1318012 - SandboxBrokerCommon::SendWithFd sends uninitialised stack-allocated data out of process

Status: RESOLVED FIXED

(Core :: Security: Process Sandboxing, defect)

Description

[Julian Seward [:jseward]]

Attachment: Valgrind complaints

I ran gtests on valgrind and noticed that SandboxBrokerCommon::SendWithFd can
send uninitialised stack allocated data out of process.  I suspect this is due
to alignment holes in char cmsg_buf[] in that function.

Partly because it seems ungood to have an information leak, and partly
to stop valgrind complaining, I'd prefer to zero initialise that array.
It's only 24 bytes, at least on x86_64-linux.

Comment 1

[Julian Seward [:jseward]]

Attachment: bug1318012-1.cset

A possible fix.

Comment 2

[Julian Seward [:jseward]]

Running gtests on valgrind is a bit tricky.  STR:

run valgrind-listener in a different shell.  This listens on port 1500.

export MOZVFLAGS="--num-transtab-sectors=24 --px-default=allregs-at-mem-access \
  --px-file-backed=unwindregs-at-mem-access --fair-sched=yes \
  --fullpath-after=/MOZ/ --trace-children=yes --show-mismatched-frees=no"

DISPLAY=:1.0 G_SLICE=always-malloc v312BRANCH $MOZVFLAGS \
  "--trace-children-skip=/usr/bin/which,/usr/bin/file,/usr/bin/uname,/sbin/ldconfig,/usr/bin/cat,/usr/bin/awk,/usr/bin/sed,/usr/bin/whoami,/usr/bin/true,/*/bin/gcc,/usr/bin/gmake" \
  --track-origins=yes --log-socket=127.0.0.1:1500 --num-callers=40 \
  ./mach gtest

Comment 3

[Julian Seward [:jseward]]

Attachment: bug1318012-2.cset

Comment 4

[Carsten Book [:Tomcat]]

https://hg.mozilla.org/mozilla-central/rev/55d9f54c1a58

Comment 5

[Jed Davis [:jld] ⟨⏰|UTC-7⟩ ⟦he/him⟧]

The sec-* keywords on this bug might be a bit on the high side — at worst it's an information leak of a few bytes (and I haven't verified that the bytes actually are copied to the other process), it always reads from the same place on the broker thread's stack, _and_ it needs to be combined with a sec-critical to do anything.

Comment 6

[Ryan VanderMeulen [:RyanVM]]

Should we consider backporting this to 52 for the next ESR?

Comment 7

[Jed Davis [:jld] ⟨⏰|UTC-7⟩ ⟦he/him⟧]

Thanks for checking, but there's no benefit to uplifting this.  The file broker is used only for content processes, and content sandboxing won't be enabled for non-nightly builds until 53 at the earliest.  (I'm not sure if "unaffected" is the right flag, because nightly 52 *was* affected, but aurora/beta/release/etc. 52 aren't.)

Comment 8

[Ryan VanderMeulen [:RyanVM]]

We generally use the disabled status for features that aren't riding that train. Thanks!