Make access to task's nsproxy liter

When someone wants to deal with some other taks's namespaces it has to lock the task and then to get the desired namespace if the one exists. This is slow on read-only paths and may be impossible in some cases.

E.g. Oleg recently noticed a race between unshare() and the (sent for review in containers) pid namespaces - when the task notifies the parent it has to know the parent's namespace, but taking the task_lock() is impossible there - the code is under write locked tasklist lock.

On the other hand switching the namespace on task (daemonize) and releasing the namespace (after the last task exit) is rather
rare operation and we can sacrifice its speed to solve the issues above.

The access to other task namespaces is proposed to be performed like this:

rcu_read_lock();
nsproxy = task_nsproxy(tsk);
if (nsproxy != NULL) {
/ *
* work with the namespaces here
* e.g. get the reference on one of them
* /
} / *
* NULL task_nsproxy() means that this task is
* almost dead (zombie)
* /
rcu_read_unlock();

This patch has passed the review by Eric and Oleg :) and, of course, tested.

Signed-off-by Pavel Emelyanov <[email protected]>

Cc Eric W. Biederman <[email protected]>

Cc Oleg Nesterov <[email protected]>

Cc Paul E. McKenney <[email protected]>

Cc Serge Hallyn <[email protected]>

---

fs/proc/base.c          |   27 +++++++++++++++++----------
include/linux/nsproxy.h |   44 ++++++++++++++++++++++++++++++++------------
kernel/exit.c           |    4 +---
kernel/fork.c           |   11 +++++------
kernel/nsproxy.c        |   39 +++++++++++++++++++++++++++++++--------
5 files changed, 86 insertions(+), 39 deletions(-)

fs/proc/base.c +17 -10

index ed2b224..614851a 100644
@@ -371,18 +371,21 @@ struct proc_mounts {
 static int mounts_open(struct inode *inode, struct file *file)
 {
 	struct task_struct *task = get_proc_task(inode);
+	struct nsproxy *nsp;
 	struct mnt_namespace *ns = NULL;
 	struct proc_mounts *p;
 	int ret = -EINVAL;
 
 	if (task) {
-		task_lock(task);
-		if (task->nsproxy) {
-			ns = task->nsproxy->mnt_ns;
+		rcu_read_lock();
+		nsp = task_nsproxy(task);
+		if (nsp) {
+			ns = nsp->mnt_ns;
 			if (ns)
 				get_mnt_ns(ns);
 		}
-		task_unlock(task);
+		rcu_read_unlock();
+
 		put_task_struct(task);
 	}
 

@@ -445,16 +448,20 @@ static int mountstats_open(struct inode 
 	if (!ret) {
 		struct seq_file *m = file->private_data;
+		struct nsproxy *nsp;
 		struct mnt_namespace *mnt_ns = NULL;
 		struct task_struct *task = get_proc_task(inode);
 
 		if (task) {
-			task_lock(task);
-			if (task->nsproxy)
-				mnt_ns = task->nsproxy->mnt_ns;
-			if (mnt_ns)
-				get_mnt_ns(mnt_ns);
-			task_unlock(task);
+			rcu_read_lock();
+			nsp = task_nsproxy(task);
+			if (nsp) {
+				mnt_ns = nsp->mnt_ns;
+				if (mnt_ns)
+					get_mnt_ns(mnt_ns);
+			}
+			rcu_read_unlock();
+
 			put_task_struct(task);
 		}
 

include/linux/nsproxy.h +12 -0

index 525d8fc..ef4e863 100644
@@ -32,8 +32,39 @@ struct nsproxy {
 };
 extern struct nsproxy init_nsproxy;
 
