Back to home page

OSCL-LXR

 
 

    


0001 /* SPDX-License-Identifier: GPL-2.0 */
0002 /*
0003  * linux/cgroup-defs.h - basic definitions for cgroup
0004  *
0005  * This file provides basic type and interface.  Include this file directly
0006  * only if necessary to avoid cyclic dependencies.
0007  */
0008 #ifndef _LINUX_CGROUP_DEFS_H
0009 #define _LINUX_CGROUP_DEFS_H
0010 
0011 #include <linux/limits.h>
0012 #include <linux/list.h>
0013 #include <linux/idr.h>
0014 #include <linux/wait.h>
0015 #include <linux/mutex.h>
0016 #include <linux/rcupdate.h>
0017 #include <linux/refcount.h>
0018 #include <linux/percpu-refcount.h>
0019 #include <linux/percpu-rwsem.h>
0020 #include <linux/u64_stats_sync.h>
0021 #include <linux/workqueue.h>
0022 #include <linux/bpf-cgroup-defs.h>
0023 #include <linux/psi_types.h>
0024 
0025 #ifdef CONFIG_CGROUPS
0026 
0027 struct cgroup;
0028 struct cgroup_root;
0029 struct cgroup_subsys;
0030 struct cgroup_taskset;
0031 struct kernfs_node;
0032 struct kernfs_ops;
0033 struct kernfs_open_file;
0034 struct seq_file;
0035 struct poll_table_struct;
0036 
0037 #define MAX_CGROUP_TYPE_NAMELEN 32
0038 #define MAX_CGROUP_ROOT_NAMELEN 64
0039 #define MAX_CFTYPE_NAME     64
0040 
0041 /* define the enumeration of all cgroup subsystems */
0042 #define SUBSYS(_x) _x ## _cgrp_id,
0043 enum cgroup_subsys_id {
0044 #include <linux/cgroup_subsys.h>
0045     CGROUP_SUBSYS_COUNT,
0046 };
0047 #undef SUBSYS
0048 
0049 /* bits in struct cgroup_subsys_state flags field */
0050 enum {
0051     CSS_NO_REF  = (1 << 0), /* no reference counting for this css */
0052     CSS_ONLINE  = (1 << 1), /* between ->css_online() and ->css_offline() */
0053     CSS_RELEASED    = (1 << 2), /* refcnt reached zero, released */
0054     CSS_VISIBLE = (1 << 3), /* css is visible to userland */
0055     CSS_DYING   = (1 << 4), /* css is dying */
0056 };
0057 
0058 /* bits in struct cgroup flags field */
0059 enum {
0060     /* Control Group requires release notifications to userspace */
0061     CGRP_NOTIFY_ON_RELEASE,
0062     /*
0063      * Clone the parent's configuration when creating a new child
0064      * cpuset cgroup.  For historical reasons, this option can be
0065      * specified at mount time and thus is implemented here.
0066      */
0067     CGRP_CPUSET_CLONE_CHILDREN,
0068 
0069     /* Control group has to be frozen. */
0070     CGRP_FREEZE,
0071 
0072     /* Cgroup is frozen. */
0073     CGRP_FROZEN,
0074 
0075     /* Control group has to be killed. */
0076     CGRP_KILL,
0077 };
0078 
0079 /* cgroup_root->flags */
0080 enum {
0081     CGRP_ROOT_NOPREFIX  = (1 << 1), /* mounted subsystems have no named prefix */
0082     CGRP_ROOT_XATTR     = (1 << 2), /* supports extended attributes */
0083 
0084     /*
0085      * Consider namespaces as delegation boundaries.  If this flag is
0086      * set, controller specific interface files in a namespace root
0087      * aren't writeable from inside the namespace.
0088      */
0089     CGRP_ROOT_NS_DELEGATE   = (1 << 3),
0090 
0091     /*
0092      * Reduce latencies on dynamic cgroup modifications such as task
0093      * migrations and controller on/offs by disabling percpu operation on
0094      * cgroup_threadgroup_rwsem. This makes hot path operations such as
0095      * forks and exits into the slow path and more expensive.
