linux/kernel/cgroup.c
<<
>>
Prefs
   1/*
   2 *  Generic process-grouping system.
   3 *
   4 *  Based originally on the cpuset system, extracted by Paul Menage
   5 *  Copyright (C) 2006 Google, Inc
   6 *
   7 *  Notifications support
   8 *  Copyright (C) 2009 Nokia Corporation
   9 *  Author: Kirill A. Shutemov
  10 *
  11 *  Copyright notices from the original cpuset code:
  12 *  --------------------------------------------------
  13 *  Copyright (C) 2003 BULL SA.
  14 *  Copyright (C) 2004-2006 Silicon Graphics, Inc.
  15 *
  16 *  Portions derived from Patrick Mochel's sysfs code.
  17 *  sysfs is Copyright (c) 2001-3 Patrick Mochel
  18 *
  19 *  2003-10-10 Written by Simon Derr.
  20 *  2003-10-22 Updates by Stephen Hemminger.
  21 *  2004 May-July Rework by Paul Jackson.
  22 *  ---------------------------------------------------
  23 *
  24 *  This file is subject to the terms and conditions of the GNU General Public
  25 *  License.  See the file COPYING in the main directory of the Linux
  26 *  distribution for more details.
  27 */
  28
  29#include <linux/cgroup.h>
  30#include <linux/cred.h>
  31#include <linux/ctype.h>
  32#include <linux/errno.h>
  33#include <linux/fs.h>
  34#include <linux/init_task.h>
  35#include <linux/kernel.h>
  36#include <linux/list.h>
  37#include <linux/mm.h>
  38#include <linux/mutex.h>
  39#include <linux/mount.h>
  40#include <linux/pagemap.h>
  41#include <linux/proc_fs.h>
  42#include <linux/rcupdate.h>
  43#include <linux/sched.h>
  44#include <linux/backing-dev.h>
  45#include <linux/seq_file.h>
  46#include <linux/slab.h>
  47#include <linux/magic.h>
  48#include <linux/spinlock.h>
  49#include <linux/string.h>
  50#include <linux/sort.h>
  51#include <linux/kmod.h>
  52#include <linux/module.h>
  53#include <linux/delayacct.h>
  54#include <linux/cgroupstats.h>
  55#include <linux/hash.h>
  56#include <linux/namei.h>
  57#include <linux/pid_namespace.h>
  58#include <linux/idr.h>
  59#include <linux/vmalloc.h> /* TODO: replace with more sophisticated array */
  60#include <linux/eventfd.h>
  61#include <linux/poll.h>
  62#include <linux/flex_array.h> /* used in cgroup_attach_proc */
  63#include <linux/kthread.h>
  64
  65#include <linux/atomic.h>
  66
  67/* css deactivation bias, makes css->refcnt negative to deny new trygets */
  68#define CSS_DEACT_BIAS          INT_MIN
  69
  70/*
  71 * cgroup_mutex is the master lock.  Any modification to cgroup or its
  72 * hierarchy must be performed while holding it.
  73 *
  74 * cgroup_root_mutex nests inside cgroup_mutex and should be held to modify
  75 * cgroupfs_root of any cgroup hierarchy - subsys list, flags,
  76 * release_agent_path and so on.  Modifying requires both cgroup_mutex and
  77 * cgroup_root_mutex.  Readers can acquire either of the two.  This is to
  78 * break the following locking order cycle.
  79 *
  80 *  A. cgroup_mutex -> cred_guard_mutex -> s_type->i_mutex_key -> namespace_sem
  81 *  B. namespace_sem -> cgroup_mutex
  82 *
  83 * B happens only through cgroup_show_options() and using cgroup_root_mutex
  84 * breaks it.
  85 */
  86static DEFINE_MUTEX(cgroup_mutex);
  87static DEFINE_MUTEX(cgroup_root_mutex);
  88
  89/*
  90 * Generate an array of cgroup subsystem pointers. At boot time, this is
  91 * populated up to CGROUP_BUILTIN_SUBSYS_COUNT, and modular subsystems are
  92 * registered after that. The mutable section of this array is protected by
  93 * cgroup_mutex.
  94 */
  95#define SUBSYS(_x) &_x ## _subsys,
  96static struct cgroup_subsys *subsys[CGROUP_SUBSYS_COUNT] = {
  97#include <linux/cgroup_subsys.h>
  98};
  99
 100#define MAX_CGROUP_ROOT_NAMELEN 64
 101
 102/*
 103 * A cgroupfs_root represents the root of a cgroup hierarchy,
 104 * and may be associated with a superblock to form an active
 105 * hierarchy
 106 */
 107struct cgroupfs_root {
 108        struct super_block *sb;
 109
 110        /*
 111         * The bitmask of subsystems intended to be attached to this
 112         * hierarchy
 113         */
 114        unsigned long subsys_bits;
 115
 116        /* Unique id for this hierarchy. */
 117        int hierarchy_id;
 118
 119        /* The bitmask of subsystems currently attached to this hierarchy */
 120        unsigned long actual_subsys_bits;
 121
 122        /* A list running through the attached subsystems */
 123        struct list_head subsys_list;
 124
 125        /* The root cgroup for this hierarchy */
 126        struct cgroup top_cgroup;
 127
 128        /* Tracks how many cgroups are currently defined in hierarchy.*/
 129        int number_of_cgroups;
 130
 131        /* A list running through the active hierarchies */
 132        struct list_head root_list;
 133
 134        /* All cgroups on this root, cgroup_mutex protected */
 135        struct list_head allcg_list;
 136
 137        /* Hierarchy-specific flags */
 138        unsigned long flags;
 139
 140        /* The path to use for release notifications. */
 141        char release_agent_path[PATH_MAX];
 142
 143        /* The name for this hierarchy - may be empty */
 144        char name[MAX_CGROUP_ROOT_NAMELEN];
 145};
 146
 147/*
 148 * The "rootnode" hierarchy is the "dummy hierarchy", reserved for the
 149 * subsystems that are otherwise unattached - it never has more than a
 150 * single cgroup, and all tasks are part of that cgroup.
 151 */
 152static struct cgroupfs_root rootnode;
 153
 154/*
 155 * cgroupfs file entry, pointed to from leaf dentry->d_fsdata.
 156 */
 157struct cfent {
 158        struct list_head                node;
 159        struct dentry                   *dentry;
 160        struct cftype                   *type;
 161};
 162
 163/*
 164 * CSS ID -- ID per subsys's Cgroup Subsys State(CSS). used only when
 165 * cgroup_subsys->use_id != 0.
 166 */
 167#define CSS_ID_MAX      (65535)
 168struct css_id {
 169        /*
 170         * The css to which this ID points. This pointer is set to valid value
 171         * after cgroup is populated. If cgroup is removed, this will be NULL.
 172         * This pointer is expected to be RCU-safe because destroy()
 173         * is called after synchronize_rcu(). But for safe use, css_is_removed()
 174         * css_tryget() should be used for avoiding race.
 175         */
 176        struct cgroup_subsys_state __rcu *css;
 177        /*
 178         * ID of this css.
 179         */
 180        unsigned short id;
 181        /*
 182         * Depth in hierarchy which this ID belongs to.
 183         */
 184        unsigned short depth;
 185        /*
 186         * ID is freed by RCU. (and lookup routine is RCU safe.)
 187         */
 188        struct rcu_head rcu_head;
 189        /*
 190         * Hierarchy of CSS ID belongs to.
 191         */
 192        unsigned short stack[0]; /* Array of Length (depth+1) */
 193};
 194
 195/*
 196 * cgroup_event represents events which userspace want to receive.
 197 */
 198struct cgroup_event {
 199        /*
 200         * Cgroup which the event belongs to.
 201         */
 202        struct cgroup *cgrp;
 203        /*
 204         * Control file which the event associated.
 205         */
 206        struct cftype *cft;
 207        /*
 208         * eventfd to signal userspace about the event.
 209         */
 210        struct eventfd_ctx *eventfd;
 211        /*
 212         * Each of these stored in a list by the cgroup.
 213         */
 214        struct list_head list;
 215        /*
 216         * All fields below needed to unregister event when
 217         * userspace closes eventfd.
 218         */
 219        poll_table pt;
 220        wait_queue_head_t *wqh;
 221        wait_queue_t wait;
 222        struct work_struct remove;
 223};
 224
 225/* The list of hierarchy roots */
 226
 227static LIST_HEAD(roots);
 228static int root_count;
 229
 230static DEFINE_IDA(hierarchy_ida);
 231static int next_hierarchy_id;
 232static DEFINE_SPINLOCK(hierarchy_id_lock);
 233
 234/* dummytop is a shorthand for the dummy hierarchy's top cgroup */
 235#define dummytop (&rootnode.top_cgroup)
 236
 237/* This flag indicates whether tasks in the fork and exit paths should
 238 * check for fork/exit handlers to call. This avoids us having to do
 239 * extra work in the fork/exit path if none of the subsystems need to
 240 * be called.
 241 */
 242static int need_forkexit_callback __read_mostly;
 243
 244#ifdef CONFIG_PROVE_LOCKING
 245int cgroup_lock_is_held(void)
 246{
 247        return lockdep_is_held(&cgroup_mutex);
 248}
 249#else /* #ifdef CONFIG_PROVE_LOCKING */
 250int cgroup_lock_is_held(void)
 251{
 252        return mutex_is_locked(&cgroup_mutex);
 253}
 254#endif /* #else #ifdef CONFIG_PROVE_LOCKING */
 255
 256EXPORT_SYMBOL_GPL(cgroup_lock_is_held);
 257
 258static int css_unbias_refcnt(int refcnt)
 259{
 260        return refcnt >= 0 ? refcnt : refcnt - CSS_DEACT_BIAS;
 261}
 262
 263/* the current nr of refs, always >= 0 whether @css is deactivated or not */
 264static int css_refcnt(struct cgroup_subsys_state *css)
 265{
 266        int v = atomic_read(&css->refcnt);
 267
 268        return css_unbias_refcnt(v);
 269}
 270
 271/* convenient tests for these bits */
 272inline int cgroup_is_removed(const struct cgroup *cgrp)
 273{
 274        return test_bit(CGRP_REMOVED, &cgrp->flags);
 275}
 276
 277/* bits in struct cgroupfs_root flags field */
 278enum {
 279        ROOT_NOPREFIX, /* mounted subsystems have no named prefix */
 280};
 281
 282static int cgroup_is_releasable(const struct cgroup *cgrp)
 283{
 284        const int bits =
 285                (1 << CGRP_RELEASABLE) |
 286                (1 << CGRP_NOTIFY_ON_RELEASE);
 287        return (cgrp->flags & bits) == bits;
 288}
 289
 290static int notify_on_release(const struct cgroup *cgrp)
 291{
 292        return test_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
 293}
 294
 295static int clone_children(const struct cgroup *cgrp)
 296{
 297        return test_bit(CGRP_CLONE_CHILDREN, &cgrp->flags);
 298}
 299
 300/*
 301 * for_each_subsys() allows you to iterate on each subsystem attached to
 302 * an active hierarchy
 303 */
 304#define for_each_subsys(_root, _ss) \
 305list_for_each_entry(_ss, &_root->subsys_list, sibling)
 306
 307/* for_each_active_root() allows you to iterate across the active hierarchies */
 308#define for_each_active_root(_root) \
 309list_for_each_entry(_root, &roots, root_list)
 310
 311static inline struct cgroup *__d_cgrp(struct dentry *dentry)
 312{
 313        return dentry->d_fsdata;
 314}
 315
 316static inline struct cfent *__d_cfe(struct dentry *dentry)
 317{
 318        return dentry->d_fsdata;
 319}
 320
 321static inline struct cftype *__d_cft(struct dentry *dentry)
 322{
 323        return __d_cfe(dentry)->type;
 324}
 325
 326/* the list of cgroups eligible for automatic release. Protected by
 327 * release_list_lock */
 328static LIST_HEAD(release_list);
 329static DEFINE_RAW_SPINLOCK(release_list_lock);
 330static void cgroup_release_agent(struct work_struct *work);
 331static DECLARE_WORK(release_agent_work, cgroup_release_agent);
 332static void check_for_release(struct cgroup *cgrp);
 333
 334/* Link structure for associating css_set objects with cgroups */
 335struct cg_cgroup_link {
 336        /*
 337         * List running through cg_cgroup_links associated with a
 338         * cgroup, anchored on cgroup->css_sets
 339         */
 340        struct list_head cgrp_link_list;
 341        struct cgroup *cgrp;
 342        /*
 343         * List running through cg_cgroup_links pointing at a
 344         * single css_set object, anchored on css_set->cg_links
 345         */
 346        struct list_head cg_link_list;
 347        struct css_set *cg;
 348};
 349
 350/* The default css_set - used by init and its children prior to any
 351 * hierarchies being mounted. It contains a pointer to the root state
 352 * for each subsystem. Also used to anchor the list of css_sets. Not
 353 * reference-counted, to improve performance when child cgroups
 354 * haven't been created.
 355 */
 356
 357static struct css_set init_css_set;
 358static struct cg_cgroup_link init_css_set_link;
 359
 360static int cgroup_init_idr(struct cgroup_subsys *ss,
 361                           struct cgroup_subsys_state *css);
 362
 363/* css_set_lock protects the list of css_set objects, and the
 364 * chain of tasks off each css_set.  Nests outside task->alloc_lock
 365 * due to cgroup_iter_start() */
 366static DEFINE_RWLOCK(css_set_lock);
 367static int css_set_count;
 368
 369/*
 370 * hash table for cgroup groups. This improves the performance to find
 371 * an existing css_set. This hash doesn't (currently) take into
 372 * account cgroups in empty hierarchies.
 373 */
 374#define CSS_SET_HASH_BITS       7
 375#define CSS_SET_TABLE_SIZE      (1 << CSS_SET_HASH_BITS)
 376static struct hlist_head css_set_table[CSS_SET_TABLE_SIZE];
 377
 378static struct hlist_head *css_set_hash(struct cgroup_subsys_state *css[])
 379{
 380        int i;
 381        int index;
 382        unsigned long tmp = 0UL;
 383
 384        for (i = 0; i < CGROUP_SUBSYS_COUNT; i++)
 385                tmp += (unsigned long)css[i];
 386        tmp = (tmp >> 16) ^ tmp;
 387
 388        index = hash_long(tmp, CSS_SET_HASH_BITS);
 389
 390        return &css_set_table[index];
 391}
 392
 393/* We don't maintain the lists running through each css_set to its
 394 * task until after the first call to cgroup_iter_start(). This
 395 * reduces the fork()/exit() overhead for people who have cgroups
 396 * compiled into their kernel but not actually in use */
 397static int use_task_css_set_links __read_mostly;
 398
 399static void __put_css_set(struct css_set *cg, int taskexit)
 400{
 401        struct cg_cgroup_link *link;
 402        struct cg_cgroup_link *saved_link;
 403        /*
 404         * Ensure that the refcount doesn't hit zero while any readers
 405         * can see it. Similar to atomic_dec_and_lock(), but for an
 406         * rwlock
 407         */
 408        if (atomic_add_unless(&cg->refcount, -1, 1))
 409                return;
 410        write_lock(&css_set_lock);
 411        if (!atomic_dec_and_test(&cg->refcount)) {
 412                write_unlock(&css_set_lock);
 413                return;
 414        }
 415
 416        /* This css_set is dead. unlink it and release cgroup refcounts */
 417        hlist_del(&cg->hlist);
 418        css_set_count--;
 419
 420        list_for_each_entry_safe(link, saved_link, &cg->cg_links,
 421                                 cg_link_list) {
 422                struct cgroup *cgrp = link->cgrp;
 423                list_del(&link->cg_link_list);
 424                list_del(&link->cgrp_link_list);
 425                if (atomic_dec_and_test(&cgrp->count) &&
 426                    notify_on_release(cgrp)) {
 427                        if (taskexit)
 428                                set_bit(CGRP_RELEASABLE, &cgrp->flags);
 429                        check_for_release(cgrp);
 430                }
 431
 432                kfree(link);
 433        }
 434
 435        write_unlock(&css_set_lock);
 436        kfree_rcu(cg, rcu_head);
 437}
 438
 439/*
 440 * refcounted get/put for css_set objects
 441 */
 442static inline void get_css_set(struct css_set *cg)
 443{
 444        atomic_inc(&cg->refcount);
 445}
 446
 447static inline void put_css_set(struct css_set *cg)
 448{
 449        __put_css_set(cg, 0);
 450}
 451
 452static inline void put_css_set_taskexit(struct css_set *cg)
 453{
 454        __put_css_set(cg, 1);
 455}
 456
 457/*
 458 * compare_css_sets - helper function for find_existing_css_set().
 459 * @cg: candidate css_set being tested
 460 * @old_cg: existing css_set for a task
 461 * @new_cgrp: cgroup that's being entered by the task
 462 * @template: desired set of css pointers in css_set (pre-calculated)
 463 *
 464 * Returns true if "cg" matches "old_cg" except for the hierarchy
 465 * which "new_cgrp" belongs to, for which it should match "new_cgrp".
 466 */
 467static bool compare_css_sets(struct css_set *cg,
 468                             struct css_set *old_cg,
 469                             struct cgroup *new_cgrp,
 470                             struct cgroup_subsys_state *template[])
 471{
 472        struct list_head *l1, *l2;
 473
 474        if (memcmp(template, cg->subsys, sizeof(cg->subsys))) {
 475                /* Not all subsystems matched */
 476                return false;
 477        }
 478
 479        /*
 480         * Compare cgroup pointers in order to distinguish between
 481         * different cgroups in heirarchies with no subsystems. We
 482         * could get by with just this check alone (and skip the
 483         * memcmp above) but on most setups the memcmp check will
 484         * avoid the need for this more expensive check on almost all
 485         * candidates.
 486         */
 487
 488        l1 = &cg->cg_links;
 489        l2 = &old_cg->cg_links;
 490        while (1) {
 491                struct cg_cgroup_link *cgl1, *cgl2;
 492                struct cgroup *cg1, *cg2;
 493
 494                l1 = l1->next;
 495                l2 = l2->next;
 496                /* See if we reached the end - both lists are equal length. */
 497                if (l1 == &cg->cg_links) {
 498                        BUG_ON(l2 != &old_cg->cg_links);
 499                        break;
 500                } else {
 501                        BUG_ON(l2 == &old_cg->cg_links);
 502                }
 503                /* Locate the cgroups associated with these links. */
 504                cgl1 = list_entry(l1, struct cg_cgroup_link, cg_link_list);
 505                cgl2 = list_entry(l2, struct cg_cgroup_link, cg_link_list);
 506                cg1 = cgl1->cgrp;
 507                cg2 = cgl2->cgrp;
 508                /* Hierarchies should be linked in the same order. */
 509                BUG_ON(cg1->root != cg2->root);
 510
 511                /*
 512                 * If this hierarchy is the hierarchy of the cgroup
 513                 * that's changing, then we need to check that this
 514                 * css_set points to the new cgroup; if it's any other
 515                 * hierarchy, then this css_set should point to the
 516                 * same cgroup as the old css_set.
 517                 */
 518                if (cg1->root == new_cgrp->root) {
 519                        if (cg1 != new_cgrp)
 520                                return false;
 521                } else {
 522                        if (cg1 != cg2)
 523                                return false;
 524                }
 525        }
 526        return true;
 527}
 528
 529/*
 530 * find_existing_css_set() is a helper for
 531 * find_css_set(), and checks to see whether an existing
 532 * css_set is suitable.
 533 *
 534 * oldcg: the cgroup group that we're using before the cgroup
 535 * transition
 536 *
 537 * cgrp: the cgroup that we're moving into
 538 *
 539 * template: location in which to build the desired set of subsystem
 540 * state objects for the new cgroup group
 541 */
 542static struct css_set *find_existing_css_set(
 543        struct css_set *oldcg,
 544        struct cgroup *cgrp,
 545        struct cgroup_subsys_state *template[])
 546{
 547        int i;
 548        struct cgroupfs_root *root = cgrp->root;
 549        struct hlist_head *hhead;
 550        struct hlist_node *node;
 551        struct css_set *cg;
 552
 553        /*
 554         * Build the set of subsystem state objects that we want to see in the
 555         * new css_set. while subsystems can change globally, the entries here
 556         * won't change, so no need for locking.
 557         */
 558        for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
 559                if (root->subsys_bits & (1UL << i)) {
 560                        /* Subsystem is in this hierarchy. So we want
 561                         * the subsystem state from the new
 562                         * cgroup */
 563                        template[i] = cgrp->subsys[i];
 564                } else {
 565                        /* Subsystem is not in this hierarchy, so we
 566                         * don't want to change the subsystem state */
 567                        template[i] = oldcg->subsys[i];
 568                }
 569        }
 570
 571        hhead = css_set_hash(template);
 572        hlist_for_each_entry(cg, node, hhead, hlist) {
 573                if (!compare_css_sets(cg, oldcg, cgrp, template))
 574                        continue;
 575
 576                /* This css_set matches what we need */
 577                return cg;
 578        }
 579
 580        /* No existing cgroup group matched */
 581        return NULL;
 582}
 583
 584static void free_cg_links(struct list_head *tmp)
 585{
 586        struct cg_cgroup_link *link;
 587        struct cg_cgroup_link *saved_link;
 588
 589        list_for_each_entry_safe(link, saved_link, tmp, cgrp_link_list) {
 590                list_del(&link->cgrp_link_list);
 591                kfree(link);
 592        }
 593}
 594
 595/*
 596 * allocate_cg_links() allocates "count" cg_cgroup_link structures
 597 * and chains them on tmp through their cgrp_link_list fields. Returns 0 on
 598 * success or a negative error
 599 */
 600static int allocate_cg_links(int count, struct list_head *tmp)
 601{
 602        struct cg_cgroup_link *link;
 603        int i;
 604        INIT_LIST_HEAD(tmp);
 605        for (i = 0; i < count; i++) {
 606                link = kmalloc(sizeof(*link), GFP_KERNEL);
 607                if (!link) {
 608                        free_cg_links(tmp);
 609                        return -ENOMEM;
 610                }
 611                list_add(&link->cgrp_link_list, tmp);
 612        }
 613        return 0;
 614}
 615
 616/**
 617 * link_css_set - a helper function to link a css_set to a cgroup
 618 * @tmp_cg_links: cg_cgroup_link objects allocated by allocate_cg_links()
 619 * @cg: the css_set to be linked
 620 * @cgrp: the destination cgroup
 621 */
 622static void link_css_set(struct list_head *tmp_cg_links,
 623                         struct css_set *cg, struct cgroup *cgrp)
 624{
 625        struct cg_cgroup_link *link;
 626
 627        BUG_ON(list_empty(tmp_cg_links));
 628        link = list_first_entry(tmp_cg_links, struct cg_cgroup_link,
 629                                cgrp_link_list);
 630        link->cg = cg;
 631        link->cgrp = cgrp;
 632        atomic_inc(&cgrp->count);
 633        list_move(&link->cgrp_link_list, &cgrp->css_sets);
 634        /*
 635         * Always add links to the tail of the list so that the list
 636         * is sorted by order of hierarchy creation
 637         */
 638        list_add_tail(&link->cg_link_list, &cg->cg_links);
 639}
 640
 641/*
 642 * find_css_set() takes an existing cgroup group and a
 643 * cgroup object, and returns a css_set object that's
 644 * equivalent to the old group, but with the given cgroup
 645 * substituted into the appropriate hierarchy. Must be called with
 646 * cgroup_mutex held
 647 */
 648static struct css_set *find_css_set(
 649        struct css_set *oldcg, struct cgroup *cgrp)
 650{
 651        struct css_set *res;
 652        struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT];
 653
 654        struct list_head tmp_cg_links;
 655
 656        struct hlist_head *hhead;
 657        struct cg_cgroup_link *link;
 658
 659        /* First see if we already have a cgroup group that matches
 660         * the desired set */
 661        read_lock(&css_set_lock);
 662        res = find_existing_css_set(oldcg, cgrp, template);
 663        if (res)
 664                get_css_set(res);
 665        read_unlock(&css_set_lock);
 666
 667        if (res)
 668                return res;
 669
 670        res = kmalloc(sizeof(*res), GFP_KERNEL);
 671        if (!res)
 672                return NULL;
 673
 674        /* Allocate all the cg_cgroup_link objects that we'll need */
 675        if (allocate_cg_links(root_count, &tmp_cg_links) < 0) {
 676                kfree(res);
 677                return NULL;
 678        }
 679
 680        atomic_set(&res->refcount, 1);
 681        INIT_LIST_HEAD(&res->cg_links);
 682        INIT_LIST_HEAD(&res->tasks);
 683        INIT_HLIST_NODE(&res->hlist);
 684
 685        /* Copy the set of subsystem state objects generated in
 686         * find_existing_css_set() */
 687        memcpy(res->subsys, template, sizeof(res->subsys));
 688
 689        write_lock(&css_set_lock);
 690        /* Add reference counts and links from the new css_set. */
 691        list_for_each_entry(link, &oldcg->cg_links, cg_link_list) {
 692                struct cgroup *c = link->cgrp;
 693                if (c->root == cgrp->root)
 694                        c = cgrp;
 695                link_css_set(&tmp_cg_links, res, c);
 696        }
 697
 698        BUG_ON(!list_empty(&tmp_cg_links));
 699
 700        css_set_count++;
 701
 702        /* Add this cgroup group to the hash table */
 703        hhead = css_set_hash(res->subsys);
 704        hlist_add_head(&res->hlist, hhead);
 705
 706        write_unlock(&css_set_lock);
 707
 708        return res;
 709}
 710
 711/*
 712 * Return the cgroup for "task" from the given hierarchy. Must be
 713 * called with cgroup_mutex held.
