This compiles and boots, but is not intended for inclusion in -mm (yet),
just as an RFC for the naming scheme to fix the bug Andrew pointed out.
Seem ok overall?
thanks,
-serge
>From 8e9b972f7482415777e982d3bc9a0d55cbaf862b Mon Sep 17 00:00:00 2001
From: Serge E. Hallyn <[email protected]>
Date: Fri, 1 Jun 2007 15:32:15 -0400
Subject: [PATCH 1/2] containers: improve automatic container naming
The automatic naming of containers created using container_clone()
is currently broken (not protected from wraparound) and inconvenient.
Add a per-container counter for use in naming children of the container.
Before two unshares in a row by one process, and a third in another,
would result in
/node1/node2
/node3
The current scheme should result in
/node1/node1
/node2
Also, keep a hash table populated with used names, to protect
against counter wrap-around.
Signed-off-by: Serge E. Hallyn <[email protected]>
---
include/linux/container.h | 8 +++
kernel/container.c | 116 +++++++++++++++++++++++++++++++++++++++++++--
2 files changed, 120 insertions(+), 4 deletions(-)
diff --git a/include/linux/container.h b/include/linux/container.h
index 37c0bdf..61c750d 100644
--- a/include/linux/container.h
+++ b/include/linux/container.h
@@ -75,6 +75,14 @@ struct container {
atomic_t count;
/*
+ * these are to support automatic naming of automatically
+ * created containers
+ */
+ int auto_cnt_set; /* 1 if thie container is using up a auto_cnt */
+ unsigned long auto_cnt; /* the auto_cnt it's using up */
+ unsigned long next_childcnt; /* next avail cnt for child containers */
+
+ /*
* We link our 'sibling' struct into our parent's 'children'.
* Our children link their 'sibling' into our 'children'.
*/
diff --git a/kernel/container.c b/kernel/container.c
index 6f80487..2b2f6f2 100644
--- a/kernel/container.c
+++ b/kernel/container.c
@@ -55,6 +55,7 @@
#include <linux/time.h>
#include <linux/backing-dev.h>
#include <linux/sort.h>
+#include <linux/hash.h>
#include <asm/uaccess.h>
#include <asm/atomic.h>
@@ -1819,6 +1820,9 @@ static long container_create(struct container *parent, struct dentry *dentry,
INIT_LIST_HEAD(&cont->css_groups);
INIT_LIST_HEAD(&cont->release_list);
+ cont->auto_cnt_set = 0;
+ cont->next_childcnt = 0;
+
cont->parent = parent;
cont->root = parent->root;
cont->top_container = parent->top_container;
@@ -1894,6 +1898,71 @@ static inline int container_has_css_refs(struct container *cont)
return 0;
}
+#define cnthash_shift 3
+#define childcnt_hash(ptr, nr) hash_long(nr+(unsigned long)ptr, cnthash_shift)
+static DEFINE_SPINLOCK(childcntlock);
+
+struct container_child_hash_elem {
+ struct container *parent;
+ unsigned long cnt;
+ struct hlist_node list;
+};
+
+static struct hlist_head cntnum_hash_list[1 << cnthash_shift];
+
+static void init_cntnum_hash(void)
+{
+ int i;
+
+ for (i=0; i<1 << cnthash_shift; i++)
+ INIT_HLIST_HEAD(&cntnum_hash_list[i]);
+}
+
+static inline struct container_child_hash_elem *find_cntnum_hash(
+ struct container *parent, unsigned long count)
+{
+ struct container_child_hash_elem *he;
+ struct hlist_head *h;
+ struct hlist_node *tmp;
+
+ h = &cntnum_hash_list[childcnt_hash(parent, count)];
+ hlist_for_each_entry(he, tmp, h, list) {
+ if (he->parent == parent && he->cnt == count)
+ return he;
+ }
+ return NULL;
+}
+
+#define UNSET_HASH 0
+#define SET_HASH 1
+static inline int set_cntnum_hash(struct container *parent,
+ unsigned long count, int set)
+{
