[PATCH 0/1] pseudo 1.7.5

[PATCH 0/1] pseudo 1.7.5

[Peter Seebach]

tarball submitted, but this won't work without it... but you should
be able to use pseudo_git if you want to see the new shiny.

Long story short, there has always been a race condition if enough
clients were trying to simultaneously start a server, but that doesn't
come up much. But a while back, I added a "feature" where "pseudo <cmd>"
would try to shut the server down cleanly after running the command,
which is great for casual command-line use, but meant that we would
actually hit the race condition.

Respawn logic drastically reworked, various other changes made to
make this less likely to come up, and so on.

Also, pseudo will now abort() if the client can't get a server, which
could conceivably break things, *but*, it should only be breaking things
which have in fact been broken all along, we just didn't know it.

Also a bug fix for cases where mknod is called with only file permission
bits to create an ordinary file, which GNU tar does if and only if trying
to extract xattrs apparently (!?!??), and then a followup workaround for
a problem making ext4 filesystems, where I'm pretty sure the behavior was
right to begin with but this workaround lets things progress in the mean
time.

Whee.

The following changes since commit d53a0d410ae2390528f1090a131cd3e16d7ebcc5:

  glibc-locale: fix QA warning (2016-02-07 22:53:22 +0000)

are available in the git repository at:

  git://git.yoctoproject.org/poky-contrib seebs/pseudo175
  http://git.yoctoproject.org/cgit.cgi/poky-contrib/log/?h=seebs/pseudo175

Peter Seebach (1):
  pseudo: uprev to 1.7.5

 meta/recipes-devtools/pseudo/{pseudo_1.7.4.bb => pseudo_1.7.5.bb} | 4 ++--
 meta/recipes-devtools/pseudo/pseudo_git.bb                        | 4 ++--
 2 files changed, 4 insertions(+), 4 deletions(-)
 rename meta/recipes-devtools/pseudo/{pseudo_1.7.4.bb => pseudo_1.7.5.bb} (77%)

-- 
2.3.1

[PATCH 1/1] pseudo: uprev to 1.7.5

[Peter Seebach]

This uprev adds various improvements with regards to the server
spawn logic, and also sorts xattrs to work around a bug in one of
the mkfs utilities.

Signed-off-by: Peter Seebach <peter.seebach@windriver.com>
---
 meta/recipes-devtools/pseudo/{pseudo_1.7.4.bb => pseudo_1.7.5.bb} | 4 ++--
 meta/recipes-devtools/pseudo/pseudo_git.bb                        | 4 ++--
 2 files changed, 4 insertions(+), 4 deletions(-)
 rename meta/recipes-devtools/pseudo/{pseudo_1.7.4.bb => pseudo_1.7.5.bb} (77%)

diff --git a/meta/recipes-devtools/pseudo/pseudo_1.7.4.bb b/meta/recipes-devtools/pseudo/pseudo_1.7.5.bb
similarity index 77%
rename from meta/recipes-devtools/pseudo/pseudo_1.7.4.bb
rename to meta/recipes-devtools/pseudo/pseudo_1.7.5.bb
index d68e0af..195a508 100644
--- a/meta/recipes-devtools/pseudo/pseudo_1.7.4.bb
+++ b/meta/recipes-devtools/pseudo/pseudo_1.7.5.bb
@@ -6,8 +6,8 @@ SRC_URI = " \
     file://fallback-group \
 "
 
-SRC_URI[md5sum] = "6e4b59a346d08d4a29133c335ea12052"
-SRC_URI[sha256sum] = "f33ff84da328f943155f22cfd49030ef4ad85ad35fc2d9419a203521b65c384c"
+SRC_URI[md5sum] = "c10209938f03128d0c193f041ff3596d"
+SRC_URI[sha256sum] = "fd89cadec984d3b8202aca465898b1bb4350e0d63ba9aa9ac899f6f50270e688"
 
 PSEUDO_EXTRA_OPTS ?= "--enable-force-async --without-passwd-fallback"
 
diff --git a/meta/recipes-devtools/pseudo/pseudo_git.bb b/meta/recipes-devtools/pseudo/pseudo_git.bb
index eb666c0..abff1cf 100644
--- a/meta/recipes-devtools/pseudo/pseudo_git.bb
+++ b/meta/recipes-devtools/pseudo/pseudo_git.bb
@@ -1,7 +1,7 @@
 require pseudo.inc
 
-SRCREV = "3bc3909fa70535c2ef876009dc58e577b10a7e0e"
-PV = "1.7.4+git${SRCPV}"
+SRCREV = "786c6d3813622d18e12d36c4aa722af6a417c8fa"
+PV = "1.7.5+git${SRCPV}"
 
 DEFAULT_PREFERENCE = "-1"
 
-- 
2.3.1

Re: [PATCH 0/1] pseudo 1.7.5 [jethro][fido]

[Mark Hatle]

Once the corresponding tarball is available, this should likely be
backported/merged to Jethro and Fido as well.

Everything remains compatible, but it does fix a subtle race condition that can
lead to files under pseudo control inheriting their host system permissions.

See bug 9031 for more details.

