285991 - leak loading site from HTTPS

Status: RESOLVED FIXED

(Core :: Networking, defect, P1)

Description

[David Baron :dbaron: (⌚️UTC-4, no longer working on Mozilla)]

I leak a good bit of stuff when loading something over https, e.g.,
https://bugzilla.mozilla.org/show_bug.cgi?id=285943 .

The problem seems to be a cycle:
 * An nsNSSSocketInfo object's mCallbacks member owns (via XPCOM proxy code) an
nsHttpConnection object
 * The nsHttpConnection object's mSocketTransport, mSocketIn, and mSocketOut
members own an nsSocketTransport (and its internal objects which aggregate its
refcount)
 * The nsSocketTransport object's mSecInfo member owns the nsNSSSocketInfo.

Comment 1

[David Baron :dbaron: (⌚️UTC-4, no longer working on Mozilla)]

Attachment: edited leaksoup output showing cycle

Comment 2

[Darin Fisher]

I think we should be null'ing out nsSocketTransport::mSecInfo at some point. 
Or, maybe since consumers may want that object after the socket has closed, it
would be better to call SetNotificationCallbacks(null) once we are closing the
connection (in OnSocketDetached most likely).

Comment 3

[Darin Fisher]

Attachment: v1 patch

Comment 4

[Darin Fisher]

biesi points out on IRC that this patch makes it so that consumers would no
longer be able to access security info for a socket after the socket has closed.
 that is probably a legitimate concern given that a consumer might not have a
chance to call GetSecurityInfo before the socket is closed in the case of a
protocol where the client writes nothing to the socket and only reads a small
chunk of data from the server before the server closes the socket.

i think we may need to look into another way to break this cycle.

Comment 5

[Christian :Biesinger (don't email me, ping me on IRC)]

Comment on attachment 179775 [details] [diff] [review]
v1 patch

please document in nsISocketTransport that the securityInfo can't be accessed
after the socket has closed

Comment 6

[Darin Fisher]

Attachment: v2 patch

Comment 7

[Christian :Biesinger (don't email me, ping me on IRC)]

Comment on attachment 179791 [details] [diff] [review]
v2 patch

ok, I like this much better, thanks.

Comment 8

[David Baron :dbaron: (⌚️UTC-4, no longer working on Mozilla)]

Comment on attachment 179791 [details] [diff] [review]
v2 patch

sr=dbaron.

(Another possibility is perhaps using weak references somewhere...)

Comment 9

[Darin Fisher]

fixed-on-trunk