+ struct container_child_hash_elem *he;
+ int idx;
+
+ he = find_cntnum_hash(parent, count);
+ if (he) {
+ if (!set) {
+ hlist_del(&he->list);
+ kfree(he);
+ return 0;
+ }
+ return -ENOENT;
+ }
+ if (!set)
+ return -ENOENT;
+ he = kmalloc(sizeof(*he), GFP_KERNEL);
+ if (!he)
+ return -ENOMEM;
+ he->parent = parent;
+ he->cnt = count;
+ INIT_HLIST_NODE(&he->list);
+ idx = childcnt_hash(parent, count);
+ hlist_add_head(&he->list, &cntnum_hash_list[idx]);
+ return 0;
+}
+
static int container_rmdir(struct inode *unused_dir, struct dentry *dentry)
{
struct container *cont = dentry->d_fsdata;
@@ -1941,6 +2010,15 @@ static int container_rmdir(struct inode *unused_dir, struct dentry *dentry)
dput(d);
root->number_of_containers--;
+ if (cont->auto_cnt_set) {
+ spin_lock(&childcntlock);
+ if (set_cntnum_hash(parent, cont->auto_cnt, UNSET_HASH))
+ printk(KERN_NOTICE "%s: could not unset %lu\n",
+ __FUNCTION__, cont->auto_cnt);
+ cont->auto_cnt_set = 0;
+ spin_unlock(&childcntlock);
+ }
+
if (!list_empty(&cont->release_list))
list_del(&cont->release_list);
set_bit(CONT_RELEASABLE, &parent->flags);
@@ -2063,6 +2141,8 @@ int __init container_init(void)
err = register_filesystem(&container_fs_type);
if (err < 0)
goto out;
+
+ init_cntnum_hash();
entry = create_proc_entry("containers", 0, NULL);
if (entry)
@@ -2302,10 +2382,31 @@ void container_exit(struct task_struct *tsk, int run_callbacks)
put_css_group_taskexit(cg);
}
-static atomic_t namecnt;
-static void get_unused_name(char *buf)
+/*
+ * the while (find_cntnum_hash) loop will only be hit after
+ * we wrap around on next_childnct, which should never happen.
+ */
+static int get_unused_name(struct container *parent, unsigned long *out_cnt,
+ char *buf)
{
- sprintf(buf, "node%d", atomic_inc_return(&namecnt));
+ unsigned long cnt;
+ int err = 0;
+
+ spin_lock(&childcntlock);
+ cnt = parent->next_childcnt;
+ while (find_cntnum_hash(parent, cnt))
+ cnt++;
+ err = set_cntnum_hash(parent, cnt, SET_HASH);
+ if (err)
+ goto out_unlock;
+
+ sprintf(buf, "node%lu", cnt);
+ parent->next_childcnt = cnt+1;
+ *out_cnt = cnt;
+
+out_unlock:
+ spin_unlock(&childcntlock);
+ return err;
}
/**
@@ -2318,6 +2419,7 @@ int container_clone(struct task_struct *tsk, struct container_subsys *subsys)
struct dentry *dentry;
int ret = 0;
char nodename[32];
+ unsigned long auto_cntnum;
struct container *parent, *child;
struct inode *inode;
struct css_group *cg;
@@ -2348,7 +2450,10 @@ int container_clone(struct task_struct *tsk, struct container_subsys *subsys)
mutex_unlock(&container_mutex);
/* Now do the VFS work to create a container */
- get_unused_name(nodename);
+ ret = get_unused_name(parent, &auto_cntnum, nodename);
+ if (ret)
+ goto out_container_mutex;
+
inode = parent->dentry->d_inode;
/* Hold the parent directory mutex across this operation to
@@ -2380,6 +2485,8 @@ int container_clone(struct task_struct *tsk, struct container_subsys *subsys)
ret = -ENOMEM;
goto out_release;
}
+ child->auto_cnt_set = 1;
+ child->auto_cnt = auto_cntnum;
/* The container now exists. Retake container_mutex and check
* that we're still in the same state that we thought we
@@ -2408,6 +2515,7 @@ int container_clone(struct task_struct *tsk, struct container_subsys *subsys)
out_release:
mutex_unlock(&inode->i_mutex);
+ out_container_mutex:
mutex_lock(&container_mutex);
put_css_group(cg);
mutex_unlock(&container_mutex);
--
1.5.1.1.GIT
Again, not compiles and boots, but not sufficiently tested for
inclusion.