 714 */
 715static struct cgroup *task_cgroup_from_root(struct task_struct *task,
 716                                            struct cgroupfs_root *root)
 717{
 718        struct css_set *css;
 719        struct cgroup *res = NULL;
 720
 721        BUG_ON(!mutex_is_locked(&cgroup_mutex));
 722        read_lock(&css_set_lock);
 723        /*
 724         * No need to lock the task - since we hold cgroup_mutex the
 725         * task can't change groups, so the only thing that can happen
 726         * is that it exits and its css is set back to init_css_set.
 727         */
 728        css = task->cgroups;
 729        if (css == &init_css_set) {
 730                res = &root->top_cgroup;
 731        } else {
 732                struct cg_cgroup_link *link;
 733                list_for_each_entry(link, &css->cg_links, cg_link_list) {
 734                        struct cgroup *c = link->cgrp;
 735                        if (c->root == root) {
 736                                res = c;
 737                                break;
 738                        }
 739                }
 740        }
 741        read_unlock(&css_set_lock);
 742        BUG_ON(!res);
 743        return res;
 744}
 745
 746/*
 747 * There is one global cgroup mutex. We also require taking
 748 * task_lock() when dereferencing a task's cgroup subsys pointers.
 749 * See "The task_lock() exception", at the end of this comment.
 750 *
 751 * A task must hold cgroup_mutex to modify cgroups.
 752 *
 753 * Any task can increment and decrement the count field without lock.
 754 * So in general, code holding cgroup_mutex can't rely on the count
 755 * field not changing.  However, if the count goes to zero, then only
 756 * cgroup_attach_task() can increment it again.  Because a count of zero
 757 * means that no tasks are currently attached, therefore there is no
 758 * way a task attached to that cgroup can fork (the other way to
 759 * increment the count).  So code holding cgroup_mutex can safely
 760 * assume that if the count is zero, it will stay zero. Similarly, if
 761 * a task holds cgroup_mutex on a cgroup with zero count, it
 762 * knows that the cgroup won't be removed, as cgroup_rmdir()
 763 * needs that mutex.
 764 *
 765 * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't
 766 * (usually) take cgroup_mutex.  These are the two most performance
 767 * critical pieces of code here.  The exception occurs on cgroup_exit(),
 768 * when a task in a notify_on_release cgroup exits.  Then cgroup_mutex
 769 * is taken, and if the cgroup count is zero, a usermode call made
 770 * to the release agent with the name of the cgroup (path relative to
 771 * the root of cgroup file system) as the argument.
 772 *
 773 * A cgroup can only be deleted if both its 'count' of using tasks
 774 * is zero, and its list of 'children' cgroups is empty.  Since all
 775 * tasks in the system use _some_ cgroup, and since there is always at
 776 * least one task in the system (init, pid == 1), therefore, top_cgroup
 777 * always has either children cgroups and/or using tasks.  So we don't
 778 * need a special hack to ensure that top_cgroup cannot be deleted.
 779 *
 780 *      The task_lock() exception
 781 *
 782 * The need for this exception arises from the action of
 783 * cgroup_attach_task(), which overwrites one tasks cgroup pointer with
 784 * another.  It does so using cgroup_mutex, however there are
 785 * several performance critical places that need to reference
 786 * task->cgroup without the expense of grabbing a system global
 787 * mutex.  Therefore except as noted below, when dereferencing or, as
 788 * in cgroup_attach_task(), modifying a task'ss cgroup pointer we use
 789 * task_lock(), which acts on a spinlock (task->alloc_lock) already in
 790 * the task_struct routinely used for such matters.
 791 *
 792 * P.S.  One more locking exception.  RCU is used to guard the
 793 * update of a tasks cgroup pointer by cgroup_attach_task()
 794 */
 795
 796/**
 797 * cgroup_lock - lock out any changes to cgroup structures
 798 *
 799 */
 800void cgroup_lock(void)
 801{
 802        mutex_lock(&cgroup_mutex);
 803}
 804EXPORT_SYMBOL_GPL(cgroup_lock);
 805
 806/**
 807 * cgroup_unlock - release lock on cgroup changes
 808 *
 809 * Undo the lock taken in a previous cgroup_lock() call.
 810 */
 811void cgroup_unlock(void)
 812{
 813        mutex_unlock(&cgroup_mutex);
 814}
 815EXPORT_SYMBOL_GPL(cgroup_unlock);
 816
 817/*
 818 * A couple of forward declarations required, due to cyclic reference loop:
 819 * cgroup_mkdir -> cgroup_create -> cgroup_populate_dir ->
 820 * cgroup_add_file -> cgroup_create_file -> cgroup_dir_inode_operations
 821 * -> cgroup_mkdir.
 822 */
 823
 824static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode);
 825static struct dentry *cgroup_lookup(struct inode *, struct dentry *, unsigned int);
 826static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry);
 827static int cgroup_populate_dir(struct cgroup *cgrp);
 828static const struct inode_operations cgroup_dir_inode_operations;
 829static const struct file_operations proc_cgroupstats_operations;
 830
 831static struct backing_dev_info cgroup_backing_dev_info = {
 832        .name           = "cgroup",
 833        .capabilities   = BDI_CAP_NO_ACCT_AND_WRITEBACK,
 834};
 835
 836static int alloc_css_id(struct cgroup_subsys *ss,
 837                        struct cgroup *parent, struct cgroup *child);
 838
 839static struct inode *cgroup_new_inode(umode_t mode, struct super_block *sb)
 840{
 841        struct inode *inode = new_inode(sb);
 842
 843        if (inode) {
 844                inode->i_ino = get_next_ino();
 845                inode->i_mode = mode;
 846                inode->i_uid = current_fsuid();
 847                inode->i_gid = current_fsgid();
 848                inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 849                inode->i_mapping->backing_dev_info = &cgroup_backing_dev_info;
 850        }
 851        return inode;
 852}
 853
 854/*
 855 * Call subsys's pre_destroy handler.
 856 * This is called before css refcnt check.
 857 */
 858static int cgroup_call_pre_destroy(struct cgroup *cgrp)
 859{
 860        struct cgroup_subsys *ss;
 861        int ret = 0;
 862
 863        for_each_subsys(cgrp->root, ss) {
 864                if (!ss->pre_destroy)
 865                        continue;
 866
 867                ret = ss->pre_destroy(cgrp);
 868                if (ret) {
 869                        /* ->pre_destroy() failure is being deprecated */
 870                        WARN_ON_ONCE(!ss->__DEPRECATED_clear_css_refs);
 871                        break;
 872                }
 873        }
 874
 875        return ret;
 876}
 877
 878static void cgroup_diput(struct dentry *dentry, struct inode *inode)
 879{
 880        /* is dentry a directory ? if so, kfree() associated cgroup */
 881        if (S_ISDIR(inode->i_mode)) {
 882                struct cgroup *cgrp = dentry->d_fsdata;
 883                struct cgroup_subsys *ss;
 884                BUG_ON(!(cgroup_is_removed(cgrp)));
 885                /* It's possible for external users to be holding css
 886                 * reference counts on a cgroup; css_put() needs to
 887                 * be able to access the cgroup after decrementing
 888                 * the reference count in order to know if it needs to
 889                 * queue the cgroup to be handled by the release
 890                 * agent */
 891                synchronize_rcu();
 892
 893                mutex_lock(&cgroup_mutex);
 894                /*
 895                 * Release the subsystem state objects.
 896                 */
 897                for_each_subsys(cgrp->root, ss)
 898                        ss->destroy(cgrp);
 899
 900                cgrp->root->number_of_cgroups--;
 901                mutex_unlock(&cgroup_mutex);
 902
 903                /*
 904                 * Drop the active superblock reference that we took when we
 905                 * created the cgroup
 906                 */
 907                deactivate_super(cgrp->root->sb);
 908
 909                /*
 910                 * if we're getting rid of the cgroup, refcount should ensure
 911                 * that there are no pidlists left.
 912                 */
 913                BUG_ON(!list_empty(&cgrp->pidlists));
 914
 915                kfree_rcu(cgrp, rcu_head);
 916        } else {
 917                struct cfent *cfe = __d_cfe(dentry);
 918                struct cgroup *cgrp = dentry->d_parent->d_fsdata;
 919
 920                WARN_ONCE(!list_empty(&cfe->node) &&
 921                          cgrp != &cgrp->root->top_cgroup,
 922                          "cfe still linked for %s\n", cfe->type->name);
 923                kfree(cfe);
 924        }
 925        iput(inode);
 926}
 927
 928static int cgroup_delete(const struct dentry *d)
 929{
 930        return 1;
 931}
 932
 933static void remove_dir(struct dentry *d)
 934{
 935        struct dentry *parent = dget(d->d_parent);
 936
 937        d_delete(d);
 938        simple_rmdir(parent->d_inode, d);
 939        dput(parent);
 940}
 941
 942static int cgroup_rm_file(struct cgroup *cgrp, const struct cftype *cft)
 943{
 944        struct cfent *cfe;
 945
 946        lockdep_assert_held(&cgrp->dentry->d_inode->i_mutex);
 947        lockdep_assert_held(&cgroup_mutex);
 948
 949        list_for_each_entry(cfe, &cgrp->files, node) {
 950                struct dentry *d = cfe->dentry;
 951
 952                if (cft && cfe->type != cft)
 953                        continue;
 954
 955                dget(d);
 956                d_delete(d);
 957                simple_unlink(cgrp->dentry->d_inode, d);
 958                list_del_init(&cfe->node);
 959                dput(d);
 960
 961                return 0;
 962        }
 963        return -ENOENT;
 964}
 965
 966static void cgroup_clear_directory(struct dentry *dir)
 967{
 968        struct cgroup *cgrp = __d_cgrp(dir);
 969
 970        while (!list_empty(&cgrp->files))
 971                cgroup_rm_file(cgrp, NULL);
 972}
 973
 974/*
 975 * NOTE : the dentry must have been dget()'ed
 976 */
 977static void cgroup_d_remove_dir(struct dentry *dentry)
 978{
 979        struct dentry *parent;
 980
 981        cgroup_clear_directory(dentry);
 982
 983        parent = dentry->d_parent;
 984        spin_lock(&parent->d_lock);
 985        spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
 986        list_del_init(&dentry->d_u.d_child);
 987        spin_unlock(&dentry->d_lock);
 988        spin_unlock(&parent->d_lock);
 989        remove_dir(dentry);
 990}
 991
 992/*
 993 * A queue for waiters to do rmdir() cgroup. A tasks will sleep when
 994 * cgroup->count == 0 && list_empty(&cgroup->children) && subsys has some
 995 * reference to css->refcnt. In general, this refcnt is expected to goes down
 996 * to zero, soon.
 997 *
 998 * CGRP_WAIT_ON_RMDIR flag is set under cgroup's inode->i_mutex;
 999 */
1000static DECLARE_WAIT_QUEUE_HEAD(cgroup_rmdir_waitq);
1001
1002static void cgroup_wakeup_rmdir_waiter(struct cgroup *cgrp)
1003{
1004        if (unlikely(test_and_clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags)))
1005                wake_up_all(&cgroup_rmdir_waitq);
1006}
1007
1008void cgroup_exclude_rmdir(struct cgroup_subsys_state *css)
1009{
1010        css_get(css);
1011}
1012
1013void cgroup_release_and_wakeup_rmdir(struct cgroup_subsys_state *css)
1014{
1015        cgroup_wakeup_rmdir_waiter(css->cgroup);
1016        css_put(css);
1017}
1018
1019/*
1020 * Call with cgroup_mutex held. Drops reference counts on modules, including
1021 * any duplicate ones that parse_cgroupfs_options took. If this function
1022 * returns an error, no reference counts are touched.
1023 */
1024static int rebind_subsystems(struct cgroupfs_root *root,
1025                              unsigned long final_bits)
1026{
1027        unsigned long added_bits, removed_bits;
1028        struct cgroup *cgrp = &root->top_cgroup;
1029        int i;
1030
1031        BUG_ON(!mutex_is_locked(&cgroup_mutex));
1032        BUG_ON(!mutex_is_locked(&cgroup_root_mutex));
1033
1034        removed_bits = root->actual_subsys_bits & ~final_bits;
1035        added_bits = final_bits & ~root->actual_subsys_bits;
1036        /* Check that any added subsystems are currently free */
1037        for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1038                unsigned long bit = 1UL << i;
1039                struct cgroup_subsys *ss = subsys[i];
1040                if (!(bit & added_bits))
1041                        continue;
1042                /*
1043                 * Nobody should tell us to do a subsys that doesn't exist:
1044                 * parse_cgroupfs_options should catch that case and refcounts
1045                 * ensure that subsystems won't disappear once selected.
1046                 */
1047                BUG_ON(ss == NULL);
1048                if (ss->root != &rootnode) {
1049                        /* Subsystem isn't free */
1050                        return -EBUSY;
1051                }
1052        }
1053
1054        /* Currently we don't handle adding/removing subsystems when
1055         * any child cgroups exist. This is theoretically supportable
1056         * but involves complex error handling, so it's being left until
1057         * later */
1058        if (root->number_of_cgroups > 1)
1059                return -EBUSY;
1060
1061        /* Process each subsystem */
1062        for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1063                struct cgroup_subsys *ss = subsys[i];
1064                unsigned long bit = 1UL << i;
1065                if (bit & added_bits) {
1066                        /* We're binding this subsystem to this hierarchy */
1067                        BUG_ON(ss == NULL);
1068                        BUG_ON(cgrp->subsys[i]);
1069                        BUG_ON(!dummytop->subsys[i]);
1070                        BUG_ON(dummytop->subsys[i]->cgroup != dummytop);
1071                        cgrp->subsys[i] = dummytop->subsys[i];
1072                        cgrp->subsys[i]->cgroup = cgrp;
1073                        list_move(&ss->sibling, &root->subsys_list);
1074                        ss->root = root;
1075                        if (ss->bind)
1076                                ss->bind(cgrp);
1077                        /* refcount was already taken, and we're keeping it */
1078                } else if (bit & removed_bits) {
1079                        /* We're removing this subsystem */
1080                        BUG_ON(ss == NULL);
1081                        BUG_ON(cgrp->subsys[i] != dummytop->subsys[i]);
1082                        BUG_ON(cgrp->subsys[i]->cgroup != cgrp);
1083                        if (ss->bind)
1084                                ss->bind(dummytop);
1085                        dummytop->subsys[i]->cgroup = dummytop;
1086                        cgrp->subsys[i] = NULL;
1087                        subsys[i]->root = &rootnode;
1088                        list_move(&ss->sibling, &rootnode.subsys_list);
1089                        /* subsystem is now free - drop reference on module */
1090                        module_put(ss->module);
1091                } else if (bit & final_bits) {
1092                        /* Subsystem state should already exist */
1093                        BUG_ON(ss == NULL);
1094                        BUG_ON(!cgrp->subsys[i]);
1095                        /*
1096                         * a refcount was taken, but we already had one, so
1097                         * drop the extra reference.
1098                         */
1099                        module_put(ss->module);
1100#ifdef CONFIG_MODULE_UNLOAD
1101                        BUG_ON(ss->module && !module_refcount(ss->module));
1102#endif
1103                } else {
1104                        /* Subsystem state shouldn't exist */
1105                        BUG_ON(cgrp->subsys[i]);
1106                }
1107        }
1108        root->subsys_bits = root->actual_subsys_bits = final_bits;
1109        synchronize_rcu();
1110
1111        return 0;
1112}
1113
1114static int cgroup_show_options(struct seq_file *seq, struct dentry *dentry)
1115{
1116        struct cgroupfs_root *root = dentry->d_sb->s_fs_info;
1117        struct cgroup_subsys *ss;
1118
1119        mutex_lock(&cgroup_root_mutex);
1120        for_each_subsys(root, ss)
1121                seq_printf(seq, ",%s", ss->name);
1122        if (test_bit(ROOT_NOPREFIX, &root->flags))
1123                seq_puts(seq, ",noprefix");
1124        if (strlen(root->release_agent_path))
1125                seq_printf(seq, ",release_agent=%s", root->release_agent_path);
1126        if (clone_children(&root->top_cgroup))
1127                seq_puts(seq, ",clone_children");
1128        if (strlen(root->name))
1129                seq_printf(seq, ",name=%s", root->name);
1130        mutex_unlock(&cgroup_root_mutex);
1131        return 0;
1132}
1133
1134struct cgroup_sb_opts {
1135        unsigned long subsys_bits;
1136        unsigned long flags;
1137        char *release_agent;
1138        bool clone_children;
1139        char *name;
1140        /* User explicitly requested empty subsystem */
1141        bool none;
1142
1143        struct cgroupfs_root *new_root;
1144
1145};
1146
1147/*
1148 * Convert a hierarchy specifier into a bitmask of subsystems and flags. Call
1149 * with cgroup_mutex held to protect the subsys[] array. This function takes
1150 * refcounts on subsystems to be used, unless it returns error, in which case
1151 * no refcounts are taken.
1152 */
1153static int parse_cgroupfs_options(char *data, struct cgroup_sb_opts *opts)
1154{
1155        char *token, *o = data;
1156        bool all_ss = false, one_ss = false;
1157        unsigned long mask = (unsigned long)-1;
1158        int i;
1159        bool module_pin_failed = false;
1160
1161        BUG_ON(!mutex_is_locked(&cgroup_mutex));
1162
1163#ifdef CONFIG_CPUSETS
1164        mask = ~(1UL << cpuset_subsys_id);
1165#endif
1166
1167        memset(opts, 0, sizeof(*opts));
1168
1169        while ((token = strsep(&o, ",")) != NULL) {
1170                if (!*token)
1171                        return -EINVAL;
1172                if (!strcmp(token, "none")) {
1173                        /* Explicitly have no subsystems */
1174                        opts->none = true;
1175                        continue;
1176                }
1177                if (!strcmp(token, "all")) {
1178                        /* Mutually exclusive option 'all' + subsystem name */
1179                        if (one_ss)
1180                                return -EINVAL;
1181                        all_ss = true;
1182                        continue;
1183                }
1184                if (!strcmp(token, "noprefix")) {
1185                        set_bit(ROOT_NOPREFIX, &opts->flags);
1186                        continue;
1187                }
1188                if (!strcmp(token, "clone_children")) {
1189                        opts->clone_children = true;
1190                        continue;
1191                }
1192                if (!strncmp(token, "release_agent=", 14)) {
1193                        /* Specifying two release agents is forbidden */
1194                        if (opts->release_agent)
1195                                return -EINVAL;
1196                        opts->release_agent =
1197                                kstrndup(token + 14, PATH_MAX - 1, GFP_KERNEL);
1198                        if (!opts->release_agent)
1199                                return -ENOMEM;
1200                        continue;
1201                }
1202                if (!strncmp(token, "name=", 5)) {
1203                        const char *name = token + 5;
1204                        /* Can't specify an empty name */
1205                        if (!strlen(name))
1206                                return -EINVAL;
1207                        /* Must match [\w.-]+ */
1208                        for (i = 0; i < strlen(name); i++) {
1209                                char c = name[i];
1210                                if (isalnum(c))
1211                                        continue;
1212                                if ((c == '.') || (c == '-') || (c == '_'))
1213                                        continue;
1214                                return -EINVAL;
1215                        }
1216                        /* Specifying two names is forbidden */
1217                        if (opts->name)
1218                                return -EINVAL;
1219                        opts->name = kstrndup(name,
1220                                              MAX_CGROUP_ROOT_NAMELEN - 1,
1221                                              GFP_KERNEL);
1222                        if (!opts->name)
1223                                return -ENOMEM;
1224
1225                        continue;
1226                }
1227
1228                for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1229                        struct cgroup_subsys *ss = subsys[i];
1230                        if (ss == NULL)
1231                                continue;
1232                        if (strcmp(token, ss->name))
1233                                continue;
1234                        if (ss->disabled)
1235                                continue;
1236
1237                        /* Mutually exclusive option 'all' + subsystem name */
1238                        if (all_ss)
1239                                return -EINVAL;
1240                        set_bit(i, &opts->subsys_bits);
1241                        one_ss = true;
1242
1243                        break;
1244                }
1245                if (i == CGROUP_SUBSYS_COUNT)
1246                        return -ENOENT;
1247        }
1248
1249        /*
1250         * If the 'all' option was specified select all the subsystems,
1251         * otherwise if 'none', 'name=' and a subsystem name options
1252         * were not specified, let's default to 'all'
1253         */
1254        if (all_ss || (!one_ss && !opts->none && !opts->name)) {
1255                for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1256                        struct cgroup_subsys *ss = subsys[i];
1257                        if (ss == NULL)
1258                                continue;
1259                        if (ss->disabled)
1260                                continue;
1261                        set_bit(i, &opts->subsys_bits);
1262                }
1263        }
1264
1265        /* Consistency checks */
1266
1267        /*
1268         * Option noprefix was introduced just for backward compatibility
1269         * with the old cpuset, so we allow noprefix only if mounting just
1270         * the cpuset subsystem.
1271         */
1272        if (test_bit(ROOT_NOPREFIX, &opts->flags) &&
1273            (opts->subsys_bits & mask))
1274                return -EINVAL;
1275
1276
1277        /* Can't specify "none" and some subsystems */
1278        if (opts->subsys_bits && opts->none)
1279                return -EINVAL;
1280
1281        /*
1282         * We either have to specify by name or by subsystems. (So all
1283         * empty hierarchies must have a name).
1284         */
1285        if (!opts->subsys_bits && !opts->name)
1286                return -EINVAL;
1287
1288        /*
1289         * Grab references on all the modules we'll need, so the subsystems
1290         * don't dance around before rebind_subsystems attaches them. This may
1291         * take duplicate reference counts on a subsystem that's already used,
1292         * but rebind_subsystems handles this case.
1293         */
1294        for (i = CGROUP_BUILTIN_SUBSYS_COUNT; i < CGROUP_SUBSYS_COUNT; i++) {
1295                unsigned long bit = 1UL << i;
1296
1297                if (!(bit & opts->subsys_bits))
1298                        continue;
1299                if (!try_module_get(subsys[i]->module)) {
1300                        module_pin_failed = true;
1301                        break;
1302                }
1303        }
1304        if (module_pin_failed) {
1305                /*
1306                 * oops, one of the modules was going away. this means that we
1307                 * raced with a module_delete call, and to the user this is
1308                 * essentially a "subsystem doesn't exist" case.
