From eaeed6cca46a0223617ead834aaa576dd5ad07ff Mon Sep 17 00:00:00 2001

From: Klaus Wenninger <klaus.wenninger@aon.at>

Date: Fri, 31 May 2019 16:11:16 +0200

Subject: [PATCH] Fix: sbd-common: query rt-budget > 0 otherwise try moving to

 root-slice

---

 src/sbd-common.c     | 110 +++++++++++++++++++++++++++++++++++++++++++++++++++

 src/sbd-inquisitor.c |  15 +++++++

 src/sbd.h            |   2 +

 src/sbd.sysconfig    |  14 +++++++

 4 files changed, 141 insertions(+)

diff --git a/src/sbd-common.c b/src/sbd-common.c

index 873a76e..ebfdaa3 100644

--- a/src/sbd-common.c

+++ b/src/sbd-common.c

@@ -662,6 +662,112 @@ static void sbd_memlock(int stackgrowK, int heapgrowK)

 #endif

 }

+static int get_realtime_budget(void)

+{

+    FILE *f;

+    char fname[PATH_MAX];

+    int res = -1, lnum = 0;

+    char *cgroup = NULL, *namespecs = NULL;

+

+    snprintf(fname, PATH_MAX, "/proc/%jd/cgroup", (intmax_t)getpid());

+    f = fopen(fname, "rt");

+    if (f == NULL) {

+        cl_log(LOG_WARNING, "Can't open cgroup file for pid=%jd",

+                            (intmax_t)getpid());

+        goto exit_res;

+    }

+    while( fscanf(f, "%d:%m[^:]:%m[^\n]", &lnum,  &namespecs, &cgroup) !=EOF ) {

+        if (namespecs && strstr(namespecs, "cpuacct")) {

+            free(namespecs);

+            break;

+        }

+        if (cgroup) {

+            free(cgroup);

+            cgroup = NULL;

+        }

+        if (namespecs) {

+            free(namespecs);

+            namespecs = NULL;

+        }

+    }

+    fclose(f);

+    if (cgroup == NULL) {

+        cl_log(LOG_WARNING, "Failed getting cgroup for pid=%jd",

+                            (intmax_t)getpid());

+        goto exit_res;

+    }

+    snprintf(fname, PATH_MAX, "/sys/fs/cgroup/cpu%s/cpu.rt_runtime_us",

+                              cgroup);

+    f = fopen(fname, "rt");

+    if (f == NULL) {

+        cl_log(LOG_WARNING, "cpu.rt_runtime_us existed for root-slice but "

+            "doesn't for '%s'", cgroup);

+        goto exit_res;

+    }

+    if (fscanf(f, "%d", &res) != 1) {

+        cl_log(LOG_WARNING, "failed reading rt-budget from %s", fname);

+    } else {

+        cl_log(LOG_INFO, "slice='%s' has rt-budget=%d", cgroup, res);

+    }

+    fclose(f);

+

+exit_res:

+    if (cgroup) {

+        free(cgroup);

+    }

+    return res;

+}

+

+/* stolen from corosync */

+static int sbd_move_to_root_cgroup(bool enforce_root_cgroup) {

+    FILE *f;

+    int res = -1;

+

+    /*

+     * /sys/fs/cgroup is hardcoded, because most of Linux distributions are now

+     * using systemd and systemd uses hardcoded path of cgroup mount point.

+     *

+     * This feature is expected to be removed as soon as systemd gets support

+     * for managing RT configuration.

+     */

+    f = fopen("/sys/fs/cgroup/cpu/cpu.rt_runtime_us", "rt");

+    if (f == NULL) {

+        cl_log(LOG_DEBUG, "cpu.rt_runtime_us doesn't exist -> "

+            "system without cgroup or with disabled CONFIG_RT_GROUP_SCHED");

+        res = 0;

+        goto exit_res;

+    }

+    fclose(f);

+

+    if ((!enforce_root_cgroup) && (get_realtime_budget() > 0)) {

+        cl_log(LOG_DEBUG, "looks as if we have rt-budget in the slice we are "

+                          "-> skip moving to root-slice");

