Skip to contentSkip to conversation

livepatch: introduce 'stack_order' sysfs interface to klp_patch

PATCH In review Wardenjohn 3 participants 9 messages Sep 25, 2024 – Sep 29, 2024 Live patching

lore
Share & tools
lore

Wardenjohn

author
#0 lore raw

cover livepatch: introduce 'stack_order' sysfs interface to klp_patch

As previous discussion, maintainers think that patch-level sysfs interface is the only acceptable way to maintain the information of the order that klp_patch is applied to the system.

However, the previous patch introduce klp_ops into klp_func is a optimization methods of the patch introducing 'using' feature to klp_func.

But now, we don't support 'using' feature to klp_func and make 'klp_ops' patch not necessary.

Therefore, this new version is only introduce the sysfs feature of klp_patch 'stack_order'.

V1 -> V2:
1. According to the suggestion from Petr, to make the meaning more clear, rename 'order' to 'stack_order'.
2. According to the suggestion from Petr and Miroslav, this patch now move the calculating process to stack_order_show function. Adding klp_mutex lock protection.

Regards.
Wardenjohn.

author
#1 lore raw

PATCH 1/1 livepatch: introduce 'stack_order' sysfs interface to klp_patch

This feature can provide livepatch patch order information. With the order of sysfs interface of one klp_patch, we can use patch order to find out which function of the patch is now activate.

After the discussion, we decided that patch-level sysfs interface is the only accaptable way to introduce this information.

This feature is like:
cat /sys/kernel/livepatch/livepatch_1/stack_order -> 1 means this livepatch_1 module is the 1st klp patch applied.

cat /sys/kernel/livepatch/livepatch_module/stack_order -> N means this lviepatch_module is the Nth klp patch applied to the system.

kernel/livepatch/core.c +25 -0
index ecbc9b6aba3a..914b7cabf8fe 100644
@@ -346,6 +346,7 @@ int klp_apply_section_relocs(struct module *pmod, Elf_Shdr *sechdrs,
  * /sys/kernel/livepatch/<patch>/enabled
  * /sys/kernel/livepatch/<patch>/transition
  * /sys/kernel/livepatch/<patch>/force
+ * /sys/kernel/livepatch/<patch>/stack_order
  * /sys/kernel/livepatch/<patch>/<object>
  * /sys/kernel/livepatch/<patch>/<object>/patched
  * /sys/kernel/livepatch/<patch>/<object>/<function,sympos>
@@ -443,13 +444,37 @@ static ssize_t force_store(struct kobject *kobj, struct kobj_attribute *attr,
 	return count;
 }
 
+static ssize_t stack_order_show(struct kobject *kobj,
+				struct kobj_attribute *attr, char *buf)
+{
+	struct klp_patch *patch, *this_patch;
+	int stack_order = 0;
+
+	this_patch = container_of(kobj, struct klp_patch, kobj);
+
+	/* make sure the calculate of patch order correct */
+	mutex_lock(&klp_mutex);
+
+	klp_for_each_patch(patch) {
+		stack_order++;
+		if (patch == this_patch)
+			break;
+	}
+
+	mutex_unlock(&klp_mutex);
+	return sysfs_emit(buf, "%d\n", stack_order);
+}
+
+
 static struct kobj_attribute enabled_kobj_attr = __ATTR_RW(enabled);
 static struct kobj_attribute transition_kobj_attr = __ATTR_RO(transition);
 static struct kobj_attribute force_kobj_attr = __ATTR_WO(force);
+static struct kobj_attribute stack_order_kobj_attr = __ATTR_RO(stack_order);
 static struct attribute *klp_patch_attrs[] = {
 	&enabled_kobj_attr.attr,
 	&transition_kobj_attr.attr,
 	&force_kobj_attr.attr,
+	&stack_order_kobj_attr.attr,
 	NULL
 };
 ATTRIBUTE_GROUPS(klp_patch);
Signature · 1 line
2.18.2
maintainer
#2 lore raw

Hi,

I would change the subject to something like

"livepatch: Add "stack_order" sysfs attribute"

to keep it somehow consistent with what we have there so far.

Quoted 9 lines · Wardenjohn

This feature can provide livepatch patch order information. With the order of sysfs interface of one klp_patch, we can use patch order to find out which function of the patch is now activate.

After the discussion, we decided that patch-level sysfs interface is the only accaptable way to introduce this information.

Quoted 7 lines

This feature is like:
cat /sys/kernel/livepatch/livepatch_1/stack_order -> 1 means this livepatch_1 module is the 1st klp patch applied.

cat /sys/kernel/livepatch/livepatch_module/stack_order -> N means this lviepatch_module is the Nth klp patch applied to the system.

Perhaps something like

"
Add "stack_order" sysfs attribute which holds the order in which a live patch module was loaded into the system. A user can then determine an active live patched version of a function.

cat /sys/kernel/livepatch/livepatch_1/stack_order -> 1

means that livepatch_1 is the first live patch applied

cat /sys/kernel/livepatch/livepatch_module/stack_order -> N

means that livepatch_module is the Nth live patch applied
"
?

