kobject: support namespace aware udev

Restrict sending uevents to only those listeners operating in the same network namespace as the system init process. This is the first step toward allowing policy control of the forwarding of events to other namespaces in userspace.

Signed-off-by Michael J. Coss <[email protected]>

---
 lib/kobject_uevent.c | 4 ++++
 1 file changed, 4 insertions(+)

lib/kobject_uevent.c +4 -0

index f6c2c1e..d791e33 100644
@@ -295,6 +295,10 @@ int kobject_uevent_env(struct kobject *kobj, enum kobject_action action,
 		if (!netlink_has_listeners(uevent_sock, 1))
 			continue;
 
+		/* forward event only to the host systems network namespaces */
+		if (!net_eq(sock_net(uevent_sock), &init_net))
+			continue;
+
 		/* allocate message with the maximum possible size */
 		len = strlen(action_string) + strlen(devpath) + 2;
 		skb = alloc_skb(len + env->buflen, GFP_KERNEL);

Quoted 4 lines · Michael J. Coss

Restrict sending uevents to only those listeners operating in the same network namespace as the system init process. This is the first step toward allowing policy control of the forwarding of events to other namespaces in userspace.

This limitation whould be better if we only skipped network namespaces where you are sending spoofed uevents.

As it sits this has the possibility to break userspace.

Eric

Restrict sending uevents to only those listeners operating in the same network namespace as the system init process. This is the first step toward allowing policy control of the forwarding of events to other namespaces in userspace.

Signed-off-by Michael J. Coss <[email protected]>

---
 lib/kobject_uevent.c | 4 ++++
 1 file changed, 4 insertions(+)

lib/kobject_uevent.c +1 -0

index f6c2c1e..d791e33 100644
@@ -295,6 +295,10 @@ int kobject_uevent_env(struct kobject *kobj, enum
 kobject_action action,
                if (!netlink_has_listeners(uevent_sock, 1))
                        continue;
 
+               /* forward event only to the host systems network

namespaces */

lib/kobject_uevent.c +3 -1

+               if (!net_eq(sock_net(uevent_sock), &init_net))
+                       continue;
+
                /* allocate message with the maximum possible size */
                len = strlen(action_string) + strlen(devpath) + 2;
                skb = alloc_skb(len + env->buflen, GFP_KERNEL);
--

2.4.6

_______________________________________________
Containers mailing list Containers-cunTk1MwBs9QetFLy7KEm3xJsTq8ys+cHZ5vskTnxNA@public.gmane.org
https://lists.linuxfoundation.org/mailman/listinfo/containers

Adds capability to allow userspace programs to forward a given event to a specific network namespace as determined by the provided pid. In addition, support for a per-namespace kobject_sequence counter was added. Sysfs was modified to return the correct event counter based on the current network namespace.

Signed-off-by Michael J. Coss <[email protected]>

---
 include/linux/kobject.h     |  3 ++
 include/net/net_namespace.h |  3 ++
 kernel/ksysfs.c             | 12 ++++++
 lib/kobject_uevent.c        | 90 +++++++++++++++++++++++++++++++++++++++++++++
 4 files changed, 108 insertions(+)

include/linux/kobject.h +3 -0

index 637f670..d1bb509 100644
@@ -215,6 +215,9 @@ extern struct kobject *firmware_kobj;
 int kobject_uevent(struct kobject *kobj, enum kobject_action action);
 int kobject_uevent_env(struct kobject *kobj, enum kobject_action action,
 			char *envp[]);
+#if defined(CONFIG_UDEVNS) || defined(CONFIG_UDEVNS_MODULE)
+int kobject_uevent_forward(char *buf, size_t len, pid_t pid);
+#endif
 
 __printf(2, 3)
 int add_uevent_var(struct kobj_uevent_env *env, const char *format, ...);

include/net/net_namespace.h +3 -0

index e951453..a4013e5 100644
@@ -134,6 +134,9 @@ struct net {
 #if IS_ENABLED(CONFIG_MPLS)
 	struct netns_mpls	mpls;
 #endif
+#if defined(CONFIG_UDEVNS) || defined(CONFIG_UDEVNS_MODULE)
+	u64				kevent_seqnum;
+#endif
 	struct sock		*diag_nlsk;
 	atomic_t		fnhe_genid;
 };

kernel/ksysfs.c +0 -0

index 6683cce..4bc15fd 100644

kernel/ksysfs.c +9 -0

 #define KERNEL_ATTR_RO(_name) \
 static struct kobj_attribute _name##_attr = __ATTR_RO(_name)

@@ -33,6 +36,15 @@ static struct kobj_attribute _name##_attr = \
 static ssize_t uevent_seqnum_show(struct kobject *kobj,
 				  struct kobj_attribute *attr, char *buf)
 {
+#if defined(CONFIG_UDEVNS) || defined(CONFIG_UDEVNS_MODULE)
+	pid_t p = task_pid_vnr(current);
+	struct net *n = get_net_ns_by_pid(p);
+
+	if (n != ERR_PTR(-ESRCH)) {
+		if (!net_eq(n, &init_net))
+			return sprintf(buf, "%llu\n", n->kevent_seqnum);
+	}
+#endif
 	return sprintf(buf, "%llu\n", (unsigned long long)uevent_seqnum);
 }
 KERNEL_ATTR_RO(uevent_seqnum);

lib/kobject_uevent.c +32 -0

index d791e33..7589745 100644
@@ -379,6 +379,96 @@ int kobject_uevent(struct kobject *kobj, enum kobject_action action)
 }
 EXPORT_SYMBOL_GPL(kobject_uevent);
 
+#if defined(CONFIG_UDEVNS) || defined(CONFIG_UDEVNS_MODULE)
+/**
+ * kobject_uevent_forward - forward event to specified network namespace
+ *
+ * @buf: event buffer
+ * @len: event length
+ * @pid: pid of network namespace
+ *
+ * Returns 0 if kobject_uevent_forward() is completed with success or the
+ * corresponding error when it fails.
+ */
+int kobject_uevent_forward(char *buf, size_t len, pid_t pid)
+{
+	int retval = 0;
+#if defined(CONFIG_NET)
+	struct uevent_sock *ue_sk;
+	struct net *pns;
+	char *p;
+	u64 num;
+
+	/* grab the network namespace of the provided pid */
+	pns = get_net_ns_by_pid(pid);
+	if (pns == ERR_PTR(-ESRCH))
+		return -ESRCH;
+
+	/* find sequence number in buffer */
+	p = buf;
+	num = 0;
+	while (p < (buf + len)) {
+		if (strncmp(p, "SEQNUM=", 7) == 0) {
+			int r;
+

Quoted 47 lines · Michael J. Coss

Adds capability to allow userspace programs to forward a given event to a specific network namespace as determined by the provided pid. In addition, support for a per-namespace kobject_sequence counter was added. Sysfs was modified to return the correct event counter based on the current network namespace.

Signed-off-by Michael J. Coss <[email protected]>

---
 include/linux/kobject.h     |  3 ++
 include/net/net_namespace.h |  3 ++
 kernel/ksysfs.c             | 12 ++++++
 lib/kobject_uevent.c        | 90 +++++++++++++++++++++++++++++++++++++++++++++
 4 files changed, 108 insertions(+)

include/linux/kobject.h +3 -0

index 637f670..d1bb509 100644
@@ -215,6 +215,9 @@ extern struct kobject *firmware_kobj;
 int kobject_uevent(struct kobject *kobj, enum kobject_action action);
 int kobject_uevent_env(struct kobject *kobj, enum kobject_action action,
 			char *envp[]);
+#if defined(CONFIG_UDEVNS) || defined(CONFIG_UDEVNS_MODULE)
+int kobject_uevent_forward(char *buf, size_t len, pid_t pid);
+#endif
 
 __printf(2, 3)
 int add_uevent_var(struct kobj_uevent_env *env, const char *format, ...);

include/net/net_namespace.h +3 -0

index e951453..a4013e5 100644
@@ -134,6 +134,9 @@ struct net {
 #if IS_ENABLED(CONFIG_MPLS)
 	struct netns_mpls	mpls;
 #endif
+#if defined(CONFIG_UDEVNS) || defined(CONFIG_UDEVNS_MODULE)
+	u64				kevent_seqnum;
+#endif
 	struct sock		*diag_nlsk;
 	atomic_t		fnhe_genid;
 };

kernel/ksysfs.c +0 -0

index 6683cce..4bc15fd 100644

kernel/ksysfs.c line 26

Quoted 6 lines

kernel/ksysfs.c +3 -0

@@ -21,6 +21,9 @@
 #include <linux/compiler.h>
 
 #include <linux/rcupdate.h>	/* rcu_expedited */
+#if defined(CONFIG_UDEVNS) || defined(CONFIG_UDEVNS_MODULE)
+#include <net/net_namespace.h>
+#endif

#if isn't needed here, right?

