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 recieve.
 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 int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry);
 768static int cgroup_populate_dir(struct cgroup *cgrp);
 769static const struct inode_operations cgroup_dir_inode_operations;
 770static const struct file_operations proc_cgroupstats_operations;
 771
 772static struct backing_dev_info cgroup_backing_dev_info = {
 773        .name           = "cgroup",
 774        .capabilities   = BDI_CAP_NO_ACCT_AND_WRITEBACK,
 775};
 776
 777static int alloc_css_id(struct cgroup_subsys *ss,
 778                        struct cgroup *parent, struct cgroup *child);
 779
 780static struct inode *cgroup_new_inode(mode_t mode, struct super_block *sb)
 781{
 782        struct inode *inode = new_inode(sb);
 783
 784        if (inode) {
 785                inode->i_ino = get_next_ino();
 786                inode->i_mode = mode;
 787                inode->i_uid = current_fsuid();
 788                inode->i_gid = current_fsgid();
 789                inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 790                inode->i_mapping->backing_dev_info = &cgroup_backing_dev_info;
 791        }
 792        return inode;
 793}
 794
 795/*
 796 * Call subsys's pre_destroy handler.
 797 * This is called before css refcnt check.
 798 */
 799static int cgroup_call_pre_destroy(struct cgroup *cgrp)
 800{
 801        struct cgroup_subsys *ss;
 802        int ret = 0;
 803
 804        for_each_subsys(cgrp->root, ss)
 805                if (ss->pre_destroy) {
 806                        ret = ss->pre_destroy(ss, cgrp);
 807                        if (ret)
 808                                break;
 809                }
 810
 811        return ret;
 812}
 813
 814static void free_cgroup_rcu(struct rcu_head *obj)
 815{
 816        struct cgroup *cgrp = container_of(obj, struct cgroup, rcu_head);
 817
 818        kfree(cgrp);
 819}
 820
 821static void cgroup_diput(struct dentry *dentry, struct inode *inode)
 822{
 823        /* is dentry a directory ? if so, kfree() associated cgroup */
 824        if (S_ISDIR(inode->i_mode)) {
 825                struct cgroup *cgrp = dentry->d_fsdata;
 826                struct cgroup_subsys *ss;
 827                BUG_ON(!(cgroup_is_removed(cgrp)));
 828                /* It's possible for external users to be holding css
 829                 * reference counts on a cgroup; css_put() needs to
 830                 * be able to access the cgroup after decrementing
 831                 * the reference count in order to know if it needs to
 832                 * queue the cgroup to be handled by the release
 833                 * agent */
 834                synchronize_rcu();
 835
 836                mutex_lock(&cgroup_mutex);
 837                /*
 838                 * Release the subsystem state objects.
 839                 */
 840                for_each_subsys(cgrp->root, ss)
 841                        ss->destroy(ss, cgrp);
 842
 843                cgrp->root->number_of_cgroups--;
 844                mutex_unlock(&cgroup_mutex);
 845
 846                /*
 847                 * Drop the active superblock reference that we took when we
 848                 * created the cgroup
 849                 */
 850                deactivate_super(cgrp->root->sb);
 851
 852                /*
 853                 * if we're getting rid of the cgroup, refcount should ensure
 854                 * that there are no pidlists left.
 855                 */
 856                BUG_ON(!list_empty(&cgrp->pidlists));
 857
 858                call_rcu(&cgrp->rcu_head, free_cgroup_rcu);
 859        }
 860        iput(inode);
 861}
 862
 863static void remove_dir(struct dentry *d)
 864{
 865        struct dentry *parent = dget(d->d_parent);
 866
 867        d_delete(d);
 868        simple_rmdir(parent->d_inode, d);
 869        dput(parent);
 870}
 871
 872static void cgroup_clear_directory(struct dentry *dentry)
 873{
 874        struct list_head *node;
 875
 876        BUG_ON(!mutex_is_locked(&dentry->d_inode->i_mutex));
 877        spin_lock(&dcache_lock);
 878        node = dentry->d_subdirs.next;
 879        while (node != &dentry->d_subdirs) {
 880                struct dentry *d = list_entry(node, struct dentry, d_u.d_child);
 881                list_del_init(node);
 882                if (d->d_inode) {
 883                        /* This should never be called on a cgroup
 884                         * directory with child cgroups */
 885                        BUG_ON(d->d_inode->i_mode & S_IFDIR);
 886                        d = dget_locked(d);
 887                        spin_unlock(&dcache_lock);
 888                        d_delete(d);
 889                        simple_unlink(dentry->d_inode, d);
 890                        dput(d);
 891                        spin_lock(&dcache_lock);
 892                }
 893                node = dentry->d_subdirs.next;
 894        }
 895        spin_unlock(&dcache_lock);
 896}
 897
 898/*
 899 * NOTE : the dentry must have been dget()'ed
 900 */
 901static void cgroup_d_remove_dir(struct dentry *dentry)
 902{
 903        cgroup_clear_directory(dentry);
 904
 905        spin_lock(&dcache_lock);
 906        list_del_init(&dentry->d_u.d_child);
 907        spin_unlock(&dcache_lock);
 908        remove_dir(dentry);
 909}
 910
 911/*
 912 * A queue for waiters to do rmdir() cgroup. A tasks will sleep when
 913 * cgroup->count == 0 && list_empty(&cgroup->children) && subsys has some
 914 * reference to css->refcnt. In general, this refcnt is expected to goes down
 915 * to zero, soon.
 916 *
 917 * CGRP_WAIT_ON_RMDIR flag is set under cgroup's inode->i_mutex;
 918 */
 919DECLARE_WAIT_QUEUE_HEAD(cgroup_rmdir_waitq);
 920
 921static void cgroup_wakeup_rmdir_waiter(struct cgroup *cgrp)
 922{
 923        if (unlikely(test_and_clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags)))
 924                wake_up_all(&cgroup_rmdir_waitq);
 925}
 926
 927void cgroup_exclude_rmdir(struct cgroup_subsys_state *css)
 928{
 929        css_get(css);
 930}
 931
 932void cgroup_release_and_wakeup_rmdir(struct cgroup_subsys_state *css)
 933{
 934        cgroup_wakeup_rmdir_waiter(css->cgroup);
 935        css_put(css);
 936}
 937
 938/*
 939 * Call with cgroup_mutex held. Drops reference counts on modules, including
 940 * any duplicate ones that parse_cgroupfs_options took. If this function
 941 * returns an error, no reference counts are touched.
