include/linux/cgroup-defs.h

   1 /* SPDX-License-Identifier: GPL-2.0 */
   2 /*
   3  * linux/cgroup-defs.h - basic definitions for cgroup
   4  *
   5  * This file provides basic type and interface.  Include this file directly
   6  * only if necessary to avoid cyclic dependencies.
   7  */
   8 #ifndef _LINUX_CGROUP_DEFS_H
   9 #define _LINUX_CGROUP_DEFS_H
  10
  11 #include <linux/limits.h>
  12 #include <linux/list.h>
  13 #include <linux/idr.h>
  14 #include <linux/wait.h>
  15 #include <linux/mutex.h>
  16 #include <linux/rcupdate.h>
  17 #include <linux/refcount.h>
  18 #include <linux/percpu-refcount.h>
  19 #include <linux/percpu-rwsem.h>
  20 #include <linux/u64_stats_sync.h>
  21 #include <linux/workqueue.h>
  22 #include <linux/bpf-cgroup.h>
  23 #include <linux/psi_types.h>
  24
  25 #ifdef CONFIG_CGROUPS
  26
  27 struct cgroup;
  28 struct cgroup_root;
  29 struct cgroup_subsys;
  30 struct cgroup_taskset;
  31 struct kernfs_node;
  32 struct kernfs_ops;
  33 struct kernfs_open_file;
  34 struct seq_file;
  35
  36 #define MAX_CGROUP_TYPE_NAMELEN 32
  37 #define MAX_CGROUP_ROOT_NAMELEN 64
  38 #define MAX_CFTYPE_NAME         64
  39
  40 /* define the enumeration of all cgroup subsystems */
  41 #define SUBSYS(_x) _x ## _cgrp_id,
  42 enum cgroup_subsys_id {
  43 #include <linux/cgroup_subsys.h>
  44         CGROUP_SUBSYS_COUNT,
  45 };
  46 #undef SUBSYS
  47
  48 /* bits in struct cgroup_subsys_state flags field */
  49 enum {
  50         CSS_NO_REF      = (1 << 0), /* no reference counting for this css */
  51         CSS_ONLINE      = (1 << 1), /* between ->css_online() and ->css_offline() */
  52         CSS_RELEASED    = (1 << 2), /* refcnt reached zero, released */
  53         CSS_VISIBLE     = (1 << 3), /* css is visible to userland */
  54         CSS_DYING       = (1 << 4), /* css is dying */
  55 };
  56
  57 /* bits in struct cgroup flags field */
  58 enum {
  59         /* Control Group requires release notifications to userspace */
  60         CGRP_NOTIFY_ON_RELEASE,
  61         /*
  62          * Clone the parent's configuration when creating a new child
  63          * cpuset cgroup.  For historical reasons, this option can be
  64          * specified at mount time and thus is implemented here.
  65          */
  66         CGRP_CPUSET_CLONE_CHILDREN,
  67 };
  68
  69 /* cgroup_root->flags */
  70 enum {
  71         CGRP_ROOT_NOPREFIX      = (1 << 1), /* mounted subsystems have no named prefix */
  72         CGRP_ROOT_XATTR         = (1 << 2), /* supports extended attributes */
  73
  74         /*
  75          * Consider namespaces as delegation boundaries.  If this flag is
  76          * set, controller specific interface files in a namespace root
  77          * aren't writeable from inside the namespace.
  78          */
  79         CGRP_ROOT_NS_DELEGATE   = (1 << 3),
  80
  81         /*
  82          * Enable cpuset controller in v1 cgroup to use v2 behavior.
