summaryrefslogtreecommitdiff
path: root/doc/plugins/po/discussion.mdwn
blob: 27683f1eae22ddce2fca8c01155fee976f35c273 (plain)

[[!toc ]]


Security review

Probable holes

(The list of things to fix.)

po4a-gettextize

  • po4a CVS 2009-01-16
  • Perl 5.10.0

po4a-gettextize uses more or less the same po4a features as our refreshpot function.

Without specifying an input charset, zzuf'ed po4a-gettextize quickly errors out, complaining it was not able to detect the input charset; it leaves no incomplete file on disk. I therefore had to pretend the input was in UTF-8, as does the po plugin.

    zzuf -c -s 13 -r 0.1 \
        po4a-gettextize -f text -o markdown -M utf-8 -L utf-8 \
         -m GPL-3 -p GPL-3.pot

Crashes with:

    Malformed UTF-8 character (UTF-16 surrogate 0xdfa4) in substitution
    iterator at /usr/share/perl5/Locale/Po4a/Po.pm line 1449.
    Malformed UTF-8 character (fatal) at /usr/share/perl5/Locale/Po4a/Po.pm
    line 1449.

An incomplete pot file is left on disk. Unfortunately Po.pm tells us nothing about the place where the crash happens.

It's fairly standard perl behavior when fed malformed utf-8. As long as it doesn't crash ikiwiki, it's probably acceptable. Ikiwiki can do some similar things itself when fed malformed utf-8 (doesn't crash tho) --[[Joey]]


Potential gotchas

(Things not to do.)

Blindly activating more po4a format modules

The format modules we want to use have to be checked, as not all are safe (e.g. the LaTeX module's behaviour is changed by commands included in the content); they may use regexps generated from the content.


Hopefully non-holes

(AKA, the assumptions that will be the root of most security holes...)

PO file features

No documented directive that can be put in po files is supposed to cause mischief (ie, include other files, run commands, crash gettext, whatever).

gettext

Security history

The only past security issue I could find in GNU gettext is CVE-2004-0966, i.e. Debian bug #278283: the autopoint and gettextize scripts in the GNU gettext package (1.14 and later versions) may allow local users to overwrite files via a symlink attack on temporary files.

This plugin would not have allowed to exploit this bug, as it does not use, either directly or indirectly, the faulty scripts.

Note: the lack of found security issues can either indicate that there are none, or reveal that no-one ever bothered to find or publish them.

msgmerge

refreshpofiles() runs this external program.

  • I was not able to crash it with zzuf.
  • I could not find any past security hole.

msgfmt

isvalidpo() runs this external program.

  • I was not able to make it behave badly using zzuf: it exits cleanly when too many errors are detected.
  • I could not find any past security hole.

po4a

Security history

The only past security issue I could find in po4a is CVE-2007-4462: lib/Locale/Po4a/Po.pm in po4a before 0.32 allowed local users to overwrite arbitrary files via a symlink attack on the gettextization.failed.po temporary file.

This plugin would not have allowed to exploit this bug, as it does not use, either directly or indirectly, the faulty gettextize function.

Note: the lack of found security issues can either indicate that there are none, or reveal that no-one ever bothered to find or publish them.

General feeling

Are there any security issues on running po4a on untrusted content?

To say the least, this issue is not well covered, at least publicly:

  • the documentation does not talk about it;
  • grep'ing the source code for security or trust gives no answer.

On the other hand, a po4a developer answered my questions in a convincing manner, stating that processing untrusted content was not an initial goal, and analysing in detail the possible issues. The following analysis was done with his help.

Details

  • the core (Po.pm, Transtractor.pm) should be safe
  • po4a source code was fully checked for other potential symlink attacks, after discovery of one such issue
  • the only external program run by the core is diff, in Po.pm (in parts of its code we don't use)
  • Locale::gettext is only used to display translated error messages
  • Nicolas François "hopes" DynaLoader is safe, and has "no reason to think that Encode is not safe"
  • Nicolas François has "no reason to think that Encode::Guess is not safe". The po plugin nevertheless avoids using it by defining the input charset (file_in_charset) before asking TransTractor to read any file. NB: this hack depends on po4a internals.
Locale::Po4a::Text
  • does not run any external program
  • only do_paragraph() builds regexp's that expand untrusted variables; according to [[Joey]], this is "Freaky code, but seems ok due to use of quotementa".
Text::WrapI18N

Text::WrapI18N can cause DoS (Debian bug #470250). It is optional, and we do not need the features it provides.