+/*
+ * the namespaces access rules are:
+ *
+ *  1. only current task is allowed to change tsk->nsproxy pointer or
+ *     any pointer on the nsproxy itself
+ *
+ *  2. when accessing (i.e. reading) current task's namespaces - no
+ *     precautions should be taken - just dereference the pointers
+ *
+ *  3. the access to other task namespaces is performed like this
+ *     rcu_read_lock();
+ *     nsproxy = task_nsproxy(tsk);

include/linux/nsproxy.h +10 -12

+ *
+ */
+
+static inline struct nsproxy *task_nsproxy(struct task_struct *tsk)
+{
+	return rcu_dereference(tsk->nsproxy);
+}
+
 int copy_namespaces(unsigned long flags, struct task_struct *tsk);
-void get_task_namespaces(struct task_struct *tsk);
+void exit_task_namespaces(struct task_struct *tsk);
+void switch_task_namespaces(struct task_struct *tsk, struct nsproxy *new);
 void free_nsproxy(struct nsproxy *ns);
 int unshare_nsproxy_namespaces(unsigned long, struct nsproxy **,
 	struct fs_struct *);

@@ -45,17 +76,6 @@ static inline void put_nsproxy(struct ns
 	}
 }
 
-static inline void exit_task_namespaces(struct task_struct *p)
-{
-	struct nsproxy *ns = p->nsproxy;
-	if (ns) {
-		task_lock(p);
-		p->nsproxy = NULL;
-		task_unlock(p);
-		put_nsproxy(ns);
-	}
-}
-
 #ifdef CONFIG_CONTAINER_NS
 int ns_container_clone(struct task_struct *tsk);
 #else

kernel/exit.c +1 -3

index 2f4f35e..a2aef1f 100644
@@ -409,9 +409,7 @@ void daemonize(const char *name, ...)
 	current->fs = fs;
 	atomic_inc(&fs->count);
 
-	exit_task_namespaces(current);
-	current->nsproxy = init_task.nsproxy;
-	get_task_namespaces(current);
+	switch_task_namespaces(current, init_task.nsproxy);
 
 	exit_files(current);
 	current->files = init_task.files;

kernel/fork.c +5 -6

index bc48e0f..0c6dd67 100644
@@ -1618,7 +1618,7 @@ asmlinkage long sys_unshare(unsigned lon
 	struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
 	struct files_struct *fd, *new_fd = NULL;
 	struct sem_undo_list *new_ulist = NULL;
-	struct nsproxy *new_nsproxy = NULL, *old_nsproxy = NULL;
+	struct nsproxy *new_nsproxy = NULL;
 
 	check_unshare_flags(&unshare_flags);
 

@@ -1647,14 +1647,13 @@ asmlinkage long sys_unshare(unsigned lon
 
 	if (new_fs ||  new_mm || new_fd || new_ulist || new_nsproxy) {
 
-		task_lock(current);
-
 		if (new_nsproxy) {
-			old_nsproxy = current->nsproxy;
-			current->nsproxy = new_nsproxy;
-			new_nsproxy = old_nsproxy;
+			switch_task_namespaces(current, new_nsproxy);
+			new_nsproxy = NULL;
 		}
 
+		task_lock(current);
+
 		if (new_fs) {
 			fs = current->fs;
 			current->fs = new_fs;

kernel/nsproxy.c +24 -8

index 58843ae..48bc070 100644
@@ -30,14 +30,6 @@ static inline void get_nsproxy(struct ns
 	atomic_inc(&ns->count);
 }
 
-void get_task_namespaces(struct task_struct *tsk)
-{
-	struct nsproxy *ns = tsk->nsproxy;
-	if (ns) {
-		get_nsproxy(ns);
-	}
-}
-
 /*
 * creates a copy of "orig" with refcount 1.
 */