--Mark

On 2/9/16 1:57 PM, Peter Seebach wrote:
> tarball submitted, but this won't work without it... but you should
> be able to use pseudo_git if you want to see the new shiny.
> 
> Long story short, there has always been a race condition if enough
> clients were trying to simultaneously start a server, but that doesn't
> come up much. But a while back, I added a "feature" where "pseudo <cmd>"
> would try to shut the server down cleanly after running the command,
> which is great for casual command-line use, but meant that we would
> actually hit the race condition.
> 
> Respawn logic drastically reworked, various other changes made to
> make this less likely to come up, and so on.
> 
> Also, pseudo will now abort() if the client can't get a server, which
> could conceivably break things, *but*, it should only be breaking things
> which have in fact been broken all along, we just didn't know it.
> 
> Also a bug fix for cases where mknod is called with only file permission
> bits to create an ordinary file, which GNU tar does if and only if trying
> to extract xattrs apparently (!?!??), and then a followup workaround for
> a problem making ext4 filesystems, where I'm pretty sure the behavior was
> right to begin with but this workaround lets things progress in the mean
> time.
> 
> Whee.
> 
> The following changes since commit d53a0d410ae2390528f1090a131cd3e16d7ebcc5:
> 
>   glibc-locale: fix QA warning (2016-02-07 22:53:22 +0000)
> 
> are available in the git repository at:
> 
>   git://git.yoctoproject.org/poky-contrib seebs/pseudo175
>   http://git.yoctoproject.org/cgit.cgi/poky-contrib/log/?h=seebs/pseudo175
> 
> Peter Seebach (1):
>   pseudo: uprev to 1.7.5
> 
>  meta/recipes-devtools/pseudo/{pseudo_1.7.4.bb => pseudo_1.7.5.bb} | 4 ++--
>  meta/recipes-devtools/pseudo/pseudo_git.bb                        | 4 ++--
>  2 files changed, 4 insertions(+), 4 deletions(-)
>  rename meta/recipes-devtools/pseudo/{pseudo_1.7.4.bb => pseudo_1.7.5.bb} (77%)
> 

Re: [PATCH 0/1] pseudo 1.7.5 [jethro][fido]

[Peter Seebach]

On Tue, 9 Feb 2016 14:11:09 -0600
Mark Hatle <mark.hatle@windriver.com> wrote:

> Everything remains compatible, but it does fix a subtle race condition that can
> lead to files under pseudo control inheriting their host system permissions.
> 
> See bug 9031 for more details.

For reference, the thing which exposed the circumstance where this was
actually remotely likely to come up was added in f1d4744, back in 2014,
and somehow the actual change isn't mentioned in ChangeLog.txt, but in
practice it's been there since 1.6.1. But it would only (probably) show
up if you ran some commands with "pseudo <cmd>" *while* running other
things using a pseudo server normally.

-s
-- 
Listen, get this.  Nobody with a good compiler needs to be justified.

Re: [PATCH 0/1] pseudo 1.7.5 [jethro][fido]

[Richard Purdie]

On Tue, 2016-02-09 at 15:16 -0600, Peter Seebach wrote:
> On Tue, 9 Feb 2016 14:11:09 -0600
> Mark Hatle <mark.hatle@windriver.com> wrote:
> 
> > Everything remains compatible, but it does fix a subtle race
> > condition that can
> > lead to files under pseudo control inheriting their host system
> > permissions.
> > 
> > See bug 9031 for more details.
> 
> For reference, the thing which exposed the circumstance where this
> was
> actually remotely likely to come up was added in f1d4744, back in
> 2014,
> and somehow the actual change isn't mentioned in ChangeLog.txt, but
> in
> practice it's been there since 1.6.1. But it would only (probably)
> show
> up if you ran some commands with "pseudo <cmd>" *while* running other
> things using a pseudo server normally.

Which we do as standard if you enable buildhistory.

(for full disclosure of the issue)

Cheers,

Richard

Re: [PATCH 0/1] pseudo 1.7.5 [jethro][fido]

[Andreas Müller]

On Tue, Feb 9, 2016 at 9:11 PM, Mark Hatle <mark.hatle@windriver.com> wrote:
> Once the corresponding tarball is available, this should likely be
> backported/merged to Jethro and Fido as well.
>
> Everything remains compatible, but it does fix a subtle race condition that can
> lead to files under pseudo control inheriting their host system permissions.
Are user IDs affected? If yes there would be a chance to reduce loads
of host contamination warnings..

Andreas

Re: [PATCH 0/1] pseudo 1.7.5 [jethro][fido]

[Mark Hatle]

On 2/9/16 3:52 PM, Andreas Müller wrote:
> On Tue, Feb 9, 2016 at 9:11 PM, Mark Hatle <mark.hatle@windriver.com> wrote:
>> Once the corresponding tarball is available, this should likely be
>> backported/merged to Jethro and Fido as well.
>>
>> Everything remains compatible, but it does fix a subtle race condition that can
>> lead to files under pseudo control inheriting their host system permissions.
> Are user IDs affected? If yes there would be a chance to reduce loads
> of host contamination warnings..

There are two pieces to this issue.

The first is the most common place people will see it.  Running a buildhistory,
the wrong file attributes are displayed.  These include xattr, owner, group and
permissions.  Note, in THIS case they are most likely correct in the package,
only buildhistory sees the wrong values.

The second is a lot less likely from what we saw, but IS possible, and we have
seen a few of these in real situations.  The problem occurs during a do_install
or other place and the wrong file attributes (see above) are inherited from the
environment.  In modern systems this, in my experience, always manifests itself
with a QA warning -- rebuilding the package the problem goes away.

Both of the above cases are believed to be fixed with the new version.

--Mark

> Andreas
> 

Re: [PATCH 0/1] pseudo 1.7.5 [jethro][fido]

[Peter Seebach]

On Tue, 9 Feb 2016 22:52:53 +0100
Andreas Müller <schnitzeltony@googlemail.com> wrote:

> Are user IDs affected? If yes there would be a chance to reduce loads
> of host contamination warnings..

They could be in some cases, especially with build history. In most cases,
packages were fine, but reporting could be wrong. But basically, it was
possible for stuff that should have been running under pseudo not to actually
be querying the database.

-s
-- 
Listen, get this.  Nobody with a good compiler needs to be justified.