Task Control Groups: shared cgroup subsystem group arrays
Changes
5 files changed · +632 −154
| File | Status | Add | Del | Content |
|---|---|---|---|---|
| Documentation/cgroups.txt | modified | +13 | −1 | View file |
| include/linux/cgroup.h | modified | +79 | −10 | View file |
| include/linux/sched.h | modified | +4 | −29 | View file |
| kernel/cgroup.c | modified | +535 | −114 | View file |
| kernel/fork.c | modified | +1 | −0 | View file |
Only the first 600 diff lines are shown. Download the complete patch.
Documentation/cgroups.txt
View filediff --git a/Documentation/cgroups.txt b/Documentation/cgroups.txt
index 553727c..98a26f8 100644
--- a/Documentation/cgroups.txt
+++ b/Documentation/cgroups.txt
@@ -176,7 +176,9 @@ Control Groups extends the kernel as follows:
subsystem state is something that's expected to happen frequently
and in performance-critical code, whereas operations that require a
task's actual cgroup assignments (in particular, moving between
- cgroups) are less common.
+ cgroups) are less common. A linked list runs through the cg_list
+ field of each task_struct using the css_set, anchored at
+ css_set->tasks.
- A cgroup hierarchy filesystem can be mounted for browsing and
manipulation from user space.
@@ -252,6 +254,16 @@ linear search to locate an appropriate existing css_set, so isn't
very efficient. A future version will use a hash table for better
performance.
+To allow access from a cgroup to the css_sets (and hence tasks)
+that comprise it, a set of cg_cgroup_link objects form a lattice;
+each cg_cgroup_link is linked into a list of cg_cgroup_links for
+a single cgroup on its cont_link_list field, and a list of
+cg_cgroup_links for a single css_set on its cg_link_list.
+
+Thus the set of tasks in a cgroup can be listed by iterating over
+each css_set that references the cgroup, and sub-iterating over
+each css_set's task set.
+
The use of a Linux virtual file system (vfs) to represent the
cgroup hierarchy provides for a familiar permission and name space
for cgroups, with a minimum of additional kernel code.
include/linux/cgroup.h
View filediff --git a/include/linux/cgroup.h b/include/linux/cgroup.h
index a955356..836b355 100644
--- a/include/linux/cgroup.h
+++ b/include/linux/cgroup.h
@@ -27,10 +27,19 @@ extern void cgroup_lock(void);
extern void cgroup_unlock(void);
extern void cgroup_fork(struct task_struct *p);
extern void cgroup_fork_callbacks(struct task_struct *p);
+extern void cgroup_post_fork(struct task_struct *p);
extern void cgroup_exit(struct task_struct *p, int run_callbacks);
extern struct file_operations proc_cgroup_operations;
+/* Define the enumeration of all cgroup subsystems */
+#define SUBSYS(_x) _x ## _subsys_id,
+enum cgroup_subsys_id {
+#include <linux/cgroup_subsys.h>
+ CGROUP_SUBSYS_COUNT
+};
+#undef SUBSYS
+
/* Per-subsystem/per-cgroup state maintained by the system. */
struct cgroup_subsys_state {
/* The cgroup that this subsystem is attached to. Useful
@@ -97,6 +106,52 @@ struct cgroup {
struct cgroupfs_root *root;
struct cgroup *top_cgroup;
+
+ /*
+ * List of cg_cgroup_links pointing at css_sets with
+ * tasks in this cgroup. Protected by css_set_lock
+ */
+ struct list_head css_sets;
+};
+
+/* A css_set is a structure holding pointers to a set of
+ * cgroup_subsys_state objects. This saves space in the task struct
+ * object and speeds up fork()/exit(), since a single inc/dec and a
+ * list_add()/del() can bump the reference count on the entire
+ * cgroup set for a task.
+ */
+
+struct css_set {
+
+ /* Reference count */
+ struct kref ref;
+
+ /*
+ * List running through all cgroup groups. Protected by
+ * css_set_lock
+ */
+ struct list_head list;
+
+ /*
+ * List running through all tasks using this cgroup
+ * group. Protected by css_set_lock
+ */
+ struct list_head tasks;
+
+ /*
+ * List of cg_cgroup_link objects on link chains from
+ * cgroups referenced from this css_set. Protected by
+ * css_set_lock
+ */
+ struct list_head cg_links;
+
+ /*
+ * Set of subsystem states, one for each subsystem. This array
+ * is immutable after creation apart from the init_css_set
+ * during subsystem registration (at boot time).