0096      *
0097      * The static usage pattern of creating a cgroup, enabling controllers,
0098      * and then seeding it with CLONE_INTO_CGROUP doesn't require write
0099      * locking cgroup_threadgroup_rwsem and thus doesn't benefit from
0100      * favordynmod.
0101      */
0102     CGRP_ROOT_FAVOR_DYNMODS = (1 << 4),
0103 
0104     /*
0105      * Enable cpuset controller in v1 cgroup to use v2 behavior.
0106      */
0107     CGRP_ROOT_CPUSET_V2_MODE = (1 << 16),
0108 
0109     /*
0110      * Enable legacy local memory.events.
0111      */
0112     CGRP_ROOT_MEMORY_LOCAL_EVENTS = (1 << 17),
0113 
0114     /*
0115      * Enable recursive subtree protection
0116      */
0117     CGRP_ROOT_MEMORY_RECURSIVE_PROT = (1 << 18),
0118 };
0119 
0120 /* cftype->flags */
0121 enum {
0122     CFTYPE_ONLY_ON_ROOT = (1 << 0), /* only create on root cgrp */
0123     CFTYPE_NOT_ON_ROOT  = (1 << 1), /* don't create on root cgrp */
0124     CFTYPE_NS_DELEGATABLE   = (1 << 2), /* writeable beyond delegation boundaries */
0125 
0126     CFTYPE_NO_PREFIX    = (1 << 3), /* (DON'T USE FOR NEW FILES) no subsys prefix */
0127     CFTYPE_WORLD_WRITABLE   = (1 << 4), /* (DON'T USE FOR NEW FILES) S_IWUGO */
0128     CFTYPE_DEBUG        = (1 << 5), /* create when cgroup_debug */
0129     CFTYPE_PRESSURE     = (1 << 6), /* only if pressure feature is enabled */
0130 
0131     /* internal flags, do not use outside cgroup core proper */
0132     __CFTYPE_ONLY_ON_DFL    = (1 << 16),    /* only on default hierarchy */
0133     __CFTYPE_NOT_ON_DFL = (1 << 17),    /* not on default hierarchy */
0134 };
0135 
0136 /*
0137  * cgroup_file is the handle for a file instance created in a cgroup which
0138  * is used, for example, to generate file changed notifications.  This can
0139  * be obtained by setting cftype->file_offset.
0140  */
0141 struct cgroup_file {
0142     /* do not access any fields from outside cgroup core */
0143     struct kernfs_node *kn;
0144     unsigned long notified_at;
0145     struct timer_list notify_timer;
0146 };
0147 
0148 /*
0149  * Per-subsystem/per-cgroup state maintained by the system.  This is the
0150  * fundamental structural building block that controllers deal with.
0151  *
0152  * Fields marked with "PI:" are public and immutable and may be accessed
0153  * directly without synchronization.
0154  */
0155 struct cgroup_subsys_state {
0156     /* PI: the cgroup that this css is attached to */
0157     struct cgroup *cgroup;
0158 
0159     /* PI: the cgroup subsystem that this css is attached to */
0160     struct cgroup_subsys *ss;
0161 
0162     /* reference count - access via css_[try]get() and css_put() */
0163     struct percpu_ref refcnt;
0164 
0165     /* siblings list anchored at the parent's ->children */
0166     struct list_head sibling;
0167     struct list_head children;
0168 
0169     /* flush target list anchored at cgrp->rstat_css_list */
0170     struct list_head rstat_css_node;
0171 
0172     /*
0173      * PI: Subsys-unique ID.  0 is unused and root is always 1.  The
0174      * matching css can be looked up using css_from_id().
0175      */
0176     int id;
0177 
0178     unsigned int flags;
0179 
0180     /*
0181      * Monotonically increasing unique serial number which defines a
0182      * uniform order among all csses.  It's guaranteed that all
0183      * ->children lists are in the ascending order of ->serial_nr and
0184      * used to allow interrupting and resuming iterations.
0185      */
0186     u64 serial_nr;
0187 
0188     /*
0189      * Incremented by online self and children.  Used to guarantee that
0190      * parents are not offlined before their children.