If a recent enough po4a (>=0.35) is installed, this module's use is fully disabled. Else, the wiki administrator is warned about this at runtime.

Term::ReadKey

Term::ReadKey is not a hard dependency in our case, i.e. po4a works nicely without it. But the po4a Debian package recommends libterm-readkey-perl, so it will probably be installed on most systems using the po plugin.

Term::ReadKey has too far reaching implications for us to be able to guarantee anything wrt. security.

If a recent enough po4a (>=2009-01-15 CVS, which will probably be released as 0.35) is installed, this module's use is fully disabled.

Fuzzing input
po4a-translate
  • po4a CVS 2009-01-16
  • Perl 5.10.0

po4a-translate uses more or less the same po4a features as our filter function.

Without specifying an input charset, same behaviour as po4a-gettextize, so let's specify UTF-8 as input charset as of now.

LICENSES is a 21M file containing 100 concatenated copies of all the files in /usr/share/common-licenses/; I had no existing PO file or translated versions at hand, which renders these tests quite incomplete.

    zzuf -cv -s 0:10 -r 0.001:0.3 \
      po4a-translate -d -f text -o markdown -M utf-8 -L utf-8 \
        -k 0 -m LICENSES -p LICENSES.fr.po -l test.fr

... seems to lose the fight, at the readpo(LICENSES.fr.po) step, against some kind of infinite loop, deadlock, or any similar beast.

The root of this bug lies in Text::WrapI18N, see the corresponding section.


Fixed holes


original contrib/po page, with old commentary

I've been working on a plugin called "po", that adds support for multi-lingual wikis, translated with gettext, using po4a.

More information:

  • It can be found in my "po" branch: git clone git://gaffer.ptitcanardnoir.org/ikiwiki.git
  • It is self-contained, i.e. it does not modify ikiwiki core at all.
  • It is documented (including TODO and plans for next work steps) in doc/plugins/po.mdwn, which can be found in the same branch.
  • No public demo site is available so far, I'm working on this.

My plan is to get this plugin clean enough to be included in ikiwiki.

The current version is a proof-of-concept, mature enough for me to dare submitting it here, but I'm prepared to hear various helpful remarks, and to rewrite parts of it as needed.

Any thoughts on this?

Well, I think it's pretty stunning what you've done here. Seems very complete and well thought out. I have not read the code in great detail yet.

Just using po files is an approach I've never seen tried with a wiki. I suspect it will work better for some wikis than others. For wikis that just want translations that match the master language as closely as possible and don't wander off and diverge, it seems perfect. (But what happens if someone edits the Discussion page of a translated page?)

Please keep me posted, when you get closer to having all issues solved and ready for merging I can do a review and hopefully help with the security items you listed. --[[Joey]]

Thanks a lot for your quick review, it's reassuring to hear such nice words from you. I did not want to design and write a full translation system, when tools such as gettext/po4a already have all the needed functionality, for cases where the master/slave languages paradigm fits. Integrating these tools into ikiwiki plugin system was a pleasure.

I'll tell you when I'm ready for merging, but in the meantime, I'd like you to review the changes I did to the core (3 added hooks). Can you please do this? If not, I'll go on and hope I'm not going to far in the wrong direction.

Sure.. I'm not completly happy with any of the hooks since they're very special purpose, and also since run_hooks is not the best interface for a hook that modifies a variable, where only the last hook run will actually do anything. It might be better to just wrap targetpage, bestlink, and beautify_urlpath. But, I noticed the other day that such wrappers around exported functions are only visible by plugins loaded after the plugin that defines them.

