't empower trusted scripts to do 
useful things.  We do and should.

I am not convinced a pref is the right way here at all.  It's more a matter of 
compile-time branding than user preference, I think.  If we could use #ifndef 
and #define to avoid hardwiring Mozilla/5.0 in a way that makes branders of the 
source have to muck with an otherwise readonly source file, that would win.

/be

Comment 7

[Ben Bucksch (:BenB)]

Posted to .netlib "Should the UA string be settable per pref?"
<news://news.mozilla.org/[e-mail]>.

Comment 8

[Ben Bucksch (:BenB)]

> I am not convinced a pref is the right way here at all.  It's more a matter of 
> compile-time branding than user preference, I think.  If we could use #ifndef 
> and #define to avoid hardwiring Mozilla/5.0 in a way that makes branders of
> the source have to muck with an otherwise readonly source file, that would
> win.

I don't think, this is a problem. *All* Mozillas of this version propably should
have "Mozilla/5.0" in the beginning, for the reasons valeski and the spec name.
We have "vendor" fields for branding. But many *users* want to change the
string, without compiling.

Comment 9

[Brendan Eich [:brendan]]

> But many *users* want to change the string, without compiling.

First I've heard of this.  I doubt Mom&Pop want to do any such thing.  Can you 
tell me more about such users?  *Why* do they want to change the beginning of 
the user-agent?

/be

Comment 10

[Ben Bucksch (:BenB)]

> I doubt Mom&Pop want to do any such thing.

lol, agreed.

> Can you tell me more about such users?  *Why* do they want to change the
> beginning of the user-agent?

(Easiest ones first)
- Change for the sake of it
- "See, I'm cool. I can hack my browser."
- Security: If a webserver admin knows details about my platform (e.g. the
kernel version, as sent out for Unix), he can start a specific attack.
  - Many exploits involve writing of certain files, which are of course
different across platforms.
  - Certain OS versions may have known bugs, e.g. the remote-root-exploit bug of
the 2.2.14 kernel.
- Privacy
  - I just may not want others to know my platform
  - I may use an usual combination and want to set it to a more usual string

Comment 11

[Ben Bucksch (:BenB)]

(Mostly talking about the whole UA string, i.e. "general.useragent.override".)

Comment 12

[Ben Bucksch (:BenB)]

> I may use an usual combination

*un*usual (sorry for the spam.)

Comment 13

[usenet]

I want that setting, too :-)

Comment 14

[nick255]

You might want to change the user_agent because some websites don't like the browser version or OS.  I have an online banking site that only likes Windows or MacOS browsers.  It doesn't even like Opera unless you set it to spoof as netscape 4.73.  Perhaps instead of fully customizable, there could be a few options known to work, such as the following prefs:

Pretend to be Netscape 4.08
Pretend to be Netscape 4.73
Pretend to be MSIE 5.0
Pretend to be Netscape 3.04
Pretend to be a Windows browser
Pretend to be a MacOS browser

Comment 15

[Ben Bucksch (:BenB)]

So, what do we do now?

Comment 16

[David Baron :dbaron: (⌚️UTC-5, no longer working on Mozilla)]

Having a general override for the whole string seems like it does have some uses
for power users (as mentioned above), especially on uncommon platforms, and also
maybe for our QA, if we want to test if a layout "bug" is really the site's bad
browser-sniffing or test UA-string changes in advance.  I don't see the value of
having separate overrides for AppName and AppVersion, unless that would affect
JS sniffing too.

However, messing with the UA string should be "unsupported" for end-users (since
we know a lot of things break if it's changed), and it shouldn't be possible to
do so accidentally or for a malicious web page to do so.

As far as the UA-string spec is concerned, I think all that matters is what the
default is that ships with the browser.  That's what most people will use. 
Furthermore, (almost) however that default is stored, it can be changed.

Comment 17

[Ben Bucksch (:BenB)]

> I don't see the value of
> having separate overrides for AppName and AppVersion

I could remove that.

> messing with the UA string should be "unsupported" for end-users

Agreed.