0191      */
0192     atomic_t online_cnt;
0193 
0194     /* percpu_ref killing and RCU release */
0195     struct work_struct destroy_work;
0196     struct rcu_work destroy_rwork;
0197 
0198     /*
0199      * PI: the parent css.  Placed here for cache proximity to following
0200      * fields of the containing structure.
0201      */
0202     struct cgroup_subsys_state *parent;
0203 };
0204 
0205 /*
0206  * A css_set is a structure holding pointers to a set of
0207  * cgroup_subsys_state objects. This saves space in the task struct
0208  * object and speeds up fork()/exit(), since a single inc/dec and a
0209  * list_add()/del() can bump the reference count on the entire cgroup
0210  * set for a task.
0211  */
0212 struct css_set {
0213     /*
0214      * Set of subsystem states, one for each subsystem. This array is
0215      * immutable after creation apart from the init_css_set during
0216      * subsystem registration (at boot time).
0217      */
0218     struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
0219 
0220     /* reference count */
0221     refcount_t refcount;
0222 
0223     /*
0224      * For a domain cgroup, the following points to self.  If threaded,
0225      * to the matching cset of the nearest domain ancestor.  The
0226      * dom_cset provides access to the domain cgroup and its csses to
0227      * which domain level resource consumptions should be charged.
0228      */
0229     struct css_set *dom_cset;
0230 
0231     /* the default cgroup associated with this css_set */
0232     struct cgroup *dfl_cgrp;
0233 
0234     /* internal task count, protected by css_set_lock */
0235     int nr_tasks;
0236 
0237     /*
0238      * Lists running through all tasks using this cgroup group.
0239      * mg_tasks lists tasks which belong to this cset but are in the
0240      * process of being migrated out or in.  Protected by
0241      * css_set_rwsem, but, during migration, once tasks are moved to
0242      * mg_tasks, it can be read safely while holding cgroup_mutex.
0243      */
0244     struct list_head tasks;
0245     struct list_head mg_tasks;
0246     struct list_head dying_tasks;
0247 
0248     /* all css_task_iters currently walking this cset */
0249     struct list_head task_iters;
0250 
0251     /*
0252      * On the default hierarchy, ->subsys[ssid] may point to a css
0253      * attached to an ancestor instead of the cgroup this css_set is
0254      * associated with.  The following node is anchored at
0255      * ->subsys[ssid]->cgroup->e_csets[ssid] and provides a way to
0256      * iterate through all css's attached to a given cgroup.
0257      */
0258     struct list_head e_cset_node[CGROUP_SUBSYS_COUNT];
0259 
0260     /* all threaded csets whose ->dom_cset points to this cset */
0261     struct list_head threaded_csets;
0262     struct list_head threaded_csets_node;
0263 
0264     /*
0265      * List running through all cgroup groups in the same hash
0266      * slot. Protected by css_set_lock
0267      */
0268     struct hlist_node hlist;
0269 
0270     /*
0271      * List of cgrp_cset_links pointing at cgroups referenced from this
0272      * css_set.  Protected by css_set_lock.
0273      */
0274     struct list_head cgrp_links;
0275 
0276     /*
0277      * List of csets participating in the on-going migration either as
0278      * source or destination.  Protected by cgroup_mutex.
0279      */
0280     struct list_head mg_src_preload_node;
0281     struct list_head mg_dst_preload_node;
0282     struct list_head mg_node;
0283 
0284     /*
0285      * If this cset is acting as the source of migration the following
0286      * two fields are set.  mg_src_cgrp and mg_dst_cgrp are
0287      * respectively the source and destination cgroups of the on-going
0288      * migration.  mg_dst_cset is the destination cset the target tasks
0289      * on this cset should be migrated to.  Protected by cgroup_mutex.
0290      */
0291     struct cgroup *mg_src_cgrp;
0292     struct cgroup *mg_dst_cgrp;
0293     struct css_set *mg_dst_cset;
0294 
0295     /* dead and being drained, ignore for migration */
0296     bool dead;
0297 
0298     /* For RCU-protected deletion */
0299     struct rcu_head rcu_head;
0300 };
0301 
0302 struct cgroup_base_stat {
0303     struct task_cputime cputime;
0304 
0305 #ifdef CONFIG_SCHED_CORE
0306     u64 forceidle_sum;
0307 #endif
0308 };
0309 
0310 /*
0311  * rstat - cgroup scalable recursive statistics.  Accounting is done
0312  * per-cpu in cgroup_rstat_cpu which is then lazily propagated up the
0313  * hierarchy on reads.