Quoted 4 lines

How do you prepare your patches?

"---" delimiter is missing here.

Quoted 15 lines
kernel/livepatch/core.c +1 -0
index ecbc9b6aba3a..914b7cabf8fe 100644
@@ -346,6 +346,7 @@ int klp_apply_section_relocs(struct module *pmod, Elf_Shdr *sechdrs,
  * /sys/kernel/livepatch/<patch>/enabled
  * /sys/kernel/livepatch/<patch>/transition
  * /sys/kernel/livepatch/<patch>/force
+ * /sys/kernel/livepatch/<patch>/stack_order
  * /sys/kernel/livepatch/<patch>/<object>
  * /sys/kernel/livepatch/<patch>/<object>/patched
  * /sys/kernel/livepatch/<patch>/<object>/<function,sympos>
@@ -443,13 +444,37 @@ static ssize_t force_store(struct kobject *kobj, struct kobj_attribute *attr,
 	return count;
 }

kernel/livepatch/core.c static ssize_t force_store(struct kobject *kobj, struct kobj_attribute *attr, · line 455

Quoted 10 lines
kernel/livepatch/core.c +9 -0
 
+static ssize_t stack_order_show(struct kobject *kobj,
+				struct kobj_attribute *attr, char *buf)
+{
+	struct klp_patch *patch, *this_patch;
+	int stack_order = 0;
+
+	this_patch = container_of(kobj, struct klp_patch, kobj);
+
+	/* make sure the calculate of patch order correct */

The comment is not necessary.

kernel/livepatch/core.c static ssize_t force_store(struct kobject *kobj, struct kobj_attribute *attr, · line 464

Quoted 9 lines
kernel/livepatch/core.c +9 -0
+	mutex_lock(&klp_mutex);
+
+	klp_for_each_patch(patch) {
+		stack_order++;
+		if (patch == this_patch)
+			break;
+	}
+
+	mutex_unlock(&klp_mutex);

Please add an empty line before return here.

kernel/livepatch/core.c static ssize_t force_store(struct kobject *kobj, struct kobj_attribute *attr, · line 466

Quoted 2 lines
kernel/livepatch/core.c +2 -0
+       return sysfs_emit(buf, "%d\n", stack_order);
+}

Miroslav

#3 lore raw

Hi, Miroslav!

Quoted 18 lines

Perhaps something like

"
Add "stack_order" sysfs attribute which holds the order in which a live patch module was loaded into the system. A user can then determine an active live patched version of a function.

cat /sys/kernel/livepatch/livepatch_1/stack_order -> 1

means that livepatch_1 is the first live patch applied

cat /sys/kernel/livepatch/livepatch_module/stack_order -> N

means that livepatch_module is the Nth live patch applied "
?

Quoted 8 lines
Quoted 4 lines

How do you prepare your patches?

"---" delimiter is missing here.

I will commit my changes with 'git commit -m' option. Then, I use 'git format-patch' to generate my patches. After my patches is ready, I would use 'git send-email' to send my directory containing my patches and cover letter.

Is there any step I missed?

Quoted 41 lines
Quoted 16 lines
kernel/livepatch/core.c +1 -0
index ecbc9b6aba3a..914b7cabf8fe 100644
@@ -346,6 +346,7 @@ int klp_apply_section_relocs(struct module *pmod, Elf_Shdr *sechdrs,
 * /sys/kernel/livepatch/<patch>/enabled
 * /sys/kernel/livepatch/<patch>/transition
 * /sys/kernel/livepatch/<patch>/force
+ * /sys/kernel/livepatch/<patch>/stack_order
 * /sys/kernel/livepatch/<patch>/<object>
 * /sys/kernel/livepatch/<patch>/<object>/patched
 * /sys/kernel/livepatch/<patch>/<object>/<function,sympos>
@@ -443,13 +444,37 @@ static ssize_t force_store(struct kobject *kobj, struct kobj_attribute *attr,
 return count;
 }

kernel/livepatch/core.c static ssize_t force_store(struct kobject *kobj, struct kobj_attribute *attr, · line 455

Quoted 10 lines
kernel/livepatch/core.c +9 -0
 
+static ssize_t stack_order_show(struct kobject *kobj,
+ struct kobj_attribute *attr, char *buf)
+{
+ struct klp_patch *patch, *this_patch;
+ int stack_order = 0;
+
+ this_patch = container_of(kobj, struct klp_patch, kobj);
+
+ /* make sure the calculate of patch order correct */

The comment is not necessary.

kernel/livepatch/core.c static ssize_t force_store(struct kobject *kobj, struct kobj_attribute *attr, · line 464

Quoted 9 lines
kernel/livepatch/core.c +9 -0
+ mutex_lock(&klp_mutex);
+
+ klp_for_each_patch(patch) {
+ stack_order++;
+ if (patch == this_patch)
+ break;
+ }
+
+ mutex_unlock(&klp_mutex);

Please add an empty line before return here.