  83          */
  84         CGRP_ROOT_CPUSET_V2_MODE = (1 << 4),
  85 };
  86
  87 /* cftype->flags */
  88 enum {
  89         CFTYPE_ONLY_ON_ROOT     = (1 << 0),     /* only create on root cgrp */
  90         CFTYPE_NOT_ON_ROOT      = (1 << 1),     /* don't create on root cgrp */
  91         CFTYPE_NS_DELEGATABLE   = (1 << 2),     /* writeable beyond delegation boundaries */
  92
  93         CFTYPE_NO_PREFIX        = (1 << 3),     /* (DON'T USE FOR NEW FILES) no subsys prefix */
  94         CFTYPE_WORLD_WRITABLE   = (1 << 4),     /* (DON'T USE FOR NEW FILES) S_IWUGO */
  95
  96         /* internal flags, do not use outside cgroup core proper */
  97         __CFTYPE_ONLY_ON_DFL    = (1 << 16),    /* only on default hierarchy */
  98         __CFTYPE_NOT_ON_DFL     = (1 << 17),    /* not on default hierarchy */
  99 };
 100
 101 /*
 102  * cgroup_file is the handle for a file instance created in a cgroup which
 103  * is used, for example, to generate file changed notifications.  This can
 104  * be obtained by setting cftype->file_offset.
 105  */
 106 struct cgroup_file {
 107         /* do not access any fields from outside cgroup core */
 108         struct kernfs_node *kn;
 109         unsigned long notified_at;
 110         struct timer_list notify_timer;
 111 };
 112
 113 /*
 114  * Per-subsystem/per-cgroup state maintained by the system.  This is the
 115  * fundamental structural building block that controllers deal with.
 116  *
 117  * Fields marked with "PI:" are public and immutable and may be accessed
 118  * directly without synchronization.
 119  */
 120 struct cgroup_subsys_state {
 121         /* PI: the cgroup that this css is attached to */
 122         struct cgroup *cgroup;
 123
 124         /* PI: the cgroup subsystem that this css is attached to */
 125         struct cgroup_subsys *ss;
 126
 127         /* reference count - access via css_[try]get() and css_put() */
 128         struct percpu_ref refcnt;
 129
 130         /* siblings list anchored at the parent's ->children */
 131         struct list_head sibling;
 132         struct list_head children;
 133
 134         /* flush target list anchored at cgrp->rstat_css_list */
 135         struct list_head rstat_css_node;
 136
 137         /*
 138          * PI: Subsys-unique ID.  0 is unused and root is always 1.  The
 139          * matching css can be looked up using css_from_id().
 140          */
 141         int id;
 142
 143         unsigned int flags;
 144
 145         /*
 146          * Monotonically increasing unique serial number which defines a
 147          * uniform order among all csses.  It's guaranteed that all
 148          * ->children lists are in the ascending order of ->serial_nr and
 149          * used to allow interrupting and resuming iterations.
 150          */
 151         u64 serial_nr;
 152
 153         /*
 154          * Incremented by online self and children.  Used to guarantee that
 155          * parents are not offlined before their children.
 156          */
 157         atomic_t online_cnt;
 158
 159         /* percpu_ref killing and RCU release */
 160         struct work_struct destroy_work;
 161         struct rcu_work destroy_rwork;
 162
 163         /*
 164          * PI: the parent css.  Placed here for cache proximity to following
 165          * fields of the containing structure.
 166          */
 167         struct cgroup_subsys_state *parent;
 168 };
 169
 170 /*
 171  * A css_set is a structure holding pointers to a set of
 172  * cgroup_subsys_state objects. This saves space in the task struct
 173  * object and speeds up fork()/exit(), since a single inc/dec and a
 174  * list_add()/del() can bump the reference count on the entire cgroup
 175  * set for a task.
 176  */
 177 struct css_set {
 178         /*
 179          * Set of subsystem states, one for each subsystem. This array is
 180          * immutable after creation apart from the init_css_set during
 181          * subsystem registration (at boot time).
 182          */
 183         struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
 184
 185         /* reference count */
 186         refcount_t refcount;
 187
 188         /*
 189          * For a domain cgroup, the following points to self.  If threaded,
 190          * to the matching cset of the nearest domain ancestor.  The
 191          * dom_cset provides access to the domain cgroup and its csses to
 192          * which domain level resource consumptions should be charged.
 193          */
 194         struct css_set *dom_cset;
 195
 196         /* the default cgroup associated with this css_set */
 197         struct cgroup *dfl_cgrp;
 198
 199         /* internal task count, protected by css_set_lock */
 200         int nr_tasks;
 201
 202         /*
 203          * Lists running through all tasks using this cgroup group.
 204          * mg_tasks lists tasks which belong to this cset but are in the
 205          * process of being migrated out or in.  Protected by
 206          * css_set_rwsem, but, during migration, once tasks are moved to
 207          * mg_tasks, it can be read safely while holding cgroup_mutex.