0314  *
0315  * When a stat gets updated, the cgroup_rstat_cpu and its ancestors are
0316  * linked into the updated tree.  On the following read, propagation only
0317  * considers and consumes the updated tree.  This makes reading O(the
0318  * number of descendants which have been active since last read) instead of
0319  * O(the total number of descendants).
0320  *
0321  * This is important because there can be a lot of (draining) cgroups which
0322  * aren't active and stat may be read frequently.  The combination can
0323  * become very expensive.  By propagating selectively, increasing reading
0324  * frequency decreases the cost of each read.
0325  *
0326  * This struct hosts both the fields which implement the above -
0327  * updated_children and updated_next - and the fields which track basic
0328  * resource statistics on top of it - bsync, bstat and last_bstat.
0329  */
0330 struct cgroup_rstat_cpu {
0331     /*
0332      * ->bsync protects ->bstat.  These are the only fields which get
0333      * updated in the hot path.
0334      */
0335     struct u64_stats_sync bsync;
0336     struct cgroup_base_stat bstat;
0337 
0338     /*
0339      * Snapshots at the last reading.  These are used to calculate the
0340      * deltas to propagate to the global counters.
0341      */
0342     struct cgroup_base_stat last_bstat;
0343 
0344     /*
0345      * Child cgroups with stat updates on this cpu since the last read
0346      * are linked on the parent's ->updated_children through
0347      * ->updated_next.
0348      *
0349      * In addition to being more compact, singly-linked list pointing
0350      * to the cgroup makes it unnecessary for each per-cpu struct to
0351      * point back to the associated cgroup.
0352      *
0353      * Protected by per-cpu cgroup_rstat_cpu_lock.
0354      */
0355     struct cgroup *updated_children;    /* terminated by self cgroup */
0356     struct cgroup *updated_next;        /* NULL iff not on the list */
0357 };
0358 
0359 struct cgroup_freezer_state {
0360     /* Should the cgroup and its descendants be frozen. */
0361     bool freeze;
0362 
0363     /* Should the cgroup actually be frozen? */
0364     int e_freeze;
0365 
0366     /* Fields below are protected by css_set_lock */
0367 
0368     /* Number of frozen descendant cgroups */
0369     int nr_frozen_descendants;
0370 
0371     /*
0372      * Number of tasks, which are counted as frozen:
0373      * frozen, SIGSTOPped, and PTRACEd.
0374      */
0375     int nr_frozen_tasks;
0376 };
0377 
0378 struct cgroup {
0379     /* self css with NULL ->ss, points back to this cgroup */
0380     struct cgroup_subsys_state self;
0381 
0382     unsigned long flags;        /* "unsigned long" so bitops work */
0383 
0384     /*
0385      * The depth this cgroup is at.  The root is at depth zero and each
0386      * step down the hierarchy increments the level.  This along with
0387      * ancestor_ids[] can determine whether a given cgroup is a
0388      * descendant of another without traversing the hierarchy.
0389      */
0390     int level;
0391 
0392     /* Maximum allowed descent tree depth */
0393     int max_depth;
0394 
0395     /*
0396      * Keep track of total numbers of visible and dying descent cgroups.
0397      * Dying cgroups are cgroups which were deleted by a user,
0398      * but are still existing because someone else is holding a reference.
0399      * max_descendants is a maximum allowed number of descent cgroups.
0400      *
0401      * nr_descendants and nr_dying_descendants are protected
0402      * by cgroup_mutex and css_set_lock. It's fine to read them holding
0403      * any of cgroup_mutex and css_set_lock; for writing both locks
0404      * should be held.
0405      */
0406     int nr_descendants;
0407     int nr_dying_descendants;
0408     int max_descendants;
0409 
0410     /*
0411      * Each non-empty css_set associated with this cgroup contributes
0412      * one to nr_populated_csets.  The counter is zero iff this cgroup
0413      * doesn't have any tasks.
