Skip to contentSkip to conversation

cgroup: Handle trailing new line in cgroup.controllers

PATCH Superseded Richard Palethorpe via ltp 4 participants 5 messages Oct 25, 2023 – Oct 26, 2023 Library code

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 cgroup: Handle trailing new line in cgroup.controllers

The last item in cgroup.controllers (misc in my case) contained a new line character which caused the controller search to fail.

---

This stops cgroup_regression_test.sh from failing on Tumbleweed. However there are other issues with that test which are not addressed by this patch.

 lib/tst_cgroup.c | 13 ++++++++++++-
 1 file changed, 12 insertions(+), 1 deletion(-)
lib/tst_cgroup.c +12 -1
index 5240aadaa..d70b5431d 100644
@@ -433,9 +433,20 @@ __attribute__ ((nonnull, warn_unused_result))
 static struct cgroup_ctrl *cgroup_find_ctrl(const char *const ctrl_name)
 {
 	struct cgroup_ctrl *ctrl;
+	int l = 0;
+
+	while (ctrl_name[l] >= 'a' && ctrl_name[l] <= 'z')
+		l++;
+
+	switch (ctrl_name[l]) {
+	case '\n': break;
+	case '\0': break;
+	default:
+		tst_brk(TBROK, "Unexpected char in %s: %c", ctrl_name, ctrl_name[l]);
+	}
 
 	for_each_ctrl(ctrl) {
-		if (!strcmp(ctrl_name, ctrl->ctrl_name))
+		if (!strncmp(ctrl_name, ctrl->ctrl_name, l))
 			return ctrl;
 	}
 
Signature · 5 lines
2.40.1


-- 
Mailing list info: https://lists.linux.it/listinfo/ltp
#1 lore raw

Am Mittwoch, 25. Oktober 2023, 13:05:33 CEST schrieb Richard Palethorpe via ltp:

lib/tst_cgroup.c __attribute__ ((nonnull, warn_unused_result))

Quoted 5 lines
lib/tst_cgroup.c +5 -0
+	switch (ctrl_name[l]) {
+	case '\n': break;
+	case '\0': break;
+	default:
+		tst_brk(TBROK, "Unexpected char in %s: %c", ctrl_name, 

ctrl_name[l]);

I'm wondering whether that's a bit too restrictive. Or is there any official documentation says that you really can only have the letters a-z in cgroup names (and not even A-Z). Otherwise it might be better to make this just a warning or allow any printable characters.

Signature · 1 line
Mailing list info: https://lists.linux.it/listinfo/ltp
#2 lore raw

Hi Richie, Martchus,

@Richie, please add Fixes: tag when commit

I suppose it should be

Fixes 310da3784 ("Add new CGroups APIs")

but please check yourself.

Why this is useful? It helps to identify which test failures were false positives. Also, you actually not just fix a line character, but also do other validation, it would be worth to mention that.

Quoted 9 lines

Am Mittwoch, 25. Oktober 2023, 13:05:33 CEST schrieb Richard Palethorpe via ltp:

lib/tst_cgroup.c __attribute__ ((nonnull, warn_unused_result))

Quoted 5 lines
lib/tst_cgroup.c +5 -0
+	switch (ctrl_name[l]) {
+	case '\n': break;
+	case '\0': break;
+	default:
+		tst_brk(TBROK, "Unexpected char in %s: %c", ctrl_name, 

ctrl_name[l]);

Quoted 4 lines

I'm wondering whether that's a bit too restrictive. Or is there any official documentation says that you really can only have the letters a-z in cgroup names (and not even A-Z). Otherwise it might be better to make this just a warning or allow any printable characters.

I guess for cgroup v1 [1]

The name should match [\w.-]+

\w Matches a "word" character (alphanumeric plus "_", plus other connector punctuation chars plus Unicode marks). Also '.' and '-' can be used. => [A-Z.-] and others are valid names in v1. Although I'm not sure if cgroup_find_ctrl() is used on systems with cgroups v1.

Also, shouldn't we check with MAX_CGROUP_TYPE_NAMELEN:

- name: should be initialized to a unique subsystem name. Should be no longer than MAX_CGROUP_TYPE_NAMELEN.

