| Commit message (Collapse) | Author | Age | Files | Lines |
|
|
|
|
|
|
|
|
|
|
| |
OpenStack was full of the add_copy_specs(string) anti-pattern.
While we should address this properly these instances all need
fixing to avoid collecting the whole file system.
They should also have been tested before being put forward as
a pull request.
Signed-off-by: Bryn M. Reeves <bmr@redhat.com>
|
|
|
|
|
|
|
|
| |
We already know we are running on RHEL or Fedora if we are
executing a plug-in class tagged with RedHatPlugin. Do not check
for /etc/*-release.
Signed-off-by: Bryn M. Reeves <bmr@redhat.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The OpenStack plug-in attempts to check for the presence of
binaries before running them. To do this it passes a relative
path to os.path.exists(); clearly this will never succeed.
Ideally plug-ins should never need to do these checks. If the
binary exists we'll run it and if not we won't (with no harmful
side-effects). This is how sos is currently designed and if there
is a need for plug-ins to ever do this we should add a method to
the base plug-in class e.g. Plugin.find_command() to do this
properly (i.e. evaluating the policy-supplied PATH).
For now, just make all of this unconditional.
Signed-off-by: Bryn M. Reeves <bmr@redhat.com>
|
|
|
|
|
|
|
|
| |
One call in openstack.py used the old addCopySpecs() method. This
will fail with current sos-3.x code. Update it to use the new
add_copy_specs() method instead.
Signed-off-by: Bryn M. Reeves <bmr@redhat.com>
|
|\
| |
| | |
cluster plugin: restrictions wrt. luci
|
| |
| |
| |
| |
| |
| |
| | |
Notably, avoid server cert being added. On the other hand, allow
collection of rotated log files for luci.
Signed-off-by: Jan Pokorný <jpokorny@redhat.com>
|
| |
| |
| |
| | |
Signed-off-by: Jan Pokorný <jpokorny@redhat.com>
|
|\ \
| | |
| | | |
Refactored OpenStack plugin
|
| | |
| | |
| | |
| | |
| | |
| | |
| | | |
contains all non-distribution specific details. Also dropped all
logrotate parts since those are collected by logrotate plugin.
Signed-off-by: Ante Karamatic <ante.karamatic@canonical.com>
|
| |/
| |
| |
| | |
Signed-off-by: Ante Karamatic <ante.karamatic@canonical.com>
|
| |
| |
| |
| | |
Signed-off-by: Pierguido Lambri <plambri@redhat.com>
|
|/
|
|
|
|
|
|
|
|
| |
Add an sos plug-in for openshift to gathers several directories
and command output sepcific to Openshift systems.
Passwords and secrets are scrubbed for privacy reasons.
Signed-off-by: Nick Harvey <niharvey at redhat dot com>
Signed-off-by: Bryn M. Reeves <bmr@redhat.com>
|
|\
| |
| | |
import tempfile into policies
|
| |
| |
| |
| | |
Signed-off-by: Adam Stokes <adam.stokes@ubuntu.com>
|
|\ \
| | |
| | | |
Make debian packaging 'non-native'
|
| |\ \ |
|
| | | |
| | | |
| | | |
| | | | |
Signed-off-by: Adam Stokes <adam.stokes@ubuntu.com>
|
| |_|/
|/| |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | | |
The SELinux plug-in can now optionally collect object lists from
the semanage command. Since this command loads the SELinux python
run time (~4s/command) it can add considerably to the overall run
time of sosreport - for this reason these commands are disabled by
default and may be enabled via the selinux.list option:
# sosreport -k selinux.list
With the option turned on we also collect:
semanage fcontext -l
semanage login -l
semanage port -l
semanage user -l
Signed-off-by: Bryn M. Reeves <bmr@redhat.com>
|
| | |
| | |
| | |
| | |
| | |
| | |
| | | |
Policy classes need to return the user-supplied temporary
directory if they decide not to override it.
Signed-off-by: Bryn M. Reeves <bmr@redhat.com>
|
| |/
|/|
| |
| |
| |
| |
| | |
We expect not to find all commands; don't output a log message on
each missing binary.