+ */
+ struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
+
};
/* struct cftype:
@@ -157,15 +212,7 @@ int cgroup_is_removed(const struct cgroup *cont);
int cgroup_path(const struct cgroup *cont, char *buf, int buflen);
-int __cgroup_task_count(const struct cgroup *cont);
-static inline int cgroup_task_count(const struct cgroup *cont)
-{
- int task_count;
- rcu_read_lock();
- task_count = __cgroup_task_count(cont);
- rcu_read_unlock();
- return task_count;
-}
+int cgroup_task_count(const struct cgroup *cont);
/* Return true if the cgroup is a descendant of the current cgroup */
int cgroup_is_descendant(const struct cgroup *cont);
@@ -213,7 +260,7 @@ static inline struct cgroup_subsys_state *cgroup_subsys_state(
static inline struct cgroup_subsys_state *task_subsys_state(
struct task_struct *task, int subsys_id)
{
- return rcu_dereference(task->cgroups.subsys[subsys_id]);
+ return rcu_dereference(task->cgroups->subsys[subsys_id]);
}
static inline struct cgroup* task_cgroup(struct task_struct *task,
@@ -226,6 +273,27 @@ int cgroup_path(const struct cgroup *cont, char *buf, int buflen);
int cgroup_clone(struct task_struct *tsk, struct cgroup_subsys *ss);
+/* A cgroup_iter should be treated as an opaque object */
+struct cgroup_iter {
+ struct list_head *cg_link;
+ struct list_head *task;
+};
+
+/* To iterate across the tasks in a cgroup:
+ *
+ * 1) call cgroup_iter_start to intialize an iterator
+ *
+ * 2) call cgroup_iter_next() to retrieve member tasks until it
+ * returns NULL or until you want to end the iteration
+ *
+ * 3) call cgroup_iter_end() to destroy the iterator.
+ */
+void cgroup_iter_start(struct cgroup *cont, struct cgroup_iter *it);
+struct task_struct *cgroup_iter_next(struct cgroup *cont,
+ struct cgroup_iter *it);
+void cgroup_iter_end(struct cgroup *cont, struct cgroup_iter *it);
+
+
#else /* !CONFIG_CGROUPS */
static inline int cgroup_init_early(void) { return 0; }
@@ -233,6 +301,7 @@ static inline int cgroup_init(void) { return 0; }
static inline void cgroup_init_smp(void) {}
static inline void cgroup_fork(struct task_struct *p) {}
static inline void cgroup_fork_callbacks(struct task_struct *p) {}
+static inline void cgroup_post_fork(struct task_struct *p) {}
static inline void cgroup_exit(struct task_struct *p, int callbacks) {}
static inline void cgroup_lock(void) {}
include/linux/sched.h
View filediff --git a/include/linux/sched.h b/include/linux/sched.h
index af2ed4b..1aa1cfa 100644
--- a/include/linux/sched.h
+++ b/include/linux/sched.h
@@ -894,34 +894,6 @@ struct sched_entity {
#endif
};
-#ifdef CONFIG_CGROUPS
-
-#define SUBSYS(_x) _x ## _subsys_id,
-enum cgroup_subsys_id {
-#include <linux/cgroup_subsys.h>
- CGROUP_SUBSYS_COUNT
-};
-#undef SUBSYS
-
-/* A css_set is a structure holding pointers to a set of
- * cgroup_subsys_state objects.