Quoted 59 lines

kernel/ksysfs.c +9 -0

 #define KERNEL_ATTR_RO(_name) \
 static struct kobj_attribute _name##_attr = __ATTR_RO(_name)

@@ -33,6 +36,15 @@ static struct kobj_attribute _name##_attr = \
 static ssize_t uevent_seqnum_show(struct kobject *kobj,
 				  struct kobj_attribute *attr, char *buf)
 {
+#if defined(CONFIG_UDEVNS) || defined(CONFIG_UDEVNS_MODULE)
+	pid_t p = task_pid_vnr(current);
+	struct net *n = get_net_ns_by_pid(p);
+
+	if (n != ERR_PTR(-ESRCH)) {
+		if (!net_eq(n, &init_net))
+			return sprintf(buf, "%llu\n", n->kevent_seqnum);
+	}
+#endif
 	return sprintf(buf, "%llu\n", (unsigned long long)uevent_seqnum);
 }
 KERNEL_ATTR_RO(uevent_seqnum);

lib/kobject_uevent.c +32 -0

index d791e33..7589745 100644
@@ -379,6 +379,96 @@ int kobject_uevent(struct kobject *kobj, enum kobject_action action)
 }
 EXPORT_SYMBOL_GPL(kobject_uevent);
 
+#if defined(CONFIG_UDEVNS) || defined(CONFIG_UDEVNS_MODULE)
+/**
+ * kobject_uevent_forward - forward event to specified network namespace
+ *
+ * @buf: event buffer
+ * @len: event length
+ * @pid: pid of network namespace
+ *
+ * Returns 0 if kobject_uevent_forward() is completed with success or the
+ * corresponding error when it fails.
+ */
+int kobject_uevent_forward(char *buf, size_t len, pid_t pid)
+{
+	int retval = 0;
+#if defined(CONFIG_NET)
+	struct uevent_sock *ue_sk;
+	struct net *pns;
+	char *p;
+	u64 num;
+
+	/* grab the network namespace of the provided pid */
+	pns = get_net_ns_by_pid(pid);
+	if (pns == ERR_PTR(-ESRCH))
+		return -ESRCH;
+
+	/* find sequence number in buffer */
+	p = buf;
+	num = 0;
+	while (p < (buf + len)) {
+		if (strncmp(p, "SEQNUM=", 7) == 0) {
+			int r;
+

lib/kobject_uevent.c int kobject_uevent(struct kobject *kobj, enum kobject_action action) · line 423

Quoted 10 lines

lib/kobject_uevent.c +10 -0

+			p += 7;
+			r = kstrtoull(p, 10, &num);
+			if (r) {
+				put_net(pns);
+				return r;
+			}
+			break;
+		}
+		p += (strlen(p) + 1);
+	}

Ok, that's crazy, replacing an existing sequence number with a sequence number of the namespace? An interesting hack, yes, but something we want to maintain for the next 20+ years, no. Surely there's a better way to do this?

But again, I'm not sold on this whole idea anyway.

greg k-h

Quoted 132 lines · Greg KH

Quoted 47 lines · Michael J. Coss

Adds capability to allow userspace programs to forward a given event to a specific network namespace as determined by the provided pid. In addition, support for a per-namespace kobject_sequence counter was added. Sysfs was modified to return the correct event counter based on the current network namespace.

Signed-off-by Michael J. Coss <[email protected]>

---
 include/linux/kobject.h     |  3 ++
 include/net/net_namespace.h |  3 ++
 kernel/ksysfs.c             | 12 ++++++
 lib/kobject_uevent.c        | 90 +++++++++++++++++++++++++++++++++++++++++++++
 4 files changed, 108 insertions(+)

include/linux/kobject.h +3 -0

index 637f670..d1bb509 100644
@@ -215,6 +215,9 @@ extern struct kobject *firmware_kobj;
 int kobject_uevent(struct kobject *kobj, enum kobject_action action);
 int kobject_uevent_env(struct kobject *kobj, enum kobject_action action,
 			char *envp[]);
+#if defined(CONFIG_UDEVNS) || defined(CONFIG_UDEVNS_MODULE)
+int kobject_uevent_forward(char *buf, size_t len, pid_t pid);
+#endif
 
 __printf(2, 3)
 int add_uevent_var(struct kobj_uevent_env *env, const char *format, ...);

include/net/net_namespace.h +3 -0

index e951453..a4013e5 100644
@@ -134,6 +134,9 @@ struct net {
 #if IS_ENABLED(CONFIG_MPLS)
 	struct netns_mpls	mpls;
 #endif
+#if defined(CONFIG_UDEVNS) || defined(CONFIG_UDEVNS_MODULE)
+	u64				kevent_seqnum;
+#endif
 	struct sock		*diag_nlsk;
 	atomic_t		fnhe_genid;
 };

kernel/ksysfs.c +0 -0

index 6683cce..4bc15fd 100644

kernel/ksysfs.c line 26

Quoted 6 lines

kernel/ksysfs.c +3 -0

@@ -21,6 +21,9 @@
 #include <linux/compiler.h>
 
 #include <linux/rcupdate.h>	/* rcu_expedited */
+#if defined(CONFIG_UDEVNS) || defined(CONFIG_UDEVNS_MODULE)
+#include <net/net_namespace.h>
+#endif

#if isn't needed here, right?

Quoted 59 lines

kernel/ksysfs.c +9 -0

 #define KERNEL_ATTR_RO(_name) \
 static struct kobj_attribute _name##_attr = __ATTR_RO(_name)

@@ -33,6 +36,15 @@ static struct kobj_attribute _name##_attr = \
 static ssize_t uevent_seqnum_show(struct kobject *kobj,
 				  struct kobj_attribute *attr, char *buf)
 {
+#if defined(CONFIG_UDEVNS) || defined(CONFIG_UDEVNS_MODULE)
+	pid_t p = task_pid_vnr(current);
+	struct net *n = get_net_ns_by_pid(p);
+
+	if (n != ERR_PTR(-ESRCH)) {
+		if (!net_eq(n, &init_net))
+			return sprintf(buf, "%llu\n", n->kevent_seqnum);
+	}
+#endif
 	return sprintf(buf, "%llu\n", (unsigned long long)uevent_seqnum);
 }
 KERNEL_ATTR_RO(uevent_seqnum);

lib/kobject_uevent.c +32 -0

index d791e33..7589745 100644
@@ -379,6 +379,96 @@ int kobject_uevent(struct kobject *kobj, enum kobject_action action)
 }
 EXPORT_SYMBOL_GPL(kobject_uevent);
 
+#if defined(CONFIG_UDEVNS) || defined(CONFIG_UDEVNS_MODULE)
+/**
+ * kobject_uevent_forward - forward event to specified network namespace
+ *
+ * @buf: event buffer
+ * @len: event length
+ * @pid: pid of network namespace
+ *
+ * Returns 0 if kobject_uevent_forward() is completed with success or the
+ * corresponding error when it fails.
+ */
+int kobject_uevent_forward(char *buf, size_t len, pid_t pid)
+{
+	int retval = 0;
+#if defined(CONFIG_NET)
+	struct uevent_sock *ue_sk;
+	struct net *pns;
+	char *p;
+	u64 num;
+
+	/* grab the network namespace of the provided pid */
+	pns = get_net_ns_by_pid(pid);
+	if (pns == ERR_PTR(-ESRCH))
+		return -ESRCH;
+
+	/* find sequence number in buffer */
+	p = buf;
+	num = 0;
+	while (p < (buf + len)) {
+		if (strncmp(p, "SEQNUM=", 7) == 0) {
+			int r;
+

lib/kobject_uevent.c int kobject_uevent(struct kobject *kobj, enum kobject_action action) · line 423

Quoted 10 lines

lib/kobject_uevent.c +10 -0

+			p += 7;
+			r = kstrtoull(p, 10, &num);
+			if (r) {
+				put_net(pns);
+				return r;
+			}
+			break;
+		}
+		p += (strlen(p) + 1);
+	}

Ok, that's crazy, replacing an existing sequence number with a sequence number of the namespace? An interesting hack, yes, but something we want to maintain for the next 20+ years, no. Surely there's a better way to do this?

But again, I'm not sold on this whole idea anyway.

Quoted 2 lines

greg k-h

Re: the #if
The #if is only needed because the new udevns code references a structure defined in that header. Obviously it could be included without consequences but I #if it to show it was part of the forwarding code.

Re: sequence #
I wanted it as transparent as possible, without this the udevd running inside the container has issues with the values of the sequence numbers seen, by making it tied to the namespace, udevd could run unchanged.

Our goal was to minimize the changes to a linux distro and still be able to run a full desktop inside a container. But even in absence of our use case, the first question is should uevents be broadcasts to every network namespace? I don't think that broadcasting is the correct action. And follow on questions are what if anything should be done with regards to uevents and containers.

Quoted 133 lines · Michael J Coss

Quoted 132 lines · Greg KH

Quoted 47 lines · Michael J. Coss

Adds capability to allow userspace programs to forward a given event to a specific network namespace as determined by the provided pid. In addition, support for a per-namespace kobject_sequence counter was added. Sysfs was modified to return the correct event counter based on the current network namespace.