1309                 */
1310                for (i--; i >= CGROUP_BUILTIN_SUBSYS_COUNT; i--) {
1311                        /* drop refcounts only on the ones we took */
1312                        unsigned long bit = 1UL << i;
1313
1314                        if (!(bit & opts->subsys_bits))
1315                                continue;
1316                        module_put(subsys[i]->module);
1317                }
1318                return -ENOENT;
1319        }
1320
1321        return 0;
1322}
1323
1324static void drop_parsed_module_refcounts(unsigned long subsys_bits)
1325{
1326        int i;
1327        for (i = CGROUP_BUILTIN_SUBSYS_COUNT; i < CGROUP_SUBSYS_COUNT; i++) {
1328                unsigned long bit = 1UL << i;
1329
1330                if (!(bit & subsys_bits))
1331                        continue;
1332                module_put(subsys[i]->module);
1333        }
1334}
1335
1336static int cgroup_remount(struct super_block *sb, int *flags, char *data)
1337{
1338        int ret = 0;
1339        struct cgroupfs_root *root = sb->s_fs_info;
1340        struct cgroup *cgrp = &root->top_cgroup;
1341        struct cgroup_sb_opts opts;
1342
1343        mutex_lock(&cgrp->dentry->d_inode->i_mutex);
1344        mutex_lock(&cgroup_mutex);
1345        mutex_lock(&cgroup_root_mutex);
1346
1347        /* See what subsystems are wanted */
1348        ret = parse_cgroupfs_options(data, &opts);
1349        if (ret)
1350                goto out_unlock;
1351
1352        /* See feature-removal-schedule.txt */
1353        if (opts.subsys_bits != root->actual_subsys_bits || opts.release_agent)
1354                pr_warning("cgroup: option changes via remount are deprecated (pid=%d comm=%s)\n",
1355                           task_tgid_nr(current), current->comm);
1356
1357        /* Don't allow flags or name to change at remount */
1358        if (opts.flags != root->flags ||
1359            (opts.name && strcmp(opts.name, root->name))) {
1360                ret = -EINVAL;
1361                drop_parsed_module_refcounts(opts.subsys_bits);
1362                goto out_unlock;
1363        }
1364
1365        ret = rebind_subsystems(root, opts.subsys_bits);
1366        if (ret) {
1367                drop_parsed_module_refcounts(opts.subsys_bits);
1368                goto out_unlock;
1369        }
1370
1371        /* clear out any existing files and repopulate subsystem files */
1372        cgroup_clear_directory(cgrp->dentry);
1373        cgroup_populate_dir(cgrp);
1374
1375        if (opts.release_agent)
1376                strcpy(root->release_agent_path, opts.release_agent);
1377 out_unlock:
1378        kfree(opts.release_agent);
1379        kfree(opts.name);
1380        mutex_unlock(&cgroup_root_mutex);
1381        mutex_unlock(&cgroup_mutex);
1382        mutex_unlock(&cgrp->dentry->d_inode->i_mutex);
1383        return ret;
1384}
1385
1386static const struct super_operations cgroup_ops = {
1387        .statfs = simple_statfs,
1388        .drop_inode = generic_delete_inode,
1389        .show_options = cgroup_show_options,
1390        .remount_fs = cgroup_remount,
1391};
1392
1393static void init_cgroup_housekeeping(struct cgroup *cgrp)
1394{
1395        INIT_LIST_HEAD(&cgrp->sibling);
1396        INIT_LIST_HEAD(&cgrp->children);
1397        INIT_LIST_HEAD(&cgrp->files);
1398        INIT_LIST_HEAD(&cgrp->css_sets);
1399        INIT_LIST_HEAD(&cgrp->release_list);
1400        INIT_LIST_HEAD(&cgrp->pidlists);
1401        mutex_init(&cgrp->pidlist_mutex);
1402        INIT_LIST_HEAD(&cgrp->event_list);
1403        spin_lock_init(&cgrp->event_list_lock);
1404}
1405
1406static void init_cgroup_root(struct cgroupfs_root *root)
1407{
1408        struct cgroup *cgrp = &root->top_cgroup;
1409
1410        INIT_LIST_HEAD(&root->subsys_list);
1411        INIT_LIST_HEAD(&root->root_list);
1412        INIT_LIST_HEAD(&root->allcg_list);
1413        root->number_of_cgroups = 1;
1414        cgrp->root = root;
1415        cgrp->top_cgroup = cgrp;
1416        list_add_tail(&cgrp->allcg_node, &root->allcg_list);
1417        init_cgroup_housekeeping(cgrp);
1418}
1419
1420static bool init_root_id(struct cgroupfs_root *root)
1421{
1422        int ret = 0;
1423
1424        do {
1425                if (!ida_pre_get(&hierarchy_ida, GFP_KERNEL))
1426                        return false;
1427                spin_lock(&hierarchy_id_lock);
1428                /* Try to allocate the next unused ID */
1429                ret = ida_get_new_above(&hierarchy_ida, next_hierarchy_id,
1430                                        &root->hierarchy_id);
1431                if (ret == -ENOSPC)
1432                        /* Try again starting from 0 */
1433                        ret = ida_get_new(&hierarchy_ida, &root->hierarchy_id);
1434                if (!ret) {
1435                        next_hierarchy_id = root->hierarchy_id + 1;
1436                } else if (ret != -EAGAIN) {
1437                        /* Can only get here if the 31-bit IDR is full ... */
1438                        BUG_ON(ret);
1439                }
1440                spin_unlock(&hierarchy_id_lock);
1441        } while (ret);
1442        return true;
1443}
1444
1445static int cgroup_test_super(struct super_block *sb, void *data)
1446{
1447        struct cgroup_sb_opts *opts = data;
1448        struct cgroupfs_root *root = sb->s_fs_info;
1449
1450        /* If we asked for a name then it must match */
1451        if (opts->name && strcmp(opts->name, root->name))
1452                return 0;
1453
1454        /*
1455         * If we asked for subsystems (or explicitly for no
1456         * subsystems) then they must match
1457         */
1458        if ((opts->subsys_bits || opts->none)
1459            && (opts->subsys_bits != root->subsys_bits))
1460                return 0;
1461
1462        return 1;
1463}
1464
1465static struct cgroupfs_root *cgroup_root_from_opts(struct cgroup_sb_opts *opts)
1466{
1467        struct cgroupfs_root *root;
1468
1469        if (!opts->subsys_bits && !opts->none)
1470                return NULL;
1471
1472        root = kzalloc(sizeof(*root), GFP_KERNEL);
1473        if (!root)
1474                return ERR_PTR(-ENOMEM);
1475
1476        if (!init_root_id(root)) {
1477                kfree(root);
1478                return ERR_PTR(-ENOMEM);
1479        }
1480        init_cgroup_root(root);
1481
1482        root->subsys_bits = opts->subsys_bits;
1483        root->flags = opts->flags;
1484        if (opts->release_agent)
1485                strcpy(root->release_agent_path, opts->release_agent);
1486        if (opts->name)
1487                strcpy(root->name, opts->name);
1488        if (opts->clone_children)
1489                set_bit(CGRP_CLONE_CHILDREN, &root->top_cgroup.flags);
1490        return root;
1491}
1492
1493static void cgroup_drop_root(struct cgroupfs_root *root)
1494{
1495        if (!root)
1496                return;
1497
1498        BUG_ON(!root->hierarchy_id);
1499        spin_lock(&hierarchy_id_lock);
1500        ida_remove(&hierarchy_ida, root->hierarchy_id);
1501        spin_unlock(&hierarchy_id_lock);
1502        kfree(root);
1503}
1504
1505static int cgroup_set_super(struct super_block *sb, void *data)
1506{
1507        int ret;
1508        struct cgroup_sb_opts *opts = data;
1509
1510        /* If we don't have a new root, we can't set up a new sb */
1511        if (!opts->new_root)
1512                return -EINVAL;
1513
1514        BUG_ON(!opts->subsys_bits && !opts->none);
1515
1516        ret = set_anon_super(sb, NULL);
1517        if (ret)
1518                return ret;
1519
1520        sb->s_fs_info = opts->new_root;
1521        opts->new_root->sb = sb;
1522
1523        sb->s_blocksize = PAGE_CACHE_SIZE;
1524        sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
1525        sb->s_magic = CGROUP_SUPER_MAGIC;
1526        sb->s_op = &cgroup_ops;
1527
1528        return 0;
1529}
1530
1531static int cgroup_get_rootdir(struct super_block *sb)
1532{
1533        static const struct dentry_operations cgroup_dops = {
1534                .d_iput = cgroup_diput,
1535                .d_delete = cgroup_delete,
1536        };
1537
1538        struct inode *inode =
1539                cgroup_new_inode(S_IFDIR | S_IRUGO | S_IXUGO | S_IWUSR, sb);
1540
1541        if (!inode)
1542                return -ENOMEM;
1543
1544        inode->i_fop = &simple_dir_operations;
1545        inode->i_op = &cgroup_dir_inode_operations;
1546        /* directories start off with i_nlink == 2 (for "." entry) */
1547        inc_nlink(inode);
1548        sb->s_root = d_make_root(inode);
1549        if (!sb->s_root)
1550                return -ENOMEM;
1551        /* for everything else we want ->d_op set */
1552        sb->s_d_op = &cgroup_dops;
1553        return 0;
1554}
1555
1556static struct dentry *cgroup_mount(struct file_system_type *fs_type,
1557                         int flags, const char *unused_dev_name,
1558                         void *data)
1559{
1560        struct cgroup_sb_opts opts;
1561        struct cgroupfs_root *root;
1562        int ret = 0;
1563        struct super_block *sb;
1564        struct cgroupfs_root *new_root;
1565        struct inode *inode;
1566
1567        /* First find the desired set of subsystems */
1568        mutex_lock(&cgroup_mutex);
1569        ret = parse_cgroupfs_options(data, &opts);
1570        mutex_unlock(&cgroup_mutex);
1571        if (ret)
1572                goto out_err;
1573
1574        /*
1575         * Allocate a new cgroup root. We may not need it if we're
1576         * reusing an existing hierarchy.
1577         */
1578        new_root = cgroup_root_from_opts(&opts);
1579        if (IS_ERR(new_root)) {
1580                ret = PTR_ERR(new_root);
1581                goto drop_modules;
1582        }
1583        opts.new_root = new_root;
1584
1585        /* Locate an existing or new sb for this hierarchy */
1586        sb = sget(fs_type, cgroup_test_super, cgroup_set_super, 0, &opts);
1587        if (IS_ERR(sb)) {
1588                ret = PTR_ERR(sb);
1589                cgroup_drop_root(opts.new_root);
1590                goto drop_modules;
1591        }
1592
1593        root = sb->s_fs_info;
1594        BUG_ON(!root);
1595        if (root == opts.new_root) {
1596                /* We used the new root structure, so this is a new hierarchy */
1597                struct list_head tmp_cg_links;
1598                struct cgroup *root_cgrp = &root->top_cgroup;
1599                struct cgroupfs_root *existing_root;
1600                const struct cred *cred;
1601                int i;
1602
1603                BUG_ON(sb->s_root != NULL);
1604
1605                ret = cgroup_get_rootdir(sb);
1606                if (ret)
1607                        goto drop_new_super;
1608                inode = sb->s_root->d_inode;
1609
1610                mutex_lock(&inode->i_mutex);
1611                mutex_lock(&cgroup_mutex);
1612                mutex_lock(&cgroup_root_mutex);
1613
1614                /* Check for name clashes with existing mounts */
1615                ret = -EBUSY;
1616                if (strlen(root->name))
1617                        for_each_active_root(existing_root)
1618                                if (!strcmp(existing_root->name, root->name))
1619                                        goto unlock_drop;
1620
1621                /*
1622                 * We're accessing css_set_count without locking
1623                 * css_set_lock here, but that's OK - it can only be
1624                 * increased by someone holding cgroup_lock, and
1625                 * that's us. The worst that can happen is that we
1626                 * have some link structures left over
1627                 */
1628                ret = allocate_cg_links(css_set_count, &tmp_cg_links);
1629                if (ret)
1630                        goto unlock_drop;
1631
1632                ret = rebind_subsystems(root, root->subsys_bits);
1633                if (ret == -EBUSY) {
1634                        free_cg_links(&tmp_cg_links);
1635                        goto unlock_drop;
1636                }
1637                /*
1638                 * There must be no failure case after here, since rebinding
1639                 * takes care of subsystems' refcounts, which are explicitly
1640                 * dropped in the failure exit path.
1641                 */
1642
1643                /* EBUSY should be the only error here */
1644                BUG_ON(ret);
1645
1646                list_add(&root->root_list, &roots);
1647                root_count++;
1648
1649                sb->s_root->d_fsdata = root_cgrp;
1650                root->top_cgroup.dentry = sb->s_root;
1651
1652                /* Link the top cgroup in this hierarchy into all
1653                 * the css_set objects */
1654                write_lock(&css_set_lock);
1655                for (i = 0; i < CSS_SET_TABLE_SIZE; i++) {
1656                        struct hlist_head *hhead = &css_set_table[i];
1657                        struct hlist_node *node;
1658                        struct css_set *cg;
1659
1660                        hlist_for_each_entry(cg, node, hhead, hlist)
1661                                link_css_set(&tmp_cg_links, cg, root_cgrp);
1662                }
1663                write_unlock(&css_set_lock);
1664
1665                free_cg_links(&tmp_cg_links);
1666
1667                BUG_ON(!list_empty(&root_cgrp->sibling));
1668                BUG_ON(!list_empty(&root_cgrp->children));
1669                BUG_ON(root->number_of_cgroups != 1);
1670
1671                cred = override_creds(&init_cred);
1672                cgroup_populate_dir(root_cgrp);
1673                revert_creds(cred);
1674                mutex_unlock(&cgroup_root_mutex);
1675                mutex_unlock(&cgroup_mutex);
1676                mutex_unlock(&inode->i_mutex);
1677        } else {
1678                /*
1679                 * We re-used an existing hierarchy - the new root (if
1680                 * any) is not needed
1681                 */
1682                cgroup_drop_root(opts.new_root);
1683                /* no subsys rebinding, so refcounts don't change */
1684                drop_parsed_module_refcounts(opts.subsys_bits);
1685        }
1686
1687        kfree(opts.release_agent);
1688        kfree(opts.name);
1689        return dget(sb->s_root);
1690
1691 unlock_drop:
1692        mutex_unlock(&cgroup_root_mutex);
1693        mutex_unlock(&cgroup_mutex);
1694        mutex_unlock(&inode->i_mutex);
1695 drop_new_super:
1696        deactivate_locked_super(sb);
1697 drop_modules:
1698        drop_parsed_module_refcounts(opts.subsys_bits);
1699 out_err:
1700        kfree(opts.release_agent);
1701        kfree(opts.name);
1702        return ERR_PTR(ret);
1703}
1704
1705static void cgroup_kill_sb(struct super_block *sb) {
1706        struct cgroupfs_root *root = sb->s_fs_info;
1707        struct cgroup *cgrp = &root->top_cgroup;
1708        int ret;
1709        struct cg_cgroup_link *link;
1710        struct cg_cgroup_link *saved_link;
1711
1712        BUG_ON(!root);
1713
1714        BUG_ON(root->number_of_cgroups != 1);
1715        BUG_ON(!list_empty(&cgrp->children));
1716        BUG_ON(!list_empty(&cgrp->sibling));
1717
1718        mutex_lock(&cgroup_mutex);
1719        mutex_lock(&cgroup_root_mutex);
1720
1721        /* Rebind all subsystems back to the default hierarchy */
1722        ret = rebind_subsystems(root, 0);
1723        /* Shouldn't be able to fail ... */
1724        BUG_ON(ret);
1725
1726        /*
1727         * Release all the links from css_sets to this hierarchy's
1728         * root cgroup
1729         */
1730        write_lock(&css_set_lock);
1731
1732        list_for_each_entry_safe(link, saved_link, &cgrp->css_sets,
1733                                 cgrp_link_list) {
1734                list_del(&link->cg_link_list);
1735                list_del(&link->cgrp_link_list);
1736                kfree(link);
1737        }
1738        write_unlock(&css_set_lock);
1739
1740        if (!list_empty(&root->root_list)) {
1741                list_del(&root->root_list);
1742                root_count--;
1743        }
1744
1745        mutex_unlock(&cgroup_root_mutex);
1746        mutex_unlock(&cgroup_mutex);
1747
1748        kill_litter_super(sb);
1749        cgroup_drop_root(root);
1750}
1751
1752static struct file_system_type cgroup_fs_type = {
1753        .name = "cgroup",
1754        .mount = cgroup_mount,
1755        .kill_sb = cgroup_kill_sb,
1756};
1757
1758static struct kobject *cgroup_kobj;
1759
1760/**
1761 * cgroup_path - generate the path of a cgroup
1762 * @cgrp: the cgroup in question
1763 * @buf: the buffer to write the path into
1764 * @buflen: the length of the buffer
1765 *
1766 * Called with cgroup_mutex held or else with an RCU-protected cgroup
1767 * reference.  Writes path of cgroup into buf.  Returns 0 on success,
1768 * -errno on error.
1769 */
1770int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen)
1771{
1772        char *start;
1773        struct dentry *dentry = rcu_dereference_check(cgrp->dentry,
1774                                                      cgroup_lock_is_held());
1775
1776        if (!dentry || cgrp == dummytop) {
1777                /*
1778                 * Inactive subsystems have no dentry for their root
1779                 * cgroup
1780                 */
1781                strcpy(buf, "/");
1782                return 0;
1783        }
1784
1785        start = buf + buflen;
1786
1787        *--start = '\0';
1788        for (;;) {
1789                int len = dentry->d_name.len;
1790
1791                if ((start -= len) < buf)
1792                        return -ENAMETOOLONG;
1793                memcpy(start, dentry->d_name.name, len);
1794                cgrp = cgrp->parent;
1795                if (!cgrp)
1796                        break;
1797
1798                dentry = rcu_dereference_check(cgrp->dentry,
1799                                               cgroup_lock_is_held());
1800                if (!cgrp->parent)
1801                        continue;
1802                if (--start < buf)
1803                        return -ENAMETOOLONG;
1804                *start = '/';
1805        }
1806        memmove(buf, start, buf + buflen - start);
1807        return 0;
1808}
1809EXPORT_SYMBOL_GPL(cgroup_path);
1810
1811/*
1812 * Control Group taskset
1813 */
1814struct task_and_cgroup {
1815        struct task_struct      *task;
1816        struct cgroup           *cgrp;
1817        struct css_set          *cg;
1818};
1819
1820struct cgroup_taskset {
1821        struct task_and_cgroup  single;
1822        struct flex_array       *tc_array;
1823        int                     tc_array_len;
1824        int                     idx;
1825        struct cgroup           *cur_cgrp;
1826};
1827
1828/**
1829 * cgroup_taskset_first - reset taskset and return the first task
1830 * @tset: taskset of interest
1831 *
1832 * @tset iteration is initialized and the first task is returned.
1833 */
1834struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset)
1835{
1836        if (tset->tc_array) {
1837                tset->idx = 0;
1838                return cgroup_taskset_next(tset);
1839        } else {
1840                tset->cur_cgrp = tset->single.cgrp;
1841                return tset->single.task;
1842        }
1843}
1844EXPORT_SYMBOL_GPL(cgroup_taskset_first);
1845
1846/**
1847 * cgroup_taskset_next - iterate to the next task in taskset
1848 * @tset: taskset of interest
1849 *
1850 * Return the next task in @tset.  Iteration must have been initialized
1851 * with cgroup_taskset_first().
1852 */
1853struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset)
1854{
1855        struct task_and_cgroup *tc;
1856
1857        if (!tset->tc_array || tset->idx >= tset->tc_array_len)
1858                return NULL;
1859
1860        tc = flex_array_get(tset->tc_array, tset->idx++);
1861        tset->cur_cgrp = tc->cgrp;
1862        return tc->task;
1863}
1864EXPORT_SYMBOL_GPL(cgroup_taskset_next);
1865
1866/**
1867 * cgroup_taskset_cur_cgroup - return the matching cgroup for the current task
1868 * @tset: taskset of interest
1869 *
1870 * Return the cgroup for the current (last returned) task of @tset.  This
1871 * function must be preceded by either cgroup_taskset_first() or
1872 * cgroup_taskset_next().
1873 */
1874struct cgroup *cgroup_taskset_cur_cgroup(struct cgroup_taskset *tset)
1875{
1876        return tset->cur_cgrp;
1877}
1878EXPORT_SYMBOL_GPL(cgroup_taskset_cur_cgroup);
1879
1880/**
1881 * cgroup_taskset_size - return the number of tasks in taskset
1882 * @tset: taskset of interest
1883 */
1884int cgroup_taskset_size(struct cgroup_taskset *tset)
1885{
1886        return tset->tc_array ? tset->tc_array_len : 1;
1887}
1888EXPORT_SYMBOL_GPL(cgroup_taskset_size);
1889
1890
1891/*
1892 * cgroup_task_migrate - move a task from one cgroup to another.
1893 *
1894 * 'guarantee' is set if the caller promises that a new css_set for the task
1895 * will already exist. If not set, this function might sleep, and can fail with
1896 * -ENOMEM. Must be called with cgroup_mutex and threadgroup locked.
1897 */
1898static void cgroup_task_migrate(struct cgroup *cgrp, struct cgroup *oldcgrp,
1899                                struct task_struct *tsk, struct css_set *newcg)
1900{
1901        struct css_set *oldcg;
1902
1903        /*
1904         * We are synchronized through threadgroup_lock() against PF_EXITING
1905         * setting such that we can't race against cgroup_exit() changing the
1906         * css_set to init_css_set and dropping the old one.
1907         */
1908        WARN_ON_ONCE(tsk->flags & PF_EXITING);
1909        oldcg = tsk->cgroups;
1910
1911        task_lock(tsk);
1912        rcu_assign_pointer(tsk->cgroups, newcg);
1913        task_unlock(tsk);
1914
1915        /* Update the css_set linked lists if we're using them */
1916        write_lock(&css_set_lock);
1917        if (!list_empty(&tsk->cg_list))
1918                list_move(&tsk->cg_list, &newcg->tasks);
1919        write_unlock(&css_set_lock);
1920
1921        /*
1922         * We just gained a reference on oldcg by taking it from the task. As
1923         * trading it for newcg is protected by cgroup_mutex, we're safe to drop
1924         * it here; it will be freed under RCU.
1925         */
1926        put_css_set(oldcg);
1927
1928        set_bit(CGRP_RELEASABLE, &oldcgrp->flags);
1929}
1930
1931/**
1932 * cgroup_attach_task - attach task 'tsk' to cgroup 'cgrp'
1933 * @cgrp: the cgroup the task is attaching to
1934 * @tsk: the task to be attached
1935 *
1936 * Call with cgroup_mutex and threadgroup locked. May take task_lock of
1937 * @tsk during call.
1938 */
1939int cgroup_attach_task(struct cgroup *cgrp, struct task_struct *tsk)
1940{
1941        int retval = 0;
1942        struct cgroup_subsys *ss, *failed_ss = NULL;
1943        struct cgroup *oldcgrp;
1944        struct cgroupfs_root *root = cgrp->root;
1945        struct cgroup_taskset tset = { };
1946        struct css_set *newcg;
1947
1948        /* @tsk either already exited or can't exit until the end */
1949        if (tsk->flags & PF_EXITING)
1950                return -ESRCH;
1951
1952        /* Nothing to do if the task is already in that cgroup */
1953        oldcgrp = task_cgroup_from_root(tsk, root);
1954        if (cgrp == oldcgrp)
1955                return 0;
1956
1957        tset.single.task = tsk;
1958        tset.single.cgrp = oldcgrp;
1959
1960        for_each_subsys(root, ss) {
1961                if (ss->can_attach) {
1962                        retval = ss->can_attach(cgrp, &tset);
1963                        if (retval) {
1964                                /*
1965                                 * Remember on which subsystem the can_attach()
1966                                 * failed, so that we only call cancel_attach()
1967                                 * against the subsystems whose can_attach()
1968                                 * succeeded. (See below)
1969                                 */
1970                                failed_ss = ss;
1971                                goto out;
1972                        }
1973                }
1974        }
1975
1976        newcg = find_css_set(tsk->cgroups, cgrp);
1977        if (!newcg) {
1978                retval = -ENOMEM;
1979                goto out;
1980        }
1981
1982        cgroup_task_migrate(cgrp, oldcgrp, tsk, newcg);
1983
1984        for_each_subsys(root, ss) {
1985                if (ss->attach)
1986                        ss->attach(cgrp, &tset);
1987        }
1988
1989        synchronize_rcu();
1990
1991        /*
1992         * wake up rmdir() waiter. the rmdir should fail since the cgroup
1993         * is no longer empty.
1994         */
1995        cgroup_wakeup_rmdir_waiter(cgrp);
1996out:
1997        if (retval) {
1998                for_each_subsys(root, ss) {
1999                        if (ss == failed_ss)
2000                                /*
2001                                 * This subsystem was the one that failed the
2002                                 * can_attach() check earlier, so we don't need
2003                                 * to call cancel_attach() against it or any
2004                                 * remaining subsystems.
2005                                 */
2006                                break;
2007                        if (ss->cancel_attach)
2008                                ss->cancel_attach(cgrp, &tset);
2009                }
2010        }
2011        return retval;
2012}
2013
2014/**
2015 * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from'
2016 * @from: attach to all cgroups of a given task
2017 * @tsk: the task to be attached
2018 */
2019int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk)
2020{
2021        struct cgroupfs_root *root;
2022        int retval = 0;
2023
2024        cgroup_lock();
2025        for_each_active_root(root) {
2026                struct cgroup *from_cg = task_cgroup_from_root(from, root);
2027
2028                retval = cgroup_attach_task(from_cg, tsk);
2029                if (retval)
2030                        break;
2031        }
2032        cgroup_unlock();
2033
2034        return retval;
2035}
2036EXPORT_SYMBOL_GPL(cgroup_attach_task_all);
2037
2038/**
2039 * cgroup_attach_proc - attach all threads in a threadgroup to a cgroup
2040 * @cgrp: the cgroup to attach to
2041 * @leader: the threadgroup leader task_struct of the group to be attached
2042 *
2043 * Call holding cgroup_mutex and the group_rwsem of the leader. Will take
2044 * task_lock of each thread in leader's threadgroup individually in turn.
