Tom Hromatka [Mon, 12 Apr 2021 15:25:00 +0000 (09:25 -0600)]
wrapper.c: Fix fprintf argument warning
Fix a warning in cgroup_add_value_string() where '%d' was
being passed to fprintf when it should have been using '%ld'.
wrapper.c: In function ‘cgroup_add_value_string’:
wrapper.c:207:51: warning: format ‘%d’ expects argument of type ‘int’,
but argument 3 has type ‘long unsigned int’ [-Wformat=]
207 | fprintf(stderr, "value exceeds the maximum of %d
characters\n",
| ~^
| |
| int
| %ld
208 | sizeof(cntl_value->value) - 1);
| ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
| |
| long unsigned int
Signed-off-by: Tom Hromatka <tom.hromatka@oracle.com>
Tom Hromatka [Mon, 12 Apr 2021 15:18:56 +0000 (09:18 -0600)]
api.c: Fix string truncation warnings
Fix two snprintf string truncation warnings in
cg_build_path_locked().
api.c:1475:38: warning: ‘snprintf’ output may be truncated before the
last format character [-Wformat-truncation=]
1475 | snprintf(path, FILENAME_MAX, "%s/",
| ^
api.c:1475:5: note: ‘snprintf’ output between 2 and 4097 bytes into a
destination of size 4096
1475 | snprintf(path, FILENAME_MAX, "%s/",
| ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1476 | cg_mount_table[i].mount.path);
| ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
api.c:1470:40: warning: ‘/’ directive output may be truncated writing 1
byte into a region of size between 0 and 4095 [-Wformat-truncation=]
1470 | snprintf(path, FILENAME_MAX, "%s/%s/",
| ^
api.c:1470:5: note: ‘snprintf’ output 3 or more bytes (assuming 4098)
into a destination of size 4096
1470 | snprintf(path, FILENAME_MAX, "%s/%s/",
| ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1471 | cg_mount_table[i].mount.path,
| ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1472 | cg_namespace_table[i]);
| ~~~~~~~~~~~~~~~~~~~~~~
Signed-off-by: Tom Hromatka <tom.hromatka@oracle.com>
Tom Hromatka [Wed, 31 Mar 2021 16:44:11 +0000 (16:44 +0000)]
tools: Create a cgset library for testing purposes
Create libcgset.la so that functions within cgset can be
unit tested.
A functional test would be preferable in this case, but I'm
looking to test a very specific scenario - where there's an
'=' sign in the value of a setting's name/value pair. To the
best of my knowledge, only the rdma (v1) and io (v2)
controllers have settings that utilize an '=' in such a fashion.
I can't think of an easy way to test rdma across our test fleet,
and our automated systems don't support the v2 io controller
yet. Thus a unit test for now.
Signed-off-by: Tom Hromatka <tom.hromatka@oracle.com>
Tom Hromatka [Tue, 30 Mar 2021 18:54:06 +0000 (18:54 +0000)]
cgset: Fix parsing of name/values with an '=' in the value
Fix parsing of name/value pairs that contain an '='
character in the value string. For example, the io.max
setting utilizes the '=' character:
cgset -r io.max="8:16 wbps=1024" foo
Fixes: https://github.com/libcgroup/libcgroup/issues/33 Reported-by: Santiago Ruano Rincón <santiago.ruano-rincon@imt-atlantique.fr> Signed-off-by: Tom Hromatka <tom.hromatka@oracle.com>
Tom Hromatka [Mon, 29 Mar 2021 16:46:52 +0000 (16:46 +0000)]
bootstrap: github: Ignore systemd hierarchy
Ignore the "name=systemd" hierarchy by setting the autoconf
--enable-opaque-hierarchy flag equal to "name=systemd". This
instructs libcgroup to ignore mounts named "name=systemd" and
thus these mounts aren't added to the cg_mount_table[].
Suggested-by: Michal Koutný <mkoutny@suse.com> Signed-off-by: Tom Hromatka <tom.hromatka@oracle.com>
Tom Hromatka [Wed, 10 Mar 2021 18:20:24 +0000 (18:20 +0000)]
api.c: Don't fail a recursive write if value isn't dirty
When cgroup_modify_cgroup() invokes cgroup_set_values_recursive(),
some of the settings within the cgroup may not be writable. Avoid
failing the entire write by ignoring write failures on settings that
do not explicitly have the dirty flag set.