Comment 18

[Ben Bucksch (:BenB)]

I'll attach a new patch, which only adds the general.useragent.override prefs
and leaves Mozilla/5.0 hardcoded. Changing SUMMARY to reflect this.

Comment 19

[Ben Bucksch (:BenB)]

Attachment: Fix, version 2 (can override the completely string only)

Comment 20

[nick255]

Some websites will not allow Mozilla/5.0.  I know, my online banking site won't 
allow netscape 4.08 for linux, only windows.  Perhaps 4.7 or 4.08 should be 
allowed as well.

Comment 21

[Brendan Eich [:brendan]]

Ben, looks good to me except that now rv is uselessly initialized to NS_OK, then 
immediately set to the result of CopyCharPref.  I would remove the useless NS_OK 
initializer.  Anyone else want to give this an r=?  I'll say a= once the NS_OK 
nit is picked.

/be

Comment 22

[Ben Bucksch (:BenB)]

Brendan,
the needless initialization is preexistant, but I'll remove it before I check
in.

Please see <http://www.bucksch.org/1/projects/mozilla/review.html> before
reviewing.

Comment 23

[Ben Bucksch (:BenB)]

gagan, would you review, please?

Comment 24

[Gagan]

Jud should review this. 

Comment 25

[Judson Valeski]

ugh. user agents suck. we're about to rewrite all this stuff and I'd like to 
deal w/ this kind of customization then/after that.

Comment 26

[Ben Bucksch (:BenB)]

Rewrite what?

Anyway, can't we just get this in now, and when you rewrite, you just maintain
that (override) functionality? I'd like to
- have this from my plate and
- make sure this is in the product

Comment 27

[Judson Valeski]

the ua stuff is being overhauled. r=valeski.

Comment 28

[Ben Bucksch (:BenB)]

> the ua stuff is being overhauled.

Can you cc me in that bug, please?

> r=valeski

Thanks

Comment 29

[Ben Bucksch (:BenB)]

Checked in.

Comment 30

[Daniel Roberts]

I realize the fix is already in, but since there seems to be quite a bit of
disagreement on its usefulness, I just thought I would mention another posible
use for it: To spoof other browsers.  For example, there is quite a bit of
content out there that is unviewable with anything other than IE.  This is a
long-running problem, but I blame it on IE in the first place, which started
this mess by pretending to be Mozilla, changing its UA string to "Mozilla/x.x
(compatible: MSIE x.x)", to work around sites in existence at the time that were
looking at the UA string and refusing to serve content to IE.  With a pref for
the UA string, this could be solved.

BTW, for a GOOD example of it in action, check out the UA-string spoofing option
at the web page backward compatibility viewer, now playing at an URL near you:

http://www.delorie.com/web/wpbcv.html

Comment 31

[benc]

VERIFIED:
I'd like to see further documentation of these preferences, but this feature is 
clearly in and used all the time.

I compared logs of user agent to the relevant contents of 
"$INSTALL/defaults/pref/*"

"Mozilla/5.0 (X11; U; Linux 2.2.16-22smp i686; en-US; rv:0.9) Gecko/20010505"

all.js:pref("general.useragent.locale", 
"chrome://navigator/locale/navigator.properties");
all.js:pref("general.useragent.misc", "rv:0.9");
security-prefs.js:pref("general.useragent.security",       "U");

"Mozilla/5.0 (X11; U; Linux 2.2.16-22smp i686; en-US; rv:0.8.1+) Gecko/20010416
Netscape6/6.5b0"

all.js:pref("general.useragent.locale", 
"chrome://navigator/locale/navigator.properties");
all.js:pref("general.useragent.misc", "rv:0.8.1+");
all-ns.js:pref("general.useragent.vendor", "Netscape6");
all-ns.js:pref("general.useragent.vendorSub", "6.5b0");
security-prefs.js:pref("general.useragent.security",       "U");

Comment 32

[Ben Bucksch (:BenB)]

benc, the new pref here is general.useragent.override. Please test *that*.

Comment 33

[Ben Bucksch (:BenB)]

Bug 46029 is about a GUI for this pref.

Comment 34

[Ben Bucksch (:BenB)]

Adding mostfreq keyword for the many bugs that are dupped against bug 46029.