- */
-
-struct css_set {
-
- /* Set of subsystem states, one for each subsystem. NULL for
- * subsystems that aren't part of this hierarchy. These
- * pointers reduce the number of dereferences required to get
- * from a task to its state for a given cgroup, but result
- * in increased space usage if tasks are in wildly different
- * groupings across different hierarchies. This array is
- * immutable after creation */
- struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
-
-};
-
-#endif /* CONFIG_CGROUPS */
-
struct task_struct {
volatile long state; /* -1 unrunnable, 0 runnable, >0 stopped */
void *stack;
@@ -1159,7 +1131,10 @@ struct task_struct {
int cpuset_mem_spread_rotor;
#endif
#ifdef CONFIG_CGROUPS
- struct css_set cgroups;
+ /* Control Group info protected by css_set_lock */
+ struct css_set *cgroups;
+ /* cg_list protected by css_set_lock and tsk->alloc_lock */
+ struct list_head cg_list;
#endif
#ifdef CONFIG_FUTEX
struct robust_list_head __user *robust_list;
kernel/cgroup.c
View filediff --git a/kernel/cgroup.c b/kernel/cgroup.c
index db245f1..883928c 100644
--- a/kernel/cgroup.c
+++ b/kernel/cgroup.c
@@ -36,6 +36,7 @@
#include <linux/proc_fs.h>
#include <linux/rcupdate.h>
#include <linux/sched.h>
+#include <linux/backing-dev.h>
#include <linux/seq_file.h>
#include <linux/slab.h>
#include <linux/magic.h>
@@ -95,6 +96,7 @@ static struct cgroupfs_root rootnode;
/* The list of hierarchy roots */
static LIST_HEAD(roots);
+static int root_count;
/* dummytop is a shorthand for the dummy hierarchy's top cgroup */
#define dummytop (&rootnode.top_cgroup)
@@ -133,12 +135,49 @@ list_for_each_entry(_ss, &_root->subsys_list, sibling)
#define for_each_root(_root) \
list_for_each_entry(_root, &roots, root_list)
-/* Each task_struct has an embedded css_set, so the get/put
- * operation simply takes a reference count on all the cgroups
- * referenced by subsystems in this css_set. This can end up
- * multiple-counting some cgroups, but that's OK - the ref-count is
- * just a busy/not-busy indicator; ensuring that we only count each
- * cgroup once would require taking a global lock to ensure that no
+/* Link structure for associating css_set objects with cgroups */
+struct cg_cgroup_link {
+ /*
+ * List running through cg_cgroup_links associated with a
+ * cgroup, anchored on cgroup->css_sets
+ */
+ struct list_head cont_link_list;
+ /*
+ * List running through cg_cgroup_links pointing at a
+ * single css_set object, anchored on css_set->cg_links
+ */
+ struct list_head cg_link_list;
+ struct css_set *cg;
+};
+
+/* The default css_set - used by init and its children prior to any
+ * hierarchies being mounted. It contains a pointer to the root state
+ * for each subsystem. Also used to anchor the list of css_sets. Not
+ * reference-counted, to improve performance when child cgroups
+ * haven't been created.
+ */
+
+static struct css_set init_css_set;
+static struct cg_cgroup_link init_css_set_link;
+
+/* css_set_lock protects the list of css_set objects, and the
+ * chain of tasks off each css_set. Nests outside task->alloc_lock
+ * due to cgroup_iter_start() */
+static DEFINE_RWLOCK(css_set_lock);
+static int css_set_count;
+
+/* We don't maintain the lists running through each css_set to its
+ * task until after the first call to cgroup_iter_start(). This
+ * reduces the fork()/exit() overhead for people who have cgroups
+ * compiled into their kernel but not actually in use */
+static int use_task_css_set_links;
+
+/* When we create or destroy a css_set, the operation simply
+ * takes/releases a reference count on all the cgroups referenced
+ * by subsystems in this css_set. This can end up multiple-counting
+ * some cgroups, but that's OK - the ref-count is just a
+ * busy/not-busy indicator; ensuring that we only count each cgroup
+ * once would require taking a global lock to ensure that no
* subsystems moved between hierarchies while we were doing so.
*
* Possible TODO: decide at boot time based on the number of
@@ -146,18 +185,230 @@ list_for_each_entry(_root, &roots, root_list)
* it's better for performance to ref-count every subsystem, or to
* take a global lock and only add one ref count to each hierarchy.
