Skip to contentSkip to conversation

livepatch: Add a new sysfs interface replace

PATCH In review Yafang Shao 3 participants 5 messages Jun 7, 2024 Live patching

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 livepatch: Add a new sysfs interface replace

When building a livepatch, a user can set it to be either an atomic-replace livepatch or a non-atomic-replace livepatch. However, it is not easy to identify whether a livepatch is atomic-replace or not until it actually replaces some old livepatches. This can lead to mistakes in a mixed atomic-replace and non-atomic-replace environment, especially when transitioning all livepatches from non-atomic-replace to atomic-replace in a large fleet of servers.

To address this issue, a new sysfs interface called 'replace' is introduced in this patch. The result after this change is as follows:

  $ cat /sys/kernel/livepatch/livepatch-non_replace/replace
  0
  $ cat /sys/kernel/livepatch/livepatch-replace/replace
  1
---
 Documentation/ABI/testing/sysfs-kernel-livepatch |  8 ++++++++
 kernel/livepatch/core.c                          | 12 ++++++++++++
 2 files changed, 20 insertions(+)
Documentation/ABI/testing/sysfs-kernel-livepatch +8 -0
index a5df9b4910dc..3735d868013d 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>/replace
+Date:		Jun 2024
+KernelVersion:	6.11.0
+Contact:	[email protected]
+Description:
+		An attribute which indicates whether the patch supports
+		atomic-replace.
+
 What:		/sys/kernel/livepatch/<patch>/<object>
 Date:		Nov 2014
 KernelVersion:	3.19.0
kernel/livepatch/core.c +12 -0
index 52426665eecc..0e9832f146f1 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>/replace
  * /sys/kernel/livepatch/<patch>/<object>
  * /sys/kernel/livepatch/<patch>/<object>/patched
  * /sys/kernel/livepatch/<patch>/<object>/<function,sympos>
@@ -443,13 +444,24 @@ static ssize_t force_store(struct kobject *kobj, struct kobj_attribute *attr,
 	return count;
 }
 
+static ssize_t replace_show(struct kobject *kobj,
+			    struct kobj_attribute *attr, char *buf)
+{
+	struct klp_patch *patch;
+
+	patch = container_of(kobj, struct klp_patch, kobj);
+	return snprintf(buf, PAGE_SIZE-1, "%d\n", patch->replace);
+}
+
 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 replace_kobj_attr = __ATTR_RO(replace);
 static struct attribute *klp_patch_attrs[] = {
 	&enabled_kobj_attr.attr,
 	&transition_kobj_attr.attr,
 	&force_kobj_attr.attr,
+	&replace_kobj_attr.attr,
 	NULL
 };
 ATTRIBUTE_GROUPS(klp_patch);
Signature · 1 line
2.39.1
maintainer
#1 lore raw

Hi,

I think the subject should be something like

"livepatch: Add "replace" sysfs attribute"

or use a different way to stress "replace" is the name.

Quoted 2 lines · Yafang Shao

When building a livepatch, a user can set it to be either an atomic-replace livepatch or a non-atomic-replace livepatch.

I am not a native speaker but I would drop '-' everywhere in this context.

Quoted 3 lines

However, it is not easy to
identify whether a livepatch is atomic-replace or not until it actually replaces some old livepatches.

Ok.

Quoted 4 lines

This can lead to mistakes in a mixed
atomic-replace and non-atomic-replace environment, especially when transitioning all livepatches from non-atomic-replace to atomic-replace in a large fleet of servers.

Out of curiosity, could you describe your setup in more detail here? It is interesting to mix different type of livepatches so I would like to learn more.

