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/ctype.h>
  31#include <linux/errno.h>
  32#include <linux/fs.h>
  33#include <linux/kernel.h>
  34#include <linux/list.h>
  35#include <linux/mm.h>
  36#include <linux/mutex.h>
  37#include <linux/mount.h>
  38#include <linux/pagemap.h>
  39#include <linux/proc_fs.h>
  40#include <linux/rcupdate.h>
  41#include <linux/sched.h>
  42#include <linux/backing-dev.h>
  43#include <linux/seq_file.h>
  44#include <linux/slab.h>
  45#include <linux/magic.h>
  46#include <linux/spinlock.h>
  47#include <linux/string.h>
  48#include <linux/sort.h>
  49#include <linux/kmod.h>
  50#include <linux/module.h>
  51#include <linux/delayacct.h>
  52#include <linux/cgroupstats.h>
  53#include <linux/hash.h>
  54#include <linux/namei.h>
  55#include <linux/pid_namespace.h>
  56#include <linux/idr.h>
  57#include <linux/vmalloc.h> /* TODO: replace with more sophisticated array */
  58#include <linux/eventfd.h>
  59#include <linux/poll.h>
  60
  61#include <asm/atomic.h>
  62
  63static DEFINE_MUTEX(cgroup_mutex);
  64
  65/*
  66 * Generate an array of cgroup subsystem pointers. At boot time, this is
  67 * populated up to CGROUP_BUILTIN_SUBSYS_COUNT, and modular subsystems are
  68 * registered after that. The mutable section of this array is protected by
  69 * cgroup_mutex.
  70 */
  71#define SUBSYS(_x) &_x ## _subsys,
  72static struct cgroup_subsys *subsys[CGROUP_SUBSYS_COUNT] = {
  73#include <linux/cgroup_subsys.h>
  74};
  75
  76#define MAX_CGROUP_ROOT_NAMELEN 64
  77
  78/*
  79 * A cgroupfs_root represents the root of a cgroup hierarchy,
  80 * and may be associated with a superblock to form an active
  81 * hierarchy
  82 */
  83struct cgroupfs_root {
  84        struct super_block *sb;
  85
  86        /*
  87         * The bitmask of subsystems intended to be attached to this
  88         * hierarchy
  89         */
  90        unsigned long subsys_bits;
  91
  92        /* Unique id for this hierarchy. */
  93        int hierarchy_id;
  94
  95        /* The bitmask of subsystems currently attached to this hierarchy */
  96        unsigned long actual_subsys_bits;
  97
  98        /* A list running through the attached subsystems */
  99        struct list_head subsys_list;
 100
 101        /* The root cgroup for this hierarchy */
 102        struct cgroup top_cgroup;
 103
 104        /* Tracks how many cgroups are currently defined in hierarchy.*/
 105        int number_of_cgroups;
 106
 107        /* A list running through the active hierarchies */
 108        struct list_head root_list;
 109
 110        /* Hierarchy-specific flags */
 111        unsigned long flags;
 112
 113        /* The path to use for release notifications. */
 114        char release_agent_path[PATH_MAX];
 115
 116        /* The name for this hierarchy - may be empty */
 117        char name[MAX_CGROUP_ROOT_NAMELEN];
 118};
 119
 120/*
 121 * The "rootnode" hierarchy is the "dummy hierarchy", reserved for the
 122 * subsystems that are otherwise unattached - it never has more than a
 123 * single cgroup, and all tasks are part of that cgroup.
 124 */
 125static struct cgroupfs_root rootnode;
 126
 127/*
 128 * CSS ID -- ID per subsys's Cgroup Subsys State(CSS). used only when
 129 * cgroup_subsys->use_id != 0.
 130 */
 131#define CSS_ID_MAX      (65535)
 132struct css_id {
 133        /*
 134         * The css to which this ID points. This pointer is set to valid value
 135         * after cgroup is populated. If cgroup is removed, this will be NULL.
 136         * This pointer is expected to be RCU-safe because destroy()
 137         * is called after synchronize_rcu(). But for safe use, css_is_removed()
 138         * css_tryget() should be used for avoiding race.
 139         */
 140        struct cgroup_subsys_state __rcu *css;
 141        /*
 142         * ID of this css.
 143         */
 144        unsigned short id;
 145        /*
 146         * Depth in hierarchy which this ID belongs to.
 147         */
 148        unsigned short depth;
 149        /*
 150         * ID is freed by RCU. (and lookup routine is RCU safe.)
 151         */
 152        struct rcu_head rcu_head;
 153        /*
 154         * Hierarchy of CSS ID belongs to.
 155         */
 156        unsigned short stack[0]; /* Array of Length (depth+1) */
 157};
 158
 159/*
 160 * cgroup_event represents events which userspace want to receive.
 161 */
 162struct cgroup_event {
 163        /*
 164         * Cgroup which the event belongs to.
 165         */
 166        struct cgroup *cgrp;
 167        /*
 168         * Control file which the event associated.
 169         */
 170        struct cftype *cft;
 171        /*
 172         * eventfd to signal userspace about the event.
 173         */
 174        struct eventfd_ctx *eventfd;
 175        /*
 176         * Each of these stored in a list by the cgroup.
 177         */
 178        struct list_head list;
 179        /*
 180         * All fields below needed to unregister event when
 181         * userspace closes eventfd.
 182         */
 183        poll_table pt;
 184        wait_queue_head_t *wqh;
 185        wait_queue_t wait;
 186        struct work_struct remove;
 187};
 188
 189/* The list of hierarchy roots */
 190
 191static LIST_HEAD(roots);
 192static int root_count;
 193
 194static DEFINE_IDA(hierarchy_ida);
 195static int next_hierarchy_id;
 196static DEFINE_SPINLOCK(hierarchy_id_lock);
 197
 198/* dummytop is a shorthand for the dummy hierarchy's top cgroup */
 199#define dummytop (&rootnode.top_cgroup)
 200
 201/* This flag indicates whether tasks in the fork and exit paths should
 202 * check for fork/exit handlers to call. This avoids us having to do
 203 * extra work in the fork/exit path if none of the subsystems need to
 204 * be called.
 205 */
 206static int need_forkexit_callback __read_mostly;
 207
 208#ifdef CONFIG_PROVE_LOCKING
 209int cgroup_lock_is_held(void)
 210{
 211        return lockdep_is_held(&cgroup_mutex);
 212}
 213#else /* #ifdef CONFIG_PROVE_LOCKING */
 214int cgroup_lock_is_held(void)
 215{
 216        return mutex_is_locked(&cgroup_mutex);
 217}
 218#endif /* #else #ifdef CONFIG_PROVE_LOCKING */
 219
 220EXPORT_SYMBOL_GPL(cgroup_lock_is_held);
 221
 222/* convenient tests for these bits */
 223inline int cgroup_is_removed(const struct cgroup *cgrp)
 224{
 225        return test_bit(CGRP_REMOVED, &cgrp->flags);
 226}
 227
 228/* bits in struct cgroupfs_root flags field */
 229enum {
 230        ROOT_NOPREFIX, /* mounted subsystems have no named prefix */
 231};
 232
 233static int cgroup_is_releasable(const struct cgroup *cgrp)
 234{
 235        const int bits =
 236                (1 << CGRP_RELEASABLE) |
 237                (1 << CGRP_NOTIFY_ON_RELEASE);
 238        return (cgrp->flags & bits) == bits;
 239}
 240
 241static int notify_on_release(const struct cgroup *cgrp)
 242{
 243        return test_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
 244}
 245
 246static int clone_children(const struct cgroup *cgrp)
 247{
 248        return test_bit(CGRP_CLONE_CHILDREN, &cgrp->flags);
 249}
 250
 251/*
 252 * for_each_subsys() allows you to iterate on each subsystem attached to
 253 * an active hierarchy
 254 */
 255#define for_each_subsys(_root, _ss) \
 256list_for_each_entry(_ss, &_root->subsys_list, sibling)
 257
 258/* for_each_active_root() allows you to iterate across the active hierarchies */
 259#define for_each_active_root(_root) \
 260list_for_each_entry(_root, &roots, root_list)
 261
 262/* the list of cgroups eligible for automatic release. Protected by
 263 * release_list_lock */
 264static LIST_HEAD(release_list);
 265static DEFINE_SPINLOCK(release_list_lock);
 266static void cgroup_release_agent(struct work_struct *work);
 267static DECLARE_WORK(release_agent_work, cgroup_release_agent);
 268static void check_for_release(struct cgroup *cgrp);
 269
 270/* Link structure for associating css_set objects with cgroups */
 271struct cg_cgroup_link {
 272        /*
 273         * List running through cg_cgroup_links associated with a
 274         * cgroup, anchored on cgroup->css_sets
 275         */
 276        struct list_head cgrp_link_list;
 277        struct cgroup *cgrp;
 278        /*
 279         * List running through cg_cgroup_links pointing at a
 280         * single css_set object, anchored on css_set->cg_links
 281         */
 282        struct list_head cg_link_list;
 283        struct css_set *cg;
 284};
 285
 286/* The default css_set - used by init and its children prior to any
 287 * hierarchies being mounted. It contains a pointer to the root state
 288 * for each subsystem. Also used to anchor the list of css_sets. Not
 289 * reference-counted, to improve performance when child cgroups
 290 * haven't been created.
 291 */
 292
 293static struct css_set init_css_set;
 294static struct cg_cgroup_link init_css_set_link;
 295
 296static int cgroup_init_idr(struct cgroup_subsys *ss,
 297                           struct cgroup_subsys_state *css);
 298
 299/* css_set_lock protects the list of css_set objects, and the
 300 * chain of tasks off each css_set.  Nests outside task->alloc_lock
 301 * due to cgroup_iter_start() */
 302static DEFINE_RWLOCK(css_set_lock);
 303static int css_set_count;
 304
 305/*
 306 * hash table for cgroup groups. This improves the performance to find
 307 * an existing css_set. This hash doesn't (currently) take into
 308 * account cgroups in empty hierarchies.
 309 */
 310#define CSS_SET_HASH_BITS       7
 311#define CSS_SET_TABLE_SIZE      (1 << CSS_SET_HASH_BITS)
 312static struct hlist_head css_set_table[CSS_SET_TABLE_SIZE];
 313
 314static struct hlist_head *css_set_hash(struct cgroup_subsys_state *css[])
 315{
 316        int i;
 317        int index;
 318        unsigned long tmp = 0UL;
 319
 320        for (i = 0; i < CGROUP_SUBSYS_COUNT; i++)
 321                tmp += (unsigned long)css[i];
 322        tmp = (tmp >> 16) ^ tmp;
 323
 324        index = hash_long(tmp, CSS_SET_HASH_BITS);
 325
 326        return &css_set_table[index];
 327}
 328
 329static void free_css_set_rcu(struct rcu_head *obj)
 330{
 331        struct css_set *cg = container_of(obj, struct css_set, rcu_head);
 332        kfree(cg);
 333}
 334
 335/* We don't maintain the lists running through each css_set to its
 336 * task until after the first call to cgroup_iter_start(). This
 337 * reduces the fork()/exit() overhead for people who have cgroups
 338 * compiled into their kernel but not actually in use */
 339static int use_task_css_set_links __read_mostly;
 340
 341static void __put_css_set(struct css_set *cg, int taskexit)
 342{
 343        struct cg_cgroup_link *link;
 344        struct cg_cgroup_link *saved_link;
 345        /*
 346         * Ensure that the refcount doesn't hit zero while any readers
 347         * can see it. Similar to atomic_dec_and_lock(), but for an
 348         * rwlock
 349         */
 350        if (atomic_add_unless(&cg->refcount, -1, 1))
 351                return;
 352        write_lock(&css_set_lock);
 353        if (!atomic_dec_and_test(&cg->refcount)) {
 354                write_unlock(&css_set_lock);
 355                return;
 356        }
 357
 358        /* This css_set is dead. unlink it and release cgroup refcounts */
 359        hlist_del(&cg->hlist);
 360        css_set_count--;
 361
 362        list_for_each_entry_safe(link, saved_link, &cg->cg_links,
 363                                 cg_link_list) {
 364                struct cgroup *cgrp = link->cgrp;
 365                list_del(&link->cg_link_list);
 366                list_del(&link->cgrp_link_list);
 367                if (atomic_dec_and_test(&cgrp->count) &&
 368                    notify_on_release(cgrp)) {
 369                        if (taskexit)
 370                                set_bit(CGRP_RELEASABLE, &cgrp->flags);
 371                        check_for_release(cgrp);
 372                }
 373
 374                kfree(link);
 375        }
 376
 377        write_unlock(&css_set_lock);
 378        call_rcu(&cg->rcu_head, free_css_set_rcu);
 379}
 380
 381/*
 382 * refcounted get/put for css_set objects
 383 */
 384static inline void get_css_set(struct css_set *cg)
 385{
 386        atomic_inc(&cg->refcount);
 387}
 388
 389static inline void put_css_set(struct css_set *cg)
 390{
 391        __put_css_set(cg, 0);
 392}
 393
 394static inline void put_css_set_taskexit(struct css_set *cg)
 395{
 396        __put_css_set(cg, 1);
 397}
 398
 399/*
 400 * compare_css_sets - helper function for find_existing_css_set().
 401 * @cg: candidate css_set being tested
 402 * @old_cg: existing css_set for a task
 403 * @new_cgrp: cgroup that's being entered by the task
 404 * @template: desired set of css pointers in css_set (pre-calculated)
 405 *
 406 * Returns true if "cg" matches "old_cg" except for the hierarchy
 407 * which "new_cgrp" belongs to, for which it should match "new_cgrp".
 408 */
 409static bool compare_css_sets(struct css_set *cg,
 410                             struct css_set *old_cg,
 411                             struct cgroup *new_cgrp,
 412                             struct cgroup_subsys_state *template[])
 413{
 414        struct list_head *l1, *l2;
 415
 416        if (memcmp(template, cg->subsys, sizeof(cg->subsys))) {
 417                /* Not all subsystems matched */
 418                return false;
 419        }
 420
 421        /*
 422         * Compare cgroup pointers in order to distinguish between
 423         * different cgroups in heirarchies with no subsystems. We
 424         * could get by with just this check alone (and skip the
 425         * memcmp above) but on most setups the memcmp check will
 426         * avoid the need for this more expensive check on almost all
 427         * candidates.
 428         */
 429
 430        l1 = &cg->cg_links;
 431        l2 = &old_cg->cg_links;
 432        while (1) {
 433                struct cg_cgroup_link *cgl1, *cgl2;
 434                struct cgroup *cg1, *cg2;
 435
 436                l1 = l1->next;
 437                l2 = l2->next;
 438                /* See if we reached the end - both lists are equal length. */
 439                if (l1 == &cg->cg_links) {
 440                        BUG_ON(l2 != &old_cg->cg_links);
 441                        break;
 442                } else {
 443                        BUG_ON(l2 == &old_cg->cg_links);
 444                }
 445                /* Locate the cgroups associated with these links. */
 446                cgl1 = list_entry(l1, struct cg_cgroup_link, cg_link_list);
 447                cgl2 = list_entry(l2, struct cg_cgroup_link, cg_link_list);
 448                cg1 = cgl1->cgrp;
 449                cg2 = cgl2->cgrp;
 450                /* Hierarchies should be linked in the same order. */
 451                BUG_ON(cg1->root != cg2->root);
 452
 453                /*
 454                 * If this hierarchy is the hierarchy of the cgroup
 455                 * that's changing, then we need to check that this
 456                 * css_set points to the new cgroup; if it's any other
 457                 * hierarchy, then this css_set should point to the
 458                 * same cgroup as the old css_set.
 459                 */
 460                if (cg1->root == new_cgrp->root) {
 461                        if (cg1 != new_cgrp)
 462                                return false;
 463                } else {
 464                        if (cg1 != cg2)
 465                                return false;
 466                }
 467        }
 468        return true;
 469}
 470
 471/*
 472 * find_existing_css_set() is a helper for
 473 * find_css_set(), and checks to see whether an existing
 474 * css_set is suitable.
 475 *
 476 * oldcg: the cgroup group that we're using before the cgroup
 477 * transition
 478 *
 479 * cgrp: the cgroup that we're moving into
 480 *
 481 * template: location in which to build the desired set of subsystem
 482 * state objects for the new cgroup group
 483 */
 484static struct css_set *find_existing_css_set(
 485        struct css_set *oldcg,
 486        struct cgroup *cgrp,
 487        struct cgroup_subsys_state *template[])
 488{
 489        int i;
 490        struct cgroupfs_root *root = cgrp->root;
 491        struct hlist_head *hhead;
 492        struct hlist_node *node;
 493        struct css_set *cg;
 494
 495        /*
 496         * Build the set of subsystem state objects that we want to see in the
 497         * new css_set. while subsystems can change globally, the entries here
 498         * won't change, so no need for locking.
 499         */
 500        for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
 501                if (root->subsys_bits & (1UL << i)) {
 502                        /* Subsystem is in this hierarchy. So we want
 503                         * the subsystem state from the new
 504                         * cgroup */
 505                        template[i] = cgrp->subsys[i];
 506                } else {
 507                        /* Subsystem is not in this hierarchy, so we
 508                         * don't want to change the subsystem state */
 509                        template[i] = oldcg->subsys[i];
 510                }
 511        }
 512
 513        hhead = css_set_hash(template);
 514        hlist_for_each_entry(cg, node, hhead, hlist) {
 515                if (!compare_css_sets(cg, oldcg, cgrp, template))
 516                        continue;
 517
 518                /* This css_set matches what we need */
 519                return cg;
 520        }
 521
 522        /* No existing cgroup group matched */
 523        return NULL;
 524}
 525
 526static void free_cg_links(struct list_head *tmp)
 527{
 528        struct cg_cgroup_link *link;
 529        struct cg_cgroup_link *saved_link;
 530
 531        list_for_each_entry_safe(link, saved_link, tmp, cgrp_link_list) {
 532                list_del(&link->cgrp_link_list);
 533                kfree(link);
 534        }
 535}
 536
 537/*
 538 * allocate_cg_links() allocates "count" cg_cgroup_link structures
 539 * and chains them on tmp through their cgrp_link_list fields. Returns 0 on
 540 * success or a negative error
 541 */
 542static int allocate_cg_links(int count, struct list_head *tmp)
 543{
 544        struct cg_cgroup_link *link;
 545        int i;
 546        INIT_LIST_HEAD(tmp);
 547        for (i = 0; i < count; i++) {
 548                link = kmalloc(sizeof(*link), GFP_KERNEL);
 549                if (!link) {
 550                        free_cg_links(tmp);
 551                        return -ENOMEM;
 552                }
 553                list_add(&link->cgrp_link_list, tmp);
 554        }
 555        return 0;
 556}
 557
 558/**
 559 * link_css_set - a helper function to link a css_set to a cgroup
 560 * @tmp_cg_links: cg_cgroup_link objects allocated by allocate_cg_links()
 561 * @cg: the css_set to be linked
 562 * @cgrp: the destination cgroup
 563 */
 564static void link_css_set(struct list_head *tmp_cg_links,
 565                         struct css_set *cg, struct cgroup *cgrp)
 566{
 567        struct cg_cgroup_link *link;
 568
 569        BUG_ON(list_empty(tmp_cg_links));
 570        link = list_first_entry(tmp_cg_links, struct cg_cgroup_link,
 571                                cgrp_link_list);
 572        link->cg = cg;
 573        link->cgrp = cgrp;
 574        atomic_inc(&cgrp->count);
 575        list_move(&link->cgrp_link_list, &cgrp->css_sets);
 576        /*
 577         * Always add links to the tail of the list so that the list
 578         * is sorted by order of hierarchy creation
 579         */
 580        list_add_tail(&link->cg_link_list, &cg->cg_links);
 581}
 582
 583/*
 584 * find_css_set() takes an existing cgroup group and a
 585 * cgroup object, and returns a css_set object that's
 586 * equivalent to the old group, but with the given cgroup
 587 * substituted into the appropriate hierarchy. Must be called with
 588 * cgroup_mutex held
 589 */
 590static struct css_set *find_css_set(
 591        struct css_set *oldcg, struct cgroup *cgrp)
 592{
 593        struct css_set *res;
 594        struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT];
 595
 596        struct list_head tmp_cg_links;
 597
 598        struct hlist_head *hhead;
 599        struct cg_cgroup_link *link;
 600
 601        /* First see if we already have a cgroup group that matches
 602         * the desired set */
 603        read_lock(&css_set_lock);
 604        res = find_existing_css_set(oldcg, cgrp, template);
 605        if (res)
 606                get_css_set(res);
 607        read_unlock(&css_set_lock);
 608
 609        if (res)
 610                return res;
 611
 612        res = kmalloc(sizeof(*res), GFP_KERNEL);
 613        if (!res)
 614                return NULL;
 615
 616        /* Allocate all the cg_cgroup_link objects that we'll need */
 617        if (allocate_cg_links(root_count, &tmp_cg_links) < 0) {
 618                kfree(res);
 619                return NULL;
 620        }
 621
 622        atomic_set(&res->refcount, 1);
 623        INIT_LIST_HEAD(&res->cg_links);
 624        INIT_LIST_HEAD(&res->tasks);
 625        INIT_HLIST_NODE(&res->hlist);
 626
 627        /* Copy the set of subsystem state objects generated in
 628         * find_existing_css_set() */
 629        memcpy(res->subsys, template, sizeof(res->subsys));
 630
 631        write_lock(&css_set_lock);
 632        /* Add reference counts and links from the new css_set. */
 633        list_for_each_entry(link, &oldcg->cg_links, cg_link_list) {
 634                struct cgroup *c = link->cgrp;
 635                if (c->root == cgrp->root)
 636                        c = cgrp;
 637                link_css_set(&tmp_cg_links, res, c);
 638        }
 639
 640        BUG_ON(!list_empty(&tmp_cg_links));
 641
 642        css_set_count++;
 643
 644        /* Add this cgroup group to the hash table */
 645        hhead = css_set_hash(res->subsys);
 646        hlist_add_head(&res->hlist, hhead);
 647
 648        write_unlock(&css_set_lock);
 649
 650        return res;
 651}
 652
 653/*
 654 * Return the cgroup for "task" from the given hierarchy. Must be
 655 * called with cgroup_mutex held.