*/
-static void get_css_set(struct css_set *cg)
+
+/*
+ * unlink a css_set from the list and free it
+ */
+static void release_css_set(struct kref *k)
{
+ struct css_set *cg = container_of(k, struct css_set, ref);
int i;
+
+ write_lock(&css_set_lock);
+ list_del(&cg->list);
+ css_set_count--;
+ while (!list_empty(&cg->cg_links)) {
+ struct cg_cgroup_link *link;
+ link = list_entry(cg->cg_links.next,
+ struct cg_cgroup_link, cg_link_list);
+ list_del(&link->cg_link_list);
+ list_del(&link->cont_link_list);
+ kfree(link);
+ }
+ write_unlock(&css_set_lock);
for (i = 0; i < CGROUP_SUBSYS_COUNT; i++)
- atomic_inc(&cg->subsys[i]->cgroup->count);
+ atomic_dec(&cg->subsys[i]->cgroup->count);
+ kfree(cg);
}
-static void put_css_set(struct css_set *cg)
+/*
+ * refcounted get/put for css_set objects
+ */
+static inline void get_css_set(struct css_set *cg)
+{
+ kref_get(&cg->ref);
+}
+
+static inline void put_css_set(struct css_set *cg)
+{
+ kref_put(&cg->ref, release_css_set);
+}
+
+/*
+ * find_existing_css_set() is a helper for
+ * find_css_set(), and checks to see whether an existing
+ * css_set is suitable. This currently walks a linked-list for
+ * simplicity; a later patch will use a hash table for better
+ * performance
+ *
+ * oldcg: the cgroup group that we're using before the cgroup
+ * transition
+ *
+ * cont: the cgroup that we're moving into
+ *
+ * template: location in which to build the desired set of subsystem
+ * state objects for the new cgroup group
+ */
+
+static struct css_set *find_existing_css_set(
+ struct css_set *oldcg,
+ struct cgroup *cont,
+ struct cgroup_subsys_state *template[])
{
int i;
- for (i = 0; i < CGROUP_SUBSYS_COUNT; i++)
- atomic_dec(&cg->subsys[i]->cgroup->count);
+ struct cgroupfs_root *root = cont->root;
+ struct list_head *l = &init_css_set.list;
+
+ /* Built the set of subsystem state objects that we want to
+ * see in the new css_set */
+ for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
+ if (root->subsys_bits & (1ull << i)) {
+ /* Subsystem is in this hierarchy. So we want
+ * the subsystem state from the new
+ * cgroup */
+ template[i] = cont->subsys[i];
+ } else {
+ /* Subsystem is not in this hierarchy, so we
+ * don't want to change the subsystem state */
+ template[i] = oldcg->subsys[i];
+ }
+ }
+
+ /* Look through existing cgroup groups to find one to reuse */
+ do {
+ struct css_set *cg =
+ list_entry(l, struct css_set, list);
+
+ if (!memcmp(template, cg->subsys, sizeof(cg->subsys))) {
+ /* All subsystems matched */
+ return cg;
+ }
+ /* Try the next cgroup group */
+ l = l->next;
+ } while (l != &init_css_set.list);
+
+ /* No existing cgroup group matched */
+ return NULL;
+}
+
+/*
+ * allocate_cg_links() allocates "count" cg_cgroup_link structures
+ * and chains them on tmp through their cont_link_list fields. Returns 0 on
+ * success or a negative error
+ */
+
+static int allocate_cg_links(int count, struct list_head *tmp)
+{
+ struct cg_cgroup_link *link;
+ int i;
+ INIT_LIST_HEAD(tmp);
+ for (i = 0; i < count; i++) {
+ link = kmalloc(sizeof(*link), GFP_KERNEL);
+ if (!link) {
+ while (!list_empty(tmp)) {
+ link = list_entry(tmp->next,
+ struct cg_cgroup_link,
+ cont_link_list);
+ list_del(&link->cont_link_list);
+ kfree(link);
+ }
+ return -ENOMEM;
+ }
+ list_add(&link->cont_link_list, tmp);
+ }
+ return 0;
+}
+
+static void free_cg_links(struct list_head *tmp)
+{
+ while (!list_empty(tmp)) {
+ struct cg_cgroup_link *link;
+ link = list_entry(tmp->next,
+ struct cg_cgroup_link,
+ cont_link_list);
+ list_del(&link->cont_link_list);
+ kfree(link);
+ }
+}
+
+/*
+ * find_css_set() takes an existing cgroup group and a
+ * cgroup object, and returns a css_set object that's
+ * equivalent to the old group, but with the given cgroup
+ * substituted into the appropriate hierarchy. Must be called with
+ * cgroup_mutex held
+ */
+
+static struct css_set *find_css_set(
+ struct css_set *oldcg, struct cgroup *cont)
+{
+ struct css_set *res;
+ struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT];