Signed-off-by: Bryn M. Reeves <bmr@redhat.com>
|
| |
| |
| |
| | |
Increment the spec file version to 3.0
|
| | |
|
| |
| |
| |
| |
| |
| |
| |
| | |
There are cases where we may attempt to add a symlink to the
archive more than once. Since this will fail for paths that
already exist test for their presence first.
Signed-off-by: Bryn M. Reeves <bmr@redhat.com>
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
The temporary trees used by sos are set up with correct
permissions but tarfile.open() uses the default mode and umask
settings.
Set the umask to deny group and other permissions for the archive
to avoid leaking data to unprivileged users.
Signed-off-by: Bryn M. Reeves <bmr@redhat.com>
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
Revert the use of POSIX PAX archive format in TarFileArchive. The
code to capture SELinux file contexts is currently disabled in
master due to the undesirable affects on archive extraction. Since
this is the main reason for using the PAX format revert to
automatic format selection to allow the widest compatibility.
Signed-off-by: Bryn M. Reeves <bmr@redhat.com>
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
As reported in Issue #140 plug-ins that need to write to an
arbitrary location in the archive are broken by the in-line tar
archive changes.
Introduce a 'get_tmp_dir()' method to the Archive classes that
must return a writable directory that is included in the archive.
Archives deriving from FileCacheArchive simply retun the root of
the archive tree.
Add a new method to Plugin to use this support, 'get_cmd_dir()',
that will return 'Archive.get_tmp_dir() + 'sos_commands'
+ self.name()'.
Fixes problems in lvm2, cloudforms, satellite and rhui.
Signed-off-by: Bryn M. Reeves <bmr@redhat.com>
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
Since some distributions may be using features like tmp-on-tmpfs
that can cause problems for sos data collection move the selection
of this directory into the policy class.
The new behaviour respects options passed on the command line but
will ignore any environment variables. Users wishing to override
the location of sos' temporary files should use the command line
option.
Signed-off-by: Bryn M. Reeves <bmr@redhat.com>
|
| |
| |
| |
| |
| |
| |
| |
| | |
Add a cleanup() method to the Archive classes and ensure that it
is called during normal and abnormal termination (unless the pdb
debugger has been invoked during --debug mode).
Signed-off-by: Bryn M. Reeves <bmr@redhat.com>
|
| |
| |
| |
| |
| |
| |
| | |
The logs module has a paramter 'logsize' but some code still uses
the old 'syslogsize' name. Make all uses consistent.
Signed-off-by: Bryn M. Reeves <bmr@redhat.com>
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
Add a new level to the Archive class hierarchy. This class may
be used by any archive class that needs to temporarily cache
files in the file system during collection.
Classes deriving from FileCacheArchive must supply
_build_archive() and _compress() methods which will be called
during archive finalization.
This fixes the TarFileArchive problems caused by attempting to
apply read-modify-write updates to content already present in the
archive.
Signed-off-by: Bryn M. Reeves <bmr@redhat.com>
|
| |
| |
| |
| |
| |
| |
| |
| | |
Policies may wish to influence the choice of temporary directory.
Move the TempFileUtil initialisation to be after the call to
load policy.
Signed-off-by: Bryn M. Reeves <bmr@redhat.com>
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
The logic in add_copy_spec_limit() only attempted to collect a
'tailed' version of a file that is over the size limit in the case
that it is the first file in the list of specs.
Change this to always collect the tailed version for the first
file to exceed the size limit and improve the variable naming
to be more consistent with related methods.
Signed-off-by: Bryn M. Reeves <bmr@redhat.com>
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
Pythons os.path.join behaviour returns only the last path given
when that path is absolute. This means that when make_path() is
given an absolute destination path (e.g. /etc/foo) it will return
an archive path of '/etc/foo' causing an error when copying into
the temporary tree.
Check for absolute paths and use lstrip to remove leading
instances of os.sep from the string.
Signed-off-by: Bryn M. Reeves <bmr@redhat.com>
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
Allow clients of the Archive classes to create arbitrary
directories. Sos needs to be able to do this to create the report
directory structure within a temporary file system tree.