@@ -205,6 +197,37 @@ out:
 	return err;
 }
 
+void switch_task_namespaces(struct task_struct *p, struct nsproxy *new)
+{
+	struct nsproxy *ns;
+
+	might_sleep();
+
+	ns = p->nsproxy;
+	if (ns == new)
+		return;
+
+	if (new)
+		get_nsproxy(new);
+	rcu_assign_pointer(p->nsproxy, new);
+
+	if (ns && atomic_dec_and_test(&ns->count)) {
+		/*
+		 * wait for others to get what they want from this nsproxy.
+		 *
+		 * cannot release this nsproxy via the call_rcu() since
+		 * put_mnt_ns() will want to sleep
+		 */
+		synchronize_rcu();
+		free_nsproxy(ns);
+	}

Quoted 21 lines · Pavel Emelyanov

+/*
+ * the namespaces access rules are:
+ *
+ * 1. only current task is allowed to change tsk->nsproxy pointer or
+ * any pointer on the nsproxy itself
+ *
+ * 2. when accessing (i.e. reading) current task's namespaces - no
+ * precautions should be taken - just dereference the pointers
+ *
+ * 3. the access to other task namespaces is performed like this

+ *     rcu_read_lock();
+ *     nsproxy = task_nsproxy(tsk);
+ *     if (nsproxy != NULL) {
+ *             / *
+ *               * work with the namespaces here
+ *               * e.g. get the reference on one of them
+ *               * /
+ *     } / *
+ *         * NULL task_nsproxy() means that this task is
+ *         * almost dead (zombie)
+ *         * /

Quoted 1 line

+ * rcu_read_unlock();

And lastly, I guess that the caller to switch_task_namespaces() has to ensure that new_nsproxy either (1) is the init namespace, (2) is a brand-new namespace to which noone else has a reference, or (3) the caller has to hold a reference to the new_nsproxy across the call to switch_task_namespaces().

As it happens the current calls fit (1) or (2). Again if we happen to jump into the game of switching a task into another task's nsproxy, we'll need to be mindful of (3) so that new_nsproxy can't be tossed into the bin between

	if (new)
		get_nsproxy(new);

Quoted 36 lines · Serge E. Hallyn

Quoted 21 lines · Pavel Emelyanov

+/*
+ * the namespaces access rules are:
+ *
+ * 1. only current task is allowed to change tsk->nsproxy pointer or
+ * any pointer on the nsproxy itself
+ *
+ * 2. when accessing (i.e. reading) current task's namespaces - no
+ * precautions should be taken - just dereference the pointers
+ *
+ * 3. the access to other task namespaces is performed like this

+ *     rcu_read_lock();
+ *     nsproxy = task_nsproxy(tsk);
+ *     if (nsproxy != NULL) {
+ *             / *
+ *               * work with the namespaces here
+ *               * e.g. get the reference on one of them
+ *               * /
+ *     } / *
+ *         * NULL task_nsproxy() means that this task is
+ *         * almost dead (zombie)
+ *         * /

Quoted 1 line

+ * rcu_read_unlock();

And lastly, I guess that the caller to switch_task_namespaces() has to ensure that new_nsproxy either (1) is the init namespace, (2) is a brand-new namespace to which noone else has a reference, or (3) the caller has to hold a reference to the new_nsproxy across the call to switch_task_namespaces().

As it happens the current calls fit (1) or (2). Again if we happen to jump into the game of switching a task into another task's nsproxy, we'll need to be mindful of (3) so that new_nsproxy can't be tossed into the bin between

Quoted 2 lines

	if (new)
		get_nsproxy(new);

4) Unless tsk == current, get_task_namespaces(tsk) and get_nsproxy(tsk) are racy even if done under rcu_read_lock().