 942 */
 943static int rebind_subsystems(struct cgroupfs_root *root,
 944                              unsigned long final_bits)
 945{
 946        unsigned long added_bits, removed_bits;
 947        struct cgroup *cgrp = &root->top_cgroup;
 948        int i;
 949
 950        BUG_ON(!mutex_is_locked(&cgroup_mutex));
 951
 952        removed_bits = root->actual_subsys_bits & ~final_bits;
 953        added_bits = final_bits & ~root->actual_subsys_bits;
 954        /* Check that any added subsystems are currently free */
 955        for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
 956                unsigned long bit = 1UL << i;
 957                struct cgroup_subsys *ss = subsys[i];
 958                if (!(bit & added_bits))
 959                        continue;
 960                /*
 961                 * Nobody should tell us to do a subsys that doesn't exist:
 962                 * parse_cgroupfs_options should catch that case and refcounts
 963                 * ensure that subsystems won't disappear once selected.
 964                 */
 965                BUG_ON(ss == NULL);
 966                if (ss->root != &rootnode) {
 967                        /* Subsystem isn't free */
 968                        return -EBUSY;
 969                }
 970        }
 971
 972        /* Currently we don't handle adding/removing subsystems when
 973         * any child cgroups exist. This is theoretically supportable
 974         * but involves complex error handling, so it's being left until
 975         * later */
 976        if (root->number_of_cgroups > 1)
 977                return -EBUSY;
 978
 979        /* Process each subsystem */
 980        for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
 981                struct cgroup_subsys *ss = subsys[i];
 982                unsigned long bit = 1UL << i;
 983                if (bit & added_bits) {
 984                        /* We're binding this subsystem to this hierarchy */
 985                        BUG_ON(ss == NULL);
 986                        BUG_ON(cgrp->subsys[i]);
 987                        BUG_ON(!dummytop->subsys[i]);
 988                        BUG_ON(dummytop->subsys[i]->cgroup != dummytop);
 989                        mutex_lock(&ss->hierarchy_mutex);
 990                        cgrp->subsys[i] = dummytop->subsys[i];
 991                        cgrp->subsys[i]->cgroup = cgrp;
 992                        list_move(&ss->sibling, &root->subsys_list);
 993                        ss->root = root;
 994                        if (ss->bind)
 995                                ss->bind(ss, cgrp);
 996                        mutex_unlock(&ss->hierarchy_mutex);
 997                        /* refcount was already taken, and we're keeping it */
 998                } else if (bit & removed_bits) {
 999                        /* We're removing this subsystem */
1000                        BUG_ON(ss == NULL);
1001                        BUG_ON(cgrp->subsys[i] != dummytop->subsys[i]);
1002                        BUG_ON(cgrp->subsys[i]->cgroup != cgrp);
1003                        mutex_lock(&ss->hierarchy_mutex);
1004                        if (ss->bind)
1005                                ss->bind(ss, dummytop);
1006                        dummytop->subsys[i]->cgroup = dummytop;
1007                        cgrp->subsys[i] = NULL;
1008                        subsys[i]->root = &rootnode;
1009                        list_move(&ss->sibling, &rootnode.subsys_list);
1010                        mutex_unlock(&ss->hierarchy_mutex);
1011                        /* subsystem is now free - drop reference on module */
1012                        module_put(ss->module);
1013                } else if (bit & final_bits) {
1014                        /* Subsystem state should already exist */
1015                        BUG_ON(ss == NULL);
1016                        BUG_ON(!cgrp->subsys[i]);
1017                        /*
1018                         * a refcount was taken, but we already had one, so
1019                         * drop the extra reference.
1020                         */
1021                        module_put(ss->module);
1022#ifdef CONFIG_MODULE_UNLOAD
1023                        BUG_ON(ss->module && !module_refcount(ss->module));
1024#endif
1025                } else {
1026                        /* Subsystem state shouldn't exist */
1027                        BUG_ON(cgrp->subsys[i]);
1028                }
1029        }
1030        root->subsys_bits = root->actual_subsys_bits = final_bits;
1031        synchronize_rcu();
1032
1033        return 0;
1034}
1035
1036static int cgroup_show_options(struct seq_file *seq, struct vfsmount *vfs)
1037{
1038        struct cgroupfs_root *root = vfs->mnt_sb->s_fs_info;
1039        struct cgroup_subsys *ss;
1040
1041        mutex_lock(&cgroup_mutex);
1042        for_each_subsys(root, ss)
1043                seq_printf(seq, ",%s", ss->name);
1044        if (test_bit(ROOT_NOPREFIX, &root->flags))
1045                seq_puts(seq, ",noprefix");
1046        if (strlen(root->release_agent_path))
1047                seq_printf(seq, ",release_agent=%s", root->release_agent_path);
1048        if (clone_children(&root->top_cgroup))
1049                seq_puts(seq, ",clone_children");
1050        if (strlen(root->name))
1051                seq_printf(seq, ",name=%s", root->name);
1052        mutex_unlock(&cgroup_mutex);
1053        return 0;
1054}
1055
1056struct cgroup_sb_opts {
1057        unsigned long subsys_bits;
1058        unsigned long flags;
1059        char *release_agent;
1060        bool clone_children;
1061        char *name;
1062        /* User explicitly requested empty subsystem */
1063        bool none;
1064
1065        struct cgroupfs_root *new_root;
1066
1067};
1068
1069/*
1070 * Convert a hierarchy specifier into a bitmask of subsystems and flags. Call
1071 * with cgroup_mutex held to protect the subsys[] array. This function takes
1072 * refcounts on subsystems to be used, unless it returns error, in which case
1073 * no refcounts are taken.
