Skip to contentSkip to conversation

tst_cgroup: Don't try to use V2 if V1 controllers are mounted

PATCH New Richard Palethorpe 4 participants 8 messages Sep 24, 2020 – Sep 25, 2020 Library code

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 tst_cgroup: Don't try to use V2 if V1 controllers are mounted

It is not possible to use a controller in V2 cgroups if it has been mounted as a V1 controller. So if V1 is mounted we use it regardless of if V2 is available.

We have to include a space in tst_is_mounted so that we do not match cgroup2.

---
 lib/tst_cgroup.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
lib/tst_cgroup.c +1 -1
index ba413d874..73ddd4b82 100644
@@ -44,7 +44,7 @@ enum tst_cgroup_ver tst_cgroup_version(void)
         enum tst_cgroup_ver cg_ver;
 
         if (tst_cgroup_check("cgroup2")) {
-                if (!tst_is_mounted("cgroup2") && tst_is_mounted("cgroup"))
+                if (tst_is_mounted("cgroup "))
                         cg_ver = TST_CGROUP_V1;
                 else
                         cg_ver = TST_CGROUP_V2;
Signature · 1 line
2.28.0
#1 lore raw

PATCH 1/1 tst_cgroup: Don't try to use V2 if V1 controllers are mounted

Hi Richie,

Quoted 6 lines

It is not possible to use a controller in V2 cgroups if it has been mounted as a V1 controller. So if V1 is mounted we use it regardless of if V2 is available.

We have to include a space in tst_is_mounted so that we do not match cgroup2.

Thanks for your fix.

Quoted 10 lines
---
 lib/tst_cgroup.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
lib/tst_cgroup.c +0 -0
index ba413d874..73ddd4b82 100644

lib/tst_cgroup.c enum tst_cgroup_ver tst_cgroup_version(void) · line 47

Quoted 5 lines
lib/tst_cgroup.c +1 -1
@@ -44,7 +44,7 @@ enum tst_cgroup_ver tst_cgroup_version(void)
         enum tst_cgroup_ver cg_ver;
 
         if (tst_cgroup_check("cgroup2")) {
-                if (!tst_is_mounted("cgroup2") && tst_is_mounted("cgroup"))
+                if (tst_is_mounted("cgroup "))

On first look the space looks like typo. But people can use git blame.

lib/tst_cgroup.c enum tst_cgroup_ver tst_cgroup_version(void) · line 50

Quoted 3 lines
lib/tst_cgroup.c +0 -0
                         cg_ver = TST_CGROUP_V1;
                 else
                         cg_ver = TST_CGROUP_V2;

Kind regards,
Petr

#2 lore raw

PATCH 1/1 tst_cgroup: Don't try to use V2 if V1 controllers are mounted

In reply to Richard Palethorpe

Hi Richard,

Quoted 17 lines · Richard Palethorpe

It is not possible to use a controller in V2 cgroups if it has been mounted as a V1 controller. So if V1 is mounted we use it regardless of if V2 is available.

We have to include a space in tst_is_mounted so that we do not match cgroup2.

---
 lib/tst_cgroup.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
lib/tst_cgroup.c +0 -0
index ba413d874..73ddd4b82 100644

lib/tst_cgroup.c enum tst_cgroup_ver tst_cgroup_version(void) · line 49

Quoted 7 lines
lib/tst_cgroup.c +1 -1
@@ -44,7 +44,7 @@ enum tst_cgroup_ver tst_cgroup_version(void)
         enum tst_cgroup_ver cg_ver;
 
         if (tst_cgroup_check("cgroup2")) {
-                if (!tst_is_mounted("cgroup2") &&
 tst_is_mounted("cgroup"))
+                if (tst_is_mounted("cgroup "))
 

Add a space in the suffix still not work as expected.

The reason is that tst_is_mounted("cgroup ") also get non-zero return if system only mount cgroup_v2, which lead to choose cgroup_v1 in LTP test.

