include/linux/cgroup.h

   1 #ifndef _LINUX_CGROUP_H
   2 #define _LINUX_CGROUP_H
   3 /*
   4  *  cgroup interface
   5  *
   6  *  Copyright (C) 2003 BULL SA
   7  *  Copyright (C) 2004-2006 Silicon Graphics, Inc.
   8  *
   9  */
  10
  11 #include <linux/sched.h>
  12 #include <linux/kref.h>
  13 #include <linux/cpumask.h>
  14 #include <linux/nodemask.h>
  15 #include <linux/rcupdate.h>
  16
  17 #ifdef CONFIG_CGROUPS
  18
  19 struct cgroupfs_root;
  20 struct cgroup_subsys;
  21 struct inode;
  22
  23 extern int cgroup_init_early(void);
  24 extern int cgroup_init(void);
  25 extern void cgroup_init_smp(void);
  26 extern void cgroup_lock(void);
  27 extern void cgroup_unlock(void);
  28 extern void cgroup_fork(struct task_struct *p);
  29 extern void cgroup_fork_callbacks(struct task_struct *p);
  30 extern void cgroup_post_fork(struct task_struct *p);
  31 extern void cgroup_exit(struct task_struct *p, int run_callbacks);
  32
  33 extern struct file_operations proc_cgroup_operations;
  34
  35 /* Define the enumeration of all cgroup subsystems */
  36 #define SUBSYS(_x) _x ## _subsys_id,
  37 enum cgroup_subsys_id {
  38 #include <linux/cgroup_subsys.h>
  39         CGROUP_SUBSYS_COUNT
  40 };
  41 #undef SUBSYS
  42
  43 /* Per-subsystem/per-cgroup state maintained by the system. */
  44 struct cgroup_subsys_state {
  45         /* The cgroup that this subsystem is attached to. Useful
  46          * for subsystems that want to know about the cgroup
  47          * hierarchy structure */
  48         struct cgroup *cgroup;
  49
  50         /* State maintained by the cgroup system to allow
  51          * subsystems to be "busy". Should be accessed via css_get()
  52          * and css_put() */
  53
  54         atomic_t refcnt;
  55
  56         unsigned long flags;
  57 };
  58
  59 /* bits in struct cgroup_subsys_state flags field */
  60 enum {
  61         CSS_ROOT, /* This CSS is the root of the subsystem */
  62 };
  63
  64 /*
  65  * Call css_get() to hold a reference on the cgroup;
  66  *
  67  */
  68
  69 static inline void css_get(struct cgroup_subsys_state *css)
  70 {
  71         /* We don't need to reference count the root state */
  72         if (!test_bit(CSS_ROOT, &css->flags))
  73                 atomic_inc(&css->refcnt);
  74 }
  75 /*
  76  * css_put() should be called to release a reference taken by
  77  * css_get()
  78  */
  79
  80 extern void __css_put(struct cgroup_subsys_state *css);
  81 static inline void css_put(struct cgroup_subsys_state *css)
  82 {
  83         if (!test_bit(CSS_ROOT, &css->flags))
  84                 __css_put(css);
  85 }
  86
  87 struct cgroup {
  88         unsigned long flags;            /* "unsigned long" so bitops work */
  89
  90         /* count users of this cgroup. >0 means busy, but doesn't
  91          * necessarily indicate the number of tasks in the
  92          * cgroup */
  93         atomic_t count;
  94
  95         /*
  96          * We link our 'sibling' struct into our parent's 'children'.
  97          * Our children link their 'sibling' into our 'children'.
  98          */
  99         struct list_head sibling;       /* my parent's children */
 100         struct list_head children;      /* my children */
 101
 102         struct cgroup *parent;  /* my parent */
 103         struct dentry *dentry;          /* cgroup fs entry */
 104
 105         /* Private pointers for each registered subsystem */
 106         struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
 107
 108         struct cgroupfs_root *root;
 109         struct cgroup *top_cgroup;
 110
 111         /*
 112          * List of cg_cgroup_links pointing at css_sets with
 113          * tasks in this cgroup. Protected by css_set_lock
 114          */
 115         struct list_head css_sets;
 116
 117         /*
 118          * Linked list running through all cgroups that can
 119          * potentially be reaped by the release agent. Protected by
 120          * release_list_lock
 121          */
 122         struct list_head release_list;
 123 };
 124
 125 /* A css_set is a structure holding pointers to a set of
 126  * cgroup_subsys_state objects. This saves space in the task struct
 127  * object and speeds up fork()/exit(), since a single inc/dec and a
 128  * list_add()/del() can bump the reference count on the entire
 129  * cgroup set for a task.