1074 */
1075static int parse_cgroupfs_options(char *data, struct cgroup_sb_opts *opts)
1076{
1077        char *token, *o = data;
1078        bool all_ss = false, one_ss = false;
1079        unsigned long mask = (unsigned long)-1;
1080        int i;
1081        bool module_pin_failed = false;
1082
1083        BUG_ON(!mutex_is_locked(&cgroup_mutex));
1084
1085#ifdef CONFIG_CPUSETS
1086        mask = ~(1UL << cpuset_subsys_id);
1087#endif
1088
1089        memset(opts, 0, sizeof(*opts));
1090
1091        while ((token = strsep(&o, ",")) != NULL) {
1092                if (!*token)
1093                        return -EINVAL;
1094                if (!strcmp(token, "none")) {
1095                        /* Explicitly have no subsystems */
1096                        opts->none = true;
1097                        continue;
1098                }
1099                if (!strcmp(token, "all")) {
1100                        /* Mutually exclusive option 'all' + subsystem name */
1101                        if (one_ss)
1102                                return -EINVAL;
1103                        all_ss = true;
1104                        continue;
1105                }
1106                if (!strcmp(token, "noprefix")) {
1107                        set_bit(ROOT_NOPREFIX, &opts->flags);
1108                        continue;
1109                }
1110                if (!strcmp(token, "clone_children")) {
1111                        opts->clone_children = true;
1112                        continue;
1113                }
1114                if (!strncmp(token, "release_agent=", 14)) {
1115                        /* Specifying two release agents is forbidden */
1116                        if (opts->release_agent)
1117                                return -EINVAL;
1118                        opts->release_agent =
1119                                kstrndup(token + 14, PATH_MAX - 1, GFP_KERNEL);
1120                        if (!opts->release_agent)
1121                                return -ENOMEM;
1122                        continue;
1123                }
1124                if (!strncmp(token, "name=", 5)) {
1125                        const char *name = token + 5;
1126                        /* Can't specify an empty name */
1127                        if (!strlen(name))
1128                                return -EINVAL;
1129                        /* Must match [\w.-]+ */
1130                        for (i = 0; i < strlen(name); i++) {
1131                                char c = name[i];
1132                                if (isalnum(c))
1133                                        continue;
1134                                if ((c == '.') || (c == '-') || (c == '_'))
1135                                        continue;
1136                                return -EINVAL;
1137                        }
1138                        /* Specifying two names is forbidden */
1139                        if (opts->name)
1140                                return -EINVAL;
1141                        opts->name = kstrndup(name,
1142                                              MAX_CGROUP_ROOT_NAMELEN - 1,
1143                                              GFP_KERNEL);
1144                        if (!opts->name)
1145                                return -ENOMEM;
1146
1147                        continue;
1148                }
1149
1150                for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1151                        struct cgroup_subsys *ss = subsys[i];
1152                        if (ss == NULL)
1153                                continue;
1154                        if (strcmp(token, ss->name))
1155                                continue;
1156                        if (ss->disabled)
1157                                continue;
1158
1159                        /* Mutually exclusive option 'all' + subsystem name */
1160                        if (all_ss)
1161                                return -EINVAL;
1162                        set_bit(i, &opts->subsys_bits);
1163                        one_ss = true;
1164
1165                        break;
1166                }
1167                if (i == CGROUP_SUBSYS_COUNT)
1168                        return -ENOENT;
1169        }
1170
1171        /*
1172         * If the 'all' option was specified select all the subsystems,
1173         * otherwise 'all, 'none' and a subsystem name options were not
1174         * specified, let's default to 'all'
1175         */
1176        if (all_ss || (!all_ss && !one_ss && !opts->none)) {
1177                for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1178                        struct cgroup_subsys *ss = subsys[i];
1179                        if (ss == NULL)
1180                                continue;
1181                        if (ss->disabled)
1182                                continue;
1183                        set_bit(i, &opts->subsys_bits);
1184                }
1185        }
1186
1187        /* Consistency checks */
1188
1189        /*
1190         * Option noprefix was introduced just for backward compatibility
1191         * with the old cpuset, so we allow noprefix only if mounting just
1192         * the cpuset subsystem.
1193         */
1194        if (test_bit(ROOT_NOPREFIX, &opts->flags) &&
1195            (opts->subsys_bits & mask))
1196                return -EINVAL;
1197
1198
1199        /* Can't specify "none" and some subsystems */
1200        if (opts->subsys_bits && opts->none)
1201                return -EINVAL;
1202
1203        /*
1204         * We either have to specify by name or by subsystems. (So all
1205         * empty hierarchies must have a name).
1206         */
1207        if (!opts->subsys_bits && !opts->name)
1208                return -EINVAL;
1209
1210        /*
1211         * Grab references on all the modules we'll need, so the subsystems
1212         * don't dance around before rebind_subsystems attaches them. This may
1213         * take duplicate reference counts on a subsystem that's already used,
1214         * but rebind_subsystems handles this case.
1215         */
1216        for (i = CGROUP_BUILTIN_SUBSYS_COUNT; i < CGROUP_SUBSYS_COUNT; i++) {
1217                unsigned long bit = 1UL << i;
1218
1219                if (!(bit & opts->subsys_bits))
1220                        continue;
1221                if (!try_module_get(subsys[i]->module)) {
1222                        module_pin_failed = true;
1223                        break;
1224                }
1225        }
1226        if (module_pin_failed) {
1227                /*
1228                 * oops, one of the modules was going away. this means that we
1229                 * raced with a module_delete call, and to the user this is
1230                 * essentially a "subsystem doesn't exist" case.
1231                 */
1232                for (i--; i >= CGROUP_BUILTIN_SUBSYS_COUNT; i--) {
1233                        /* drop refcounts only on the ones we took */
1234                        unsigned long bit = 1UL << i;
1235
1236                        if (!(bit & opts->subsys_bits))
1237                                continue;
1238                        module_put(subsys[i]->module);
1239                }
1240                return -ENOENT;
1241        }
1242
1243        return 0;
1244}
1245
1246static void drop_parsed_module_refcounts(unsigned long subsys_bits)
1247{
1248        int i;
1249        for (i = CGROUP_BUILTIN_SUBSYS_COUNT; i < CGROUP_SUBSYS_COUNT; i++) {
1250                unsigned long bit = 1UL << i;
1251
1252                if (!(bit & subsys_bits))
1253                        continue;
1254                module_put(subsys[i]->module);
1255        }
1256}
1257
1258static int cgroup_remount(struct super_block *sb, int *flags, char *data)