Signed-off-by: Tom Hromatka <tom.hromatka@oracle.com>
Tom Hromatka [Wed, 10 Feb 2021 17:25:10 +0000 (17:25 +0000)]
cgget: Major refactor
Previously cgget used data structures unlike anywhere else
within libcgroup. cgget had arrays of strings to hold the
list of controllers, settings, and values.
Update cgget to utilize struct cgroup and struct
cgroup_controller.
This patch is intended to have no functional changes.
Signed-off-by: Tom Hromatka <tom.hromatka@oracle.com>
Tom Hromatka [Mon, 8 Feb 2021 23:05:24 +0000 (16:05 -0700)]
wrapper.c: Add support for empty values in cgroup_add_value_string()
cgroup_add_value_string() is the fundamental building block
for adding a name/value pair to a cgroup_controller struct.
Add support for a NULL value field.
Currently it only supports adding a valid value, but in a
subsequent commit, cgget will utilize this function to build
up a hierarchy of cgroups, controllers, and setting names.
The values are NULL at the time the struct are populated and
will be read from sysfs at a later time.
Signed-off-by: Tom Hromatka <tom.hromatka@oracle.com>
Tom Hromatka [Tue, 23 Feb 2021 20:47:38 +0000 (13:47 -0700)]
api.c: Fix erroneous error handling in cgroup_set_values_recursive()
Ignore the positive length returned by asprintf() and only act
upon the error code returned from cg_set_control_value(). Previously
cgroup_set_values_recursive() mishandled the error handling when
the value's dirty flag was set.
Reported-by: Bharani viswas <bharani.viswas@gmail.com> Fixes: 092e4aeb0416f ("api.c: Refactor duplicate logic in cgroup modify and create") Signed-off-by: Tom Hromatka <tom.hromatka@oracle.com>
Tom Hromatka [Thu, 17 Dec 2020 18:39:11 +0000 (11:39 -0700)]
cgsnapshot: Replace hardcoded "tasks" file
Replace the hardcoded building of the tasks path in
cgsnapshot with a call to cgroup_build_tasks_procs_path().
By calling cgroup_build_tasks_procs_path(), cgsnapshot
now supports both cgroup v1 and cgroup v2.
Signed-off-by: Tom Hromatka <tom.hromatka@oracle.com> Reviewed-by: Dhaval Giani <dhaval.giani@oracle.com>
Tom Hromatka [Thu, 17 Dec 2020 18:24:13 +0000 (11:24 -0700)]
github: Display ftests.sh.log in the CI GUI
Now that we invoke ftests.sh in the functional tests'
Makefile.am, test-suite.log isn't that useful. Display
ftests.sh.log in the Github Actions GUI since
ftests.sh.log contains pass/fail info for each test.
Signed-off-by: Tom Hromatka <tom.hromatka@oracle.com>
Tom Hromatka [Thu, 17 Dec 2020 15:03:32 +0000 (08:03 -0700)]
github: Improve logging and reporting
Display the high-level log, test-suite.log, in
the github work for all jobs. Also, archive the
functional test logs to facilitate debugging a
failed run.
Signed-off-by: Tom Hromatka <tom.hromatka@oracle.com>
Tom Hromatka [Tue, 15 Dec 2020 22:11:01 +0000 (15:11 -0700)]
README.md: Update the coveralls badge
The coveralls.io badge wasn't consistently updating
after a successful run. Remove the "?branch=master"
parameter and switch to using a png rather than an
svg.
Signed-off-by: Tom Hromatka <tom.hromatka@oracle.com>
Add a Github Actions continuous integration workflow. This
workflow runs the unit tests, the cgroup v1 functional tests,
and the cgroup v2 functional tests. The code coverage results
from all of the runs are collated at coveralls.io.
Like Travis CI, Github Actions currently does not provide a
distro that natively runs cgroup v2. (And it also doesn't
expose the kernel boot command line.) So, like our previous
Travis CI solution, unmount the cpuset controller and enable
it in cgroup v2.
Signed-off-by: Tom Hromatka <tom.hromatka@oracle.com>
Tom Hromatka [Tue, 15 Dec 2020 20:45:46 +0000 (13:45 -0700)]
github: Add a Github Action to setup the libcgroup directory
Add a Github Action to setup the libcgroup directory.
This action installs the required dependencies, run
bootstrap.sh, runs configure, and invokes make.
Since the action is invoked locally (and not via Github
Action's published action mechanism), the repository must
be git cloned prior to running this action.
Signed-off-by: Tom Hromatka <tom.hromatka@oracle.com>
Tom Hromatka [Wed, 28 Oct 2020 14:57:09 +0000 (14:57 +0000)]
travis: Add test job on cgroup v2 Ubuntu 20.04
Add a test job on Ubuntu 20.04. Prior to running the tests, unmount
the cpuset cgroup controller and enable it in the cgroup v2 unified
hierarchy. Once each job is complete, it notifies coveralls via a
REST call.