 130  */
 131
 132 struct css_set {
 133
 134         /* Reference count */
 135         struct kref ref;
 136
 137         /*
 138          * List running through all cgroup groups. Protected by
 139          * css_set_lock
 140          */
 141         struct list_head list;
 142
 143         /*
 144          * List running through all tasks using this cgroup
 145          * group. Protected by css_set_lock
 146          */
 147         struct list_head tasks;
 148
 149         /*
 150          * List of cg_cgroup_link objects on link chains from
 151          * cgroups referenced from this css_set. Protected by
 152          * css_set_lock
 153          */
 154         struct list_head cg_links;
 155
 156         /*
 157          * Set of subsystem states, one for each subsystem. This array
 158          * is immutable after creation apart from the init_css_set
 159          * during subsystem registration (at boot time).
 160          */
 161         struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
 162
 163 };
 164
 165 /* struct cftype:
 166  *
 167  * The files in the cgroup filesystem mostly have a very simple read/write
 168  * handling, some common function will take care of it. Nevertheless some cases
 169  * (read tasks) are special and therefore I define this structure for every
 170  * kind of file.
 171  *
 172  *
 173  * When reading/writing to a file:
 174  *      - the cgroup to use in file->f_dentry->d_parent->d_fsdata
 175  *      - the 'cftype' of the file is file->f_dentry->d_fsdata
 176  */
 177
 178 #define MAX_CFTYPE_NAME 64
 179 struct cftype {
 180         /* By convention, the name should begin with the name of the
 181          * subsystem, followed by a period */
 182         char name[MAX_CFTYPE_NAME];
 183         int private;
 184         int (*open) (struct inode *inode, struct file *file);
 185         ssize_t (*read) (struct cgroup *cont, struct cftype *cft,
 186                          struct file *file,
 187                          char __user *buf, size_t nbytes, loff_t *ppos);
 188         /*
 189          * read_uint() is a shortcut for the common case of returning a
 190          * single integer. Use it in place of read()
 191          */
 192         u64 (*read_uint) (struct cgroup *cont, struct cftype *cft);
 193         ssize_t (*write) (struct cgroup *cont, struct cftype *cft,
 194                           struct file *file,
 195                           const char __user *buf, size_t nbytes, loff_t *ppos);
 196
 197         /*
 198          * write_uint() is a shortcut for the common case of accepting
 199          * a single integer (as parsed by simple_strtoull) from
 200          * userspace. Use in place of write(); return 0 or error.