Signed-off-by Michael J. Coss <[email protected]>

---
 include/linux/kobject.h     |  3 ++
 include/net/net_namespace.h |  3 ++
 kernel/ksysfs.c             | 12 ++++++
 lib/kobject_uevent.c        | 90 +++++++++++++++++++++++++++++++++++++++++++++
 4 files changed, 108 insertions(+)

include/linux/kobject.h +3 -0

index 637f670..d1bb509 100644
@@ -215,6 +215,9 @@ extern struct kobject *firmware_kobj;
 int kobject_uevent(struct kobject *kobj, enum kobject_action action);
 int kobject_uevent_env(struct kobject *kobj, enum kobject_action action,
                   char *envp[]);
+#if defined(CONFIG_UDEVNS) || defined(CONFIG_UDEVNS_MODULE)
+int kobject_uevent_forward(char *buf, size_t len, pid_t pid);
+#endif
 
 __printf(2, 3)
 int add_uevent_var(struct kobj_uevent_env *env, const char *format, ...);

include/net/net_namespace.h +3 -0

index e951453..a4013e5 100644
@@ -134,6 +134,9 @@ struct net {
 #if IS_ENABLED(CONFIG_MPLS)
   struct netns_mpls       mpls;
 #endif
+#if defined(CONFIG_UDEVNS) || defined(CONFIG_UDEVNS_MODULE)
+  u64                             kevent_seqnum;
+#endif
   struct sock             *diag_nlsk;
   atomic_t                fnhe_genid;
 };

kernel/ksysfs.c +0 -0

index 6683cce..4bc15fd 100644

kernel/ksysfs.c line 26

Quoted 6 lines

kernel/ksysfs.c +3 -0

@@ -21,6 +21,9 @@
 #include <linux/compiler.h>
 
 #include <linux/rcupdate.h>       /* rcu_expedited */
+#if defined(CONFIG_UDEVNS) || defined(CONFIG_UDEVNS_MODULE)
+#include <net/net_namespace.h>
+#endif

#if isn't needed here, right?

Quoted 59 lines

kernel/ksysfs.c +9 -0

 #define KERNEL_ATTR_RO(_name) \
 static struct kobj_attribute _name##_attr = __ATTR_RO(_name)

@@ -33,6 +36,15 @@ static struct kobj_attribute _name##_attr = \
 static ssize_t uevent_seqnum_show(struct kobject *kobj,
                             struct kobj_attribute *attr, char *buf)
 {
+#if defined(CONFIG_UDEVNS) || defined(CONFIG_UDEVNS_MODULE)
+  pid_t p = task_pid_vnr(current);
+  struct net *n = get_net_ns_by_pid(p);
+
+  if (n != ERR_PTR(-ESRCH)) {
+          if (!net_eq(n, &init_net))
+                  return sprintf(buf, "%llu\n", n->kevent_seqnum);
+  }
+#endif
   return sprintf(buf, "%llu\n", (unsigned long long)uevent_seqnum);
 }
 KERNEL_ATTR_RO(uevent_seqnum);

lib/kobject_uevent.c +32 -0

index d791e33..7589745 100644
@@ -379,6 +379,96 @@ int kobject_uevent(struct kobject *kobj, enum kobject_action action)
 }
 EXPORT_SYMBOL_GPL(kobject_uevent);
 
+#if defined(CONFIG_UDEVNS) || defined(CONFIG_UDEVNS_MODULE)
+/**
+ * kobject_uevent_forward - forward event to specified network namespace
+ *
+ * @buf: event buffer
+ * @len: event length
+ * @pid: pid of network namespace
+ *
+ * Returns 0 if kobject_uevent_forward() is completed with success or the
+ * corresponding error when it fails.
+ */
+int kobject_uevent_forward(char *buf, size_t len, pid_t pid)
+{
+  int retval = 0;
+#if defined(CONFIG_NET)
+  struct uevent_sock *ue_sk;
+  struct net *pns;
+  char *p;
+  u64 num;
+
+  /* grab the network namespace of the provided pid */
+  pns = get_net_ns_by_pid(pid);
+  if (pns == ERR_PTR(-ESRCH))
+          return -ESRCH;
+
+  /* find sequence number in buffer */
+  p = buf;
+  num = 0;
+  while (p < (buf + len)) {
+          if (strncmp(p, "SEQNUM=", 7) == 0) {
+                  int r;
+

lib/kobject_uevent.c int kobject_uevent(struct kobject *kobj, enum kobject_action action) · line 423

Quoted 10 lines

lib/kobject_uevent.c +10 -0

+                  p += 7;
+                  r = kstrtoull(p, 10, &num);
+                  if (r) {
+                          put_net(pns);
+                          return r;
+                  }
+                  break;
+          }
+          p += (strlen(p) + 1);
+  }

Ok, that's crazy, replacing an existing sequence number with a sequence number of the namespace? An interesting hack, yes, but something we want to maintain for the next 20+ years, no. Surely there's a better way to do this?

But again, I'm not sold on this whole idea anyway.

Quoted 7 lines

Quoted 2 lines

greg k-h

Re: the #if
The #if is only needed because the new udevns code references a structure defined in that header. Obviously it could be included without consequences but I #if it to show it was part of the forwarding code.

That's not an issue, we don't put #ifdefs in .c code if at all possible, you will note that you added a bunch of them :(

Quoted 4 lines

Re: sequence #
I wanted it as transparent as possible, without this the udevd running inside the container has issues with the values of the sequence numbers seen, by making it tied to the namespace, udevd could run unchanged.

Oh I know why you did it, I just don't like it :)

Quoted 6 lines

Our goal was to minimize the changes to a linux distro and still be able to run a full desktop inside a container. But even in absence of our use case, the first question is should uevents be broadcasts to every network namespace? I don't think that broadcasting is the correct action. And follow on questions are what if anything should be done with regards to uevents and containers.

I don't think you should be running udevd within a container, as a device is never bound to a namespace (network devices are the only exception), it's a false thing to think that a uevent is only for a single namespace as well. I understand your wish to change only the kernel to get unmodified userspace to run, but I suggest modifying your userspace instead :)

sorry,

greg k-h
_______________________________________________
Containers mailing list Containers-cunTk1MwBs9QetFLy7KEm3xJsTq8ys+cHZ5vskTnxNA@public.gmane.org
https://lists.linuxfoundation.org/mailman/listinfo/containers

I fully agree with you that running unmodified distro inside a container is not a justified
cause for kernel changes.

However, I would like to highlight the point that udev is not the only consumer of uevents, so changing userspace, as you rightfully suggested, may not be as simple as you imagine.

For example, in our Android use case, we have no intention of running unmodified Android
inside container and modifying Android's ueventd is not an issue for us. Android userspace uses uevents extensively. For example, vold uses uevents to be notified of SDCARD insert event and that is just one example.
We can get around vold and maybe we can get around every other open source Android tool
that make use of uevents, but phone vendors tend to use uevents to communicate messages between their drivers and proprietary software and this is were we must have
a way to filter those uevents in the kernel.

You argue that "device is never bound to a namespace" (and that it never will be) and I don't disagree with that.

However, as Eric said, the "broadcast everything" logic does not make much sense.

In a way, the broadcast logic is opposite to your suggestion to modify userspace.

In almost every existing implementation of containers, only the root namespace owns responsibility for booting the machine and configuring devices, so if all containers were running modified userspace, there would have been no need to broadcast uevents to all namespaces.

At the very least, you should be able to consent with the idea of having the broadcast policy decided by userspace.

Adds capability to allow userspace programs to forward a given event to a specific network namespace as determined by the provided pid. In addition, support for a per-namespace kobject_sequence counter was added. Sysfs was modified to return the correct event counter based on the current network namespace.