This work also paves the way for improviyng the robustness of the
tree copying code in Plugin (which presently relies on the Archive
class to recurse up to root, adding parent directories that do not
already exist in the archive to ensure the corret permissions and
context are stored in the archive). This would also allow the
Archive.add_parent() hack to be removed.
Signed-off-by: Bryn M. Reeves <bmr@redhat.com>
|
| |
| |
| |
| |
| |
| |
| | |
Improve code formatting and readability in _set_archive() ahead
of new temporary tree archive code.
Signed-off-by: Bryn M. Reeves <bmr@redhat.com>
|
| |
| |
| |
| |
| |
| |
| | |
The RedHatSANLock initialiser used the wrong class name when
determining the parent class initialiser to invoke.
Signed-off-by: Bryn M. Reeves <bmr@redhat.com>
|
| |
| |
| |
| |
| |
| | |
s/)[/])/
Signed-off-by: Bryn M. Reeves <bmr@redhat.com>
|
|/
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The system plug-in tries to pass a naked string to add_copy_specs().
This causes each character of the string to be passed down to an
invocation of add_copy_spec():
add_copy_specs(copyspecs=/proc/sys, sub=None)
add_copy_spec(copyspec=/, sub=None)
add_copy_spec(copyspec=p, sub=None)
add_copy_spec(copyspec=r, sub=None)
[...]
Collecting '/' is clearly a bad idea (see Issue #141).
Signed-off-by: Bryn M. Reeves <bmr@redhat.com>
|
|\
| |
| | |
Update system.py, Typo
|
|/
|
| |
Typo, IPv6 not IPv8
|
|\
| |
| | |
Patch debian fixes
|
| |
| |
| |
| | |
Signed-off-by: Adam Stokes <adam.stokes@ubuntu.com>
|
|/
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
1) lintian information, you should run lintian with '-IE --pedantic',
lintian will tell you more if there are any problem. There are still
some lintian information, but I don't care of them. you may fix them
when you polish your package.
2) You use old format copyright files, it is better to switch to
DEP5[1] compatible format. lintian shows this information.
3) The section is python, but I think admin should be better, after
all, sosreport is a sysadmin tool
4) Please move LICENSE file to usr/share/doc/sosreport, rather than
remove it.
A note about 4. I have removed the actual installing of the LICENSE file
in the Makefile in favor of the distribution specific packaging handle it.
For example, within Debian everything is placed in debian/copyright which
removes the need for our Makefile to do anything with the LICENSE file.
Signed-off-by: Adam Stokes <adam.stokes@ubuntu.com>
|
|
|
|
| |
Signed-off-by: Bryn M. Reeves <bmr@redhat.com>
|
|
|
|
| |
Signed-off-by: Bryn M. Reeves <bmr@redhat.com>
|
|
|
|
|
|
|
|
| |
Move all the syslog collection from the general plug-in to a new
logs plug-in and factor out the all_logs support from the
RedHatLogs class into the Logs superclass.
Signed-off-by: Bryn M. Reeves <bmr@redhat.com>
|
|
|
|
|
|
|
|
|
|
| |
The kernel ring buffer is certainly more 'kernel' than 'general'.
Put it where it belongs and remove the duplicate dmesg call and
'dmesg_now' suggested filename (there's no need as it's in a
separate directory to the saved version at var/log/dmesg and the
normal action of dmesg is to print the current content).
Signed-off-by: Bryn M. Reeves <bmr@redhat.com>
|
|
|
|
|
|
|
|
| |
The hardware plug-in has been split up into separate ata, block,
pci, scsi, and usb plug-ins remove the data collected for these
subsystems from the hardware module.
Signed-off-by: Bryn M. Reeves <bmr@redhat.com>
|
|
|
|
|
|
|
| |
Add a new plug-in, processor, to capture CPU related information
and separate out all the CPU data from the old hardware plug-in.
Signed-off-by: Bryn M. Reeves <bmr@redhat.com>
|
|
|
|
|
|
| |
Separate out the scsi data from the hardware plug-in.
Signed-off-by: Bryn M. Reeves <bmr@redhat.com>
|