# cat /proc/mounts |grep cgroup
cgroup2 /sys/fs/cgroup cgroup2 rw,seclabel,nosuid,nodev,noexec,relatime 0 0
Signature · 5 lines
Regards,
Li Wang
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.linux.it/pipermail/ltp/attachments/20200924/04c3bfa6/attachment.htm>
#3 lore raw

PATCH 1/1 tst_cgroup: Don't try to use V2 if V1 controllers are mounted

Hello Li,

Quoted 32 lines · Li Wang

Hi Richard,

Quoted 17 lines · Richard Palethorpe

It is not possible to use a controller in V2 cgroups if it has been mounted as a V1 controller. So if V1 is mounted we use it regardless of if V2 is available.

We have to include a space in tst_is_mounted so that we do not match cgroup2.

---
 lib/tst_cgroup.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
lib/tst_cgroup.c +0 -0
index ba413d874..73ddd4b82 100644

lib/tst_cgroup.c enum tst_cgroup_ver tst_cgroup_version(void) · line 49

Quoted 7 lines
lib/tst_cgroup.c +1 -1
@@ -44,7 +44,7 @@ enum tst_cgroup_ver tst_cgroup_version(void)
         enum tst_cgroup_ver cg_ver;
 
         if (tst_cgroup_check("cgroup2")) {
-                if (!tst_is_mounted("cgroup2") &&
 tst_is_mounted("cgroup"))
+                if (tst_is_mounted("cgroup "))
 

Add a space in the suffix still not work as expected.

Quoted 5 lines

The reason is that tst_is_mounted("cgroup ") also get non-zero return if system only mount cgroup_v2, which lead to choose cgroup_v1 in LTP test.

# cat /proc/mounts |grep cgroup
cgroup2 /sys/fs/cgroup cgroup2 rw,seclabel,nosuid,nodev,noexec,relatime 0 0

I wonder if it would be better to simply try mounting/using V2 and if that fails try V1?

Signature · 2 lines
Thank you,
Richard.
#4 lore raw

PATCH 1/1 tst_cgroup: Don't try to use V2 if V1 controllers are mounted

Quoted 40 lines · Richard Palethorpe

Hello Li,

Quoted 36 lines · Li Wang

Hi Richard,

On Thu, Sep 24, 2020 at 7:11 PM Richard Palethorpe <[email protected]

Quoted 17 lines · wrote:

It is not possible to use a controller in V2 cgroups if it has been mounted as a V1 controller. So if V1 is mounted we use it regardless of if V2 is available.

We have to include a space in tst_is_mounted so that we do not match cgroup2.

---
 lib/tst_cgroup.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
lib/tst_cgroup.c +0 -0
index ba413d874..73ddd4b82 100644

lib/tst_cgroup.c enum tst_cgroup_ver tst_cgroup_version(void) · line 49

Quoted 7 lines
lib/tst_cgroup.c +1 -1
@@ -44,7 +44,7 @@ enum tst_cgroup_ver tst_cgroup_version(void)
         enum tst_cgroup_ver cg_ver;
 
         if (tst_cgroup_check("cgroup2")) {
-                if (!tst_is_mounted("cgroup2") &&
 tst_is_mounted("cgroup"))
+                if (tst_is_mounted("cgroup "))
 

Add a space in the suffix still not work as expected.

The reason is that tst_is_mounted("cgroup ") also get non-zero return if system only mount cgroup_v2, which lead to choose cgroup_v1 in LTP test.

Quoted 7 lines
Quoted 2 lines
# cat /proc/mounts |grep cgroup
cgroup2 /sys/fs/cgroup cgroup2 rw,seclabel,nosuid,nodev,noexec,relatime

0 0

I wonder if it would be better to simply try mounting/using V2 and if that fails try V1?

That will be work but not good, because if cgroup mount fails, how do we know it is a bug or not support?