 656 */
 657static struct cgroup *task_cgroup_from_root(struct task_struct *task,
 658                                            struct cgroupfs_root *root)
 659{
 660        struct css_set *css;
 661        struct cgroup *res = NULL;
 662
 663        BUG_ON(!mutex_is_locked(&cgroup_mutex));
 664        read_lock(&css_set_lock);
 665        /*
 666         * No need to lock the task - since we hold cgroup_mutex the
 667         * task can't change groups, so the only thing that can happen
 668         * is that it exits and its css is set back to init_css_set.
 669         */
 670        css = task->cgroups;
 671        if (css == &init_css_set) {
 672                res = &root->top_cgroup;
 673        } else {
 674                struct cg_cgroup_link *link;
 675                list_for_each_entry(link, &css->cg_links, cg_link_list) {
 676                        struct cgroup *c = link->cgrp;
 677                        if (c->root == root) {
 678                                res = c;
 679                                break;
 680                        }
 681                }
 682        }
 683        read_unlock(&css_set_lock);
 684        BUG_ON(!res);
 685        return res;
 686}
 687
 688/*
 689 * There is one global cgroup mutex. We also require taking
 690 * task_lock() when dereferencing a task's cgroup subsys pointers.
 691 * See "The task_lock() exception", at the end of this comment.
 692 *
 693 * A task must hold cgroup_mutex to modify cgroups.
 694 *
 695 * Any task can increment and decrement the count field without lock.
 696 * So in general, code holding cgroup_mutex can't rely on the count
 697 * field not changing.  However, if the count goes to zero, then only
 698 * cgroup_attach_task() can increment it again.  Because a count of zero
 699 * means that no tasks are currently attached, therefore there is no
 700 * way a task attached to that cgroup can fork (the other way to
 701 * increment the count).  So code holding cgroup_mutex can safely
 702 * assume that if the count is zero, it will stay zero. Similarly, if
 703 * a task holds cgroup_mutex on a cgroup with zero count, it
 704 * knows that the cgroup won't be removed, as cgroup_rmdir()
 705 * needs that mutex.
 706 *
 707 * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't
 708 * (usually) take cgroup_mutex.  These are the two most performance
 709 * critical pieces of code here.  The exception occurs on cgroup_exit(),
 710 * when a task in a notify_on_release cgroup exits.  Then cgroup_mutex
 711 * is taken, and if the cgroup count is zero, a usermode call made
 712 * to the release agent with the name of the cgroup (path relative to
 713 * the root of cgroup file system) as the argument.
 714 *
 715 * A cgroup can only be deleted if both its 'count' of using tasks
 716 * is zero, and its list of 'children' cgroups is empty.  Since all
 717 * tasks in the system use _some_ cgroup, and since there is always at
 718 * least one task in the system (init, pid == 1), therefore, top_cgroup
 719 * always has either children cgroups and/or using tasks.  So we don't
 720 * need a special hack to ensure that top_cgroup cannot be deleted.
 721 *
 722 *      The task_lock() exception
 723 *
 724 * The need for this exception arises from the action of
 725 * cgroup_attach_task(), which overwrites one tasks cgroup pointer with
 726 * another.  It does so using cgroup_mutex, however there are
 727 * several performance critical places that need to reference
 728 * task->cgroup without the expense of grabbing a system global
 729 * mutex.  Therefore except as noted below, when dereferencing or, as
 730 * in cgroup_attach_task(), modifying a task'ss cgroup pointer we use
 731 * task_lock(), which acts on a spinlock (task->alloc_lock) already in
 732 * the task_struct routinely used for such matters.
 733 *
 734 * P.S.  One more locking exception.  RCU is used to guard the
 735 * update of a tasks cgroup pointer by cgroup_attach_task()
 736 */
 737
 738/**
 739 * cgroup_lock - lock out any changes to cgroup structures
 740 *
 741 */
 742void cgroup_lock(void)
 743{
 744        mutex_lock(&cgroup_mutex);
 745}
 746EXPORT_SYMBOL_GPL(cgroup_lock);
 747
 748/**
 749 * cgroup_unlock - release lock on cgroup changes
 750 *
 751 * Undo the lock taken in a previous cgroup_lock() call.
 752 */
 753void cgroup_unlock(void)
 754{
 755        mutex_unlock(&cgroup_mutex);
 756}
 757EXPORT_SYMBOL_GPL(cgroup_unlock);
 758
 759/*
 760 * A couple of forward declarations required, due to cyclic reference loop:
 761 * cgroup_mkdir -> cgroup_create -> cgroup_populate_dir ->
 762 * cgroup_add_file -> cgroup_create_file -> cgroup_dir_inode_operations
 763 * -> cgroup_mkdir.
 764 */
 765
 766static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, int mode);
 767static struct dentry *cgroup_lookup(struct inode *, struct dentry *, struct nameidata *);
 768static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry);
 769static int cgroup_populate_dir(struct cgroup *cgrp);
 770static const struct inode_operations cgroup_dir_inode_operations;
 771static const struct file_operations proc_cgroupstats_operations;
 772
 773static struct backing_dev_info cgroup_backing_dev_info = {
 774        .name           = "cgroup",
 775        .capabilities   = BDI_CAP_NO_ACCT_AND_WRITEBACK,
 776};
 777
 778static int alloc_css_id(struct cgroup_subsys *ss,
 779                        struct cgroup *parent, struct cgroup *child);
 780
 781static struct inode *cgroup_new_inode(mode_t mode, struct super_block *sb)
 782{
 783        struct inode *inode = new_inode(sb);
 784
 785        if (inode) {
 786                inode->i_ino = get_next_ino();
 787                inode->i_mode = mode;
 788                inode->i_uid = current_fsuid();
 789                inode->i_gid = current_fsgid();
 790                inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 791                inode->i_mapping->backing_dev_info = &cgroup_backing_dev_info;
 792        }
 793        return inode;
 794}
 795
 796/*
 797 * Call subsys's pre_destroy handler.
 798 * This is called before css refcnt check.
 799 */
 800static int cgroup_call_pre_destroy(struct cgroup *cgrp)
 801{
 802        struct cgroup_subsys *ss;
 803        int ret = 0;
 804
 805        for_each_subsys(cgrp->root, ss)
 806                if (ss->pre_destroy) {
 807                        ret = ss->pre_destroy(ss, cgrp);
 808                        if (ret)
 809                                break;
 810                }
 811
 812        return ret;
 813}
 814
 815static void free_cgroup_rcu(struct rcu_head *obj)
 816{
 817        struct cgroup *cgrp = container_of(obj, struct cgroup, rcu_head);
 818
 819        kfree(cgrp);
 820}
 821
 822static void cgroup_diput(struct dentry *dentry, struct inode *inode)
 823{
 824        /* is dentry a directory ? if so, kfree() associated cgroup */
 825        if (S_ISDIR(inode->i_mode)) {
 826                struct cgroup *cgrp = dentry->d_fsdata;
 827                struct cgroup_subsys *ss;
 828                BUG_ON(!(cgroup_is_removed(cgrp)));
 829                /* It's possible for external users to be holding css
 830                 * reference counts on a cgroup; css_put() needs to
 831                 * be able to access the cgroup after decrementing
 832                 * the reference count in order to know if it needs to
 833                 * queue the cgroup to be handled by the release
 834                 * agent */
 835                synchronize_rcu();
 836
 837                mutex_lock(&cgroup_mutex);
 838                /*
 839                 * Release the subsystem state objects.
 840                 */
 841                for_each_subsys(cgrp->root, ss)
 842                        ss->destroy(ss, cgrp);
 843
 844                cgrp->root->number_of_cgroups--;
 845                mutex_unlock(&cgroup_mutex);
 846
 847                /*
 848                 * Drop the active superblock reference that we took when we
 849                 * created the cgroup
 850                 */
 851                deactivate_super(cgrp->root->sb);
 852
 853                /*
 854                 * if we're getting rid of the cgroup, refcount should ensure
 855                 * that there are no pidlists left.
 856                 */
 857                BUG_ON(!list_empty(&cgrp->pidlists));
 858
 859                call_rcu(&cgrp->rcu_head, free_cgroup_rcu);
 860        }
 861        iput(inode);
 862}
 863
 864static int cgroup_delete(const struct dentry *d)
 865{
 866        return 1;
 867}
 868
 869static void remove_dir(struct dentry *d)
 870{
 871        struct dentry *parent = dget(d->d_parent);
 872
 873        d_delete(d);
 874        simple_rmdir(parent->d_inode, d);
 875        dput(parent);
 876}
 877
 878static void cgroup_clear_directory(struct dentry *dentry)
 879{
 880        struct list_head *node;
 881
 882        BUG_ON(!mutex_is_locked(&dentry->d_inode->i_mutex));
 883        spin_lock(&dentry->d_lock);
 884        node = dentry->d_subdirs.next;
 885        while (node != &dentry->d_subdirs) {
 886                struct dentry *d = list_entry(node, struct dentry, d_u.d_child);
 887
 888                spin_lock_nested(&d->d_lock, DENTRY_D_LOCK_NESTED);
 889                list_del_init(node);
 890                if (d->d_inode) {
 891                        /* This should never be called on a cgroup
 892                         * directory with child cgroups */
 893                        BUG_ON(d->d_inode->i_mode & S_IFDIR);
 894                        dget_dlock(d);
 895                        spin_unlock(&d->d_lock);
 896                        spin_unlock(&dentry->d_lock);
 897                        d_delete(d);
 898                        simple_unlink(dentry->d_inode, d);
 899                        dput(d);
 900                        spin_lock(&dentry->d_lock);
 901                } else
 902                        spin_unlock(&d->d_lock);
 903                node = dentry->d_subdirs.next;
 904        }
 905        spin_unlock(&dentry->d_lock);
 906}
 907
 908/*
 909 * NOTE : the dentry must have been dget()'ed
 910 */
 911static void cgroup_d_remove_dir(struct dentry *dentry)
 912{
 913        struct dentry *parent;
 914
 915        cgroup_clear_directory(dentry);
 916
 917        parent = dentry->d_parent;
 918        spin_lock(&parent->d_lock);
 919        spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
 920        list_del_init(&dentry->d_u.d_child);
 921        spin_unlock(&dentry->d_lock);
 922        spin_unlock(&parent->d_lock);
 923        remove_dir(dentry);
 924}
 925
 926/*
 927 * A queue for waiters to do rmdir() cgroup. A tasks will sleep when
 928 * cgroup->count == 0 && list_empty(&cgroup->children) && subsys has some
 929 * reference to css->refcnt. In general, this refcnt is expected to goes down
 930 * to zero, soon.
 931 *
 932 * CGRP_WAIT_ON_RMDIR flag is set under cgroup's inode->i_mutex;
 933 */
 934DECLARE_WAIT_QUEUE_HEAD(cgroup_rmdir_waitq);
 935
 936static void cgroup_wakeup_rmdir_waiter(struct cgroup *cgrp)
 937{
 938        if (unlikely(test_and_clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags)))
 939                wake_up_all(&cgroup_rmdir_waitq);
 940}
 941
 942void cgroup_exclude_rmdir(struct cgroup_subsys_state *css)
 943{
 944        css_get(css);
 945}
 946
 947void cgroup_release_and_wakeup_rmdir(struct cgroup_subsys_state *css)
 948{
 949        cgroup_wakeup_rmdir_waiter(css->cgroup);
 950        css_put(css);
 951}
 952
 953/*
 954 * Call with cgroup_mutex held. Drops reference counts on modules, including
 955 * any duplicate ones that parse_cgroupfs_options took. If this function
 956 * returns an error, no reference counts are touched.
 957 */
 958static int rebind_subsystems(struct cgroupfs_root *root,
 959                              unsigned long final_bits)
 960{
 961        unsigned long added_bits, removed_bits;
 962        struct cgroup *cgrp = &root->top_cgroup;
 963        int i;
 964
 965        BUG_ON(!mutex_is_locked(&cgroup_mutex));
 966
 967        removed_bits = root->actual_subsys_bits & ~final_bits;
 968        added_bits = final_bits & ~root->actual_subsys_bits;
 969        /* Check that any added subsystems are currently free */
 970        for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
 971                unsigned long bit = 1UL << i;
 972                struct cgroup_subsys *ss = subsys[i];
 973                if (!(bit & added_bits))
 974                        continue;
 975                /*
 976                 * Nobody should tell us to do a subsys that doesn't exist:
 977                 * parse_cgroupfs_options should catch that case and refcounts
 978                 * ensure that subsystems won't disappear once selected.
 979                 */
 980                BUG_ON(ss == NULL);
 981                if (ss->root != &rootnode) {
 982                        /* Subsystem isn't free */
 983                        return -EBUSY;
 984                }
 985        }
 986
 987        /* Currently we don't handle adding/removing subsystems when
 988         * any child cgroups exist. This is theoretically supportable
 989         * but involves complex error handling, so it's being left until
 990         * later */
 991        if (root->number_of_cgroups > 1)
 992                return -EBUSY;
 993
 994        /* Process each subsystem */
 995        for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
 996                struct cgroup_subsys *ss = subsys[i];
 997                unsigned long bit = 1UL << i;
 998                if (bit & added_bits) {
 999                        /* We're binding this subsystem to this hierarchy */
1000                        BUG_ON(ss == NULL);
1001                        BUG_ON(cgrp->subsys[i]);
1002                        BUG_ON(!dummytop->subsys[i]);
1003                        BUG_ON(dummytop->subsys[i]->cgroup != dummytop);
1004                        mutex_lock(&ss->hierarchy_mutex);
1005                        cgrp->subsys[i] = dummytop->subsys[i];
1006                        cgrp->subsys[i]->cgroup = cgrp;
1007                        list_move(&ss->sibling, &root->subsys_list);
1008                        ss->root = root;
1009                        if (ss->bind)
1010                                ss->bind(ss, cgrp);
1011                        mutex_unlock(&ss->hierarchy_mutex);
1012                        /* refcount was already taken, and we're keeping it */
1013                } else if (bit & removed_bits) {
1014                        /* We're removing this subsystem */
1015                        BUG_ON(ss == NULL);
1016                        BUG_ON(cgrp->subsys[i] != dummytop->subsys[i]);
1017                        BUG_ON(cgrp->subsys[i]->cgroup != cgrp);
1018                        mutex_lock(&ss->hierarchy_mutex);
1019                        if (ss->bind)
1020                                ss->bind(ss, dummytop);
1021                        dummytop->subsys[i]->cgroup = dummytop;
1022                        cgrp->subsys[i] = NULL;
1023                        subsys[i]->root = &rootnode;
1024                        list_move(&ss->sibling, &rootnode.subsys_list);
1025                        mutex_unlock(&ss->hierarchy_mutex);
1026                        /* subsystem is now free - drop reference on module */
1027                        module_put(ss->module);
1028                } else if (bit & final_bits) {
1029                        /* Subsystem state should already exist */
1030                        BUG_ON(ss == NULL);
1031                        BUG_ON(!cgrp->subsys[i]);
1032                        /*
1033                         * a refcount was taken, but we already had one, so
1034                         * drop the extra reference.
1035                         */
1036                        module_put(ss->module);
1037#ifdef CONFIG_MODULE_UNLOAD
1038                        BUG_ON(ss->module && !module_refcount(ss->module));
1039#endif
1040                } else {
1041                        /* Subsystem state shouldn't exist */
1042                        BUG_ON(cgrp->subsys[i]);
1043                }
1044        }
1045        root->subsys_bits = root->actual_subsys_bits = final_bits;
1046        synchronize_rcu();
1047
1048        return 0;
1049}
1050
1051static int cgroup_show_options(struct seq_file *seq, struct vfsmount *vfs)
1052{
1053        struct cgroupfs_root *root = vfs->mnt_sb->s_fs_info;
1054        struct cgroup_subsys *ss;
1055
1056        mutex_lock(&cgroup_mutex);
1057        for_each_subsys(root, ss)
1058                seq_printf(seq, ",%s", ss->name);
1059        if (test_bit(ROOT_NOPREFIX, &root->flags))
1060                seq_puts(seq, ",noprefix");
1061        if (strlen(root->release_agent_path))
1062                seq_printf(seq, ",release_agent=%s", root->release_agent_path);
1063        if (clone_children(&root->top_cgroup))
1064                seq_puts(seq, ",clone_children");
1065        if (strlen(root->name))
1066                seq_printf(seq, ",name=%s", root->name);
1067        mutex_unlock(&cgroup_mutex);
1068        return 0;
1069}
1070
1071struct cgroup_sb_opts {
1072        unsigned long subsys_bits;
1073        unsigned long flags;
1074        char *release_agent;
1075        bool clone_children;
1076        char *name;
1077        /* User explicitly requested empty subsystem */
1078        bool none;
1079
1080        struct cgroupfs_root *new_root;
1081
1082};
1083
1084/*
1085 * Convert a hierarchy specifier into a bitmask of subsystems and flags. Call
1086 * with cgroup_mutex held to protect the subsys[] array. This function takes
1087 * refcounts on subsystems to be used, unless it returns error, in which case
1088 * no refcounts are taken.
1089 */
1090static int parse_cgroupfs_options(char *data, struct cgroup_sb_opts *opts)
1091{
1092        char *token, *o = data;
1093        bool all_ss = false, one_ss = false;
1094        unsigned long mask = (unsigned long)-1;
1095        int i;
1096        bool module_pin_failed = false;
1097
1098        BUG_ON(!mutex_is_locked(&cgroup_mutex));
1099
1100#ifdef CONFIG_CPUSETS
1101        mask = ~(1UL << cpuset_subsys_id);
1102#endif
1103
1104        memset(opts, 0, sizeof(*opts));
1105
1106        while ((token = strsep(&o, ",")) != NULL) {
1107                if (!*token)
1108                        return -EINVAL;
1109                if (!strcmp(token, "none")) {
1110                        /* Explicitly have no subsystems */
1111                        opts->none = true;
1112                        continue;
1113                }
1114                if (!strcmp(token, "all")) {
1115                        /* Mutually exclusive option 'all' + subsystem name */
1116                        if (one_ss)
1117                                return -EINVAL;
1118                        all_ss = true;
1119                        continue;
1120                }
1121                if (!strcmp(token, "noprefix")) {
1122                        set_bit(ROOT_NOPREFIX, &opts->flags);
1123                        continue;
1124                }
1125                if (!strcmp(token, "clone_children")) {
1126                        opts->clone_children = true;
1127                        continue;
1128                }
1129                if (!strncmp(token, "release_agent=", 14)) {
1130                        /* Specifying two release agents is forbidden */
1131                        if (opts->release_agent)
1132                                return -EINVAL;
1133                        opts->release_agent =
1134                                kstrndup(token + 14, PATH_MAX - 1, GFP_KERNEL);
1135                        if (!opts->release_agent)
1136                                return -ENOMEM;
1137                        continue;
1138                }
1139                if (!strncmp(token, "name=", 5)) {
1140                        const char *name = token + 5;
1141                        /* Can't specify an empty name */
1142                        if (!strlen(name))
1143                                return -EINVAL;
1144                        /* Must match [\w.-]+ */
1145                        for (i = 0; i < strlen(name); i++) {
1146                                char c = name[i];
1147                                if (isalnum(c))
1148                                        continue;
1149                                if ((c == '.') || (c == '-') || (c == '_'))
1150                                        continue;
1151                                return -EINVAL;
1152                        }
1153                        /* Specifying two names is forbidden */
1154                        if (opts->name)
1155                                return -EINVAL;
1156                        opts->name = kstrndup(name,
1157                                              MAX_CGROUP_ROOT_NAMELEN - 1,
1158                                              GFP_KERNEL);
1159                        if (!opts->name)
1160                                return -ENOMEM;
1161
1162                        continue;
1163                }
1164
1165                for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1166                        struct cgroup_subsys *ss = subsys[i];
1167                        if (ss == NULL)
1168                                continue;
1169                        if (strcmp(token, ss->name))
1170                                continue;
1171                        if (ss->disabled)
1172                                continue;
1173
1174                        /* Mutually exclusive option 'all' + subsystem name */
1175                        if (all_ss)
1176                                return -EINVAL;
1177                        set_bit(i, &opts->subsys_bits);
1178                        one_ss = true;
1179
1180                        break;
1181                }
1182                if (i == CGROUP_SUBSYS_COUNT)
1183                        return -ENOENT;
1184        }
1185
1186        /*
1187         * If the 'all' option was specified select all the subsystems,
1188         * otherwise 'all, 'none' and a subsystem name options were not
1189         * specified, let's default to 'all'
1190         */
1191        if (all_ss || (!all_ss && !one_ss && !opts->none)) {
1192                for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1193                        struct cgroup_subsys *ss = subsys[i];
1194                        if (ss == NULL)
1195                                continue;
1196                        if (ss->disabled)
1197                                continue;
1198                        set_bit(i, &opts->subsys_bits);
1199                }
1200        }
1201
1202        /* Consistency checks */
1203
1204        /*
1205         * Option noprefix was introduced just for backward compatibility
1206         * with the old cpuset, so we allow noprefix only if mounting just
1207         * the cpuset subsystem.
1208         */
1209        if (test_bit(ROOT_NOPREFIX, &opts->flags) &&
1210            (opts->subsys_bits & mask))
1211                return -EINVAL;
1212
1213
1214        /* Can't specify "none" and some subsystems */
1215        if (opts->subsys_bits && opts->none)
1216                return -EINVAL;
1217
1218        /*
1219         * We either have to specify by name or by subsystems. (So all
1220         * empty hierarchies must have a name).