 201          */
 202         int (*write_uint) (struct cgroup *cont, struct cftype *cft, u64 val);
 203
 204         int (*release) (struct inode *inode, struct file *file);
 205 };
 206
 207 /* Add a new file to the given cgroup directory. Should only be
 208  * called by subsystems from within a populate() method */
 209 int cgroup_add_file(struct cgroup *cont, struct cgroup_subsys *subsys,
 210                        const struct cftype *cft);
 211
 212 /* Add a set of new files to the given cgroup directory. Should
 213  * only be called by subsystems from within a populate() method */
 214 int cgroup_add_files(struct cgroup *cont,
 215                         struct cgroup_subsys *subsys,
 216                         const struct cftype cft[],
 217                         int count);
 218
 219 int cgroup_is_removed(const struct cgroup *cont);
 220
 221 int cgroup_path(const struct cgroup *cont, char *buf, int buflen);
 222
 223 int cgroup_task_count(const struct cgroup *cont);
 224
 225 /* Return true if the cgroup is a descendant of the current cgroup */
 226 int cgroup_is_descendant(const struct cgroup *cont);
 227
 228 /* Control Group subsystem type. See Documentation/cgroups.txt for details */
 229
 230 struct cgroup_subsys {
 231         struct cgroup_subsys_state *(*create)(struct cgroup_subsys *ss,
 232                                                   struct cgroup *cont);
 233         void (*destroy)(struct cgroup_subsys *ss, struct cgroup *cont);
 234         int (*can_attach)(struct cgroup_subsys *ss,
 235                           struct cgroup *cont, struct task_struct *tsk);
 236         void (*attach)(struct cgroup_subsys *ss, struct cgroup *cont,
 237                         struct cgroup *old_cont, struct task_struct *tsk);
 238         void (*fork)(struct cgroup_subsys *ss, struct task_struct *task);
 239         void (*exit)(struct cgroup_subsys *ss, struct task_struct *task);
 240         int (*populate)(struct cgroup_subsys *ss,
 241                         struct cgroup *cont);
 242         void (*post_clone)(struct cgroup_subsys *ss, struct cgroup *cont);
 243         void (*bind)(struct cgroup_subsys *ss, struct cgroup *root);
 244         int subsys_id;
 245         int active;
 246         int early_init;
 247 #define MAX_CGROUP_TYPE_NAMELEN 32
 248         const char *name;
 249
 250         /* Protected by RCU */
 251         struct cgroupfs_root *root;
 252
 253         struct list_head sibling;
 254
 255         void *private;
 256 };
 257
 258 #define SUBSYS(_x) extern struct cgroup_subsys _x ## _subsys;
 259 #include <linux/cgroup_subsys.h>
 260 #undef SUBSYS
 261
 262 static inline struct cgroup_subsys_state *cgroup_subsys_state(
 263         struct cgroup *cont, int subsys_id)
 264 {
 265         return cont->subsys[subsys_id];
 266 }
 267
 268 static inline struct cgroup_subsys_state *task_subsys_state(
 269         struct task_struct *task, int subsys_id)
 270 {
 271         return rcu_dereference(task->cgroups->subsys[subsys_id]);
 272 }
 273
 274 static inline struct cgroup* task_cgroup(struct task_struct *task,
 275                                                int subsys_id)
 276 {
 277         return task_subsys_state(task, subsys_id)->cgroup;
 278 }
 279
 280 int cgroup_path(const struct cgroup *cont, char *buf, int buflen);
 281
 282 int cgroup_clone(struct task_struct *tsk, struct cgroup_subsys *ss);
 283
 284 /* A cgroup_iter should be treated as an opaque object */
 285 struct cgroup_iter {
 286         struct list_head *cg_link;
 287         struct list_head *task;
 288 };
 289
 290 /* To iterate across the tasks in a cgroup:
 291  *
 292  * 1) call cgroup_iter_start to intialize an iterator
 293  *
 294  * 2) call cgroup_iter_next() to retrieve member tasks until it
 295  *    returns NULL or until you want to end the iteration
 296  *
 297  * 3) call cgroup_iter_end() to destroy the iterator.
 298  */
 299 void cgroup_iter_start(struct cgroup *cont, struct cgroup_iter *it);
 300 struct task_struct *cgroup_iter_next(struct cgroup *cont,
 301                                         struct cgroup_iter *it);
 302 void cgroup_iter_end(struct cgroup *cont, struct cgroup_iter *it);
 303
 304 #else /* !CONFIG_CGROUPS */
 305
 306 static inline int cgroup_init_early(void) { return 0; }
 307 static inline int cgroup_init(void) { return 0; }
 308 static inline void cgroup_init_smp(void) {}
 309 static inline void cgroup_fork(struct task_struct *p) {}
 310 static inline void cgroup_fork_callbacks(struct task_struct *p) {}
 311 static inline void cgroup_post_fork(struct task_struct *p) {}
 312 static inline void cgroup_exit(struct task_struct *p, int callbacks) {}
 313
 314 static inline void cgroup_lock(void) {}
 315 static inline void cgroup_unlock(void) {}
 316
 317 #endif /* !CONFIG_CGROUPS */
 318
 319 #endif /* _LINUX_CGROUP_H */