1259{
1260        int ret = 0;
1261        struct cgroupfs_root *root = sb->s_fs_info;
1262        struct cgroup *cgrp = &root->top_cgroup;
1263        struct cgroup_sb_opts opts;
1264
1265        mutex_lock(&cgrp->dentry->d_inode->i_mutex);
1266        mutex_lock(&cgroup_mutex);
1267
1268        /* See what subsystems are wanted */
1269        ret = parse_cgroupfs_options(data, &opts);
1270        if (ret)
1271                goto out_unlock;
1272
1273        /* Don't allow flags or name to change at remount */
1274        if (opts.flags != root->flags ||
1275            (opts.name && strcmp(opts.name, root->name))) {
1276                ret = -EINVAL;
1277                drop_parsed_module_refcounts(opts.subsys_bits);
1278                goto out_unlock;
1279        }
1280
1281        ret = rebind_subsystems(root, opts.subsys_bits);
1282        if (ret) {
1283                drop_parsed_module_refcounts(opts.subsys_bits);
1284                goto out_unlock;
1285        }
1286
1287        /* (re)populate subsystem files */
1288        cgroup_populate_dir(cgrp);
1289
1290        if (opts.release_agent)
1291                strcpy(root->release_agent_path, opts.release_agent);
1292 out_unlock:
1293        kfree(opts.release_agent);
1294        kfree(opts.name);
1295        mutex_unlock(&cgroup_mutex);
1296        mutex_unlock(&cgrp->dentry->d_inode->i_mutex);
1297        return ret;
1298}
1299
1300static const struct super_operations cgroup_ops = {
1301        .statfs = simple_statfs,
1302        .drop_inode = generic_delete_inode,
1303        .show_options = cgroup_show_options,
1304        .remount_fs = cgroup_remount,
1305};
1306
1307static void init_cgroup_housekeeping(struct cgroup *cgrp)
1308{
1309        INIT_LIST_HEAD(&cgrp->sibling);
1310        INIT_LIST_HEAD(&cgrp->children);
1311        INIT_LIST_HEAD(&cgrp->css_sets);
1312        INIT_LIST_HEAD(&cgrp->release_list);
1313        INIT_LIST_HEAD(&cgrp->pidlists);
1314        mutex_init(&cgrp->pidlist_mutex);
1315        INIT_LIST_HEAD(&cgrp->event_list);
1316        spin_lock_init(&cgrp->event_list_lock);
1317}
1318
1319static void init_cgroup_root(struct cgroupfs_root *root)
1320{
1321        struct cgroup *cgrp = &root->top_cgroup;
1322        INIT_LIST_HEAD(&root->subsys_list);
1323        INIT_LIST_HEAD(&root->root_list);
1324        root->number_of_cgroups = 1;
1325        cgrp->root = root;
1326        cgrp->top_cgroup = cgrp;
1327        init_cgroup_housekeeping(cgrp);
1328}
1329
1330static bool init_root_id(struct cgroupfs_root *root)
1331{
1332        int ret = 0;
1333
1334        do {
1335                if (!ida_pre_get(&hierarchy_ida, GFP_KERNEL))
1336                        return false;
1337                spin_lock(&hierarchy_id_lock);
1338                /* Try to allocate the next unused ID */
1339                ret = ida_get_new_above(&hierarchy_ida, next_hierarchy_id,
1340                                        &root->hierarchy_id);
1341                if (ret == -ENOSPC)
1342                        /* Try again starting from 0 */
1343                        ret = ida_get_new(&hierarchy_ida, &root->hierarchy_id);
1344                if (!ret) {
1345                        next_hierarchy_id = root->hierarchy_id + 1;
1346                } else if (ret != -EAGAIN) {
1347                        /* Can only get here if the 31-bit IDR is full ... */
1348                        BUG_ON(ret);
1349                }
1350                spin_unlock(&hierarchy_id_lock);
1351        } while (ret);
1352        return true;
1353}
1354
1355static int cgroup_test_super(struct super_block *sb, void *data)
1356{
1357        struct cgroup_sb_opts *opts = data;
1358        struct cgroupfs_root *root = sb->s_fs_info;
1359
1360        /* If we asked for a name then it must match */
1361        if (opts->name && strcmp(opts->name, root->name))
1362                return 0;
1363
1364        /*
1365         * If we asked for subsystems (or explicitly for no
1366         * subsystems) then they must match
1367         */
1368        if ((opts->subsys_bits || opts->none)
1369            && (opts->subsys_bits != root->subsys_bits))
1370                return 0;
1371
1372        return 1;
1373}
1374
1375static struct cgroupfs_root *cgroup_root_from_opts(struct cgroup_sb_opts *opts)
1376{
1377        struct cgroupfs_root *root;
1378
1379        if (!opts->subsys_bits && !opts->none)
1380                return NULL;
1381
1382        root = kzalloc(sizeof(*root), GFP_KERNEL);
1383        if (!root)
1384                return ERR_PTR(-ENOMEM);
1385
1386        if (!init_root_id(root)) {
1387                kfree(root);
1388                return ERR_PTR(-ENOMEM);
1389        }
1390        init_cgroup_root(root);
1391
1392        root->subsys_bits = opts->subsys_bits;
1393        root->flags = opts->flags;
1394        if (opts->release_agent)
1395                strcpy(root->release_agent_path, opts->release_agent);
1396        if (opts->name)
1397                strcpy(root->name, opts->name);
1398        if (opts->clone_children)
1399                set_bit(CGRP_CLONE_CHILDREN, &root->top_cgroup.flags);
1400        return root;
1401}
1402
1403static void cgroup_drop_root(struct cgroupfs_root *root)
1404{
1405        if (!root)
1406                return;
1407
1408        BUG_ON(!root->hierarchy_id);
1409        spin_lock(&hierarchy_id_lock);
1410        ida_remove(&hierarchy_ida, root->hierarchy_id);
1411        spin_unlock(&hierarchy_id_lock);
1412        kfree(root);
1413}
1414
1415static int cgroup_set_super(struct super_block *sb, void *data)
1416{
1417        int ret;
1418        struct cgroup_sb_opts *opts = data;
1419
1420        /* If we don't have a new root, we can't set up a new sb */
1421        if (!opts->new_root)
1422                return -EINVAL;
1423
1424        BUG_ON(!opts->subsys_bits && !opts->none);
1425
1426        ret = set_anon_super(sb, NULL);
1427        if (ret)
1428                return ret;
1429
1430        sb->s_fs_info = opts->new_root;
1431        opts->new_root->sb = sb;
1432
1433        sb->s_blocksize = PAGE_CACHE_SIZE;
1434        sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
1435        sb->s_magic = CGROUP_SUPER_MAGIC;
1436        sb->s_op = &cgroup_ops;
1437
1438        return 0;
1439}
1440
1441static int cgroup_get_rootdir(struct super_block *sb)
1442{
1443        struct inode *inode =
1444                cgroup_new_inode(S_IFDIR | S_IRUGO | S_IXUGO | S_IWUSR, sb);
1445        struct dentry *dentry;
1446
1447        if (!inode)
1448                return -ENOMEM;
1449
1450        inode->i_fop = &simple_dir_operations;
1451        inode->i_op = &cgroup_dir_inode_operations;
1452        /* directories start off with i_nlink == 2 (for "." entry) */
1453        inc_nlink(inode);
1454        dentry = d_alloc_root(inode);
1455        if (!dentry) {
1456                iput(inode);
1457                return -ENOMEM;
1458        }
1459        sb->s_root = dentry;
1460        return 0;
1461}
1462
1463static struct dentry *cgroup_mount(struct file_system_type *fs_type,
1464                         int flags, const char *unused_dev_name,
1465                         void *data)
1466{
1467        struct cgroup_sb_opts opts;
1468        struct cgroupfs_root *root;
1469        int ret = 0;
1470        struct super_block *sb;
1471        struct cgroupfs_root *new_root;
1472
1473        /* First find the desired set of subsystems */
1474        mutex_lock(&cgroup_mutex);
1475        ret = parse_cgroupfs_options(data, &opts);
1476        mutex_unlock(&cgroup_mutex);
1477        if (ret)
1478                goto out_err;
1479
1480        /*
1481         * Allocate a new cgroup root. We may not need it if we're
1482         * reusing an existing hierarchy.