 208          */
 209         struct list_head tasks;
 210         struct list_head mg_tasks;
 211
 212         /* all css_task_iters currently walking this cset */
 213         struct list_head task_iters;
 214
 215         /*
 216          * On the default hierarhcy, ->subsys[ssid] may point to a css
 217          * attached to an ancestor instead of the cgroup this css_set is
 218          * associated with.  The following node is anchored at
 219          * ->subsys[ssid]->cgroup->e_csets[ssid] and provides a way to
 220          * iterate through all css's attached to a given cgroup.
 221          */
 222         struct list_head e_cset_node[CGROUP_SUBSYS_COUNT];
 223
 224         /* all threaded csets whose ->dom_cset points to this cset */
 225         struct list_head threaded_csets;
 226         struct list_head threaded_csets_node;
 227
 228         /*
 229          * List running through all cgroup groups in the same hash
 230          * slot. Protected by css_set_lock
 231          */
 232         struct hlist_node hlist;
 233
 234         /*
 235          * List of cgrp_cset_links pointing at cgroups referenced from this
 236          * css_set.  Protected by css_set_lock.
 237          */
 238         struct list_head cgrp_links;
 239
 240         /*
 241          * List of csets participating in the on-going migration either as
 242          * source or destination.  Protected by cgroup_mutex.
 243          */
 244         struct list_head mg_preload_node;
 245         struct list_head mg_node;
 246
 247         /*
 248          * If this cset is acting as the source of migration the following
 249          * two fields are set.  mg_src_cgrp and mg_dst_cgrp are
 250          * respectively the source and destination cgroups of the on-going
 251          * migration.  mg_dst_cset is the destination cset the target tasks
 252          * on this cset should be migrated to.  Protected by cgroup_mutex.
 253          */
 254         struct cgroup *mg_src_cgrp;
 255         struct cgroup *mg_dst_cgrp;
 256         struct css_set *mg_dst_cset;
 257
 258         /* dead and being drained, ignore for migration */
 259         bool dead;
 260
 261         /* For RCU-protected deletion */
 262         struct rcu_head rcu_head;
 263 };
 264
 265 struct cgroup_base_stat {
 266         struct task_cputime cputime;
 267 };
 268
 269 /*
 270  * rstat - cgroup scalable recursive statistics.  Accounting is done
 271  * per-cpu in cgroup_rstat_cpu which is then lazily propagated up the
 272  * hierarchy on reads.
 273  *
 274  * When a stat gets updated, the cgroup_rstat_cpu and its ancestors are
 275  * linked into the updated tree.  On the following read, propagation only
 276  * considers and consumes the updated tree.  This makes reading O(the
 277  * number of descendants which have been active since last read) instead of
 278  * O(the total number of descendants).
 279  *
 280  * This is important because there can be a lot of (draining) cgroups which
 281  * aren't active and stat may be read frequently.  The combination can
 282  * become very expensive.  By propagating selectively, increasing reading
 283  * frequency decreases the cost of each read.
 284  *
 285  * This struct hosts both the fields which implement the above -
 286  * updated_children and updated_next - and the fields which track basic
 287  * resource statistics on top of it - bsync, bstat and last_bstat.
 288  */
 289 struct cgroup_rstat_cpu {
 290         /*
 291          * ->bsync protects ->bstat.  These are the only fields which get
 292          * updated in the hot path.
 293          */
 294         struct u64_stats_sync bsync;
 295         struct cgroup_base_stat bstat;
 296
 297         /*
 298          * Snapshots at the last reading.  These are used to calculate the
 299          * deltas to propagate to the global counters.
 300          */
 301         struct cgroup_base_stat last_bstat;
 302
 303         /*
 304          * Child cgroups with stat updates on this cpu since the last read
 305          * are linked on the parent's ->updated_children through
 306          * ->updated_next.
 307          *
 308          * In addition to being more compact, singly-linked list pointing
 309          * to the cgroup makes it unnecessary for each per-cpu struct to
 310          * point back to the associated cgroup.
 311          *
 312          * Protected by per-cpu cgroup_rstat_cpu_lock.