0414      *
0415      * All children which have non-zero nr_populated_csets and/or
0416      * nr_populated_children of their own contribute one to either
0417      * nr_populated_domain_children or nr_populated_threaded_children
0418      * depending on their type.  Each counter is zero iff all cgroups
0419      * of the type in the subtree proper don't have any tasks.
0420      */
0421     int nr_populated_csets;
0422     int nr_populated_domain_children;
0423     int nr_populated_threaded_children;
0424 
0425     int nr_threaded_children;   /* # of live threaded child cgroups */
0426 
0427     struct kernfs_node *kn;     /* cgroup kernfs entry */
0428     struct cgroup_file procs_file;  /* handle for "cgroup.procs" */
0429     struct cgroup_file events_file; /* handle for "cgroup.events" */
0430 
0431     /*
0432      * The bitmask of subsystems enabled on the child cgroups.
0433      * ->subtree_control is the one configured through
0434      * "cgroup.subtree_control" while ->subtree_ss_mask is the effective
0435      * one which may have more subsystems enabled.  Controller knobs
0436      * are made available iff it's enabled in ->subtree_control.
0437      */
0438     u16 subtree_control;
0439     u16 subtree_ss_mask;
0440     u16 old_subtree_control;
0441     u16 old_subtree_ss_mask;
0442 
0443     /* Private pointers for each registered subsystem */
0444     struct cgroup_subsys_state __rcu *subsys[CGROUP_SUBSYS_COUNT];
0445 
0446     struct cgroup_root *root;
0447 
0448     /*
0449      * List of cgrp_cset_links pointing at css_sets with tasks in this
0450      * cgroup.  Protected by css_set_lock.
0451      */
0452     struct list_head cset_links;
0453 
0454     /*
0455      * On the default hierarchy, a css_set for a cgroup with some
0456      * susbsys disabled will point to css's which are associated with
0457      * the closest ancestor which has the subsys enabled.  The
0458      * following lists all css_sets which point to this cgroup's css
0459      * for the given subsystem.
0460      */
0461     struct list_head e_csets[CGROUP_SUBSYS_COUNT];
0462 
0463     /*
0464      * If !threaded, self.  If threaded, it points to the nearest
0465      * domain ancestor.  Inside a threaded subtree, cgroups are exempt
0466      * from process granularity and no-internal-task constraint.
0467      * Domain level resource consumptions which aren't tied to a
0468      * specific task are charged to the dom_cgrp.
0469      */
0470     struct cgroup *dom_cgrp;
0471     struct cgroup *old_dom_cgrp;        /* used while enabling threaded */
0472 
0473     /* per-cpu recursive resource statistics */
0474     struct cgroup_rstat_cpu __percpu *rstat_cpu;
0475     struct list_head rstat_css_list;
0476 
0477     /* cgroup basic resource statistics */
0478     struct cgroup_base_stat last_bstat;
0479     struct cgroup_base_stat bstat;
0480     struct prev_cputime prev_cputime;   /* for printing out cputime */
0481 
0482     /*
0483      * list of pidlists, up to two for each namespace (one for procs, one
0484      * for tasks); created on demand.
0485      */
0486     struct list_head pidlists;
0487     struct mutex pidlist_mutex;
0488 
0489     /* used to wait for offlining of csses */
0490     wait_queue_head_t offline_waitq;
0491 
0492     /* used to schedule release agent */
0493     struct work_struct release_agent_work;
0494 
0495     /* used to track pressure stalls */
0496     struct psi_group *psi;
0497 
0498     /* used to store eBPF programs */
0499     struct cgroup_bpf bpf;
0500 
0501     /* If there is block congestion on this cgroup. */
0502     atomic_t congestion_count;
0503 
0504     /* Used to store internal freezer state */
0505     struct cgroup_freezer_state freezer;
0506 
0507     /* ids of the ancestors at each level including self */
0508     u64 ancestor_ids[];
0509 };
0510 
0511 /*
0512  * A cgroup_root represents the root of a cgroup hierarchy, and may be
0513  * associated with a kernfs_root to form an active hierarchy.  This is
0514  * internal to cgroup core.  Don't access directly from controllers.