2045 */
2046static int cgroup_attach_proc(struct cgroup *cgrp, struct task_struct *leader)
2047{
2048        int retval, i, group_size;
2049        struct cgroup_subsys *ss, *failed_ss = NULL;
2050        /* guaranteed to be initialized later, but the compiler needs this */
2051        struct cgroupfs_root *root = cgrp->root;
2052        /* threadgroup list cursor and array */
2053        struct task_struct *tsk;
2054        struct task_and_cgroup *tc;
2055        struct flex_array *group;
2056        struct cgroup_taskset tset = { };
2057
2058        /*
2059         * step 0: in order to do expensive, possibly blocking operations for
2060         * every thread, we cannot iterate the thread group list, since it needs
2061         * rcu or tasklist locked. instead, build an array of all threads in the
2062         * group - group_rwsem prevents new threads from appearing, and if
2063         * threads exit, this will just be an over-estimate.
2064         */
2065        group_size = get_nr_threads(leader);
2066        /* flex_array supports very large thread-groups better than kmalloc. */
2067        group = flex_array_alloc(sizeof(*tc), group_size, GFP_KERNEL);
2068        if (!group)
2069                return -ENOMEM;
2070        /* pre-allocate to guarantee space while iterating in rcu read-side. */
2071        retval = flex_array_prealloc(group, 0, group_size - 1, GFP_KERNEL);
2072        if (retval)
2073                goto out_free_group_list;
2074
2075        tsk = leader;
2076        i = 0;
2077        /*
2078         * Prevent freeing of tasks while we take a snapshot. Tasks that are
2079         * already PF_EXITING could be freed from underneath us unless we
2080         * take an rcu_read_lock.
2081         */
2082        rcu_read_lock();
2083        do {
2084                struct task_and_cgroup ent;
2085
2086                /* @tsk either already exited or can't exit until the end */
2087                if (tsk->flags & PF_EXITING)
2088                        continue;
2089
2090                /* as per above, nr_threads may decrease, but not increase. */
2091                BUG_ON(i >= group_size);
2092                ent.task = tsk;
2093                ent.cgrp = task_cgroup_from_root(tsk, root);
2094                /* nothing to do if this task is already in the cgroup */
2095                if (ent.cgrp == cgrp)
2096                        continue;
2097                /*
2098                 * saying GFP_ATOMIC has no effect here because we did prealloc
2099                 * earlier, but it's good form to communicate our expectations.
2100                 */
2101                retval = flex_array_put(group, i, &ent, GFP_ATOMIC);
2102                BUG_ON(retval != 0);
2103                i++;
2104        } while_each_thread(leader, tsk);
2105        rcu_read_unlock();
2106        /* remember the number of threads in the array for later. */
2107        group_size = i;
2108        tset.tc_array = group;
2109        tset.tc_array_len = group_size;
2110
2111        /* methods shouldn't be called if no task is actually migrating */
2112        retval = 0;
2113        if (!group_size)
2114                goto out_free_group_list;
2115
2116        /*
2117         * step 1: check that we can legitimately attach to the cgroup.
2118         */
2119        for_each_subsys(root, ss) {
2120                if (ss->can_attach) {
2121                        retval = ss->can_attach(cgrp, &tset);
2122                        if (retval) {
2123                                failed_ss = ss;
2124                                goto out_cancel_attach;
2125                        }
2126                }
2127        }
2128
2129        /*
2130         * step 2: make sure css_sets exist for all threads to be migrated.
2131         * we use find_css_set, which allocates a new one if necessary.
2132         */
2133        for (i = 0; i < group_size; i++) {
2134                tc = flex_array_get(group, i);
2135                tc->cg = find_css_set(tc->task->cgroups, cgrp);
2136                if (!tc->cg) {
2137                        retval = -ENOMEM;
2138                        goto out_put_css_set_refs;
2139                }
2140        }
2141
2142        /*
2143         * step 3: now that we're guaranteed success wrt the css_sets,
2144         * proceed to move all tasks to the new cgroup.  There are no
2145         * failure cases after here, so this is the commit point.
2146         */
2147        for (i = 0; i < group_size; i++) {
2148                tc = flex_array_get(group, i);
2149                cgroup_task_migrate(cgrp, tc->cgrp, tc->task, tc->cg);
2150        }
2151        /* nothing is sensitive to fork() after this point. */
2152
2153        /*
2154         * step 4: do subsystem attach callbacks.
2155         */
2156        for_each_subsys(root, ss) {
2157                if (ss->attach)
2158                        ss->attach(cgrp, &tset);
2159        }
2160
2161        /*
2162         * step 5: success! and cleanup
2163         */
2164        synchronize_rcu();
2165        cgroup_wakeup_rmdir_waiter(cgrp);
2166        retval = 0;
2167out_put_css_set_refs:
2168        if (retval) {
2169                for (i = 0; i < group_size; i++) {
2170                        tc = flex_array_get(group, i);
2171                        if (!tc->cg)
2172                                break;
2173                        put_css_set(tc->cg);
2174                }
2175        }
2176out_cancel_attach:
2177        if (retval) {
2178                for_each_subsys(root, ss) {
2179                        if (ss == failed_ss)
2180                                break;
2181                        if (ss->cancel_attach)
2182                                ss->cancel_attach(cgrp, &tset);
2183                }
2184        }
2185out_free_group_list:
2186        flex_array_free(group);
2187        return retval;
2188}
2189
2190/*
2191 * Find the task_struct of the task to attach by vpid and pass it along to the
2192 * function to attach either it or all tasks in its threadgroup. Will lock
2193 * cgroup_mutex and threadgroup; may take task_lock of task.
2194 */
2195static int attach_task_by_pid(struct cgroup *cgrp, u64 pid, bool threadgroup)
2196{
2197        struct task_struct *tsk;
2198        const struct cred *cred = current_cred(), *tcred;
2199        int ret;
2200
2201        if (!cgroup_lock_live_group(cgrp))
2202                return -ENODEV;
2203
2204retry_find_task:
2205        rcu_read_lock();
2206        if (pid) {
2207                tsk = find_task_by_vpid(pid);
2208                if (!tsk) {
2209                        rcu_read_unlock();
2210                        ret= -ESRCH;
2211                        goto out_unlock_cgroup;
2212                }
2213                /*
2214                 * even if we're attaching all tasks in the thread group, we
2215                 * only need to check permissions on one of them.
2216                 */
2217                tcred = __task_cred(tsk);
2218                if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) &&
2219                    !uid_eq(cred->euid, tcred->uid) &&
2220                    !uid_eq(cred->euid, tcred->suid)) {
2221                        rcu_read_unlock();
2222                        ret = -EACCES;
2223                        goto out_unlock_cgroup;
2224                }
2225        } else
2226                tsk = current;
2227
2228        if (threadgroup)
2229                tsk = tsk->group_leader;
2230
2231        /*
2232         * Workqueue threads may acquire PF_THREAD_BOUND and become
2233         * trapped in a cpuset, or RT worker may be born in a cgroup
2234         * with no rt_runtime allocated.  Just say no.
2235         */
2236        if (tsk == kthreadd_task || (tsk->flags & PF_THREAD_BOUND)) {
2237                ret = -EINVAL;
2238                rcu_read_unlock();
2239                goto out_unlock_cgroup;
2240        }
2241
2242        get_task_struct(tsk);
2243        rcu_read_unlock();
2244
2245        threadgroup_lock(tsk);
2246        if (threadgroup) {
2247                if (!thread_group_leader(tsk)) {
2248                        /*
2249                         * a race with de_thread from another thread's exec()
2250                         * may strip us of our leadership, if this happens,
2251                         * there is no choice but to throw this task away and
2252                         * try again; this is
2253                         * "double-double-toil-and-trouble-check locking".
2254                         */
2255                        threadgroup_unlock(tsk);
2256                        put_task_struct(tsk);
2257                        goto retry_find_task;
2258                }
2259                ret = cgroup_attach_proc(cgrp, tsk);
2260        } else
2261                ret = cgroup_attach_task(cgrp, tsk);
2262        threadgroup_unlock(tsk);
2263
2264        put_task_struct(tsk);
2265out_unlock_cgroup:
2266        cgroup_unlock();
2267        return ret;
2268}
2269
2270static int cgroup_tasks_write(struct cgroup *cgrp, struct cftype *cft, u64 pid)
2271{
2272        return attach_task_by_pid(cgrp, pid, false);
2273}
2274
2275static int cgroup_procs_write(struct cgroup *cgrp, struct cftype *cft, u64 tgid)
2276{
2277        return attach_task_by_pid(cgrp, tgid, true);
2278}
2279
2280/**
2281 * cgroup_lock_live_group - take cgroup_mutex and check that cgrp is alive.
2282 * @cgrp: the cgroup to be checked for liveness
2283 *
2284 * On success, returns true; the lock should be later released with
2285 * cgroup_unlock(). On failure returns false with no lock held.
2286 */
2287bool cgroup_lock_live_group(struct cgroup *cgrp)
2288{
2289        mutex_lock(&cgroup_mutex);
2290        if (cgroup_is_removed(cgrp)) {
2291                mutex_unlock(&cgroup_mutex);
2292                return false;
2293        }
2294        return true;
2295}
2296EXPORT_SYMBOL_GPL(cgroup_lock_live_group);
2297
2298static int cgroup_release_agent_write(struct cgroup *cgrp, struct cftype *cft,
2299                                      const char *buffer)
2300{
2301        BUILD_BUG_ON(sizeof(cgrp->root->release_agent_path) < PATH_MAX);
2302        if (strlen(buffer) >= PATH_MAX)
2303                return -EINVAL;
2304        if (!cgroup_lock_live_group(cgrp))
2305                return -ENODEV;
2306        mutex_lock(&cgroup_root_mutex);
2307        strcpy(cgrp->root->release_agent_path, buffer);
2308        mutex_unlock(&cgroup_root_mutex);
2309        cgroup_unlock();
2310        return 0;
2311}
2312
2313static int cgroup_release_agent_show(struct cgroup *cgrp, struct cftype *cft,
2314                                     struct seq_file *seq)
2315{
2316        if (!cgroup_lock_live_group(cgrp))
2317                return -ENODEV;
2318        seq_puts(seq, cgrp->root->release_agent_path);
2319        seq_putc(seq, '\n');
2320        cgroup_unlock();
2321        return 0;
2322}
2323
2324/* A buffer size big enough for numbers or short strings */
2325#define CGROUP_LOCAL_BUFFER_SIZE 64
2326
2327static ssize_t cgroup_write_X64(struct cgroup *cgrp, struct cftype *cft,
2328                                struct file *file,
2329                                const char __user *userbuf,
2330                                size_t nbytes, loff_t *unused_ppos)
2331{
2332        char buffer[CGROUP_LOCAL_BUFFER_SIZE];
2333        int retval = 0;
2334        char *end;
2335
2336        if (!nbytes)
2337                return -EINVAL;
2338        if (nbytes >= sizeof(buffer))
2339                return -E2BIG;
2340        if (copy_from_user(buffer, userbuf, nbytes))
2341                return -EFAULT;
2342
2343        buffer[nbytes] = 0;     /* nul-terminate */
2344        if (cft->write_u64) {
2345                u64 val = simple_strtoull(strstrip(buffer), &end, 0);
2346                if (*end)
2347                        return -EINVAL;
2348                retval = cft->write_u64(cgrp, cft, val);
2349        } else {
2350                s64 val = simple_strtoll(strstrip(buffer), &end, 0);
2351                if (*end)
2352                        return -EINVAL;
2353                retval = cft->write_s64(cgrp, cft, val);
2354        }
2355        if (!retval)
2356                retval = nbytes;
2357        return retval;
2358}
2359
2360static ssize_t cgroup_write_string(struct cgroup *cgrp, struct cftype *cft,
2361                                   struct file *file,
2362                                   const char __user *userbuf,
2363                                   size_t nbytes, loff_t *unused_ppos)
2364{
2365        char local_buffer[CGROUP_LOCAL_BUFFER_SIZE];
2366        int retval = 0;
2367        size_t max_bytes = cft->max_write_len;
2368        char *buffer = local_buffer;
2369
2370        if (!max_bytes)
2371                max_bytes = sizeof(local_buffer) - 1;
2372        if (nbytes >= max_bytes)
2373                return -E2BIG;
2374        /* Allocate a dynamic buffer if we need one */
2375        if (nbytes >= sizeof(local_buffer)) {
2376                buffer = kmalloc(nbytes + 1, GFP_KERNEL);
2377                if (buffer == NULL)
2378                        return -ENOMEM;
2379        }
2380        if (nbytes && copy_from_user(buffer, userbuf, nbytes)) {
2381                retval = -EFAULT;
2382                goto out;
2383        }
2384
2385        buffer[nbytes] = 0;     /* nul-terminate */
2386        retval = cft->write_string(cgrp, cft, strstrip(buffer));
2387        if (!retval)
2388                retval = nbytes;
2389out:
2390        if (buffer != local_buffer)
2391                kfree(buffer);
2392        return retval;
2393}
2394
2395static ssize_t cgroup_file_write(struct file *file, const char __user *buf,
2396                                                size_t nbytes, loff_t *ppos)
2397{
2398        struct cftype *cft = __d_cft(file->f_dentry);
2399        struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
2400
2401        if (cgroup_is_removed(cgrp))
2402                return -ENODEV;
2403        if (cft->write)
2404                return cft->write(cgrp, cft, file, buf, nbytes, ppos);
2405        if (cft->write_u64 || cft->write_s64)
2406                return cgroup_write_X64(cgrp, cft, file, buf, nbytes, ppos);
2407        if (cft->write_string)
2408                return cgroup_write_string(cgrp, cft, file, buf, nbytes, ppos);
2409        if (cft->trigger) {
2410                int ret = cft->trigger(cgrp, (unsigned int)cft->private);
2411                return ret ? ret : nbytes;
2412        }
2413        return -EINVAL;
2414}
2415
2416static ssize_t cgroup_read_u64(struct cgroup *cgrp, struct cftype *cft,
2417                               struct file *file,
2418                               char __user *buf, size_t nbytes,
2419                               loff_t *ppos)
2420{
2421        char tmp[CGROUP_LOCAL_BUFFER_SIZE];
2422        u64 val = cft->read_u64(cgrp, cft);
2423        int len = sprintf(tmp, "%llu\n", (unsigned long long) val);
2424
2425        return simple_read_from_buffer(buf, nbytes, ppos, tmp, len);
2426}
2427
2428static ssize_t cgroup_read_s64(struct cgroup *cgrp, struct cftype *cft,
2429                               struct file *file,
2430                               char __user *buf, size_t nbytes,
2431                               loff_t *ppos)
2432{
2433        char tmp[CGROUP_LOCAL_BUFFER_SIZE];
2434        s64 val = cft->read_s64(cgrp, cft);
2435        int len = sprintf(tmp, "%lld\n", (long long) val);
2436
2437        return simple_read_from_buffer(buf, nbytes, ppos, tmp, len);
2438}
2439
2440static ssize_t cgroup_file_read(struct file *file, char __user *buf,
2441                                   size_t nbytes, loff_t *ppos)
2442{
2443        struct cftype *cft = __d_cft(file->f_dentry);
2444        struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
2445
2446        if (cgroup_is_removed(cgrp))
2447                return -ENODEV;
2448
2449        if (cft->read)
2450                return cft->read(cgrp, cft, file, buf, nbytes, ppos);
2451        if (cft->read_u64)
2452                return cgroup_read_u64(cgrp, cft, file, buf, nbytes, ppos);
2453        if (cft->read_s64)
2454                return cgroup_read_s64(cgrp, cft, file, buf, nbytes, ppos);
2455        return -EINVAL;
2456}
2457
2458/*
2459 * seqfile ops/methods for returning structured data. Currently just
2460 * supports string->u64 maps, but can be extended in future.
2461 */
2462
2463struct cgroup_seqfile_state {
2464        struct cftype *cft;
2465        struct cgroup *cgroup;
2466};
2467
2468static int cgroup_map_add(struct cgroup_map_cb *cb, const char *key, u64 value)
2469{
2470        struct seq_file *sf = cb->state;
2471        return seq_printf(sf, "%s %llu\n", key, (unsigned long long)value);
2472}
2473
2474static int cgroup_seqfile_show(struct seq_file *m, void *arg)
2475{
2476        struct cgroup_seqfile_state *state = m->private;
2477        struct cftype *cft = state->cft;
2478        if (cft->read_map) {
2479                struct cgroup_map_cb cb = {
2480                        .fill = cgroup_map_add,
2481                        .state = m,
2482                };
2483                return cft->read_map(state->cgroup, cft, &cb);
2484        }
2485        return cft->read_seq_string(state->cgroup, cft, m);
2486}
2487
2488static int cgroup_seqfile_release(struct inode *inode, struct file *file)
2489{
2490        struct seq_file *seq = file->private_data;
2491        kfree(seq->private);
2492        return single_release(inode, file);
2493}
2494
2495static const struct file_operations cgroup_seqfile_operations = {
2496        .read = seq_read,
2497        .write = cgroup_file_write,
2498        .llseek = seq_lseek,
2499        .release = cgroup_seqfile_release,
2500};
2501
2502static int cgroup_file_open(struct inode *inode, struct file *file)
2503{
2504        int err;
2505        struct cftype *cft;
2506
2507        err = generic_file_open(inode, file);
2508        if (err)
2509                return err;
2510        cft = __d_cft(file->f_dentry);
2511
2512        if (cft->read_map || cft->read_seq_string) {
2513                struct cgroup_seqfile_state *state =
2514                        kzalloc(sizeof(*state), GFP_USER);
2515                if (!state)
2516                        return -ENOMEM;
2517                state->cft = cft;
2518                state->cgroup = __d_cgrp(file->f_dentry->d_parent);
2519                file->f_op = &cgroup_seqfile_operations;
2520                err = single_open(file, cgroup_seqfile_show, state);
2521                if (err < 0)
2522                        kfree(state);
2523        } else if (cft->open)
2524                err = cft->open(inode, file);
2525        else
2526                err = 0;
2527
2528        return err;
2529}
2530
2531static int cgroup_file_release(struct inode *inode, struct file *file)
2532{
2533        struct cftype *cft = __d_cft(file->f_dentry);
2534        if (cft->release)
2535                return cft->release(inode, file);
2536        return 0;
2537}
2538
2539/*
2540 * cgroup_rename - Only allow simple rename of directories in place.
2541 */
2542static int cgroup_rename(struct inode *old_dir, struct dentry *old_dentry,
2543                            struct inode *new_dir, struct dentry *new_dentry)
2544{
2545        if (!S_ISDIR(old_dentry->d_inode->i_mode))
2546                return -ENOTDIR;
2547        if (new_dentry->d_inode)
2548                return -EEXIST;
2549        if (old_dir != new_dir)
2550                return -EIO;
2551        return simple_rename(old_dir, old_dentry, new_dir, new_dentry);
2552}
2553
2554static const struct file_operations cgroup_file_operations = {
2555        .read = cgroup_file_read,
2556        .write = cgroup_file_write,
2557        .llseek = generic_file_llseek,
2558        .open = cgroup_file_open,
2559        .release = cgroup_file_release,
2560};
2561
2562static const struct inode_operations cgroup_dir_inode_operations = {
2563        .lookup = cgroup_lookup,
2564        .mkdir = cgroup_mkdir,
2565        .rmdir = cgroup_rmdir,
2566        .rename = cgroup_rename,
2567};
2568
2569static struct dentry *cgroup_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
2570{
2571        if (dentry->d_name.len > NAME_MAX)
2572                return ERR_PTR(-ENAMETOOLONG);
2573        d_add(dentry, NULL);
2574        return NULL;
2575}
2576
2577/*
2578 * Check if a file is a control file
2579 */
2580static inline struct cftype *__file_cft(struct file *file)
2581{
2582        if (file->f_dentry->d_inode->i_fop != &cgroup_file_operations)
2583                return ERR_PTR(-EINVAL);
2584        return __d_cft(file->f_dentry);
2585}
2586
2587static int cgroup_create_file(struct dentry *dentry, umode_t mode,
2588                                struct super_block *sb)
2589{
2590        struct inode *inode;
2591
2592        if (!dentry)
2593                return -ENOENT;
2594        if (dentry->d_inode)
2595                return -EEXIST;
2596
2597        inode = cgroup_new_inode(mode, sb);
2598        if (!inode)
2599                return -ENOMEM;
2600
2601        if (S_ISDIR(mode)) {
2602                inode->i_op = &cgroup_dir_inode_operations;
2603                inode->i_fop = &simple_dir_operations;
2604
2605                /* start off with i_nlink == 2 (for "." entry) */
2606                inc_nlink(inode);
2607
2608                /* start with the directory inode held, so that we can
2609                 * populate it without racing with another mkdir */
2610                mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
2611        } else if (S_ISREG(mode)) {
2612                inode->i_size = 0;
2613                inode->i_fop = &cgroup_file_operations;
2614        }
2615        d_instantiate(dentry, inode);
2616        dget(dentry);   /* Extra count - pin the dentry in core */
2617        return 0;
2618}
2619
2620/*
2621 * cgroup_create_dir - create a directory for an object.
2622 * @cgrp: the cgroup we create the directory for. It must have a valid
2623 *        ->parent field. And we are going to fill its ->dentry field.
2624 * @dentry: dentry of the new cgroup
2625 * @mode: mode to set on new directory.
2626 */
2627static int cgroup_create_dir(struct cgroup *cgrp, struct dentry *dentry,
2628                                umode_t mode)
2629{
2630        struct dentry *parent;
2631        int error = 0;
2632
2633        parent = cgrp->parent->dentry;
2634        error = cgroup_create_file(dentry, S_IFDIR | mode, cgrp->root->sb);
2635        if (!error) {
2636                dentry->d_fsdata = cgrp;
2637                inc_nlink(parent->d_inode);
2638                rcu_assign_pointer(cgrp->dentry, dentry);
2639                dget(dentry);
2640        }
2641        dput(dentry);
2642
2643        return error;
2644}
2645
2646/**
2647 * cgroup_file_mode - deduce file mode of a control file
2648 * @cft: the control file in question
2649 *
2650 * returns cft->mode if ->mode is not 0
2651 * returns S_IRUGO|S_IWUSR if it has both a read and a write handler
2652 * returns S_IRUGO if it has only a read handler
2653 * returns S_IWUSR if it has only a write hander
2654 */
2655static umode_t cgroup_file_mode(const struct cftype *cft)
2656{
2657        umode_t mode = 0;
2658
2659        if (cft->mode)
2660                return cft->mode;
2661
2662        if (cft->read || cft->read_u64 || cft->read_s64 ||
2663            cft->read_map || cft->read_seq_string)
2664                mode |= S_IRUGO;
2665
2666        if (cft->write || cft->write_u64 || cft->write_s64 ||
2667            cft->write_string || cft->trigger)
2668                mode |= S_IWUSR;
2669
2670        return mode;
2671}
2672
2673static int cgroup_add_file(struct cgroup *cgrp, struct cgroup_subsys *subsys,
2674                           const struct cftype *cft)
2675{
2676        struct dentry *dir = cgrp->dentry;
2677        struct cgroup *parent = __d_cgrp(dir);
2678        struct dentry *dentry;
2679        struct cfent *cfe;
2680        int error;
2681        umode_t mode;
2682        char name[MAX_CGROUP_TYPE_NAMELEN + MAX_CFTYPE_NAME + 2] = { 0 };
2683
2684        /* does @cft->flags tell us to skip creation on @cgrp? */
2685        if ((cft->flags & CFTYPE_NOT_ON_ROOT) && !cgrp->parent)
2686                return 0;
2687        if ((cft->flags & CFTYPE_ONLY_ON_ROOT) && cgrp->parent)
2688                return 0;
2689
2690        if (subsys && !test_bit(ROOT_NOPREFIX, &cgrp->root->flags)) {
2691                strcpy(name, subsys->name);
2692                strcat(name, ".");
2693        }
2694        strcat(name, cft->name);
2695
2696        BUG_ON(!mutex_is_locked(&dir->d_inode->i_mutex));
2697
2698        cfe = kzalloc(sizeof(*cfe), GFP_KERNEL);
2699        if (!cfe)
2700                return -ENOMEM;
2701
2702        dentry = lookup_one_len(name, dir, strlen(name));
2703        if (IS_ERR(dentry)) {
2704                error = PTR_ERR(dentry);
2705                goto out;
2706        }
2707
2708        mode = cgroup_file_mode(cft);
2709        error = cgroup_create_file(dentry, mode | S_IFREG, cgrp->root->sb);
2710        if (!error) {
2711                cfe->type = (void *)cft;
2712                cfe->dentry = dentry;
2713                dentry->d_fsdata = cfe;
2714                list_add_tail(&cfe->node, &parent->files);
2715                cfe = NULL;
2716        }
2717        dput(dentry);
2718out:
2719        kfree(cfe);
2720        return error;
2721}
2722
2723static int cgroup_addrm_files(struct cgroup *cgrp, struct cgroup_subsys *subsys,
2724                              const struct cftype cfts[], bool is_add)
2725{
2726        const struct cftype *cft;
2727        int err, ret = 0;
2728
2729        for (cft = cfts; cft->name[0] != '\0'; cft++) {
2730                if (is_add)
2731                        err = cgroup_add_file(cgrp, subsys, cft);
2732                else
2733                        err = cgroup_rm_file(cgrp, cft);
2734                if (err) {
2735                        pr_warning("cgroup_addrm_files: failed to %s %s, err=%d\n",
2736                                   is_add ? "add" : "remove", cft->name, err);
2737                        ret = err;
2738                }
2739        }
2740        return ret;
2741}
2742
2743static DEFINE_MUTEX(cgroup_cft_mutex);
2744
2745static void cgroup_cfts_prepare(void)
2746        __acquires(&cgroup_cft_mutex) __acquires(&cgroup_mutex)
2747{
2748        /*
2749         * Thanks to the entanglement with vfs inode locking, we can't walk
2750         * the existing cgroups under cgroup_mutex and create files.