Update: Take a look at the new "Function overriding" section of [[plugins/write]]. I think you can just inject wrappers about a few ikiwiki functions, rather than adding hooks. The inject function is pretty insane^Wlow level, but seems to work great. --[[Joey]]

Thanks a lot, it seems to be a nice interface for what I was trying to achieve. I may be forced to wait two long weeks before I have a chance to confirm this. Stay tuned. --[[intrigeri]]

I've updated the plugin to use inject. It is now fully self-contained, and does not modify the core anymore. --[[intrigeri]]

The Discussion pages issue is something I am not sure about yet. But I will probably decide that "slave" pages, being only translations, don't deserve a discussion page: the discussion should happen in the language in which the pages are written for real, which is the "master" one. --[[intrigeri]]

I think that's a good decision, you don't want to translate discussion, and if the discussion page turns out multilingual, well, se la vi. ;-)

Relatedly, what happens if a translated page has a broken link, and you click on it to edit it? Seems you'd first have to create a master page and could only then translate it, right? I wonder if this will be clear though to the user.

Right: a broken link points to the URL that allows to create a page that can either be a new master page or a non-translatable page, depending on po_translatable_pages value. The best solution I can thing of is to use [[plugins/edittemplate]] to insert something like "Warning: this is a master page, that must be written in $MASTER_LANGUAGE" into newly created master pages, and maybe another warning message on newly created non-translatable pages. It seems quite doable to me, but in order to avoid breaking existing functionality, it implies to hack a bit [[plugins/edittemplate]] so that multiple templates can be inserted at page creation time. [[--intrigeri]]

I implemented such a warning using the formbuilder_setup hook. --[[intrigeri]]

And also, is there any way to start a translation of a page into a new lanauge using the web interface?

When a new language is added to po_slave_languages, a rebuild is triggered, and all missing PO files are created and checked into VCS. An unpriviledged wiki user can not add a new language to po_slave_languages, though. One could think of adding the needed interface to translate a page into a yet-unsupported slave language, and this would automagically add this new language to po_slave_languages. It would probably be useful in some usecases, but I'm not comfortable with letting unpriviledged wiki users change the wiki configuration as a side effect of their actions; if this were to be implemented, special care would be needed. [[--intrigeri]]

Actually I meant into any of the currently supported languages. I guess that if the template modification is made, it will list those languages on the page, and if a translation to a language is missing, the link will allow creating it?

Any translation page always exist for every supported slave language, even if no string at all have been translated yet. This implies the po plugin is especially friendly to people who prefer reading in their native language if available, but don't mind reading in English else.

While I'm at it, there is a remaining issue that needs to be sorted out: how painful it could be for non-English speakers (assuming the master language is English) to be perfectly able to navigate between translation pages supposed to be written in their own language, when their translation level is most often low.

(It is currently easy to display this status on the translation page itself, but then it's too late, and how frustrating to load a page just to realize it's actually not translated enough for you. The "other languages" loop also allows displaying this information, but it is generally not the primary navigation tool.)

IMHO, this is actually a social problem (i.e. it's no use adding a language to the supported slave ones if you don't have the manpower to actually do the translations), that can't be fully solved by technical solutions, but I can think of some hacks that would limit the negative impact: a given translation's status (currently = percent translated) could be displayed next to the link that leads to it; a color code could as well be used ("just" a matter of adding a CSS id or class to the links, depending on this variable). As there is already work to be done to have the links text generation more customizable through plugins, I could do both at the same time if we consider this matter to be important enough. --[[intrigeri]]

The translation status in links is now implemented in my pobranch. It requires my meta branch changes to work, though. I consider the latter to be mature enough to be merged. --[[intrigeri]]

FWIW, I'm tracking your po branch in ikiwiki master git in the po branch. One thing I'd like to try in there is setting up a translated basewiki, which seems like it should be pretty easy to do, and would be a great demo! --[[Joey]]

I have a complete translation of basewiki into danish, available merged into ikiwiki at git://source.jones.dk/ikiwiki-upstream (branch underlay-da), and am working with others on preparing one in german. For a complete translated user experience, however, you will also need templates translated (there are a few translatable strings there too). My most recent po4a Markdown improvements adopted upstream but not yet in Debian (see bug#530574) correctly handles multiple files in a single PO which might be relevant for template translation handling. --[[JonasSmedegaard]]

I've merged your changes into my own branch, and made great progress on the various todo items. Please note my repository location has changed a few days ago, my user page was updated accordingly, but I forgot to update this page at the same time. Hoping it's not too complicated to relocated an existing remote... (never done that, I'm a Git beginner as well as a Perl newbie) --[[intrigeri]]

Just a matter of editing .git/config, thanks for the heads up.