0515  */
0516 struct cgroup_root {
0517     struct kernfs_root *kf_root;
0518 
0519     /* The bitmask of subsystems attached to this hierarchy */
0520     unsigned int subsys_mask;
0521 
0522     /* Unique id for this hierarchy. */
0523     int hierarchy_id;
0524 
0525     /* The root cgroup.  Root is destroyed on its release. */
0526     struct cgroup cgrp;
0527 
0528     /* for cgrp->ancestor_ids[0] */
0529     u64 cgrp_ancestor_id_storage;
0530 
0531     /* Number of cgroups in the hierarchy, used only for /proc/cgroups */
0532     atomic_t nr_cgrps;
0533 
0534     /* A list running through the active hierarchies */
0535     struct list_head root_list;
0536 
0537     /* Hierarchy-specific flags */
0538     unsigned int flags;
0539 
0540     /* The path to use for release notifications. */
0541     char release_agent_path[PATH_MAX];
0542 
0543     /* The name for this hierarchy - may be empty */
0544     char name[MAX_CGROUP_ROOT_NAMELEN];
0545 };
0546 
0547 /*
0548  * struct cftype: handler definitions for cgroup control files
0549  *
0550  * When reading/writing to a file:
0551  *  - the cgroup to use is file->f_path.dentry->d_parent->d_fsdata
0552  *  - the 'cftype' of the file is file->f_path.dentry->d_fsdata
0553  */
0554 struct cftype {
0555     /*
0556      * By convention, the name should begin with the name of the
0557      * subsystem, followed by a period.  Zero length string indicates
0558      * end of cftype array.
0559      */
0560     char name[MAX_CFTYPE_NAME];
0561     unsigned long private;
0562 
0563     /*
0564      * The maximum length of string, excluding trailing nul, that can
0565      * be passed to write.  If < PAGE_SIZE-1, PAGE_SIZE-1 is assumed.
0566      */
0567     size_t max_write_len;
0568 
0569     /* CFTYPE_* flags */
0570     unsigned int flags;
0571 
0572     /*
0573      * If non-zero, should contain the offset from the start of css to
0574      * a struct cgroup_file field.  cgroup will record the handle of
0575      * the created file into it.  The recorded handle can be used as
0576      * long as the containing css remains accessible.
0577      */
0578     unsigned int file_offset;
0579 
0580     /*
0581      * Fields used for internal bookkeeping.  Initialized automatically
0582      * during registration.
0583      */
0584     struct cgroup_subsys *ss;   /* NULL for cgroup core files */
0585     struct list_head node;      /* anchored at ss->cfts */
0586     struct kernfs_ops *kf_ops;
0587 
0588     int (*open)(struct kernfs_open_file *of);
0589     void (*release)(struct kernfs_open_file *of);
0590 
0591     /*
0592      * read_u64() is a shortcut for the common case of returning a
0593      * single integer. Use it in place of read()
0594      */
0595     u64 (*read_u64)(struct cgroup_subsys_state *css, struct cftype *cft);
0596     /*
0597      * read_s64() is a signed version of read_u64()
0598      */
0599     s64 (*read_s64)(struct cgroup_subsys_state *css, struct cftype *cft);
0600 
0601     /* generic seq_file read interface */
0602     int (*seq_show)(struct seq_file *sf, void *v);
0603 
0604     /* optional ops, implement all or none */
0605     void *(*seq_start)(struct seq_file *sf, loff_t *ppos);
0606     void *(*seq_next)(struct seq_file *sf, void *v, loff_t *ppos);
0607     void (*seq_stop)(struct seq_file *sf, void *v);
0608 
0609     /*
0610      * write_u64() is a shortcut for the common case of accepting
0611      * a single integer (as parsed by simple_strtoull) from
0612      * userspace. Use in place of write(); return 0 or error.
0613      */
0614     int (*write_u64)(struct cgroup_subsys_state *css, struct cftype *cft,
0615              u64 val);
0616     /*
0617      * write_s64() is a signed version of write_u64()
0618      */
0619     int (*write_s64)(struct cgroup_subsys_state *css, struct cftype *cft,
0620              s64 val);
0621 
0622     /*
0623      * write() is the generic write callback which maps directly to
0624      * kernfs write operation and overrides all other operations.