thanks,
-serge
>From 6c03518dbda04bf20a9ddeefb291bf1c63e32ec1 Mon Sep 17 00:00:00 2001
From: Serge E. Hallyn <[email protected]>
Date: Fri, 1 Jun 2007 17:30:16 -0400
Subject: [PATCH 2/2] Container interface to nsproxy subsystem
When a task enters a new namespace via a clone() or unshare(), a new
container is created and the task moves into it.
Signed-off-by: Serge E. Hallyn <[email protected]>
---
include/linux/container_subsys.h | 6 ++
include/linux/nsproxy.h | 7 +++
init/Kconfig | 9 ++++
kernel/Makefile | 1 +
kernel/container.c | 15 +++++-
kernel/ns_container.c | 96 ++++++++++++++++++++++++++++++++++++++
kernel/nsproxy.c | 16 ++++++
7 files changed, 148 insertions(+), 2 deletions(-)
create mode 100644 kernel/ns_container.c
diff --git a/include/linux/container_subsys.h b/include/linux/container_subsys.h
index 8fea7cf..9861751 100644
--- a/include/linux/container_subsys.h
+++ b/include/linux/container_subsys.h
@@ -24,3 +24,9 @@ SUBSYS(debug)
#endif
/* */
+
+#ifdef CONFIG_CONTAINER_NS
+SUBSYS(ns)
+#endif
+
+/* */
diff --git a/include/linux/nsproxy.h b/include/linux/nsproxy.h
index 189e0dc..8be975b 100644
--- a/include/linux/nsproxy.h
+++ b/include/linux/nsproxy.h
@@ -54,4 +54,11 @@ static inline void exit_task_namespaces(struct task_struct *p)
put_nsproxy(ns);
}
}
+
+#ifdef CONFIG_CONTAINER_NS
+int ns_container_clone(struct task_struct *tsk);
+#else
+static inline int ns_container_clone(struct task_struct *tsk) { return 0; }
+#endif
+
#endif
diff --git a/init/Kconfig b/init/Kconfig
index 5861ad9..d79c505 100644
--- a/init/Kconfig
+++ b/init/Kconfig
@@ -355,6 +355,15 @@ config CONTAINER_CPUACCT
Provides a simple Resource Controller for monitoring the
total CPU consumed by the tasks in a container
+config CONTAINER_NS
+ bool "Namespace container subsystem"
+ select CONTAINERS
+ help
+ Provides a simple namespace container subsystem to
+ provide hierarchical naming of sets of namespaces,
+ for instance virtual servers and checkpoint/restart
+ jobs.
+
config PROC_PID_CPUSET
bool "Include legacy /proc/<pid>/cpuset file"
depends on CPUSETS
diff --git a/kernel/Makefile b/kernel/Makefile
index f73b3d3..34f2345 100644
--- a/kernel/Makefile
+++ b/kernel/Makefile
@@ -40,6 +40,7 @@ obj-$(CONFIG_CONTAINERS) += container.o
obj-$(CONFIG_CONTAINER_DEBUG) += container_debug.o
obj-$(CONFIG_CPUSETS) += cpuset.o
obj-$(CONFIG_CONTAINER_CPUACCT) += cpu_acct.o
+obj-$(CONFIG_CONTAINER_NS) += ns_container.o
obj-$(CONFIG_IKCONFIG) += configs.o
obj-$(CONFIG_STOP_MACHINE) += stop_machine.o
obj-$(CONFIG_AUDIT) += audit.o auditfilter.o
diff --git a/kernel/container.c b/kernel/container.c
index 2b2f6f2..ac5879f 100644
--- a/kernel/container.c
+++ b/kernel/container.c
@@ -2523,14 +2523,25 @@ int container_clone(struct task_struct *tsk, struct container_subsys *subsys)
return ret;
}
-/* See if "cont" is a descendant of the current task's container in
- * the appropriate hierarchy */
+/*
+ * See if "cont" is a descendant of the current task's container in
+ * the appropriate hierarchy
+ *
+ * If we are sending in dummytop, then presumably we are creating
+ * the top container in the subsystem.
