975387 - TextComposition should clear mPresContext when the PresContext is destroyed

Status: RESOLVED FIXED

(Core :: DOM: UI Events & Focus Handling, defect)

Description

[Masayuki Nakano [:masayuki] (he/him)(JST, +0900)]

This may be dangerous.

The TextComposition::mPresContext may be referred after destroyed.

We should fix this bug on Aurora too.

Comment 1

[Masayuki Nakano [:masayuki] (he/him)(JST, +0900)]

Attachment: Patch

Comment 2

[Masayuki Nakano [:masayuki] (he/him)(JST, +0900)]

Attachment: Patch

I realized this bug when I was writing a patch for another bug. Therefore, I'm not sure this bug actually causes crash or something.

At compositionstart event handler, JS may do something to destroy its presContext. Then, mPresContext may be already released.

However, when this method is called, the caller's PresShell is held by nsViewManager with local variable:
http://mxr.mozilla.org/mozilla-central/source/view/src/nsViewManager.cpp#751

And PresShell::mPresContext must be alive while PresShell instance is alive because its destructor releases mPresContext:
http://mxr.mozilla.org/mozilla-central/source/layout/base/nsPresShell.cpp#769

Although, I don't know the reason why PresShell checks mPresContext at a lot of plases...

So, I think that this is safe, but may not be so in the future. So, we should make a wall with this patch.

Comment 3

[Masayuki Nakano [:masayuki] (he/him)(JST, +0900)]

Comment on attachment 8379751 [details] [diff] [review]
Patch

[Security approval request comment]
How easily could an exploit be constructed based on the patch?

As I said in my previous comment, the mPresContext is always held luckily.

Do comments in the patch, the check-in comment, or tests included in the patch paint a bulls-eye on the security problem?

The check-in comment might be so. I'd like you to suggest other commit message if you think it's bad.

Which older supported branches are affected by this flaw?

Aurora has this bug too. Beta and Release don't have this.

If not all supported branches, which bug introduced the flaw?

The referencing code of mPresContext after dispatching an event is added by bug 935821.

Do you have backports for the affected branches? If not, how different, hard to create, and risky will they be?

It's easy. I'll post the patch for Aurora soon.

How likely is this patch to cause regressions; how much testing does it need?

Even if this causes regressions, it's only affected to e10s. But e10s hasn't supported IME completely yet.

Comment 4

[Masayuki Nakano [:masayuki] (he/him)(JST, +0900)]

Attachment: Patch for Aurora

[Approval Request Comment]
Bug caused by (feature/regressing bug #): Bug 935821
User impact if declined: As far as I researched, this doesn't cause actual trouble. But this makes a wall for the dangerous scenario.
Testing completed (on m-c, etc.): Not yet landed on m-c. Waiting sec-approval.
Risk to taking this patch (and alternatives if risky): No idea. After nsIMEStateManager stops managing TextComposition instances, they shouldn't be used for dispatching any events. (Others cannot retrieve the destroyed instance in expected scenarios)
String or IDL/UUID changes made by this patch: Nothing.

Comment 5

[Masayuki Nakano [:masayuki] (he/him)(JST, +0900)]

Comment on attachment 8379751 [details] [diff] [review]
Patch

Hmm, I'm sorry. I found this patch is not enough.

Comment 6

[Masayuki Nakano [:masayuki] (he/him)(JST, +0900)]

Comment on attachment 8380472 [details] [diff] [review]
Patch for Aurora

cancelling the request because I found that this is not enough.

Comment 7

[Masayuki Nakano [:masayuki] (he/him)(JST, +0900)]

Comment on attachment 8379751 [details] [diff] [review]
Patch

Sorry for the spam. I misunderstood as that nsIMEStateManager::DispatchCompositionEvent() will create new composition with destroyed PresContext. However, I confirmed that my concern is wrong.

So, this patch's wall works as I expected.

Please see comment 3 for the sec-approval.

Comment 8

[Masayuki Nakano [:masayuki] (he/him)(JST, +0900)]

Comment on attachment 8380472 [details] [diff] [review]
Patch for Aurora

Sorry for the spam.

See comment 4 and comment 7. This patch is enough for avoiding the reported bug completely.

Comment 9

[Masayuki Nakano [:masayuki] (he/him)(JST, +0900)]

Anyway, we need to redesign the cases at destroying a document from composition event handlers. Even with these patches, IME state is still confused even in the next focused editor.

Comment 10

[Al Billings [:abillings - ex-MoCo]]

Comment on attachment 8379751 [details] [diff] [review]
Patch

sec-approval+ for trunk. I'm not concerned about the commit message as long as we take this on Aurora as well.

Comment 11

[Masayuki Nakano [:masayuki] (he/him)(JST, +0900)]

(In reply to Al Billings [:abillings] from comment #10)
> sec-approval+ for trunk. I'm not concerned about the commit message as long
> as we take this on Aurora as well.

Thank you for the sec-review.

You said "sec-approval+ for trunk", but you gives + for approval-mozilla-aurora too. Can I land them now? Or do I need to request sec-review for tha patch for Aurora again?

Comment 12

[Andrew McCreight [:mccr8]]

Once you have sec-approval for trunk, the patches can be landed anywhere, as long as they go through the usual approval process.  So the patch can be landed immediately on trunk and Aurora.  The idea is that once a patch lands on any branch, a bad guy could see it, so it doesn't matter much from a security perspective if it lands on other branches, too.

Comment 13

[Masayuki Nakano [:masayuki] (he/him)(JST, +0900)]

Thank you, I'll land them after lunch.

Comment 14

[Masayuki Nakano [:masayuki] (he/him)(JST, +0900)]

https://hg.mozilla.org/integration/mozilla-inbound/rev/f2fcc1dea688
https://hg.mozilla.org/releases/mozilla-aurora/rev/f8ec94d3ab7c

Comment 15

[Carsten Book [:Tomcat]]

landed on https://hg.mozilla.org/mozilla-central/rev/f2fcc1dea688

Comment 16

[Ryan VanderMeulen [:RyanVM]]

The tree rules are clear that you are not to double-land patches on Aurora and inbound. Please wait until the patch has stuck on trunk before uplifting. Also, when landing on the release branches, please set the status flag appropriately.

Comment 17

[Masayuki Nakano [:masayuki] (he/him)(JST, +0900)]

(In reply to Ryan VanderMeulen [:RyanVM UTC-5] from comment #16)
> The tree rules are clear that you are not to double-land patches on Aurora
> and inbound. Please wait until the patch has stuck on trunk before
> uplifting. Also, when landing on the release branches, please set the status
> flag appropriately.

I see. Thank you for your explanation.

Comment 18

[Al Billings [:abillings - ex-MoCo]]

Yes, generally "sec-approval+" is an approval for trunk landing. After it lands on trunk, you can get approval for any branch patches as necessary. We want it to go green on trunk though.

Comment 19

[Matt Wobensmith [:mwobensmith][:matt:]]

Marking [qa-] due to lack of test case, but feel free to provide one if you'd like verification. Thanks.