Joey, please have a look at my branch, your help would be really welcome for the security research, as I'm almost done with what I am able to do myself in this area. --[[intrigeri]]

I came up with a patch for the WrapI18N issue --[[Joey]]

I've set this plugin development aside for a while. I will be back and finish it at some point in the first quarter of 2009. --[[intrigeri]]

Abstract: Joey, please have a look at my po and meta branches.

Detailed progress report:

  • it seems the po branch in your repository has not been tracking my own po branch for two months. any config issue?
  • all the plugin's todo items have been completed, robustness tests done
  • I've finished the detailed security audit, and the fix for po4a bugs has entered upstream CVS last week
  • I've merged your new checkcontent hook with the cansave hook I previously introduced in my own branch; blogspam plugin updated accordingly
  • the rename hook changes we discussed elsewhere are also part of my branch
  • I've introduced two new hooks (canremove and canrename), not a big deal; IMHO, they extend quite logically the plugin interface
  • as highlighted on [[bugs/pagetitle_function_does_not_respect_meta_titles]], my meta branch contains a new feature that is really useful in a translatable wiki

As a conclusion, I'm feeling that my branches are ready to be merged; only thing missing, I guess, are a bit of discussion and subsequent adjustments.

--[[intrigeri]]

I've looked it over and updated my branch with some (untested) changes.

I've merged your changes into my branch. Only one was buggy.

Sorry, I'd forgotten about your cansave hook.. sorry for the duplicate work there.

Reviewing the changes, mostly outside of po.pm, I have the following issues.

  • renamepage to renamelink change would break the ikiwiki 3.x API, which I've promised not to do, so needs to be avoided somehow. (Sorry, I guess I dropped the ball on not getting this API change in before cutting 3.0..)

Fixed, see [[todo/need_global_renamepage_hook]].

  • I don't understand the parentlinks code change and need to figure it out. Can you explain what is going on there?

I'm calling bestlink there so that po's injected bestlink is run. This way, the parent links of a page link to the parent page version in the proper language, depending on the po_link_to=current and po_link_to=negotiated settings. Moreover, when using my meta branch enhancements plus meta title to make pages titles translatable, this small patch is needed to get the translated titles into parentlinks.

  • canrename's mix of positional and named parameters is way too ugly to get into an ikiwiki API. Use named parameters entirely. Also probably should just use named parameters for canremove.
  • skeleton.pm.example's canrename needs fixing to use either the current or my suggested parameters.

Done.

  • I don't like the exporting of %backlinks and $backlinks_calculated (the latter is exported but not used).

The commit message for 85f865b5d98e0122934d11e3f3eb6703e4f4c620 contains the rationale for this change. I guess I don't understand the subtleties of our use, and perldoc does not help me a lot. IIRC, I actually did not use our to "export" these variables, but rather to have them shared between Render.pm uses.

My wording was unclear, I meant exposing. --[[Joey]]

I guess I still don't know Perl's our enough to understand clearly. No matter whether these variables are declared with my or our, any plugin can use IkiWiki::Render and then access $IkiWiki::backlinks, as already does e.g. the pagestat plugin. So I guess your problem is not with letting plugins use these variables, but with them being visible for every piece of (possibly external) code called from Render.pm. Am I right? If I understand clearly, using a brace block to lexically enclose these two our declarations, alongside with the calculate_backlinks and backlinks subs definitions, would be a proper solution, wouldn't it? --[[intrigeri]]

No, %backlinks and the backlinks() function are not the same thing. The variable is lexically scoped; only accessible from inside Render.pm --[[Joey]]

  • What is this IkiWiki::nicepagetitle and why are you injecting it into that namespace when only your module uses it? Actually, I can't even find a caller of it in your module.