1221         */
1222        if (!opts->subsys_bits && !opts->name)
1223                return -EINVAL;
1224
1225        /*
1226         * Grab references on all the modules we'll need, so the subsystems
1227         * don't dance around before rebind_subsystems attaches them. This may
1228         * take duplicate reference counts on a subsystem that's already used,
1229         * but rebind_subsystems handles this case.
1230         */
1231        for (i = CGROUP_BUILTIN_SUBSYS_COUNT; i < CGROUP_SUBSYS_COUNT; i++) {
1232                unsigned long bit = 1UL << i;
1233
1234                if (!(bit & opts->subsys_bits))
1235                        continue;
1236                if (!try_module_get(subsys[i]->module)) {
1237                        module_pin_failed = true;
1238                        break;
1239                }
1240        }
1241        if (module_pin_failed) {
1242                /*
1243                 * oops, one of the modules was going away. this means that we
1244                 * raced with a module_delete call, and to the user this is
1245                 * essentially a "subsystem doesn't exist" case.
1246                 */
1247                for (i--; i >= CGROUP_BUILTIN_SUBSYS_COUNT; i--) {
1248                        /* drop refcounts only on the ones we took */
1249                        unsigned long bit = 1UL << i;
1250
1251                        if (!(bit & opts->subsys_bits))
1252                                continue;
1253                        module_put(subsys[i]->module);
1254                }
1255                return -ENOENT;
1256        }
1257
1258        return 0;
1259}
1260
1261static void drop_parsed_module_refcounts(unsigned long subsys_bits)
1262{
1263        int i;
1264        for (i = CGROUP_BUILTIN_SUBSYS_COUNT; i < CGROUP_SUBSYS_COUNT; i++) {
1265                unsigned long bit = 1UL << i;
1266
1267                if (!(bit & subsys_bits))
1268                        continue;
1269                module_put(subsys[i]->module);
1270        }
1271}
1272
1273static int cgroup_remount(struct super_block *sb, int *flags, char *data)
1274{
1275        int ret = 0;
1276        struct cgroupfs_root *root = sb->s_fs_info;
1277        struct cgroup *cgrp = &root->top_cgroup;
1278        struct cgroup_sb_opts opts;
1279
1280        mutex_lock(&cgrp->dentry->d_inode->i_mutex);
1281        mutex_lock(&cgroup_mutex);
1282
1283        /* See what subsystems are wanted */
1284        ret = parse_cgroupfs_options(data, &opts);
1285        if (ret)
1286                goto out_unlock;
1287
1288        /* Don't allow flags or name to change at remount */
1289        if (opts.flags != root->flags ||
1290            (opts.name && strcmp(opts.name, root->name))) {
1291                ret = -EINVAL;
1292                drop_parsed_module_refcounts(opts.subsys_bits);
1293                goto out_unlock;
1294        }
1295
1296        ret = rebind_subsystems(root, opts.subsys_bits);
1297        if (ret) {
1298                drop_parsed_module_refcounts(opts.subsys_bits);
1299                goto out_unlock;
1300        }
1301
1302        /* (re)populate subsystem files */
1303        cgroup_populate_dir(cgrp);
1304
1305        if (opts.release_agent)
1306                strcpy(root->release_agent_path, opts.release_agent);
1307 out_unlock:
1308        kfree(opts.release_agent);
1309        kfree(opts.name);
1310        mutex_unlock(&cgroup_mutex);
1311        mutex_unlock(&cgrp->dentry->d_inode->i_mutex);
1312        return ret;
1313}
1314
1315static const struct super_operations cgroup_ops = {
1316        .statfs = simple_statfs,
1317        .drop_inode = generic_delete_inode,
1318        .show_options = cgroup_show_options,
1319        .remount_fs = cgroup_remount,
1320};
1321
1322static void init_cgroup_housekeeping(struct cgroup *cgrp)
1323{
1324        INIT_LIST_HEAD(&cgrp->sibling);
1325        INIT_LIST_HEAD(&cgrp->children);
1326        INIT_LIST_HEAD(&cgrp->css_sets);
1327        INIT_LIST_HEAD(&cgrp->release_list);
1328        INIT_LIST_HEAD(&cgrp->pidlists);
1329        mutex_init(&cgrp->pidlist_mutex);
1330        INIT_LIST_HEAD(&cgrp->event_list);
1331        spin_lock_init(&cgrp->event_list_lock);
1332}
1333
1334static void init_cgroup_root(struct cgroupfs_root *root)
1335{
1336        struct cgroup *cgrp = &root->top_cgroup;
1337        INIT_LIST_HEAD(&root->subsys_list);
1338        INIT_LIST_HEAD(&root->root_list);
1339        root->number_of_cgroups = 1;
1340        cgrp->root = root;
1341        cgrp->top_cgroup = cgrp;
1342        init_cgroup_housekeeping(cgrp);
1343}
1344
1345static bool init_root_id(struct cgroupfs_root *root)
1346{
1347        int ret = 0;
1348
1349        do {
1350                if (!ida_pre_get(&hierarchy_ida, GFP_KERNEL))
1351                        return false;
1352                spin_lock(&hierarchy_id_lock);
1353                /* Try to allocate the next unused ID */
1354                ret = ida_get_new_above(&hierarchy_ida, next_hierarchy_id,
1355                                        &root->hierarchy_id);
1356                if (ret == -ENOSPC)
1357                        /* Try again starting from 0 */
1358                        ret = ida_get_new(&hierarchy_ida, &root->hierarchy_id);
1359                if (!ret) {
1360                        next_hierarchy_id = root->hierarchy_id + 1;
1361                } else if (ret != -EAGAIN) {
1362                        /* Can only get here if the 31-bit IDR is full ... */
1363                        BUG_ON(ret);
1364                }
1365                spin_unlock(&hierarchy_id_lock);
1366        } while (ret);
1367        return true;
1368}
1369
1370static int cgroup_test_super(struct super_block *sb, void *data)
1371{
1372        struct cgroup_sb_opts *opts = data;
1373        struct cgroupfs_root *root = sb->s_fs_info;
1374
1375        /* If we asked for a name then it must match */
1376        if (opts->name && strcmp(opts->name, root->name))
1377                return 0;
1378
1379        /*
1380         * If we asked for subsystems (or explicitly for no
1381         * subsystems) then they must match
1382         */
1383        if ((opts->subsys_bits || opts->none)
1384            && (opts->subsys_bits != root->subsys_bits))
1385                return 0;
1386
1387        return 1;
1388}
1389
1390static struct cgroupfs_root *cgroup_root_from_opts(struct cgroup_sb_opts *opts)
1391{
1392        struct cgroupfs_root *root;
1393
1394        if (!opts->subsys_bits && !opts->none)
1395                return NULL;
1396
1397        root = kzalloc(sizeof(*root), GFP_KERNEL);
1398        if (!root)
1399                return ERR_PTR(-ENOMEM);
1400
1401        if (!init_root_id(root)) {
1402                kfree(root);
1403                return ERR_PTR(-ENOMEM);
1404        }
1405        init_cgroup_root(root);
1406
1407        root->subsys_bits = opts->subsys_bits;
1408        root->flags = opts->flags;
1409        if (opts->release_agent)
1410                strcpy(root->release_agent_path, opts->release_agent);
1411        if (opts->name)
1412                strcpy(root->name, opts->name);
1413        if (opts->clone_children)
1414                set_bit(CGRP_CLONE_CHILDREN, &root->top_cgroup.flags);
1415        return root;
1416}
1417
1418static void cgroup_drop_root(struct cgroupfs_root *root)
1419{
1420        if (!root)
1421                return;
1422
1423        BUG_ON(!root->hierarchy_id);
1424        spin_lock(&hierarchy_id_lock);
1425        ida_remove(&hierarchy_ida, root->hierarchy_id);
1426        spin_unlock(&hierarchy_id_lock);
1427        kfree(root);
1428}
1429
1430static int cgroup_set_super(struct super_block *sb, void *data)
1431{
1432        int ret;
1433        struct cgroup_sb_opts *opts = data;
1434
1435        /* If we don't have a new root, we can't set up a new sb */
1436        if (!opts->new_root)
1437                return -EINVAL;
1438
1439        BUG_ON(!opts->subsys_bits && !opts->none);
1440
1441        ret = set_anon_super(sb, NULL);
1442        if (ret)
1443                return ret;
1444
1445        sb->s_fs_info = opts->new_root;
1446        opts->new_root->sb = sb;
1447
1448        sb->s_blocksize = PAGE_CACHE_SIZE;
1449        sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
1450        sb->s_magic = CGROUP_SUPER_MAGIC;
1451        sb->s_op = &cgroup_ops;
1452
1453        return 0;
1454}
1455
1456static int cgroup_get_rootdir(struct super_block *sb)
1457{
1458        static const struct dentry_operations cgroup_dops = {
1459                .d_iput = cgroup_diput,
1460                .d_delete = cgroup_delete,
1461        };
1462
1463        struct inode *inode =
1464                cgroup_new_inode(S_IFDIR | S_IRUGO | S_IXUGO | S_IWUSR, sb);
1465        struct dentry *dentry;
1466
1467        if (!inode)
1468                return -ENOMEM;
1469
1470        inode->i_fop = &simple_dir_operations;
1471        inode->i_op = &cgroup_dir_inode_operations;
1472        /* directories start off with i_nlink == 2 (for "." entry) */
1473        inc_nlink(inode);
1474        dentry = d_alloc_root(inode);
1475        if (!dentry) {
1476                iput(inode);
1477                return -ENOMEM;
1478        }
1479        sb->s_root = dentry;
1480        /* for everything else we want ->d_op set */
1481        sb->s_d_op = &cgroup_dops;
1482        return 0;
1483}
1484
1485static struct dentry *cgroup_mount(struct file_system_type *fs_type,
1486                         int flags, const char *unused_dev_name,
1487                         void *data)
1488{
1489        struct cgroup_sb_opts opts;
1490        struct cgroupfs_root *root;
1491        int ret = 0;
1492        struct super_block *sb;
1493        struct cgroupfs_root *new_root;
1494
1495        /* First find the desired set of subsystems */
1496        mutex_lock(&cgroup_mutex);
1497        ret = parse_cgroupfs_options(data, &opts);
1498        mutex_unlock(&cgroup_mutex);
1499        if (ret)
1500                goto out_err;
1501
1502        /*
1503         * Allocate a new cgroup root. We may not need it if we're
1504         * reusing an existing hierarchy.
1505         */
1506        new_root = cgroup_root_from_opts(&opts);
1507        if (IS_ERR(new_root)) {
1508                ret = PTR_ERR(new_root);
1509                goto drop_modules;
1510        }
1511        opts.new_root = new_root;
1512
1513        /* Locate an existing or new sb for this hierarchy */
1514        sb = sget(fs_type, cgroup_test_super, cgroup_set_super, &opts);
1515        if (IS_ERR(sb)) {
1516                ret = PTR_ERR(sb);
1517                cgroup_drop_root(opts.new_root);
1518                goto drop_modules;
1519        }
1520
1521        root = sb->s_fs_info;
1522        BUG_ON(!root);
1523        if (root == opts.new_root) {
1524                /* We used the new root structure, so this is a new hierarchy */
1525                struct list_head tmp_cg_links;
1526                struct cgroup *root_cgrp = &root->top_cgroup;
1527                struct inode *inode;
1528                struct cgroupfs_root *existing_root;
1529                int i;
1530
1531                BUG_ON(sb->s_root != NULL);
1532
1533                ret = cgroup_get_rootdir(sb);
1534                if (ret)
1535                        goto drop_new_super;
1536                inode = sb->s_root->d_inode;
1537
1538                mutex_lock(&inode->i_mutex);
1539                mutex_lock(&cgroup_mutex);
1540
1541                if (strlen(root->name)) {
1542                        /* Check for name clashes with existing mounts */
1543                        for_each_active_root(existing_root) {
1544                                if (!strcmp(existing_root->name, root->name)) {
1545                                        ret = -EBUSY;
1546                                        mutex_unlock(&cgroup_mutex);
1547                                        mutex_unlock(&inode->i_mutex);
1548                                        goto drop_new_super;
1549                                }
1550                        }
1551                }
1552
1553                /*
1554                 * We're accessing css_set_count without locking
1555                 * css_set_lock here, but that's OK - it can only be
1556                 * increased by someone holding cgroup_lock, and
1557                 * that's us. The worst that can happen is that we
1558                 * have some link structures left over
1559                 */
1560                ret = allocate_cg_links(css_set_count, &tmp_cg_links);
1561                if (ret) {
1562                        mutex_unlock(&cgroup_mutex);
1563                        mutex_unlock(&inode->i_mutex);
1564                        goto drop_new_super;
1565                }
1566
1567                ret = rebind_subsystems(root, root->subsys_bits);
1568                if (ret == -EBUSY) {
1569                        mutex_unlock(&cgroup_mutex);
1570                        mutex_unlock(&inode->i_mutex);
1571                        free_cg_links(&tmp_cg_links);
1572                        goto drop_new_super;
1573                }
1574                /*
1575                 * There must be no failure case after here, since rebinding
1576                 * takes care of subsystems' refcounts, which are explicitly
1577                 * dropped in the failure exit path.
1578                 */
1579
1580                /* EBUSY should be the only error here */
1581                BUG_ON(ret);
1582
1583                list_add(&root->root_list, &roots);
1584                root_count++;
1585
1586                sb->s_root->d_fsdata = root_cgrp;
1587                root->top_cgroup.dentry = sb->s_root;
1588
1589                /* Link the top cgroup in this hierarchy into all
1590                 * the css_set objects */
1591                write_lock(&css_set_lock);
1592                for (i = 0; i < CSS_SET_TABLE_SIZE; i++) {
1593                        struct hlist_head *hhead = &css_set_table[i];
1594                        struct hlist_node *node;
1595                        struct css_set *cg;
1596
1597                        hlist_for_each_entry(cg, node, hhead, hlist)
1598                                link_css_set(&tmp_cg_links, cg, root_cgrp);
1599                }
1600                write_unlock(&css_set_lock);
1601
1602                free_cg_links(&tmp_cg_links);
1603
1604                BUG_ON(!list_empty(&root_cgrp->sibling));
1605                BUG_ON(!list_empty(&root_cgrp->children));
1606                BUG_ON(root->number_of_cgroups != 1);
1607
1608                cgroup_populate_dir(root_cgrp);
1609                mutex_unlock(&cgroup_mutex);
1610                mutex_unlock(&inode->i_mutex);
1611        } else {
1612                /*
1613                 * We re-used an existing hierarchy - the new root (if
1614                 * any) is not needed
1615                 */
1616                cgroup_drop_root(opts.new_root);
1617                /* no subsys rebinding, so refcounts don't change */
1618                drop_parsed_module_refcounts(opts.subsys_bits);
1619        }
1620
1621        kfree(opts.release_agent);
1622        kfree(opts.name);
1623        return dget(sb->s_root);
1624
1625 drop_new_super:
1626        deactivate_locked_super(sb);
1627 drop_modules:
1628        drop_parsed_module_refcounts(opts.subsys_bits);
1629 out_err:
1630        kfree(opts.release_agent);
1631        kfree(opts.name);
1632        return ERR_PTR(ret);
1633}
1634
1635static void cgroup_kill_sb(struct super_block *sb) {
1636        struct cgroupfs_root *root = sb->s_fs_info;
1637        struct cgroup *cgrp = &root->top_cgroup;
1638        int ret;
1639        struct cg_cgroup_link *link;
1640        struct cg_cgroup_link *saved_link;
1641
1642        BUG_ON(!root);
1643
1644        BUG_ON(root->number_of_cgroups != 1);
1645        BUG_ON(!list_empty(&cgrp->children));
1646        BUG_ON(!list_empty(&cgrp->sibling));
1647
1648        mutex_lock(&cgroup_mutex);
1649
1650        /* Rebind all subsystems back to the default hierarchy */
1651        ret = rebind_subsystems(root, 0);
1652        /* Shouldn't be able to fail ... */
1653        BUG_ON(ret);
1654
1655        /*
1656         * Release all the links from css_sets to this hierarchy's
1657         * root cgroup
1658         */
1659        write_lock(&css_set_lock);
1660
1661        list_for_each_entry_safe(link, saved_link, &cgrp->css_sets,
1662                                 cgrp_link_list) {
1663                list_del(&link->cg_link_list);
1664                list_del(&link->cgrp_link_list);
1665                kfree(link);
1666        }
1667        write_unlock(&css_set_lock);
1668
1669        if (!list_empty(&root->root_list)) {
1670                list_del(&root->root_list);
1671                root_count--;
1672        }
1673
1674        mutex_unlock(&cgroup_mutex);
1675
1676        kill_litter_super(sb);
1677        cgroup_drop_root(root);
1678}
1679
1680static struct file_system_type cgroup_fs_type = {
1681        .name = "cgroup",
1682        .mount = cgroup_mount,
1683        .kill_sb = cgroup_kill_sb,
1684};
1685
1686static struct kobject *cgroup_kobj;
1687
1688static inline struct cgroup *__d_cgrp(struct dentry *dentry)
1689{
1690        return dentry->d_fsdata;
1691}
1692
1693static inline struct cftype *__d_cft(struct dentry *dentry)
1694{
1695        return dentry->d_fsdata;
1696}
1697
1698/**
1699 * cgroup_path - generate the path of a cgroup
1700 * @cgrp: the cgroup in question
1701 * @buf: the buffer to write the path into
1702 * @buflen: the length of the buffer
1703 *
1704 * Called with cgroup_mutex held or else with an RCU-protected cgroup
1705 * reference.  Writes path of cgroup into buf.  Returns 0 on success,
1706 * -errno on error.
1707 */
1708int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen)
1709{
1710        char *start;
1711        struct dentry *dentry = rcu_dereference_check(cgrp->dentry,
1712                                                      rcu_read_lock_held() ||
1713                                                      cgroup_lock_is_held());
1714
1715        if (!dentry || cgrp == dummytop) {
1716                /*
1717                 * Inactive subsystems have no dentry for their root
1718                 * cgroup
1719                 */
1720                strcpy(buf, "/");
1721                return 0;
1722        }
1723
1724        start = buf + buflen;
1725
1726        *--start = '\0';
1727        for (;;) {
1728                int len = dentry->d_name.len;
1729
1730                if ((start -= len) < buf)
1731                        return -ENAMETOOLONG;
1732                memcpy(start, dentry->d_name.name, len);
1733                cgrp = cgrp->parent;
1734                if (!cgrp)
1735                        break;
1736
1737                dentry = rcu_dereference_check(cgrp->dentry,
1738                                               rcu_read_lock_held() ||
1739                                               cgroup_lock_is_held());
1740                if (!cgrp->parent)
1741                        continue;
1742                if (--start < buf)
1743                        return -ENAMETOOLONG;
1744                *start = '/';
1745        }
1746        memmove(buf, start, buf + buflen - start);
1747        return 0;
1748}
1749EXPORT_SYMBOL_GPL(cgroup_path);
1750
1751/**
1752 * cgroup_attach_task - attach task 'tsk' to cgroup 'cgrp'
1753 * @cgrp: the cgroup the task is attaching to
1754 * @tsk: the task to be attached
1755 *
1756 * Call holding cgroup_mutex. May take task_lock of
1757 * the task 'tsk' during call.
1758 */
1759int cgroup_attach_task(struct cgroup *cgrp, struct task_struct *tsk)
1760{
1761        int retval = 0;
1762        struct cgroup_subsys *ss, *failed_ss = NULL;
1763        struct cgroup *oldcgrp;
1764        struct css_set *cg;
1765        struct css_set *newcg;
1766        struct cgroupfs_root *root = cgrp->root;
1767
1768        /* Nothing to do if the task is already in that cgroup */
1769        oldcgrp = task_cgroup_from_root(tsk, root);
1770        if (cgrp == oldcgrp)
1771                return 0;
1772
1773        for_each_subsys(root, ss) {
1774                if (ss->can_attach) {
1775                        retval = ss->can_attach(ss, cgrp, tsk, false);
1776                        if (retval) {
1777                                /*
1778                                 * Remember on which subsystem the can_attach()
1779                                 * failed, so that we only call cancel_attach()
1780                                 * against the subsystems whose can_attach()
1781                                 * succeeded. (See below)
1782                                 */
1783                                failed_ss = ss;
1784                                goto out;
1785                        }
1786                }
1787        }
1788
1789        task_lock(tsk);
1790        cg = tsk->cgroups;
1791        get_css_set(cg);
1792        task_unlock(tsk);
1793        /*
1794         * Locate or allocate a new css_set for this task,
1795         * based on its final set of cgroups
1796         */
1797        newcg = find_css_set(cg, cgrp);
1798        put_css_set(cg);
1799        if (!newcg) {
1800                retval = -ENOMEM;
1801                goto out;
1802        }
1803
1804        task_lock(tsk);
1805        if (tsk->flags & PF_EXITING) {
1806                task_unlock(tsk);
1807                put_css_set(newcg);
1808                retval = -ESRCH;
1809                goto out;
1810        }
1811        rcu_assign_pointer(tsk->cgroups, newcg);
1812        task_unlock(tsk);
1813
1814        /* Update the css_set linked lists if we're using them */
1815        write_lock(&css_set_lock);
1816        if (!list_empty(&tsk->cg_list))
1817                list_move(&tsk->cg_list, &newcg->tasks);
1818        write_unlock(&css_set_lock);
1819
1820        for_each_subsys(root, ss) {
1821                if (ss->attach)
1822                        ss->attach(ss, cgrp, oldcgrp, tsk, false);
1823        }
1824        set_bit(CGRP_RELEASABLE, &oldcgrp->flags);
1825        synchronize_rcu();
1826        put_css_set(cg);
1827
1828        /*
1829         * wake up rmdir() waiter. the rmdir should fail since the cgroup
1830         * is no longer empty.