+ *
+ * Called only by the ns (nsproxy) container.
+ */
int container_is_descendant(const struct container *cont)
{
int ret;
struct container *target;
int subsys_id;
+
+ if (cont == dummytop)
+ return 1;
+
get_first_subsys(cont, NULL, &subsys_id);
target = task_container(current, subsys_id);
while (cont != target && cont!= cont->top_container) {
diff --git a/kernel/ns_container.c b/kernel/ns_container.c
new file mode 100644
index 0000000..afa50d9
--- /dev/null
+++ b/kernel/ns_container.c
@@ -0,0 +1,96 @@
+/*
+ * ns_container.c - namespace container subsystem
+ *
+ * Copyright 2006, 2007 IBM Corp
+ */
+
+#include <linux/module.h>
+#include <linux/container.h>
+#include <linux/fs.h>
+
+struct nscont {
+ struct container_subsys_state css;
+ spinlock_t lock;
+};
+
+struct container_subsys ns_subsys;
+
+static inline struct nscont *container_nscont(struct container *cont)
+{
+ return container_of(container_subsys_state(cont, ns_subsys_id),
+ struct nscont, css);
+}
+
+int ns_container_clone(struct task_struct *tsk)
+{
+ return container_clone(tsk, &ns_subsys);
+}
+
+/*
+ * Rules:
+ * 1. you can only enter a container which is a child of your current
+ * container
+ * 2. you can only place another process into a container if
+ * a. you have CAP_SYS_ADMIN
+ * b. your container is an ancestor of tsk's destination container
+ * (hence either you are in the same container as tsk, or in an
+ * ancestor container thereof)
+ */
+int ns_can_attach(struct container_subsys *ss,
+ struct container *cont, struct task_struct *tsk)
+{
+ struct container *c;
+
+ if (current != tsk) {
+ if (!capable(CAP_SYS_ADMIN))
+ return -EPERM;
+
+ if (!container_is_descendant(cont))
+ return -EPERM;
+ }
+
+ if (atomic_read(&cont->count) != 0)
+ return -EPERM;
+
+ c = task_container(tsk, ns_subsys_id);
+ if (c && c != cont->parent)
+ return -EPERM;
+
+ return 0;
+}
+
+/*
+ * Rules: you can only create a container if
+ * 1. you are capable(CAP_SYS_ADMIN)
+ * 2. the target container is a descendant of your own container
+ */
+static int ns_create(struct container_subsys *ss, struct container *cont)
+{
+ struct nscont *ns;
+
+ if (!capable(CAP_SYS_ADMIN))
+ return -EPERM;
+ if (!container_is_descendant(cont))
+ return -EPERM;
+
+ ns = kzalloc(sizeof(*ns), GFP_KERNEL);
+ if (!ns) return -ENOMEM;
+ spin_lock_init(&ns->lock);
+ cont->subsys[ns_subsys.subsys_id] = &ns->css;
+ return 0;
+}
+
+static void ns_destroy(struct container_subsys *ss,
+ struct container *cont)
+{
+ struct nscont *ns = container_nscont(cont);
+ kfree(ns);
+}
+
+struct container_subsys ns_subsys = {
+ .name = "ns",
+ .create = ns_create,
+ .destroy = ns_destroy,
+ .can_attach = ns_can_attach,
+ .subsys_id = ns_subsys_id,
+};
diff --git a/kernel/nsproxy.c b/kernel/nsproxy.c
index 1bc4b55..afce808 100644
--- a/kernel/nsproxy.c
+++ b/kernel/nsproxy.c
@@ -124,7 +124,14 @@ int copy_namespaces(int flags, struct task_struct *tsk)
goto out;
}
+ err = ns_container_clone(tsk);
+ if (err) {
+ put_nsproxy(new_ns);
+ goto out;
+ }
+
tsk->nsproxy = new_ns;
+
out:
put_nsproxy(old_ns);
return err;
@@ -177,6 +184,15 @@ int unshare_nsproxy_namespaces(unsigned long unshare_flags,
if (IS_ERR(*new_nsp)) {
err = PTR_ERR(*new_nsp);
put_nsproxy(old_ns);
+ goto out;
+ }
+
+ err = ns_container_clone(current);
+ if (err) {
+ put_nsproxy(*new_nsp);
+ put_nsproxy(old_ns);
}
+
+out:
return err;
}
--
1.5.1.1.GIT
On Fri, 1 Jun 2007 16:48:09 -0500
"Serge E. Hallyn" <[email protected]> wrote:
> This compiles and boots, but is not intended for inclusion in -mm (yet),
> just as an RFC for the naming scheme to fix the bug Andrew pointed out.
