2023-09-20 14:30:55

by Kamalesh Babulal

[permalink] [raw]
Subject: [PATCH] cgroup: In cgroup_no_v1() check v1 controllers only

cgroup v1 or v2 or both controller names can be passed as arguments to
the 'cgroup_no_v1' kernel parameter, though most of the controller's
names are the same for both cgroup versions. This can be confusing when
both versions are used interchangeably, i.e., passing cgroup_no_v1=io

$ sudo dmesg |grep cgroup
...
cgroup: Disabling io control group subsystem in v1 mounts
cgroup: Disabled controller 'blkio'

Restrict the check to ss->legacy_name only, when parsing the controller
names passed to cgroup_no_v1 parameter and also, use ss->legacy_name in
pr_info(), that prints the controller disabling information.

Signed-off-by: Kamalesh Babulal <[email protected]>
---
kernel/cgroup/cgroup-v1.c | 3 +--
kernel/cgroup/cgroup.c | 2 +-
2 files changed, 2 insertions(+), 3 deletions(-)

diff --git a/kernel/cgroup/cgroup-v1.c b/kernel/cgroup/cgroup-v1.c
index c487ffef6652..942872b83bf2 100644
--- a/kernel/cgroup/cgroup-v1.c
+++ b/kernel/cgroup/cgroup-v1.c
@@ -1297,8 +1297,7 @@ static int __init cgroup_no_v1(char *str)
}

for_each_subsys(ss, i) {
- if (strcmp(token, ss->name) &&
- strcmp(token, ss->legacy_name))
+ if (strcmp(token, ss->legacy_name))
continue;

cgroup_no_v1_mask |= 1 << i;
diff --git a/kernel/cgroup/cgroup.c b/kernel/cgroup/cgroup.c
index 1fb7f562289d..622fb53a806b 100644
--- a/kernel/cgroup/cgroup.c
+++ b/kernel/cgroup/cgroup.c
@@ -6121,7 +6121,7 @@ int __init cgroup_init(void)

if (cgroup1_ssid_disabled(ssid))
pr_info("Disabling %s control group subsystem in v1 mounts\n",
- ss->name);
+ ss->legacy_name);

cgrp_dfl_root.subsys_mask |= 1 << ss->id;


base-commit: 2cf0f715623872823a72e451243bbf555d10d032
--
2.41.0


2023-10-04 19:19:36

by Tejun Heo

[permalink] [raw]
Subject: Re: [PATCH] cgroup: In cgroup_no_v1() check v1 controllers only

On Wed, Sep 20, 2023 at 03:55:12PM +0530, Kamalesh Babulal wrote:
> cgroup v1 or v2 or both controller names can be passed as arguments to
> the 'cgroup_no_v1' kernel parameter, though most of the controller's
> names are the same for both cgroup versions. This can be confusing when
> both versions are used interchangeably, i.e., passing cgroup_no_v1=io
>
> $ sudo dmesg |grep cgroup
> ...
> cgroup: Disabling io control group subsystem in v1 mounts
> cgroup: Disabled controller 'blkio'

So, making the printed names consistent makes sense but I'm not sure about
not matching "io" anymore. That's gonna break users who already use them,
right?

Thanks.

--
tejun

2023-10-05 14:41:48

by Kamalesh Babulal

[permalink] [raw]
Subject: Re: [External] : Re: [PATCH] cgroup: In cgroup_no_v1() check v1 controllers only



On 10/5/23 00:47, Tejun Heo wrote:
> On Wed, Sep 20, 2023 at 03:55:12PM +0530, Kamalesh Babulal wrote:
>> cgroup v1 or v2 or both controller names can be passed as arguments to
>> the 'cgroup_no_v1' kernel parameter, though most of the controller's
>> names are the same for both cgroup versions. This can be confusing when
>> both versions are used interchangeably, i.e., passing cgroup_no_v1=io
>>
>> $ sudo dmesg |grep cgroup
>> ...
>> cgroup: Disabling io control group subsystem in v1 mounts
>> cgroup: Disabled controller 'blkio'
>
> So, making the printed names consistent makes sense but I'm not sure about
> not matching "io" anymore. That's gonna break users who already use them,
> right?

Agreed, users might pass "io" in place of "blkio", I was thinking
in terms of cgroup v1 users, for whom it is "blkio" controller but
as an argument, "cgroup_no_v1" makes the controllers passed to it,
available under cgroup v2, where the controller is named "io",
which the user may use interchangeably.

Shall I re-send the patch to print the controller legacy_name only?

--
Thanks,
Kamalesh

2023-10-05 18:12:54

by Tejun Heo

[permalink] [raw]
Subject: Re: [External] : Re: [PATCH] cgroup: In cgroup_no_v1() check v1 controllers only

On Thu, Oct 05, 2023 at 04:51:52PM +0530, Kamalesh Babulal wrote:
>
>
> On 10/5/23 00:47, Tejun Heo wrote:
> > On Wed, Sep 20, 2023 at 03:55:12PM +0530, Kamalesh Babulal wrote:
> >> cgroup v1 or v2 or both controller names can be passed as arguments to
> >> the 'cgroup_no_v1' kernel parameter, though most of the controller's
> >> names are the same for both cgroup versions. This can be confusing when
> >> both versions are used interchangeably, i.e., passing cgroup_no_v1=io
> >>
> >> $ sudo dmesg |grep cgroup
> >> ...
> >> cgroup: Disabling io control group subsystem in v1 mounts
> >> cgroup: Disabled controller 'blkio'
> >
> > So, making the printed names consistent makes sense but I'm not sure about
> > not matching "io" anymore. That's gonna break users who already use them,
> > right?
>
> Agreed, users might pass "io" in place of "blkio", I was thinking
> in terms of cgroup v1 users, for whom it is "blkio" controller but
> as an argument, "cgroup_no_v1" makes the controllers passed to it,
> available under cgroup v2, where the controller is named "io",
> which the user may use interchangeably.
>
> Shall I re-send the patch to print the controller legacy_name only?

Yes, let's do that.

Thanks.

--
tejun