1831         */
1832        cgroup_wakeup_rmdir_waiter(cgrp);
1833out:
1834        if (retval) {
1835                for_each_subsys(root, ss) {
1836                        if (ss == failed_ss)
1837                                /*
1838                                 * This subsystem was the one that failed the
1839                                 * can_attach() check earlier, so we don't need
1840                                 * to call cancel_attach() against it or any
1841                                 * remaining subsystems.
1842                                 */
1843                                break;
1844                        if (ss->cancel_attach)
1845                                ss->cancel_attach(ss, cgrp, tsk, false);
1846                }
1847        }
1848        return retval;
1849}
1850
1851/**
1852 * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from'
1853 * @from: attach to all cgroups of a given task
1854 * @tsk: the task to be attached
1855 */
1856int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk)
1857{
1858        struct cgroupfs_root *root;
1859        int retval = 0;
1860
1861        cgroup_lock();
1862        for_each_active_root(root) {
1863                struct cgroup *from_cg = task_cgroup_from_root(from, root);
1864
1865                retval = cgroup_attach_task(from_cg, tsk);
1866                if (retval)
1867                        break;
1868        }
1869        cgroup_unlock();
1870
1871        return retval;
1872}
1873EXPORT_SYMBOL_GPL(cgroup_attach_task_all);
1874
1875/*
1876 * Attach task with pid 'pid' to cgroup 'cgrp'. Call with cgroup_mutex
1877 * held. May take task_lock of task
1878 */
1879static int attach_task_by_pid(struct cgroup *cgrp, u64 pid)
1880{
1881        struct task_struct *tsk;
1882        const struct cred *cred = current_cred(), *tcred;
1883        int ret;
1884
1885        if (pid) {
1886                rcu_read_lock();
1887                tsk = find_task_by_vpid(pid);
1888                if (!tsk || tsk->flags & PF_EXITING) {
1889                        rcu_read_unlock();
1890                        return -ESRCH;
1891                }
1892
1893                tcred = __task_cred(tsk);
1894                if (cred->euid &&
1895                    cred->euid != tcred->uid &&
1896                    cred->euid != tcred->suid) {
1897                        rcu_read_unlock();
1898                        return -EACCES;
1899                }
1900                get_task_struct(tsk);
1901                rcu_read_unlock();
1902        } else {
1903                tsk = current;
1904                get_task_struct(tsk);
1905        }
1906
1907        ret = cgroup_attach_task(cgrp, tsk);
1908        put_task_struct(tsk);
1909        return ret;
1910}
1911
1912static int cgroup_tasks_write(struct cgroup *cgrp, struct cftype *cft, u64 pid)
1913{
1914        int ret;
1915        if (!cgroup_lock_live_group(cgrp))
1916                return -ENODEV;
1917        ret = attach_task_by_pid(cgrp, pid);
1918        cgroup_unlock();
1919        return ret;
1920}
1921
1922/**
1923 * cgroup_lock_live_group - take cgroup_mutex and check that cgrp is alive.
1924 * @cgrp: the cgroup to be checked for liveness
1925 *
1926 * On success, returns true; the lock should be later released with
1927 * cgroup_unlock(). On failure returns false with no lock held.
1928 */
1929bool cgroup_lock_live_group(struct cgroup *cgrp)
1930{
1931        mutex_lock(&cgroup_mutex);
1932        if (cgroup_is_removed(cgrp)) {
1933                mutex_unlock(&cgroup_mutex);
1934                return false;
1935        }
1936        return true;
1937}
1938EXPORT_SYMBOL_GPL(cgroup_lock_live_group);
1939
1940static int cgroup_release_agent_write(struct cgroup *cgrp, struct cftype *cft,
1941                                      const char *buffer)
1942{
1943        BUILD_BUG_ON(sizeof(cgrp->root->release_agent_path) < PATH_MAX);
1944        if (strlen(buffer) >= PATH_MAX)
1945                return -EINVAL;
1946        if (!cgroup_lock_live_group(cgrp))
1947                return -ENODEV;
1948        strcpy(cgrp->root->release_agent_path, buffer);
1949        cgroup_unlock();
1950        return 0;
1951}
1952
1953static int cgroup_release_agent_show(struct cgroup *cgrp, struct cftype *cft,
1954                                     struct seq_file *seq)
1955{
1956        if (!cgroup_lock_live_group(cgrp))
1957                return -ENODEV;
1958        seq_puts(seq, cgrp->root->release_agent_path);
1959        seq_putc(seq, '\n');
1960        cgroup_unlock();
1961        return 0;
1962}
1963
1964/* A buffer size big enough for numbers or short strings */
1965#define CGROUP_LOCAL_BUFFER_SIZE 64
1966
1967static ssize_t cgroup_write_X64(struct cgroup *cgrp, struct cftype *cft,
1968                                struct file *file,
1969                                const char __user *userbuf,
1970                                size_t nbytes, loff_t *unused_ppos)
1971{
1972        char buffer[CGROUP_LOCAL_BUFFER_SIZE];
1973        int retval = 0;
1974        char *end;
1975
1976        if (!nbytes)
1977                return -EINVAL;
1978        if (nbytes >= sizeof(buffer))
1979                return -E2BIG;
1980        if (copy_from_user(buffer, userbuf, nbytes))
1981                return -EFAULT;
1982
1983        buffer[nbytes] = 0;     /* nul-terminate */
1984        if (cft->write_u64) {
1985                u64 val = simple_strtoull(strstrip(buffer), &end, 0);
1986                if (*end)
1987                        return -EINVAL;
1988                retval = cft->write_u64(cgrp, cft, val);
1989        } else {
1990                s64 val = simple_strtoll(strstrip(buffer), &end, 0);
1991                if (*end)
1992                        return -EINVAL;
1993                retval = cft->write_s64(cgrp, cft, val);
1994        }
1995        if (!retval)
1996                retval = nbytes;
1997        return retval;
1998}
1999
2000static ssize_t cgroup_write_string(struct cgroup *cgrp, struct cftype *cft,
2001                                   struct file *file,
2002                                   const char __user *userbuf,
2003                                   size_t nbytes, loff_t *unused_ppos)
2004{
2005        char local_buffer[CGROUP_LOCAL_BUFFER_SIZE];
2006        int retval = 0;
2007        size_t max_bytes = cft->max_write_len;
2008        char *buffer = local_buffer;
2009
2010        if (!max_bytes)
2011                max_bytes = sizeof(local_buffer) - 1;
2012        if (nbytes >= max_bytes)
2013                return -E2BIG;
2014        /* Allocate a dynamic buffer if we need one */
2015        if (nbytes >= sizeof(local_buffer)) {
2016                buffer = kmalloc(nbytes + 1, GFP_KERNEL);
2017                if (buffer == NULL)
2018                        return -ENOMEM;
2019        }
2020        if (nbytes && copy_from_user(buffer, userbuf, nbytes)) {
2021                retval = -EFAULT;
2022                goto out;
2023        }
2024
2025        buffer[nbytes] = 0;     /* nul-terminate */
2026        retval = cft->write_string(cgrp, cft, strstrip(buffer));
2027        if (!retval)
2028                retval = nbytes;
2029out:
2030        if (buffer != local_buffer)
2031                kfree(buffer);
2032        return retval;
2033}
2034
2035static ssize_t cgroup_file_write(struct file *file, const char __user *buf,
2036                                                size_t nbytes, loff_t *ppos)
2037{
2038        struct cftype *cft = __d_cft(file->f_dentry);
2039        struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
2040
2041        if (cgroup_is_removed(cgrp))
2042                return -ENODEV;
2043        if (cft->write)
2044                return cft->write(cgrp, cft, file, buf, nbytes, ppos);
2045        if (cft->write_u64 || cft->write_s64)
2046                return cgroup_write_X64(cgrp, cft, file, buf, nbytes, ppos);
2047        if (cft->write_string)
2048                return cgroup_write_string(cgrp, cft, file, buf, nbytes, ppos);
2049        if (cft->trigger) {
2050                int ret = cft->trigger(cgrp, (unsigned int)cft->private);
2051                return ret ? ret : nbytes;
2052        }
2053        return -EINVAL;
2054}
2055
2056static ssize_t cgroup_read_u64(struct cgroup *cgrp, struct cftype *cft,
2057                               struct file *file,
2058                               char __user *buf, size_t nbytes,
2059                               loff_t *ppos)
2060{
2061        char tmp[CGROUP_LOCAL_BUFFER_SIZE];
2062        u64 val = cft->read_u64(cgrp, cft);
2063        int len = sprintf(tmp, "%llu\n", (unsigned long long) val);
2064
2065        return simple_read_from_buffer(buf, nbytes, ppos, tmp, len);
2066}
2067
2068static ssize_t cgroup_read_s64(struct cgroup *cgrp, struct cftype *cft,
2069                               struct file *file,
2070                               char __user *buf, size_t nbytes,
2071                               loff_t *ppos)
2072{
2073        char tmp[CGROUP_LOCAL_BUFFER_SIZE];
2074        s64 val = cft->read_s64(cgrp, cft);
2075        int len = sprintf(tmp, "%lld\n", (long long) val);
2076
2077        return simple_read_from_buffer(buf, nbytes, ppos, tmp, len);
2078}
2079
2080static ssize_t cgroup_file_read(struct file *file, char __user *buf,
2081                                   size_t nbytes, loff_t *ppos)
2082{
2083        struct cftype *cft = __d_cft(file->f_dentry);
2084        struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
2085
2086        if (cgroup_is_removed(cgrp))
2087                return -ENODEV;
2088
2089        if (cft->read)
2090                return cft->read(cgrp, cft, file, buf, nbytes, ppos);
2091        if (cft->read_u64)
2092                return cgroup_read_u64(cgrp, cft, file, buf, nbytes, ppos);
2093        if (cft->read_s64)
2094                return cgroup_read_s64(cgrp, cft, file, buf, nbytes, ppos);
2095        return -EINVAL;
2096}
2097
2098/*
2099 * seqfile ops/methods for returning structured data. Currently just
2100 * supports string->u64 maps, but can be extended in future.
2101 */
2102
2103struct cgroup_seqfile_state {
2104        struct cftype *cft;
2105        struct cgroup *cgroup;
2106};
2107
2108static int cgroup_map_add(struct cgroup_map_cb *cb, const char *key, u64 value)
2109{
2110        struct seq_file *sf = cb->state;
2111        return seq_printf(sf, "%s %llu\n", key, (unsigned long long)value);
2112}
2113
2114static int cgroup_seqfile_show(struct seq_file *m, void *arg)
2115{
2116        struct cgroup_seqfile_state *state = m->private;
2117        struct cftype *cft = state->cft;
2118        if (cft->read_map) {
2119                struct cgroup_map_cb cb = {
2120                        .fill = cgroup_map_add,
2121                        .state = m,
2122                };
2123                return cft->read_map(state->cgroup, cft, &cb);
2124        }
2125        return cft->read_seq_string(state->cgroup, cft, m);
2126}
2127
2128static int cgroup_seqfile_release(struct inode *inode, struct file *file)
2129{
2130        struct seq_file *seq = file->private_data;
2131        kfree(seq->private);
2132        return single_release(inode, file);
2133}
2134
2135static const struct file_operations cgroup_seqfile_operations = {
2136        .read = seq_read,
2137        .write = cgroup_file_write,
2138        .llseek = seq_lseek,
2139        .release = cgroup_seqfile_release,
2140};
2141
2142static int cgroup_file_open(struct inode *inode, struct file *file)
2143{
2144        int err;
2145        struct cftype *cft;
2146
2147        err = generic_file_open(inode, file);
2148        if (err)
2149                return err;
2150        cft = __d_cft(file->f_dentry);
2151
2152        if (cft->read_map || cft->read_seq_string) {
2153                struct cgroup_seqfile_state *state =
2154                        kzalloc(sizeof(*state), GFP_USER);
2155                if (!state)
2156                        return -ENOMEM;
2157                state->cft = cft;
2158                state->cgroup = __d_cgrp(file->f_dentry->d_parent);
2159                file->f_op = &cgroup_seqfile_operations;
2160                err = single_open(file, cgroup_seqfile_show, state);
2161                if (err < 0)
2162                        kfree(state);
2163        } else if (cft->open)
2164                err = cft->open(inode, file);
2165        else
2166                err = 0;
2167
2168        return err;
2169}
2170
2171static int cgroup_file_release(struct inode *inode, struct file *file)
2172{
2173        struct cftype *cft = __d_cft(file->f_dentry);
2174        if (cft->release)
2175                return cft->release(inode, file);
2176        return 0;
2177}
2178
2179/*
2180 * cgroup_rename - Only allow simple rename of directories in place.
2181 */
2182static int cgroup_rename(struct inode *old_dir, struct dentry *old_dentry,
2183                            struct inode *new_dir, struct dentry *new_dentry)
2184{
2185        if (!S_ISDIR(old_dentry->d_inode->i_mode))
2186                return -ENOTDIR;
2187        if (new_dentry->d_inode)
2188                return -EEXIST;
2189        if (old_dir != new_dir)
2190                return -EIO;
2191        return simple_rename(old_dir, old_dentry, new_dir, new_dentry);
2192}
2193
2194static const struct file_operations cgroup_file_operations = {
2195        .read = cgroup_file_read,
2196        .write = cgroup_file_write,
2197        .llseek = generic_file_llseek,
2198        .open = cgroup_file_open,
2199        .release = cgroup_file_release,
2200};
2201
2202static const struct inode_operations cgroup_dir_inode_operations = {
2203        .lookup = cgroup_lookup,
2204        .mkdir = cgroup_mkdir,
2205        .rmdir = cgroup_rmdir,
2206        .rename = cgroup_rename,
2207};
2208
2209static struct dentry *cgroup_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
2210{
2211        if (dentry->d_name.len > NAME_MAX)
2212                return ERR_PTR(-ENAMETOOLONG);
2213        d_add(dentry, NULL);
2214        return NULL;
2215}
2216
2217/*
2218 * Check if a file is a control file
2219 */
2220static inline struct cftype *__file_cft(struct file *file)
2221{
2222        if (file->f_dentry->d_inode->i_fop != &cgroup_file_operations)
2223                return ERR_PTR(-EINVAL);
2224        return __d_cft(file->f_dentry);
2225}
2226
2227static int cgroup_create_file(struct dentry *dentry, mode_t mode,
2228                                struct super_block *sb)
2229{
2230        struct inode *inode;
2231
2232        if (!dentry)
2233                return -ENOENT;
2234        if (dentry->d_inode)
2235                return -EEXIST;
2236
2237        inode = cgroup_new_inode(mode, sb);
2238        if (!inode)
2239                return -ENOMEM;
2240
2241        if (S_ISDIR(mode)) {
2242                inode->i_op = &cgroup_dir_inode_operations;
2243                inode->i_fop = &simple_dir_operations;
2244
2245                /* start off with i_nlink == 2 (for "." entry) */
2246                inc_nlink(inode);
2247
2248                /* start with the directory inode held, so that we can
2249                 * populate it without racing with another mkdir */
2250                mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
2251        } else if (S_ISREG(mode)) {
2252                inode->i_size = 0;
2253                inode->i_fop = &cgroup_file_operations;
2254        }
2255        d_instantiate(dentry, inode);
2256        dget(dentry);   /* Extra count - pin the dentry in core */
2257        return 0;
2258}
2259
2260/*
2261 * cgroup_create_dir - create a directory for an object.
2262 * @cgrp: the cgroup we create the directory for. It must have a valid
2263 *        ->parent field. And we are going to fill its ->dentry field.
2264 * @dentry: dentry of the new cgroup
2265 * @mode: mode to set on new directory.
2266 */
2267static int cgroup_create_dir(struct cgroup *cgrp, struct dentry *dentry,
2268                                mode_t mode)
2269{
2270        struct dentry *parent;
2271        int error = 0;
2272
2273        parent = cgrp->parent->dentry;
2274        error = cgroup_create_file(dentry, S_IFDIR | mode, cgrp->root->sb);
2275        if (!error) {
2276                dentry->d_fsdata = cgrp;
2277                inc_nlink(parent->d_inode);
2278                rcu_assign_pointer(cgrp->dentry, dentry);
2279                dget(dentry);
2280        }
2281        dput(dentry);
2282
2283        return error;
2284}
2285
2286/**
2287 * cgroup_file_mode - deduce file mode of a control file
2288 * @cft: the control file in question
2289 *
2290 * returns cft->mode if ->mode is not 0
2291 * returns S_IRUGO|S_IWUSR if it has both a read and a write handler
2292 * returns S_IRUGO if it has only a read handler
2293 * returns S_IWUSR if it has only a write hander
2294 */
2295static mode_t cgroup_file_mode(const struct cftype *cft)
2296{
2297        mode_t mode = 0;
2298
2299        if (cft->mode)
2300                return cft->mode;
2301
2302        if (cft->read || cft->read_u64 || cft->read_s64 ||
2303            cft->read_map || cft->read_seq_string)
2304                mode |= S_IRUGO;
2305
2306        if (cft->write || cft->write_u64 || cft->write_s64 ||
2307            cft->write_string || cft->trigger)
2308                mode |= S_IWUSR;
2309
2310        return mode;
2311}
2312
2313int cgroup_add_file(struct cgroup *cgrp,
2314                       struct cgroup_subsys *subsys,
2315                       const struct cftype *cft)
2316{
2317        struct dentry *dir = cgrp->dentry;
2318        struct dentry *dentry;
2319        int error;
2320        mode_t mode;
2321
2322        char name[MAX_CGROUP_TYPE_NAMELEN + MAX_CFTYPE_NAME + 2] = { 0 };
2323        if (subsys && !test_bit(ROOT_NOPREFIX, &cgrp->root->flags)) {
2324                strcpy(name, subsys->name);
2325                strcat(name, ".");
2326        }
2327        strcat(name, cft->name);
2328        BUG_ON(!mutex_is_locked(&dir->d_inode->i_mutex));
2329        dentry = lookup_one_len(name, dir, strlen(name));
2330        if (!IS_ERR(dentry)) {
2331                mode = cgroup_file_mode(cft);
2332                error = cgroup_create_file(dentry, mode | S_IFREG,
2333                                                cgrp->root->sb);
2334                if (!error)
2335                        dentry->d_fsdata = (void *)cft;
2336                dput(dentry);
2337        } else
2338                error = PTR_ERR(dentry);
2339        return error;
2340}
2341EXPORT_SYMBOL_GPL(cgroup_add_file);
2342
2343int cgroup_add_files(struct cgroup *cgrp,
2344                        struct cgroup_subsys *subsys,
2345                        const struct cftype cft[],
2346                        int count)
2347{
2348        int i, err;
2349        for (i = 0; i < count; i++) {
2350                err = cgroup_add_file(cgrp, subsys, &cft[i]);
2351                if (err)
2352                        return err;
2353        }
2354        return 0;
2355}
2356EXPORT_SYMBOL_GPL(cgroup_add_files);
2357
2358/**
2359 * cgroup_task_count - count the number of tasks in a cgroup.
2360 * @cgrp: the cgroup in question
2361 *
2362 * Return the number of tasks in the cgroup.
2363 */
2364int cgroup_task_count(const struct cgroup *cgrp)
2365{
2366        int count = 0;
2367        struct cg_cgroup_link *link;
2368
2369        read_lock(&css_set_lock);
2370        list_for_each_entry(link, &cgrp->css_sets, cgrp_link_list) {
2371                count += atomic_read(&link->cg->refcount);
2372        }
2373        read_unlock(&css_set_lock);
2374        return count;
2375}
2376
2377/*
2378 * Advance a list_head iterator.  The iterator should be positioned at
2379 * the start of a css_set
2380 */
2381static void cgroup_advance_iter(struct cgroup *cgrp,
2382                                struct cgroup_iter *it)
2383{
2384        struct list_head *l = it->cg_link;
2385        struct cg_cgroup_link *link;
2386        struct css_set *cg;
2387
2388        /* Advance to the next non-empty css_set */
2389        do {
2390                l = l->next;
2391                if (l == &cgrp->css_sets) {
2392                        it->cg_link = NULL;
2393                        return;
2394                }
2395                link = list_entry(l, struct cg_cgroup_link, cgrp_link_list);
2396                cg = link->cg;
2397        } while (list_empty(&cg->tasks));
2398        it->cg_link = l;
2399        it->task = cg->tasks.next;
2400}
2401
2402/*
2403 * To reduce the fork() overhead for systems that are not actually
2404 * using their cgroups capability, we don't maintain the lists running
2405 * through each css_set to its tasks until we see the list actually
2406 * used - in other words after the first call to cgroup_iter_start().
2407 *
2408 * The tasklist_lock is not held here, as do_each_thread() and
2409 * while_each_thread() are protected by RCU.
2410 */
2411static void cgroup_enable_task_cg_lists(void)
2412{
2413        struct task_struct *p, *g;
2414        write_lock(&css_set_lock);
2415        use_task_css_set_links = 1;
2416        do_each_thread(g, p) {
2417                task_lock(p);
2418                /*
2419                 * We should check if the process is exiting, otherwise
2420                 * it will race with cgroup_exit() in that the list
2421                 * entry won't be deleted though the process has exited.
2422                 */
2423                if (!(p->flags & PF_EXITING) && list_empty(&p->cg_list))
2424                        list_add(&p->cg_list, &p->cgroups->tasks);
2425                task_unlock(p);
2426        } while_each_thread(g, p);
2427        write_unlock(&css_set_lock);
2428}
2429
2430void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it)
2431{
2432        /*
2433         * The first time anyone tries to iterate across a cgroup,
2434         * we need to enable the list linking each css_set to its
2435         * tasks, and fix up all existing tasks.