Unfortunately, Travis CI doesn't provide a cgroup v2 system image
and also doesn't allow us to modify the kernel boot parameters.
Signed-off-by: Tom Hromatka <tom.hromatka@oracle.com>
Tom Hromatka [Thu, 15 Oct 2020 21:58:19 +0000 (15:58 -0600)]
api.c: Replace hardcoded "/tasks" in delete functions
Replace the hardcoded building of the tasks path in
cg_delete_cgroup_controller() and cgroup_delete_cgroup_ext()
with a call to cgroup_build_tasks_procs_path(). By calling
cgroup_build_tasks_procs_path(), both delete functions now
support cgroup v1 and v2.
cgdelete now supports cgroup v2.
Signed-off-by: Tom Hromatka <tom.hromatka@oracle.com>
Tom Hromatka [Wed, 23 Sep 2020 16:41:32 +0000 (10:41 -0600)]
api.c: Verify if a controller is enabled for a v2 cgroup
Add a function, cgroupv2_controller_enabled(), that checks if the
requested cgroup v2 controller can be used for the specified path.
Again, in cgroup v2, to use a controller, it must be added to the
parent cgroup's cgroup.subtree_control file.
This check is required to ensure that the cgexec and cgclassify
operations will succeed. And with this commit, cgexec and cgclassify
work on both cgroup v1 and cgroup v2 hierarchies.
Signed-off-by: Tom Hromatka <tom.hromatka@oracle.com>
Tom Hromatka [Wed, 23 Sep 2020 16:41:00 +0000 (10:41 -0600)]
api.c: Add function to read the cgroup.subtree_control file
Add a function, cgroupv2_get_subtree_control(), that verifies
if a controller is enabled in the cgroup.subtree_control file
for the requested path.
This function will be used in the next commit - "api.c: Add function
to check if a controller is enabled for a cgroup v2 cgroup" - to
determine if the requested cgroup v2 controller is enabled for a
cgroup. This is a required check to ensure that the
cgclassify/cgexec operation will succeed.
Signed-off-by: Tom Hromatka <tom.hromatka@oracle.com>
Tom Hromatka [Wed, 23 Sep 2020 16:39:33 +0000 (10:39 -0600)]
api.c: Add function to build tasks/procs path
Replace the inline logic in multiple functions that concatenates
the path and cgroup v1 tasks file with a function -
cgroup_build_tasks_procs_path() - that works for both cgroup v1
and v2. In the v1 case, it appends tasks, and in the v2 case, it
appends cgroup.procs.
Signed-off-by: Tom Hromatka <tom.hromatka@oracle.com>
Tom Hromatka [Wed, 29 Jul 2020 16:52:49 +0000 (16:52 +0000)]
api.c: Add cgroup v2 support to cgroup_create_cgroup()
This commit adds cgroup v2 support go cgroup_create_cgroup().
If the controller is mounted via cgroup v2, then the
subtree_control file in the parent directory will be updated
to enable the controller in children cgroups. The tasks
file is only updated for cgroup v1 mounts.
Here's an example of creating a cgroup hierarchy in a single command:
sudo cgcreate -g io:a/b/c/d
The above cgcreate example is equivalent to:
mkdir -p {root_cg}/a/b/c/d
echo +io > {root_cg}/cgroup.subtree_control
echo +io > {root_cg}/a/cgroup.subtree_control
echo +io > {root_cg}/a/b/cgroup.subtree_control
echo +io > {root_cg}/a/b/c/cgroup.subtree_control
Signed-off-by: Tom Hromatka <tom.hromatka@oracle.com>
Tom Hromatka [Wed, 29 Jul 2020 16:46:44 +0000 (16:46 +0000)]
api.c: Add function to recursively manage the cgroup v2 subtree
This commit adds the function cgroupv2_subtree_control_recursive().
cgroupv2_subtree_control_recursive() creates a cgroup hierarchy with
the subtree_control enabled in each parent cgroup for the requested
cgroup controller.