1483         */
1484        new_root = cgroup_root_from_opts(&opts);
1485        if (IS_ERR(new_root)) {
1486                ret = PTR_ERR(new_root);
1487                goto drop_modules;
1488        }
1489        opts.new_root = new_root;
1490
1491        /* Locate an existing or new sb for this hierarchy */
1492        sb = sget(fs_type, cgroup_test_super, cgroup_set_super, &opts);
1493        if (IS_ERR(sb)) {
1494                ret = PTR_ERR(sb);
1495                cgroup_drop_root(opts.new_root);
1496                goto drop_modules;
1497        }
1498
1499        root = sb->s_fs_info;
1500        BUG_ON(!root);
1501        if (root == opts.new_root) {
1502                /* We used the new root structure, so this is a new hierarchy */
1503                struct list_head tmp_cg_links;
1504                struct cgroup *root_cgrp = &root->top_cgroup;
1505                struct inode *inode;
1506                struct cgroupfs_root *existing_root;
1507                int i;
1508
1509                BUG_ON(sb->s_root != NULL);
1510
1511                ret = cgroup_get_rootdir(sb);
1512                if (ret)
1513                        goto drop_new_super;
1514                inode = sb->s_root->d_inode;
1515
1516                mutex_lock(&inode->i_mutex);
1517                mutex_lock(&cgroup_mutex);
1518
1519                if (strlen(root->name)) {
1520                        /* Check for name clashes with existing mounts */
1521                        for_each_active_root(existing_root) {
1522                                if (!strcmp(existing_root->name, root->name)) {
1523                                        ret = -EBUSY;
1524                                        mutex_unlock(&cgroup_mutex);
1525                                        mutex_unlock(&inode->i_mutex);
1526                                        goto drop_new_super;
1527                                }
1528                        }
1529                }
1530
1531                /*
1532                 * We're accessing css_set_count without locking
1533                 * css_set_lock here, but that's OK - it can only be
1534                 * increased by someone holding cgroup_lock, and
1535                 * that's us. The worst that can happen is that we
1536                 * have some link structures left over
1537                 */
1538                ret = allocate_cg_links(css_set_count, &tmp_cg_links);
1539                if (ret) {
1540                        mutex_unlock(&cgroup_mutex);
1541                        mutex_unlock(&inode->i_mutex);
1542                        goto drop_new_super;
1543                }
1544
1545                ret = rebind_subsystems(root, root->subsys_bits);
1546                if (ret == -EBUSY) {
1547                        mutex_unlock(&cgroup_mutex);
1548                        mutex_unlock(&inode->i_mutex);
1549                        free_cg_links(&tmp_cg_links);
1550                        goto drop_new_super;
1551                }
1552                /*
1553                 * There must be no failure case after here, since rebinding
1554                 * takes care of subsystems' refcounts, which are explicitly
1555                 * dropped in the failure exit path.
1556                 */
1557
1558                /* EBUSY should be the only error here */
1559                BUG_ON(ret);
1560
1561                list_add(&root->root_list, &roots);
1562                root_count++;
1563
1564                sb->s_root->d_fsdata = root_cgrp;
1565                root->top_cgroup.dentry = sb->s_root;
1566
1567                /* Link the top cgroup in this hierarchy into all
1568                 * the css_set objects */
1569                write_lock(&css_set_lock);
1570                for (i = 0; i < CSS_SET_TABLE_SIZE; i++) {
1571                        struct hlist_head *hhead = &css_set_table[i];
1572                        struct hlist_node *node;
1573                        struct css_set *cg;
1574
1575                        hlist_for_each_entry(cg, node, hhead, hlist)
1576                                link_css_set(&tmp_cg_links, cg, root_cgrp);
1577                }
1578                write_unlock(&css_set_lock);
1579
1580                free_cg_links(&tmp_cg_links);
1581
1582                BUG_ON(!list_empty(&root_cgrp->sibling));
1583                BUG_ON(!list_empty(&root_cgrp->children));
1584                BUG_ON(root->number_of_cgroups != 1);
1585
1586                cgroup_populate_dir(root_cgrp);
1587                mutex_unlock(&cgroup_mutex);
1588                mutex_unlock(&inode->i_mutex);
1589        } else {
1590                /*
1591                 * We re-used an existing hierarchy - the new root (if
1592                 * any) is not needed
1593                 */
1594                cgroup_drop_root(opts.new_root);
1595                /* no subsys rebinding, so refcounts don't change */
1596                drop_parsed_module_refcounts(opts.subsys_bits);
1597        }
1598
1599        kfree(opts.release_agent);
1600        kfree(opts.name);
1601        return dget(sb->s_root);
1602
1603 drop_new_super:
1604        deactivate_locked_super(sb);
1605 drop_modules:
1606        drop_parsed_module_refcounts(opts.subsys_bits);
1607 out_err:
1608        kfree(opts.release_agent);
1609        kfree(opts.name);
1610        return ERR_PTR(ret);
1611}
1612
1613static void cgroup_kill_sb(struct super_block *sb) {
1614        struct cgroupfs_root *root = sb->s_fs_info;
1615        struct cgroup *cgrp = &root->top_cgroup;
1616        int ret;
1617        struct cg_cgroup_link *link;
1618        struct cg_cgroup_link *saved_link;
1619
1620        BUG_ON(!root);
1621
1622        BUG_ON(root->number_of_cgroups != 1);
1623        BUG_ON(!list_empty(&cgrp->children));
1624        BUG_ON(!list_empty(&cgrp->sibling));
1625
1626        mutex_lock(&cgroup_mutex);
1627
1628        /* Rebind all subsystems back to the default hierarchy */
1629        ret = rebind_subsystems(root, 0);
1630        /* Shouldn't be able to fail ... */
1631        BUG_ON(ret);
1632
1633        /*
1634         * Release all the links from css_sets to this hierarchy's
1635         * root cgroup
1636         */
1637        write_lock(&css_set_lock);
1638
1639        list_for_each_entry_safe(link, saved_link, &cgrp->css_sets,
1640                                 cgrp_link_list) {
1641                list_del(&link->cg_link_list);
1642                list_del(&link->cgrp_link_list);
1643                kfree(link);
1644        }
1645        write_unlock(&css_set_lock);
1646
1647        if (!list_empty(&root->root_list)) {
1648                list_del(&root->root_list);
1649                root_count--;
1650        }
1651
1652        mutex_unlock(&cgroup_mutex);
1653
1654        kill_litter_super(sb);
1655        cgroup_drop_root(root);
1656}
1657
1658static struct file_system_type cgroup_fs_type = {
1659        .name = "cgroup",
1660        .mount = cgroup_mount,
1661        .kill_sb = cgroup_kill_sb,
1662};
1663
1664static struct kobject *cgroup_kobj;
1665
1666static inline struct cgroup *__d_cgrp(struct dentry *dentry)
1667{
1668        return dentry->d_fsdata;
1669}
1670
1671static inline struct cftype *__d_cft(struct dentry *dentry)
1672{
1673        return dentry->d_fsdata;
1674}
1675
1676/**
1677 * cgroup_path - generate the path of a cgroup
1678 * @cgrp: the cgroup in question
1679 * @buf: the buffer to write the path into
1680 * @buflen: the length of the buffer
1681 *
1682 * Called with cgroup_mutex held or else with an RCU-protected cgroup
1683 * reference.  Writes path of cgroup into buf.  Returns 0 on success,
1684 * -errno on error.