For cgroup v2 [2] it looks to be:

All cgroup core interface files are prefixed with "cgroup." and each controller's interface files are prefixed with the controller name and a dot. A controller's name is composed of lower case alphabets and '_'s but never begins with an '_' so it can be used as the prefix character for collision avoidance. Also, interface file names won't start or end with terms which are often used in categorizing workloads such as job, service, slice, unit or workload.

=> It matches ^[a-z][a-z_]. At least "_" is missing. Also this validation should specify somewhere if it's for v2 only or for both.

Kind regards,
Petr

[1] https://www.kernel.org/doc/Documentation/cgroup-v1/cgroups.txt
[2] https://docs.kernel.org/admin-guide/cgroup-v2.html#avoid-name-collisions

Signature · 1 line
Mailing list info: https://lists.linux.it/listinfo/ltp
#3 lore raw

Hello,

Quoted 7 lines · Petr Vorel

Hi Richie, Martchus,

@Richie, please add Fixes: tag when commit

I suppose it should be

Fixes 310da3784 ("Add new CGroups APIs")

but please check yourself.

Yes that is right.

Quoted 4 lines

Why this is useful? It helps to identify which test failures were false positives. Also, you actually not just fix a line character, but also do other validation, it would be worth to mention that.

The validation is primarily checking my assumptions. We don't want to cut the name off at a '_' then get more confusing errors in the future.

Quoted 11 lines
Quoted 10 lines

Am Mittwoch, 25. Oktober 2023, 13:05:33 CEST schrieb Richard Palethorpe via ltp:

lib/tst_cgroup.c __attribute__ ((nonnull, warn_unused_result))

Quoted 5 lines
lib/tst_cgroup.c +5 -0
+	switch (ctrl_name[l]) {
+	case '\n': break;
+	case '\0': break;
+	default:
+		tst_brk(TBROK, "Unexpected char in %s: %c", ctrl_name, 

ctrl_name[l]);

Quoted 4 lines
Quoted 4 lines

I'm wondering whether that's a bit too restrictive. Or is there any official documentation says that you really can only have the letters a-z in cgroup names (and not even A-Z). Otherwise it might be better to make this just a warning or allow any printable characters.

Well I assumed there wasn't, but it seems this was actually thought about and specified to some extent. I should have scanned the docs.

Quoted 4 lines

I guess for cgroup v1 [1]

The name should match [\w.-]+

Thats the name of the "hierarchy" AFAICT. I don't think that is the controller/subsystem name. Those characters would be a pain for naming things in C.

Quoted 5 lines

\w Matches a "word" character (alphanumeric plus "_", plus other connector punctuation chars plus Unicode marks). Also '.' and '-' can be used. => [A-Z.-] and others are valid names in v1. Although I'm not sure if cgroup_find_ctrl() is used on systems with cgroups v1.

None of the existing upstream controllers contain a _ unless we are missing one or more. However we should allow it, so I'll add it.

Quoted 6 lines

Also, shouldn't we check with MAX_CGROUP_TYPE_NAMELEN:

- name: should be initialized to a unique subsystem name. Should be no longer than MAX_CGROUP_TYPE_NAMELEN.

It's not exposed to userland or specified in the docs. I suppose I could issue a WARN if it is over 32 though. Mostly likely if that happens then there is a parsing error.

The kernel will also issue a WARN if the subsystem name is over 32.

Quoted 18 lines

For cgroup v2 [2] it looks to be:

All cgroup core interface files are prefixed with "cgroup." and each controller's interface files are prefixed with the controller name and a dot. A controller's name is composed of lower case alphabets and '_'s but never begins with an '_' so it can be used as the prefix character for collision avoidance. Also, interface file names won't start or end with terms which are often used in categorizing workloads such as job, service, slice, unit or workload.

=> It matches ^[a-z][a-z_]. At least "_" is missing. Also this validation should specify somewhere if it's for v2 only or for both.