Signed-off-by: Tom Hromatka <tom.hromatka@oracle.com>
Tom Hromatka [Thu, 16 Jul 2020 20:26:49 +0000 (14:26 -0600)]
api.c: Add function to manage the cgroup v2 subtree_control file
This commit adds the function cgroupv2_subtree_control().
cgroupv2_subtree_control() can be used to enable or disable a
controller in the subtree_control file. The equivalent command
line is:
echo +{ctrl_name} > {path}/cgroup.subtree_control
echo -{ctrl_name} > {path}/cgroup.subtree_control
Signed-off-by: Tom Hromatka <tom.hromatka@oracle.com>
Tom Hromatka [Wed, 15 Jul 2020 18:44:20 +0000 (12:44 -0600)]
api.c: Refactor tasks file permissions code into its own function
Within cgroup_create_cgroup(), there was a block of code to set the
owner, group, and file permissions on the cgroup v1 tasks file. This
commit moves that logic into its own function -
cgroup_chown_chmod_tasks().
Signed-off-by: Tom Hromatka <tom.hromatka@oracle.com>
Tom Hromatka [Wed, 15 Jul 2020 17:06:47 +0000 (11:06 -0600)]
api.c: Refactor duplicate logic in cgroup modify and create
cgroup_modify_cgroup() and cgroup_create_cgroup() have nearly
identical logic to walk the settings beneath the controller
and update them. This commit refactors the duplicate logic in
these two functions into one location - cgroup_set_values_recursive().
Signed-off-by: Tom Hromatka <tom.hromatka@oracle.com>
Tom Hromatka [Wed, 15 Jul 2020 18:06:15 +0000 (12:06 -0600)]
travis: Show the gunit log on failure
When a failure occurs in TravisCI, we currently show only the
test-suite.log from the functional tests. This commit also displays
the googletest test-suite.log.
Signed-off-by: Tom Hromatka <tom.hromatka@oracle.com>
Koji Builder [Sat, 11 Jul 2020 15:22:48 +0000 (16:22 +0100)]
Paralell build fix #7
All source soce files which first needs to be generated needs to be listed in
BUILT_SOURCES.
This patch adds to that automake variable parse.c and parse.h because they are
listed below in libcgroup_la_SOURCES.
Documentetion of the BUILT_SOURCES is on:
https://www.gnu.org/software/automake/manual/html_node/Sources.html
Reported-by: Tomasz Kłoczko <kloczek@giithub.com> Signed-off-by: Tomasz Kłoczko <kloczek@giithub.com>
Tom Hromatka [Mon, 11 May 2020 02:09:44 +0000 (02:09 +0000)]
api.c: cgget: cgset: Add cgroup v2 support
This commit adds cgroup v2 support to cgroup_init(). With these
changes, cgget and cgset now work on a cgroup v1 mount, a cgroup
v2 mount, or a cgroup v2 unified mount hierarchy.
Signed-off-by: Tom Hromatka <tom.hromatka@oracle.com>
Tom Hromatka [Mon, 11 May 2020 02:20:40 +0000 (02:20 +0000)]
api.c: Add support for cgroup v2 to cg_test_mounted_fs()
This commit adds a string comparison of "cgroup2" to
cg_test_mounted_fs(). This allows cgroup v2 mounts to pass the
filesystem check and thus cgroups can be created.
Signed-off-by: Tom Hromatka <tom.hromatka@oracle.com> Acked-by: Dhaval Giani <dhaval.giani@oracle.com>
Tom Hromatka [Mon, 11 May 2020 02:00:37 +0000 (02:00 +0000)]
api.c: Refactor cgroup v1 logic in cgroup_init()
In preparation for adding support for cgroup v2, this commit
moves the cgroup v1 mount logic out of cgroup_init() and into
its own function. No functional changes were made.
Signed-off-by: Tom Hromatka <tom.hromatka@oracle.com> Acked-by: Dhaval Giani <dhaval.giani@oracle.com>
Tom Hromatka [Tue, 7 Apr 2020 14:08:25 +0000 (08:08 -0600)]
tests: Remove the tests folder from this git repo
The test code has been moved to a separate git repo,
libcgroup-tests. But before that repo can be added as a
submodule, the tests directory must first be deleted from
this repo.
Signed-off-by: Tom Hromatka <tom.hromatka@oracle.com>
Tom Hromatka [Mon, 6 Apr 2020 14:43:43 +0000 (08:43 -0600)]
tests: Remove the tests folder from this git repo
The test code has been moved to a separate git repo,
libcgroup-tests. Before we can add that repo as a submodule,
we must first delete the tests directory.
Signed-off-by: Tom Hromatka <tom.hromatka@oracle.com>
Nikola Forró [Thu, 16 Jan 2020 17:43:54 +0000 (18:43 +0100)]
Increase maximal size of controller values
Maximal length of a controller value is determined by CG_VALUE_MAX,
which is equal to 100. That is not sufficient in some cases.