2436         */
2437        if (!use_task_css_set_links)
2438                cgroup_enable_task_cg_lists();
2439
2440        read_lock(&css_set_lock);
2441        it->cg_link = &cgrp->css_sets;
2442        cgroup_advance_iter(cgrp, it);
2443}
2444
2445struct task_struct *cgroup_iter_next(struct cgroup *cgrp,
2446                                        struct cgroup_iter *it)
2447{
2448        struct task_struct *res;
2449        struct list_head *l = it->task;
2450        struct cg_cgroup_link *link;
2451
2452        /* If the iterator cg is NULL, we have no tasks */
2453        if (!it->cg_link)
2454                return NULL;
2455        res = list_entry(l, struct task_struct, cg_list);
2456        /* Advance iterator to find next entry */
2457        l = l->next;
2458        link = list_entry(it->cg_link, struct cg_cgroup_link, cgrp_link_list);
2459        if (l == &link->cg->tasks) {
2460                /* We reached the end of this task list - move on to
2461                 * the next cg_cgroup_link */
2462                cgroup_advance_iter(cgrp, it);
2463        } else {
2464                it->task = l;
2465        }
2466        return res;
2467}
2468
2469void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it)
2470{
2471        read_unlock(&css_set_lock);
2472}
2473
2474static inline int started_after_time(struct task_struct *t1,
2475                                     struct timespec *time,
2476                                     struct task_struct *t2)
2477{
2478        int start_diff = timespec_compare(&t1->start_time, time);
2479        if (start_diff > 0) {
2480                return 1;
2481        } else if (start_diff < 0) {
2482                return 0;
2483        } else {
2484                /*
2485                 * Arbitrarily, if two processes started at the same
2486                 * time, we'll say that the lower pointer value
2487                 * started first. Note that t2 may have exited by now
2488                 * so this may not be a valid pointer any longer, but
2489                 * that's fine - it still serves to distinguish
2490                 * between two tasks started (effectively) simultaneously.
2491                 */
2492                return t1 > t2;
2493        }
2494}
2495
2496/*
2497 * This function is a callback from heap_insert() and is used to order
2498 * the heap.
2499 * In this case we order the heap in descending task start time.
2500 */
2501static inline int started_after(void *p1, void *p2)
2502{
2503        struct task_struct *t1 = p1;
2504        struct task_struct *t2 = p2;
2505        return started_after_time(t1, &t2->start_time, t2);
2506}
2507
2508/**
2509 * cgroup_scan_tasks - iterate though all the tasks in a cgroup
2510 * @scan: struct cgroup_scanner containing arguments for the scan
2511 *
2512 * Arguments include pointers to callback functions test_task() and
2513 * process_task().
2514 * Iterate through all the tasks in a cgroup, calling test_task() for each,
2515 * and if it returns true, call process_task() for it also.
2516 * The test_task pointer may be NULL, meaning always true (select all tasks).
2517 * Effectively duplicates cgroup_iter_{start,next,end}()
2518 * but does not lock css_set_lock for the call to process_task().
2519 * The struct cgroup_scanner may be embedded in any structure of the caller's
2520 * creation.
2521 * It is guaranteed that process_task() will act on every task that
2522 * is a member of the cgroup for the duration of this call. This
2523 * function may or may not call process_task() for tasks that exit
2524 * or move to a different cgroup during the call, or are forked or
2525 * move into the cgroup during the call.
2526 *
2527 * Note that test_task() may be called with locks held, and may in some
2528 * situations be called multiple times for the same task, so it should
2529 * be cheap.
2530 * If the heap pointer in the struct cgroup_scanner is non-NULL, a heap has been
2531 * pre-allocated and will be used for heap operations (and its "gt" member will
2532 * be overwritten), else a temporary heap will be used (allocation of which
2533 * may cause this function to fail).
2534 */
2535int cgroup_scan_tasks(struct cgroup_scanner *scan)
2536{
2537        int retval, i;
2538        struct cgroup_iter it;
2539        struct task_struct *p, *dropped;
2540        /* Never dereference latest_task, since it's not refcounted */
2541        struct task_struct *latest_task = NULL;
2542        struct ptr_heap tmp_heap;
2543        struct ptr_heap *heap;
2544        struct timespec latest_time = { 0, 0 };
2545
2546        if (scan->heap) {
2547                /* The caller supplied our heap and pre-allocated its memory */
2548                heap = scan->heap;
2549                heap->gt = &started_after;
2550        } else {
2551                /* We need to allocate our own heap memory */
2552                heap = &tmp_heap;
2553                retval = heap_init(heap, PAGE_SIZE, GFP_KERNEL, &started_after);
2554                if (retval)
2555                        /* cannot allocate the heap */
2556                        return retval;
2557        }
2558
2559 again:
2560        /*
2561         * Scan tasks in the cgroup, using the scanner's "test_task" callback
2562         * to determine which are of interest, and using the scanner's
2563         * "process_task" callback to process any of them that need an update.
2564         * Since we don't want to hold any locks during the task updates,
2565         * gather tasks to be processed in a heap structure.
2566         * The heap is sorted by descending task start time.
2567         * If the statically-sized heap fills up, we overflow tasks that
2568         * started later, and in future iterations only consider tasks that
2569         * started after the latest task in the previous pass. This
2570         * guarantees forward progress and that we don't miss any tasks.
2571         */
2572        heap->size = 0;
2573        cgroup_iter_start(scan->cg, &it);
2574        while ((p = cgroup_iter_next(scan->cg, &it))) {
2575                /*
2576                 * Only affect tasks that qualify per the caller's callback,
2577                 * if he provided one
2578                 */
2579                if (scan->test_task && !scan->test_task(p, scan))
2580                        continue;
2581                /*
2582                 * Only process tasks that started after the last task
2583                 * we processed
2584                 */
2585                if (!started_after_time(p, &latest_time, latest_task))
2586                        continue;
2587                dropped = heap_insert(heap, p);
2588                if (dropped == NULL) {
2589                        /*
2590                         * The new task was inserted; the heap wasn't
2591                         * previously full
2592                         */
2593                        get_task_struct(p);
2594                } else if (dropped != p) {
2595                        /*
2596                         * The new task was inserted, and pushed out a
2597                         * different task
2598                         */
2599                        get_task_struct(p);
2600                        put_task_struct(dropped);
2601                }
2602                /*
2603                 * Else the new task was newer than anything already in
2604                 * the heap and wasn't inserted
2605                 */
2606        }
2607        cgroup_iter_end(scan->cg, &it);
2608
2609        if (heap->size) {
2610                for (i = 0; i < heap->size; i++) {
2611                        struct task_struct *q = heap->ptrs[i];
2612                        if (i == 0) {
2613                                latest_time = q->start_time;
2614                                latest_task = q;
2615                        }
2616                        /* Process the task per the caller's callback */
2617                        scan->process_task(q, scan);
2618                        put_task_struct(q);
2619                }
2620                /*
2621                 * If we had to process any tasks at all, scan again
2622                 * in case some of them were in the middle of forking
2623                 * children that didn't get processed.
2624                 * Not the most efficient way to do it, but it avoids
2625                 * having to take callback_mutex in the fork path
2626                 */
2627                goto again;
2628        }
2629        if (heap == &tmp_heap)
2630                heap_free(&tmp_heap);
2631        return 0;
2632}
2633
2634/*
2635 * Stuff for reading the 'tasks'/'procs' files.
2636 *
2637 * Reading this file can return large amounts of data if a cgroup has
2638 * *lots* of attached tasks. So it may need several calls to read(),
2639 * but we cannot guarantee that the information we produce is correct
2640 * unless we produce it entirely atomically.
2641 *
2642 */
2643
2644/*
2645 * The following two functions "fix" the issue where there are more pids
2646 * than kmalloc will give memory for; in such cases, we use vmalloc/vfree.
2647 * TODO: replace with a kernel-wide solution to this problem
2648 */
2649#define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2))
2650static void *pidlist_allocate(int count)
2651{
2652        if (PIDLIST_TOO_LARGE(count))
2653                return vmalloc(count * sizeof(pid_t));
2654        else
2655                return kmalloc(count * sizeof(pid_t), GFP_KERNEL);
2656}
2657static void pidlist_free(void *p)
2658{
2659        if (is_vmalloc_addr(p))
2660                vfree(p);
2661        else
2662                kfree(p);
2663}
2664static void *pidlist_resize(void *p, int newcount)
2665{
2666        void *newlist;
2667        /* note: if new alloc fails, old p will still be valid either way */
2668        if (is_vmalloc_addr(p)) {
2669                newlist = vmalloc(newcount * sizeof(pid_t));
2670                if (!newlist)
2671                        return NULL;
2672                memcpy(newlist, p, newcount * sizeof(pid_t));
2673                vfree(p);
2674        } else {
2675                newlist = krealloc(p, newcount * sizeof(pid_t), GFP_KERNEL);
2676        }
2677        return newlist;
2678}
2679
2680/*
2681 * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
2682 * If the new stripped list is sufficiently smaller and there's enough memory
2683 * to allocate a new buffer, will let go of the unneeded memory. Returns the
2684 * number of unique elements.
2685 */
2686/* is the size difference enough that we should re-allocate the array? */
2687#define PIDLIST_REALLOC_DIFFERENCE(old, new) ((old) - PAGE_SIZE >= (new))
2688static int pidlist_uniq(pid_t **p, int length)
2689{
2690        int src, dest = 1;
2691        pid_t *list = *p;
2692        pid_t *newlist;
2693
2694        /*
2695         * we presume the 0th element is unique, so i starts at 1. trivial
2696         * edge cases first; no work needs to be done for either
2697         */
2698        if (length == 0 || length == 1)
2699                return length;
2700        /* src and dest walk down the list; dest counts unique elements */
2701        for (src = 1; src < length; src++) {
2702                /* find next unique element */
2703                while (list[src] == list[src-1]) {
2704                        src++;
2705                        if (src == length)
2706                                goto after;
2707                }
2708                /* dest always points to where the next unique element goes */
2709                list[dest] = list[src];
2710                dest++;
2711        }
2712after:
2713        /*
2714         * if the length difference is large enough, we want to allocate a
2715         * smaller buffer to save memory. if this fails due to out of memory,
2716         * we'll just stay with what we've got.
2717         */
2718        if (PIDLIST_REALLOC_DIFFERENCE(length, dest)) {
2719                newlist = pidlist_resize(list, dest);
2720                if (newlist)
2721                        *p = newlist;
2722        }
2723        return dest;
2724}
2725
2726static int cmppid(const void *a, const void *b)
2727{
2728        return *(pid_t *)a - *(pid_t *)b;
2729}
2730
2731/*
2732 * find the appropriate pidlist for our purpose (given procs vs tasks)
2733 * returns with the lock on that pidlist already held, and takes care
2734 * of the use count, or returns NULL with no locks held if we're out of
2735 * memory.
2736 */
2737static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp,
2738                                                  enum cgroup_filetype type)
2739{
2740        struct cgroup_pidlist *l;
2741        /* don't need task_nsproxy() if we're looking at ourself */
2742        struct pid_namespace *ns = current->nsproxy->pid_ns;
2743
2744        /*
2745         * We can't drop the pidlist_mutex before taking the l->mutex in case
2746         * the last ref-holder is trying to remove l from the list at the same
2747         * time. Holding the pidlist_mutex precludes somebody taking whichever
2748         * list we find out from under us - compare release_pid_array().
2749         */
2750        mutex_lock(&cgrp->pidlist_mutex);
2751        list_for_each_entry(l, &cgrp->pidlists, links) {
2752                if (l->key.type == type && l->key.ns == ns) {
2753                        /* make sure l doesn't vanish out from under us */
2754                        down_write(&l->mutex);
2755                        mutex_unlock(&cgrp->pidlist_mutex);
2756                        return l;
2757                }
2758        }
2759        /* entry not found; create a new one */
2760        l = kmalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL);
2761        if (!l) {
2762                mutex_unlock(&cgrp->pidlist_mutex);
2763                return l;
2764        }
2765        init_rwsem(&l->mutex);
2766        down_write(&l->mutex);
2767        l->key.type = type;
2768        l->key.ns = get_pid_ns(ns);
2769        l->use_count = 0; /* don't increment here */
2770        l->list = NULL;
2771        l->owner = cgrp;
2772        list_add(&l->links, &cgrp->pidlists);
2773        mutex_unlock(&cgrp->pidlist_mutex);
2774        return l;
2775}
2776
2777/*
2778 * Load a cgroup's pidarray with either procs' tgids or tasks' pids
2779 */
2780static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type,
2781                              struct cgroup_pidlist **lp)
2782{
2783        pid_t *array;
2784        int length;
2785        int pid, n = 0; /* used for populating the array */
2786        struct cgroup_iter it;
2787        struct task_struct *tsk;
2788        struct cgroup_pidlist *l;
2789
2790        /*
2791         * If cgroup gets more users after we read count, we won't have
2792         * enough space - tough.  This race is indistinguishable to the
2793         * caller from the case that the additional cgroup users didn't
2794         * show up until sometime later on.
2795         */
2796        length = cgroup_task_count(cgrp);
2797        array = pidlist_allocate(length);
2798        if (!array)
2799                return -ENOMEM;
2800        /* now, populate the array */
2801        cgroup_iter_start(cgrp, &it);
2802        while ((tsk = cgroup_iter_next(cgrp, &it))) {
2803                if (unlikely(n == length))
2804                        break;
2805                /* get tgid or pid for procs or tasks file respectively */
2806                if (type == CGROUP_FILE_PROCS)
2807                        pid = task_tgid_vnr(tsk);
2808                else
2809                        pid = task_pid_vnr(tsk);
2810                if (pid > 0) /* make sure to only use valid results */
2811                        array[n++] = pid;
2812        }
2813        cgroup_iter_end(cgrp, &it);
2814        length = n;
2815        /* now sort & (if procs) strip out duplicates */
2816        sort(array, length, sizeof(pid_t), cmppid, NULL);
2817        if (type == CGROUP_FILE_PROCS)
2818                length = pidlist_uniq(&array, length);
2819        l = cgroup_pidlist_find(cgrp, type);
2820        if (!l) {
2821                pidlist_free(array);
2822                return -ENOMEM;
2823        }
2824        /* store array, freeing old if necessary - lock already held */
2825        pidlist_free(l->list);
2826        l->list = array;
2827        l->length = length;
2828        l->use_count++;
2829        up_write(&l->mutex);
2830        *lp = l;
2831        return 0;
2832}
2833
2834/**
2835 * cgroupstats_build - build and fill cgroupstats
2836 * @stats: cgroupstats to fill information into
2837 * @dentry: A dentry entry belonging to the cgroup for which stats have
2838 * been requested.
2839 *
2840 * Build and fill cgroupstats so that taskstats can export it to user
2841 * space.
2842 */
2843int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry)
2844{
2845        int ret = -EINVAL;
2846        struct cgroup *cgrp;
2847        struct cgroup_iter it;
2848        struct task_struct *tsk;
2849
2850        /*
2851         * Validate dentry by checking the superblock operations,
2852         * and make sure it's a directory.
2853         */
2854        if (dentry->d_sb->s_op != &cgroup_ops ||
2855            !S_ISDIR(dentry->d_inode->i_mode))
2856                 goto err;
2857
2858        ret = 0;
2859        cgrp = dentry->d_fsdata;
2860
2861        cgroup_iter_start(cgrp, &it);
2862        while ((tsk = cgroup_iter_next(cgrp, &it))) {
2863                switch (tsk->state) {
2864                case TASK_RUNNING:
2865                        stats->nr_running++;
2866                        break;
2867                case TASK_INTERRUPTIBLE:
2868                        stats->nr_sleeping++;
2869                        break;
2870                case TASK_UNINTERRUPTIBLE:
2871                        stats->nr_uninterruptible++;
2872                        break;
2873                case TASK_STOPPED:
2874                        stats->nr_stopped++;
2875                        break;
2876                default:
2877                        if (delayacct_is_task_waiting_on_io(tsk))
2878                                stats->nr_io_wait++;
2879                        break;
2880                }
2881        }
2882        cgroup_iter_end(cgrp, &it);
2883
2884err:
2885        return ret;
2886}
2887
2888
2889/*
2890 * seq_file methods for the tasks/procs files. The seq_file position is the
2891 * next pid to display; the seq_file iterator is a pointer to the pid
2892 * in the cgroup->l->list array.
2893 */
2894
2895static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos)
2896{
2897        /*
2898         * Initially we receive a position value that corresponds to
2899         * one more than the last pid shown (or 0 on the first call or
2900         * after a seek to the start). Use a binary-search to find the
2901         * next pid to display, if any
2902         */
2903        struct cgroup_pidlist *l = s->private;
2904        int index = 0, pid = *pos;
2905        int *iter;
2906
2907        down_read(&l->mutex);
2908        if (pid) {
2909                int end = l->length;
2910
2911                while (index < end) {
2912                        int mid = (index + end) / 2;
2913                        if (l->list[mid] == pid) {
2914                                index = mid;
2915                                break;
2916                        } else if (l->list[mid] <= pid)
2917                                index = mid + 1;
2918                        else
2919                                end = mid;
2920                }
2921        }
2922        /* If we're off the end of the array, we're done */
2923        if (index >= l->length)
2924                return NULL;
2925        /* Update the abstract position to be the actual pid that we found */
2926        iter = l->list + index;
2927        *pos = *iter;
2928        return iter;
2929}
2930
2931static void cgroup_pidlist_stop(struct seq_file *s, void *v)
2932{
2933        struct cgroup_pidlist *l = s->private;
2934        up_read(&l->mutex);
2935}
2936
2937static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos)
2938{
2939        struct cgroup_pidlist *l = s->private;
2940        pid_t *p = v;
2941        pid_t *end = l->list + l->length;
2942        /*
2943         * Advance to the next pid in the array. If this goes off the
2944         * end, we're done
2945         */
2946        p++;
2947        if (p >= end) {
2948                return NULL;
2949        } else {
2950                *pos = *p;
2951                return p;
2952        }
2953}
2954
2955static int cgroup_pidlist_show(struct seq_file *s, void *v)
2956{
2957        return seq_printf(s, "%d\n", *(int *)v);
2958}
2959
2960/*
2961 * seq_operations functions for iterating on pidlists through seq_file -
2962 * independent of whether it's tasks or procs
2963 */
2964static const struct seq_operations cgroup_pidlist_seq_operations = {
2965        .start = cgroup_pidlist_start,
2966        .stop = cgroup_pidlist_stop,
2967        .next = cgroup_pidlist_next,
2968        .show = cgroup_pidlist_show,
2969};
2970
2971static void cgroup_release_pid_array(struct cgroup_pidlist *l)
2972{
2973        /*
2974         * the case where we're the last user of this particular pidlist will
2975         * have us remove it from the cgroup's list, which entails taking the
2976         * mutex. since in pidlist_find the pidlist->lock depends on cgroup->
2977         * pidlist_mutex, we have to take pidlist_mutex first.
2978         */
2979        mutex_lock(&l->owner->pidlist_mutex);
2980        down_write(&l->mutex);
2981        BUG_ON(!l->use_count);
2982        if (!--l->use_count) {
2983                /* we're the last user if refcount is 0; remove and free */
2984                list_del(&l->links);
2985                mutex_unlock(&l->owner->pidlist_mutex);
2986                pidlist_free(l->list);
2987                put_pid_ns(l->key.ns);
2988                up_write(&l->mutex);
2989                kfree(l);
2990                return;
2991        }
2992        mutex_unlock(&l->owner->pidlist_mutex);
2993        up_write(&l->mutex);
2994}
2995
2996static int cgroup_pidlist_release(struct inode *inode, struct file *file)
2997{
2998        struct cgroup_pidlist *l;
2999        if (!(file->f_mode & FMODE_READ))
3000                return 0;
3001        /*
3002         * the seq_file will only be initialized if the file was opened for
3003         * reading; hence we check if it's not null only in that case.
3004         */
3005        l = ((struct seq_file *)file->private_data)->private;
3006        cgroup_release_pid_array(l);
3007        return seq_release(inode, file);
3008}
3009
3010static const struct file_operations cgroup_pidlist_operations = {
3011        .read = seq_read,
3012        .llseek = seq_lseek,
3013        .write = cgroup_file_write,
3014        .release = cgroup_pidlist_release,
3015};
3016
3017/*
3018 * The following functions handle opens on a file that displays a pidlist
3019 * (tasks or procs). Prepare an array of the process/thread IDs of whoever's
3020 * in the cgroup.
3021 */
3022/* helper function for the two below it */
3023static int cgroup_pidlist_open(struct file *file, enum cgroup_filetype type)
3024{
3025        struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
3026        struct cgroup_pidlist *l;
3027        int retval;
3028
3029        /* Nothing to do for write-only files */
3030        if (!(file->f_mode & FMODE_READ))
3031                return 0;
3032
3033        /* have the array populated */
3034        retval = pidlist_array_load(cgrp, type, &l);
3035        if (retval)
3036                return retval;
3037        /* configure file information */
3038        file->f_op = &cgroup_pidlist_operations;
3039
3040        retval = seq_open(file, &cgroup_pidlist_seq_operations);
3041        if (retval) {
3042                cgroup_release_pid_array(l);
3043                return retval;
3044        }
3045        ((struct seq_file *)file->private_data)->private = l;
3046        return 0;
3047}
3048static int cgroup_tasks_open(struct inode *unused, struct file *file)
3049{
3050        return cgroup_pidlist_open(file, CGROUP_FILE_TASKS);
3051}
3052static int cgroup_procs_open(struct inode *unused, struct file *file)
3053{
3054        return cgroup_pidlist_open(file, CGROUP_FILE_PROCS);
3055}
3056
3057static u64 cgroup_read_notify_on_release(struct cgroup *cgrp,
3058                                            struct cftype *cft)
3059{
3060        return notify_on_release(cgrp);
3061}
3062
3063static int cgroup_write_notify_on_release(struct cgroup *cgrp,
3064                                          struct cftype *cft,
3065                                          u64 val)
3066{
3067        clear_bit(CGRP_RELEASABLE, &cgrp->flags);
3068        if (val)
3069                set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
3070        else
3071                clear_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
3072        return 0;
3073}
3074
3075/*
3076 * Unregister event and free resources.