 313          */
 314         struct cgroup *updated_children;        /* terminated by self cgroup */
 315         struct cgroup *updated_next;            /* NULL iff not on the list */
 316 };
 317
 318 struct cgroup {
 319         /* self css with NULL ->ss, points back to this cgroup */
 320         struct cgroup_subsys_state self;
 321
 322         unsigned long flags;            /* "unsigned long" so bitops work */
 323
 324         /*
 325          * idr allocated in-hierarchy ID.
 326          *
 327          * ID 0 is not used, the ID of the root cgroup is always 1, and a
 328          * new cgroup will be assigned with a smallest available ID.
 329          *
 330          * Allocating/Removing ID must be protected by cgroup_mutex.
 331          */
 332         int id;
 333
 334         /*
 335          * The depth this cgroup is at.  The root is at depth zero and each
 336          * step down the hierarchy increments the level.  This along with
 337          * ancestor_ids[] can determine whether a given cgroup is a
 338          * descendant of another without traversing the hierarchy.
 339          */
 340         int level;
 341
 342         /* Maximum allowed descent tree depth */
 343         int max_depth;
 344
 345         /*
 346          * Keep track of total numbers of visible and dying descent cgroups.
 347          * Dying cgroups are cgroups which were deleted by a user,
 348          * but are still existing because someone else is holding a reference.
 349          * max_descendants is a maximum allowed number of descent cgroups.
 350          */
 351         int nr_descendants;
 352         int nr_dying_descendants;
 353         int max_descendants;
 354
 355         /*
 356          * Each non-empty css_set associated with this cgroup contributes
 357          * one to nr_populated_csets.  The counter is zero iff this cgroup
 358          * doesn't have any tasks.
 359          *
 360          * All children which have non-zero nr_populated_csets and/or
 361          * nr_populated_children of their own contribute one to either
 362          * nr_populated_domain_children or nr_populated_threaded_children
 363          * depending on their type.  Each counter is zero iff all cgroups
 364          * of the type in the subtree proper don't have any tasks.
 365          */
 366         int nr_populated_csets;
 367         int nr_populated_domain_children;
 368         int nr_populated_threaded_children;
 369
 370         int nr_threaded_children;       /* # of live threaded child cgroups */
 371
 372         struct kernfs_node *kn;         /* cgroup kernfs entry */
 373         struct cgroup_file procs_file;  /* handle for "cgroup.procs" */
 374         struct cgroup_file events_file; /* handle for "cgroup.events" */
 375
 376         /*
 377          * The bitmask of subsystems enabled on the child cgroups.
 378          * ->subtree_control is the one configured through
 379          * "cgroup.subtree_control" while ->child_ss_mask is the effective
 380          * one which may have more subsystems enabled.  Controller knobs
 381          * are made available iff it's enabled in ->subtree_control.
 382          */
 383         u16 subtree_control;
 384         u16 subtree_ss_mask;
 385         u16 old_subtree_control;
 386         u16 old_subtree_ss_mask;
 387
 388         /* Private pointers for each registered subsystem */
 389         struct cgroup_subsys_state __rcu *subsys[CGROUP_SUBSYS_COUNT];
 390
 391         struct cgroup_root *root;
 392
 393         /*
 394          * List of cgrp_cset_links pointing at css_sets with tasks in this
 395          * cgroup.  Protected by css_set_lock.
 396          */
 397         struct list_head cset_links;
 398
 399         /*
 400          * On the default hierarchy, a css_set for a cgroup with some
 401          * susbsys disabled will point to css's which are associated with
 402          * the closest ancestor which has the subsys enabled.  The
 403          * following lists all css_sets which point to this cgroup's css
 404          * for the given subsystem.
 405          */
 406         struct list_head e_csets[CGROUP_SUBSYS_COUNT];
 407
 408         /*
 409          * If !threaded, self.  If threaded, it points to the nearest
 410          * domain ancestor.  Inside a threaded subtree, cgroups are exempt
 411          * from process granularity and no-internal-task constraint.
 412          * Domain level resource consumptions which aren't tied to a
 413          * specific task are charged to the dom_cgrp.