1685 */
1686int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen)
1687{
1688        char *start;
1689        struct dentry *dentry = rcu_dereference_check(cgrp->dentry,
1690                                                      rcu_read_lock_held() ||
1691                                                      cgroup_lock_is_held());
1692
1693        if (!dentry || cgrp == dummytop) {
1694                /*
1695                 * Inactive subsystems have no dentry for their root
1696                 * cgroup
1697                 */
1698                strcpy(buf, "/");
1699                return 0;
1700        }
1701
1702        start = buf + buflen;
1703
1704        *--start = '\0';
1705        for (;;) {
1706                int len = dentry->d_name.len;
1707
1708                if ((start -= len) < buf)
1709                        return -ENAMETOOLONG;
1710                memcpy(start, dentry->d_name.name, len);
1711                cgrp = cgrp->parent;
1712                if (!cgrp)
1713                        break;
1714
1715                dentry = rcu_dereference_check(cgrp->dentry,
1716                                               rcu_read_lock_held() ||
1717                                               cgroup_lock_is_held());
1718                if (!cgrp->parent)
1719                        continue;
1720                if (--start < buf)
1721                        return -ENAMETOOLONG;
1722                *start = '/';
1723        }
1724        memmove(buf, start, buf + buflen - start);
1725        return 0;
1726}
1727EXPORT_SYMBOL_GPL(cgroup_path);
1728
1729/**
1730 * cgroup_attach_task - attach task 'tsk' to cgroup 'cgrp'
1731 * @cgrp: the cgroup the task is attaching to
1732 * @tsk: the task to be attached
1733 *
1734 * Call holding cgroup_mutex. May take task_lock of
1735 * the task 'tsk' during call.
1736 */
1737int cgroup_attach_task(struct cgroup *cgrp, struct task_struct *tsk)
1738{
1739        int retval = 0;
1740        struct cgroup_subsys *ss, *failed_ss = NULL;
1741        struct cgroup *oldcgrp;
1742        struct css_set *cg;
1743        struct css_set *newcg;
1744        struct cgroupfs_root *root = cgrp->root;
1745
1746        /* Nothing to do if the task is already in that cgroup */
1747        oldcgrp = task_cgroup_from_root(tsk, root);
1748        if (cgrp == oldcgrp)
1749                return 0;
1750
1751        for_each_subsys(root, ss) {
1752                if (ss->can_attach) {
1753                        retval = ss->can_attach(ss, cgrp, tsk, false);
1754                        if (retval) {
1755                                /*
1756                                 * Remember on which subsystem the can_attach()
1757                                 * failed, so that we only call cancel_attach()
1758                                 * against the subsystems whose can_attach()
1759                                 * succeeded. (See below)
1760                                 */
1761                                failed_ss = ss;
1762                                goto out;
1763                        }
1764                }
1765        }
1766
1767        task_lock(tsk);
1768        cg = tsk->cgroups;
1769        get_css_set(cg);
1770        task_unlock(tsk);
1771        /*
1772         * Locate or allocate a new css_set for this task,
1773         * based on its final set of cgroups
1774         */
1775        newcg = find_css_set(cg, cgrp);
1776        put_css_set(cg);
1777        if (!newcg) {
1778                retval = -ENOMEM;
1779                goto out;
1780        }
1781
1782        task_lock(tsk);
1783        if (tsk->flags & PF_EXITING) {
1784                task_unlock(tsk);
1785                put_css_set(newcg);
1786                retval = -ESRCH;
1787                goto out;
1788        }
1789        rcu_assign_pointer(tsk->cgroups, newcg);
1790        task_unlock(tsk);
1791
1792        /* Update the css_set linked lists if we're using them */
1793        write_lock(&css_set_lock);
1794        if (!list_empty(&tsk->cg_list))
1795                list_move(&tsk->cg_list, &newcg->tasks);
1796        write_unlock(&css_set_lock);
1797
1798        for_each_subsys(root, ss) {
1799                if (ss->attach)
1800                        ss->attach(ss, cgrp, oldcgrp, tsk, false);
1801        }
1802        set_bit(CGRP_RELEASABLE, &oldcgrp->flags);
1803        synchronize_rcu();
1804        put_css_set(cg);
1805
1806        /*
1807         * wake up rmdir() waiter. the rmdir should fail since the cgroup
1808         * is no longer empty.
1809         */
1810        cgroup_wakeup_rmdir_waiter(cgrp);
1811out:
1812        if (retval) {
1813                for_each_subsys(root, ss) {
1814                        if (ss == failed_ss)
1815                                /*
1816                                 * This subsystem was the one that failed the
1817                                 * can_attach() check earlier, so we don't need
1818                                 * to call cancel_attach() against it or any
1819                                 * remaining subsystems.