3077 *
3078 * Gets called from workqueue.
3079 */
3080static void cgroup_event_remove(struct work_struct *work)
3081{
3082        struct cgroup_event *event = container_of(work, struct cgroup_event,
3083                        remove);
3084        struct cgroup *cgrp = event->cgrp;
3085
3086        event->cft->unregister_event(cgrp, event->cft, event->eventfd);
3087
3088        eventfd_ctx_put(event->eventfd);
3089        kfree(event);
3090        dput(cgrp->dentry);
3091}
3092
3093/*
3094 * Gets called on POLLHUP on eventfd when user closes it.
3095 *
3096 * Called with wqh->lock held and interrupts disabled.
3097 */
3098static int cgroup_event_wake(wait_queue_t *wait, unsigned mode,
3099                int sync, void *key)
3100{
3101        struct cgroup_event *event = container_of(wait,
3102                        struct cgroup_event, wait);
3103        struct cgroup *cgrp = event->cgrp;
3104        unsigned long flags = (unsigned long)key;
3105
3106        if (flags & POLLHUP) {
3107                __remove_wait_queue(event->wqh, &event->wait);
3108                spin_lock(&cgrp->event_list_lock);
3109                list_del(&event->list);
3110                spin_unlock(&cgrp->event_list_lock);
3111                /*
3112                 * We are in atomic context, but cgroup_event_remove() may
3113                 * sleep, so we have to call it in workqueue.
3114                 */
3115                schedule_work(&event->remove);
3116        }
3117
3118        return 0;
3119}
3120
3121static void cgroup_event_ptable_queue_proc(struct file *file,
3122                wait_queue_head_t *wqh, poll_table *pt)
3123{
3124        struct cgroup_event *event = container_of(pt,
3125                        struct cgroup_event, pt);
3126
3127        event->wqh = wqh;
3128        add_wait_queue(wqh, &event->wait);
3129}
3130
3131/*
3132 * Parse input and register new cgroup event handler.
3133 *
3134 * Input must be in format '<event_fd> <control_fd> <args>'.
3135 * Interpretation of args is defined by control file implementation.
3136 */
3137static int cgroup_write_event_control(struct cgroup *cgrp, struct cftype *cft,
3138                                      const char *buffer)
3139{
3140        struct cgroup_event *event = NULL;
3141        unsigned int efd, cfd;
3142        struct file *efile = NULL;
3143        struct file *cfile = NULL;
3144        char *endp;
3145        int ret;
3146
3147        efd = simple_strtoul(buffer, &endp, 10);
3148        if (*endp != ' ')
3149                return -EINVAL;
3150        buffer = endp + 1;
3151
3152        cfd = simple_strtoul(buffer, &endp, 10);
3153        if ((*endp != ' ') && (*endp != '\0'))
3154                return -EINVAL;
3155        buffer = endp + 1;
3156
3157        event = kzalloc(sizeof(*event), GFP_KERNEL);
3158        if (!event)
3159                return -ENOMEM;
3160        event->cgrp = cgrp;
3161        INIT_LIST_HEAD(&event->list);
3162        init_poll_funcptr(&event->pt, cgroup_event_ptable_queue_proc);
3163        init_waitqueue_func_entry(&event->wait, cgroup_event_wake);
3164        INIT_WORK(&event->remove, cgroup_event_remove);
3165
3166        efile = eventfd_fget(efd);
3167        if (IS_ERR(efile)) {
3168                ret = PTR_ERR(efile);
3169                goto fail;
3170        }
3171
3172        event->eventfd = eventfd_ctx_fileget(efile);
3173        if (IS_ERR(event->eventfd)) {
3174                ret = PTR_ERR(event->eventfd);
3175                goto fail;
3176        }
3177
3178        cfile = fget(cfd);
3179        if (!cfile) {
3180                ret = -EBADF;
3181                goto fail;
3182        }
3183
3184        /* the process need read permission on control file */
3185        ret = file_permission(cfile, MAY_READ);
3186        if (ret < 0)
3187                goto fail;
3188
3189        event->cft = __file_cft(cfile);
3190        if (IS_ERR(event->cft)) {
3191                ret = PTR_ERR(event->cft);
3192                goto fail;
3193        }
3194
3195        if (!event->cft->register_event || !event->cft->unregister_event) {
3196                ret = -EINVAL;
3197                goto fail;
3198        }
3199
3200        ret = event->cft->register_event(cgrp, event->cft,
3201                        event->eventfd, buffer);
3202        if (ret)
3203                goto fail;
3204
3205        if (efile->f_op->poll(efile, &event->pt) & POLLHUP) {
3206                event->cft->unregister_event(cgrp, event->cft, event->eventfd);
3207                ret = 0;
3208                goto fail;
3209        }
3210
3211        /*
3212         * Events should be removed after rmdir of cgroup directory, but before
3213         * destroying subsystem state objects. Let's take reference to cgroup
3214         * directory dentry to do that.
3215         */
3216        dget(cgrp->dentry);
3217
3218        spin_lock(&cgrp->event_list_lock);
3219        list_add(&event->list, &cgrp->event_list);
3220        spin_unlock(&cgrp->event_list_lock);
3221
3222        fput(cfile);
3223        fput(efile);
3224
3225        return 0;
3226
3227fail:
3228        if (cfile)
3229                fput(cfile);
3230
3231        if (event && event->eventfd && !IS_ERR(event->eventfd))
3232                eventfd_ctx_put(event->eventfd);
3233
3234        if (!IS_ERR_OR_NULL(efile))
3235                fput(efile);
3236
3237        kfree(event);
3238
3239        return ret;
3240}
3241
3242static u64 cgroup_clone_children_read(struct cgroup *cgrp,
3243                                    struct cftype *cft)
3244{
3245        return clone_children(cgrp);
3246}
3247
3248static int cgroup_clone_children_write(struct cgroup *cgrp,
3249                                     struct cftype *cft,
3250                                     u64 val)
3251{
3252        if (val)
3253                set_bit(CGRP_CLONE_CHILDREN, &cgrp->flags);
3254        else
3255                clear_bit(CGRP_CLONE_CHILDREN, &cgrp->flags);
3256        return 0;
3257}
3258
3259/*
3260 * for the common functions, 'private' gives the type of file
3261 */
3262/* for hysterical raisins, we can't put this on the older files */
3263#define CGROUP_FILE_GENERIC_PREFIX "cgroup."
3264static struct cftype files[] = {
3265        {
3266                .name = "tasks",
3267                .open = cgroup_tasks_open,
3268                .write_u64 = cgroup_tasks_write,
3269                .release = cgroup_pidlist_release,
3270                .mode = S_IRUGO | S_IWUSR,
3271        },
3272        {
3273                .name = CGROUP_FILE_GENERIC_PREFIX "procs",
3274                .open = cgroup_procs_open,
3275                /* .write_u64 = cgroup_procs_write, TODO */
3276                .release = cgroup_pidlist_release,
3277                .mode = S_IRUGO,
3278        },
3279        {
3280                .name = "notify_on_release",
3281                .read_u64 = cgroup_read_notify_on_release,
3282                .write_u64 = cgroup_write_notify_on_release,
3283        },
3284        {
3285                .name = CGROUP_FILE_GENERIC_PREFIX "event_control",
3286                .write_string = cgroup_write_event_control,
3287                .mode = S_IWUGO,
3288        },
3289        {
3290                .name = "cgroup.clone_children",
3291                .read_u64 = cgroup_clone_children_read,
3292                .write_u64 = cgroup_clone_children_write,
3293        },
3294};
3295
3296static struct cftype cft_release_agent = {
3297        .name = "release_agent",
3298        .read_seq_string = cgroup_release_agent_show,
3299        .write_string = cgroup_release_agent_write,
3300        .max_write_len = PATH_MAX,
3301};
3302
3303static int cgroup_populate_dir(struct cgroup *cgrp)
3304{
3305        int err;
3306        struct cgroup_subsys *ss;
3307
3308        /* First clear out any existing files */
3309        cgroup_clear_directory(cgrp->dentry);
3310
3311        err = cgroup_add_files(cgrp, NULL, files, ARRAY_SIZE(files));
3312        if (err < 0)
3313                return err;
3314
3315        if (cgrp == cgrp->top_cgroup) {
3316                if ((err = cgroup_add_file(cgrp, NULL, &cft_release_agent)) < 0)
3317                        return err;
3318        }
3319
3320        for_each_subsys(cgrp->root, ss) {
3321                if (ss->populate && (err = ss->populate(ss, cgrp)) < 0)
3322                        return err;
3323        }
3324        /* This cgroup is ready now */
3325        for_each_subsys(cgrp->root, ss) {
3326                struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
3327                /*
3328                 * Update id->css pointer and make this css visible from
3329                 * CSS ID functions. This pointer will be dereferened
3330                 * from RCU-read-side without locks.
3331                 */
3332                if (css->id)
3333                        rcu_assign_pointer(css->id->css, css);
3334        }
3335
3336        return 0;
3337}
3338
3339static void init_cgroup_css(struct cgroup_subsys_state *css,
3340                               struct cgroup_subsys *ss,
3341                               struct cgroup *cgrp)
3342{
3343        css->cgroup = cgrp;
3344        atomic_set(&css->refcnt, 1);
3345        css->flags = 0;
3346        css->id = NULL;
3347        if (cgrp == dummytop)
3348                set_bit(CSS_ROOT, &css->flags);
3349        BUG_ON(cgrp->subsys[ss->subsys_id]);
3350        cgrp->subsys[ss->subsys_id] = css;
3351}
3352
3353static void cgroup_lock_hierarchy(struct cgroupfs_root *root)
3354{
3355        /* We need to take each hierarchy_mutex in a consistent order */
3356        int i;
3357
3358        /*
3359         * No worry about a race with rebind_subsystems that might mess up the
3360         * locking order, since both parties are under cgroup_mutex.
3361         */
3362        for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
3363                struct cgroup_subsys *ss = subsys[i];
3364                if (ss == NULL)
3365                        continue;
3366                if (ss->root == root)
3367                        mutex_lock(&ss->hierarchy_mutex);
3368        }
3369}
3370
3371static void cgroup_unlock_hierarchy(struct cgroupfs_root *root)
3372{
3373        int i;
3374
3375        for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
3376                struct cgroup_subsys *ss = subsys[i];
3377                if (ss == NULL)
3378                        continue;
3379                if (ss->root == root)
3380                        mutex_unlock(&ss->hierarchy_mutex);
3381        }
3382}
3383
3384/*
3385 * cgroup_create - create a cgroup
3386 * @parent: cgroup that will be parent of the new cgroup
3387 * @dentry: dentry of the new cgroup
3388 * @mode: mode to set on new inode
3389 *
3390 * Must be called with the mutex on the parent inode held
3391 */
3392static long cgroup_create(struct cgroup *parent, struct dentry *dentry,
3393                             mode_t mode)
3394{
3395        struct cgroup *cgrp;
3396        struct cgroupfs_root *root = parent->root;
3397        int err = 0;
3398        struct cgroup_subsys *ss;
3399        struct super_block *sb = root->sb;
3400
3401        cgrp = kzalloc(sizeof(*cgrp), GFP_KERNEL);
3402        if (!cgrp)
3403                return -ENOMEM;
3404
3405        /* Grab a reference on the superblock so the hierarchy doesn't
3406         * get deleted on unmount if there are child cgroups.  This
3407         * can be done outside cgroup_mutex, since the sb can't
3408         * disappear while someone has an open control file on the
3409         * fs */
3410        atomic_inc(&sb->s_active);
3411
3412        mutex_lock(&cgroup_mutex);
3413
3414        init_cgroup_housekeeping(cgrp);
3415
3416        cgrp->parent = parent;
3417        cgrp->root = parent->root;
3418        cgrp->top_cgroup = parent->top_cgroup;
3419
3420        if (notify_on_release(parent))
3421                set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
3422
3423        if (clone_children(parent))
3424                set_bit(CGRP_CLONE_CHILDREN, &cgrp->flags);
3425
3426        for_each_subsys(root, ss) {
3427                struct cgroup_subsys_state *css = ss->create(ss, cgrp);
3428
3429                if (IS_ERR(css)) {
3430                        err = PTR_ERR(css);
3431                        goto err_destroy;
3432                }
3433                init_cgroup_css(css, ss, cgrp);
3434                if (ss->use_id) {
3435                        err = alloc_css_id(ss, parent, cgrp);
3436                        if (err)
3437                                goto err_destroy;
3438                }
3439                /* At error, ->destroy() callback has to free assigned ID. */
3440                if (clone_children(parent) && ss->post_clone)
3441                        ss->post_clone(ss, cgrp);
3442        }
3443
3444        cgroup_lock_hierarchy(root);
3445        list_add(&cgrp->sibling, &cgrp->parent->children);
3446        cgroup_unlock_hierarchy(root);
3447        root->number_of_cgroups++;
3448
3449        err = cgroup_create_dir(cgrp, dentry, mode);
3450        if (err < 0)
3451                goto err_remove;
3452
3453        /* The cgroup directory was pre-locked for us */
3454        BUG_ON(!mutex_is_locked(&cgrp->dentry->d_inode->i_mutex));
3455
3456        err = cgroup_populate_dir(cgrp);
3457        /* If err < 0, we have a half-filled directory - oh well ;) */
3458
3459        mutex_unlock(&cgroup_mutex);
3460        mutex_unlock(&cgrp->dentry->d_inode->i_mutex);
3461
3462        return 0;
3463
3464 err_remove:
3465
3466        cgroup_lock_hierarchy(root);
3467        list_del(&cgrp->sibling);
3468        cgroup_unlock_hierarchy(root);
3469        root->number_of_cgroups--;
3470
3471 err_destroy:
3472
3473        for_each_subsys(root, ss) {
3474                if (cgrp->subsys[ss->subsys_id])
3475                        ss->destroy(ss, cgrp);
3476        }
3477
3478        mutex_unlock(&cgroup_mutex);
3479
3480        /* Release the reference count that we took on the superblock */
3481        deactivate_super(sb);
3482
3483        kfree(cgrp);
3484        return err;
3485}
3486
3487static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, int mode)
3488{
3489        struct cgroup *c_parent = dentry->d_parent->d_fsdata;
3490
3491        /* the vfs holds inode->i_mutex already */
3492        return cgroup_create(c_parent, dentry, mode | S_IFDIR);
3493}
3494
3495static int cgroup_has_css_refs(struct cgroup *cgrp)
3496{
3497        /* Check the reference count on each subsystem. Since we
3498         * already established that there are no tasks in the
3499         * cgroup, if the css refcount is also 1, then there should
3500         * be no outstanding references, so the subsystem is safe to
3501         * destroy. We scan across all subsystems rather than using
3502         * the per-hierarchy linked list of mounted subsystems since
3503         * we can be called via check_for_release() with no
3504         * synchronization other than RCU, and the subsystem linked
3505         * list isn't RCU-safe */
3506        int i;
3507        /*
3508         * We won't need to lock the subsys array, because the subsystems
3509         * we're concerned about aren't going anywhere since our cgroup root
3510         * has a reference on them.
3511         */
3512        for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
3513                struct cgroup_subsys *ss = subsys[i];
3514                struct cgroup_subsys_state *css;
3515                /* Skip subsystems not present or not in this hierarchy */
3516                if (ss == NULL || ss->root != cgrp->root)
3517                        continue;
3518                css = cgrp->subsys[ss->subsys_id];
3519                /* When called from check_for_release() it's possible
3520                 * that by this point the cgroup has been removed
3521                 * and the css deleted. But a false-positive doesn't
3522                 * matter, since it can only happen if the cgroup
3523                 * has been deleted and hence no longer needs the
3524                 * release agent to be called anyway. */
3525                if (css && (atomic_read(&css->refcnt) > 1))
3526                        return 1;
3527        }
3528        return 0;
3529}
3530
3531/*
3532 * Atomically mark all (or else none) of the cgroup's CSS objects as
3533 * CSS_REMOVED. Return true on success, or false if the cgroup has
3534 * busy subsystems. Call with cgroup_mutex held
3535 */
3536
3537static int cgroup_clear_css_refs(struct cgroup *cgrp)
3538{
3539        struct cgroup_subsys *ss;
3540        unsigned long flags;
3541        bool failed = false;
3542        local_irq_save(flags);
3543        for_each_subsys(cgrp->root, ss) {
3544                struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
3545                int refcnt;
3546                while (1) {
3547                        /* We can only remove a CSS with a refcnt==1 */
3548                        refcnt = atomic_read(&css->refcnt);
3549                        if (refcnt > 1) {
3550                                failed = true;
3551                                goto done;
3552                        }
3553                        BUG_ON(!refcnt);
3554                        /*
3555                         * Drop the refcnt to 0 while we check other
3556                         * subsystems. This will cause any racing
3557                         * css_tryget() to spin until we set the
3558                         * CSS_REMOVED bits or abort
3559                         */
3560                        if (atomic_cmpxchg(&css->refcnt, refcnt, 0) == refcnt)
3561                                break;
3562                        cpu_relax();
3563                }
3564        }
3565 done:
3566        for_each_subsys(cgrp->root, ss) {
3567                struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
3568                if (failed) {
3569                        /*
3570                         * Restore old refcnt if we previously managed
3571                         * to clear it from 1 to 0
3572                         */
3573                        if (!atomic_read(&css->refcnt))
3574                                atomic_set(&css->refcnt, 1);
3575                } else {
3576                        /* Commit the fact that the CSS is removed */
3577                        set_bit(CSS_REMOVED, &css->flags);
3578                }
3579        }
3580        local_irq_restore(flags);
3581        return !failed;
3582}
3583
3584static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry)
3585{
3586        struct cgroup *cgrp = dentry->d_fsdata;
3587        struct dentry *d;
3588        struct cgroup *parent;
3589        DEFINE_WAIT(wait);
3590        struct cgroup_event *event, *tmp;
3591        int ret;
3592
3593        /* the vfs holds both inode->i_mutex already */
3594again:
3595        mutex_lock(&cgroup_mutex);
3596        if (atomic_read(&cgrp->count) != 0) {
3597                mutex_unlock(&cgroup_mutex);
3598                return -EBUSY;
3599        }
3600        if (!list_empty(&cgrp->children)) {
3601                mutex_unlock(&cgroup_mutex);
3602                return -EBUSY;
3603        }
3604        mutex_unlock(&cgroup_mutex);
3605
3606        /*
3607         * In general, subsystem has no css->refcnt after pre_destroy(). But
3608         * in racy cases, subsystem may have to get css->refcnt after
3609         * pre_destroy() and it makes rmdir return with -EBUSY. This sometimes
3610         * make rmdir return -EBUSY too often. To avoid that, we use waitqueue
3611         * for cgroup's rmdir. CGRP_WAIT_ON_RMDIR is for synchronizing rmdir
3612         * and subsystem's reference count handling. Please see css_get/put
3613         * and css_tryget() and cgroup_wakeup_rmdir_waiter() implementation.
3614         */
3615        set_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
3616
3617        /*
3618         * Call pre_destroy handlers of subsys. Notify subsystems
3619         * that rmdir() request comes.
3620         */
3621        ret = cgroup_call_pre_destroy(cgrp);
3622        if (ret) {
3623                clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
3624                return ret;
3625        }
3626
3627        mutex_lock(&cgroup_mutex);
3628        parent = cgrp->parent;
3629        if (atomic_read(&cgrp->count) || !list_empty(&cgrp->children)) {
3630                clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
3631                mutex_unlock(&cgroup_mutex);
3632                return -EBUSY;
3633        }
3634        prepare_to_wait(&cgroup_rmdir_waitq, &wait, TASK_INTERRUPTIBLE);
3635        if (!cgroup_clear_css_refs(cgrp)) {
3636                mutex_unlock(&cgroup_mutex);
3637                /*
3638                 * Because someone may call cgroup_wakeup_rmdir_waiter() before
3639                 * prepare_to_wait(), we need to check this flag.
3640                 */
3641                if (test_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags))
3642                        schedule();
3643                finish_wait(&cgroup_rmdir_waitq, &wait);
3644                clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
3645                if (signal_pending(current))
3646                        return -EINTR;
3647                goto again;
3648        }
3649        /* NO css_tryget() can success after here. */
3650        finish_wait(&cgroup_rmdir_waitq, &wait);
3651        clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
3652
3653        spin_lock(&release_list_lock);
3654        set_bit(CGRP_REMOVED, &cgrp->flags);
3655        if (!list_empty(&cgrp->release_list))
3656                list_del_init(&cgrp->release_list);
3657        spin_unlock(&release_list_lock);
3658
3659        cgroup_lock_hierarchy(cgrp->root);
3660        /* delete this cgroup from parent->children */
3661        list_del_init(&cgrp->sibling);
3662        cgroup_unlock_hierarchy(cgrp->root);
3663
3664        d = dget(cgrp->dentry);
3665
3666        cgroup_d_remove_dir(d);
3667        dput(d);
3668
3669        set_bit(CGRP_RELEASABLE, &parent->flags);
3670        check_for_release(parent);
3671
3672        /*
3673         * Unregister events and notify userspace.