Quoted 41 lines · Oleg Nesterov

Quoted 38 lines · Serge E. Hallyn

Quoted 21 lines · Pavel Emelyanov

+/*
+ * the namespaces access rules are:
+ *
+ * 1. only current task is allowed to change tsk->nsproxy pointer or
+ * any pointer on the nsproxy itself
+ *
+ * 2. when accessing (i.e. reading) current task's namespaces - no
+ * precautions should be taken - just dereference the pointers
+ *
+ * 3. the access to other task namespaces is performed like this

+ *     rcu_read_lock();
+ *     nsproxy = task_nsproxy(tsk);
+ *     if (nsproxy != NULL) {
+ *             / *
+ *               * work with the namespaces here
+ *               * e.g. get the reference on one of them
+ *               * /
+ *     } / *
+ *         * NULL task_nsproxy() means that this task is
+ *         * almost dead (zombie)
+ *         * /

Quoted 1 line

+ * rcu_read_unlock();

And lastly, I guess that the caller to switch_task_namespaces() has to ensure that new_nsproxy either (1) is the init namespace, (2) is a brand-new namespace to which noone else has a reference, or (3) the caller has to hold a reference to the new_nsproxy across the call to switch_task_namespaces().

As it happens the current calls fit (1) or (2). Again if we happen to jump into the game of switching a task into another task's nsproxy, we'll need to be mindful of (3) so that new_nsproxy can't be tossed into the bin between

	if (new)
		get_nsproxy(new);

Quoted 2 lines

4) Unless tsk == current, get_task_namespaces(tsk) and get_nsproxy(tsk) are racy even if done under rcu_read_lock().

(sorry for noise, but I'm afraid I was not clear again...)

This looks OK, we don't do get_nsproxy(not_a_current), but perhaps it is not immediately obvious that we shouldn't.

Quoted 34 lines · Serge E. Hallyn

Quoted 21 lines · Pavel Emelyanov

+/*
+ * the namespaces access rules are:
+ *
+ * 1. only current task is allowed to change tsk->nsproxy pointer or
+ * any pointer on the nsproxy itself
+ *
+ * 2. when accessing (i.e. reading) current task's namespaces - no
+ * precautions should be taken - just dereference the pointers
+ *
+ * 3. the access to other task namespaces is performed like this

+ *     rcu_read_lock();
+ *     nsproxy = task_nsproxy(tsk);
+ *     if (nsproxy != NULL) {
+ *             / *
+ *               * work with the namespaces here
+ *               * e.g. get the reference on one of them
+ *               * /
+ *     } / *
+ *         * NULL task_nsproxy() means that this task is
+ *         * almost dead (zombie)
+ *         * /

Quoted 1 line

+ * rcu_read_unlock();

And lastly, I guess that the caller to switch_task_namespaces() has to ensure that new_nsproxy either (1) is the init namespace, (2) is a brand-new namespace to which noone else has a reference, or (3) the caller has to hold a reference to the new_nsproxy across the call to switch_task_namespaces().

As it happens the current calls fit (1) or (2). Again if we happen to jump into the game of switching a task into another task's nsproxy, we'll need to be mindful of (3) so that new_nsproxy can't be tossed into the bin between

Quoted 2 lines

	if (new)
		get_nsproxy(new);

4) Unless tsk == current, get_task_namespaces(tsk) and get_nsproxy(tsk) are racy even if done under rcu_read_lock().

Oleg.

-
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 38 lines · Oleg Nesterov

Quoted 36 lines · Serge E. Hallyn

Quoted 21 lines · Pavel Emelyanov

+/*
+ * the namespaces access rules are:
+ *
+ * 1. only current task is allowed to change tsk->nsproxy pointer or
+ * any pointer on the nsproxy itself
+ *
+ * 2. when accessing (i.e. reading) current task's namespaces - no
+ * precautions should be taken - just dereference the pointers
+ *
+ * 3. the access to other task namespaces is performed like this

+ *     rcu_read_lock();
+ *     nsproxy = task_nsproxy(tsk);
+ *     if (nsproxy != NULL) {
+ *             / *
+ *               * work with the namespaces here
+ *               * e.g. get the reference on one of them
+ *               * /
+ *     } / *
+ *         * NULL task_nsproxy() means that this task is
+ *         * almost dead (zombie)
+ *         * /

Quoted 1 line

+ * rcu_read_unlock();

And lastly, I guess that the caller to switch_task_namespaces() has to ensure that new_nsproxy either (1) is the init namespace, (2) is a brand-new namespace to which noone else has a reference, or (3) the caller has to hold a reference to the new_nsproxy across the call to switch_task_namespaces().