Kind regards,
Petr

[1] https://www.kernel.org/doc/Documentation/cgroup-v1/cgroups.txt
[2] https://docs.kernel.org/admin-guide/cgroup-v2.html#avoid-name-collisions

Signature · 5 lines
Thank you,
Richard.

-- 
Mailing list info: https://lists.linux.it/listinfo/ltp
#4 lore raw
Quoted 6 lines

Hello,

Quoted 1 line · Petr Vorel

Hi Richie, Martchus,

Quoted 7 lines
Quoted 5 lines

@Richie, please add Fixes: tag when commit

I suppose it should be

Fixes 310da3784 ("Add new CGroups APIs")

but please check yourself.

Yes that is right.

+1

Quoted 6 lines
Quoted 3 lines

Why this is useful? It helps to identify which test failures were false positives. Also, you actually not just fix a line character, but also do other validation, it would be worth to mention that.

The validation is primarily checking my assumptions. We don't want to cut the name off at a '_' then get more confusing errors in the future.

+1, looking into the code and description once more, it makes sense.

Quoted 18 lines
Quoted 9 lines
Quoted 8 lines

Am Mittwoch, 25. Oktober 2023, 13:05:33 CEST schrieb Richard Palethorpe via ltp:

lib/tst_cgroup.c __attribute__ ((nonnull, warn_unused_result))

Quoted 5 lines
lib/tst_cgroup.c +5 -0
+	switch (ctrl_name[l]) {
+	case '\n': break;
+	case '\0': break;
+	default:
+		tst_brk(TBROK, "Unexpected char in %s: %c", ctrl_name, 

ctrl_name[l]);

Quoted 4 lines
Quoted 4 lines

I'm wondering whether that's a bit too restrictive. Or is there any official documentation says that you really can only have the letters a-z in cgroup names (and not even A-Z). Otherwise it might be better to make this just a warning or allow any printable characters.

Well I assumed there wasn't, but it seems this was actually thought about and specified to some extent. I should have scanned the docs.

Quoted 7 lines
Quoted 3 lines

I guess for cgroup v1 [1]

The name should match [\w.-]+

Thats the name of the "hierarchy" AFAICT. I don't think that is the controller/subsystem name. Those characters would be a pain for naming things in C.

Ah, right, I'm sorry.

Quoted 7 lines
Quoted 4 lines

\w Matches a "word" character (alphanumeric plus "_", plus other connector punctuation chars plus Unicode marks). Also '.' and '-' can be used. => [A-Z.-] and others are valid names in v1. Although I'm not sure if cgroup_find_ctrl() is used on systems with cgroups v1.

None of the existing upstream controllers contain a _ unless we are missing one or more. However we should allow it, so I'll add it.

+1

Quoted 9 lines
Quoted 4 lines

Also, shouldn't we check with MAX_CGROUP_TYPE_NAMELEN:

- name: should be initialized to a unique subsystem name. Should be no longer than MAX_CGROUP_TYPE_NAMELEN.

It's not exposed to userland or specified in the docs. I suppose I could issue a WARN if it is over 32 though. Mostly likely if that happens then there is a parsing error.

+1

Quoted 1 line

The kernel will also issue a WARN if the subsystem name is over 32.

Good. But warning in LTP will be more visible (not everybody parses dmesg).

Kind regards,
Petr

Quoted 18 lines
Quoted 18 lines

For cgroup v2 [2] it looks to be:

All cgroup core interface files are prefixed with "cgroup." and each controller's interface files are prefixed with the controller name and a dot. A controller's name is composed of lower case alphabets and '_'s but never begins with an '_' so it can be used as the prefix character for collision avoidance. Also, interface file names won't start or end with terms which are often used in categorizing workloads such as job, service, slice, unit or workload.

=> It matches ^[a-z][a-z_]. At least "_" is missing. Also this validation should specify somewhere if it's for v2 only or for both.

Kind regards,
Petr

[1] https://www.kernel.org/doc/Documentation/cgroup-v1/cgroups.txt
[2] https://docs.kernel.org/admin-guide/cgroup-v2.html#avoid-name-collisions

Signature · 1 line
Mailing list info: https://lists.linux.it/listinfo/ltp