Linux cgroup

include/linux/cgroup-defs.h

/*
 * 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 */
    atomic_t refcount;

    /*
     * List running through all cgroup groups in the same hash
     * slot. Protected by css_set_lock
     */
    struct hlist_node hlist;

    /*
     * Lists running through all tasks using this cgroup group.
     * mg_tasks lists tasks which belong to this cset but are in the
     * process of being migrated out or in.  Protected by
     * css_set_rwsem, but, during migration, once tasks are moved to
     * mg_tasks, it can be read safely while holding cgroup_mutex.
     */
    struct list_head tasks;
    struct list_head mg_tasks;

    /*
     * List of cgrp_cset_links pointing at cgroups referenced from this
     * css_set.  Protected by css_set_lock.
     */
    struct list_head cgrp_links;

    /* the default cgroup associated with this css_set */
    struct cgroup *dfl_cgrp;

    /*
     * 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];

    /*
     * List of csets participating in the on-going migration either as
     * source or destination.  Protected by cgroup_mutex.
     */
    struct list_head mg_preload_node;
    struct list_head mg_node;

    /*
     * If this cset is acting as the source of migration the following
     * two fields are set.  mg_src_cgrp and mg_dst_cgrp are
     * respectively the source and destination cgroups of the on-going
     * migration.  mg_dst_cset is the destination cset the target tasks
     * on this cset should be migrated to.  Protected by cgroup_mutex.
     */
    struct cgroup *mg_src_cgrp;
    struct cgroup *mg_dst_cgrp;
    struct css_set *mg_dst_cset;

    /*
     * On the default hierarhcy, ->subsys[ssid] may point to a css
     * attached to an ancestor instead of the cgroup this css_set is
     * associated with.  The following node is anchored at
     * ->subsys[ssid]->cgroup->e_csets[ssid] and provides a way to
     * iterate through all css's attached to a given cgroup.
     */
    struct list_head e_cset_node[CGROUP_SUBSYS_COUNT];

    /* all css_task_iters currently walking this cset */
    struct list_head task_iters;

    /* dead and being drained, ignore for migration */
    bool dead;

    /* For RCU-protected deletion */
    struct rcu_head rcu_head;
};

struct cgroup {
    /* self css with NULL ->ss, points back to this cgroup */
    struct cgroup_subsys_state self;

    unsigned long flags;        /* "unsigned long" so bitops work */

    /*
     * idr allocated in-hierarchy ID.
     *
     * ID 0 is not used, the ID of the root cgroup is always 1, and a
     * new cgroup will be assigned with a smallest available ID.
     *
     * Allocating/Removing ID must be protected by cgroup_mutex.
     */
    int id;

    /*
     * The depth this cgroup is at.  The root is at depth zero and each
     * step down the hierarchy increments the level.  This along with
     * ancestor_ids[] can determine whether a given cgroup is a
     * descendant of another without traversing the hierarchy.
     */
    int level;

    /*
     * Each non-empty css_set associated with this cgroup contributes
     * one to populated_cnt.  All children with non-zero popuplated_cnt
     * of their own contribute one.  The count is zero iff there's no
     * task in this cgroup or its subtree.
     */
    int populated_cnt;

    struct kernfs_node *kn;        /* cgroup kernfs entry */
    struct cgroup_file procs_file;    /* handle for "cgroup.procs" */
    struct cgroup_file events_file;    /* handle for "cgroup.events" */

    /*
     * The bitmask of subsystems enabled on the child cgroups.
     * ->subtree_control is the one configured through
     * "cgroup.subtree_control" while ->child_ss_mask is the effective
     * one which may have more subsystems enabled.  Controller knobs
     * are made available iff it's enabled in ->subtree_control.
     */
    u16 subtree_control;
    u16 subtree_ss_mask;
    u16 old_subtree_control;
    u16 old_subtree_ss_mask;

    /* Private pointers for each registered subsystem */
    struct cgroup_subsys_state __rcu *subsys[CGROUP_SUBSYS_COUNT];

    struct cgroup_root *root;

    /*
     * List of cgrp_cset_links pointing at css_sets with tasks in this
     * cgroup.  Protected by css_set_lock.
     */
    struct list_head cset_links;

    /*
     * On the default hierarchy, a css_set for a cgroup with some
     * susbsys disabled will point to css's which are associated with
     * the closest ancestor which has the subsys enabled.  The
     * following lists all css_sets which point to this cgroup's css
     * for the given subsystem.
     */
    struct list_head e_csets[CGROUP_SUBSYS_COUNT];

    /*
     * list of pidlists, up to two for each namespace (one for procs, one
     * for tasks); created on demand.
     */
    struct list_head pidlists;
    struct mutex pidlist_mutex;

    /* used to wait for offlining of csses */
    wait_queue_head_t offline_waitq;

    /* used to schedule release agent */
    struct work_struct release_agent_work;

    /* ids of the ancestors at each level including self */
    int ancestor_ids[];
};


/*
 * Per-subsystem/per-cgroup state maintained by the system.  This is the
 * fundamental structural building block that controllers deal with.
 *
 * Fields marked with "PI:" are public and immutable and may be accessed
 * directly without synchronization.
 */
struct cgroup_subsys_state {
    /* PI: the cgroup that this css is attached to */
    struct cgroup *cgroup;

    /* PI: the cgroup subsystem that this css is attached to */
    struct cgroup_subsys *ss;

    /* reference count - access via css_[try]get() and css_put() */
    struct percpu_ref refcnt;

