RecvStoreAndBroadcastBlobURLRegistration allows blob URL registration under any origin
Categories
(Core :: DOM: Content Processes, defect)
Tracking
()
People
(Reporter: syd9363, Unassigned)
Details
(Keywords: ai-involved, reporter-external)
User Agent: Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/146.0.0.0 Safari/537.36
Steps to reproduce:
Built Firefox locally from source tag FIREFOX_BETA_149_BASE-4581-gc1e060fc9eab (mozilla-unified) using the following configuration:
mozconfig:
ac_add_options --enable-debug
ac_add_options --disable-optimize
ac_add_options --enable-tests
ac_add_options --disable-crashreporter
ac_add_options --disable-updater
mk_add_options MOZ_OBJDIR=@TOPSRCDIR@/../firefox-research/build/obj-debug
mk_add_options MOZ_MAKE_FLAGS=-j8
Build command:
cd targets/firefox
export MOZCONFIG=$(pwd)/../firefox-research/build/mozconfig-debug
./mach build
Reproduces in a new Firefox profile. Two browser tabs are needed for the full attack: one controlled by the attacker, one with the victim origin open.
- Open Firefox with two tabs:
https://victim.com(target) and any attacker-controlled page. - From the compromised renderer process, send a forged
StoreAndBroadcastBlobURLRegistrationIPC message with aContentPrincipalforvictim.com:
// Compromised renderer (debug build or post-RCE):
ContentChild::GetSingleton()->SendStoreAndBroadcastBlobURLRegistration(
"blob:https://victim.com/aaaaaaaa-bbbb-cccc-dddd-eeeeeeeeeeee"_ns,
attackerBlob,
ContentPrincipalInfo{ /* origin: https://victim.com */ },
""_ns
);
- From the
victim.comtab, fetch the forged blob URL:
fetch("blob:https://victim.com/aaaaaaaa-bbbb-cccc-dddd-eeeeeeeeeeee")
.then(r => r.text())
.then(t => console.log("Got:", t));
The relevant source is at dom/ipc/ContentParent.cpp:5836. Principal validation failure is log-only (ContentParent.cpp:1326–1337) — execution always reaches BlobURLProtocolHandler::AddDataEntry and BroadcastBlobURLRegistration.
Actual results:
The parent-side handler RecvStoreAndBroadcastBlobURLRegistration registered the blob under the forged victim.com principal and broadcast the registration to all content processes. Any code running in a victim.com context that fetches the blob URL receives attacker-controlled content, with the browser treating the response as same-origin to victim.com.
This is caused by ContentParent.cpp:5836 where validation failure has no return:
if (!ValidatePrincipal(aPrincipal, {ValidatePrincipalOptions::AllowSystem})) {
LogAndAssertFailedPrincipalValidationInfo(aPrincipal, __func__);
// NO return — registration proceeds regardless
}
BlobURLProtocolHandler::AddDataEntry(aURI, aPrincipal, ...); // stored under forged principal
BroadcastBlobURLRegistration(aURI, blobImpl, aPrincipal, ...); // broadcast to all processes
A compromised renderer can inject attacker-controlled content into any origin's blob URL namespace. A victim.com page that fetches the blob URL receives attacker data while believing it came from its own origin — same-origin checks pass. If a victim.com service worker caches the response, the cache is poisoned. The registration is broadcast to all open tabs.
Expected results:
RecvStoreAndBroadcastBlobURLRegistration should return IPC_FAIL when validation fails rather than continuing. It should also verify that the supplied principal's origin matches the origin of the sending content process. A renderer at evil.com should not be able to register a blob URL under victim.com. The blob should not be stored or broadcast if the principal does not match the process.
Updated•7 months ago
|
Updated•7 months ago
|
Updated•7 months ago
|
Description
•