2751         * Instead, we increment reference on all cgroups and build list of
2752         * them using @cgrp->cft_q_node.  Grab cgroup_cft_mutex to ensure
2753         * exclusive access to the field.
2754         */
2755        mutex_lock(&cgroup_cft_mutex);
2756        mutex_lock(&cgroup_mutex);
2757}
2758
2759static void cgroup_cfts_commit(struct cgroup_subsys *ss,
2760                               const struct cftype *cfts, bool is_add)
2761        __releases(&cgroup_mutex) __releases(&cgroup_cft_mutex)
2762{
2763        LIST_HEAD(pending);
2764        struct cgroup *cgrp, *n;
2765
2766        /* %NULL @cfts indicates abort and don't bother if @ss isn't attached */
2767        if (cfts && ss->root != &rootnode) {
2768                list_for_each_entry(cgrp, &ss->root->allcg_list, allcg_node) {
2769                        dget(cgrp->dentry);
2770                        list_add_tail(&cgrp->cft_q_node, &pending);
2771                }
2772        }
2773
2774        mutex_unlock(&cgroup_mutex);
2775
2776        /*
2777         * All new cgroups will see @cfts update on @ss->cftsets.  Add/rm
2778         * files for all cgroups which were created before.
2779         */
2780        list_for_each_entry_safe(cgrp, n, &pending, cft_q_node) {
2781                struct inode *inode = cgrp->dentry->d_inode;
2782
2783                mutex_lock(&inode->i_mutex);
2784                mutex_lock(&cgroup_mutex);
2785                if (!cgroup_is_removed(cgrp))
2786                        cgroup_addrm_files(cgrp, ss, cfts, is_add);
2787                mutex_unlock(&cgroup_mutex);
2788                mutex_unlock(&inode->i_mutex);
2789
2790                list_del_init(&cgrp->cft_q_node);
2791                dput(cgrp->dentry);
2792        }
2793
2794        mutex_unlock(&cgroup_cft_mutex);
2795}
2796
2797/**
2798 * cgroup_add_cftypes - add an array of cftypes to a subsystem
2799 * @ss: target cgroup subsystem
2800 * @cfts: zero-length name terminated array of cftypes
2801 *
2802 * Register @cfts to @ss.  Files described by @cfts are created for all
2803 * existing cgroups to which @ss is attached and all future cgroups will
2804 * have them too.  This function can be called anytime whether @ss is
2805 * attached or not.
2806 *
2807 * Returns 0 on successful registration, -errno on failure.  Note that this
2808 * function currently returns 0 as long as @cfts registration is successful
2809 * even if some file creation attempts on existing cgroups fail.
2810 */
2811int cgroup_add_cftypes(struct cgroup_subsys *ss, const struct cftype *cfts)
2812{
2813        struct cftype_set *set;
2814
2815        set = kzalloc(sizeof(*set), GFP_KERNEL);
2816        if (!set)
2817                return -ENOMEM;
2818
2819        cgroup_cfts_prepare();
2820        set->cfts = cfts;
2821        list_add_tail(&set->node, &ss->cftsets);
2822        cgroup_cfts_commit(ss, cfts, true);
2823
2824        return 0;
2825}
2826EXPORT_SYMBOL_GPL(cgroup_add_cftypes);
2827
2828/**
2829 * cgroup_rm_cftypes - remove an array of cftypes from a subsystem
2830 * @ss: target cgroup subsystem
2831 * @cfts: zero-length name terminated array of cftypes
2832 *
2833 * Unregister @cfts from @ss.  Files described by @cfts are removed from
2834 * all existing cgroups to which @ss is attached and all future cgroups
2835 * won't have them either.  This function can be called anytime whether @ss
2836 * is attached or not.
2837 *
2838 * Returns 0 on successful unregistration, -ENOENT if @cfts is not
2839 * registered with @ss.
2840 */
2841int cgroup_rm_cftypes(struct cgroup_subsys *ss, const struct cftype *cfts)
2842{
2843        struct cftype_set *set;
2844
2845        cgroup_cfts_prepare();
2846
2847        list_for_each_entry(set, &ss->cftsets, node) {
2848                if (set->cfts == cfts) {
2849                        list_del_init(&set->node);
2850                        cgroup_cfts_commit(ss, cfts, false);
2851                        return 0;
2852                }
2853        }
2854
2855        cgroup_cfts_commit(ss, NULL, false);
2856        return -ENOENT;
2857}
2858
2859/**
2860 * cgroup_task_count - count the number of tasks in a cgroup.
2861 * @cgrp: the cgroup in question
2862 *
2863 * Return the number of tasks in the cgroup.
2864 */
2865int cgroup_task_count(const struct cgroup *cgrp)
2866{
2867        int count = 0;
2868        struct cg_cgroup_link *link;
2869
2870        read_lock(&css_set_lock);
2871        list_for_each_entry(link, &cgrp->css_sets, cgrp_link_list) {
2872                count += atomic_read(&link->cg->refcount);
2873        }
2874        read_unlock(&css_set_lock);
2875        return count;
2876}
2877
2878/*
2879 * Advance a list_head iterator.  The iterator should be positioned at
2880 * the start of a css_set
2881 */
2882static void cgroup_advance_iter(struct cgroup *cgrp,
2883                                struct cgroup_iter *it)
2884{
2885        struct list_head *l = it->cg_link;
2886        struct cg_cgroup_link *link;
2887        struct css_set *cg;
2888
2889        /* Advance to the next non-empty css_set */
2890        do {
2891                l = l->next;
2892                if (l == &cgrp->css_sets) {
2893                        it->cg_link = NULL;
2894                        return;
2895                }
2896                link = list_entry(l, struct cg_cgroup_link, cgrp_link_list);
2897                cg = link->cg;
2898        } while (list_empty(&cg->tasks));
2899        it->cg_link = l;
2900        it->task = cg->tasks.next;
2901}
2902
2903/*
2904 * To reduce the fork() overhead for systems that are not actually
2905 * using their cgroups capability, we don't maintain the lists running
2906 * through each css_set to its tasks until we see the list actually
2907 * used - in other words after the first call to cgroup_iter_start().
2908 */
2909static void cgroup_enable_task_cg_lists(void)
2910{
2911        struct task_struct *p, *g;
2912        write_lock(&css_set_lock);
2913        use_task_css_set_links = 1;
2914        /*
2915         * We need tasklist_lock because RCU is not safe against
2916         * while_each_thread(). Besides, a forking task that has passed
2917         * cgroup_post_fork() without seeing use_task_css_set_links = 1
2918         * is not guaranteed to have its child immediately visible in the
2919         * tasklist if we walk through it with RCU.
2920         */
2921        read_lock(&tasklist_lock);
2922        do_each_thread(g, p) {
2923                task_lock(p);
2924                /*
2925                 * We should check if the process is exiting, otherwise
2926                 * it will race with cgroup_exit() in that the list
2927                 * entry won't be deleted though the process has exited.
2928                 */
2929                if (!(p->flags & PF_EXITING) && list_empty(&p->cg_list))
2930                        list_add(&p->cg_list, &p->cgroups->tasks);
2931                task_unlock(p);
2932        } while_each_thread(g, p);
2933        read_unlock(&tasklist_lock);
2934        write_unlock(&css_set_lock);
2935}
2936
2937void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it)
2938        __acquires(css_set_lock)
2939{
2940        /*
2941         * The first time anyone tries to iterate across a cgroup,
2942         * we need to enable the list linking each css_set to its
2943         * tasks, and fix up all existing tasks.
2944         */
2945        if (!use_task_css_set_links)
2946                cgroup_enable_task_cg_lists();
2947
2948        read_lock(&css_set_lock);
2949        it->cg_link = &cgrp->css_sets;
2950        cgroup_advance_iter(cgrp, it);
2951}
2952
2953struct task_struct *cgroup_iter_next(struct cgroup *cgrp,
2954                                        struct cgroup_iter *it)
2955{
2956        struct task_struct *res;
2957        struct list_head *l = it->task;
2958        struct cg_cgroup_link *link;
2959
2960        /* If the iterator cg is NULL, we have no tasks */
2961        if (!it->cg_link)
2962                return NULL;
2963        res = list_entry(l, struct task_struct, cg_list);
2964        /* Advance iterator to find next entry */
2965        l = l->next;
2966        link = list_entry(it->cg_link, struct cg_cgroup_link, cgrp_link_list);
2967        if (l == &link->cg->tasks) {
2968                /* We reached the end of this task list - move on to
2969                 * the next cg_cgroup_link */
2970                cgroup_advance_iter(cgrp, it);
2971        } else {
2972                it->task = l;
2973        }
2974        return res;
2975}
2976
2977void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it)
2978        __releases(css_set_lock)
2979{
2980        read_unlock(&css_set_lock);
2981}
2982
2983static inline int started_after_time(struct task_struct *t1,
2984                                     struct timespec *time,
2985                                     struct task_struct *t2)
2986{
2987        int start_diff = timespec_compare(&t1->start_time, time);
2988        if (start_diff > 0) {
2989                return 1;
2990        } else if (start_diff < 0) {
2991                return 0;
2992        } else {
2993                /*
2994                 * Arbitrarily, if two processes started at the same
2995                 * time, we'll say that the lower pointer value
2996                 * started first. Note that t2 may have exited by now
2997                 * so this may not be a valid pointer any longer, but
2998                 * that's fine - it still serves to distinguish
2999                 * between two tasks started (effectively) simultaneously.
3000                 */
3001                return t1 > t2;
3002        }
3003}
3004
3005/*
3006 * This function is a callback from heap_insert() and is used to order
3007 * the heap.
3008 * In this case we order the heap in descending task start time.
3009 */
3010static inline int started_after(void *p1, void *p2)
3011{
3012        struct task_struct *t1 = p1;
3013        struct task_struct *t2 = p2;
3014        return started_after_time(t1, &t2->start_time, t2);
3015}
3016
3017/**
3018 * cgroup_scan_tasks - iterate though all the tasks in a cgroup
3019 * @scan: struct cgroup_scanner containing arguments for the scan
3020 *
3021 * Arguments include pointers to callback functions test_task() and
3022 * process_task().
3023 * Iterate through all the tasks in a cgroup, calling test_task() for each,
3024 * and if it returns true, call process_task() for it also.
3025 * The test_task pointer may be NULL, meaning always true (select all tasks).
3026 * Effectively duplicates cgroup_iter_{start,next,end}()
3027 * but does not lock css_set_lock for the call to process_task().
3028 * The struct cgroup_scanner may be embedded in any structure of the caller's
3029 * creation.
3030 * It is guaranteed that process_task() will act on every task that
3031 * is a member of the cgroup for the duration of this call. This
3032 * function may or may not call process_task() for tasks that exit
3033 * or move to a different cgroup during the call, or are forked or
3034 * move into the cgroup during the call.
3035 *
3036 * Note that test_task() may be called with locks held, and may in some
3037 * situations be called multiple times for the same task, so it should
3038 * be cheap.
3039 * If the heap pointer in the struct cgroup_scanner is non-NULL, a heap has been
3040 * pre-allocated and will be used for heap operations (and its "gt" member will
3041 * be overwritten), else a temporary heap will be used (allocation of which
3042 * may cause this function to fail).
3043 */
3044int cgroup_scan_tasks(struct cgroup_scanner *scan)
3045{
3046        int retval, i;
3047        struct cgroup_iter it;
3048        struct task_struct *p, *dropped;
3049        /* Never dereference latest_task, since it's not refcounted */
3050        struct task_struct *latest_task = NULL;
3051        struct ptr_heap tmp_heap;
3052        struct ptr_heap *heap;
3053        struct timespec latest_time = { 0, 0 };
3054
3055        if (scan->heap) {
3056                /* The caller supplied our heap and pre-allocated its memory */
3057                heap = scan->heap;
3058                heap->gt = &started_after;
3059        } else {
3060                /* We need to allocate our own heap memory */
3061                heap = &tmp_heap;
3062                retval = heap_init(heap, PAGE_SIZE, GFP_KERNEL, &started_after);
3063                if (retval)
3064                        /* cannot allocate the heap */
3065                        return retval;
3066        }
3067
3068 again:
3069        /*
3070         * Scan tasks in the cgroup, using the scanner's "test_task" callback
3071         * to determine which are of interest, and using the scanner's
3072         * "process_task" callback to process any of them that need an update.
3073         * Since we don't want to hold any locks during the task updates,
3074         * gather tasks to be processed in a heap structure.
3075         * The heap is sorted by descending task start time.
3076         * If the statically-sized heap fills up, we overflow tasks that
3077         * started later, and in future iterations only consider tasks that
3078         * started after the latest task in the previous pass. This
3079         * guarantees forward progress and that we don't miss any tasks.
3080         */
3081        heap->size = 0;
3082        cgroup_iter_start(scan->cg, &it);
3083        while ((p = cgroup_iter_next(scan->cg, &it))) {
3084                /*
3085                 * Only affect tasks that qualify per the caller's callback,
3086                 * if he provided one
3087                 */
3088                if (scan->test_task && !scan->test_task(p, scan))
3089                        continue;
3090                /*
3091                 * Only process tasks that started after the last task
3092                 * we processed
3093                 */
3094                if (!started_after_time(p, &latest_time, latest_task))
3095                        continue;
3096                dropped = heap_insert(heap, p);
3097                if (dropped == NULL) {
3098                        /*
3099                         * The new task was inserted; the heap wasn't
3100                         * previously full
3101                         */
3102                        get_task_struct(p);
3103                } else if (dropped != p) {
3104                        /*
3105                         * The new task was inserted, and pushed out a
3106                         * different task
3107                         */
3108                        get_task_struct(p);
3109                        put_task_struct(dropped);
3110                }
3111                /*
3112                 * Else the new task was newer than anything already in
3113                 * the heap and wasn't inserted
3114                 */
3115        }
3116        cgroup_iter_end(scan->cg, &it);
3117
3118        if (heap->size) {
3119                for (i = 0; i < heap->size; i++) {
3120                        struct task_struct *q = heap->ptrs[i];
3121                        if (i == 0) {
3122                                latest_time = q->start_time;
3123                                latest_task = q;
3124                        }
3125                        /* Process the task per the caller's callback */
3126                        scan->process_task(q, scan);
3127                        put_task_struct(q);
3128                }
3129                /*
3130                 * If we had to process any tasks at all, scan again
3131                 * in case some of them were in the middle of forking
3132                 * children that didn't get processed.
3133                 * Not the most efficient way to do it, but it avoids
3134                 * having to take callback_mutex in the fork path
3135                 */
3136                goto again;
3137        }
3138        if (heap == &tmp_heap)
3139                heap_free(&tmp_heap);
3140        return 0;
3141}
3142
3143/*
3144 * Stuff for reading the 'tasks'/'procs' files.
3145 *
3146 * Reading this file can return large amounts of data if a cgroup has
3147 * *lots* of attached tasks. So it may need several calls to read(),
3148 * but we cannot guarantee that the information we produce is correct
3149 * unless we produce it entirely atomically.
3150 *
3151 */
3152
3153/* which pidlist file are we talking about? */
3154enum cgroup_filetype {
3155        CGROUP_FILE_PROCS,
3156        CGROUP_FILE_TASKS,
3157};
3158
3159/*
3160 * A pidlist is a list of pids that virtually represents the contents of one
3161 * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists,
3162 * a pair (one each for procs, tasks) for each pid namespace that's relevant
3163 * to the cgroup.
3164 */
3165struct cgroup_pidlist {
3166        /*
3167         * used to find which pidlist is wanted. doesn't change as long as
3168         * this particular list stays in the list.
3169        */
3170        struct { enum cgroup_filetype type; struct pid_namespace *ns; } key;
3171        /* array of xids */
3172        pid_t *list;
3173        /* how many elements the above list has */
3174        int length;
3175        /* how many files are using the current array */
3176        int use_count;
3177        /* each of these stored in a list by its cgroup */
3178        struct list_head links;
3179        /* pointer to the cgroup we belong to, for list removal purposes */
3180        struct cgroup *owner;
3181        /* protects the other fields */
3182        struct rw_semaphore mutex;
3183};
3184
3185/*
3186 * The following two functions "fix" the issue where there are more pids
3187 * than kmalloc will give memory for; in such cases, we use vmalloc/vfree.
3188 * TODO: replace with a kernel-wide solution to this problem
3189 */
3190#define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2))
3191static void *pidlist_allocate(int count)
3192{
3193        if (PIDLIST_TOO_LARGE(count))
3194                return vmalloc(count * sizeof(pid_t));
3195        else
3196                return kmalloc(count * sizeof(pid_t), GFP_KERNEL);
3197}
3198static void pidlist_free(void *p)
3199{
3200        if (is_vmalloc_addr(p))
3201                vfree(p);
3202        else
3203                kfree(p);
3204}
3205static void *pidlist_resize(void *p, int newcount)
3206{
3207        void *newlist;
3208        /* note: if new alloc fails, old p will still be valid either way */
3209        if (is_vmalloc_addr(p)) {
3210                newlist = vmalloc(newcount * sizeof(pid_t));
3211                if (!newlist)
3212                        return NULL;
3213                memcpy(newlist, p, newcount * sizeof(pid_t));
3214                vfree(p);
3215        } else {
3216                newlist = krealloc(p, newcount * sizeof(pid_t), GFP_KERNEL);
3217        }
3218        return newlist;
3219}
3220
3221/*
3222 * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
3223 * If the new stripped list is sufficiently smaller and there's enough memory
3224 * to allocate a new buffer, will let go of the unneeded memory. Returns the
3225 * number of unique elements.
3226 */
3227/* is the size difference enough that we should re-allocate the array? */
3228#define PIDLIST_REALLOC_DIFFERENCE(old, new) ((old) - PAGE_SIZE >= (new))
3229static int pidlist_uniq(pid_t **p, int length)
3230{
3231        int src, dest = 1;
3232        pid_t *list = *p;
3233        pid_t *newlist;
3234
3235        /*
3236         * we presume the 0th element is unique, so i starts at 1. trivial
3237         * edge cases first; no work needs to be done for either
3238         */
3239        if (length == 0 || length == 1)
3240                return length;
3241        /* src and dest walk down the list; dest counts unique elements */
3242        for (src = 1; src < length; src++) {
3243                /* find next unique element */
3244                while (list[src] == list[src-1]) {
3245                        src++;
3246                        if (src == length)
3247                                goto after;
3248                }
3249                /* dest always points to where the next unique element goes */
3250                list[dest] = list[src];
3251                dest++;
3252        }
3253after:
3254        /*
3255         * if the length difference is large enough, we want to allocate a
3256         * smaller buffer to save memory. if this fails due to out of memory,
3257         * we'll just stay with what we've got.
3258         */
3259        if (PIDLIST_REALLOC_DIFFERENCE(length, dest)) {
3260                newlist = pidlist_resize(list, dest);
3261                if (newlist)
3262                        *p = newlist;
3263        }
3264        return dest;
3265}
3266
3267static int cmppid(const void *a, const void *b)
3268{
3269        return *(pid_t *)a - *(pid_t *)b;
3270}
3271
3272/*
3273 * find the appropriate pidlist for our purpose (given procs vs tasks)
3274 * returns with the lock on that pidlist already held, and takes care
3275 * of the use count, or returns NULL with no locks held if we're out of
3276 * memory.
3277 */
3278static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp,
3279                                                  enum cgroup_filetype type)
3280{
3281        struct cgroup_pidlist *l;
3282        /* don't need task_nsproxy() if we're looking at ourself */
3283        struct pid_namespace *ns = current->nsproxy->pid_ns;
3284
3285        /*
3286         * We can't drop the pidlist_mutex before taking the l->mutex in case
3287         * the last ref-holder is trying to remove l from the list at the same
3288         * time. Holding the pidlist_mutex precludes somebody taking whichever
3289         * list we find out from under us - compare release_pid_array().
3290         */
3291        mutex_lock(&cgrp->pidlist_mutex);
3292        list_for_each_entry(l, &cgrp->pidlists, links) {
3293                if (l->key.type == type && l->key.ns == ns) {
3294                        /* make sure l doesn't vanish out from under us */
3295                        down_write(&l->mutex);
3296                        mutex_unlock(&cgrp->pidlist_mutex);
3297                        return l;
3298                }
3299        }
3300        /* entry not found; create a new one */
3301        l = kmalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL);
3302        if (!l) {
3303                mutex_unlock(&cgrp->pidlist_mutex);
3304                return l;
3305        }
3306        init_rwsem(&l->mutex);
3307        down_write(&l->mutex);
3308        l->key.type = type;
3309        l->key.ns = get_pid_ns(ns);
3310        l->use_count = 0; /* don't increment here */
3311        l->list = NULL;
3312        l->owner = cgrp;
3313        list_add(&l->links, &cgrp->pidlists);
3314        mutex_unlock(&cgrp->pidlist_mutex);
3315        return l;
3316}
3317
3318/*
3319 * Load a cgroup's pidarray with either procs' tgids or tasks' pids
3320 */
3321static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type,
3322                              struct cgroup_pidlist **lp)
3323{
3324        pid_t *array;
3325        int length;
3326        int pid, n = 0; /* used for populating the array */
3327        struct cgroup_iter it;
3328        struct task_struct *tsk;
3329        struct cgroup_pidlist *l;
3330
3331        /*
3332         * If cgroup gets more users after we read count, we won't have
3333         * enough space - tough.  This race is indistinguishable to the
3334         * caller from the case that the additional cgroup users didn't
3335         * show up until sometime later on.
3336         */
3337        length = cgroup_task_count(cgrp);
3338        array = pidlist_allocate(length);
3339        if (!array)
3340                return -ENOMEM;
3341        /* now, populate the array */
3342        cgroup_iter_start(cgrp, &it);
3343        while ((tsk = cgroup_iter_next(cgrp, &it))) {
3344                if (unlikely(n == length))
3345                        break;
3346                /* get tgid or pid for procs or tasks file respectively */
3347                if (type == CGROUP_FILE_PROCS)
3348                        pid = task_tgid_vnr(tsk);
3349                else
3350                        pid = task_pid_vnr(tsk);
3351                if (pid > 0) /* make sure to only use valid results */
3352                        array[n++] = pid;
3353        }
3354        cgroup_iter_end(cgrp, &it);
3355        length = n;
3356        /* now sort & (if procs) strip out duplicates */
3357        sort(array, length, sizeof(pid_t), cmppid, NULL);
3358        if (type == CGROUP_FILE_PROCS)
3359                length = pidlist_uniq(&array, length);
3360        l = cgroup_pidlist_find(cgrp, type);
3361        if (!l) {
3362                pidlist_free(array);
3363                return -ENOMEM;
3364        }
3365        /* store array, freeing old if necessary - lock already held */
3366        pidlist_free(l->list);
3367        l->list = array;
3368        l->length = length;
3369        l->use_count++;
3370        up_write(&l->mutex);
3371        *lp = l;
3372        return 0;
3373}
3374
3375/**
3376 * cgroupstats_build - build and fill cgroupstats
3377 * @stats: cgroupstats to fill information into
3378 * @dentry: A dentry entry belonging to the cgroup for which stats have
3379 * been requested.
3380 *
3381 * Build and fill cgroupstats so that taskstats can export it to user
3382 * space.
3383 */
3384int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry)
3385{
3386        int ret = -EINVAL;
3387        struct cgroup *cgrp;
3388        struct cgroup_iter it;
3389        struct task_struct *tsk;
3390
3391        /*
3392         * Validate dentry by checking the superblock operations,
3393         * and make sure it's a directory.