I guess you should have a look to my meta branch and to [[bugs/pagetitle_function_does_not_respect_meta_titles]] in order to understand this :)

It would probably be good if I could merge this branch without having to worry about also immediatly merging that one. --[[Joey]]

I removed all dependencies on my meta branch from the po one. This implied removing the po_translation_status_in_links and po_strictly_refresh_backlinks features, and every link text is now displayed in the master language. I believe the removed features really enhance user experience of a translatable wiki, that's why I was initially supposing the meta branch would be merged first. IMHO, we'll need to come back to this quite soon after po is merged. --[[intrigeri]]

Maybe you should keep those features in a meta-po branch? I did a cursory review of your meta last night, have some issues with it, but this page isn't the place for a detailed review. --[[Joey]]

Done. --[[intrigeri]]

  • I'm very fearful of the add_depends in indexhtml. Does this make every page depend on every page that links to it? Won't this absurdly bloat the dependency pagespecs and slow everything down? And since nicepagetitle is given as the reason for doing it, and nicepagetitle isn't used, why do it?

As explained in the 85f865b5d98e0122934d11e3f3eb6703e4f4c620 log: this feature hits performance a bit. Its cost was quite small in my real-world use-cases (a few percents bigger refresh time), but could be bigger in worst cases. When using the po plugin with my meta branch changes (i.e. the nicepagetitle thing), and having enabled the option to display translation status in links, this maintains the translation status up-to-date in backlinks. Same when using meta title to make the pages titles translatable. It does help having a nice and consistent translated wiki, but as it can also involve problems, I just turned it into an option.

This has been completely removed for now due to the removal of the dependency on my meta branch. --[[intrigeri]]

  • The po4a Suggests should be versioned to the first version that can be used safely, and that version documented in plugins/po.mdwn.

Done.

--[[intrigeri]]

--[[Joey]]

I reverted the %backlinks and $backlinks_calculated exposing. The issue they were solving probably will arise again when I'll work on my meta branch again (i.e. when the simplified po one is merged), but the po thing is supposed to work without these ugly our. Seems like it was the last unaddressed item from Joey's review, so I'm daring a timid "please pull"... or rather, please review again :) --[[intrigeri]]

Ok, I've reviewed and merged into my own po branch. It's looking very mergeable.

  • Is it worth trying to fix compatability with indexpages?

Supporting usedirs being enabled or disabled was already quite hard IIRC, so supporting all four combinations of usedirs and indexpages settings will probably be painful. I propose we forget about it until someone reports he/she badly needs it, and then we'll see what can be done.

  • Would it make sense to go ahead and modify page.tmpl to use OTHERLANGUAGES and PERCENTTRANSLATED, instead of documenting how to modify it?

Done in my branch.

  • Would it be better to disable po support for pages that use unsupported or poorly-supported markup languages?

I prefer keeping it enabled, as:

  • most wiki markups "almost work"
  • when someone needs one of these to be fully supported, it's not that hard to add dedicated support for it to po4a; if it were disabled, I fear the ones who could do this would maybe think it's blandly impossible and give up.
  • What's the reasoning behind checking that the link plugin is enabled? AFAICS, the same code in the scan hook should also work when other link plugins like camelcase are used.

That's right, fixed.

  • In pagetemplate there is a comment that claims the code relies on genpage, but I don't see how it does; it seems to always add a discussion link?

It relies on IkiWiki::Render's genpage as this function sets the discussionlink template param iff it considers a discussion link should appear on the current page. That's why I'm testing $template->param('discussionlink').

Maybe I was really wondering why it says it could lead to a broken link if the cgiurl is disabled. I think I see why now: Discussionlink will be set to a link to an existing disucssion page, even if cgi is disabled -- but there's no guarantee of a translated discussion page existing in that case. However, htmllink actually checks for this case, and will avoid generating a broken link so AFAICS, the comment is actually innacurate.. what will really happen in this case is discussionlink will be set to a non-link translation of "discussion". Also, I consider $config{cgi} and %links (etc) documented parts of the plugin interface, which won't change; po could rely on them to avoid this minor problem. --[[Joey]]

Done in my branch. --[[intrigeri]]

  • Is there any real reason not to allow removing a translation? I'm imagining a spammy translation, which an admin might not be able to fix, but could remove.