Quoted 47 lines

To address this issue, a new sysfs interface called 'replace' is introduced in this patch. The result after this change is as follows:

  $ cat /sys/kernel/livepatch/livepatch-non_replace/replace
  0
  $ cat /sys/kernel/livepatch/livepatch-replace/replace
  1
---
 Documentation/ABI/testing/sysfs-kernel-livepatch |  8 ++++++++
 kernel/livepatch/core.c                          | 12 ++++++++++++
 2 files changed, 20 insertions(+)
Documentation/ABI/testing/sysfs-kernel-livepatch +8 -0
index a5df9b4910dc..3735d868013d 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>/replace
+Date:		Jun 2024
+KernelVersion:	6.11.0
+Contact:	[email protected]
+Description:
+		An attribute which indicates whether the patch supports
+		atomic-replace.
+
 What:		/sys/kernel/livepatch/<patch>/<object>
 Date:		Nov 2014
 KernelVersion:	3.19.0
kernel/livepatch/core.c +1 -0
index 52426665eecc..0e9832f146f1 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>/replace
  * /sys/kernel/livepatch/<patch>/<object>
  * /sys/kernel/livepatch/<patch>/<object>/patched
  * /sys/kernel/livepatch/<patch>/<object>/<function,sympos>

kernel/livepatch/core.c static ssize_t force_store(struct kobject *kobj, struct kobj_attribute *attr, · line 453

Quoted 10 lines
kernel/livepatch/core.c +7 -0
@@ -443,13 +444,24 @@ static ssize_t force_store(struct kobject *kobj, struct kobj_attribute *attr,
 	return count;
 }
 
+static ssize_t replace_show(struct kobject *kobj,
+			    struct kobj_attribute *attr, char *buf)
+{
+	struct klp_patch *patch;
+
+	patch = container_of(kobj, struct klp_patch, kobj);
+	return snprintf(buf, PAGE_SIZE-1, "%d\n", patch->replace);

It might be better to use sysfs_emit() here. See patched_show() in the same file. There are still snprintf() occurrences and it might be a separate cleanup patch if you are interested.

Regards,
Miroslav

#2 lore raw
Quoted 97 lines · Miroslav Benes

Hi,

I think the subject should be something like

"livepatch: Add "replace" sysfs attribute"

or use a different way to stress "replace" is the name.

Quoted 3 lines · Yafang Shao

When building a livepatch, a user can set it to be either an atomic-replace
livepatch or a non-atomic-replace livepatch.

I am not a native speaker but I would drop '-' everywhere in this context.

Quoted 4 lines

However, it is not easy to
identify whether a livepatch is atomic-replace or not until it actually
replaces some old livepatches.

Ok.

Quoted 5 lines

This can lead to mistakes in a mixed
atomic-replace and non-atomic-replace environment, especially when transitioning all livepatches from non-atomic-replace to atomic- replace in
a large fleet of servers.

Out of curiosity, could you describe your setup in more detail here? It is
interesting to mix different type of livepatches so I would like to learn
more.

Quoted 61 lines

To address this issue, a new sysfs interface called 'replace' is introduced
in this patch. The result after this change is as follows:

  $ cat /sys/kernel/livepatch/livepatch-non_replace/replace
  0
  $ cat /sys/kernel/livepatch/livepatch-replace/replace
  1

---
 Documentation/ABI/testing/sysfs-kernel-livepatch |  8 ++++++++ kernel/livepatch/core.c                          | 12 ++++++++++++

 2 files changed, 20 insertions(+)
Documentation/ABI/testing/sysfs-kernel-livepatch +0 -0
index a5df9b4910dc..3735d868013d 100644
@@ -47,6 +47,14 @@ Description:

  disabled when the feature is used. See Documentation/livepatch/livepatch.rst for more information.

Documentation/ABI/testing/sysfs-kernel-livepatch +6 -0
+What:		/sys/kernel/livepatch/<patch>/replace
+Date:		Jun 2024
+KernelVersion:	6.11.0
+Contact:	[email protected]
+Description:
+		An attribute which indicates whether the patch

supports

Documentation/ABI/testing/sysfs-kernel-livepatch +2 -0
+		atomic-replace.
+

 What: /sys/kernel/livepatch/<patch>/<object>
 Date: Nov 2014
 KernelVersion: 3.19.0

kernel/livepatch/core.c +1 -0
index 52426665eecc..0e9832f146f1 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>/replace
   * /sys/kernel/livepatch/<patch>/<object>
   * /sys/kernel/livepatch/<patch>/<object>/patched

  * /sys/kernel/livepatch/<patch>/<object>/<function,sympos>

kernel/livepatch/core.c +0 -0
@@ -443,13 +444,24 @@ static ssize_t force_store(struct kobject
 *kobj, struct kobj_attribute *attr,
  	return count;
  }

kernel/livepatch/core.c +2 -0
+static ssize_t replace_show(struct kobject *kobj,
+			    struct kobj_attribute *attr, char