+ int i;
+
+ struct list_head tmp_cg_links;
+ struct cg_cgroup_link *link;
+
+ /* First see if we already have a cgroup group that matches
+ * the desired set */
+ write_lock(&css_set_lock);
+ res = find_existing_css_set(oldcg, cont, template);
+ if (res)
+ get_css_set(res);
+ write_unlock(&css_set_lock);
+
+ if (res)
+ return res;
+
+ res = kmalloc(sizeof(*res), GFP_KERNEL);
+ if (!res)
+ return NULL;
+
+ /* Allocate all the cg_cgroup_link objects that we'll need */
+ if (allocate_cg_links(root_count, &tmp_cg_links) < 0) {
+ kfree(res);
+ return NULL;
+ }
+
+ kref_init(&res->ref);
+ INIT_LIST_HEAD(&res->cg_links);
+ INIT_LIST_HEAD(&res->tasks);
+
+ /* Copy the set of subsystem state objects generated in
+ * find_existing_css_set() */
+ memcpy(res->subsys, template, sizeof(res->subsys));
+
+ write_lock(&css_set_lock);
+ /* Add reference counts and links from the new css_set. */
+ for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
+ struct cgroup *cont = res->subsys[i]->cgroup;
+ struct cgroup_subsys *ss = subsys[i];
+ atomic_inc(&cont->count);
+ /*
+ * We want to add a link once per cgroup, so we
+ * only do it for the first subsystem in each
+ * hierarchy
+ */
+ if (ss->root->subsys_list.next == &ss->sibling) {
+ BUG_ON(list_empty(&tmp_cg_links));
+ link = list_entry(tmp_cg_links.next,
+ struct cg_cgroup_link,
+ cont_link_list);
+ list_del(&link->cont_link_list);
+ list_add(&link->cont_link_list, &cont->css_sets);
+ link->cg = res;
+ list_add(&link->cg_link_list, &res->cg_links);
+ }
+ }
+ if (list_empty(&rootnode.subsys_list)) {
+ link = list_entry(tmp_cg_links.next,
+ struct cg_cgroup_link,
+ cont_link_list);
+ list_del(&link->cont_link_list);
+ list_add(&link->cont_link_list, &dummytop->css_sets);
+ link->cg = res;
+ list_add(&link->cg_link_list, &res->cg_links);
+ }
+
+ BUG_ON(!list_empty(&tmp_cg_links));
+
+ /* Link this cgroup group into the list */
+ list_add(&res->list, &init_css_set.list);
+ css_set_count++;
+ INIT_LIST_HEAD(&res->tasks);
+ write_unlock(&css_set_lock);
+
+ return res;
}
/*
@@ -504,6 +755,7 @@ static void init_cgroup_root(struct cgroupfs_root *root)
cont->top_cgroup = cont;
INIT_LIST_HEAD(&cont->sibling);
INIT_LIST_HEAD(&cont->children);
+ INIT_LIST_HEAD(&cont->css_sets);
}
static int cgroup_test_super(struct super_block *sb, void *data)
@@ -573,6 +825,8 @@ static int cgroup_get_sb(struct file_system_type *fs_type,
int ret = 0;
struct super_block *sb;
struct cgroupfs_root *root;
+ struct list_head tmp_cg_links, *l;
+ INIT_LIST_HEAD(&tmp_cg_links);
/* First find the desired set of subsystems */
ret = parse_cgroupfs_options(data, &opts);
@@ -602,18 +856,36 @@ static int cgroup_get_sb(struct file_system_type *fs_type,
} else {
/* New superblock */
struct cgroup *cont = &root->top_cgroup;
+ struct inode *inode;
BUG_ON(sb->s_root != NULL);
ret = cgroup_get_rootdir(sb);
if (ret)
goto drop_new_super;
+ inode = sb->s_root->d_inode;
+ mutex_lock(&inode->i_mutex);
mutex_lock(&cgroup_mutex);
+ /*
+ * We're accessing css_set_count without locking
+ * css_set_lock here, but that's OK - it can only be
+ * increased by someone holding cgroup_lock, and
+ * that's us. The worst that can happen is that we
+ * have some link structures left over
+ */
+ ret = allocate_cg_links(css_set_count, &tmp_cg_links);
+ if (ret) {
+ mutex_unlock(&cgroup_mutex);
+ mutex_unlock(&inode->i_mutex);
+ goto drop_new_super;
+ }
+
ret = rebind_subsystems(root, root->subsys_bits);
if (ret == -EBUSY) {
mutex_unlock(&cgroup_mutex);
+ mutex_unlock(&inode->i_mutex);
goto drop_new_super;
}
@@ -621,24 +893,40 @@ static int cgroup_get_sb(struct file_system_type *fs_type,
BUG_ON(ret);
list_add(&root->root_list, &roots);
+ root_count++;
sb->s_root->d_fsdata = &root->top_cgroup;
root->top_cgroup.dentry = sb->s_root;
+ /* Link the top cgroup in this hierarchy into all
+ * the css_set objects */
+ write_lock(&css_set_lock);
This diff is incomplete. Download the complete patch.
kernel/fork.c
View fileThis diff is incomplete. Download the complete patch.