3674         * Notify userspace about cgroup removing only after rmdir of cgroup
3675         * directory to avoid race between userspace and kernelspace
3676         */
3677        spin_lock(&cgrp->event_list_lock);
3678        list_for_each_entry_safe(event, tmp, &cgrp->event_list, list) {
3679                list_del(&event->list);
3680                remove_wait_queue(event->wqh, &event->wait);
3681                eventfd_signal(event->eventfd, 1);
3682                schedule_work(&event->remove);
3683        }
3684        spin_unlock(&cgrp->event_list_lock);
3685
3686        mutex_unlock(&cgroup_mutex);
3687        return 0;
3688}
3689
3690static void __init cgroup_init_subsys(struct cgroup_subsys *ss)
3691{
3692        struct cgroup_subsys_state *css;
3693
3694        printk(KERN_INFO "Initializing cgroup subsys %s\n", ss->name);
3695
3696        /* Create the top cgroup state for this subsystem */
3697        list_add(&ss->sibling, &rootnode.subsys_list);
3698        ss->root = &rootnode;
3699        css = ss->create(ss, dummytop);
3700        /* We don't handle early failures gracefully */
3701        BUG_ON(IS_ERR(css));
3702        init_cgroup_css(css, ss, dummytop);
3703
3704        /* Update the init_css_set to contain a subsys
3705         * pointer to this state - since the subsystem is
3706         * newly registered, all tasks and hence the
3707         * init_css_set is in the subsystem's top cgroup. */
3708        init_css_set.subsys[ss->subsys_id] = dummytop->subsys[ss->subsys_id];
3709
3710        need_forkexit_callback |= ss->fork || ss->exit;
3711
3712        /* At system boot, before all subsystems have been
3713         * registered, no tasks have been forked, so we don't
3714         * need to invoke fork callbacks here. */
3715        BUG_ON(!list_empty(&init_task.tasks));
3716
3717        mutex_init(&ss->hierarchy_mutex);
3718        lockdep_set_class(&ss->hierarchy_mutex, &ss->subsys_key);
3719        ss->active = 1;
3720
3721        /* this function shouldn't be used with modular subsystems, since they
3722         * need to register a subsys_id, among other things */
3723        BUG_ON(ss->module);
3724}
3725
3726/**
3727 * cgroup_load_subsys: load and register a modular subsystem at runtime
3728 * @ss: the subsystem to load
3729 *
3730 * This function should be called in a modular subsystem's initcall. If the
3731 * subsystem is built as a module, it will be assigned a new subsys_id and set
3732 * up for use. If the subsystem is built-in anyway, work is delegated to the
3733 * simpler cgroup_init_subsys.
3734 */
3735int __init_or_module cgroup_load_subsys(struct cgroup_subsys *ss)
3736{
3737        int i;
3738        struct cgroup_subsys_state *css;
3739
3740        /* check name and function validity */
3741        if (ss->name == NULL || strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN ||
3742            ss->create == NULL || ss->destroy == NULL)
3743                return -EINVAL;
3744
3745        /*
3746         * we don't support callbacks in modular subsystems. this check is
3747         * before the ss->module check for consistency; a subsystem that could
3748         * be a module should still have no callbacks even if the user isn't
3749         * compiling it as one.
3750         */
3751        if (ss->fork || ss->exit)
3752                return -EINVAL;
3753
3754        /*
3755         * an optionally modular subsystem is built-in: we want to do nothing,
3756         * since cgroup_init_subsys will have already taken care of it.
3757         */
3758        if (ss->module == NULL) {
3759                /* a few sanity checks */
3760                BUG_ON(ss->subsys_id >= CGROUP_BUILTIN_SUBSYS_COUNT);
3761                BUG_ON(subsys[ss->subsys_id] != ss);
3762                return 0;
3763        }
3764
3765        /*
3766         * need to register a subsys id before anything else - for example,
3767         * init_cgroup_css needs it.
3768         */
3769        mutex_lock(&cgroup_mutex);
3770        /* find the first empty slot in the array */
3771        for (i = CGROUP_BUILTIN_SUBSYS_COUNT; i < CGROUP_SUBSYS_COUNT; i++) {
3772                if (subsys[i] == NULL)
3773                        break;
3774        }
3775        if (i == CGROUP_SUBSYS_COUNT) {
3776                /* maximum number of subsystems already registered! */
3777                mutex_unlock(&cgroup_mutex);
3778                return -EBUSY;
3779        }
3780        /* assign ourselves the subsys_id */
3781        ss->subsys_id = i;
3782        subsys[i] = ss;
3783
3784        /*
3785         * no ss->create seems to need anything important in the ss struct, so
3786         * this can happen first (i.e. before the rootnode attachment).
3787         */
3788        css = ss->create(ss, dummytop);
3789        if (IS_ERR(css)) {
3790                /* failure case - need to deassign the subsys[] slot. */
3791                subsys[i] = NULL;
3792                mutex_unlock(&cgroup_mutex);
3793                return PTR_ERR(css);
3794        }
3795
3796        list_add(&ss->sibling, &rootnode.subsys_list);
3797        ss->root = &rootnode;
3798
3799        /* our new subsystem will be attached to the dummy hierarchy. */
3800        init_cgroup_css(css, ss, dummytop);
3801        /* init_idr must be after init_cgroup_css because it sets css->id. */
3802        if (ss->use_id) {
3803                int ret = cgroup_init_idr(ss, css);
3804                if (ret) {
3805                        dummytop->subsys[ss->subsys_id] = NULL;
3806                        ss->destroy(ss, dummytop);
3807                        subsys[i] = NULL;
3808                        mutex_unlock(&cgroup_mutex);
3809                        return ret;
3810                }
3811        }
3812
3813        /*
3814         * Now we need to entangle the css into the existing css_sets. unlike
3815         * in cgroup_init_subsys, there are now multiple css_sets, so each one
3816         * will need a new pointer to it; done by iterating the css_set_table.
3817         * furthermore, modifying the existing css_sets will corrupt the hash
3818         * table state, so each changed css_set will need its hash recomputed.
3819         * this is all done under the css_set_lock.
3820         */
3821        write_lock(&css_set_lock);
3822        for (i = 0; i < CSS_SET_TABLE_SIZE; i++) {
3823                struct css_set *cg;
3824                struct hlist_node *node, *tmp;
3825                struct hlist_head *bucket = &css_set_table[i], *new_bucket;
3826
3827                hlist_for_each_entry_safe(cg, node, tmp, bucket, hlist) {
3828                        /* skip entries that we already rehashed */
3829                        if (cg->subsys[ss->subsys_id])
3830                                continue;
3831                        /* remove existing entry */
3832                        hlist_del(&cg->hlist);
3833                        /* set new value */
3834                        cg->subsys[ss->subsys_id] = css;
3835                        /* recompute hash and restore entry */
3836                        new_bucket = css_set_hash(cg->subsys);
3837                        hlist_add_head(&cg->hlist, new_bucket);
3838                }
3839        }
3840        write_unlock(&css_set_lock);
3841
3842        mutex_init(&ss->hierarchy_mutex);
3843        lockdep_set_class(&ss->hierarchy_mutex, &ss->subsys_key);
3844        ss->active = 1;
3845
3846        /* success! */
3847        mutex_unlock(&cgroup_mutex);
3848        return 0;
3849}
3850EXPORT_SYMBOL_GPL(cgroup_load_subsys);
3851
3852/**
3853 * cgroup_unload_subsys: unload a modular subsystem
3854 * @ss: the subsystem to unload
3855 *
3856 * This function should be called in a modular subsystem's exitcall. When this
3857 * function is invoked, the refcount on the subsystem's module will be 0, so
3858 * the subsystem will not be attached to any hierarchy.
3859 */
3860void cgroup_unload_subsys(struct cgroup_subsys *ss)
3861{
3862        struct cg_cgroup_link *link;
3863        struct hlist_head *hhead;
3864
3865        BUG_ON(ss->module == NULL);
3866
3867        /*
3868         * we shouldn't be called if the subsystem is in use, and the use of
3869         * try_module_get in parse_cgroupfs_options should ensure that it
3870         * doesn't start being used while we're killing it off.
3871         */
3872        BUG_ON(ss->root != &rootnode);
3873
3874        mutex_lock(&cgroup_mutex);
3875        /* deassign the subsys_id */
3876        BUG_ON(ss->subsys_id < CGROUP_BUILTIN_SUBSYS_COUNT);
3877        subsys[ss->subsys_id] = NULL;
3878
3879        /* remove subsystem from rootnode's list of subsystems */
3880        list_del_init(&ss->sibling);
3881
3882        /*
3883         * disentangle the css from all css_sets attached to the dummytop. as
3884         * in loading, we need to pay our respects to the hashtable gods.
3885         */
3886        write_lock(&css_set_lock);
3887        list_for_each_entry(link, &dummytop->css_sets, cgrp_link_list) {
3888                struct css_set *cg = link->cg;
3889
3890                hlist_del(&cg->hlist);
3891                BUG_ON(!cg->subsys[ss->subsys_id]);
3892                cg->subsys[ss->subsys_id] = NULL;
3893                hhead = css_set_hash(cg->subsys);
3894                hlist_add_head(&cg->hlist, hhead);
3895        }
3896        write_unlock(&css_set_lock);
3897
3898        /*
3899         * remove subsystem's css from the dummytop and free it - need to free
3900         * before marking as null because ss->destroy needs the cgrp->subsys
3901         * pointer to find their state. note that this also takes care of
3902         * freeing the css_id.
3903         */
3904        ss->destroy(ss, dummytop);
3905        dummytop->subsys[ss->subsys_id] = NULL;
3906
3907        mutex_unlock(&cgroup_mutex);
3908}
3909EXPORT_SYMBOL_GPL(cgroup_unload_subsys);
3910
3911/**
3912 * cgroup_init_early - cgroup initialization at system boot
3913 *
3914 * Initialize cgroups at system boot, and initialize any
3915 * subsystems that request early init.
3916 */
3917int __init cgroup_init_early(void)
3918{
3919        int i;
3920        atomic_set(&init_css_set.refcount, 1);
3921        INIT_LIST_HEAD(&init_css_set.cg_links);
3922        INIT_LIST_HEAD(&init_css_set.tasks);
3923        INIT_HLIST_NODE(&init_css_set.hlist);
3924        css_set_count = 1;
3925        init_cgroup_root(&rootnode);
3926        root_count = 1;
3927        init_task.cgroups = &init_css_set;
3928
3929        init_css_set_link.cg = &init_css_set;
3930        init_css_set_link.cgrp = dummytop;
3931        list_add(&init_css_set_link.cgrp_link_list,
3932                 &rootnode.top_cgroup.css_sets);
3933        list_add(&init_css_set_link.cg_link_list,
3934                 &init_css_set.cg_links);
3935
3936        for (i = 0; i < CSS_SET_TABLE_SIZE; i++)
3937                INIT_HLIST_HEAD(&css_set_table[i]);
3938
3939        /* at bootup time, we don't worry about modular subsystems */
3940        for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
3941                struct cgroup_subsys *ss = subsys[i];
3942
3943                BUG_ON(!ss->name);
3944                BUG_ON(strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN);
3945                BUG_ON(!ss->create);
3946                BUG_ON(!ss->destroy);
3947                if (ss->subsys_id != i) {
3948                        printk(KERN_ERR "cgroup: Subsys %s id == %d\n",
3949                               ss->name, ss->subsys_id);
3950                        BUG();
3951                }
3952
3953                if (ss->early_init)
3954                        cgroup_init_subsys(ss);
3955        }
3956        return 0;
3957}
3958
3959/**
3960 * cgroup_init - cgroup initialization
3961 *
3962 * Register cgroup filesystem and /proc file, and initialize
3963 * any subsystems that didn't request early init.
3964 */
3965int __init cgroup_init(void)
3966{
3967        int err;
3968        int i;
3969        struct hlist_head *hhead;
3970
3971        err = bdi_init(&cgroup_backing_dev_info);
3972        if (err)
3973                return err;
3974
3975        /* at bootup time, we don't worry about modular subsystems */
3976        for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
3977                struct cgroup_subsys *ss = subsys[i];
3978                if (!ss->early_init)
3979                        cgroup_init_subsys(ss);
3980                if (ss->use_id)
3981                        cgroup_init_idr(ss, init_css_set.subsys[ss->subsys_id]);
3982        }
3983
3984        /* Add init_css_set to the hash table */
3985        hhead = css_set_hash(init_css_set.subsys);
3986        hlist_add_head(&init_css_set.hlist, hhead);
3987        BUG_ON(!init_root_id(&rootnode));
3988
3989        cgroup_kobj = kobject_create_and_add("cgroup", fs_kobj);
3990        if (!cgroup_kobj) {
3991                err = -ENOMEM;
3992                goto out;
3993        }
3994
3995        err = register_filesystem(&cgroup_fs_type);
3996        if (err < 0) {
3997                kobject_put(cgroup_kobj);
3998                goto out;
3999        }
4000
4001        proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations);
4002
4003out:
4004        if (err)
4005                bdi_destroy(&cgroup_backing_dev_info);
4006
4007        return err;
4008}
4009
4010/*
4011 * proc_cgroup_show()
4012 *  - Print task's cgroup paths into seq_file, one line for each hierarchy
4013 *  - Used for /proc/<pid>/cgroup.
4014 *  - No need to task_lock(tsk) on this tsk->cgroup reference, as it
4015 *    doesn't really matter if tsk->cgroup changes after we read it,
4016 *    and we take cgroup_mutex, keeping cgroup_attach_task() from changing it
4017 *    anyway.  No need to check that tsk->cgroup != NULL, thanks to
4018 *    the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks
4019 *    cgroup to top_cgroup.
4020 */
4021
4022/* TODO: Use a proper seq_file iterator */
4023static int proc_cgroup_show(struct seq_file *m, void *v)
4024{
4025        struct pid *pid;
4026        struct task_struct *tsk;
4027        char *buf;
4028        int retval;
4029        struct cgroupfs_root *root;
4030
4031        retval = -ENOMEM;
4032        buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
4033        if (!buf)
4034                goto out;
4035
4036        retval = -ESRCH;
4037        pid = m->private;
4038        tsk = get_pid_task(pid, PIDTYPE_PID);
4039        if (!tsk)
4040                goto out_free;
4041
4042        retval = 0;
4043
4044        mutex_lock(&cgroup_mutex);
4045
4046        for_each_active_root(root) {
4047                struct cgroup_subsys *ss;
4048                struct cgroup *cgrp;
4049                int count = 0;
4050
4051                seq_printf(m, "%d:", root->hierarchy_id);
4052                for_each_subsys(root, ss)
4053                        seq_printf(m, "%s%s", count++ ? "," : "", ss->name);
4054                if (strlen(root->name))
4055                        seq_printf(m, "%sname=%s", count ? "," : "",
4056                                   root->name);
4057                seq_putc(m, ':');
4058                cgrp = task_cgroup_from_root(tsk, root);
4059                retval = cgroup_path(cgrp, buf, PAGE_SIZE);
4060                if (retval < 0)
4061                        goto out_unlock;
4062                seq_puts(m, buf);
4063                seq_putc(m, '\n');
4064        }
4065
4066out_unlock:
4067        mutex_unlock(&cgroup_mutex);
4068        put_task_struct(tsk);
4069out_free:
4070        kfree(buf);
4071out:
4072        return retval;
4073}
4074
4075static int cgroup_open(struct inode *inode, struct file *file)
4076{
4077        struct pid *pid = PROC_I(inode)->pid;
4078        return single_open(file, proc_cgroup_show, pid);
4079}
4080
4081const struct file_operations proc_cgroup_operations = {
4082        .open           = cgroup_open,
4083        .read           = seq_read,
4084        .llseek         = seq_lseek,
4085        .release        = single_release,
4086};
4087
4088/* Display information about each subsystem and each hierarchy */
4089static int proc_cgroupstats_show(struct seq_file *m, void *v)
4090{
4091        int i;
4092
4093        seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
4094        /*
4095         * ideally we don't want subsystems moving around while we do this.
4096         * cgroup_mutex is also necessary to guarantee an atomic snapshot of
4097         * subsys/hierarchy state.
4098         */
4099        mutex_lock(&cgroup_mutex);
4100        for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
4101                struct cgroup_subsys *ss = subsys[i];
4102                if (ss == NULL)
4103                        continue;
4104                seq_printf(m, "%s\t%d\t%d\t%d\n",
4105                           ss->name, ss->root->hierarchy_id,
4106                           ss->root->number_of_cgroups, !ss->disabled);
4107        }
4108        mutex_unlock(&cgroup_mutex);
4109        return 0;
4110}
4111
4112static int cgroupstats_open(struct inode *inode, struct file *file)
4113{
4114        return single_open(file, proc_cgroupstats_show, NULL);
4115}
4116
4117static const struct file_operations proc_cgroupstats_operations = {
4118        .open = cgroupstats_open,
4119        .read = seq_read,
4120        .llseek = seq_lseek,
4121        .release = single_release,
4122};
4123
4124/**
4125 * cgroup_fork - attach newly forked task to its parents cgroup.
4126 * @child: pointer to task_struct of forking parent process.
4127 *
4128 * Description: A task inherits its parent's cgroup at fork().
4129 *
4130 * A pointer to the shared css_set was automatically copied in
4131 * fork.c by dup_task_struct().  However, we ignore that copy, since
4132 * it was not made under the protection of RCU or cgroup_mutex, so
4133 * might no longer be a valid cgroup pointer.  cgroup_attach_task() might
4134 * have already changed current->cgroups, allowing the previously
4135 * referenced cgroup group to be removed and freed.
4136 *
4137 * At the point that cgroup_fork() is called, 'current' is the parent
4138 * task, and the passed argument 'child' points to the child task.
4139 */
4140void cgroup_fork(struct task_struct *child)
4141{
4142        task_lock(current);
4143        child->cgroups = current->cgroups;
4144        get_css_set(child->cgroups);
4145        task_unlock(current);
4146        INIT_LIST_HEAD(&child->cg_list);
4147}
4148
4149/**
4150 * cgroup_fork_callbacks - run fork callbacks
4151 * @child: the new task
4152 *
4153 * Called on a new task very soon before adding it to the
4154 * tasklist. No need to take any locks since no-one can
4155 * be operating on this task.
4156 */
4157void cgroup_fork_callbacks(struct task_struct *child)
4158{
4159        if (need_forkexit_callback) {
4160                int i;
4161                /*
4162                 * forkexit callbacks are only supported for builtin
4163                 * subsystems, and the builtin section of the subsys array is
4164                 * immutable, so we don't need to lock the subsys array here.
4165                 */
4166                for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
4167                        struct cgroup_subsys *ss = subsys[i];
4168                        if (ss->fork)
4169                                ss->fork(ss, child);
4170                }
4171        }
4172}
4173
4174/**
4175 * cgroup_post_fork - called on a new task after adding it to the task list
4176 * @child: the task in question
4177 *
4178 * Adds the task to the list running through its css_set if necessary.
4179 * Has to be after the task is visible on the task list in case we race
4180 * with the first call to cgroup_iter_start() - to guarantee that the
4181 * new task ends up on its list.
4182 */
4183void cgroup_post_fork(struct task_struct *child)
4184{
4185        if (use_task_css_set_links) {
4186                write_lock(&css_set_lock);
4187                task_lock(child);
4188                if (list_empty(&child->cg_list))
4189                        list_add(&child->cg_list, &child->cgroups->tasks);
4190                task_unlock(child);
4191                write_unlock(&css_set_lock);
4192        }
4193}
4194/**
4195 * cgroup_exit - detach cgroup from exiting task
4196 * @tsk: pointer to task_struct of exiting process
4197 * @run_callback: run exit callbacks?
4198 *
4199 * Description: Detach cgroup from @tsk and release it.
4200 *
4201 * Note that cgroups marked notify_on_release force every task in
4202 * them to take the global cgroup_mutex mutex when exiting.
4203 * This could impact scaling on very large systems.  Be reluctant to
4204 * use notify_on_release cgroups where very high task exit scaling
4205 * is required on large systems.
4206 *
4207 * the_top_cgroup_hack:
4208 *
4209 *    Set the exiting tasks cgroup to the root cgroup (top_cgroup).
4210 *
4211 *    We call cgroup_exit() while the task is still competent to
4212 *    handle notify_on_release(), then leave the task attached to the
4213 *    root cgroup in each hierarchy for the remainder of its exit.
4214 *
4215 *    To do this properly, we would increment the reference count on
4216 *    top_cgroup, and near the very end of the kernel/exit.c do_exit()
4217 *    code we would add a second cgroup function call, to drop that
4218 *    reference.  This would just create an unnecessary hot spot on
4219 *    the top_cgroup reference count, to no avail.
4220 *
4221 *    Normally, holding a reference to a cgroup without bumping its
4222 *    count is unsafe.   The cgroup could go away, or someone could
4223 *    attach us to a different cgroup, decrementing the count on
4224 *    the first cgroup that we never incremented.  But in this case,
4225 *    top_cgroup isn't going away, and either task has PF_EXITING set,
4226 *    which wards off any cgroup_attach_task() attempts, or task is a failed
4227 *    fork, never visible to cgroup_attach_task.
4228 */
4229void cgroup_exit(struct task_struct *tsk, int run_callbacks)
4230{
4231        struct css_set *cg;
4232        int i;
4233
4234        /*
4235         * Unlink from the css_set task list if necessary.