Signed-off-by Michael J. Coss <[email protected]>

---
 include/linux/kobject.h     |  3 ++
 include/net/net_namespace.h |  3 ++
 kernel/ksysfs.c             | 12 ++++++
 lib/kobject_uevent.c        | 90 +++++++++++++++++++++++++++++++++++++++++++++
 4 files changed, 108 insertions(+)

include/linux/kobject.h +3 -0

index 637f670..d1bb509 100644
@@ -215,6 +215,9 @@ extern struct kobject *firmware_kobj;
 int kobject_uevent(struct kobject *kobj, enum kobject_action action);
 int kobject_uevent_env(struct kobject *kobj, enum kobject_action action,
 			char *envp[]);
+#if defined(CONFIG_UDEVNS) || defined(CONFIG_UDEVNS_MODULE)
+int kobject_uevent_forward(char *buf, size_t len, pid_t pid);
+#endif
 
 __printf(2, 3)
 int add_uevent_var(struct kobj_uevent_env *env, const char *format, ...);

include/net/net_namespace.h +3 -0

index e951453..a4013e5 100644
@@ -134,6 +134,9 @@ struct net {
 #if IS_ENABLED(CONFIG_MPLS)
 	struct netns_mpls	mpls;
 #endif
+#if defined(CONFIG_UDEVNS) || defined(CONFIG_UDEVNS_MODULE)
+	u64				kevent_seqnum;
+#endif
 	struct sock		*diag_nlsk;
 	atomic_t		fnhe_genid;
 };

kernel/ksysfs.c +3 -0

index 6683cce..4bc15fd 100644
@@ -21,6 +21,9 @@
 #include <linux/compiler.h>
 
 #include <linux/rcupdate.h>	/* rcu_expedited */
+#if defined(CONFIG_UDEVNS) || defined(CONFIG_UDEVNS_MODULE)
+#include <net/net_namespace.h>
+#endif
 
 #define KERNEL_ATTR_RO(_name) \
 static struct kobj_attribute _name##_attr = __ATTR_RO(_name)

@@ -33,6 +36,15 @@ static struct kobj_attribute _name##_attr = \
 static ssize_t uevent_seqnum_show(struct kobject *kobj,
 				  struct kobj_attribute *attr, char *buf)

kernel/ksysfs.c +0 -0

 	return sprintf(buf, "%llu\n", (unsigned long long)uevent_seqnum);
 }
 KERNEL_ATTR_RO(uevent_seqnum);

lib/kobject_uevent.c +0 -0

index d791e33..7589745 100644

lib/kobject_uevent.c +24 -0

+ * Returns 0 if kobject_uevent_forward() is completed with success or the
+ * corresponding error when it fails.
+ */
+int kobject_uevent_forward(char *buf, size_t len, pid_t pid)
+{
+	int retval = 0;
+#if defined(CONFIG_NET)
+	struct uevent_sock *ue_sk;
+	struct net *pns;
+	char *p;
+	u64 num;
+
+	/* grab the network namespace of the provided pid */
+	pns = get_net_ns_by_pid(pid);
+	if (pns == ERR_PTR(-ESRCH))
+		return -ESRCH;
+
+	/* find sequence number in buffer */
+	p = buf;
+	num = 0;
+	while (p < (buf + len)) {
+		if (strncmp(p, "SEQNUM=", 7) == 0) {
+			int r;
+

lib/kobject_uevent.c +12 -0

+
+	/* if we didn't see a valid seqnum, or none was present, return error */
+	if (num == 0) {
+		put_net(pns);
+		return -EINVAL;
+	}
+	/* update per namespace sequence number as needed */
+	if (pns->kevent_seqnum < num)
+		pns->kevent_seqnum = num;
+
+	list_for_each_entry(ue_sk, &uevent_sock_list, list) {
+		struct sock *uevent_sock = ue_sk->sk;

lib/kobject_uevent.c +25 -0

+		/* allocate message with the maximum possible size */
+		skb = alloc_skb(len, GFP_KERNEL);
+		if (skb) {
+			char *p;
+
+			p = skb_put(skb, len);
+			memcpy(p, buf, len);
+			NETLINK_CB(skb).dst_group = 1;
+			retval = netlink_broadcast(uevent_sock, skb, 0, 1,
+						   GFP_KERNEL);
+
+			/* ENOBUFS should be handled in userspace */
+			if (retval == -ENOBUFS || retval == -ESRCH)
+				retval = 0;
+		} else {
+			retval = -ENOMEM;
+		}
+	}
+	put_net(pns);
+#endif
+	return retval;
+}
+EXPORT_SYMBOL_GPL(kobject_uevent_forward);
+#endif
+
 /**
  * add_uevent_var - add key value string to the environment buffer
  * @env: environment buffer structure

Quoted 5 lines · Michael J. Coss

Adds capability to allow userspace programs to forward a given event to a specific network namespace as determined by the provided pid. In addition, support for a per-namespace kobject_sequence counter was added. Sysfs was modified to return the correct event counter based on the current network namespace.

So this patch is buggy.

Quoted 53 lines

Signed-off-by Michael J. Coss <[email protected]>

---
 include/linux/kobject.h     |  3 ++
 include/net/net_namespace.h |  3 ++
 kernel/ksysfs.c             | 12 ++++++
 lib/kobject_uevent.c        | 90 +++++++++++++++++++++++++++++++++++++++++++++
 4 files changed, 108 insertions(+)

include/linux/kobject.h +3 -0

index 637f670..d1bb509 100644
@@ -215,6 +215,9 @@ extern struct kobject *firmware_kobj;
 int kobject_uevent(struct kobject *kobj, enum kobject_action action);
 int kobject_uevent_env(struct kobject *kobj, enum kobject_action action,
 			char *envp[]);
+#if defined(CONFIG_UDEVNS) || defined(CONFIG_UDEVNS_MODULE)
+int kobject_uevent_forward(char *buf, size_t len, pid_t pid);
+#endif
 
 __printf(2, 3)
 int add_uevent_var(struct kobj_uevent_env *env, const char *format, ...);

include/net/net_namespace.h +3 -0

index e951453..a4013e5 100644
@@ -134,6 +134,9 @@ struct net {
 #if IS_ENABLED(CONFIG_MPLS)
 	struct netns_mpls	mpls;
 #endif
+#if defined(CONFIG_UDEVNS) || defined(CONFIG_UDEVNS_MODULE)
+	u64				kevent_seqnum;
+#endif
 	struct sock		*diag_nlsk;
 	atomic_t		fnhe_genid;
 };

kernel/ksysfs.c +3 -0

index 6683cce..4bc15fd 100644
@@ -21,6 +21,9 @@
 #include <linux/compiler.h>
 
 #include <linux/rcupdate.h>	/* rcu_expedited */
+#if defined(CONFIG_UDEVNS) || defined(CONFIG_UDEVNS_MODULE)
+#include <net/net_namespace.h>
+#endif
 
 #define KERNEL_ATTR_RO(_name) \
 static struct kobj_attribute _name##_attr = __ATTR_RO(_name)

@@ -33,6 +36,15 @@ static struct kobj_attribute _name##_attr = \
 static ssize_t uevent_seqnum_show(struct kobject *kobj,
 				  struct kobj_attribute *attr, char *buf)

kernel/ksysfs.c static struct kobj_attribute _name##_attr = \ · line 47

Quoted 10 lines

kernel/ksysfs.c +9 -0

 {
+#if defined(CONFIG_UDEVNS) || defined(CONFIG_UDEVNS_MODULE)
+	pid_t p = task_pid_vnr(current);
+	struct net *n = get_net_ns_by_pid(p);
+
+	if (n != ERR_PTR(-ESRCH)) {
+		if (!net_eq(n, &init_net))
+			return sprintf(buf, "%llu\n", n->kevent_seqnum);
+	}
+#endif

The test here is completely wrong.
a) Each sysfs instance already has a network namespace attached to it's superblock so using the pid of the caller is wrong.

b) You return kevent_seqnum even in network namespaces where you are not sending uevents from userspace which is concerning.

c) Is this all we need to do to sysfs? Otherwise it may be best to simply fake this file, and remove the need to play games with the sequence number in kobject_uevent_forward.

Quoted 8 lines

kernel/ksysfs.c +0 -0

 	return sprintf(buf, "%llu\n", (unsigned long long)uevent_seqnum);
 }
 KERNEL_ATTR_RO(uevent_seqnum);

lib/kobject_uevent.c +0 -0

index d791e33..7589745 100644

lib/kobject_uevent.c int kobject_uevent(struct kobject *kobj, enum kobject_action action) · line 388

Quoted 10 lines

lib/kobject_uevent.c +7 -0

@@ -379,6 +379,96 @@ int kobject_uevent(struct kobject *kobj, enum kobject_action action)
 }
 EXPORT_SYMBOL_GPL(kobject_uevent);
 
+#if defined(CONFIG_UDEVNS) || defined(CONFIG_UDEVNS_MODULE)
+/**
+ * kobject_uevent_forward - forward event to specified network namespace
+ *
+ * @buf: event buffer
+ * @len: event length
+ * @pid: pid of network namespace

This function should just take a network namespace all of the weird bits should be left to the user/kernel interface.

The pid should be translated into a network namespace at the edge of user space. Not down here in a helper function.

Quoted 29 lines

lib/kobject_uevent.c +29 -0

+ * Returns 0 if kobject_uevent_forward() is completed with success or the
+ * corresponding error when it fails.
+ */
+int kobject_uevent_forward(char *buf, size_t len, pid_t pid)
+{
+	int retval = 0;
+#if defined(CONFIG_NET)
+	struct uevent_sock *ue_sk;
+	struct net *pns;
+	char *p;
+	u64 num;
+
+	/* grab the network namespace of the provided pid */
+	pns = get_net_ns_by_pid(pid);
+	if (pns == ERR_PTR(-ESRCH))
+		return -ESRCH;
+
+	/* find sequence number in buffer */
+	p = buf;
+	num = 0;
+	while (p < (buf + len)) {
+		if (strncmp(p, "SEQNUM=", 7) == 0) {
+			int r;
+
+			p += 7;
+			r = kstrtoull(p, 10, &num);
+			if (r) {
+				put_net(pns);
+				return r;

lib/kobject_uevent.c int kobject_uevent(struct kobject *kobj, enum kobject_action action) · line 423

Quoted 5 lines

lib/kobject_uevent.c +5 -0

+			}
+			break;
+		}
+		p += (strlen(p) + 1);
+	}

Do we really need to parse the sequence number out of the packet? I suspect it would be easier to simply have a sequence number that increments every time a message passes through.