 414          */
 415         struct cgroup *dom_cgrp;
 416         struct cgroup *old_dom_cgrp;            /* used while enabling threaded */
 417
 418         /* per-cpu recursive resource statistics */
 419         struct cgroup_rstat_cpu __percpu *rstat_cpu;
 420         struct list_head rstat_css_list;
 421
 422         /* cgroup basic resource statistics */
 423         struct cgroup_base_stat pending_bstat;  /* pending from children */
 424         struct cgroup_base_stat bstat;
 425         struct prev_cputime prev_cputime;       /* for printing out cputime */
 426
 427         /*
 428          * list of pidlists, up to two for each namespace (one for procs, one
 429          * for tasks); created on demand.
 430          */
 431         struct list_head pidlists;
 432         struct mutex pidlist_mutex;
 433
 434         /* used to wait for offlining of csses */
 435         wait_queue_head_t offline_waitq;
 436
 437         /* used to schedule release agent */
 438         struct work_struct release_agent_work;
 439
 440         /* used to track pressure stalls */
 441         struct psi_group psi;
 442
 443         /* used to store eBPF programs */
 444         struct cgroup_bpf bpf;
 445
 446         /* If there is block congestion on this cgroup. */
 447         atomic_t congestion_count;
 448
 449         /* ids of the ancestors at each level including self */
 450         int ancestor_ids[];
 451 };
 452
 453 /*
 454  * A cgroup_root represents the root of a cgroup hierarchy, and may be
 455  * associated with a kernfs_root to form an active hierarchy.  This is
 456  * internal to cgroup core.  Don't access directly from controllers.
 457  */
 458 struct cgroup_root {
 459         struct kernfs_root *kf_root;
 460
 461         /* The bitmask of subsystems attached to this hierarchy */
 462         unsigned int subsys_mask;
 463
 464         /* Unique id for this hierarchy. */
 465         int hierarchy_id;
 466
 467         /* The root cgroup.  Root is destroyed on its release. */
 468         struct cgroup cgrp;
 469
 470         /* for cgrp->ancestor_ids[0] */
 471         int cgrp_ancestor_id_storage;
 472
 473         /* Number of cgroups in the hierarchy, used only for /proc/cgroups */
 474         atomic_t nr_cgrps;
 475
 476         /* A list running through the active hierarchies */
 477         struct list_head root_list;
 478
 479         /* Hierarchy-specific flags */
 480         unsigned int flags;
 481
 482         /* IDs for cgroups in this hierarchy */
 483         struct idr cgroup_idr;
 484
 485         /* The path to use for release notifications. */
 486         char release_agent_path[PATH_MAX];
 487
 488         /* The name for this hierarchy - may be empty */
 489         char name[MAX_CGROUP_ROOT_NAMELEN];
 490 };
 491
 492 /*
 493  * struct cftype: handler definitions for cgroup control files
 494  *
 495  * When reading/writing to a file:
 496  *      - the cgroup to use is file->f_path.dentry->d_parent->d_fsdata
 497  *      - the 'cftype' of the file is file->f_path.dentry->d_fsdata
 498  */
 499 struct cftype {
 500         /*
 501          * By convention, the name should begin with the name of the
 502          * subsystem, followed by a period.  Zero length string indicates
 503          * end of cftype array.
 504          */
 505         char name[MAX_CFTYPE_NAME];
 506         unsigned long private;
 507
 508         /*
 509          * The maximum length of string, excluding trailing nul, that can
 510          * be passed to write.  If < PAGE_SIZE-1, PAGE_SIZE-1 is assumed.
 511          */
 512         size_t max_write_len;
 513
 514         /* CFTYPE_* flags */
 515         unsigned int flags;
 516
 517         /*
 518          * If non-zero, should contain the offset from the start of css to
 519          * a struct cgroup_file field.  cgroup will record the handle of
 520          * the created file into it.  The recorded handle can be used as
 521          * long as the containing css remains accessible.
 522          */
 523         unsigned int file_offset;
 524
 525         /*
 526          * Fields used for internal bookkeeping.  Initialized automatically
 527          * during registration.