    /* PI: the parent css */
    struct cgroup_subsys_state *parent;

    /* siblings list anchored at the parent's ->children */
    struct list_head sibling;
    struct list_head children;

    /*
     * PI: Subsys-unique ID.  0 is unused and root is always 1.  The
     * matching css can be looked up using css_from_id().
     */
    int id;

    unsigned int flags;

    /*
     * Monotonically increasing unique serial number which defines a
     * uniform order among all csses.  It's guaranteed that all
     * ->children lists are in the ascending order of ->serial_nr and
     * used to allow interrupting and resuming iterations.
     */
    u64 serial_nr;

    /*
     * Incremented by online self and children.  Used to guarantee that
     * parents are not offlined before their children.
     */
    atomic_t online_cnt;

    /* percpu_ref killing and RCU release */
    struct rcu_head rcu_head;
    struct work_struct destroy_work;
};


/*
 * A cgroup_root represents the root of a cgroup hierarchy, and may be
 * associated with a kernfs_root to form an active hierarchy.  This is
 * internal to cgroup core.  Don't access directly from controllers.
 */
struct cgroup_root {
    struct kernfs_root *kf_root;

    /* The bitmask of subsystems attached to this hierarchy */
    unsigned int subsys_mask;

    /* Unique id for this hierarchy. */
    int hierarchy_id;

    /* The root cgroup.  Root is destroyed on its release. */
    struct cgroup cgrp;

    /* for cgrp->ancestor_ids[0] */
    int cgrp_ancestor_id_storage;

    /* Number of cgroups in the hierarchy, used only for /proc/cgroups */
    atomic_t nr_cgrps;

    /* A list running through the active hierarchies */
    struct list_head root_list;

    /* Hierarchy-specific flags */
    unsigned int flags;

    /* IDs for cgroups in this hierarchy */
    struct idr cgroup_idr;

    /* The path to use for release notifications. */
    char release_agent_path[PATH_MAX];

    /* The name for this hierarchy - may be empty */
    char name[MAX_CGROUP_ROOT_NAMELEN];
};


/*
 * Control Group subsystem type.
 * See Documentation/cgroups/cgroups.txt for details
 */
struct cgroup_subsys {
    struct cgroup_subsys_state *(*css_alloc)(struct cgroup_subsys_state *parent_css);
    int (*css_online)(struct cgroup_subsys_state *css);
    void (*css_offline)(struct cgroup_subsys_state *css);
    void (*css_released)(struct cgroup_subsys_state *css);
    void (*css_free)(struct cgroup_subsys_state *css);
    void (*css_reset)(struct cgroup_subsys_state *css);

    int (*can_attach)(struct cgroup_taskset *tset);
    void (*cancel_attach)(struct cgroup_taskset *tset);
    void (*attach)(struct cgroup_taskset *tset);
    void (*post_attach)(void);
    int (*can_fork)(struct task_struct *task);
    void (*cancel_fork)(struct task_struct *task);
    void (*fork)(struct task_struct *task);
    void (*exit)(struct task_struct *task);
    void (*free)(struct task_struct *task);
    void (*bind)(struct cgroup_subsys_state *root_css);

    bool early_init:1;
    bool implicit_on_dfl:1;

    bool broken_hierarchy:1;
    bool warned_broken_hierarchy:1;

    /* the following two fields are initialized automtically during boot */
    int id;
    const char *name;

    /* optional, initialized automatically during boot if not set */
    const char *legacy_name;

    /* link to parent, protected by cgroup_lock() */
    struct cgroup_root *root;

    /* idr for css->id */
    struct idr css_idr;

    struct list_head cfts;
    struct cftype *dfl_cftypes;    /* for the default hierarchy */
    struct cftype *legacy_cftypes;    /* for the legacy hierarchies */
    unsigned int depends_on;
};

/*
 * struct cftype: handler definitions for cgroup control files
 * When reading/writing to a file:
 *    - the cgroup to use is file->f_path.dentry->d_parent->d_fsdata
 *    - the 'cftype' of the file is file->f_path.dentry->d_fsdata
 */
struct cftype {
    char name[MAX_CFTYPE_NAME];
    unsigned long private;
    size_t max_write_len;

    /* CFTYPE_* flags */
    unsigned int flags;
    unsigned int file_offset;
    struct cgroup_subsys *ss;    /* NULL for cgroup core files */
    struct list_head node;        /* anchored at ss->cfts */
    struct kernfs_ops *kf_ops;

    u64 (*read_u64)(struct cgroup_subsys_state *css, struct cftype *cft);
    s64 (*read_s64)(struct cgroup_subsys_state *css, struct cftype *cft);

    /* generic seq_file read interface */
    int (*seq_show)(struct seq_file *sf, void *v);

    /* optional ops, implement all or none */
    void *(*seq_start)(struct seq_file *sf, loff_t *ppos);
    void *(*seq_next)(struct seq_file *sf, void *v, loff_t *ppos);
    void (*seq_stop)(struct seq_file *sf, void *v);

    int (*write_u64)(struct cgroup_subsys_state *css, struct cftype *cft,u64 val);
    int (*write_s64)(struct cgroup_subsys_state *css, struct cftype *cft, s64 val);
    ssize_t (*write)(struct kernfs_open_file *of, char *buf, size_t nbytes, loff_t off);

#ifdef CONFIG_DEBUG_LOCK_ALLOC
    struct lock_class_key    lockdep_key;
#endif
};