Quoted 12 lines

lib/kobject_uevent.c +12 -0

+
+	/* if we didn't see a valid seqnum, or none was present, return error */
+	if (num == 0) {
+		put_net(pns);
+		return -EINVAL;
+	}
+	/* update per namespace sequence number as needed */
+	if (pns->kevent_seqnum < num)
+		pns->kevent_seqnum = num;
+
+	list_for_each_entry(ue_sk, &uevent_sock_list, list) {
+		struct sock *uevent_sock = ue_sk->sk;

lib/kobject_uevent.c int kobject_uevent(struct kobject *kobj, enum kobject_action action) · line 445

Quoted 10 lines

lib/kobject_uevent.c +10 -0

+		struct sk_buff *skb;
+
+		if (!netlink_has_listeners(uevent_sock, 1))
+			continue;
+		/*
+		 * only send to sockets share the same network namespace
+		 * as the passed pid
+		 */
+		if (!net_eq(sock_net(uevent_sock), pns))
+			continue;

This look is terribly inefficient. You could just go directly to the network namespace uevent_sock that you want from your network namespace.

I wonder if we could arrange things so that the same skb you pass in is the skb that gets broadcast out. That would simplify a lot of things.

Quoted 25 lines

lib/kobject_uevent.c +25 -0

+		/* allocate message with the maximum possible size */
+		skb = alloc_skb(len, GFP_KERNEL);
+		if (skb) {
+			char *p;
+
+			p = skb_put(skb, len);
+			memcpy(p, buf, len);
+			NETLINK_CB(skb).dst_group = 1;
+			retval = netlink_broadcast(uevent_sock, skb, 0, 1,
+						   GFP_KERNEL);
+
+			/* ENOBUFS should be handled in userspace */
+			if (retval == -ENOBUFS || retval == -ESRCH)
+				retval = 0;
+		} else {
+			retval = -ENOMEM;
+		}
+	}
+	put_net(pns);
+#endif
+	return retval;
+}
+EXPORT_SYMBOL_GPL(kobject_uevent_forward);
+#endif
+

lib/kobject_uevent.c int kobject_uevent(struct kobject *kobj, enum kobject_action action) · line 474

Quoted 3 lines

lib/kobject_uevent.c +0 -0

 /**
  * add_uevent_var - add key value string to the environment buffer
  * @env: environment buffer structure

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

Adds capability to allow userspace programs to forward a given event to a specific network namespace as determined by the provided pid. In addition, support for a per-namespace kobject_sequence counter was added. Sysfs was modified to return the correct event counter based on the current network namespace.

Signed-off-by Michael J. Coss <[email protected]>

---
 include/linux/kobject.h     |  3 ++
 include/net/net_namespace.h |  3 ++
 kernel/ksysfs.c             | 12 ++++++
 lib/kobject_uevent.c        | 90
+++++++++++++++++++++++++++++++++++++++++++++
 4 files changed, 108 insertions(+)

include/linux/kobject.h +3 -0

index 637f670..d1bb509 100644
@@ -215,6 +215,9 @@ extern struct kobject *firmware_kobj;
 int kobject_uevent(struct kobject *kobj, enum kobject_action action);
 int kobject_uevent_env(struct kobject *kobj, enum kobject_action action,
                        char *envp[]);
+#if defined(CONFIG_UDEVNS) || defined(CONFIG_UDEVNS_MODULE)
+int kobject_uevent_forward(char *buf, size_t len, pid_t pid);
+#endif
 
 __printf(2, 3)
 int add_uevent_var(struct kobj_uevent_env *env, const char *format, ...);

include/net/net_namespace.h +3 -0

index e951453..a4013e5 100644
@@ -134,6 +134,9 @@ struct net {
 #if IS_ENABLED(CONFIG_MPLS)
        struct netns_mpls       mpls;
 #endif
+#if defined(CONFIG_UDEVNS) || defined(CONFIG_UDEVNS_MODULE)
+       u64                             kevent_seqnum;
+#endif
        struct sock             *diag_nlsk;
        atomic_t                fnhe_genid;
 };

kernel/ksysfs.c +12 -0

index 6683cce..4bc15fd 100644
@@ -21,6 +21,9 @@
 #include <linux/compiler.h>
 
 #include <linux/rcupdate.h>    /* rcu_expedited */
+#if defined(CONFIG_UDEVNS) || defined(CONFIG_UDEVNS_MODULE)
+#include <net/net_namespace.h>
+#endif
 
 #define KERNEL_ATTR_RO(_name) \
 static struct kobj_attribute _name##_attr = __ATTR_RO(_name)

@@ -33,6 +36,15 @@ static struct kobj_attribute _name##_attr = \
 static ssize_t uevent_seqnum_show(struct kobject *kobj,
                                  struct kobj_attribute *attr, char *buf)
 {
+#if defined(CONFIG_UDEVNS) || defined(CONFIG_UDEVNS_MODULE)
+       pid_t p = task_pid_vnr(current);
+       struct net *n = get_net_ns_by_pid(p);
+
+       if (n != ERR_PTR(-ESRCH)) {
+               if (!net_eq(n, &init_net))
+                       return sprintf(buf, "%llu\n", n->kevent_seqnum);
+       }
+#endif
        return sprintf(buf, "%llu\n", (unsigned long long)uevent_seqnum);
 }
 KERNEL_ATTR_RO(uevent_seqnum);

lib/kobject_uevent.c +44 -0

index d791e33..7589745 100644
@@ -379,6 +379,96 @@ int kobject_uevent(struct kobject *kobj, enum
 kobject_action action)
 }
 EXPORT_SYMBOL_GPL(kobject_uevent);
 
+#if defined(CONFIG_UDEVNS) || defined(CONFIG_UDEVNS_MODULE)
+/**
+ * kobject_uevent_forward - forward event to specified network namespace
+ *
+ * @buf: event buffer
+ * @len: event length
+ * @pid: pid of network namespace
+ *
+ * Returns 0 if kobject_uevent_forward() is completed with success or the
+ * corresponding error when it fails.
+ */
+int kobject_uevent_forward(char *buf, size_t len, pid_t pid)
+{
+       int retval = 0;
+#if defined(CONFIG_NET)
+       struct uevent_sock *ue_sk;
+       struct net *pns;
+       char *p;
+       u64 num;
+
+       /* grab the network namespace of the provided pid */
+       pns = get_net_ns_by_pid(pid);
+       if (pns == ERR_PTR(-ESRCH))
+               return -ESRCH;
+
+       /* find sequence number in buffer */
+       p = buf;
+       num = 0;
+       while (p < (buf + len)) {
+               if (strncmp(p, "SEQNUM=", 7) == 0) {
+                       int r;
+
+                       p += 7;
+                       r = kstrtoull(p, 10, &num);
+                       if (r) {
+                               put_net(pns);
+                               return r;
+                       }
+                       break;
+               }
+               p += (strlen(p) + 1);
+       }
+
+       /* if we didn't see a valid seqnum, or none was present, return

error */

lib/kobject_uevent.c +25 -0

+       if (num == 0) {
+               put_net(pns);
+               return -EINVAL;
+       }
+       /* update per namespace sequence number as needed */
+       if (pns->kevent_seqnum < num)
+               pns->kevent_seqnum = num;
+
+       list_for_each_entry(ue_sk, &uevent_sock_list, list) {
+               struct sock *uevent_sock = ue_sk->sk;
+               struct sk_buff *skb;
+
+               if (!netlink_has_listeners(uevent_sock, 1))
+                       continue;
+               /*
+                * only send to sockets share the same network namespace
+                * as the passed pid
+                */
+               if (!net_eq(sock_net(uevent_sock), pns))
+                       continue;
+
+               /* allocate message with the maximum possible size */
+               skb = alloc_skb(len, GFP_KERNEL);
+               if (skb) {
+                       char *p;

lib/kobject_uevent.c +3 -1

+EXPORT_SYMBOL_GPL(kobject_uevent_forward);
+#endif
+
 /**
  * add_uevent_var - add key value string to the environment buffer
  * @env: environment buffer structure
--

2.4.6

_______________________________________________
Containers mailing list Containers-cunTk1MwBs9QetFLy7KEm3xJsTq8ys+cHZ5vskTnxNA@public.gmane.org
https://lists.linuxfoundation.org/mailman/listinfo/containers

Quoted 159 lines · Michael J. Coss

Adds capability to allow userspace programs to forward a given event to a specific network namespace as determined by the provided pid. In addition, support for a per-namespace kobject_sequence counter was added. Sysfs was modified to return the correct event counter based on the current network namespace.