 528          */
 529         struct cgroup_subsys *ss;       /* NULL for cgroup core files */
 530         struct list_head node;          /* anchored at ss->cfts */
 531         struct kernfs_ops *kf_ops;
 532
 533         int (*open)(struct kernfs_open_file *of);
 534         void (*release)(struct kernfs_open_file *of);
 535
 536         /*
 537          * read_u64() is a shortcut for the common case of returning a
 538          * single integer. Use it in place of read()
 539          */
 540         u64 (*read_u64)(struct cgroup_subsys_state *css, struct cftype *cft);
 541         /*
 542          * read_s64() is a signed version of read_u64()
 543          */
 544         s64 (*read_s64)(struct cgroup_subsys_state *css, struct cftype *cft);
 545
 546         /* generic seq_file read interface */
 547         int (*seq_show)(struct seq_file *sf, void *v);
 548
 549         /* optional ops, implement all or none */
 550         void *(*seq_start)(struct seq_file *sf, loff_t *ppos);
 551         void *(*seq_next)(struct seq_file *sf, void *v, loff_t *ppos);
 552         void (*seq_stop)(struct seq_file *sf, void *v);
 553
 554         /*
 555          * write_u64() is a shortcut for the common case of accepting
 556          * a single integer (as parsed by simple_strtoull) from
 557          * userspace. Use in place of write(); return 0 or error.
 558          */
 559         int (*write_u64)(struct cgroup_subsys_state *css, struct cftype *cft,
 560                          u64 val);
 561         /*
 562          * write_s64() is a signed version of write_u64()
 563          */
 564         int (*write_s64)(struct cgroup_subsys_state *css, struct cftype *cft,
 565                          s64 val);
 566
 567         /*
 568          * write() is the generic write callback which maps directly to
 569          * kernfs write operation and overrides all other operations.
 570          * Maximum write size is determined by ->max_write_len.  Use
 571          * of_css/cft() to access the associated css and cft.
 572          */
 573         ssize_t (*write)(struct kernfs_open_file *of,
 574                          char *buf, size_t nbytes, loff_t off);
 575
 576 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 577         struct lock_class_key   lockdep_key;
 578 #endif
 579 };
 580
 581 /*
 582  * Control Group subsystem type.
 583  * See Documentation/cgroup-v1/cgroups.txt for details
 584  */
 585 struct cgroup_subsys {
 586         struct cgroup_subsys_state *(*css_alloc)(struct cgroup_subsys_state *parent_css);
 587         int (*css_online)(struct cgroup_subsys_state *css);
 588         void (*css_offline)(struct cgroup_subsys_state *css);
 589         void (*css_released)(struct cgroup_subsys_state *css);
 590         void (*css_free)(struct cgroup_subsys_state *css);
 591         void (*css_reset)(struct cgroup_subsys_state *css);
 592         void (*css_rstat_flush)(struct cgroup_subsys_state *css, int cpu);
 593         int (*css_extra_stat_show)(struct seq_file *seq,
 594                                    struct cgroup_subsys_state *css);
 595
 596         int (*can_attach)(struct cgroup_taskset *tset);
 597         void (*cancel_attach)(struct cgroup_taskset *tset);
 598         void (*attach)(struct cgroup_taskset *tset);
 599         void (*post_attach)(void);
 600         int (*can_fork)(struct task_struct *task);
 601         void (*cancel_fork)(struct task_struct *task);
 602         void (*fork)(struct task_struct *task);
 603         void (*exit)(struct task_struct *task);
 604         void (*free)(struct task_struct *task);
 605         void (*bind)(struct cgroup_subsys_state *root_css);
 606
 607         bool early_init:1;
 608
 609         /*
 610          * If %true, the controller, on the default hierarchy, doesn't show
 611          * up in "cgroup.controllers" or "cgroup.subtree_control", is
 612          * implicitly enabled on all cgroups on the default hierarchy, and
 613          * bypasses the "no internal process" constraint.  This is for
 614          * utility type controllers which is transparent to userland.
 615          *
 616          * An implicit controller can be stolen from the default hierarchy
 617          * anytime and thus must be okay with offline csses from previous
 618          * hierarchies coexisting with csses for the current one.
 619          */
 620         bool implicit_on_dfl:1;
 621
 622         /*
 623          * If %true, the controller, supports threaded mode on the default
 624          * hierarchy.  In a threaded subtree, both process granularity and
 625          * no-internal-process constraint are ignored and a threaded
 626          * controllers should be able to handle that.