On the other hand, allowing one to "remove" a translation would probably lead to misunderstandings, as such a "removed" translation page would appear back as soon as it is "removed" (with no strings translated, though). I think an admin would be in a position to delete the spammy .po file by hand using whatever VCS is in use. Not that I'd really care, but I am slightly in favour of the way it currently works.

That would definitly be confusing. It sounds to me like if we end up needing to allow web-based deletion of spammy translations, it will need improvements to the deletion UI to de-confuse that. It's fine to put that off until needed --[[Joey]]

  • Re the meta title escaping issue worked around by change. I suppose this does not only affect meta, but other things at scan time too. Also, handling it only on rebuild feels suspicious -- a refresh could involve changes to multiple pages and trigger the same problem, I think. Also, exposing this rebuild to the user seems really ugly, not confidence inducing.

    So I wonder if there's a better way. Such as making po, at scan time, re-run the scan hooks, passing them modified content (either converted from po to mdwn or with the escaped stuff cheaply de-escaped). (Of course the scan hook would need to avoid calling itself!)

    (This doesn't need to block the merge, but I hope it can be addressed eventually..)

--[[Joey]]

I'll think about it soon.

--[[intrigeri]]

Did you get a chance to? --[[Joey]]

  • As discussed at [[todo/l10n]] the templates needs to be translatable too. They should be treated properly by po4a using the markdown option - at least with my later patches in bug#530574) applied.

  • It seems to me that the po plugin (and possibly other parts of ikiwiki) wrongly uses gettext. As I understand it, gettext (as used currently in ikiwiki) always lookup a single language, That might make sense for a single-language site, but multilingual sites should emit all strings targeted at the web output in each own language.

    So generally the system language (used for e.g. compile warnings) should be separated from both master language and slave languages.

    Preferrably the gettext subroutine could be extended to pass locale as optional secondary parameter overriding the default locale (for messages like "N/A" as percentage in po plugin). Alternatively (with above mentioned template support) all such strings could be externalized as templates that can then be localized.

Robustness tests

Enabling/disabling the plugin

  • enabling the plugin with po_translatable_pages set to blacklist: OK
  • enabling the plugin with po_translatable_pages set to whitelist: OK
  • enabling the plugin without po_translatable_pages set: OK
  • disabling the plugin: OK

Changing the plugin config

  • adding existing pages to po_translatable_pages: OK
  • removing existing pages from po_translatable_pages: OK
  • adding a language to po_slave_languages: OK
  • removing a language from po_slave_languages: OK
  • changing po_master_language: OK
  • replacing po_master_language with a language previously part of po_slave_languages: needs two rebuilds, but OK (this is quite a perverse test actually)

Creating/deleting/renaming pages

All cases of master/slave page creation/deletion/rename, both via RCS and via CGI, have been tested.

Misc

  • general test with usedirs disabled: OK
  • general test with indexpages enabled: not OK
  • general test with po_link_to=default with userdirs enabled: OK
  • general test with po_link_to=default with userdirs disabled: OK

Duplicate %links ?

I notice code in the scan hook that seems to assume that %links will accumulate duplicate links for a page. That used to be so, but the bug was fixed. Does this mean that po might be replacing the only link on a page, in error? --[[Joey]]

It would replace it. The only problematic case is when another plugin has its own reasons, in its scan hook, to add a page that is already there to $links{$page}. This other plugin's effect might then be changed by po's scan hook... which could be either good (better overall l10n) or bad (break the other plugin's goal). --[[intrigeri]]

Right.. well, the cases where links are added is very small. Grepping for add_link, it's just done by link, camelcase, meta, and tag. All of these are supposed to work just link regular links so I'd think that is ok. We could probably remove the currently scary comment about only wanting to change the first link. --[[Joey]]

Commit 3c2bffe21b91684 in my po branch does this. --[[intrigeri]]

Cherry-picked --[[Joey]]