Signed-off-by Michael J. Coss <[email protected]>

---
 include/linux/kobject.h     |  3 ++
 include/net/net_namespace.h |  3 ++
 kernel/ksysfs.c             | 12 ++++++
 lib/kobject_uevent.c        | 90
+++++++++++++++++++++++++++++++++++++++++++++
 4 files changed, 108 insertions(+)

include/linux/kobject.h +3 -0

index 637f670..d1bb509 100644
@@ -215,6 +215,9 @@ extern struct kobject *firmware_kobj;
 int kobject_uevent(struct kobject *kobj, enum kobject_action action);
 int kobject_uevent_env(struct kobject *kobj, enum kobject_action action,
                        char *envp[]);
+#if defined(CONFIG_UDEVNS) || defined(CONFIG_UDEVNS_MODULE)
+int kobject_uevent_forward(char *buf, size_t len, pid_t pid);
+#endif
 
 __printf(2, 3)
 int add_uevent_var(struct kobj_uevent_env *env, const char *format, ...);

include/net/net_namespace.h +3 -0

index e951453..a4013e5 100644
@@ -134,6 +134,9 @@ struct net {
 #if IS_ENABLED(CONFIG_MPLS)
        struct netns_mpls       mpls;
 #endif
+#if defined(CONFIG_UDEVNS) || defined(CONFIG_UDEVNS_MODULE)
+       u64                             kevent_seqnum;
+#endif
        struct sock             *diag_nlsk;
        atomic_t                fnhe_genid;
 };

kernel/ksysfs.c +12 -0

index 6683cce..4bc15fd 100644
@@ -21,6 +21,9 @@
 #include <linux/compiler.h>
 
 #include <linux/rcupdate.h>    /* rcu_expedited */
+#if defined(CONFIG_UDEVNS) || defined(CONFIG_UDEVNS_MODULE)
+#include <net/net_namespace.h>
+#endif
 
 #define KERNEL_ATTR_RO(_name) \
 static struct kobj_attribute _name##_attr = __ATTR_RO(_name)

@@ -33,6 +36,15 @@ static struct kobj_attribute _name##_attr = \
 static ssize_t uevent_seqnum_show(struct kobject *kobj,
                                  struct kobj_attribute *attr, char *buf)
 {
+#if defined(CONFIG_UDEVNS) || defined(CONFIG_UDEVNS_MODULE)
+       pid_t p = task_pid_vnr(current);
+       struct net *n = get_net_ns_by_pid(p);
+
+       if (n != ERR_PTR(-ESRCH)) {
+               if (!net_eq(n, &init_net))
+                       return sprintf(buf, "%llu\n", n->kevent_seqnum);
+       }
+#endif
        return sprintf(buf, "%llu\n", (unsigned long long)uevent_seqnum);
 }
 KERNEL_ATTR_RO(uevent_seqnum);

lib/kobject_uevent.c +44 -0

index d791e33..7589745 100644
@@ -379,6 +379,96 @@ int kobject_uevent(struct kobject *kobj, enum
 kobject_action action)
 }
 EXPORT_SYMBOL_GPL(kobject_uevent);
 
+#if defined(CONFIG_UDEVNS) || defined(CONFIG_UDEVNS_MODULE)
+/**
+ * kobject_uevent_forward - forward event to specified network namespace
+ *
+ * @buf: event buffer
+ * @len: event length
+ * @pid: pid of network namespace
+ *
+ * Returns 0 if kobject_uevent_forward() is completed with success or the
+ * corresponding error when it fails.
+ */
+int kobject_uevent_forward(char *buf, size_t len, pid_t pid)
+{
+       int retval = 0;
+#if defined(CONFIG_NET)
+       struct uevent_sock *ue_sk;
+       struct net *pns;
+       char *p;
+       u64 num;
+
+       /* grab the network namespace of the provided pid */
+       pns = get_net_ns_by_pid(pid);
+       if (pns == ERR_PTR(-ESRCH))
+               return -ESRCH;
+
+       /* find sequence number in buffer */
+       p = buf;
+       num = 0;
+       while (p < (buf + len)) {
+               if (strncmp(p, "SEQNUM=", 7) == 0) {
+                       int r;
+
+                       p += 7;
+                       r = kstrtoull(p, 10, &num);
+                       if (r) {
+                               put_net(pns);
+                               return r;
+                       }
+                       break;
+               }
+               p += (strlen(p) + 1);
+       }
+
+       /* if we didn't see a valid seqnum, or none was present, return

error */

lib/kobject_uevent.c +32 -0

+       if (num == 0) {
+               put_net(pns);
+               return -EINVAL;
+       }
+       /* update per namespace sequence number as needed */
+       if (pns->kevent_seqnum < num)
+               pns->kevent_seqnum = num;
+
+       list_for_each_entry(ue_sk, &uevent_sock_list, list) {
+               struct sock *uevent_sock = ue_sk->sk;
+               struct sk_buff *skb;
+
+               if (!netlink_has_listeners(uevent_sock, 1))
+                       continue;
+               /*
+                * only send to sockets share the same network namespace
+                * as the passed pid
+                */
+               if (!net_eq(sock_net(uevent_sock), pns))
+                       continue;
+
+               /* allocate message with the maximum possible size */
+               skb = alloc_skb(len, GFP_KERNEL);
+               if (skb) {
+                       char *p;
+
+                       p = skb_put(skb, len);
+                       memcpy(p, buf, len);
+                       NETLINK_CB(skb).dst_group = 1;
+                       retval = netlink_broadcast(uevent_sock, skb, 0, 1,
+                                                  GFP_KERNEL);
+

lib/kobject_uevent.c int kobject_uevent(struct kobject *kobj, enum kobject_action action) · line 460

Quoted 4 lines

lib/kobject_uevent.c +3 -0

+                       /* ENOBUFS should be handled in userspace */
+                       if (retval == -ENOBUFS || retval == -ESRCH)
+                               retval = 0;

This may mask an error from an earlier send (to a distinct listener socket). Instead, we should retain and report either the first error seen or some error seen.

New generic netlink module to provide an interface with the new forwarding interface for uevent. The driver allows a user to direct a uevent as read from the kernel to a specific network namespace by providing the uevent message, and a target process id. The uapi header file provides the message format.

Signed-off-by Michael J. Coss <[email protected]>

---
 include/uapi/linux/Kbuild   |   1 +
 include/uapi/linux/udevns.h |  19 ++++++++
 net/Kconfig                 |   1 +
 net/Makefile                |   1 +
 net/udevns/Kconfig          |   9 ++++
 net/udevns/Makefile         |   5 ++
 net/udevns/udevns.c         | 112 ++++++++++++++++++++++++++++++++++++++++++++
 net/udevns/udevns.h         |  19 ++++++++
 8 files changed, 167 insertions(+)
 create mode 100644 include/uapi/linux/udevns.h
 create mode 100644 net/udevns/Kconfig
 create mode 100644 net/udevns/Makefile
 create mode 100644 net/udevns/udevns.c
 create mode 100644 net/udevns/udevns.h

include/uapi/linux/Kbuild +1 -0

index 1ff9942..9fb9c59 100644
@@ -404,6 +404,7 @@ header-y += toshiba.h
 header-y += tty_flags.h
 header-y += tty.h
 header-y += types.h
+header-y += udevns.h
 header-y += udf_fs_i.h
 header-y += udp.h
 header-y += uhid.h

include/uapi/linux/udevns.h new file +19 -0

new file mode 100644
index 0000000..f5702f5
@@ -0,0 +1,19 @@
+#ifndef _UDEVNS_H_
+#define _UDEVNS_H_
+
+enum udevns_msg_types {
+	UDEVNS_FORWARD_MSG = 0x1,
+	UDEVNS_CMD_MAX,
+};
+
+enum udevns_attr {
+	UDEVNS_UNSPEC,
+	UDEVNS_PID,
+	UDEVNS_MSG,
+	__UDEVNS_ATTR_MAX,
+};
+
+#define UDEVNS_ATTR_MAX (__UDEVNS_ATTR_MAX - 1)
+#define UDEVNS_VERSION 0x1
+
+#endif

net/Kconfig +1 -0

index 57a7c5a..465e288 100644
@@ -54,6 +54,7 @@ source "net/packet/Kconfig"
 source "net/unix/Kconfig"
 source "net/xfrm/Kconfig"
 source "net/iucv/Kconfig"
+source "net/udevns/Kconfig"
 
 config INET
 	bool "TCP/IP networking"

net/Makefile +1 -0

index 3995613..bde7775 100644
@@ -74,3 +74,4 @@ obj-$(CONFIG_HSR)		+= hsr/
 ifneq ($(CONFIG_NET_SWITCHDEV),)
 obj-y				+= switchdev/
 endif
+obj-$(CONFIG_UDEVNS)	+= udevns/

net/udevns/Kconfig new file +9 -0

new file mode 100644
index 0000000..367e650
@@ -0,0 +1,9 @@
+config UDEVNS
+	tristate "UDEV namespace bridge"
+	depends on SYSFS
+	default n
+	help
+		This option enables support for explicit forwarding of UDEV events to
+		other network namespaces
+
+		If unsure, say N.

