[libvirt] [RFC PATCH] cgroup: Use system reported "unlimited" value for comparison

[Viktor Mihajlovski]

With kernel 3.18 (since commit 3e32cb2e0a12b6915056ff04601cf1bb9b44f967)
the "unlimited" value for cgroup memory limits has changed once again as
its byte value is now computed from a page counter.
The new "unlimited" value reported by the cgroup fs is therefore 2**51-1
pages which is (VIR_DOMAIN_MEMORY_PARAM_UNLIMITED - 3072). This results
e.g. in virsh memtune displaying 9007199254740988 instead of unlimited
for the limits.

This patch uses the value of memory.limit_in_bytes from the cgroup
memory root which is the system's "real" unlimited value for comparison.

See also libvirt commit 231656bbeb9e4d3bedc44362784c35eee21cf0f4 for the
history for kernel 3.12 and before.

I've tested this on F24 with the following configurations:
- no memory cgroup controller mounted
- memory cgroup controller mounted but not configured for libvirt
- memory cgroup controller mounted and configured
The first two fail as expected (and as before), the third case
works as expected.

Testing on other kernel versions highly welcome!

Not perfect yet in that we still provide a fallback to the old value.
We might consider failing right away if we can't get the system
value. I'd be inclined to do that, since we're probably facing
principal cgroup issues in this case.

Further, it's not the most efficient implementation. Obviously, the
unlimited value can be read once and cached. However, I'd like to see
the question above resolved first.

Signed-off-by: Viktor Mihajlovski <mihajlov at linux.vnet.ibm.com>
---
 src/util/vircgroup.c | 61 ++++++++++++++++++++++++++++++++++++++++++++++------
 1 file changed, 55 insertions(+), 6 deletions(-)

diff --git a/src/util/vircgroup.c b/src/util/vircgroup.c
index f151193..969dca5 100644
--- a/src/util/vircgroup.c
+++ b/src/util/vircgroup.c
@@ -2452,6 +2452,40 @@ virCgroupGetBlkioDeviceWeight(virCgroupPtr group,
 }
 
 
+/*
+ * Retrieve the "memory.limit_in_bytes" value from the memory controller
+ * root dir. This value cannot be modified by userspace and therefore
+ * is the maximum limit value supported by cgroups on the local system.
+ */
+static int
+virCgroupGetMemoryUnlimited(unsigned long long int * mem_unlimited)
+{
+    int ret = -1;
+    virCgroupPtr group;
+
+    if (VIR_ALLOC(group))
+        goto cleanup;
+
+    if (virCgroupDetectMounts(group))
+        goto cleanup;
+
+    if (!group->controllers[VIR_CGROUP_CONTROLLER_MEMORY].mountPoint)
+        goto cleanup;
+
+    if (VIR_STRDUP(group->controllers[VIR_CGROUP_CONTROLLER_MEMORY].placement,
+                   "/.") < 0)
+        goto cleanup;
+
+    ret = virCgroupGetValueU64(group,
+                               VIR_CGROUP_CONTROLLER_MEMORY,
+                               "memory.limit_in_bytes",
+                               mem_unlimited);
+ cleanup:
+    virCgroupFree(&group);
+    return ret;
+}
+
+
 /**
  * virCgroupSetMemory:
  *
@@ -2534,6 +2568,7 @@ int
 virCgroupGetMemoryHardLimit(virCgroupPtr group, unsigned long long *kb)
 {
     long long unsigned int limit_in_bytes;
+    long long unsigned int unlimited_in_bytes;
     int ret = -1;
 
     if (virCgroupGetValueU64(group,
@@ -2541,9 +2576,13 @@ virCgroupGetMemoryHardLimit(virCgroupPtr group, unsigned long long *kb)
                              "memory.limit_in_bytes", &limit_in_bytes) < 0)
         goto cleanup;
 
-    *kb = limit_in_bytes >> 10;
-    if (*kb > VIR_DOMAIN_MEMORY_PARAM_UNLIMITED)
+    if (virCgroupGetMemoryUnlimited(&unlimited_in_bytes) < 0)
+        unlimited_in_bytes = VIR_DOMAIN_MEMORY_PARAM_UNLIMITED << 10;
+
+    if (limit_in_bytes == unlimited_in_bytes)
         *kb = VIR_DOMAIN_MEMORY_PARAM_UNLIMITED;
+    else
+        *kb = limit_in_bytes >> 10;
 
     ret = 0;
  cleanup:
@@ -2596,6 +2635,7 @@ int
 virCgroupGetMemorySoftLimit(virCgroupPtr group, unsigned long long *kb)
 {
     long long unsigned int limit_in_bytes;
+    long long unsigned int unlimited_in_bytes;
     int ret = -1;
 
     if (virCgroupGetValueU64(group,
@@ -2603,9 +2643,13 @@ virCgroupGetMemorySoftLimit(virCgroupPtr group, unsigned long long *kb)
                              "memory.soft_limit_in_bytes", &limit_in_bytes) < 0)
         goto cleanup;
 
-    *kb = limit_in_bytes >> 10;
-    if (*kb > VIR_DOMAIN_MEMORY_PARAM_UNLIMITED)
+    if (virCgroupGetMemoryUnlimited(&unlimited_in_bytes) < 0)
+        unlimited_in_bytes = VIR_DOMAIN_MEMORY_PARAM_UNLIMITED << 10;
+
+    if (limit_in_bytes == unlimited_in_bytes)
         *kb = VIR_DOMAIN_MEMORY_PARAM_UNLIMITED;
+    else
+        *kb = limit_in_bytes >> 10;
 
     ret = 0;
  cleanup:
@@ -2658,6 +2702,7 @@ int
 virCgroupGetMemSwapHardLimit(virCgroupPtr group, unsigned long long *kb)
 {
     long long unsigned int limit_in_bytes;
+    long long unsigned int unlimited_in_bytes;
     int ret = -1;
 
     if (virCgroupGetValueU64(group,
@@ -2665,9 +2710,13 @@ virCgroupGetMemSwapHardLimit(virCgroupPtr group, unsigned long long *kb)
                              "memory.memsw.limit_in_bytes", &limit_in_bytes) < 0)
         goto cleanup;
 
-    *kb = limit_in_bytes >> 10;
-    if (*kb > VIR_DOMAIN_MEMORY_PARAM_UNLIMITED)
+     if (virCgroupGetMemoryUnlimited(&unlimited_in_bytes) < 0)
+        unlimited_in_bytes = VIR_DOMAIN_MEMORY_PARAM_UNLIMITED << 10;
+
+    if (limit_in_bytes == unlimited_in_bytes)
         *kb = VIR_DOMAIN_MEMORY_PARAM_UNLIMITED;
+    else
+        *kb = limit_in_bytes >> 10;
 
     ret = 0;
  cleanup:
-- 
1.9.1