0625      * Maximum write size is determined by ->max_write_len.  Use
0626      * of_css/cft() to access the associated css and cft.
0627      */
0628     ssize_t (*write)(struct kernfs_open_file *of,
0629              char *buf, size_t nbytes, loff_t off);
0630 
0631     __poll_t (*poll)(struct kernfs_open_file *of,
0632              struct poll_table_struct *pt);
0633 
0634 #ifdef CONFIG_DEBUG_LOCK_ALLOC
0635     struct lock_class_key   lockdep_key;
0636 #endif
0637 };
0638 
0639 /*
0640  * Control Group subsystem type.
0641  * See Documentation/admin-guide/cgroup-v1/cgroups.rst for details
0642  */
0643 struct cgroup_subsys {
0644     struct cgroup_subsys_state *(*css_alloc)(struct cgroup_subsys_state *parent_css);
0645     int (*css_online)(struct cgroup_subsys_state *css);
0646     void (*css_offline)(struct cgroup_subsys_state *css);
0647     void (*css_released)(struct cgroup_subsys_state *css);
0648     void (*css_free)(struct cgroup_subsys_state *css);
0649     void (*css_reset)(struct cgroup_subsys_state *css);
0650     void (*css_rstat_flush)(struct cgroup_subsys_state *css, int cpu);
0651     int (*css_extra_stat_show)(struct seq_file *seq,
0652                    struct cgroup_subsys_state *css);
0653 
0654     int (*can_attach)(struct cgroup_taskset *tset);
0655     void (*cancel_attach)(struct cgroup_taskset *tset);
0656     void (*attach)(struct cgroup_taskset *tset);
0657     void (*post_attach)(void);
0658     int (*can_fork)(struct task_struct *task,
0659             struct css_set *cset);
0660     void (*cancel_fork)(struct task_struct *task, struct css_set *cset);
0661     void (*fork)(struct task_struct *task);
0662     void (*exit)(struct task_struct *task);
0663     void (*release)(struct task_struct *task);
0664     void (*bind)(struct cgroup_subsys_state *root_css);
0665 
0666     bool early_init:1;
0667 
0668     /*
0669      * If %true, the controller, on the default hierarchy, doesn't show
0670      * up in "cgroup.controllers" or "cgroup.subtree_control", is
0671      * implicitly enabled on all cgroups on the default hierarchy, and
0672      * bypasses the "no internal process" constraint.  This is for
0673      * utility type controllers which is transparent to userland.
0674      *
0675      * An implicit controller can be stolen from the default hierarchy
0676      * anytime and thus must be okay with offline csses from previous
0677      * hierarchies coexisting with csses for the current one.
0678      */
0679     bool implicit_on_dfl:1;
0680 
0681     /*
0682      * If %true, the controller, supports threaded mode on the default
0683      * hierarchy.  In a threaded subtree, both process granularity and
0684      * no-internal-process constraint are ignored and a threaded
0685      * controllers should be able to handle that.
0686      *
0687      * Note that as an implicit controller is automatically enabled on
0688      * all cgroups on the default hierarchy, it should also be
0689      * threaded.  implicit && !threaded is not supported.
0690      */
0691     bool threaded:1;
0692 
0693     /* the following two fields are initialized automatically during boot */
0694     int id;
0695     const char *name;
0696 
0697     /* optional, initialized automatically during boot if not set */
0698     const char *legacy_name;
0699 
0700     /* link to parent, protected by cgroup_lock() */
0701     struct cgroup_root *root;
0702 
0703     /* idr for css->id */
0704     struct idr css_idr;
0705 
0706     /*
0707      * List of cftypes.  Each entry is the first entry of an array
0708      * terminated by zero length name.
0709      */
0710     struct list_head cfts;
0711 
0712     /*
0713      * Base cftypes which are automatically registered.  The two can
0714      * point to the same array.
0715      */
0716     struct cftype *dfl_cftypes; /* for the default hierarchy */
0717     struct cftype *legacy_cftypes;  /* for the legacy hierarchies */
0718 
0719     /*
0720      * A subsystem may depend on other subsystems.  When such subsystem
0721      * is enabled on a cgroup, the depended-upon subsystems are enabled
0722      * together if available.  Subsystems enabled due to dependency are
0723      * not visible to userland until explicitly enabled.  The following
0724      * specifies the mask of subsystems that this one depends on.