net/udevns/Makefile new file +5 -0

new file mode 100644
index 0000000..44c6b12
@@ -0,0 +1,5 @@
+#
+# Makefile for the uevent namespace aware forwarder
+#
+#
+obj-$(CONFIG_UDEVNS) += udevns.o

net/udevns/udevns.c new file +112 -0

new file mode 100644
index 0000000..8b23751
@@ -0,0 +1,112 @@
+#include <linux/module.h>
+#include <linux/moduleparam.h>
+#include <net/sock.h>
+#include <net/genetlink.h>
+#include <linux/netlink.h>
+#include <linux/skbuff.h>
+#include <linux/fs.h>
+#include <linux/list.h>
+#include <linux/slab.h>
+#include <linux/uaccess.h>
+#include <linux/kobject.h>
+#include <linux/sysfs.h>
+
+#include "udevns.h"
+
+#define DRIVER_AUTHOR "Michael J Coss <[email protected]>"
+#define DRIVER_DESC   "new udev namespace bridge"
+#define DEVICE_NAME   "udevns"
+
+#ifdef MODULE
+#define UDEVNS_NAME (THIS_MODULE->name)
+#else
+#define UDEVNS_NAME "udevns"
+#endif
+
+#define UDEVNS_INFO(fmt, args...)                                       \
+	pr_info("[%s] " fmt, UDEVNS_NAME, ## args)
+
+#define UDEVNS_ERROR(fmt, args...)                                      \
+	pr_err("[ERROR:%s:%s] " fmt, UDEVNS_NAME, __func__, ## args)
+
+#define UDEVNS_WARNING(fmt, args...)                                    \
+	pr_warn("[WARNING:%s:%s] " fmt, UDEVNS_NAME, __func__, ## args)
+
+static struct genl_family udevns_genl_family = {
+	.id      = GENL_ID_GENERATE,
+	.name    = DEVICE_NAME,
+	.hdrsize = 0,
+	.version = UDEVNS_VERSION,
+	.maxattr = UDEVNS_ATTR_MAX,
+};
+
+static const struct nla_policy udevns_fmsgpolicy[UDEVNS_ATTR_MAX + 1] = {
+	[UDEVNS_PID] = { .type = NLA_U32 },
+	[UDEVNS_MSG] = { .type = NLA_STRING, .len =  UEVENT_BUFFER_SIZE},
+};
+
+static int udevns_forwardmsg(struct sk_buff *skb, struct genl_info *info)
+{
+	pid_t pid;
+	char *msg;
+	int msglen;
+	int err;
+
+	if (!info->attrs[UDEVNS_PID]) {
+		UDEVNS_WARNING("missing PID from UDEVNS_FORWARD_MSG.\n");
+		return -EINVAL;
+	}
+
+	if (!info->attrs[UDEVNS_MSG]) {
+		UDEVNS_WARNING("missing uevent from UDEVNS_FORWARD_MSG.\n");
+		return -EINVAL;
+	}
+
+	pid = nla_get_u32(info->attrs[UDEVNS_PID]);
+	msg = nla_data(info->attrs[UDEVNS_MSG]);
+	msglen = nla_len(info->attrs[UDEVNS_MSG]);
+
+	if (msglen < 0) {
+		UDEVNS_ERROR("Malformed uevent from UDEVNS_FORWARD_MSG.\n");
+		return -EINVAL;
+	}
+
+	err = kobject_uevent_forward(msg, msglen, pid);
+	return err;
+}
+
+static struct genl_ops udevns_genl_ops[] = {
+	{
+		.cmd    = UDEVNS_FORWARD_MSG,
+		.flags  = GENL_ADMIN_PERM,
+		.doit   = udevns_forwardmsg,
+		.policy = udevns_fmsgpolicy,
+	},
+};
+
+static int __init udevns_init(void)
+{
+	int rc;
+
+	UDEVNS_INFO("Starting udevns module\n");
+	rc = genl_register_family_with_ops(&udevns_genl_family,
+					   udevns_genl_ops);
+	if (rc) {
+		UDEVNS_ERROR("Failed to register netlink interface\n");
+		return rc;
+	}
+	return 0;
+}
+
+static void __exit udevns_exit(void)
+{
+	UDEVNS_INFO("Exiting udevns module\n");
+	genl_unregister_family(&udevns_genl_family);
+}
+
+module_init(udevns_init);
+module_exit(udevns_exit);
+
+MODULE_LICENSE("GPL");
+MODULE_DESCRIPTION(DRIVER_DESC);
+MODULE_AUTHOR(DRIVER_AUTHOR);

net/udevns/udevns.h new file +19 -0

new file mode 100644
index 0000000..f5702f5
@@ -0,0 +1,19 @@
+#ifndef _UDEVNS_H_
+#define _UDEVNS_H_
+
+enum udevns_msg_types {
+	UDEVNS_FORWARD_MSG = 0x1,
+	UDEVNS_CMD_MAX,
+};
+
+enum udevns_attr {
+	UDEVNS_UNSPEC,
+	UDEVNS_PID,
+	UDEVNS_MSG,
+	__UDEVNS_ATTR_MAX,
+};
+
+#define UDEVNS_ATTR_MAX (__UDEVNS_ATTR_MAX - 1)
+#define UDEVNS_VERSION 0x1
+
+#endif

Quoted 5 lines · Michael J. Coss

New generic netlink module to provide an interface with the new forwarding interface for uevent. The driver allows a user to direct a uevent as read from the kernel to a specific network namespace by providing the uevent message, and a target process id. The uapi header file provides the message format.

If we can't just pass the message thourgh I don't expect genetlink is a particularly good interface for this.

It would be nice if we could open some appropriate thing and the act of getting a file descriptor ould suppress all of the uevent broadcast messages in that network namespace.

Further GENL_ADMIN_PERM is an unfortunate choice for a permission check. I don't see it as exploitable but I am not certain CAP_SYS_ADMIN is the
best capability to check. Beyond that we probably want to arrange
things so that we can use ns_capable so we can allow containers to hand off their devices to child containers.

Implementations that do not allow for containers to nest bother me.

Currently when a uevent occurs, the event is replicated and sent to every listener on the kernel netlink socket, ignoring network namespaces boundaries, forwarding events to every listener in every network namespace.

With the expanded use of containers, it would be useful to be able to regulate this flow of events to specific containers. By restricting the events to only the host network namespace, it allows for a userspace program to provide a system wide policy on which events are routed where.

Quoted 8 lines · Michael J. Coss

Currently when a uevent occurs, the event is replicated and sent to every listener on the kernel netlink socket, ignoring network namespaces boundaries, forwarding events to every listener in every network namespace.

With the expanded use of containers, it would be useful to be able to regulate this flow of events to specific containers. By restricting the events to only the host network namespace, it allows for a userspace program to provide a system wide policy on which events are routed where.

Interesting, but why do you need a container to get a uevent at all? What uevents do a container care about?

thanks,

greg k-h

Quoted 12 lines · Greg KH

Quoted 8 lines · Michael J. Coss

Currently when a uevent occurs, the event is replicated and sent to every listener on the kernel netlink socket, ignoring network namespaces boundaries, forwarding events to every listener in every network namespace.

With the expanded use of containers, it would be useful to be able to regulate this flow of events to specific containers. By restricting the events to only the host network namespace, it allows for a userspace program to provide a system wide policy on which events are routed where.

Interesting, but why do you need a container to get a uevent at all? What uevents do a container care about?

Quoted 4 lines

thanks,

greg k-h

In our use case, we run a full desktop inside the container, including X.

We run the Xserver in headless mode, and forward a uevent to the container to allow binding/unbinding of remote keyboard, mice, and
displays. So I want the add/del keyboard events, add/del mouse events,
and add/del display events. This is just one use case, I could image others. The bottom line is that the current behavior is to broadcast to everyone all uevents, and I don't see that as correct as it crosses the network namespace boundaries. It seems to me that you would want to provide controls as to where you want to forward those uevents, and that is not a policy that I believe should be in the kernel but rather in user space.

Quoted 13 lines · Michael J Coss

Quoted 12 lines · Greg KH

Quoted 8 lines · Michael J. Coss

Currently when a uevent occurs, the event is replicated and sent to every listener on the kernel netlink socket, ignoring network namespaces boundaries, forwarding events to every listener in every network namespace.

With the expanded use of containers, it would be useful to be able to regulate this flow of events to specific containers. By restricting the events to only the host network namespace, it allows for a userspace program to provide a system wide policy on which events are routed where.

Interesting, but why do you need a container to get a uevent at all? What uevents do a container care about?

Quoted 6 lines

Quoted 4 lines

thanks,

greg k-h

In our use case, we run a full desktop inside the container, including X.