+        res = 0;

+        goto exit_res;

+    }

+

+    f = fopen("/sys/fs/cgroup/cpu/tasks", "w");

+    if (f == NULL) {

+        cl_log(LOG_WARNING, "Can't open cgroups tasks file for writing");

+

+        goto exit_res;

+    }

+

+    if (fprintf(f, "%jd\n", (intmax_t)getpid()) <= 0) {

+        cl_log(LOG_WARNING, "Can't write sbd pid into cgroups tasks file");

+        goto close_and_exit_res;

+    }

+

+close_and_exit_res:

+    if (fclose(f) != 0) {

+        cl_log(LOG_WARNING, "Can't close cgroups tasks file");

+        goto exit_res;

+    }

+

+exit_res:

+    return (res);

+}

+

 void

 sbd_make_realtime(int priority, int stackgrowK, int heapgrowK)

 {

@@ -670,6 +776,10 @@ sbd_make_realtime(int priority, int stackgrowK, int heapgrowK)

     }

 #ifdef SCHED_RR

+    if (move_to_root_cgroup) {

+        sbd_move_to_root_cgroup(enforce_moving_to_root_cgroup);

+    }

+

     {

         int pcurrent = 0;

         int pmin = sched_get_priority_min(SCHED_RR);

diff --git a/src/sbd-inquisitor.c b/src/sbd-inquisitor.c

index abde4e5..cef5cc7 100644

--- a/src/sbd-inquisitor.c

+++ b/src/sbd-inquisitor.c

@@ -33,6 +33,8 @@ int	start_mode = 0;

 char*	pidfile = NULL;

 bool do_flush = true;

 char timeout_sysrq_char = 'b';

+bool move_to_root_cgroup = true;

+bool enforce_moving_to_root_cgroup = false;

 int parse_device_line(const char *line);

@@ -965,6 +967,19 @@ int main(int argc, char **argv, char **envp)

             timeout_action = strdup(value);

         }

+        value = getenv("SBD_MOVE_TO_ROOT_CGROUP");

+        if(value) {

+            move_to_root_cgroup = crm_is_true(value);

+

+            if (move_to_root_cgroup) {

+               enforce_moving_to_root_cgroup = true;

+            } else {

+                if (strcmp(value, "auto") == 0) {

+                    move_to_root_cgroup = true;

+                }

+            }

+        }

+

 	while ((c = getopt(argc, argv, "czC:DPRTWZhvw:d:n:p:1:2:3:4:5:t:I:F:S:s:r:")) != -1) {

 		switch (c) {

 		case 'D':

diff --git a/src/sbd.h b/src/sbd.h

index 3b05a11..ac30ec7 100644

--- a/src/sbd.h

+++ b/src/sbd.h

@@ -159,6 +159,8 @@ extern bool watchdogdev_is_default;

 extern char*  local_uname;

 extern bool do_flush;

 extern char timeout_sysrq_char;

+extern bool move_to_root_cgroup;

+extern bool enforce_moving_to_root_cgroup;

 /* Global, non-tunable variables: */

 extern int  sector_size;

diff --git a/src/sbd.sysconfig b/src/sbd.sysconfig

index f163f21..e1a60ed 100644

--- a/src/sbd.sysconfig

+++ b/src/sbd.sysconfig

@@ -91,6 +91,20 @@ SBD_WATCHDOG_TIMEOUT=5

 #

 SBD_TIMEOUT_ACTION=flush,reboot

+## Type: yesno / auto

+## Default: auto

+#

+# If CPUAccounting is enabled default is not to assign any RT-budget

+# to the system.slice which prevents sbd from running RR-scheduled.

+#

+# One way to escape that issue is to move sbd-processes from the

+# slice they were originally started to root-slice.

+# Of course starting sbd in a certain slice might be intentional.

+# Thus in auto-mode sbd will check if the slice has RT-budget assigned.

+# If that is the case sbd will stay in that slice while it will

+# be moved to root-slice otherwise.

+SBD_MOVE_TO_ROOT_CGROUP=auto

+

 ## Type: string

 ## Default: ""

 #

-- 

1.8.3.1