Quoted 4 lines

kernel/livepatch/core.c static ssize_t force_store(struct kobject *kobj, struct kobj_attribute *attr, · line 466

Quoted 2 lines
kernel/livepatch/core.c +2 -0
+       return sysfs_emit(buf, "%d\n", stack_order);
+}

Miroslav

And the rest of the suggestions will be fix in the next version.

Regards.
Wardenjohn.

#4 lore raw
In reply to Miroslav Benes

Hi, Miroslav!

Quoted 6 lines · Miroslav Benes

How do you prepare your patches?

"---" delimiter is missing here.

I seem to found out what cause this problem.

I seemed to use 'git format-patch' with '-p' option which will make my patch have no diff state.

Sorry for this. A newer version was sent again.
Please review the V3 patch.

Thank you!
Wardenjohn.

author
#5 lore raw

PATCH 1/1 Documentation: Add description to stack_order interface

In reply to Wardenjohn

Update description of klp_patch stack_order sysfs interface to livepatch ABI documentation.

Documentation/ABI/testing/sysfs-kernel-livepatch +8 -0
index a5df9b4910dc..9cad725a69c7 100644
@@ -47,6 +47,14 @@ Description:
 		disabled when the feature is used. See
 		Documentation/livepatch/livepatch.rst for more information.
 
+What:           /sys/kernel/livepatch/<patch>/stack_order
+Date:           Sep 2024
+KernelVersion:  6.12.0
+Contact:        [email protected]
+Description:
+		This attribute record the stack order of this livepatch module
+		applied to the running system.
+
 What:		/sys/kernel/livepatch/<patch>/<object>
 Date:		Nov 2014
 KernelVersion:	3.19.0
Signature · 1 line
2.18.2
maintainer
#6 lore raw

Re: [PATCH] Documentation: Add description to stack_order interface

Quoted 10 lines · Wardenjohn

Update description of klp_patch stack_order sysfs interface to livepatch ABI documentation.

Documentation/ABI/testing/sysfs-kernel-livepatch +0 -0
index a5df9b4910dc..9cad725a69c7 100644

Documentation/ABI/testing/sysfs-kernel-livepatch Description: · line 56

Quoted 10 lines
Documentation/ABI/testing/sysfs-kernel-livepatch +7 -0
@@ -47,6 +47,14 @@ Description:
 		disabled when the feature is used. See
 		Documentation/livepatch/livepatch.rst for more information.
 
+What:           /sys/kernel/livepatch/<patch>/stack_order
+Date:           Sep 2024
+KernelVersion:  6.12.0
+Contact:        [email protected]
+Description:
+		This attribute record the stack order of this livepatch module
+		applied to the running system.

"The attribute holds the stack order of a live patch module applied to the running system." ?

Please also squash the patch into the previous one. It belongs there.

Thank you,
Miroslav

Acked-by on the v1 series
#7 lore raw
In reply to Wardenjohn
Quoted 10 lines · Wardenjohn

As previous discussion, maintainers think that patch-level sysfs interface is the
only acceptable way to maintain the information of the order that klp_patch is
applied to the system.

However, the previous patch introduce klp_ops into klp_func is a optimization
methods of the patch introducing 'using' feature to klp_func.

Quoted 7 lines

But now, we don't support 'using' feature to klp_func and make 'klp_ops' patch
not necessary.

Therefore, this new version is only introduce the sysfs feature of klp_patch
'stack_order'.

The approach seems ok to me, but I would like to see selftests for this new attribute. We have been trying to add more and more selftests for existing known behavior, so IMO adding a new attribute should contain a new test to exercise the correct behavior.

Other than that, for the series:

Quoted 13 lines

V1 -> V2:
1. According to the suggestion from Petr, to make the meaning more clear, rename
'order' to 'stack_order'.
2. According to the suggestion from Petr and Miroslav, this patch now move the
calculating process to stack_order_show function. Adding klp_mutex lock protection.

Regards.
Wardenjohn.

#8 lore raw
Quoted 22 lines · Marcos Paulo de Souza
Quoted 10 lines · Wardenjohn

As previous discussion, maintainers think that patch-level sysfs interface is the
only acceptable way to maintain the information of the order that klp_patch is
applied to the system.

However, the previous patch introduce klp_ops into klp_func is a optimization
methods of the patch introducing 'using' feature to klp_func.

Quoted 7 lines

But now, we don't support 'using' feature to klp_func and make 'klp_ops' patch
not necessary.

Therefore, this new version is only introduce the sysfs feature of klp_patch
'stack_order'.

Quoted 9 lines

The approach seems ok to me, but I would like to see selftests for this new attribute. We have been trying to add more and more selftests for existing known behavior, so IMO adding a new attribute should contain a new test to exercise the correct behavior.

Other than that, for the series:

Hi, Macros!

Thanks a lot.

I will add selftest case for it as soon as possible.

Regards.
Wardenjohn.

(This email is resent because it seemed not sent to LKML...)