systemd always invokes rmdir() on pids cgroup

[vikaschoudhary16]

There is a patch in systemd-219 which is generated from the sources fetched using rhpkg, 0475-core-add-support-for-the-pids-cgroup-controller.patch. This patch adds CGROUP_PIDS to CGroupControllerMask (along with some other changes) but does not update _CGROUP_CONTROLLER_MASK_ALL from 31 to 63. This mistake makes the check at L#1628 to fail for pids and therefore for pids, cg_trim() is invoked always.

Here is the relevant code:

1605         "cpu\0"
1606         "cpuacct\0"
1607         "blkio\0"
1608         "memory\0"
1609         "devices\0"
1610         "pids\0";
1611  
1612 int cg_create_everywhere(CGroupControllerMask supported, CGroupControllerMask mask, const char *path) {
1613         CGroupControllerMask bit = 1;
1614         const char *n;
1615         int r;
1616                 
1617         /* This one will create a cgroup in our private tree, but also
1618          * duplicate it in the trees specified in mask, and remove it
1619          * in all others */
1620  
1621         /* First create the cgroup in our own hierarchy. */
1622         r = cg_create(SYSTEMD_CGROUP_CONTROLLER, path);
1623         if (r < 0)
1624                 return r;
1625                         
1626         /* Then, do the same in the other hierarchies */
1627         NULSTR_FOREACH(n, mask_names) {
1628                 if (mask & bit)
1629                         cg_create(n, path);
1630                 else if (supported & bit)
1631                         cg_trim(n, path, true);
1632 
1633                 bit <<= 1;
1634         }
1635 
1636         return 0;
1637 }

src/shared/cgroup-util.h
32 typedef enum CGroupControllerMask {
 33         CGROUP_CPU = 1,
 34         CGROUP_CPUACCT = 2,
 35         CGROUP_BLKIO = 4,
 36         CGROUP_MEMORY = 8,
 37         CGROUP_DEVICE = 16,
 38         CGROUP_PIDS = 32,
 39         _CGROUP_CONTROLLER_MASK_ALL = 31
 40 } CGroupControllerMask;

At L39, there should be 63

Also Please take a look at openshift/origin#16246 (comment)

[lnykryn]

Yep, looks like we made a mistake during the backport. Is there an RH Bugzilla for this?

[lnykryn]

https://bugzilla.redhat.com/show_bug.cgi?id=1532586

[msekletar]

LGTM

@vikaschoudhary16 thanks for tracking this down!

[jeremyeder]

👍