1820                                 */
1821                                break;
1822                        if (ss->cancel_attach)
1823                                ss->cancel_attach(ss, cgrp, tsk, false);
1824                }
1825        }
1826        return retval;
1827}
1828
1829/**
1830 * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from'
1831 * @from: attach to all cgroups of a given task
1832 * @tsk: the task to be attached
1833 */
1834int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk)
1835{
1836        struct cgroupfs_root *root;
1837        int retval = 0;
1838
1839        cgroup_lock();
1840        for_each_active_root(root) {
1841                struct cgroup *from_cg = task_cgroup_from_root(from, root);
1842
1843                retval = cgroup_attach_task(from_cg, tsk);
1844                if (retval)
1845                        break;
1846        }
1847        cgroup_unlock();
1848
1849        return retval;
1850}
1851EXPORT_SYMBOL_GPL(cgroup_attach_task_all);
1852
1853/*
1854 * Attach task with pid 'pid' to cgroup 'cgrp'. Call with cgroup_mutex
1855 * held. May take task_lock of task
1856 */
1857static int attach_task_by_pid(struct cgroup *cgrp, u64 pid)
1858{
1859        struct task_struct *tsk;
1860        const struct cred *cred = current_cred(), *tcred;
1861        int ret;
1862
1863        if (pid) {
1864                rcu_read_lock();
1865                tsk = find_task_by_vpid(pid);
1866                if (!tsk || tsk->flags & PF_EXITING) {
1867                        rcu_read_unlock();
1868                        return -ESRCH;
1869                }
1870
1871                tcred = __task_cred(tsk);
1872                if (cred->euid &&
1873                    cred->euid != tcred->uid &&
1874                    cred->euid != tcred->suid) {
1875                        rcu_read_unlock();
1876                        return -EACCES;
1877                }
1878                get_task_struct(tsk);
1879                rcu_read_unlock();
1880        } else {
1881                tsk = current;
1882                get_task_struct(tsk);
1883        }
1884
1885        ret = cgroup_attach_task(cgrp, tsk);
1886        put_task_struct(tsk);
1887        return ret;
1888}
1889
1890static int cgroup_tasks_write(struct cgroup *cgrp, struct cftype *cft, u64 pid)
1891{
1892        int ret;
1893        if (!cgroup_lock_live_group(cgrp))
1894                return -ENODEV;
1895        ret = attach_task_by_pid(cgrp, pid);
1896        cgroup_unlock();
1897        return ret;
1898}
1899
1900/**
1901 * cgroup_lock_live_group - take cgroup_mutex and check that cgrp is alive.
1902 * @cgrp: the cgroup to be checked for liveness
1903 *
1904 * On success, returns true; the lock should be later released with
1905 * cgroup_unlock(). On failure returns false with no lock held.
1906 */
1907bool cgroup_lock_live_group(struct cgroup *cgrp)
1908{
1909        mutex_lock(&cgroup_mutex);
1910        if (cgroup_is_removed(cgrp)) {
1911                mutex_unlock(&cgroup_mutex);
1912                return false;
1913        }
1914        return true;
1915}
1916EXPORT_SYMBOL_GPL(cgroup_lock_live_group);
1917
1918static int cgroup_release_agent_write(struct cgroup *cgrp, struct cftype *cft,
1919                                      const char *buffer)
1920{
1921        BUILD_BUG_ON(sizeof(cgrp->root->release_agent_path) < PATH_MAX);
1922        if (strlen(buffer) >= PATH_MAX)
1923                return -EINVAL;
1924        if (!cgroup_lock_live_group(cgrp))
1925                return -ENODEV;
1926        strcpy(cgrp->root->release_agent_path, buffer);
1927        cgroup_unlock();
1928        return 0;
1929}
1930
1931static int cgroup_release_agent_show(struct cgroup *cgrp, struct cftype *cft,
1932                                     struct seq_file *seq)
1933{
1934        if (!cgroup_lock_live_group(cgrp))
1935                return -ENODEV;
1936        seq_puts(seq, cgrp->root->release_agent_path);
1937        seq_putc(seq, '\n');
1938        cgroup_unlock();
1939        return 0;
1940}
1941
1942/* A buffer size big enough for numbers or short strings */
1943#define CGROUP_LOCAL_BUFFER_SIZE 64
1944
1945static ssize_t cgroup_write_X64(struct cgroup *cgrp, struct cftype *cft,
1946                                struct file *file,
1947                                const char __user *userbuf,
1948                                size_t nbytes, loff_t *unused_ppos)
1949{
1950        char buffer[CGROUP_LOCAL_BUFFER_SIZE];
1951        int retval = 0;
1952        char *end;
1953
1954        if (!nbytes)
1955                return -EINVAL;
1956        if (nbytes >= sizeof(buffer))
1957                return -E2BIG;
1958        if (copy_from_user(buffer, userbuf, nbytes))
1959                return -EFAULT;
1960
1961        buffer[nbytes] = 0;     /* nul-terminate */
1962        if (cft->write_u64) {
1963                u64 val = simple_strtoull(strstrip(buffer), &end, 0);
1964                if (*end)
1965                        return -EINVAL;
1966                retval = cft->write_u64(cgrp, cft, val);
1967        } else {
1968                s64 val = simple_strtoll(strstrip(buffer), &end, 0);
1969                if (*end)
1970                        return -EINVAL;
1971                retval = cft->write_s64(cgrp, cft, val);
1972        }
1973        if (!retval)
1974                retval = nbytes;
1975        return retval;
1976}
1977
1978static ssize_t cgroup_write_string(struct cgroup *cgrp, struct cftype *cft,
1979                                   struct file *file,
1980                                   const char __user *userbuf,
1981                                   size_t nbytes, loff_t *unused_ppos)
1982{
1983        char local_buffer[CGROUP_LOCAL_BUFFER_SIZE];
1984        int retval = 0;
1985        size_t max_bytes = cft->max_write_len;
1986        char *buffer = local_buffer;
1987
1988        if (!max_bytes)
1989                max_bytes = sizeof(local_buffer) - 1;
1990        if (nbytes >= max_bytes)
1991                return -E2BIG;
1992        /* Allocate a dynamic buffer if we need one */
1993        if (nbytes >= sizeof(local_buffer)) {
1994                buffer = kmalloc(nbytes + 1, GFP_KERNEL);
1995                if (buffer == NULL)
1996                        return -ENOMEM;
1997        }
1998        if (nbytes && copy_from_user(buffer, userbuf, nbytes)) {
1999                retval = -EFAULT;
2000                goto out;
2001        }
2002
2003        buffer[nbytes] = 0;     /* nul-terminate */
2004        retval = cft->write_string(cgrp, cft, strstrip(buffer));
2005        if (!retval)
2006                retval = nbytes;
2007out:
2008        if (buffer != local_buffer)
2009                kfree(buffer);
2010        return retval;
2011}
2012
2013static ssize_t cgroup_file_write(struct file *file, const char __user *buf,
2014                                                size_t nbytes, loff_t *ppos)
2015{
2016        struct cftype *cft = __d_cft(file->f_dentry);