>
> Seem ok overall?
>
> thanks,
> -serge
>
> >From 8e9b972f7482415777e982d3bc9a0d55cbaf862b Mon Sep 17 00:00:00 2001
> From: Serge E. Hallyn <[email protected]>
> Date: Fri, 1 Jun 2007 15:32:15 -0400
> Subject: [PATCH 1/2] containers: improve automatic container naming
>
> The automatic naming of containers created using container_clone()
> is currently broken (not protected from wraparound) and inconvenient.
>
> Add a per-container counter for use in naming children of the container.
> Before two unshares in a row by one process, and a third in another,
> would result in
>
> /node1/node2
> /node3
>
> The current scheme should result in
>
> /node1/node1
> /node2
>
> Also, keep a hash table populated with used names, to protect
> against counter wrap-around.
>
> ...
>
> include/linux/container.h | 8 +++
> kernel/container.c | 116 +++++++++++++++++++++++++++++++++++++++++++--
gad, what's all this stuff?
I think an IDR tree would get you what you're after in much less code.
Although it means that container IDs would get recycled quickly across a
remove+add.
Be aware that there are IDR enhancements in Greg's driver tree (and hence
in -mm) which are relevant to this application.
> + if (cont->auto_cnt_set) {
Can we please stop using "cnt" and "cont" to refer to containers? Let's
use "container", OK?
Quoting Andrew Morton ([email protected]):
> On Fri, 1 Jun 2007 16:48:09 -0500
> "Serge E. Hallyn" <[email protected]> wrote:
>
> > This compiles and boots, but is not intended for inclusion in -mm (yet),
> > just as an RFC for the naming scheme to fix the bug Andrew pointed out.
> >
> > Seem ok overall?
> >
> > thanks,
> > -serge
> >
> > >From 8e9b972f7482415777e982d3bc9a0d55cbaf862b Mon Sep 17 00:00:00 2001
> > From: Serge E. Hallyn <[email protected]>
> > Date: Fri, 1 Jun 2007 15:32:15 -0400
> > Subject: [PATCH 1/2] containers: improve automatic container naming
> >
> > The automatic naming of containers created using container_clone()
> > is currently broken (not protected from wraparound) and inconvenient.
> >
> > Add a per-container counter for use in naming children of the container.
> > Before two unshares in a row by one process, and a third in another,
> > would result in
> >
> > /node1/node2
> > /node3
> >
> > The current scheme should result in
> >
> > /node1/node1
> > /node2
> >
> > Also, keep a hash table populated with used names, to protect
> > against counter wrap-around.
> >
> > ...
> >
> > include/linux/container.h | 8 +++
> > kernel/container.c | 116 +++++++++++++++++++++++++++++++++++++++++++--
>
> gad, what's all this stuff?
The sound of me starting over.
> I think an IDR tree would get you what you're after in much less code.
Ok, will look at that. Thanks for the tip!
> Although it means that container IDs would get recycled quickly across a
> remove+add.
>
> Be aware that there are IDR enhancements in Greg's driver tree (and hence
> in -mm) which are relevant to this application.
>
>
> > + if (cont->auto_cnt_set) {
>
> Can we please stop using "cnt" and "cont" to refer to containers? Let's
> use "container", OK?
Ok.
thanks,
-serge