3394         */
3395        if (dentry->d_sb->s_op != &cgroup_ops ||
3396            !S_ISDIR(dentry->d_inode->i_mode))
3397                 goto err;
3398
3399        ret = 0;
3400        cgrp = dentry->d_fsdata;
3401
3402        cgroup_iter_start(cgrp, &it);
3403        while ((tsk = cgroup_iter_next(cgrp, &it))) {
3404                switch (tsk->state) {
3405                case TASK_RUNNING:
3406                        stats->nr_running++;
3407                        break;
3408                case TASK_INTERRUPTIBLE:
3409                        stats->nr_sleeping++;
3410                        break;
3411                case TASK_UNINTERRUPTIBLE:
3412                        stats->nr_uninterruptible++;
3413                        break;
3414                case TASK_STOPPED:
3415                        stats->nr_stopped++;
3416                        break;
3417                default:
3418                        if (delayacct_is_task_waiting_on_io(tsk))
3419                                stats->nr_io_wait++;
3420                        break;
3421                }
3422        }
3423        cgroup_iter_end(cgrp, &it);
3424
3425err:
3426        return ret;
3427}
3428
3429
3430/*
3431 * seq_file methods for the tasks/procs files. The seq_file position is the
3432 * next pid to display; the seq_file iterator is a pointer to the pid
3433 * in the cgroup->l->list array.
3434 */
3435
3436static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos)
3437{
3438        /*
3439         * Initially we receive a position value that corresponds to
3440         * one more than the last pid shown (or 0 on the first call or
3441         * after a seek to the start). Use a binary-search to find the
3442         * next pid to display, if any
3443         */
3444        struct cgroup_pidlist *l = s->private;
3445        int index = 0, pid = *pos;
3446        int *iter;
3447
3448        down_read(&l->mutex);
3449        if (pid) {
3450                int end = l->length;
3451
3452                while (index < end) {
3453                        int mid = (index + end) / 2;
3454                        if (l->list[mid] == pid) {
3455                                index = mid;
3456                                break;
3457                        } else if (l->list[mid] <= pid)
3458                                index = mid + 1;
3459                        else
3460                                end = mid;
3461                }
3462        }
3463        /* If we're off the end of the array, we're done */
3464        if (index >= l->length)
3465                return NULL;
3466        /* Update the abstract position to be the actual pid that we found */
3467        iter = l->list + index;
3468        *pos = *iter;
3469        return iter;
3470}
3471
3472static void cgroup_pidlist_stop(struct seq_file *s, void *v)
3473{
3474        struct cgroup_pidlist *l = s->private;
3475        up_read(&l->mutex);
3476}
3477
3478static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos)
3479{
3480        struct cgroup_pidlist *l = s->private;
3481        pid_t *p = v;
3482        pid_t *end = l->list + l->length;
3483        /*
3484         * Advance to the next pid in the array. If this goes off the
3485         * end, we're done
3486         */
3487        p++;
3488        if (p >= end) {
3489                return NULL;
3490        } else {
3491                *pos = *p;
3492                return p;
3493        }
3494}
3495
3496static int cgroup_pidlist_show(struct seq_file *s, void *v)
3497{
3498        return seq_printf(s, "%d\n", *(int *)v);
3499}
3500
3501/*
3502 * seq_operations functions for iterating on pidlists through seq_file -
3503 * independent of whether it's tasks or procs
3504 */
3505static const struct seq_operations cgroup_pidlist_seq_operations = {
3506        .start = cgroup_pidlist_start,
3507        .stop = cgroup_pidlist_stop,
3508        .next = cgroup_pidlist_next,
3509        .show = cgroup_pidlist_show,
3510};
3511
3512static void cgroup_release_pid_array(struct cgroup_pidlist *l)
3513{
3514        /*
3515         * the case where we're the last user of this particular pidlist will
3516         * have us remove it from the cgroup's list, which entails taking the
3517         * mutex. since in pidlist_find the pidlist->lock depends on cgroup->
3518         * pidlist_mutex, we have to take pidlist_mutex first.
3519         */
3520        mutex_lock(&l->owner->pidlist_mutex);
3521        down_write(&l->mutex);
3522        BUG_ON(!l->use_count);
3523        if (!--l->use_count) {
3524                /* we're the last user if refcount is 0; remove and free */
3525                list_del(&l->links);
3526                mutex_unlock(&l->owner->pidlist_mutex);
3527                pidlist_free(l->list);
3528                put_pid_ns(l->key.ns);
3529                up_write(&l->mutex);
3530                kfree(l);
3531                return;
3532        }
3533        mutex_unlock(&l->owner->pidlist_mutex);
3534        up_write(&l->mutex);
3535}
3536
3537static int cgroup_pidlist_release(struct inode *inode, struct file *file)
3538{
3539        struct cgroup_pidlist *l;
3540        if (!(file->f_mode & FMODE_READ))
3541                return 0;
3542        /*
3543         * the seq_file will only be initialized if the file was opened for
3544         * reading; hence we check if it's not null only in that case.
3545         */
3546        l = ((struct seq_file *)file->private_data)->private;
3547        cgroup_release_pid_array(l);
3548        return seq_release(inode, file);
3549}
3550
3551static const struct file_operations cgroup_pidlist_operations = {
3552        .read = seq_read,
3553        .llseek = seq_lseek,
3554        .write = cgroup_file_write,
3555        .release = cgroup_pidlist_release,
3556};
3557
3558/*
3559 * The following functions handle opens on a file that displays a pidlist
3560 * (tasks or procs). Prepare an array of the process/thread IDs of whoever's
3561 * in the cgroup.
3562 */
3563/* helper function for the two below it */
3564static int cgroup_pidlist_open(struct file *file, enum cgroup_filetype type)
3565{
3566        struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
3567        struct cgroup_pidlist *l;
3568        int retval;
3569
3570        /* Nothing to do for write-only files */
3571        if (!(file->f_mode & FMODE_READ))
3572                return 0;
3573
3574        /* have the array populated */
3575        retval = pidlist_array_load(cgrp, type, &l);
3576        if (retval)
3577                return retval;
3578        /* configure file information */
3579        file->f_op = &cgroup_pidlist_operations;
3580
3581        retval = seq_open(file, &cgroup_pidlist_seq_operations);
3582        if (retval) {
3583                cgroup_release_pid_array(l);
3584                return retval;
3585        }
3586        ((struct seq_file *)file->private_data)->private = l;
3587        return 0;
3588}
3589static int cgroup_tasks_open(struct inode *unused, struct file *file)
3590{
3591        return cgroup_pidlist_open(file, CGROUP_FILE_TASKS);
3592}
3593static int cgroup_procs_open(struct inode *unused, struct file *file)
3594{
3595        return cgroup_pidlist_open(file, CGROUP_FILE_PROCS);
3596}
3597
3598static u64 cgroup_read_notify_on_release(struct cgroup *cgrp,
3599                                            struct cftype *cft)
3600{
3601        return notify_on_release(cgrp);
3602}
3603
3604static int cgroup_write_notify_on_release(struct cgroup *cgrp,
3605                                          struct cftype *cft,
3606                                          u64 val)
3607{
3608        clear_bit(CGRP_RELEASABLE, &cgrp->flags);
3609        if (val)
3610                set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
3611        else
3612                clear_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
3613        return 0;
3614}
3615
3616/*
3617 * Unregister event and free resources.
3618 *
3619 * Gets called from workqueue.
3620 */
3621static void cgroup_event_remove(struct work_struct *work)
3622{
3623        struct cgroup_event *event = container_of(work, struct cgroup_event,
3624                        remove);
3625        struct cgroup *cgrp = event->cgrp;
3626
3627        event->cft->unregister_event(cgrp, event->cft, event->eventfd);
3628
3629        eventfd_ctx_put(event->eventfd);
3630        kfree(event);
3631        dput(cgrp->dentry);
3632}
3633
3634/*
3635 * Gets called on POLLHUP on eventfd when user closes it.
3636 *
3637 * Called with wqh->lock held and interrupts disabled.
3638 */
3639static int cgroup_event_wake(wait_queue_t *wait, unsigned mode,
3640                int sync, void *key)
3641{
3642        struct cgroup_event *event = container_of(wait,
3643                        struct cgroup_event, wait);
3644        struct cgroup *cgrp = event->cgrp;
3645        unsigned long flags = (unsigned long)key;
3646
3647        if (flags & POLLHUP) {
3648                __remove_wait_queue(event->wqh, &event->wait);
3649                spin_lock(&cgrp->event_list_lock);
3650                list_del(&event->list);
3651                spin_unlock(&cgrp->event_list_lock);
3652                /*
3653                 * We are in atomic context, but cgroup_event_remove() may
3654                 * sleep, so we have to call it in workqueue.
3655                 */
3656                schedule_work(&event->remove);
3657        }
3658
3659        return 0;
3660}
3661
3662static void cgroup_event_ptable_queue_proc(struct file *file,
3663                wait_queue_head_t *wqh, poll_table *pt)
3664{
3665        struct cgroup_event *event = container_of(pt,
3666                        struct cgroup_event, pt);
3667
3668        event->wqh = wqh;
3669        add_wait_queue(wqh, &event->wait);
3670}
3671
3672/*
3673 * Parse input and register new cgroup event handler.
3674 *
3675 * Input must be in format '<event_fd> <control_fd> <args>'.
3676 * Interpretation of args is defined by control file implementation.
3677 */
3678static int cgroup_write_event_control(struct cgroup *cgrp, struct cftype *cft,
3679                                      const char *buffer)
3680{
3681        struct cgroup_event *event = NULL;
3682        unsigned int efd, cfd;
3683        struct file *efile = NULL;
3684        struct file *cfile = NULL;
3685        char *endp;
3686        int ret;
3687
3688        efd = simple_strtoul(buffer, &endp, 10);
3689        if (*endp != ' ')
3690                return -EINVAL;
3691        buffer = endp + 1;
3692
3693        cfd = simple_strtoul(buffer, &endp, 10);
3694        if ((*endp != ' ') && (*endp != '\0'))
3695                return -EINVAL;
3696        buffer = endp + 1;
3697
3698        event = kzalloc(sizeof(*event), GFP_KERNEL);
3699        if (!event)
3700                return -ENOMEM;
3701        event->cgrp = cgrp;
3702        INIT_LIST_HEAD(&event->list);
3703        init_poll_funcptr(&event->pt, cgroup_event_ptable_queue_proc);
3704        init_waitqueue_func_entry(&event->wait, cgroup_event_wake);
3705        INIT_WORK(&event->remove, cgroup_event_remove);
3706
3707        efile = eventfd_fget(efd);
3708        if (IS_ERR(efile)) {
3709                ret = PTR_ERR(efile);
3710                goto fail;
3711        }
3712
3713        event->eventfd = eventfd_ctx_fileget(efile);
3714        if (IS_ERR(event->eventfd)) {
3715                ret = PTR_ERR(event->eventfd);
3716                goto fail;
3717        }
3718
3719        cfile = fget(cfd);
3720        if (!cfile) {
3721                ret = -EBADF;
3722                goto fail;
3723        }
3724
3725        /* the process need read permission on control file */
3726        /* AV: shouldn't we check that it's been opened for read instead? */
3727        ret = inode_permission(cfile->f_path.dentry->d_inode, MAY_READ);
3728        if (ret < 0)
3729                goto fail;
3730
3731        event->cft = __file_cft(cfile);
3732        if (IS_ERR(event->cft)) {
3733                ret = PTR_ERR(event->cft);
3734                goto fail;
3735        }
3736
3737        if (!event->cft->register_event || !event->cft->unregister_event) {
3738                ret = -EINVAL;
3739                goto fail;
3740        }
3741
3742        ret = event->cft->register_event(cgrp, event->cft,
3743                        event->eventfd, buffer);
3744        if (ret)
3745                goto fail;
3746
3747        if (efile->f_op->poll(efile, &event->pt) & POLLHUP) {
3748                event->cft->unregister_event(cgrp, event->cft, event->eventfd);
3749                ret = 0;
3750                goto fail;
3751        }
3752
3753        /*
3754         * Events should be removed after rmdir of cgroup directory, but before
3755         * destroying subsystem state objects. Let's take reference to cgroup
3756         * directory dentry to do that.
3757         */
3758        dget(cgrp->dentry);
3759
3760        spin_lock(&cgrp->event_list_lock);
3761        list_add(&event->list, &cgrp->event_list);
3762        spin_unlock(&cgrp->event_list_lock);
3763
3764        fput(cfile);
3765        fput(efile);
3766
3767        return 0;
3768
3769fail:
3770        if (cfile)
3771                fput(cfile);
3772
3773        if (event && event->eventfd && !IS_ERR(event->eventfd))
3774                eventfd_ctx_put(event->eventfd);
3775
3776        if (!IS_ERR_OR_NULL(efile))
3777                fput(efile);
3778
3779        kfree(event);
3780
3781        return ret;
3782}
3783
3784static u64 cgroup_clone_children_read(struct cgroup *cgrp,
3785                                    struct cftype *cft)
3786{
3787        return clone_children(cgrp);
3788}
3789
3790static int cgroup_clone_children_write(struct cgroup *cgrp,
3791                                     struct cftype *cft,
3792                                     u64 val)
3793{
3794        if (val)
3795                set_bit(CGRP_CLONE_CHILDREN, &cgrp->flags);
3796        else
3797                clear_bit(CGRP_CLONE_CHILDREN, &cgrp->flags);
3798        return 0;
3799}
3800
3801/*
3802 * for the common functions, 'private' gives the type of file
3803 */
3804/* for hysterical raisins, we can't put this on the older files */
3805#define CGROUP_FILE_GENERIC_PREFIX "cgroup."
3806static struct cftype files[] = {
3807        {
3808                .name = "tasks",
3809                .open = cgroup_tasks_open,
3810                .write_u64 = cgroup_tasks_write,
3811                .release = cgroup_pidlist_release,
3812                .mode = S_IRUGO | S_IWUSR,
3813        },
3814        {
3815                .name = CGROUP_FILE_GENERIC_PREFIX "procs",
3816                .open = cgroup_procs_open,
3817                .write_u64 = cgroup_procs_write,
3818                .release = cgroup_pidlist_release,
3819                .mode = S_IRUGO | S_IWUSR,
3820        },
3821        {
3822                .name = "notify_on_release",
3823                .read_u64 = cgroup_read_notify_on_release,
3824                .write_u64 = cgroup_write_notify_on_release,
3825        },
3826        {
3827                .name = CGROUP_FILE_GENERIC_PREFIX "event_control",
3828                .write_string = cgroup_write_event_control,
3829                .mode = S_IWUGO,
3830        },
3831        {
3832                .name = "cgroup.clone_children",
3833                .read_u64 = cgroup_clone_children_read,
3834                .write_u64 = cgroup_clone_children_write,
3835        },
3836        {
3837                .name = "release_agent",
3838                .flags = CFTYPE_ONLY_ON_ROOT,
3839                .read_seq_string = cgroup_release_agent_show,
3840                .write_string = cgroup_release_agent_write,
3841                .max_write_len = PATH_MAX,
3842        },
3843        { }     /* terminate */
3844};
3845
3846static int cgroup_populate_dir(struct cgroup *cgrp)
3847{
3848        int err;
3849        struct cgroup_subsys *ss;
3850
3851        err = cgroup_addrm_files(cgrp, NULL, files, true);
3852        if (err < 0)
3853                return err;
3854
3855        /* process cftsets of each subsystem */
3856        for_each_subsys(cgrp->root, ss) {
3857                struct cftype_set *set;
3858
3859                list_for_each_entry(set, &ss->cftsets, node)
3860                        cgroup_addrm_files(cgrp, ss, set->cfts, true);
3861        }
3862
3863        /* This cgroup is ready now */
3864        for_each_subsys(cgrp->root, ss) {
3865                struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
3866                /*
3867                 * Update id->css pointer and make this css visible from
3868                 * CSS ID functions. This pointer will be dereferened
3869                 * from RCU-read-side without locks.
3870                 */
3871                if (css->id)
3872                        rcu_assign_pointer(css->id->css, css);
3873        }
3874
3875        return 0;
3876}
3877
3878static void css_dput_fn(struct work_struct *work)
3879{
3880        struct cgroup_subsys_state *css =
3881                container_of(work, struct cgroup_subsys_state, dput_work);
3882        struct dentry *dentry = css->cgroup->dentry;
3883        struct super_block *sb = dentry->d_sb;
3884
3885        atomic_inc(&sb->s_active);
3886        dput(dentry);
3887        deactivate_super(sb);
3888}
3889
3890static void init_cgroup_css(struct cgroup_subsys_state *css,
3891                               struct cgroup_subsys *ss,
3892                               struct cgroup *cgrp)
3893{
3894        css->cgroup = cgrp;
3895        atomic_set(&css->refcnt, 1);
3896        css->flags = 0;
3897        css->id = NULL;
3898        if (cgrp == dummytop)
3899                set_bit(CSS_ROOT, &css->flags);
3900        BUG_ON(cgrp->subsys[ss->subsys_id]);
3901        cgrp->subsys[ss->subsys_id] = css;
3902
3903        /*
3904         * If !clear_css_refs, css holds an extra ref to @cgrp->dentry
3905         * which is put on the last css_put().  dput() requires process
3906         * context, which css_put() may be called without.  @css->dput_work
3907         * will be used to invoke dput() asynchronously from css_put().
3908         */
3909        INIT_WORK(&css->dput_work, css_dput_fn);
3910        if (ss->__DEPRECATED_clear_css_refs)
3911                set_bit(CSS_CLEAR_CSS_REFS, &css->flags);
3912}
3913
3914/*
3915 * cgroup_create - create a cgroup
3916 * @parent: cgroup that will be parent of the new cgroup
3917 * @dentry: dentry of the new cgroup
3918 * @mode: mode to set on new inode
3919 *
3920 * Must be called with the mutex on the parent inode held
3921 */
3922static long cgroup_create(struct cgroup *parent, struct dentry *dentry,
3923                             umode_t mode)
3924{
3925        struct cgroup *cgrp;
3926        struct cgroupfs_root *root = parent->root;
3927        int err = 0;
3928        struct cgroup_subsys *ss;
3929        struct super_block *sb = root->sb;
3930
3931        cgrp = kzalloc(sizeof(*cgrp), GFP_KERNEL);
3932        if (!cgrp)
3933                return -ENOMEM;
3934
3935        /* Grab a reference on the superblock so the hierarchy doesn't
3936         * get deleted on unmount if there are child cgroups.  This
3937         * can be done outside cgroup_mutex, since the sb can't
3938         * disappear while someone has an open control file on the
3939         * fs */
3940        atomic_inc(&sb->s_active);
3941
3942        mutex_lock(&cgroup_mutex);
3943
3944        init_cgroup_housekeeping(cgrp);
3945
3946        cgrp->parent = parent;
3947        cgrp->root = parent->root;
3948        cgrp->top_cgroup = parent->top_cgroup;
3949
3950        if (notify_on_release(parent))
3951                set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
3952
3953        if (clone_children(parent))
3954                set_bit(CGRP_CLONE_CHILDREN, &cgrp->flags);
3955
3956        for_each_subsys(root, ss) {
3957                struct cgroup_subsys_state *css = ss->create(cgrp);
3958
3959                if (IS_ERR(css)) {
3960                        err = PTR_ERR(css);
3961                        goto err_destroy;
3962                }
3963                init_cgroup_css(css, ss, cgrp);
3964                if (ss->use_id) {
3965                        err = alloc_css_id(ss, parent, cgrp);
3966                        if (err)
3967                                goto err_destroy;
3968                }
3969                /* At error, ->destroy() callback has to free assigned ID. */
3970                if (clone_children(parent) && ss->post_clone)
3971                        ss->post_clone(cgrp);
3972        }
3973
3974        list_add(&cgrp->sibling, &cgrp->parent->children);
3975        root->number_of_cgroups++;
3976
3977        err = cgroup_create_dir(cgrp, dentry, mode);
3978        if (err < 0)
3979                goto err_remove;
3980
3981        /* If !clear_css_refs, each css holds a ref to the cgroup's dentry */
3982        for_each_subsys(root, ss)
3983                if (!ss->__DEPRECATED_clear_css_refs)
3984                        dget(dentry);
3985
3986        /* The cgroup directory was pre-locked for us */
3987        BUG_ON(!mutex_is_locked(&cgrp->dentry->d_inode->i_mutex));
3988
3989        list_add_tail(&cgrp->allcg_node, &root->allcg_list);
3990
3991        err = cgroup_populate_dir(cgrp);
3992        /* If err < 0, we have a half-filled directory - oh well ;) */
3993
3994        mutex_unlock(&cgroup_mutex);
3995        mutex_unlock(&cgrp->dentry->d_inode->i_mutex);
3996
3997        return 0;
3998
3999 err_remove:
4000
4001        list_del(&cgrp->sibling);
4002        root->number_of_cgroups--;
4003
4004 err_destroy:
4005
4006        for_each_subsys(root, ss) {
4007                if (cgrp->subsys[ss->subsys_id])
4008                        ss->destroy(cgrp);
4009        }
4010
4011        mutex_unlock(&cgroup_mutex);
4012
4013        /* Release the reference count that we took on the superblock */
4014        deactivate_super(sb);
4015
4016        kfree(cgrp);
4017        return err;
4018}
4019
4020static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
4021{
4022        struct cgroup *c_parent = dentry->d_parent->d_fsdata;
4023
4024        /* the vfs holds inode->i_mutex already */
4025        return cgroup_create(c_parent, dentry, mode | S_IFDIR);
4026}
4027
4028/*
4029 * Check the reference count on each subsystem. Since we already
4030 * established that there are no tasks in the cgroup, if the css refcount
4031 * is also 1, then there should be no outstanding references, so the
4032 * subsystem is safe to destroy. We scan across all subsystems rather than
4033 * using the per-hierarchy linked list of mounted subsystems since we can
4034 * be called via check_for_release() with no synchronization other than
4035 * RCU, and the subsystem linked list isn't RCU-safe.
4036 */
4037static int cgroup_has_css_refs(struct cgroup *cgrp)
4038{
4039        int i;
4040
4041        /*
4042         * We won't need to lock the subsys array, because the subsystems
4043         * we're concerned about aren't going anywhere since our cgroup root
4044         * has a reference on them.
4045         */
4046        for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
4047                struct cgroup_subsys *ss = subsys[i];
4048                struct cgroup_subsys_state *css;
4049
4050                /* Skip subsystems not present or not in this hierarchy */
4051                if (ss == NULL || ss->root != cgrp->root)
4052                        continue;
4053
4054                css = cgrp->subsys[ss->subsys_id];
4055                /*
4056                 * When called from check_for_release() it's possible
4057                 * that by this point the cgroup has been removed
4058                 * and the css deleted. But a false-positive doesn't
4059                 * matter, since it can only happen if the cgroup
4060                 * has been deleted and hence no longer needs the
4061                 * release agent to be called anyway.
4062                 */
4063                if (css && css_refcnt(css) > 1)
4064                        return 1;
4065        }
4066        return 0;
4067}
4068
4069/*
4070 * Atomically mark all (or else none) of the cgroup's CSS objects as
4071 * CSS_REMOVED. Return true on success, or false if the cgroup has
4072 * busy subsystems. Call with cgroup_mutex held
4073 *
4074 * Depending on whether a subsys has __DEPRECATED_clear_css_refs set or
4075 * not, cgroup removal behaves differently.
4076 *
4077 * If clear is set, css refcnt for the subsystem should be zero before
4078 * cgroup removal can be committed.  This is implemented by
4079 * CGRP_WAIT_ON_RMDIR and retry logic around ->pre_destroy(), which may be
4080 * called multiple times until all css refcnts reach zero and is allowed to
4081 * veto removal on any invocation.  This behavior is deprecated and will be
4082 * removed as soon as the existing user (memcg) is updated.
4083 *
4084 * If clear is not set, each css holds an extra reference to the cgroup's
4085 * dentry and cgroup removal proceeds regardless of css refs.
4086 * ->pre_destroy() will be called at least once and is not allowed to fail.
4087 * On the last put of each css, whenever that may be, the extra dentry ref
4088 * is put so that dentry destruction happens only after all css's are
4089 * released.
4090 */
4091static int cgroup_clear_css_refs(struct cgroup *cgrp)
4092{
4093        struct cgroup_subsys *ss;
4094        unsigned long flags;
4095        bool failed = false;
4096
4097        local_irq_save(flags);
4098
4099        /*
4100         * Block new css_tryget() by deactivating refcnt.  If all refcnts
4101         * for subsystems w/ clear_css_refs set were 1 at the moment of
4102         * deactivation, we succeeded.
4103         */
4104        for_each_subsys(cgrp->root, ss) {
4105                struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
4106
4107                WARN_ON(atomic_read(&css->refcnt) < 0);
4108                atomic_add(CSS_DEACT_BIAS, &css->refcnt);
4109
4110                if (ss->__DEPRECATED_clear_css_refs)
4111                        failed |= css_refcnt(css) != 1;
4112        }
4113
4114        /*
4115         * If succeeded, set REMOVED and put all the base refs; otherwise,
4116         * restore refcnts to positive values.  Either way, all in-progress
4117         * css_tryget() will be released.