*buf)

kernel/livepatch/core.c +2 -0
+{
+	struct klp_patch *patch;
Quoted 8 lines

kernel/livepatch/core.c static ssize_t force_store(struct kobject *kobj, struct kobj_attribute *attr, · line 453

Quoted 3 lines
kernel/livepatch/core.c +3 -0
+
+	patch = container_of(kobj, struct klp_patch, kobj);
+	return snprintf(buf, PAGE_SIZE-1, "%d\n", patch->replace);

It might be better to use sysfs_emit() here. See patched_show() in the
same file. There are still snprintf() occurrences and it might be a separate cleanup patch if you are interested.

Hi Yafang, thanks for the patch!

Besides Miroslav's comments, I think that a new kselftest exercising this attribute would be good, since we have tests for other attributes as well.

Thanks,
Marcos

Quoted 4 lines

Regards,
Miroslav

#3 lore raw
Quoted 108 lines · Marcos Paulo de Souza
Quoted 92 lines · Miroslav Benes

Hi,

I think the subject should be something like

"livepatch: Add "replace" sysfs attribute"

or use a different way to stress "replace" is the name.

Quoted 3 lines · Yafang Shao

When building a livepatch, a user can set it to be either an atomic-replace
livepatch or a non-atomic-replace livepatch.

I am not a native speaker but I would drop '-' everywhere in this context.

Quoted 4 lines

However, it is not easy to
identify whether a livepatch is atomic-replace or not until it actually
replaces some old livepatches.

Ok.

Quoted 5 lines

This can lead to mistakes in a mixed
atomic-replace and non-atomic-replace environment, especially when transitioning all livepatches from non-atomic-replace to atomic- replace in
a large fleet of servers.

Out of curiosity, could you describe your setup in more detail here? It is
interesting to mix different type of livepatches so I would like to learn
more.

Quoted 56 lines

To address this issue, a new sysfs interface called 'replace' is introduced
in this patch. The result after this change is as follows:

  $ cat /sys/kernel/livepatch/livepatch-non_replace/replace
  0
  $ cat /sys/kernel/livepatch/livepatch-replace/replace
  1
---
 Documentation/ABI/testing/sysfs-kernel-livepatch |  8 ++++++++
 kernel/livepatch/core.c                          | 12 ++++++++++++
 2 files changed, 20 insertions(+)
Documentation/ABI/testing/sysfs-kernel-livepatch +0 -0
index a5df9b4910dc..3735d868013d 100644
@@ -47,6 +47,14 @@ Description:
            disabled when the feature is used. See
            Documentation/livepatch/livepatch.rst for more

information.

Documentation/ABI/testing/sysfs-kernel-livepatch +6 -0
+What:              /sys/kernel/livepatch/<patch>/replace
+Date:              Jun 2024
+KernelVersion:     6.11.0
+Contact:   [email protected]
+Description:
+           An attribute which indicates whether the patch

supports

Documentation/ABI/testing/sysfs-kernel-livepatch +2 -0
+           atomic-replace.
+
 What:              /sys/kernel/livepatch/<patch>/<object>
 Date:              Nov 2014
 KernelVersion:     3.19.0
kernel/livepatch/core.c +1 -0
index 52426665eecc..0e9832f146f1 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>/replace
  * /sys/kernel/livepatch/<patch>/<object>
  * /sys/kernel/livepatch/<patch>/<object>/patched

* /sys/kernel/livepatch/<patch>/<object>/<function,sympos>

kernel/livepatch/core.c +0 -0
@@ -443,13 +444,24 @@ static ssize_t force_store(struct kobject
 *kobj, struct kobj_attribute *attr,
    return count;
 }
 
Quoted 13 lines

kernel/livepatch/core.c static ssize_t force_store(struct kobject *kobj, struct kobj_attribute *attr, · line 453

Quoted 8 lines
kernel/livepatch/core.c +7 -0
+static ssize_t replace_show(struct kobject *kobj,
+                       struct kobj_attribute *attr, char
 *buf)
+{
+   struct klp_patch *patch;
+
+   patch = container_of(kobj, struct klp_patch, kobj);
+   return snprintf(buf, PAGE_SIZE-1, "%d\n", patch->replace);

It might be better to use sysfs_emit() here. See patched_show() in the
same file. There are still snprintf() occurrences and it might be a separate cleanup patch if you are interested.