Ugh, I was worried you were going to say that :(

Quoted 10 lines

We run the Xserver in headless mode, and forward a uevent to the container to allow binding/unbinding of remote keyboard, mice, and
displays. So I want the add/del keyboard events, add/del mouse events,
and add/del display events. This is just one use case, I could image others. The bottom line is that the current behavior is to broadcast to everyone all uevents, and I don't see that as correct as it crosses the network namespace boundaries. It seems to me that you would want to provide controls as to where you want to forward those uevents, and that is not a policy that I believe should be in the kernel but rather in user space.

devices are not in namespaces, which is why we don't partition them off at all. And that's why I really don't want to add this type of filtering either. It's up to the "master" container/process/whatever to send uevents to child containers if it really wants to. If we were to ever have devices bound only to namespaces, then it would make sense to only send the uevents for those devices to that namespace.

But as that's never going to happen, I don't want to give people a false sense of "separation" here that isn't really there at all.

sorry,

greg k-h

Quoted 42 lines · Greg KH

Quoted 13 lines · Michael J Coss

Quoted 12 lines · Greg KH

Quoted 8 lines · Michael J. Coss

Currently when a uevent occurs, the event is replicated and sent to every listener on the kernel netlink socket, ignoring network namespaces boundaries, forwarding events to every listener in every network namespace.

With the expanded use of containers, it would be useful to be able to regulate this flow of events to specific containers. By restricting the events to only the host network namespace, it allows for a userspace program to provide a system wide policy on which events are routed where.

Interesting, but why do you need a container to get a uevent at all? What uevents do a container care about?

Quoted 6 lines

Quoted 4 lines

thanks,

greg k-h

In our use case, we run a full desktop inside the container, including X.

Ugh, I was worried you were going to say that :(

Quoted 10 lines

We run the Xserver in headless mode, and forward a uevent to the container to allow binding/unbinding of remote keyboard, mice, and
displays. So I want the add/del keyboard events, add/del mouse events,
and add/del display events. This is just one use case, I could image others. The bottom line is that the current behavior is to broadcast to everyone all uevents, and I don't see that as correct as it crosses the network namespace boundaries. It seems to me that you would want to provide controls as to where you want to forward those uevents, and that is not a policy that I believe should be in the kernel but rather in user space.

devices are not in namespaces, which is why we don't partition them off at all. And that's why I really don't want to add this type of filtering either. It's up to the "master" container/process/whatever to send uevents to child containers if it really wants to. If we were to ever have devices bound only to namespaces, then it would make sense to only send the uevents for those devices to that namespace.

But as that's never going to happen, I don't want to give people a false sense of "separation" here that isn't really there at all.

Quoted 4 lines

sorry,

greg k-h

Agreed that devices are not in namespaces, but the events are, or at least could be.

That master is the host, and to do that I want to
forward events that the host receives to those individual containers. But since the kernel is broadcasting them, I can't have that policy on the host, and would have to filter on each container. Or I can do as you say and have the master forward events. I don't see this as putting the devices into a namespace, but rather managing devices from the outside and notifying the container of the event. Just like plugging in a monitor to the container.

Quoted 43 lines · Michael J Coss

Quoted 42 lines · Greg KH

Quoted 13 lines · Michael J Coss

Quoted 12 lines · Greg KH

Quoted 8 lines · Michael J. Coss

Currently when a uevent occurs, the event is replicated and sent to every listener on the kernel netlink socket, ignoring network namespaces boundaries, forwarding events to every listener in every network namespace.

With the expanded use of containers, it would be useful to be able to regulate this flow of events to specific containers. By restricting the events to only the host network namespace, it allows for a userspace program to provide a system wide policy on which events are routed where.

Interesting, but why do you need a container to get a uevent at all? What uevents do a container care about?

Quoted 6 lines

Quoted 4 lines

thanks,

greg k-h

In our use case, we run a full desktop inside the container, including X.

Ugh, I was worried you were going to say that :(

Quoted 10 lines

We run the Xserver in headless mode, and forward a uevent to the container to allow binding/unbinding of remote keyboard, mice, and
displays. So I want the add/del keyboard events, add/del mouse events,
and add/del display events. This is just one use case, I could image others. The bottom line is that the current behavior is to broadcast to everyone all uevents, and I don't see that as correct as it crosses the network namespace boundaries. It seems to me that you would want to provide controls as to where you want to forward those uevents, and that is not a policy that I believe should be in the kernel but rather in user space.

devices are not in namespaces, which is why we don't partition them off at all. And that's why I really don't want to add this type of filtering either. It's up to the "master" container/process/whatever to send uevents to child containers if it really wants to. If we were to ever have devices bound only to namespaces, then it would make sense to only send the uevents for those devices to that namespace.

But as that's never going to happen, I don't want to give people a false sense of "separation" here that isn't really there at all.

Quoted 6 lines

Quoted 4 lines

sorry,

greg k-h

Agreed that devices are not in namespaces, but the events are, or at least could be.

No, there's no way to tell which event for which device goes to which namespace, as devices are not in a namespace.

Quoted 8 lines

That master is the host, and to do that I want to
forward events that the host receives to those individual containers. But since the kernel is broadcasting them, I can't have that policy on the host, and would have to filter on each container. Or I can do as you say and have the master forward events. I don't see this as putting the devices into a namespace, but rather managing devices from the outside and notifying the container of the event. Just like plugging in a monitor to the container.

But you can't "plug a monitor into a container". Nor can you "add a keyboard to a container". Or a tty device. Or anything else (except for network devices). Don't try to fake things out as that's not what is happening here. The kernel shouldn't be allowing things to be sent only to specific namespaces, as that's a lie, the devices are "global" and not in a namespace at all.

sorry,

greg k-h
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

Quoted 46 lines · Greg KH

Quoted 45 lines · Michael J Coss

Quoted 44 lines · Greg KH

Quoted 13 lines · Michael J Coss

Quoted 12 lines · Greg KH

Quoted 8 lines · Michael J. Coss

Currently when a uevent occurs, the event is replicated and sent to every listener on the kernel netlink socket, ignoring network namespaces boundaries, forwarding events to every listener in every network namespace.

With the expanded use of containers, it would be useful to be able to regulate this flow of events to specific containers. By restricting the events to only the host network namespace, it allows for a userspace program to provide a system wide policy on which events are routed where.

Interesting, but why do you need a container to get a uevent at all? What uevents do a container care about?

Quoted 6 lines

Quoted 4 lines

thanks,

greg k-h

In our use case, we run a full desktop inside the container, including X.

Ugh, I was worried you were going to say that :(

Quoted 10 lines

We run the Xserver in headless mode, and forward a uevent to the container to allow binding/unbinding of remote keyboard, mice, and
displays. So I want the add/del keyboard events, add/del mouse events,
and add/del display events. This is just one use case, I could image others. The bottom line is that the current behavior is to broadcast to everyone all uevents, and I don't see that as correct as it crosses the network namespace boundaries. It seems to me that you would want to provide controls as to where you want to forward those uevents, and that is not a policy that I believe should be in the kernel but rather in user space.

devices are not in namespaces, which is why we don't partition them off at all. And that's why I really don't want to add this type of filtering either. It's up to the "master" container/process/whatever to send uevents to child containers if it really wants to. If we were to ever have devices bound only to namespaces, then it would make sense to only send the uevents for those devices to that namespace.

But as that's never going to happen, I don't want to give people a false sense of "separation" here that isn't really there at all.

sorry,

Quoted 6 lines

Quoted 4 lines

Quoted 2 lines

greg k-h

Agreed that devices are not in namespaces, but the events are, or at least could be.

No, there's no way to tell which event for which device goes to which namespace, as devices are not in a namespace.

Why? The host certainly can have a policy for what devices go to which
container.

And as such knows which events goes to which container.

The container *is* a set on namespace, and control groups.

So a user program reads the events on the master, looks in a database and forwards it to that container. The uevents represent the device add/del so it seems natural that it should be the mechanism by which that communication happens. I just want to see it controlled by a policy on the host.

Quoted 8 lines

Quoted 8 lines

That master is the host, and to do that I want to
forward events that the host receives to those individual containers. But since the kernel is broadcasting them, I can't have that policy on the host, and would have to filter on each container. Or I can do as you say and have the master forward events. I don't see this as putting the devices into a namespace, but rather managing devices from the outside and notifying the container of the event. Just like plugging in a monitor to the container.

Quoted 6 lines

But you can't "plug a monitor into a container". Nor can you "add a keyboard to a container". Or a tty device. Or anything else (except for network devices). Don't try to fake things out as that's not what is happening here. The kernel shouldn't be allowing things to be sent only to specific namespaces, as that's a lie, the devices are "global" and not in a namespace at all.

Again why? Why are network devices *different*?

They are a resources that is bound to the container, not to a namespace per se, but the container is a construct. A collection of namespaces, and cgroups. Again, I don't see why you can't add a keyboard to the container.