4236         * Optimistically check cg_list before taking
4237         * css_set_lock
4238         */
4239        if (!list_empty(&tsk->cg_list)) {
4240                write_lock(&css_set_lock);
4241                if (!list_empty(&tsk->cg_list))
4242                        list_del_init(&tsk->cg_list);
4243                write_unlock(&css_set_lock);
4244        }
4245
4246        /* Reassign the task to the init_css_set. */
4247        task_lock(tsk);
4248        cg = tsk->cgroups;
4249        tsk->cgroups = &init_css_set;
4250
4251        if (run_callbacks && need_forkexit_callback) {
4252                /*
4253                 * modular subsystems can't use callbacks, so no need to lock
4254                 * the subsys array
4255                 */
4256                for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
4257                        struct cgroup_subsys *ss = subsys[i];
4258                        if (ss->exit) {
4259                                struct cgroup *old_cgrp =
4260                                        rcu_dereference_raw(cg->subsys[i])->cgroup;
4261                                struct cgroup *cgrp = task_cgroup(tsk, i);
4262                                ss->exit(ss, cgrp, old_cgrp, tsk);
4263                        }
4264                }
4265        }
4266        task_unlock(tsk);
4267
4268        if (cg)
4269                put_css_set_taskexit(cg);
4270}
4271
4272/**
4273 * cgroup_clone - clone the cgroup the given subsystem is attached to
4274 * @tsk: the task to be moved
4275 * @subsys: the given subsystem
4276 * @nodename: the name for the new cgroup
4277 *
4278 * Duplicate the current cgroup in the hierarchy that the given
4279 * subsystem is attached to, and move this task into the new
4280 * child.
4281 */
4282int cgroup_clone(struct task_struct *tsk, struct cgroup_subsys *subsys,
4283                                                        char *nodename)
4284{
4285        struct dentry *dentry;
4286        int ret = 0;
4287        struct cgroup *parent, *child;
4288        struct inode *inode;
4289        struct css_set *cg;
4290        struct cgroupfs_root *root;
4291        struct cgroup_subsys *ss;
4292
4293        /* We shouldn't be called by an unregistered subsystem */
4294        BUG_ON(!subsys->active);
4295
4296        /* First figure out what hierarchy and cgroup we're dealing
4297         * with, and pin them so we can drop cgroup_mutex */
4298        mutex_lock(&cgroup_mutex);
4299 again:
4300        root = subsys->root;
4301        if (root == &rootnode) {
4302                mutex_unlock(&cgroup_mutex);
4303                return 0;
4304        }
4305
4306        /* Pin the hierarchy */
4307        if (!atomic_inc_not_zero(&root->sb->s_active)) {
4308                /* We race with the final deactivate_super() */
4309                mutex_unlock(&cgroup_mutex);
4310                return 0;
4311        }
4312
4313        /* Keep the cgroup alive */
4314        task_lock(tsk);
4315        parent = task_cgroup(tsk, subsys->subsys_id);
4316        cg = tsk->cgroups;
4317        get_css_set(cg);
4318        task_unlock(tsk);
4319
4320        mutex_unlock(&cgroup_mutex);
4321
4322        /* Now do the VFS work to create a cgroup */
4323        inode = parent->dentry->d_inode;
4324
4325        /* Hold the parent directory mutex across this operation to
4326         * stop anyone else deleting the new cgroup */
4327        mutex_lock(&inode->i_mutex);
4328        dentry = lookup_one_len(nodename, parent->dentry, strlen(nodename));
4329        if (IS_ERR(dentry)) {
4330                printk(KERN_INFO
4331                       "cgroup: Couldn't allocate dentry for %s: %ld\n", nodename,
4332                       PTR_ERR(dentry));
4333                ret = PTR_ERR(dentry);
4334                goto out_release;
4335        }
4336
4337        /* Create the cgroup directory, which also creates the cgroup */
4338        ret = vfs_mkdir(inode, dentry, 0755);
4339        child = __d_cgrp(dentry);
4340        dput(dentry);
4341        if (ret) {
4342                printk(KERN_INFO
4343                       "Failed to create cgroup %s: %d\n", nodename,
4344                       ret);
4345                goto out_release;
4346        }
4347
4348        /* The cgroup now exists. Retake cgroup_mutex and check
4349         * that we're still in the same state that we thought we
4350         * were. */
4351        mutex_lock(&cgroup_mutex);
4352        if ((root != subsys->root) ||
4353            (parent != task_cgroup(tsk, subsys->subsys_id))) {
4354                /* Aargh, we raced ... */
4355                mutex_unlock(&inode->i_mutex);
4356                put_css_set(cg);
4357
4358                deactivate_super(root->sb);
4359                /* The cgroup is still accessible in the VFS, but
4360                 * we're not going to try to rmdir() it at this
4361                 * point. */
4362                printk(KERN_INFO
4363                       "Race in cgroup_clone() - leaking cgroup %s\n",
4364                       nodename);
4365                goto again;
4366        }
4367
4368        /* do any required auto-setup */
4369        for_each_subsys(root, ss) {
4370                if (ss->post_clone)
4371                        ss->post_clone(ss, child);
4372        }
4373
4374        /* All seems fine. Finish by moving the task into the new cgroup */
4375        ret = cgroup_attach_task(child, tsk);
4376        mutex_unlock(&cgroup_mutex);
4377
4378 out_release:
4379        mutex_unlock(&inode->i_mutex);
4380
4381        mutex_lock(&cgroup_mutex);
4382        put_css_set(cg);
4383        mutex_unlock(&cgroup_mutex);
4384        deactivate_super(root->sb);
4385        return ret;
4386}
4387
4388/**
4389 * cgroup_is_descendant - see if @cgrp is a descendant of @task's cgrp
4390 * @cgrp: the cgroup in question
4391 * @task: the task in question
4392 *
4393 * See if @cgrp is a descendant of @task's cgroup in the appropriate
4394 * hierarchy.
4395 *
4396 * If we are sending in dummytop, then presumably we are creating
4397 * the top cgroup in the subsystem.
4398 *
4399 * Called only by the ns (nsproxy) cgroup.
4400 */
4401int cgroup_is_descendant(const struct cgroup *cgrp, struct task_struct *task)
4402{
4403        int ret;
4404        struct cgroup *target;
4405
4406        if (cgrp == dummytop)
4407                return 1;
4408
4409        target = task_cgroup_from_root(task, cgrp->root);
4410        while (cgrp != target && cgrp!= cgrp->top_cgroup)
4411                cgrp = cgrp->parent;
4412        ret = (cgrp == target);
4413        return ret;
4414}
4415
4416static void check_for_release(struct cgroup *cgrp)
4417{
4418        /* All of these checks rely on RCU to keep the cgroup
4419         * structure alive */
4420        if (cgroup_is_releasable(cgrp) && !atomic_read(&cgrp->count)
4421            && list_empty(&cgrp->children) && !cgroup_has_css_refs(cgrp)) {
4422                /* Control Group is currently removeable. If it's not
4423                 * already queued for a userspace notification, queue
4424                 * it now */
4425                int need_schedule_work = 0;
4426                spin_lock(&release_list_lock);
4427                if (!cgroup_is_removed(cgrp) &&
4428                    list_empty(&cgrp->release_list)) {
4429                        list_add(&cgrp->release_list, &release_list);
4430                        need_schedule_work = 1;
4431                }
4432                spin_unlock(&release_list_lock);
4433                if (need_schedule_work)
4434                        schedule_work(&release_agent_work);
4435        }
4436}
4437
4438/* Caller must verify that the css is not for root cgroup */
4439void __css_put(struct cgroup_subsys_state *css, int count)
4440{
4441        struct cgroup *cgrp = css->cgroup;
4442        int val;
4443        rcu_read_lock();
4444        val = atomic_sub_return(count, &css->refcnt);
4445        if (val == 1) {
4446                if (notify_on_release(cgrp)) {
4447                        set_bit(CGRP_RELEASABLE, &cgrp->flags);
4448                        check_for_release(cgrp);
4449                }
4450                cgroup_wakeup_rmdir_waiter(cgrp);
4451        }
4452        rcu_read_unlock();
4453        WARN_ON_ONCE(val < 1);
4454}
4455EXPORT_SYMBOL_GPL(__css_put);
4456
4457/*
4458 * Notify userspace when a cgroup is released, by running the
4459 * configured release agent with the name of the cgroup (path
4460 * relative to the root of cgroup file system) as the argument.
4461 *
4462 * Most likely, this user command will try to rmdir this cgroup.
4463 *
4464 * This races with the possibility that some other task will be
4465 * attached to this cgroup before it is removed, or that some other
4466 * user task will 'mkdir' a child cgroup of this cgroup.  That's ok.
4467 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
4468 * unused, and this cgroup will be reprieved from its death sentence,
4469 * to continue to serve a useful existence.  Next time it's released,
4470 * we will get notified again, if it still has 'notify_on_release' set.
4471 *
4472 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
4473 * means only wait until the task is successfully execve()'d.  The
4474 * separate release agent task is forked by call_usermodehelper(),
4475 * then control in this thread returns here, without waiting for the
4476 * release agent task.  We don't bother to wait because the caller of
4477 * this routine has no use for the exit status of the release agent
4478 * task, so no sense holding our caller up for that.
4479 */
4480static void cgroup_release_agent(struct work_struct *work)
4481{
4482        BUG_ON(work != &release_agent_work);
4483        mutex_lock(&cgroup_mutex);
4484        spin_lock(&release_list_lock);
4485        while (!list_empty(&release_list)) {
4486                char *argv[3], *envp[3];
4487                int i;
4488                char *pathbuf = NULL, *agentbuf = NULL;
4489                struct cgroup *cgrp = list_entry(release_list.next,
4490                                                    struct cgroup,
4491                                                    release_list);
4492                list_del_init(&cgrp->release_list);
4493                spin_unlock(&release_list_lock);
4494                pathbuf = kmalloc(PAGE_SIZE, GFP_KERNEL);
4495                if (!pathbuf)
4496                        goto continue_free;
4497                if (cgroup_path(cgrp, pathbuf, PAGE_SIZE) < 0)
4498                        goto continue_free;
4499                agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL);
4500                if (!agentbuf)
4501                        goto continue_free;
4502
4503                i = 0;
4504                argv[i++] = agentbuf;
4505                argv[i++] = pathbuf;
4506                argv[i] = NULL;
4507
4508                i = 0;
4509                /* minimal command environment */
4510                envp[i++] = "HOME=/";
4511                envp[i++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
4512                envp[i] = NULL;
4513
4514                /* Drop the lock while we invoke the usermode helper,
4515                 * since the exec could involve hitting disk and hence
4516                 * be a slow process */
4517                mutex_unlock(&cgroup_mutex);
4518                call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC);
4519                mutex_lock(&cgroup_mutex);
4520 continue_free:
4521                kfree(pathbuf);
4522                kfree(agentbuf);
4523                spin_lock(&release_list_lock);
4524        }
4525        spin_unlock(&release_list_lock);
4526        mutex_unlock(&cgroup_mutex);
4527}
4528
4529static int __init cgroup_disable(char *str)
4530{
4531        int i;
4532        char *token;
4533
4534        while ((token = strsep(&str, ",")) != NULL) {
4535                if (!*token)
4536                        continue;
4537                /*
4538                 * cgroup_disable, being at boot time, can't know about module
4539                 * subsystems, so we don't worry about them.
4540                 */
4541                for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
4542                        struct cgroup_subsys *ss = subsys[i];
4543
4544                        if (!strcmp(token, ss->name)) {
4545                                ss->disabled = 1;
4546                                printk(KERN_INFO "Disabling %s control group"
4547                                        " subsystem\n", ss->name);
4548                                break;
4549                        }
4550                }
4551        }
4552        return 1;
4553}
4554__setup("cgroup_disable=", cgroup_disable);
4555
4556/*
4557 * Functons for CSS ID.
4558 */
4559
4560/*
4561 *To get ID other than 0, this should be called when !cgroup_is_removed().
4562 */
4563unsigned short css_id(struct cgroup_subsys_state *css)
4564{
4565        struct css_id *cssid;
4566
4567        /*
4568         * This css_id() can return correct value when somone has refcnt
4569         * on this or this is under rcu_read_lock(). Once css->id is allocated,
4570         * it's unchanged until freed.
4571         */
4572        cssid = rcu_dereference_check(css->id,
4573                        rcu_read_lock_held() || atomic_read(&css->refcnt));
4574
4575        if (cssid)
4576                return cssid->id;
4577        return 0;
4578}
4579EXPORT_SYMBOL_GPL(css_id);
4580
4581unsigned short css_depth(struct cgroup_subsys_state *css)
4582{
4583        struct css_id *cssid;
4584
4585        cssid = rcu_dereference_check(css->id,
4586                        rcu_read_lock_held() || atomic_read(&css->refcnt));
4587
4588        if (cssid)
4589                return cssid->depth;
4590        return 0;
4591}
4592EXPORT_SYMBOL_GPL(css_depth);
4593
4594/**
4595 *  css_is_ancestor - test "root" css is an ancestor of "child"
4596 * @child: the css to be tested.
4597 * @root: the css supporsed to be an ancestor of the child.
4598 *
4599 * Returns true if "root" is an ancestor of "child" in its hierarchy. Because
4600 * this function reads css->id, this use rcu_dereference() and rcu_read_lock().
4601 * But, considering usual usage, the csses should be valid objects after test.
4602 * Assuming that the caller will do some action to the child if this returns
4603 * returns true, the caller must take "child";s reference count.
4604 * If "child" is valid object and this returns true, "root" is valid, too.
4605 */
4606
4607bool css_is_ancestor(struct cgroup_subsys_state *child,
4608                    const struct cgroup_subsys_state *root)
4609{
4610        struct css_id *child_id;
4611        struct css_id *root_id;
4612        bool ret = true;
4613
4614        rcu_read_lock();
4615        child_id  = rcu_dereference(child->id);
4616        root_id = rcu_dereference(root->id);
4617        if (!child_id
4618            || !root_id
4619            || (child_id->depth < root_id->depth)
4620            || (child_id->stack[root_id->depth] != root_id->id))
4621                ret = false;
4622        rcu_read_unlock();
4623        return ret;
4624}
4625
4626static void __free_css_id_cb(struct rcu_head *head)
4627{
4628        struct css_id *id;
4629
4630        id = container_of(head, struct css_id, rcu_head);
4631        kfree(id);
4632}
4633
4634void free_css_id(struct cgroup_subsys *ss, struct cgroup_subsys_state *css)
4635{
4636        struct css_id *id = css->id;
4637        /* When this is called before css_id initialization, id can be NULL */
4638        if (!id)
4639                return;
4640
4641        BUG_ON(!ss->use_id);
4642
4643        rcu_assign_pointer(id->css, NULL);
4644        rcu_assign_pointer(css->id, NULL);
4645        spin_lock(&ss->id_lock);
4646        idr_remove(&ss->idr, id->id);
4647        spin_unlock(&ss->id_lock);
4648        call_rcu(&id->rcu_head, __free_css_id_cb);
4649}
4650EXPORT_SYMBOL_GPL(free_css_id);
4651
4652/*
4653 * This is called by init or create(). Then, calls to this function are
4654 * always serialized (By cgroup_mutex() at create()).
4655 */
4656
4657static struct css_id *get_new_cssid(struct cgroup_subsys *ss, int depth)
4658{
4659        struct css_id *newid;
4660        int myid, error, size;
4661
4662        BUG_ON(!ss->use_id);
4663
4664        size = sizeof(*newid) + sizeof(unsigned short) * (depth + 1);
4665        newid = kzalloc(size, GFP_KERNEL);
4666        if (!newid)
4667                return ERR_PTR(-ENOMEM);
4668        /* get id */
4669        if (unlikely(!idr_pre_get(&ss->idr, GFP_KERNEL))) {
4670                error = -ENOMEM;
4671                goto err_out;
4672        }
4673        spin_lock(&ss->id_lock);
4674        /* Don't use 0. allocates an ID of 1-65535 */
4675        error = idr_get_new_above(&ss->idr, newid, 1, &myid);
4676        spin_unlock(&ss->id_lock);
4677
4678        /* Returns error when there are no free spaces for new ID.*/
4679        if (error) {
4680                error = -ENOSPC;
4681                goto err_out;
4682        }
4683        if (myid > CSS_ID_MAX)
4684                goto remove_idr;
4685
4686        newid->id = myid;
4687        newid->depth = depth;
4688        return newid;
4689remove_idr:
4690        error = -ENOSPC;
4691        spin_lock(&ss->id_lock);
4692        idr_remove(&ss->idr, myid);
4693        spin_unlock(&ss->id_lock);
4694err_out:
4695        kfree(newid);
4696        return ERR_PTR(error);
4697
4698}
4699
4700static int __init_or_module cgroup_init_idr(struct cgroup_subsys *ss,
4701                                            struct cgroup_subsys_state *rootcss)
4702{
4703        struct css_id *newid;
4704
4705        spin_lock_init(&ss->id_lock);
4706        idr_init(&ss->idr);
4707
4708        newid = get_new_cssid(ss, 0);
4709        if (IS_ERR(newid))
4710                return PTR_ERR(newid);
4711
4712        newid->stack[0] = newid->id;
4713        newid->css = rootcss;
4714        rootcss->id = newid;
4715        return 0;
4716}
4717
4718static int alloc_css_id(struct cgroup_subsys *ss, struct cgroup *parent,
4719                        struct cgroup *child)
4720{
4721        int subsys_id, i, depth = 0;
4722        struct cgroup_subsys_state *parent_css, *child_css;
4723        struct css_id *child_id, *parent_id;
4724
4725        subsys_id = ss->subsys_id;
4726        parent_css = parent->subsys[subsys_id];
4727        child_css = child->subsys[subsys_id];
4728        parent_id = parent_css->id;
4729        depth = parent_id->depth + 1;
4730
4731        child_id = get_new_cssid(ss, depth);
4732        if (IS_ERR(child_id))
4733                return PTR_ERR(child_id);
4734
4735        for (i = 0; i < depth; i++)
4736                child_id->stack[i] = parent_id->stack[i];
4737        child_id->stack[depth] = child_id->id;
4738        /*
4739         * child_id->css pointer will be set after this cgroup is available
4740         * see cgroup_populate_dir()
4741         */
4742        rcu_assign_pointer(child_css->id, child_id);
4743
4744        return 0;
4745}
4746
4747/**
4748 * css_lookup - lookup css by id
4749 * @ss: cgroup subsys to be looked into.
4750 * @id: the id
4751 *
4752 * Returns pointer to cgroup_subsys_state if there is valid one with id.
4753 * NULL if not. Should be called under rcu_read_lock()
4754 */
4755struct cgroup_subsys_state *css_lookup(struct cgroup_subsys *ss, int id)
4756{
4757        struct css_id *cssid = NULL;
4758
4759        BUG_ON(!ss->use_id);
4760        cssid = idr_find(&ss->idr, id);
4761
4762        if (unlikely(!cssid))
4763                return NULL;
4764
4765        return rcu_dereference(cssid->css);
4766}
4767EXPORT_SYMBOL_GPL(css_lookup);
4768
4769/**
4770 * css_get_next - lookup next cgroup under specified hierarchy.
4771 * @ss: pointer to subsystem
4772 * @id: current position of iteration.
4773 * @root: pointer to css. search tree under this.
4774 * @foundid: position of found object.
4775 *
4776 * Search next css under the specified hierarchy of rootid. Calling under
4777 * rcu_read_lock() is necessary. Returns NULL if it reaches the end.
4778 */
4779struct cgroup_subsys_state *
4780css_get_next(struct cgroup_subsys *ss, int id,
4781             struct cgroup_subsys_state *root, int *foundid)
4782{
4783        struct cgroup_subsys_state *ret = NULL;
4784        struct css_id *tmp;
4785        int tmpid;
4786        int rootid = css_id(root);
4787        int depth = css_depth(root);
4788
4789        if (!rootid)
4790                return NULL;
4791
4792        BUG_ON(!ss->use_id);
4793        /* fill start point for scan */
4794        tmpid = id;
4795        while (1) {
4796                /*
4797                 * scan next entry from bitmap(tree), tmpid is updated after
4798                 * idr_get_next().
4799                 */
4800                spin_lock(&ss->id_lock);
4801                tmp = idr_get_next(&ss->idr, &tmpid);
4802                spin_unlock(&ss->id_lock);
4803
4804                if (!tmp)
4805                        break;
4806                if (tmp->depth >= depth && tmp->stack[depth] == rootid) {
4807                        ret = rcu_dereference(tmp->css);
4808                        if (ret) {
4809                                *foundid = tmpid;
4810                                break;
4811                        }
4812                }
4813                /* continue to scan from next id */
4814                tmpid = tmpid + 1;
4815        }
4816        return ret;
4817}
4818
4819/*
4820 * get corresponding css from file open on cgroupfs directory
4821 */
4822struct cgroup_subsys_state *cgroup_css_from_dir(struct file *f, int id)
4823{
4824        struct cgroup *cgrp;
4825        struct inode *inode;
4826        struct cgroup_subsys_state *css;
4827
4828        inode = f->f_dentry->d_inode;
4829        /* check in cgroup filesystem dir */
4830        if (inode->i_op != &cgroup_dir_inode_operations)
4831                return ERR_PTR(-EBADF);
4832
4833        if (id < 0 || id >= CGROUP_SUBSYS_COUNT)
4834                return ERR_PTR(-EINVAL);
4835
4836        /* get cgroup */
4837        cgrp = __d_cgrp(f->f_dentry);
4838        css = cgrp->subsys[id];
4839        return css ? css : ERR_PTR(-ENOENT);
4840}
4841
4842#ifdef CONFIG_CGROUP_DEBUG
4843static struct cgroup_subsys_state *debug_create(struct cgroup_subsys *ss,
4844                                                   struct cgroup *cont)
4845{
4846        struct cgroup_subsys_state *css = kzalloc(sizeof(*css), GFP_KERNEL);
4847
4848        if (!css)
4849                return ERR_PTR(-ENOMEM);
4850
4851        return css;
4852}
4853
4854static void debug_destroy(struct cgroup_subsys *ss, struct cgroup *cont)
4855{
4856        kfree(cont->subsys[debug_subsys_id]);
4857}
4858
4859static u64 cgroup_refcount_read(struct cgroup *cont, struct cftype *cft)
4860{
4861        return atomic_read(&cont->count);
4862}
4863
4864static u64 debug_taskcount_read(struct cgroup *cont, struct cftype *cft)
4865{
4866        return cgroup_task_count(cont);