As it happens the current calls fit (1) or (2). Again if we happen to jump into the game of switching a task into another task's nsproxy, we'll need to be mindful of (3) so that new_nsproxy can't be tossed into the bin between

	if (new)
		get_nsproxy(new);

Quoted 2 lines

4) Unless tsk == current, get_task_namespaces(tsk) and get_nsproxy(tsk) are racy even if done under rcu_read_lock().

Yup :)

It is already written in comment that only the current is allowed to change its nsproxy. I.e. when switch_task_nsproxy() is called for tsk other than current it's a BUG

Quoted 38 lines · Oleg Nesterov

Quoted 36 lines · Serge E. Hallyn

Quoted 21 lines · Pavel Emelyanov

+/*
+ * the namespaces access rules are:
+ *
+ * 1. only current task is allowed to change tsk->nsproxy pointer or
+ * any pointer on the nsproxy itself
+ *
+ * 2. when accessing (i.e. reading) current task's namespaces - no
+ * precautions should be taken - just dereference the pointers
+ *
+ * 3. the access to other task namespaces is performed like this

+ *     rcu_read_lock();
+ *     nsproxy = task_nsproxy(tsk);
+ *     if (nsproxy != NULL) {
+ *             / *
+ *               * work with the namespaces here
+ *               * e.g. get the reference on one of them
+ *               * /
+ *     } / *
+ *         * NULL task_nsproxy() means that this task is
+ *         * almost dead (zombie)
+ *         * /

Quoted 1 line

+ * rcu_read_unlock();

And lastly, I guess that the caller to switch_task_namespaces() has to ensure that new_nsproxy either (1) is the init namespace, (2) is a brand-new namespace to which noone else has a reference, or (3) the caller has to hold a reference to the new_nsproxy across the call to switch_task_namespaces().

As it happens the current calls fit (1) or (2). Again if we happen to jump into the game of switching a task into another task's nsproxy, we'll need to be mindful of (3) so that new_nsproxy can't be tossed into the bin between

	if (new)
		get_nsproxy(new);

Quoted 2 lines

4) Unless tsk == current, get_task_namespaces(tsk) and get_nsproxy(tsk) are racy even if done under rcu_read_lock().

Yup :)

It is already written in comment that only the current is allowed to change its nsproxy. I.e. when switch_task_nsproxy() is called for tsk other than current it's a BUG

Quoted 40 lines · Pavel Emelyanov

Quoted 38 lines · Oleg Nesterov

Quoted 36 lines · Serge E. Hallyn

Quoted 21 lines · Pavel Emelyanov

+/*
+ * the namespaces access rules are:
+ *
+ * 1. only current task is allowed to change tsk->nsproxy pointer or
+ * any pointer on the nsproxy itself
+ *
+ * 2. when accessing (i.e. reading) current task's namespaces - no
+ * precautions should be taken - just dereference the pointers
+ *
+ * 3. the access to other task namespaces is performed like this

+ *     rcu_read_lock();
+ *     nsproxy = task_nsproxy(tsk);
+ *     if (nsproxy != NULL) {
+ *             / *
+ *               * work with the namespaces here
+ *               * e.g. get the reference on one of them
+ *               * /
+ *     } / *
+ *         * NULL task_nsproxy() means that this task is
+ *         * almost dead (zombie)
+ *         * /

Quoted 1 line

+ * rcu_read_unlock();

And lastly, I guess that the caller to switch_task_namespaces() has to ensure that new_nsproxy either (1) is the init namespace, (2) is a brand-new namespace to which noone else has a reference, or (3) the caller has to hold a reference to the new_nsproxy across the call to switch_task_namespaces().