 627          *
 628          * Note that as an implicit controller is automatically enabled on
 629          * all cgroups on the default hierarchy, it should also be
 630          * threaded.  implicit && !threaded is not supported.
 631          */
 632         bool threaded:1;
 633
 634         /*
 635          * If %false, this subsystem is properly hierarchical -
 636          * configuration, resource accounting and restriction on a parent
 637          * cgroup cover those of its children.  If %true, hierarchy support
 638          * is broken in some ways - some subsystems ignore hierarchy
 639          * completely while others are only implemented half-way.
 640          *
 641          * It's now disallowed to create nested cgroups if the subsystem is
 642          * broken and cgroup core will emit a warning message on such
 643          * cases.  Eventually, all subsystems will be made properly
 644          * hierarchical and this will go away.
 645          */
 646         bool broken_hierarchy:1;
 647         bool warned_broken_hierarchy:1;
 648
 649         /* the following two fields are initialized automtically during boot */
 650         int id;
 651         const char *name;
 652
 653         /* optional, initialized automatically during boot if not set */
 654         const char *legacy_name;
 655
 656         /* link to parent, protected by cgroup_lock() */
 657         struct cgroup_root *root;
 658
 659         /* idr for css->id */
 660         struct idr css_idr;
 661
 662         /*
 663          * List of cftypes.  Each entry is the first entry of an array
 664          * terminated by zero length name.
 665          */
 666         struct list_head cfts;
 667
 668         /*
 669          * Base cftypes which are automatically registered.  The two can
 670          * point to the same array.
 671          */
 672         struct cftype *dfl_cftypes;     /* for the default hierarchy */
 673         struct cftype *legacy_cftypes;  /* for the legacy hierarchies */
 674
 675         /*
 676          * A subsystem may depend on other subsystems.  When such subsystem
 677          * is enabled on a cgroup, the depended-upon subsystems are enabled
 678          * together if available.  Subsystems enabled due to dependency are
 679          * not visible to userland until explicitly enabled.  The following
 680          * specifies the mask of subsystems that this one depends on.
 681          */
 682         unsigned int depends_on;
 683 };
 684
 685 extern struct percpu_rw_semaphore cgroup_threadgroup_rwsem;
 686
 687 /**
 688  * cgroup_threadgroup_change_begin - threadgroup exclusion for cgroups
 689  * @tsk: target task
 690  *
 691  * Allows cgroup operations to synchronize against threadgroup changes
 692  * using a percpu_rw_semaphore.
 693  */
 694 static inline void cgroup_threadgroup_change_begin(struct task_struct *tsk)
 695 {
 696         percpu_down_read(&cgroup_threadgroup_rwsem);
 697 }
 698
 699 /**
 700  * cgroup_threadgroup_change_end - threadgroup exclusion for cgroups
 701  * @tsk: target task
 702  *
 703  * Counterpart of cgroup_threadcgroup_change_begin().
 704  */
 705 static inline void cgroup_threadgroup_change_end(struct task_struct *tsk)
 706 {
 707         percpu_up_read(&cgroup_threadgroup_rwsem);
 708 }
 709
 710 #else   /* CONFIG_CGROUPS */
 711
 712 #define CGROUP_SUBSYS_COUNT 0
 713
 714 static inline void cgroup_threadgroup_change_begin(struct task_struct *tsk)
 715 {
 716         might_sleep();
 717 }
 718
 719 static inline void cgroup_threadgroup_change_end(struct task_struct *tsk) {}
 720
 721 #endif  /* CONFIG_CGROUPS */
 722
 723 #ifdef CONFIG_SOCK_CGROUP_DATA
 724
 725 /*
 726  * sock_cgroup_data is embedded at sock->sk_cgrp_data and contains
 727  * per-socket cgroup information except for memcg association.
 728  *
 729  * On legacy hierarchies, net_prio and net_cls controllers directly set
 730  * attributes on each sock which can then be tested by the network layer.
 731  * On the default hierarchy, each sock is associated with the cgroup it was
 732  * created in and the networking layer can match the cgroup directly.
