On Sun, Mar 21, 2021 at 02:55:12PM +0100, Pierre Labastie via blfs-book wrote:
> On Fri, 2021-03-19 at 05:45 +0000, Ken Moffat via blfs-book wrote:
> > On Fri, Mar 19, 2021 at 12:04:56AM -0500, Douglas R. Reno via blfs-book
> > wrote:
> > > On Thu, Mar 18, 2021 at 11:33 PM ken--- via blfs-book <
> > > blfs-book@lists.linuxfromscratch.org> wrote:
> > >
> > > > Author: ken
> > > > Date: Thu Mar 18 21:32:57 2021
> > > > New Revision: 24382
> > > >
> > > > Log:
> > > > Update qtwebengine to 5.15.3 from git.
> > > >
> > > > This is a lot bigger and slower to build.
> > > >
> > > > Downgrade the 'Caution's in qtwebengien and falkon to 'Warning's.
> > > > I am not yet suggesting we should deprecate those two packages,
> > > > but users should start to ask themselves whether the want to use
> > > > a package where the upstreams are happy to stick with python2
> > > > and have no interest in running on current glibc.
> > > >
> > [...]
> > > > + <sect2 role="kernel" id="qtwebengine-kernel">
> > > > + <title>Kernel Configuration</title>
> > > > +
> > > > + <para>
> > > > + Enable the following options in the kernel configuration and
> > > > recompile the
> > > > + kernel if necessary:
> > > > + </para>
> > > > +
> > > > +<!-- Spaces are significant in <screen> sections -->
> > > > +<screen><literal>General setup --->
> > > > + -*- Namespaces support --->
> > > > + [*] UTS namespace [CONFIG_UTS_NS]
> > > > + [*] TIME namespace [CONFIG_TIME_NS]
> > > > + [*] IPC namespace [CONFIG_IPC_NS]
> > > > + [ ] User namespace #CONFIG_USER_NS is not
> > > > set
> > > > + [*] PID namespace [CONFIG_PID_NS]
> > > > + [*] Network namespace
> > > > [CONFIG_NET_NS]</literal></screen>
> > > > +
> > > > + <para>
> > > > + These are now the default options. Do <emphasis>NOT</emphasis>
> > > > enable
> > > > + User namespace (CONFIG_USER_NS), it <emphasis>will</emphasis>
> > > > cause
> > > > + libQtWebengineCore to crash.
> > > > + </para>
> > > > +
> > > > + <indexterm zone="qtwebengine qtwebengine-kernel">
> > > > + <primary sortas="d-qtwebengine">qtwebengine</primary>
> > > > + </indexterm>
> > > > + </sect2>
> > > >
> > > >
> > > This is going to be a problem. UPower requires CONFIG_USER_NS to function
> > > properly.
> > >
> >
> > Ouch. After adding '-config debug' to the qconf invocation (in
> > front of the '--') I was able to make a debug build (needs 25GB for
> > source and install). Trying to run falkon from gdb gave me a series
> > of backtraces (multiple threads).
> >
> > [26147:26147:0318/162343.989885:FATAL:sandbox_linux.cc(555)] Check failed:
> > NamespaceSandbox::InNewPidNamespace().
> >
> > bool SandboxLinux::EngageNamespaceSandboxInternal(bool from_zygote) {
> > CHECK(pre_initialized_);
> > CHECK(IsSingleThreaded())
> > << "The process cannot have multiple threads when engaging the
> > namespace
> > "
> > "sandbox, because the thread engaging the sandbox cannot ensure
> > that
> > "
> > "other threads close all their open directories.";
> >
> > if (from_zygote) {
> > // Check being in a new PID namespace created by the namespace sandbox
> > and
> > // being the init process.
> > CHECK(NamespaceSandbox::InNewPidNamespace());
> > ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
> > const pid_t pid = getpid();
> > CHECK_EQ(1, pid);
> > }
> >
> > Note that on one of my machines where this works I do not have PID
> > namespaces enabled at the moment, but those settings above are all
> > the defaults and I obviously never saw any reason to enable that
> > when making oldconfig.
> >
> > That failing code seems to have been called from
> >
> > [26137:26137:0318/162346.234871:FATAL:zygote_host_impl_linux.cc(204)] Check
> > failed: ReceiveFixedMessage(fds[0], kZygoteHelloMessage,
> > sizeof(kZygoteHelloMessage), &real_pid).
> >
> > But I was none the wiser about the rest of it, and after checking
> > that the chromium glibc-2.33 patch at Arch matched the qtwebengine
> > patch I tried turning off CONFIG_USER_NS just in case, and was
> > relieved whn falkon worked.
> >
> > > I'll spend some time looking at this when i'm done with video editing work
> > > (Monday, which is also when I'll start on GNOME-40 and other tickets).
> >
> > Sorry about that. But I'm surprised - the kernel help says (end of
> > lines trimmed for length)
> >
> > CONFIG_USER_NS:
> > │
> > │ This allows containers, i.e. vservers, to use user namespaces
> > │ to provide different user info for different servers.
> > │
> > │ When user namespaces are enabled in the kernel it is
> > │ recommended that the MEMCG option also be enabled and that
> > │ user-space use the memory control groups to limit the amount
> > │ of memory a memory unprivileged users can use.
> > │
> > │ If unsure, say N.
> >
>
> FWIIW, i've built qtwebengine 5.15.3 by the book, and I have CONFIG_USER_NS
> set
> in my kernel config. Using the previously compiled falkon (against 5.15.2),
> I've
> not seen any problem (not sure I've tried many things, videos on youtube, and
> random web sites). I've been able to compile falkon against 5.15.3, and to
> install it. Still no failure up to now.
>
> Pierre
>
If it works, it works - my problem was that it crashed when
starting.
When falkon failed to run, and clearing out the cache had not
helped, I tried rebuilding it (no change) and then I built both
falkon and webengine with debugging enabled. Then I went back to
the altready-rebuilt non-debug version. So on this system this is a
fresh build of falkon.
But I still think *I'm* missing something: on my old 10.0 systems
(and 9.1) it built and ran without problems, on 10.1 the builds were
nothing but trouble even after adding a python symlink. And the
code where it crashed was not obviously related to USER_NS (possibly
for PID_NS, but one of my machines does not have that enabled and
still works).
If it is working for you with USER_NS enabled then the whole kernel
config section, and the reference to it in the previous section, can
be removed.
ĸen
--
On average, the Panda feeds for 15 hours a day. This is the
same as an adult at home under quarantine, which is why we call
it a "Pandemic".
--
http://lists.linuxfromscratch.org/listinfo/blfs-book
FAQ: http://www.linuxfromscratch.org/blfs/faq.html
Unsubscribe: See the above information page
