The code got removed by accident during the cleanup in commit 9c438bcac6. So
this patch is needed to bring it back and make things work like documented.
Capability dropping when changing the user in the system
mode was previously implemented by calling
prctl(PR_SET_KEEPCAPS, 0), but that doesn't necessarily
work. It's possible that the KEEPCAPS flag is locked to 1,
in which case the prctl() call fails with EPERM (this
happens at least on Harmattan). This patch implements
explicit capability dropping after changing the user.
This makes updating of permissions on existing directories optional with
pa_make_secure_dir() and pa_make_secure_parent_dir(). This makes sure
that the recursive directory creation doesn't end up modifying existing
directories, and also fixes a problem where creating an auth cookie
(specifically ~/.esd_auth) would end up modifying permissions on ~.
Thanks to Frédéric Danis for reporting this.
In most cases, we use dbus from more than one thread, as we
e.g. enable real-time scheduling from the ALSA threads.
Therefore set dbus to thread-safe mode by default, as recommended
in https://bugs.freedesktop.org/show_bug.cgi?id=47060#c5
This fixes a bug where PulseAudio could crash in two parallel
calls to pa_make_realtime.
BugLink: https://bugs.launchpad.net/bugs/937933
Signed-off-by: David Henningsson <david.henningsson@canonical.com>
This just covers Lennart's concern over the terminology used.
The majority of this change is simply the following command:
grep -rli sync[-_]volume . | xargs sed -i 's/sync_volume/deferred_volume/g;s/PA_SINK_SYNC_VOLUME/PA_SINK_DEFERRED_VOLUME/g;s/PA_SOURCE_SYNC_VOLUME/PA_SOURCE_DEFERRED_VOLUME/g;s/sync-volume/deferred-volume/g'
Some minor tweaks were added on top to tidy up formatting and
a couple of phrases were clarified too.
I don't know the exact cause for someone to submit a bug report for
this error message: if someone is truly offended by it (if so it is
a reminder that some people are more sensitive than others, and I do
want those people to feel welcome as well), or if it's a system's
check (if this goes through, it shows the system works, and the person
might put more work into his/her next patch), or if it's just a bug
(after all, it's not that weird to run two instances of PulseAudio?).
Either one could be reason enough to apply IMO.
Signed-off-by: David Henningsson <david.henningsson@canonical.com>
As various modules can subscribe to unlink callbacks unloading some modules
may trigger hooks in other modules.
The callbacks associated with these hooks could in turn need to use the core
in some capacity (e.g. perhaps they are module loading modules
(e.g. *-discover, filter-apply or gconf etc. and need to use the core to
unload modules they've loaded).
This change simply ensures that all modules and cached samples are unloaded
before freeing the core.
With Tanu's patch, the server no longer starts when a server is configured.
While this is sensible in most circumstances there is a corner case where
we still want to start.
In a typical X11 login, module-x11-publish will be loaded and will thus
set the PULSE_SERVER X11 property on the root window. This then hits the
check introduced in f1d1447e and exits. If PA had previously crashed
(thus leaving behind it's X11 properties) then this means that we will not
autospawn nor even allow ourselves to be started manually until
pax11publish -r is run to clear out the X11 properties. This is obviously
not desirable.
This patch introduces a more in-depth check of the server. If it looks like
a local unix domain socket, then we do not exit straight away and instead
probe further. This should not pose any problems with e.g. remote SSH
usage as the DBus Machine ID is used in the server string.
As spotted by Tanu Kaskinen:
The first process: daemon_pipe is not closed if the first fork() call
fails. Even if it doesn't fail, the first process never closes
daemon_pipe[0].
The second process: daemon_pipe[1] is not closed if anything fails
between the first and the second fork() call. Also, if the second fork
fails, then the finish section writes to daemon_pipe2[1], even though
only the third process should do that. Also, if anything fails between
the first and the second fork, then the second process never writes
anything to daemon_pipe[1]. I don't know what happens in the first
process in this case - does it get an error or does pa_loop_read() get
stuck.
The third process: No problems :)
The previous commit intoduced a double fork which caused a more or less immediate
successful return prior to the hard work of actually starting a daemon.
This patch simply used pipe() to only signal our father when the daemon really
has finished starting.
This adds volume scaling for 1- and 2-channel software volume scaling
using Orc. While testing the MMX and SSE backends on a Core2, I see an
~2x performance benefit over the hand-rolled MMX and SSE code. Since I
haven't been able to test on other architectures, the Orc code is only
used when MMX/SSE* is present. This can be changed in the future after
testing on AMD and ARM machines.
The check whether POSIX socket.h or WIN32 winsock2.h must be included can be
made centrally. The downside is that some functionality of e.g. arpa/inet.h is
also implemented in winsock.h, so that some files that don't use socket
functions, but do use inet.h functions, must also include pulsecore/socket.h.
(as well as arpa/inet.h)
Instead <pulsecore/poll.h> should be included. That file includes poll.h on
platform where it is appropriate. Also remove some unnecessary <ioctl.h>
includes.
This retains CPU information (processor type and supported features) in
pa_core, so that this information can be used by modules at init time to
figure out what optimisations may be used.
Lennart,
Apparently I was debugging this at the same time as you. I can't figure out
why my Fedora 11 install with glibc-2.10 has a glibc realpath that doesn't
match the gnu documentation and returns null. But it does.
Your commit aa8ce5bb9b almost fixed my
problem, but it needs a tweak.
Thanks,
David Yoder
There are multiple package management systems out there which implement
packages using symlinks. The recent (otherwise useful) check to ensure that
a re-executed pulseaudio is actually reexecuting itself unfortunately breaks
in the presence of all these packaging systems, because PA_BINARY refers
to its installed location (e.g. /usr/local/bin/pulseaudio), which is a
symlink to the binary (e.g. /usr/local/stow/pulseaudio-0.9.18/bin/pulseaudio),
because /proc/self/exe always contains the canonical path of the executable,
with all symlinks resolved.
(At least one distribution uses a symlink-based packaging system, so
will be forced to apply this locally in any case.)
The fix is simple: canonicalize PA_BINARY before equality-testing. (This
should be completely safe, because the OS does just that when PA_BINARY
is executed.)
The patch is against 0.9.18, but applies without fuzz to current master.
Get rid of the liboil dependency and reimplement the liboil functions with an
equivalent C implementation. Note that most of these functions are deprecated in
liboil and that none of them had any optimisations. We can further specialize
our handrolled versions for some extra speedups.