Signature · 5 lines
Regards,
Li Wang
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.linux.it/pipermail/ltp/attachments/20200925/67cf14da/attachment-0001.htm>
#5 lore raw

PATCH 1/1 tst_cgroup: Don't try to use V2 if V1 controllers are mounted

Hello Li,

Quoted 45 lines · Li Wang
Quoted 42 lines · Richard Palethorpe

Hello Li,

Quoted 38 lines · Li Wang

Hi Richard,

On Thu, Sep 24, 2020 at 7:11 PM Richard Palethorpe <[email protected]

Quoted 17 lines · wrote:

It is not possible to use a controller in V2 cgroups if it has been mounted as a V1 controller. So if V1 is mounted we use it regardless of if V2 is available.

We have to include a space in tst_is_mounted so that we do not match cgroup2.

---
 lib/tst_cgroup.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
lib/tst_cgroup.c +0 -0
index ba413d874..73ddd4b82 100644

lib/tst_cgroup.c enum tst_cgroup_ver tst_cgroup_version(void) · line 49

Quoted 7 lines
lib/tst_cgroup.c +1 -1
@@ -44,7 +44,7 @@ enum tst_cgroup_ver tst_cgroup_version(void)
         enum tst_cgroup_ver cg_ver;
 
         if (tst_cgroup_check("cgroup2")) {
-                if (!tst_is_mounted("cgroup2") &&
 tst_is_mounted("cgroup"))
+                if (tst_is_mounted("cgroup "))
 

Add a space in the suffix still not work as expected.

The reason is that tst_is_mounted("cgroup ") also get non-zero return if system only mount cgroup_v2, which lead to choose cgroup_v1 in LTP test.

# cat /proc/mounts |grep cgroup
cgroup2 /sys/fs/cgroup cgroup2 rw,seclabel,nosuid,nodev,noexec,relatime
Quoted 8 lines
Quoted 5 lines

0 0

I wonder if it would be better to simply try mounting/using V2 and if that fails try V1?

That will be work but not good, because if cgroup mount fails, how do we know it is a bug or not support?

I think this is a valid point if you are writing a test for mounting cgroups. However if we are testing something else then trying to guess if cgroups should be available before using them, makes the test fragile. I suppose we could check *after* trying to use the cgroups so we can report some diagnostic info.

Signature · 2 lines
Thank you,
Richard.
#6 lore raw

PATCH 1/1 tst_cgroup: Don't try to use V2 if V1 controllers are mounted

Richard Palethorpe <[email protected]> wrote:

...

Quoted 4 lines
Quoted 4 lines
Quoted 3 lines

I wonder if it would be better to simply try mounting/using V2 and if that fails try V1?

Quoted 9 lines
Quoted 2 lines

That will be work but not good, because if cgroup mount fails, how do we know it is a bug or not support?

I think this is a valid point if you are writing a test for mounting cgroups. However if we are testing something else then trying to guess if cgroups should be available before using them, makes the test fragile. I suppose we could check *after* trying to use the cgroups so we can report some diagnostic info.

This sounds practicable, please feel free to work out the patch.

Signature · 5 lines
Regards,
Li Wang
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.linux.it/pipermail/ltp/attachments/20200925/aa8c31f1/attachment.htm>
#7 lore raw

PATCH 1/1 tst_cgroup: Don't try to use V2 if V1 controllers are mounted

Hello Li,

Quoted 10 lines · Li Wang

Richard Palethorpe <[email protected]> wrote:

...

Quoted 7 lines
Quoted 6 lines
Quoted 3 lines

I wonder if it would be better to simply try mounting/using V2 and if that fails try V1?

That will be work but not good, because if cgroup mount fails, how do we know it is a bug or not support?

Quoted 8 lines
Quoted 6 lines

I think this is a valid point if you are writing a test for mounting cgroups. However if we are testing something else then trying to guess if cgroups should be available before using them, makes the test fragile. I suppose we could check *after* trying to use the cgroups so we can report some diagnostic info.

This sounds practicable, please feel free to work out the patch.

OK, I just sent a new patch. However there is no diagnostic info yet, I'm not sure if actually it would be better to make a test specifically checking if cgroups can be mounted and checking what is in filesystems and mounts.

Signature · 2 lines
Thank you,
Richard.