Bug 154837

Summary: Modern IDB: WebKit 2 IPC layer easily confused about multiple web processes being connected
Product: WebKit Reporter: Brady Eidson <beidson>
Component: WebKit2Assignee: Brady Eidson <beidson>
Status: RESOLVED FIXED    
Severity: Normal CC: achristensen
Priority: P2 Keywords: InRadar
Version: Safari 9   
Hardware: Unspecified   
OS: Unspecified   
Attachments:
Description Flags
Patch v1 achristensen: review+

Brady Eidson
Reported 2016-02-29 14:23:49 PST
Modern IDB: WebKit 2 IPC layer easily confused about multiple web processes being connected Can result in weird behavior, including (most obviously) ASSERTs in debug builds. <rdar://problem/24857351>
Attachments
Patch v1 (22.86 KB, patch)
2016-02-29 14:26 PST, Brady Eidson 		achristensen: review+
DetailsFormatted DiffDiff
View All Add attachment proposed patch, testcase, etc.
Brady Eidson
Comment 1 2016-02-29 14:26:40 PST
Created attachment 272518 [details] Patch v1
Alex Christensen
Comment 2 2016-02-29 14:50:55 PST
Comment on attachment 272518 [details] Patch v1 View in context: https://bugs.webkit.org/attachment.cgi?id=272518&action=review > Source/WebKit2/DatabaseProcess/DatabaseToWebProcessConnection.cpp:99 > +static uint64_t generateConnectionToServerIdentifier() It makes sense to have this in the DatabaseProcess. Do we have any protection against invalid identifiers being used if the DatabaseProcess crashes? > Source/WebKit2/WebProcess/Databases/IndexedDB/WebIDBConnectionToServer.cpp:62 > + m_isOpenInServer = sendSync(Messages::DatabaseToWebProcessConnection::EstablishIDBConnectionToServer(), m_identifier); Why can't we send asynchronously?
Brady Eidson
Comment 3 2016-02-29 15:00:19 PST
(In reply to comment #2) > Comment on attachment 272518 [details] > Patch v1 > > View in context: > https://bugs.webkit.org/attachment.cgi?id=272518&action=review > > > Source/WebKit2/DatabaseProcess/DatabaseToWebProcessConnection.cpp:99 > > +static uint64_t generateConnectionToServerIdentifier() > > It makes sense to have this in the DatabaseProcess. Do we have any > protection against invalid identifiers being used if the DatabaseProcess > crashes? DatabaseProcess goes away, all related objects in the web process are invalidated. > > Source/WebKit2/WebProcess/Databases/IndexedDB/WebIDBConnectionToServer.cpp:62 > > + m_isOpenInServer = sendSync(Messages::DatabaseToWebProcessConnection::EstablishIDBConnectionToServer(), m_identifier); > > Why can't we send asynchronously? New connections can start queuing up operations in the WebProcess via javascript immediately. This is at most one sync message per tab without any complicated/long-running work behind it. NBD.
Alex Christensen
Comment 4 2016-02-29 15:08:05 PST
Comment on attachment 272518 [details] Patch v1 View in context: https://bugs.webkit.org/attachment.cgi?id=272518&action=review >>> Source/WebKit2/DatabaseProcess/DatabaseToWebProcessConnection.cpp:99 >>> +static uint64_t generateConnectionToServerIdentifier() >> >> It makes sense to have this in the DatabaseProcess. Do we have any protection against invalid identifiers being used if the DatabaseProcess crashes? > > DatabaseProcess goes away, all related objects in the web process are invalidated. It makes sense to have this in the DatabaseProcess. Do we have any protection against invalid identifiers being used if the DatabaseProcess crashes? >>> Source/WebKit2/WebProcess/Databases/IndexedDB/WebIDBConnectionToServer.cpp:62 >>> + m_isOpenInServer = sendSync(Messages::DatabaseToWebProcessConnection::EstablishIDBConnectionToServer(), m_identifier); >> >> Why can't we send asynchronously? > > New connections can start queuing up operations in the WebProcess via javascript immediately. > > This is at most one sync message per tab without any complicated/long-running work behind it. NBD. Why can't we send asynchronously?
Brady Eidson
Comment 5 2016-02-29 15:13:48 PST
http://trac.webkit.org/changeset/197372
Note You need to log in before you can comment on or make changes to this bug.