As it happens the current calls fit (1) or (2). Again if we happen to jump into the game of switching a task into another task's nsproxy, we'll need to be mindful of (3) so that new_nsproxy can't be tossed into the bin between

	if (new)
		get_nsproxy(new);

Quoted 8 lines

Quoted 2 lines

4) Unless tsk == current, get_task_namespaces(tsk) and get_nsproxy(tsk) are racy even if done under rcu_read_lock().

Yup :)

It is already written in comment that only the current is allowed to change its nsproxy. I.e. when switch_task_nsproxy() is called for tsk other than current it's a BUG

Yes, but what I meant is that this code

        rcu_read_lock();
        nsproxy = task_nsproxy(tsk);
        if (nsproxy != NULL)
                get_nsproxy(nsproxy);
        rcu_read_unlock();

	if (nsproxy) {
		use_it(nsproxy);
		put_nsproxy(nsproxy);
	}

is not safe despite the fact we are _not_ changing tsk->nsproxy.

The patch itself is correct because we don't do that, and the comment is right. Just it is not immediately obvious.

Quoted 63 lines · Oleg Nesterov

Quoted 40 lines · Pavel Emelyanov

Quoted 38 lines · Oleg Nesterov

Quoted 36 lines · Serge E. Hallyn

Quoted 21 lines · Pavel Emelyanov

+/*
+ * the namespaces access rules are:
+ *
+ * 1. only current task is allowed to change tsk->nsproxy pointer or
+ * any pointer on the nsproxy itself
+ *
+ * 2. when accessing (i.e. reading) current task's namespaces - no
+ * precautions should be taken - just dereference the pointers
+ *
+ * 3. the access to other task namespaces is performed like this

+ *     rcu_read_lock();
+ *     nsproxy = task_nsproxy(tsk);
+ *     if (nsproxy != NULL) {
+ *             / *
+ *               * work with the namespaces here
+ *               * e.g. get the reference on one of them
+ *               * /
+ *     } / *
+ *         * NULL task_nsproxy() means that this task is
+ *         * almost dead (zombie)
+ *         * /

Quoted 1 line

+ * rcu_read_unlock();

And lastly, I guess that the caller to switch_task_namespaces() has to ensure that new_nsproxy either (1) is the init namespace, (2) is a brand-new namespace to which noone else has a reference, or (3) the caller has to hold a reference to the new_nsproxy across the call to switch_task_namespaces().

As it happens the current calls fit (1) or (2). Again if we happen to jump into the game of switching a task into another task's nsproxy, we'll need to be mindful of (3) so that new_nsproxy can't be tossed into the bin between

	if (new)
		get_nsproxy(new);

Quoted 8 lines

Quoted 2 lines

4) Unless tsk == current, get_task_namespaces(tsk) and get_nsproxy(tsk) are racy even if done under rcu_read_lock().

Yup :)

It is already written in comment that only the current is allowed to change its nsproxy. I.e. when switch_task_nsproxy() is called for tsk other than current it's a BUG

Yes, but what I meant is that this code

        rcu_read_lock();
        nsproxy = task_nsproxy(tsk);
        if (nsproxy != NULL)
                get_nsproxy(nsproxy);
        rcu_read_unlock();

	if (nsproxy) {
		use_it(nsproxy);
		put_nsproxy(nsproxy);
	}

Quoted 4 lines

is not safe despite the fact we are _not_ changing tsk->nsproxy.

The patch itself is correct because we don't do that, and the comment is right. Just it is not immediately obvious.

Ugh. That is nasty, non obvious and almost a problem. I don't want to do get_net(nsproxy->net_ns) from another task so I can migrate network between namespaces.

But thinking about it because we don't do the other decrements until later we can still increment the counts on the individual namespaces. We just can't share nsproxy.

So if you did want to do an enter thing you could copy the nsproxy object of a task under the rcu_read_lock(), and you would be fine.