 733  *
 734  * To avoid carrying all three cgroup related fields separately in sock,
 735  * sock_cgroup_data overloads (prioidx, classid) and the cgroup pointer.
 736  * On boot, sock_cgroup_data records the cgroup that the sock was created
 737  * in so that cgroup2 matches can be made; however, once either net_prio or
 738  * net_cls starts being used, the area is overriden to carry prioidx and/or
 739  * classid.  The two modes are distinguished by whether the lowest bit is
 740  * set.  Clear bit indicates cgroup pointer while set bit prioidx and
 741  * classid.
 742  *
 743  * While userland may start using net_prio or net_cls at any time, once
 744  * either is used, cgroup2 matching no longer works.  There is no reason to
 745  * mix the two and this is in line with how legacy and v2 compatibility is
 746  * handled.  On mode switch, cgroup references which are already being
 747  * pointed to by socks may be leaked.  While this can be remedied by adding
 748  * synchronization around sock_cgroup_data, given that the number of leaked
 749  * cgroups is bound and highly unlikely to be high, this seems to be the
 750  * better trade-off.
 751  */
 752 struct sock_cgroup_data {
 753         union {
 754 #ifdef __LITTLE_ENDIAN
 755                 struct {
 756                         u8      is_data;
 757                         u8      padding;
 758                         u16     prioidx;
 759                         u32     classid;
 760                 } __packed;
 761 #else
 762                 struct {
 763                         u32     classid;
 764                         u16     prioidx;
 765                         u8      padding;
 766                         u8      is_data;
 767                 } __packed;
 768 #endif
 769                 u64             val;
 770         };
 771 };
 772
 773 /*
 774  * There's a theoretical window where the following accessors race with
 775  * updaters and return part of the previous pointer as the prioidx or
 776  * classid.  Such races are short-lived and the result isn't critical.
 777  */
 778 static inline u16 sock_cgroup_prioidx(const struct sock_cgroup_data *skcd)
 779 {
 780         /* fallback to 1 which is always the ID of the root cgroup */
 781         return (skcd->is_data & 1) ? skcd->prioidx : 1;
 782 }
 783
 784 static inline u32 sock_cgroup_classid(const struct sock_cgroup_data *skcd)
 785 {
 786         /* fallback to 0 which is the unconfigured default classid */
 787         return (skcd->is_data & 1) ? skcd->classid : 0;
 788 }
 789
 790 /*
 791  * If invoked concurrently, the updaters may clobber each other.  The
 792  * caller is responsible for synchronization.
 793  */
 794 static inline void sock_cgroup_set_prioidx(struct sock_cgroup_data *skcd,
 795                                            u16 prioidx)
 796 {
 797         struct sock_cgroup_data skcd_buf = {{ .val = READ_ONCE(skcd->val) }};
 798
 799         if (sock_cgroup_prioidx(&skcd_buf) == prioidx)
 800                 return;
 801
 802         if (!(skcd_buf.is_data & 1)) {
 803                 skcd_buf.val = 0;
 804                 skcd_buf.is_data = 1;
 805         }
 806
 807         skcd_buf.prioidx = prioidx;
 808         WRITE_ONCE(skcd->val, skcd_buf.val);    /* see sock_cgroup_ptr() */
 809 }
 810
 811 static inline void sock_cgroup_set_classid(struct sock_cgroup_data *skcd,
 812                                            u32 classid)
 813 {
 814         struct sock_cgroup_data skcd_buf = {{ .val = READ_ONCE(skcd->val) }};
 815
 816         if (sock_cgroup_classid(&skcd_buf) == classid)
 817                 return;
 818
 819         if (!(skcd_buf.is_data & 1)) {
 820                 skcd_buf.val = 0;
 821                 skcd_buf.is_data = 1;
 822         }
 823
 824         skcd_buf.classid = classid;
 825         WRITE_ONCE(skcd->val, skcd_buf.val);    /* see sock_cgroup_ptr() */
 826 }
 827
 828 #else   /* CONFIG_SOCK_CGROUP_DATA */
 829
 830 struct sock_cgroup_data {
 831 };
 832
 833 #endif  /* CONFIG_SOCK_CGROUP_DATA */
 834
 835 #endif  /* _LINUX_CGROUP_DEFS_H */