4118         */
4119        for_each_subsys(cgrp->root, ss) {
4120                struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
4121
4122                if (!failed) {
4123                        set_bit(CSS_REMOVED, &css->flags);
4124                        css_put(css);
4125                } else {
4126                        atomic_sub(CSS_DEACT_BIAS, &css->refcnt);
4127                }
4128        }
4129
4130        local_irq_restore(flags);
4131        return !failed;
4132}
4133
4134static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry)
4135{
4136        struct cgroup *cgrp = dentry->d_fsdata;
4137        struct dentry *d;
4138        struct cgroup *parent;
4139        DEFINE_WAIT(wait);
4140        struct cgroup_event *event, *tmp;
4141        int ret;
4142
4143        /* the vfs holds both inode->i_mutex already */
4144again:
4145        mutex_lock(&cgroup_mutex);
4146        if (atomic_read(&cgrp->count) != 0) {
4147                mutex_unlock(&cgroup_mutex);
4148                return -EBUSY;
4149        }
4150        if (!list_empty(&cgrp->children)) {
4151                mutex_unlock(&cgroup_mutex);
4152                return -EBUSY;
4153        }
4154        mutex_unlock(&cgroup_mutex);
4155
4156        /*
4157         * In general, subsystem has no css->refcnt after pre_destroy(). But
4158         * in racy cases, subsystem may have to get css->refcnt after
4159         * pre_destroy() and it makes rmdir return with -EBUSY. This sometimes
4160         * make rmdir return -EBUSY too often. To avoid that, we use waitqueue
4161         * for cgroup's rmdir. CGRP_WAIT_ON_RMDIR is for synchronizing rmdir
4162         * and subsystem's reference count handling. Please see css_get/put
4163         * and css_tryget() and cgroup_wakeup_rmdir_waiter() implementation.
4164         */
4165        set_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
4166
4167        /*
4168         * Call pre_destroy handlers of subsys. Notify subsystems
4169         * that rmdir() request comes.
4170         */
4171        ret = cgroup_call_pre_destroy(cgrp);
4172        if (ret) {
4173                clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
4174                return ret;
4175        }
4176
4177        mutex_lock(&cgroup_mutex);
4178        parent = cgrp->parent;
4179        if (atomic_read(&cgrp->count) || !list_empty(&cgrp->children)) {
4180                clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
4181                mutex_unlock(&cgroup_mutex);
4182                return -EBUSY;
4183        }
4184        prepare_to_wait(&cgroup_rmdir_waitq, &wait, TASK_INTERRUPTIBLE);
4185        if (!cgroup_clear_css_refs(cgrp)) {
4186                mutex_unlock(&cgroup_mutex);
4187                /*
4188                 * Because someone may call cgroup_wakeup_rmdir_waiter() before
4189                 * prepare_to_wait(), we need to check this flag.
4190                 */
4191                if (test_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags))
4192                        schedule();
4193                finish_wait(&cgroup_rmdir_waitq, &wait);
4194                clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
4195                if (signal_pending(current))
4196                        return -EINTR;
4197                goto again;
4198        }
4199        /* NO css_tryget() can success after here. */
4200        finish_wait(&cgroup_rmdir_waitq, &wait);
4201        clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
4202
4203        raw_spin_lock(&release_list_lock);
4204        set_bit(CGRP_REMOVED, &cgrp->flags);
4205        if (!list_empty(&cgrp->release_list))
4206                list_del_init(&cgrp->release_list);
4207        raw_spin_unlock(&release_list_lock);
4208
4209        /* delete this cgroup from parent->children */
4210        list_del_init(&cgrp->sibling);
4211
4212        list_del_init(&cgrp->allcg_node);
4213
4214        d = dget(cgrp->dentry);
4215
4216        cgroup_d_remove_dir(d);
4217        dput(d);
4218
4219        set_bit(CGRP_RELEASABLE, &parent->flags);
4220        check_for_release(parent);
4221
4222        /*
4223         * Unregister events and notify userspace.
4224         * Notify userspace about cgroup removing only after rmdir of cgroup
4225         * directory to avoid race between userspace and kernelspace
4226         */
4227        spin_lock(&cgrp->event_list_lock);
4228        list_for_each_entry_safe(event, tmp, &cgrp->event_list, list) {
4229                list_del(&event->list);
4230                remove_wait_queue(event->wqh, &event->wait);
4231                eventfd_signal(event->eventfd, 1);
4232                schedule_work(&event->remove);
4233        }
4234        spin_unlock(&cgrp->event_list_lock);
4235
4236        mutex_unlock(&cgroup_mutex);
4237        return 0;
4238}
4239
4240static void __init_or_module cgroup_init_cftsets(struct cgroup_subsys *ss)
4241{
4242        INIT_LIST_HEAD(&ss->cftsets);
4243
4244        /*
4245         * base_cftset is embedded in subsys itself, no need to worry about
4246         * deregistration.
4247         */
4248        if (ss->base_cftypes) {
4249                ss->base_cftset.cfts = ss->base_cftypes;
4250                list_add_tail(&ss->base_cftset.node, &ss->cftsets);
4251        }
4252}
4253
4254static void __init cgroup_init_subsys(struct cgroup_subsys *ss)
4255{
4256        struct cgroup_subsys_state *css;
4257
4258        printk(KERN_INFO "Initializing cgroup subsys %s\n", ss->name);
4259
4260        /* init base cftset */
4261        cgroup_init_cftsets(ss);
4262
4263        /* Create the top cgroup state for this subsystem */
4264        list_add(&ss->sibling, &rootnode.subsys_list);
4265        ss->root = &rootnode;
4266        css = ss->create(dummytop);
4267        /* We don't handle early failures gracefully */
4268        BUG_ON(IS_ERR(css));
4269        init_cgroup_css(css, ss, dummytop);
4270
4271        /* Update the init_css_set to contain a subsys
4272         * pointer to this state - since the subsystem is
4273         * newly registered, all tasks and hence the
4274         * init_css_set is in the subsystem's top cgroup. */
4275        init_css_set.subsys[ss->subsys_id] = dummytop->subsys[ss->subsys_id];
4276
4277        need_forkexit_callback |= ss->fork || ss->exit;
4278
4279        /* At system boot, before all subsystems have been
4280         * registered, no tasks have been forked, so we don't
4281         * need to invoke fork callbacks here. */
4282        BUG_ON(!list_empty(&init_task.tasks));
4283
4284        ss->active = 1;
4285
4286        /* this function shouldn't be used with modular subsystems, since they
4287         * need to register a subsys_id, among other things */
4288        BUG_ON(ss->module);
4289}
4290
4291/**
4292 * cgroup_load_subsys: load and register a modular subsystem at runtime
4293 * @ss: the subsystem to load
4294 *
4295 * This function should be called in a modular subsystem's initcall. If the
4296 * subsystem is built as a module, it will be assigned a new subsys_id and set
4297 * up for use. If the subsystem is built-in anyway, work is delegated to the
4298 * simpler cgroup_init_subsys.
4299 */
4300int __init_or_module cgroup_load_subsys(struct cgroup_subsys *ss)
4301{
4302        int i;
4303        struct cgroup_subsys_state *css;
4304
4305        /* check name and function validity */
4306        if (ss->name == NULL || strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN ||
4307            ss->create == NULL || ss->destroy == NULL)
4308                return -EINVAL;
4309
4310        /*
4311         * we don't support callbacks in modular subsystems. this check is
4312         * before the ss->module check for consistency; a subsystem that could
4313         * be a module should still have no callbacks even if the user isn't
4314         * compiling it as one.
4315         */
4316        if (ss->fork || ss->exit)
4317                return -EINVAL;
4318
4319        /*
4320         * an optionally modular subsystem is built-in: we want to do nothing,
4321         * since cgroup_init_subsys will have already taken care of it.
4322         */
4323        if (ss->module == NULL) {
4324                /* a few sanity checks */
4325                BUG_ON(ss->subsys_id >= CGROUP_BUILTIN_SUBSYS_COUNT);
4326                BUG_ON(subsys[ss->subsys_id] != ss);
4327                return 0;
4328        }
4329
4330        /* init base cftset */
4331        cgroup_init_cftsets(ss);
4332
4333        /*
4334         * need to register a subsys id before anything else - for example,
4335         * init_cgroup_css needs it.
4336         */
4337        mutex_lock(&cgroup_mutex);
4338        /* find the first empty slot in the array */
4339        for (i = CGROUP_BUILTIN_SUBSYS_COUNT; i < CGROUP_SUBSYS_COUNT; i++) {
4340                if (subsys[i] == NULL)
4341                        break;
4342        }
4343        if (i == CGROUP_SUBSYS_COUNT) {
4344                /* maximum number of subsystems already registered! */
4345                mutex_unlock(&cgroup_mutex);
4346                return -EBUSY;
4347        }
4348        /* assign ourselves the subsys_id */
4349        ss->subsys_id = i;
4350        subsys[i] = ss;
4351
4352        /*
4353         * no ss->create seems to need anything important in the ss struct, so
4354         * this can happen first (i.e. before the rootnode attachment).
4355         */
4356        css = ss->create(dummytop);
4357        if (IS_ERR(css)) {
4358                /* failure case - need to deassign the subsys[] slot. */
4359                subsys[i] = NULL;
4360                mutex_unlock(&cgroup_mutex);
4361                return PTR_ERR(css);
4362        }
4363
4364        list_add(&ss->sibling, &rootnode.subsys_list);
4365        ss->root = &rootnode;
4366
4367        /* our new subsystem will be attached to the dummy hierarchy. */
4368        init_cgroup_css(css, ss, dummytop);
4369        /* init_idr must be after init_cgroup_css because it sets css->id. */
4370        if (ss->use_id) {
4371                int ret = cgroup_init_idr(ss, css);
4372                if (ret) {
4373                        dummytop->subsys[ss->subsys_id] = NULL;
4374                        ss->destroy(dummytop);
4375                        subsys[i] = NULL;
4376                        mutex_unlock(&cgroup_mutex);
4377                        return ret;
4378                }
4379        }
4380
4381        /*
4382         * Now we need to entangle the css into the existing css_sets. unlike
4383         * in cgroup_init_subsys, there are now multiple css_sets, so each one
4384         * will need a new pointer to it; done by iterating the css_set_table.
4385         * furthermore, modifying the existing css_sets will corrupt the hash
4386         * table state, so each changed css_set will need its hash recomputed.
4387         * this is all done under the css_set_lock.
4388         */
4389        write_lock(&css_set_lock);
4390        for (i = 0; i < CSS_SET_TABLE_SIZE; i++) {
4391                struct css_set *cg;
4392                struct hlist_node *node, *tmp;
4393                struct hlist_head *bucket = &css_set_table[i], *new_bucket;
4394
4395                hlist_for_each_entry_safe(cg, node, tmp, bucket, hlist) {
4396                        /* skip entries that we already rehashed */
4397                        if (cg->subsys[ss->subsys_id])
4398                                continue;
4399                        /* remove existing entry */
4400                        hlist_del(&cg->hlist);
4401                        /* set new value */
4402                        cg->subsys[ss->subsys_id] = css;
4403                        /* recompute hash and restore entry */
4404                        new_bucket = css_set_hash(cg->subsys);
4405                        hlist_add_head(&cg->hlist, new_bucket);
4406                }
4407        }
4408        write_unlock(&css_set_lock);
4409
4410        ss->active = 1;
4411
4412        /* success! */
4413        mutex_unlock(&cgroup_mutex);
4414        return 0;
4415}
4416EXPORT_SYMBOL_GPL(cgroup_load_subsys);
4417
4418/**
4419 * cgroup_unload_subsys: unload a modular subsystem
4420 * @ss: the subsystem to unload
4421 *
4422 * This function should be called in a modular subsystem's exitcall. When this
4423 * function is invoked, the refcount on the subsystem's module will be 0, so
4424 * the subsystem will not be attached to any hierarchy.
4425 */
4426void cgroup_unload_subsys(struct cgroup_subsys *ss)
4427{
4428        struct cg_cgroup_link *link;
4429        struct hlist_head *hhead;
4430
4431        BUG_ON(ss->module == NULL);
4432
4433        /*
4434         * we shouldn't be called if the subsystem is in use, and the use of
4435         * try_module_get in parse_cgroupfs_options should ensure that it
4436         * doesn't start being used while we're killing it off.
4437         */
4438        BUG_ON(ss->root != &rootnode);
4439
4440        mutex_lock(&cgroup_mutex);
4441        /* deassign the subsys_id */
4442        BUG_ON(ss->subsys_id < CGROUP_BUILTIN_SUBSYS_COUNT);
4443        subsys[ss->subsys_id] = NULL;
4444
4445        /* remove subsystem from rootnode's list of subsystems */
4446        list_del_init(&ss->sibling);
4447
4448        /*
4449         * disentangle the css from all css_sets attached to the dummytop. as
4450         * in loading, we need to pay our respects to the hashtable gods.
4451         */
4452        write_lock(&css_set_lock);
4453        list_for_each_entry(link, &dummytop->css_sets, cgrp_link_list) {
4454                struct css_set *cg = link->cg;
4455
4456                hlist_del(&cg->hlist);
4457                BUG_ON(!cg->subsys[ss->subsys_id]);
4458                cg->subsys[ss->subsys_id] = NULL;
4459                hhead = css_set_hash(cg->subsys);
4460                hlist_add_head(&cg->hlist, hhead);
4461        }
4462        write_unlock(&css_set_lock);
4463
4464        /*
4465         * remove subsystem's css from the dummytop and free it - need to free
4466         * before marking as null because ss->destroy needs the cgrp->subsys
4467         * pointer to find their state. note that this also takes care of
4468         * freeing the css_id.
4469         */
4470        ss->destroy(dummytop);
4471        dummytop->subsys[ss->subsys_id] = NULL;
4472
4473        mutex_unlock(&cgroup_mutex);
4474}
4475EXPORT_SYMBOL_GPL(cgroup_unload_subsys);
4476
4477/**
4478 * cgroup_init_early - cgroup initialization at system boot
4479 *
4480 * Initialize cgroups at system boot, and initialize any
4481 * subsystems that request early init.
4482 */
4483int __init cgroup_init_early(void)
4484{
4485        int i;
4486        atomic_set(&init_css_set.refcount, 1);
4487        INIT_LIST_HEAD(&init_css_set.cg_links);
4488        INIT_LIST_HEAD(&init_css_set.tasks);
4489        INIT_HLIST_NODE(&init_css_set.hlist);
4490        css_set_count = 1;
4491        init_cgroup_root(&rootnode);
4492        root_count = 1;
4493        init_task.cgroups = &init_css_set;
4494
4495        init_css_set_link.cg = &init_css_set;
4496        init_css_set_link.cgrp = dummytop;
4497        list_add(&init_css_set_link.cgrp_link_list,
4498                 &rootnode.top_cgroup.css_sets);
4499        list_add(&init_css_set_link.cg_link_list,
4500                 &init_css_set.cg_links);
4501
4502        for (i = 0; i < CSS_SET_TABLE_SIZE; i++)
4503                INIT_HLIST_HEAD(&css_set_table[i]);
4504
4505        /* at bootup time, we don't worry about modular subsystems */
4506        for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
4507                struct cgroup_subsys *ss = subsys[i];
4508
4509                BUG_ON(!ss->name);
4510                BUG_ON(strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN);
4511                BUG_ON(!ss->create);
4512                BUG_ON(!ss->destroy);
4513                if (ss->subsys_id != i) {
4514                        printk(KERN_ERR "cgroup: Subsys %s id == %d\n",
4515                               ss->name, ss->subsys_id);
4516                        BUG();
4517                }
4518
4519                if (ss->early_init)
4520                        cgroup_init_subsys(ss);
4521        }
4522        return 0;
4523}
4524
4525/**
4526 * cgroup_init - cgroup initialization
4527 *
4528 * Register cgroup filesystem and /proc file, and initialize
4529 * any subsystems that didn't request early init.
4530 */
4531int __init cgroup_init(void)
4532{
4533        int err;
4534        int i;
4535        struct hlist_head *hhead;
4536
4537        err = bdi_init(&cgroup_backing_dev_info);
4538        if (err)
4539                return err;
4540
4541        /* at bootup time, we don't worry about modular subsystems */
4542        for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
4543                struct cgroup_subsys *ss = subsys[i];
4544                if (!ss->early_init)
4545                        cgroup_init_subsys(ss);
4546                if (ss->use_id)
4547                        cgroup_init_idr(ss, init_css_set.subsys[ss->subsys_id]);
4548        }
4549
4550        /* Add init_css_set to the hash table */
4551        hhead = css_set_hash(init_css_set.subsys);
4552        hlist_add_head(&init_css_set.hlist, hhead);
4553        BUG_ON(!init_root_id(&rootnode));
4554
4555        cgroup_kobj = kobject_create_and_add("cgroup", fs_kobj);
4556        if (!cgroup_kobj) {
4557                err = -ENOMEM;
4558                goto out;
4559        }
4560
4561        err = register_filesystem(&cgroup_fs_type);
4562        if (err < 0) {
4563                kobject_put(cgroup_kobj);
4564                goto out;
4565        }
4566
4567        proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations);
4568
4569out:
4570        if (err)
4571                bdi_destroy(&cgroup_backing_dev_info);
4572
4573        return err;
4574}
4575
4576/*
4577 * proc_cgroup_show()
4578 *  - Print task's cgroup paths into seq_file, one line for each hierarchy
4579 *  - Used for /proc/<pid>/cgroup.
4580 *  - No need to task_lock(tsk) on this tsk->cgroup reference, as it
4581 *    doesn't really matter if tsk->cgroup changes after we read it,
4582 *    and we take cgroup_mutex, keeping cgroup_attach_task() from changing it
4583 *    anyway.  No need to check that tsk->cgroup != NULL, thanks to
4584 *    the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks
4585 *    cgroup to top_cgroup.
4586 */
4587
4588/* TODO: Use a proper seq_file iterator */
4589static int proc_cgroup_show(struct seq_file *m, void *v)
4590{
4591        struct pid *pid;
4592        struct task_struct *tsk;
4593        char *buf;
4594        int retval;
4595        struct cgroupfs_root *root;
4596
4597        retval = -ENOMEM;
4598        buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
4599        if (!buf)
4600                goto out;
4601
4602        retval = -ESRCH;
4603        pid = m->private;
4604        tsk = get_pid_task(pid, PIDTYPE_PID);
4605        if (!tsk)
4606                goto out_free;
4607
4608        retval = 0;
4609
4610        mutex_lock(&cgroup_mutex);
4611
4612        for_each_active_root(root) {
4613                struct cgroup_subsys *ss;
4614                struct cgroup *cgrp;
4615                int count = 0;
4616
4617                seq_printf(m, "%d:", root->hierarchy_id);
4618                for_each_subsys(root, ss)
4619                        seq_printf(m, "%s%s", count++ ? "," : "", ss->name);
4620                if (strlen(root->name))
4621                        seq_printf(m, "%sname=%s", count ? "," : "",
4622                                   root->name);
4623                seq_putc(m, ':');
4624                cgrp = task_cgroup_from_root(tsk, root);
4625                retval = cgroup_path(cgrp, buf, PAGE_SIZE);
4626                if (retval < 0)
4627                        goto out_unlock;
4628                seq_puts(m, buf);
4629                seq_putc(m, '\n');
4630        }
4631
4632out_unlock:
4633        mutex_unlock(&cgroup_mutex);
4634        put_task_struct(tsk);
4635out_free:
4636        kfree(buf);
4637out:
4638        return retval;
4639}
4640
4641static int cgroup_open(struct inode *inode, struct file *file)
4642{
4643        struct pid *pid = PROC_I(inode)->pid;
4644        return single_open(file, proc_cgroup_show, pid);
4645}
4646
4647const struct file_operations proc_cgroup_operations = {
4648        .open           = cgroup_open,
4649        .read           = seq_read,
4650        .llseek         = seq_lseek,
4651        .release        = single_release,
4652};
4653
4654/* Display information about each subsystem and each hierarchy */
4655static int proc_cgroupstats_show(struct seq_file *m, void *v)
4656{
4657        int i;
4658
4659        seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
4660        /*
4661         * ideally we don't want subsystems moving around while we do this.
4662         * cgroup_mutex is also necessary to guarantee an atomic snapshot of
4663         * subsys/hierarchy state.
4664         */
4665        mutex_lock(&cgroup_mutex);
4666        for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
4667                struct cgroup_subsys *ss = subsys[i];
4668                if (ss == NULL)
4669                        continue;
4670                seq_printf(m, "%s\t%d\t%d\t%d\n",
4671                           ss->name, ss->root->hierarchy_id,
4672                           ss->root->number_of_cgroups, !ss->disabled);
4673        }
4674        mutex_unlock(&cgroup_mutex);
4675        return 0;
4676}
4677
4678static int cgroupstats_open(struct inode *inode, struct file *file)
4679{
4680        return single_open(file, proc_cgroupstats_show, NULL);
4681}
4682
4683static const struct file_operations proc_cgroupstats_operations = {
4684        .open = cgroupstats_open,
4685        .read = seq_read,
4686        .llseek = seq_lseek,
4687        .release = single_release,
4688};
4689
4690/**
4691 * cgroup_fork - attach newly forked task to its parents cgroup.
4692 * @child: pointer to task_struct of forking parent process.
4693 *
4694 * Description: A task inherits its parent's cgroup at fork().
4695 *
4696 * A pointer to the shared css_set was automatically copied in
4697 * fork.c by dup_task_struct().  However, we ignore that copy, since
4698 * it was not made under the protection of RCU, cgroup_mutex or
4699 * threadgroup_change_begin(), so it might no longer be a valid
4700 * cgroup pointer.  cgroup_attach_task() might have already changed
4701 * current->cgroups, allowing the previously referenced cgroup
4702 * group to be removed and freed.
4703 *
4704 * Outside the pointer validity we also need to process the css_set
4705 * inheritance between threadgoup_change_begin() and
4706 * threadgoup_change_end(), this way there is no leak in any process
4707 * wide migration performed by cgroup_attach_proc() that could otherwise
4708 * miss a thread because it is too early or too late in the fork stage.
4709 *
4710 * At the point that cgroup_fork() is called, 'current' is the parent
4711 * task, and the passed argument 'child' points to the child task.
4712 */
4713void cgroup_fork(struct task_struct *child)
4714{
4715        /*
4716         * We don't need to task_lock() current because current->cgroups
4717         * can't be changed concurrently here. The parent obviously hasn't
4718         * exited and called cgroup_exit(), and we are synchronized against
4719         * cgroup migration through threadgroup_change_begin().
4720         */
4721        child->cgroups = current->cgroups;
4722        get_css_set(child->cgroups);
4723        INIT_LIST_HEAD(&child->cg_list);
4724}
4725
4726/**
4727 * cgroup_fork_callbacks - run fork callbacks
4728 * @child: the new task
4729 *
4730 * Called on a new task very soon before adding it to the
4731 * tasklist. No need to take any locks since no-one can
4732 * be operating on this task.
4733 */
4734void cgroup_fork_callbacks(struct task_struct *child)
4735{
4736        if (need_forkexit_callback) {
4737                int i;
4738                /*
4739                 * forkexit callbacks are only supported for builtin
4740                 * subsystems, and the builtin section of the subsys array is
4741                 * immutable, so we don't need to lock the subsys array here.
4742                 */
4743                for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
4744                        struct cgroup_subsys *ss = subsys[i];
4745                        if (ss->fork)
4746                                ss->fork(child);
4747                }
4748        }
4749}
4750
4751/**
4752 * cgroup_post_fork - called on a new task after adding it to the task list
4753 * @child: the task in question
4754 *
4755 * Adds the task to the list running through its css_set if necessary.
4756 * Has to be after the task is visible on the task list in case we race
4757 * with the first call to cgroup_iter_start() - to guarantee that the
4758 * new task ends up on its list.
4759 */
4760void cgroup_post_fork(struct task_struct *child)
4761{
4762        /*
4763         * use_task_css_set_links is set to 1 before we walk the tasklist
4764         * under the tasklist_lock and we read it here after we added the child
4765         * to the tasklist under the tasklist_lock as well. If the child wasn't
4766         * yet in the tasklist when we walked through it from
4767         * cgroup_enable_task_cg_lists(), then use_task_css_set_links value
4768         * should be visible now due to the paired locking and barriers implied
4769         * by LOCK/UNLOCK: it is written before the tasklist_lock unlock
4770         * in cgroup_enable_task_cg_lists() and read here after the tasklist_lock
4771         * lock on fork.