0725      */
0726     unsigned int depends_on;
0727 };
0728 
0729 extern struct percpu_rw_semaphore cgroup_threadgroup_rwsem;
0730 
0731 /**
0732  * cgroup_threadgroup_change_begin - threadgroup exclusion for cgroups
0733  * @tsk: target task
0734  *
0735  * Allows cgroup operations to synchronize against threadgroup changes
0736  * using a percpu_rw_semaphore.
0737  */
0738 static inline void cgroup_threadgroup_change_begin(struct task_struct *tsk)
0739 {
0740     percpu_down_read(&cgroup_threadgroup_rwsem);
0741 }
0742 
0743 /**
0744  * cgroup_threadgroup_change_end - threadgroup exclusion for cgroups
0745  * @tsk: target task
0746  *
0747  * Counterpart of cgroup_threadcgroup_change_begin().
0748  */
0749 static inline void cgroup_threadgroup_change_end(struct task_struct *tsk)
0750 {
0751     percpu_up_read(&cgroup_threadgroup_rwsem);
0752 }
0753 
0754 #else   /* CONFIG_CGROUPS */
0755 
0756 #define CGROUP_SUBSYS_COUNT 0
0757 
0758 static inline void cgroup_threadgroup_change_begin(struct task_struct *tsk)
0759 {
0760     might_sleep();
0761 }
0762 
0763 static inline void cgroup_threadgroup_change_end(struct task_struct *tsk) {}
0764 
0765 #endif  /* CONFIG_CGROUPS */
0766 
0767 #ifdef CONFIG_SOCK_CGROUP_DATA
0768 
0769 /*
0770  * sock_cgroup_data is embedded at sock->sk_cgrp_data and contains
0771  * per-socket cgroup information except for memcg association.
0772  *
0773  * On legacy hierarchies, net_prio and net_cls controllers directly
0774  * set attributes on each sock which can then be tested by the network
0775  * layer. On the default hierarchy, each sock is associated with the
0776  * cgroup it was created in and the networking layer can match the
0777  * cgroup directly.
0778  */
0779 struct sock_cgroup_data {
0780     struct cgroup   *cgroup; /* v2 */
0781 #ifdef CONFIG_CGROUP_NET_CLASSID
0782     u32     classid; /* v1 */
0783 #endif
0784 #ifdef CONFIG_CGROUP_NET_PRIO
0785     u16     prioidx; /* v1 */
0786 #endif
0787 };
0788 
0789 static inline u16 sock_cgroup_prioidx(const struct sock_cgroup_data *skcd)
0790 {
0791 #ifdef CONFIG_CGROUP_NET_PRIO
0792     return READ_ONCE(skcd->prioidx);
0793 #else
0794     return 1;
0795 #endif
0796 }
0797 
0798 static inline u32 sock_cgroup_classid(const struct sock_cgroup_data *skcd)
0799 {
0800 #ifdef CONFIG_CGROUP_NET_CLASSID
0801     return READ_ONCE(skcd->classid);
0802 #else
0803     return 0;
0804 #endif
0805 }
0806 
0807 static inline void sock_cgroup_set_prioidx(struct sock_cgroup_data *skcd,
0808                        u16 prioidx)
0809 {
0810 #ifdef CONFIG_CGROUP_NET_PRIO
0811     WRITE_ONCE(skcd->prioidx, prioidx);
0812 #endif
0813 }
0814 
0815 static inline void sock_cgroup_set_classid(struct sock_cgroup_data *skcd,
0816                        u32 classid)
0817 {
0818 #ifdef CONFIG_CGROUP_NET_CLASSID
0819     WRITE_ONCE(skcd->classid, classid);
0820 #endif
0821 }
0822 
0823 #else   /* CONFIG_SOCK_CGROUP_DATA */
0824 
0825 struct sock_cgroup_data {
0826 };
0827 
0828 #endif  /* CONFIG_SOCK_CGROUP_DATA */
0829 
0830 #endif  /* _LINUX_CGROUP_DEFS_H */