Quoted 5 lines

Hi Yafang, thanks for the patch!

Besides Miroslav's comments, I think that a new kselftest exercising this attribute would be good, since we have tests for other attributes as well.

I will add it. Thanks for your suggestion.

Signature · 2 lines
Regards
Yafang
#4 lore raw
In reply to Miroslav Benes
Quoted 6 lines · Miroslav Benes

Hi,

I think the subject should be something like

"livepatch: Add "replace" sysfs attribute"

agreed.

Quoted 9 lines

or use a different way to stress "replace" is the name.

Quoted 2 lines · Yafang Shao

When building a livepatch, a user can set it to be either an atomic-replace livepatch or a non-atomic-replace livepatch.

I am not a native speaker but I would drop '-' everywhere in this context.

agreed. I'm not a native speaker either :)

Quoted 7 lines
Quoted 4 lines

However, it is not easy to
identify whether a livepatch is atomic-replace or not until it actually replaces some old livepatches.

Ok.

Quoted 8 lines
Quoted 4 lines

This can lead to mistakes in a mixed
atomic-replace and non-atomic-replace environment, especially when transitioning all livepatches from non-atomic-replace to atomic-replace in a large fleet of servers.

Out of curiosity, could you describe your setup in more detail here? It is interesting to mix different type of livepatches so I would like to learn more.

It seems I didn't describe it clearly. We are transitioning non atomic replace livepatches to atomic replace livepatches, which will take some time to complete in a large fleet of servers. In other words, some servers are still running non atomic replace livepatches while others are running atomic replace livepatches. As a result, the sysadmins have complained that they can't find a way to identify whether a livepatch is an atomic replace or not. It will be beneficial to show it directly.

Quoted 51 lines
Quoted 51 lines

To address this issue, a new sysfs interface called 'replace' is introduced in this patch. The result after this change is as follows:

  $ cat /sys/kernel/livepatch/livepatch-non_replace/replace
  0
  $ cat /sys/kernel/livepatch/livepatch-replace/replace
  1
---
 Documentation/ABI/testing/sysfs-kernel-livepatch |  8 ++++++++
 kernel/livepatch/core.c                          | 12 ++++++++++++
 2 files changed, 20 insertions(+)
Documentation/ABI/testing/sysfs-kernel-livepatch +8 -0
index a5df9b4910dc..3735d868013d 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>/replace
+Date:                Jun 2024
+KernelVersion:       6.11.0
+Contact:     [email protected]
+Description:
+             An attribute which indicates whether the patch supports
+             atomic-replace.
+
 What:                /sys/kernel/livepatch/<patch>/<object>
 Date:                Nov 2014
 KernelVersion:       3.19.0
kernel/livepatch/core.c +1 -0
index 52426665eecc..0e9832f146f1 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>/replace
  * /sys/kernel/livepatch/<patch>/<object>
  * /sys/kernel/livepatch/<patch>/<object>/patched
  * /sys/kernel/livepatch/<patch>/<object>/<function,sympos>
@@ -443,13 +444,24 @@ static ssize_t force_store(struct kobject *kobj, struct kobj_attribute *attr,
      return count;
 }
 
Quoted 11 lines

kernel/livepatch/core.c static ssize_t force_store(struct kobject *kobj, struct kobj_attribute *attr, · line 453

Quoted 7 lines
kernel/livepatch/core.c +7 -0
+static ssize_t replace_show(struct kobject *kobj,
+                         struct kobj_attribute *attr, char *buf)
+{
+     struct klp_patch *patch;
+
+     patch = container_of(kobj, struct klp_patch, kobj);
+     return snprintf(buf, PAGE_SIZE-1, "%d\n", patch->replace);

It might be better to use sysfs_emit() here. See patched_show() in the same file. There are still snprintf() occurrences and it might be a separate cleanup patch if you are interested.

I will do it. Thanks for your suggestion.

Signature · 2 lines
Regards
Yafang