Add new constant CG_CONTROL_VALUE_MAX (to prevent breaking current API)
and set it to 4096, which is usually equal to the amount of bytes that
can be written to a sysctl file directly.
Add warning message about exceeding the limit while parsing
configuration file.
Signed-off-by: Nikola Forró <nforro@redhat.com> Reviewed-by: Tom Hromatka <tom.hromatka@oracle.com>
Nikola Forró [Thu, 16 Jan 2020 17:43:51 +0000 (18:43 +0100)]
api.c: Preserve dirty flag when copying controller values
When setting cgroup parameters with cgset fails, no error is reported.
This is caused by the fact that cgroup_copy_controller_values is not
preserving dirty flags of the values, so it's making all errors
considered non-fatal.
Signed-off-by: Nikola Forró <nforro@redhat.com> Reviewed-by: Tom Hromatka <tom.hromatka@oracle.com>
Nikola Forró [Thu, 16 Jan 2020 17:43:50 +0000 (18:43 +0100)]
api.c: Fix potential buffer overflow
It is assumed that arguments read from /proc/<pid>/cmdline don't exceed
buf_pname buffer size, which is FILENAME_MAX - 1 characters, but that's
not always the case.
Add check to prevent buffer overflow and discard the excessive part of
an argument.
Signed-off-by: Nikola Forró <nforro@redhat.com> Reviewed-by: Tom Hromatka <tom.hromatka@oracle.com>
Nikola Forró [Thu, 16 Jan 2020 17:43:49 +0000 (18:43 +0100)]
api.c: Fix order of memory subsystem parameters generated by cgsnapshot
Order of parameters usually doesn't matter, but that's not the case with
memory.limit_in_bytes and memory.memsw.limit_in_bytes. When the latter
is first in the list of parameters, the resulting configuration is not
loadable with cgconfigparser.
This happens because when a cgroup is created, both memory.limit_in_bytes
and memory.memsw.limit_in_bytes parameters are initialized to highest
value possible (RESOURCE_MAX). And because memory.memsw.limit_in_bytes
must be always higher or equal to memory.limit_in_bytes, it's impossible
to change its value first.
Make sure that after constructing parameter list of memory subsystem,
the mentioned parameters are in correct order.
Signed-off-by: Nikola Forró <nforro@redhat.com> Reviewed-by: Tom Hromatka <tom.hromatka@oracle.com>
Tom Hromatka [Mon, 13 Jan 2020 23:34:56 +0000 (23:34 +0000)]
tests: Fix test failure on Fedora 30
Unit test 004-cgroup_compare_ignore_rule.cpp segfaults on
test CgroupCompareIgnoreRuleTest.CombinedControllers(). This
is due to rule.procname being uninitialized. This commit
initializes the entire rule structure to zero.
Signed-off-by: Tom Hromatka <tom.hromatka@oracle.com>
Tom Hromatka [Mon, 13 Jan 2020 23:31:20 +0000 (23:31 +0000)]
autotools: gtest: Distribute the googletest *.so files
googletest strongly recommends that their library is
rebuilt on each platform, but this runs counter to a
tagged release of code that utilizes gtest. This commit
enables `make dist` to bundle the googletest *.so files
into the resultant zip file.
Note that these googletest *.so files are dependent upon
a certain version of glibc (and other libraries) and may
not work on distros with very old versions of those
tools.
Signed-off-by: Tom Hromatka <tom.hromatka@oracle.com>
Tom Hromatka [Fri, 10 Jan 2020 22:46:26 +0000 (22:46 +0000)]
gtest: Cleanup gtest make targets
Prior to this commit the unit test map file was being
included in the source directory. It has now been moved
to the tests/gunit directory. This commit also properly
connects the tests/gunit Makefile to the googletest *.la
so that a recompile of googletest is not required.
Signed-off-by: Tom Hromatka <tom.hromatka@oracle.com> Reviewed-by: Dhaval Giani <dhaval.giani@oracle.com>
Tom Hromatka [Fri, 10 Jan 2020 22:44:04 +0000 (22:44 +0000)]
bootstrap: Actually build gtest
googletest should be entirely compiled into an *.la
prior to ./configure. The *.la should be added as
an extra distribution item to the dist tarball. This
will make it so that the distribution vendors do not
need to rebuild googletest.
Signed-off-by: Tom Hromatka <tom.hromatka@oracle.com> Reviewed-by: Dhaval Giani <dhaval.giani@oracle.com>
projects / thirdparty / libcgroup.git / log
