700512 - Workers + Files exposes threadsafety assertions with DataOwner

Status: RESOLVED FIXED

(Core :: DOM: Core & HTML, defect)

Description

[Kyle Huey (Exited; not receiving bugmail, old account, do not use)]

Jonas reports that bent has seen threadsafety assertions from DataOwner when manipulating File objects in Workers in certain ways.

Comment 1

[Jonas Sicking (:sicking) No longer reading bugmail consistently]

Yeah, we need to make the DataOwner refcounting threadsafe. As things stand now, if a memory-blob is used on both the main thread and a worker thread, the refcount can end up both too high and too low due to races.

Comment 2

[Kyle Huey (Exited; not receiving bugmail, old account, do not use)]

Attachment: Patch

Comment 3

[Kyle Huey (Exited; not receiving bugmail, old account, do not use)]

Attachment: Patch

Comment 4

[Jonas Sicking (:sicking) No longer reading bugmail consistently]

Comment on attachment 572820 [details] [diff] [review]
Patch

Err.. you only want to delete if the refcount goes to 0, right?

Comment 5

[Kyle Huey (Exited; not receiving bugmail, old account, do not use)]

Attachment: Patch

Comment 6

[Kyle Huey (Exited; not receiving bugmail, old account, do not use)]

If we're respinning 8.0, perhaps we should consider including this?  It's a pretty easy fix to a new bug in 8.0.

Comment 7

[Alex Keybl [:akeybl]]

[Triage Comment]
This is sg:high and therefore doesn't meet the criteria for landing on the release branch for a chemspill.

Comment 8

[Kyle Huey (Exited; not receiving bugmail, old account, do not use)]

Review ping.

Comment 9

[Ben Turner (not reading bugmail, use the needinfo flag!)]

Wait... I don't like that this doesn't stabilize the refcount. Having inconsistent implementations of Release seems like a recipe for headaches, and having no stabilization seems like a recipe for double-free.

Comment 10

[Benjamin Smedberg]

refcount stabilization was a terrible idea. We should assert if you try to addref a dying object, not propagate stabilization to new macros where it's not already potentially used.

Comment 11

[Ben Turner (not reading bugmail, use the needinfo flag!)]

Fatally asserting sounds fine, but I imagine we'd only do that in debug builds since you have to have some other state bit to track that. For opt builds we'd just have no protection against double frees?

Comment 12

[Benjamin Smedberg]

You really don't need another state bit, just use PR_UINT32_MAX as a sentinel value. I'm not sure that the extra branch is worth the protection in release builds, but we should probably measure to see how much it actually costs.

Comment 13

[Kyle Huey (Exited; not receiving bugmail, old account, do not use)]

Checked in without a test case.

https://hg.mozilla.org/mozilla-central/rev/d20b9d389776

Comment 14

[Kyle Huey (Exited; not receiving bugmail, old account, do not use)]

Comment on attachment 572824 [details] [diff] [review]
Patch

Requesting aurora and beta approval for this security regression introduced in Firefox 8.

Comment 15

[LegNeato]

Comment on attachment 572824 [details] [diff] [review]
Patch

[triage comment]
Please land on aurora and beta today if at all possible.

Comment 16

[Kyle Huey (Exited; not receiving bugmail, old account, do not use)]

https://hg.mozilla.org/releases/mozilla-aurora/rev/a45dd5ac1f90
https://hg.mozilla.org/releases/mozilla-beta/rev/6858e50813de

Comment 17

[Daniel Veditz [:dveditz]]

Does this affect the 1.9.2 branch? (qawanted)

Comment 18

[Kyle Huey (Exited; not receiving bugmail, old account, do not use)]

No, File objects are main thread only in 1.9.2.

Comment 19

[u279076]

Is there something QA can do to verify this fix?

Comment 20

[Kyle Huey (Exited; not receiving bugmail, old account, do not use)]

The tests for this landed in Bug 698621 for reasons that I don't understand.