4772         */
4773        if (use_task_css_set_links) {
4774                write_lock(&css_set_lock);
4775                if (list_empty(&child->cg_list)) {
4776                        /*
4777                         * It's safe to use child->cgroups without task_lock()
4778                         * here because we are protected through
4779                         * threadgroup_change_begin() against concurrent
4780                         * css_set change in cgroup_task_migrate(). Also
4781                         * the task can't exit at that point until
4782                         * wake_up_new_task() is called, so we are protected
4783                         * against cgroup_exit() setting child->cgroup to
4784                         * init_css_set.
4785                         */
4786                        list_add(&child->cg_list, &child->cgroups->tasks);
4787                }
4788                write_unlock(&css_set_lock);
4789        }
4790}
4791/**
4792 * cgroup_exit - detach cgroup from exiting task
4793 * @tsk: pointer to task_struct of exiting process
4794 * @run_callback: run exit callbacks?
4795 *
4796 * Description: Detach cgroup from @tsk and release it.
4797 *
4798 * Note that cgroups marked notify_on_release force every task in
4799 * them to take the global cgroup_mutex mutex when exiting.
4800 * This could impact scaling on very large systems.  Be reluctant to
4801 * use notify_on_release cgroups where very high task exit scaling
4802 * is required on large systems.
4803 *
4804 * the_top_cgroup_hack:
4805 *
4806 *    Set the exiting tasks cgroup to the root cgroup (top_cgroup).
4807 *
4808 *    We call cgroup_exit() while the task is still competent to
4809 *    handle notify_on_release(), then leave the task attached to the
4810 *    root cgroup in each hierarchy for the remainder of its exit.
4811 *
4812 *    To do this properly, we would increment the reference count on
4813 *    top_cgroup, and near the very end of the kernel/exit.c do_exit()
4814 *    code we would add a second cgroup function call, to drop that
4815 *    reference.  This would just create an unnecessary hot spot on
4816 *    the top_cgroup reference count, to no avail.
4817 *
4818 *    Normally, holding a reference to a cgroup without bumping its
4819 *    count is unsafe.   The cgroup could go away, or someone could
4820 *    attach us to a different cgroup, decrementing the count on
4821 *    the first cgroup that we never incremented.  But in this case,
4822 *    top_cgroup isn't going away, and either task has PF_EXITING set,
4823 *    which wards off any cgroup_attach_task() attempts, or task is a failed
4824 *    fork, never visible to cgroup_attach_task.
4825 */
4826void cgroup_exit(struct task_struct *tsk, int run_callbacks)
4827{
4828        struct css_set *cg;
4829        int i;
4830
4831        /*
4832         * Unlink from the css_set task list if necessary.
4833         * Optimistically check cg_list before taking
4834         * css_set_lock
4835         */
4836        if (!list_empty(&tsk->cg_list)) {
4837                write_lock(&css_set_lock);
4838                if (!list_empty(&tsk->cg_list))
4839                        list_del_init(&tsk->cg_list);
4840                write_unlock(&css_set_lock);
4841        }
4842
4843        /* Reassign the task to the init_css_set. */
4844        task_lock(tsk);
4845        cg = tsk->cgroups;
4846        tsk->cgroups = &init_css_set;
4847
4848        if (run_callbacks && need_forkexit_callback) {
4849                /*
4850                 * modular subsystems can't use callbacks, so no need to lock
4851                 * the subsys array
4852                 */
4853                for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
4854                        struct cgroup_subsys *ss = subsys[i];
4855                        if (ss->exit) {
4856                                struct cgroup *old_cgrp =
4857                                        rcu_dereference_raw(cg->subsys[i])->cgroup;
4858                                struct cgroup *cgrp = task_cgroup(tsk, i);
4859                                ss->exit(cgrp, old_cgrp, tsk);
4860                        }
4861                }
4862        }
4863        task_unlock(tsk);
4864
4865        if (cg)
4866                put_css_set_taskexit(cg);
4867}
4868
4869/**
4870 * cgroup_is_descendant - see if @cgrp is a descendant of @task's cgrp
4871 * @cgrp: the cgroup in question
4872 * @task: the task in question
4873 *
4874 * See if @cgrp is a descendant of @task's cgroup in the appropriate
4875 * hierarchy.
4876 *
4877 * If we are sending in dummytop, then presumably we are creating
4878 * the top cgroup in the subsystem.
4879 *
4880 * Called only by the ns (nsproxy) cgroup.
4881 */
4882int cgroup_is_descendant(const struct cgroup *cgrp, struct task_struct *task)
4883{
4884        int ret;
4885        struct cgroup *target;
4886
4887        if (cgrp == dummytop)
4888                return 1;
4889
4890        target = task_cgroup_from_root(task, cgrp->root);
4891        while (cgrp != target && cgrp!= cgrp->top_cgroup)
4892                cgrp = cgrp->parent;
4893        ret = (cgrp == target);
4894        return ret;
4895}
4896
4897static void check_for_release(struct cgroup *cgrp)
4898{
4899        /* All of these checks rely on RCU to keep the cgroup
4900         * structure alive */
4901        if (cgroup_is_releasable(cgrp) && !atomic_read(&cgrp->count)
4902            && list_empty(&cgrp->children) && !cgroup_has_css_refs(cgrp)) {
4903                /* Control Group is currently removeable. If it's not
4904                 * already queued for a userspace notification, queue
4905                 * it now */
4906                int need_schedule_work = 0;
4907                raw_spin_lock(&release_list_lock);
4908                if (!cgroup_is_removed(cgrp) &&
4909                    list_empty(&cgrp->release_list)) {
4910                        list_add(&cgrp->release_list, &release_list);
4911                        need_schedule_work = 1;
4912                }
4913                raw_spin_unlock(&release_list_lock);
4914                if (need_schedule_work)
4915                        schedule_work(&release_agent_work);
4916        }
4917}
4918
4919/* Caller must verify that the css is not for root cgroup */
4920bool __css_tryget(struct cgroup_subsys_state *css)
4921{
4922        do {
4923                int v = css_refcnt(css);
4924
4925                if (atomic_cmpxchg(&css->refcnt, v, v + 1) == v)
4926                        return true;
4927                cpu_relax();
4928        } while (!test_bit(CSS_REMOVED, &css->flags));
4929
4930        return false;
4931}
4932EXPORT_SYMBOL_GPL(__css_tryget);
4933
4934/* Caller must verify that the css is not for root cgroup */
4935void __css_put(struct cgroup_subsys_state *css)
4936{
4937        struct cgroup *cgrp = css->cgroup;
4938        int v;
4939
4940        rcu_read_lock();
4941        v = css_unbias_refcnt(atomic_dec_return(&css->refcnt));
4942
4943        switch (v) {
4944        case 1:
4945                if (notify_on_release(cgrp)) {
4946                        set_bit(CGRP_RELEASABLE, &cgrp->flags);
4947                        check_for_release(cgrp);
4948                }
4949                cgroup_wakeup_rmdir_waiter(cgrp);
4950                break;
4951        case 0:
4952                if (!test_bit(CSS_CLEAR_CSS_REFS, &css->flags))
4953                        schedule_work(&css->dput_work);
4954                break;
4955        }
4956        rcu_read_unlock();
4957}
4958EXPORT_SYMBOL_GPL(__css_put);
4959
4960/*
4961 * Notify userspace when a cgroup is released, by running the
4962 * configured release agent with the name of the cgroup (path
4963 * relative to the root of cgroup file system) as the argument.
4964 *
4965 * Most likely, this user command will try to rmdir this cgroup.
4966 *
4967 * This races with the possibility that some other task will be
4968 * attached to this cgroup before it is removed, or that some other
4969 * user task will 'mkdir' a child cgroup of this cgroup.  That's ok.
4970 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
4971 * unused, and this cgroup will be reprieved from its death sentence,
4972 * to continue to serve a useful existence.  Next time it's released,
4973 * we will get notified again, if it still has 'notify_on_release' set.
4974 *
4975 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
4976 * means only wait until the task is successfully execve()'d.  The
4977 * separate release agent task is forked by call_usermodehelper(),
4978 * then control in this thread returns here, without waiting for the
4979 * release agent task.  We don't bother to wait because the caller of
4980 * this routine has no use for the exit status of the release agent
4981 * task, so no sense holding our caller up for that.
4982 */
4983static void cgroup_release_agent(struct work_struct *work)
4984{
4985        BUG_ON(work != &release_agent_work);
4986        mutex_lock(&cgroup_mutex);
4987        raw_spin_lock(&release_list_lock);
4988        while (!list_empty(&release_list)) {
4989                char *argv[3], *envp[3];
4990                int i;
4991                char *pathbuf = NULL, *agentbuf = NULL;
4992                struct cgroup *cgrp = list_entry(release_list.next,
4993                                                    struct cgroup,
4994                                                    release_list);
4995                list_del_init(&cgrp->release_list);
4996                raw_spin_unlock(&release_list_lock);
4997                pathbuf = kmalloc(PAGE_SIZE, GFP_KERNEL);
4998                if (!pathbuf)
4999                        goto continue_free;
5000                if (cgroup_path(cgrp, pathbuf, PAGE_SIZE) < 0)
5001                        goto continue_free;
5002                agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL);
5003                if (!agentbuf)
5004                        goto continue_free;
5005
5006                i = 0;
5007                argv[i++] = agentbuf;
5008                argv[i++] = pathbuf;
5009                argv[i] = NULL;
5010
5011                i = 0;
5012                /* minimal command environment */
5013                envp[i++] = "HOME=/";
5014                envp[i++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
5015                envp[i] = NULL;
5016
5017                /* Drop the lock while we invoke the usermode helper,
5018                 * since the exec could involve hitting disk and hence
5019                 * be a slow process */
5020                mutex_unlock(&cgroup_mutex);
5021                call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC);
5022                mutex_lock(&cgroup_mutex);
5023 continue_free:
5024                kfree(pathbuf);
5025                kfree(agentbuf);
5026                raw_spin_lock(&release_list_lock);
5027        }
5028        raw_spin_unlock(&release_list_lock);
5029        mutex_unlock(&cgroup_mutex);
5030}
5031
5032static int __init cgroup_disable(char *str)
5033{
5034        int i;
5035        char *token;
5036
5037        while ((token = strsep(&str, ",")) != NULL) {
5038                if (!*token)
5039                        continue;
5040                /*
5041                 * cgroup_disable, being at boot time, can't know about module
5042                 * subsystems, so we don't worry about them.
5043                 */
5044                for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
5045                        struct cgroup_subsys *ss = subsys[i];
5046
5047                        if (!strcmp(token, ss->name)) {
5048                                ss->disabled = 1;
5049                                printk(KERN_INFO "Disabling %s control group"
5050                                        " subsystem\n", ss->name);
5051                                break;
5052                        }
5053                }
5054        }
5055        return 1;
5056}
5057__setup("cgroup_disable=", cgroup_disable);
5058
5059/*
5060 * Functons for CSS ID.
5061 */
5062
5063/*
5064 *To get ID other than 0, this should be called when !cgroup_is_removed().
5065 */
5066unsigned short css_id(struct cgroup_subsys_state *css)
5067{
5068        struct css_id *cssid;
5069
5070        /*
5071         * This css_id() can return correct value when somone has refcnt
5072         * on this or this is under rcu_read_lock(). Once css->id is allocated,
5073         * it's unchanged until freed.
5074         */
5075        cssid = rcu_dereference_check(css->id, css_refcnt(css));
5076
5077        if (cssid)
5078                return cssid->id;
5079        return 0;
5080}
5081EXPORT_SYMBOL_GPL(css_id);
5082
5083unsigned short css_depth(struct cgroup_subsys_state *css)
5084{
5085        struct css_id *cssid;
5086
5087        cssid = rcu_dereference_check(css->id, css_refcnt(css));
5088
5089        if (cssid)
5090                return cssid->depth;
5091        return 0;
5092}
5093EXPORT_SYMBOL_GPL(css_depth);
5094
5095/**
5096 *  css_is_ancestor - test "root" css is an ancestor of "child"
5097 * @child: the css to be tested.
5098 * @root: the css supporsed to be an ancestor of the child.
5099 *
5100 * Returns true if "root" is an ancestor of "child" in its hierarchy. Because
5101 * this function reads css->id, the caller must hold rcu_read_lock().
5102 * But, considering usual usage, the csses should be valid objects after test.
5103 * Assuming that the caller will do some action to the child if this returns
5104 * returns true, the caller must take "child";s reference count.
5105 * If "child" is valid object and this returns true, "root" is valid, too.
5106 */
5107
5108bool css_is_ancestor(struct cgroup_subsys_state *child,
5109                    const struct cgroup_subsys_state *root)
5110{
5111        struct css_id *child_id;
5112        struct css_id *root_id;
5113
5114        child_id  = rcu_dereference(child->id);
5115        if (!child_id)
5116                return false;
5117        root_id = rcu_dereference(root->id);
5118        if (!root_id)
5119                return false;
5120        if (child_id->depth < root_id->depth)
5121                return false;
5122        if (child_id->stack[root_id->depth] != root_id->id)
5123                return false;
5124        return true;
5125}
5126
5127void free_css_id(struct cgroup_subsys *ss, struct cgroup_subsys_state *css)
5128{
5129        struct css_id *id = css->id;
5130        /* When this is called before css_id initialization, id can be NULL */
5131        if (!id)
5132                return;
5133
5134        BUG_ON(!ss->use_id);
5135
5136        rcu_assign_pointer(id->css, NULL);
5137        rcu_assign_pointer(css->id, NULL);
5138        spin_lock(&ss->id_lock);
5139        idr_remove(&ss->idr, id->id);
5140        spin_unlock(&ss->id_lock);
5141        kfree_rcu(id, rcu_head);
5142}
5143EXPORT_SYMBOL_GPL(free_css_id);
5144
5145/*
5146 * This is called by init or create(). Then, calls to this function are
5147 * always serialized (By cgroup_mutex() at create()).
5148 */
5149
5150static struct css_id *get_new_cssid(struct cgroup_subsys *ss, int depth)
5151{
5152        struct css_id *newid;
5153        int myid, error, size;
5154
5155        BUG_ON(!ss->use_id);
5156
5157        size = sizeof(*newid) + sizeof(unsigned short) * (depth + 1);
5158        newid = kzalloc(size, GFP_KERNEL);
5159        if (!newid)
5160                return ERR_PTR(-ENOMEM);
5161        /* get id */
5162        if (unlikely(!idr_pre_get(&ss->idr, GFP_KERNEL))) {
5163                error = -ENOMEM;
5164                goto err_out;
5165        }
5166        spin_lock(&ss->id_lock);
5167        /* Don't use 0. allocates an ID of 1-65535 */
5168        error = idr_get_new_above(&ss->idr, newid, 1, &myid);
5169        spin_unlock(&ss->id_lock);
5170
5171        /* Returns error when there are no free spaces for new ID.*/
5172        if (error) {
5173                error = -ENOSPC;
5174                goto err_out;
5175        }
5176        if (myid > CSS_ID_MAX)
5177                goto remove_idr;
5178
5179        newid->id = myid;
5180        newid->depth = depth;
5181        return newid;
5182remove_idr:
5183        error = -ENOSPC;
5184        spin_lock(&ss->id_lock);
5185        idr_remove(&ss->idr, myid);
5186        spin_unlock(&ss->id_lock);
5187err_out:
5188        kfree(newid);
5189        return ERR_PTR(error);
5190
5191}
5192
5193static int __init_or_module cgroup_init_idr(struct cgroup_subsys *ss,
5194                                            struct cgroup_subsys_state *rootcss)
5195{
5196        struct css_id *newid;
5197
5198        spin_lock_init(&ss->id_lock);
5199        idr_init(&ss->idr);
5200
5201        newid = get_new_cssid(ss, 0);
5202        if (IS_ERR(newid))
5203                return PTR_ERR(newid);
5204
5205        newid->stack[0] = newid->id;
5206        newid->css = rootcss;
5207        rootcss->id = newid;
5208        return 0;
5209}
5210
5211static int alloc_css_id(struct cgroup_subsys *ss, struct cgroup *parent,
5212                        struct cgroup *child)
5213{
5214        int subsys_id, i, depth = 0;
5215        struct cgroup_subsys_state *parent_css, *child_css;
5216        struct css_id *child_id, *parent_id;
5217
5218        subsys_id = ss->subsys_id;
5219        parent_css = parent->subsys[subsys_id];
5220        child_css = child->subsys[subsys_id];
5221        parent_id = parent_css->id;
5222        depth = parent_id->depth + 1;
5223
5224        child_id = get_new_cssid(ss, depth);
5225        if (IS_ERR(child_id))
5226                return PTR_ERR(child_id);
5227
5228        for (i = 0; i < depth; i++)
5229                child_id->stack[i] = parent_id->stack[i];
5230        child_id->stack[depth] = child_id->id;
5231        /*
5232         * child_id->css pointer will be set after this cgroup is available
5233         * see cgroup_populate_dir()
5234         */
5235        rcu_assign_pointer(child_css->id, child_id);
5236
5237        return 0;
5238}
5239
5240/**
5241 * css_lookup - lookup css by id
5242 * @ss: cgroup subsys to be looked into.
5243 * @id: the id
5244 *
5245 * Returns pointer to cgroup_subsys_state if there is valid one with id.
5246 * NULL if not. Should be called under rcu_read_lock()
5247 */
5248struct cgroup_subsys_state *css_lookup(struct cgroup_subsys *ss, int id)
5249{
5250        struct css_id *cssid = NULL;
5251
5252        BUG_ON(!ss->use_id);
5253        cssid = idr_find(&ss->idr, id);
5254
5255        if (unlikely(!cssid))
5256                return NULL;
5257
5258        return rcu_dereference(cssid->css);
5259}
5260EXPORT_SYMBOL_GPL(css_lookup);
5261
5262/**
5263 * css_get_next - lookup next cgroup under specified hierarchy.
5264 * @ss: pointer to subsystem
5265 * @id: current position of iteration.
5266 * @root: pointer to css. search tree under this.
5267 * @foundid: position of found object.
5268 *
5269 * Search next css under the specified hierarchy of rootid. Calling under
5270 * rcu_read_lock() is necessary. Returns NULL if it reaches the end.
5271 */
5272struct cgroup_subsys_state *
5273css_get_next(struct cgroup_subsys *ss, int id,
5274             struct cgroup_subsys_state *root, int *foundid)
5275{
5276        struct cgroup_subsys_state *ret = NULL;
5277        struct css_id *tmp;
5278        int tmpid;
5279        int rootid = css_id(root);
5280        int depth = css_depth(root);
5281
5282        if (!rootid)
5283                return NULL;
5284
5285        BUG_ON(!ss->use_id);
5286        WARN_ON_ONCE(!rcu_read_lock_held());
5287
5288        /* fill start point for scan */
5289        tmpid = id;
5290        while (1) {
5291                /*
5292                 * scan next entry from bitmap(tree), tmpid is updated after
5293                 * idr_get_next().
5294                 */
5295                tmp = idr_get_next(&ss->idr, &tmpid);
5296                if (!tmp)
5297                        break;
5298                if (tmp->depth >= depth && tmp->stack[depth] == rootid) {
5299                        ret = rcu_dereference(tmp->css);
5300                        if (ret) {
5301                                *foundid = tmpid;
5302                                break;
5303                        }
5304                }
5305                /* continue to scan from next id */
5306                tmpid = tmpid + 1;
5307        }
5308        return ret;
5309}
5310
5311/*
5312 * get corresponding css from file open on cgroupfs directory
5313 */
5314struct cgroup_subsys_state *cgroup_css_from_dir(struct file *f, int id)
5315{
5316        struct cgroup *cgrp;
5317        struct inode *inode;
5318        struct cgroup_subsys_state *css;
5319
5320        inode = f->f_dentry->d_inode;
5321        /* check in cgroup filesystem dir */
5322        if (inode->i_op != &cgroup_dir_inode_operations)
5323                return ERR_PTR(-EBADF);
5324
5325        if (id < 0 || id >= CGROUP_SUBSYS_COUNT)
5326                return ERR_PTR(-EINVAL);
5327
5328        /* get cgroup */
5329        cgrp = __d_cgrp(f->f_dentry);
5330        css = cgrp->subsys[id];
5331        return css ? css : ERR_PTR(-ENOENT);
5332}
5333
5334#ifdef CONFIG_CGROUP_DEBUG
5335static struct cgroup_subsys_state *debug_create(struct cgroup *cont)
5336{
5337        struct cgroup_subsys_state *css = kzalloc(sizeof(*css), GFP_KERNEL);
5338
5339        if (!css)
5340                return ERR_PTR(-ENOMEM);
5341
5342        return css;
5343}
5344
5345static void debug_destroy(struct cgroup *cont)
5346{
5347        kfree(cont->subsys[debug_subsys_id]);
5348}
5349
5350static u64 cgroup_refcount_read(struct cgroup *cont, struct cftype *cft)
5351{
5352        return atomic_read(&cont->count);
5353}
5354
5355static u64 debug_taskcount_read(struct cgroup *cont, struct cftype *cft)
5356{
5357        return cgroup_task_count(cont);
5358}
5359
5360static u64 current_css_set_read(struct cgroup *cont, struct cftype *cft)
5361{
5362        return (u64)(unsigned long)current->cgroups;
5363}
5364
5365static u64 current_css_set_refcount_read(struct cgroup *cont,
5366                                           struct cftype *cft)
5367{
5368        u64 count;
5369
5370        rcu_read_lock();
5371        count = atomic_read(&current->cgroups->refcount);
5372        rcu_read_unlock();
5373        return count;
5374}
5375
5376static int current_css_set_cg_links_read(struct cgroup *cont,
5377                                         struct cftype *cft,
5378                                         struct seq_file *seq)
5379{
5380        struct cg_cgroup_link *link;
5381        struct css_set *cg;
5382
5383        read_lock(&css_set_lock);
5384        rcu_read_lock();
5385        cg = rcu_dereference(current->cgroups);
5386        list_for_each_entry(link, &cg->cg_links, cg_link_list) {
5387                struct cgroup *c = link->cgrp;
5388                const char *name;
5389
5390                if (c->dentry)
5391                        name = c->dentry->d_name.name;
5392                else
5393                        name = "?";
5394                seq_printf(seq, "Root %d group %s\n",
5395                           c->root->hierarchy_id, name);
5396        }
5397        rcu_read_unlock();
5398        read_unlock(&css_set_lock);
5399        return 0;
5400}
5401
5402#define MAX_TASKS_SHOWN_PER_CSS 25
5403static int cgroup_css_links_read(struct cgroup *cont,
5404                                 struct cftype *cft,
5405                                 struct seq_file *seq)
5406{
5407        struct cg_cgroup_link *link;
5408
5409        read_lock(&css_set_lock);
5410        list_for_each_entry(link, &cont->css_sets, cgrp_link_list) {
5411                struct css_set *cg = link->cg;
5412                struct task_struct *task;
5413                int count = 0;
5414                seq_printf(seq, "css_set %p\n", cg);
5415                list_for_each_entry(task, &cg->tasks, cg_list) {
5416                        if (count++ > MAX_TASKS_SHOWN_PER_CSS) {
5417                                seq_puts(seq, "  ...\n");
5418                                break;
5419                        } else {
5420                                seq_printf(seq, "  task %d\n",
5421                                           task_pid_vnr(task));
5422                        }
5423                }
5424        }
5425        read_unlock(&css_set_lock);
5426        return 0;
5427}
5428
5429static u64 releasable_read(struct cgroup *cgrp, struct cftype *cft)
5430{
5431        return test_bit(CGRP_RELEASABLE, &cgrp->flags);
5432}
5433
5434static struct cftype debug_files[] =  {
5435        {
5436                .name = "cgroup_refcount",
5437                .read_u64 = cgroup_refcount_read,
5438        },
5439        {
5440                .name = "taskcount",
5441                .read_u64 = debug_taskcount_read,
5442        },
5443
5444        {
5445                .name = "current_css_set",
5446                .read_u64 = current_css_set_read,
5447        },
5448
5449        {
5450                .name = "current_css_set_refcount",
5451                .read_u64 = current_css_set_refcount_read,
5452        },
5453
5454        {
5455                .name = "current_css_set_cg_links",
5456                .read_seq_string = current_css_set_cg_links_read,
5457        },
5458
5459        {
5460                .name = "cgroup_css_links",
5461                .read_seq_string = cgroup_css_links_read,
5462        },
5463
5464        {
5465                .name = "releasable",
5466                .read_u64 = releasable_read,
5467        },
5468
5469        { }     /* terminate */
5470};
5471
5472struct cgroup_subsys debug_subsys = {
5473        .name = "debug",
5474        .create = debug_create,
5475        .destroy = debug_destroy,
5476        .subsys_id = debug_subsys_id,
5477        .base_cftypes = debug_files,
5478};
5479#endif /* CONFIG_CGROUP_DEBUG */
5480
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.