2017        struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
2018
2019        if (cgroup_is_removed(cgrp))
2020                return -ENODEV;
2021        if (cft->write)
2022                return cft->write(cgrp, cft, file, buf, nbytes, ppos);
2023        if (cft->write_u64 || cft->write_s64)
2024                return cgroup_write_X64(cgrp, cft, file, buf, nbytes, ppos);
2025        if (cft->write_string)
2026                return cgroup_write_string(cgrp, cft, file, buf, nbytes, ppos);
2027        if (cft->trigger) {
2028                int ret = cft->trigger(cgrp, (unsigned int)cft->private);
2029                return ret ? ret : nbytes;
2030        }
2031        return -EINVAL;
2032}
2033
2034static ssize_t cgroup_read_u64(struct cgroup *cgrp, struct cftype *cft,
2035                               struct file *file,
2036                               char __user *buf, size_t nbytes,
2037                               loff_t *ppos)
2038{
2039        char tmp[CGROUP_LOCAL_BUFFER_SIZE];
2040        u64 val = cft->read_u64(cgrp, cft);
2041        int len = sprintf(tmp, "%llu\n", (unsigned long long) val);
2042
2043        return simple_read_from_buffer(buf, nbytes, ppos, tmp, len);
2044}
2045
2046static ssize_t cgroup_read_s64(struct cgroup *cgrp, struct cftype *cft,
2047                               struct file *file,
2048                               char __user *buf, size_t nbytes,
2049                               loff_t *ppos)
2050{
2051        char tmp[CGROUP_LOCAL_BUFFER_SIZE];
2052        s64 val = cft->read_s64(cgrp, cft);
2053        int len = sprintf(tmp, "%lld\n", (long long) val);
2054
2055        return simple_read_from_buffer(buf, nbytes, ppos, tmp, len);
2056}
2057
2058static ssize_t cgroup_file_read(struct file *file, char __user *buf,
2059                                   size_t nbytes, loff_t *ppos)
2060{
2061        struct cftype *cft = __d_cft(file->f_dentry);
2062        struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
2063
2064        if (cgroup_is_removed(cgrp))
2065                return -ENODEV;
2066
2067        if (cft->read)
2068                return cft->read(cgrp, cft, file, buf, nbytes, ppos);
2069        if (cft->read_u64)
2070                return cgroup_read_u64(cgrp, cft, file, buf, nbytes, ppos);
2071        if (cft->read_s64)
2072                return cgroup_read_s64(cgrp, cft, file, buf, nbytes, ppos);
2073        return -EINVAL;
2074}
2075
2076/*
2077 * seqfile ops/methods for returning structured data. Currently just
2078 * supports string->u64 maps, but can be extended in future.
2079 */
2080
2081struct cgroup_seqfile_state {
2082        struct cftype *cft;
2083        struct cgroup *cgroup;
2084};
2085
2086static int cgroup_map_add(struct cgroup_map_cb *cb, const char *key, u64 value)
2087{
2088        struct seq_file *sf = cb->state;
2089        return seq_printf(sf, "%s %llu\n", key, (unsigned long long)value);
2090}
2091
2092static int cgroup_seqfile_show(struct seq_file *m, void *arg)
2093{
2094        struct cgroup_seqfile_state *state = m->private;
2095        struct cftype *cft = state->cft;
2096        if (cft->read_map) {
2097                struct cgroup_map_cb cb = {
2098                        .fill = cgroup_map_add,
2099                        .state = m,
2100                };
2101                return cft->read_map(state->cgroup, cft, &cb);
2102        }
2103        return cft->read_seq_string(state->cgroup, cft, m);
2104}
2105
2106static int cgroup_seqfile_release(struct inode *inode, struct file *file)
2107{
2108        struct seq_file *seq = file->private_data;
2109        kfree(seq->private);
2110        return single_release(inode, file);
2111}
2112
2113static const struct file_operations cgroup_seqfile_operations = {
2114        .read = seq_read,
2115        .write = cgroup_file_write,
2116        .llseek = seq_lseek,
2117        .release = cgroup_seqfile_release,
2118};
2119
2120static int cgroup_file_open(struct inode *inode, struct file *file)
2121{
2122        int err;
2123        struct cftype *cft;
2124
2125        err = generic_file_open(inode, file);
2126        if (err)
2127                return err;
2128        cft = __d_cft(file->f_dentry);
2129
2130        if (cft->read_map || cft->read_seq_string) {
2131                struct cgroup_seqfile_state *state =
2132                        kzalloc(sizeof(*state), GFP_USER);
2133                if (!state)
2134                        return -ENOMEM;
2135                state->cft = cft;
2136                state->cgroup = __d_cgrp(file->f_dentry->d_parent);
2137                file->f_op = &cgroup_seqfile_operations;
2138                err = single_open(file, cgroup_seqfile_show, state);
2139                if (err < 0)
2140                        kfree(state);
2141        } else if (cft->open)
2142                err = cft->open(inode, file);
2143        else
2144                err = 0;
2145
2146        return err;
2147}
2148
2149static int cgroup_file_release(struct inode *inode, struct file *file)
2150{
2151        struct cftype *cft = __d_cft(file->f_dentry);
2152        if (cft->release)
2153                return cft->release(inode, file);
2154        return 0;
2155}
2156
2157/*
2158 * cgroup_rename - Only allow simple rename of directories in place.
2159 */
2160static int cgroup_rename(struct inode *old_dir, struct dentry *old_dentry,
2161                            struct inode *new_dir, struct dentry *new_dentry)
2162{
2163        if (!S_ISDIR(old_dentry->d_inode->i_mode))
2164                return -ENOTDIR;
2165        if (new_dentry->d_inode)
2166                return -EEXIST;
2167        if (old_dir != new_dir)
2168                return -EIO;
2169        return simple_rename(old_dir, old_dentry, new_dir, new_dentry);
2170}
2171
2172static const struct file_operations cgroup_file_operations = {
2173        .read = cgroup_file_read,
2174        .write = cgroup_file_write,
2175        .llseek = generic_file_llseek,
2176        .open = cgroup_file_open,
2177        .release = cgroup_file_release,
2178};
2179
2180static const struct inode_operations cgroup_dir_inode_operations = {
2181        .lookup = simple_lookup,
2182        .mkdir = cgroup_mkdir,
2183        .rmdir = cgroup_rmdir,
2184        .rename = cgroup_rename,
2185};
2186
2187/*
2188 * Check if a file is a control file
2189 */
2190static inline struct cftype *__file_cft(struct file *file)
2191{
2192        if (file->f_dentry->d_inode->i_fop != &cgroup_file_operations)
2193                return ERR_PTR(-EINVAL);
2194        return __d_cft(file->f_dentry);
2195}
2196
2197static int cgroup_create_file(struct dentry *dentry, mode_t mode,
2198                                struct super_block *sb)
2199{
2200        static const struct dentry_operations cgroup_dops = {
2201                .d_iput = cgroup_diput,
2202        };
2203
2204        struct inode *inode;
2205
2206        if (!dentry)
2207                return -ENOENT;
2208        if (dentry->d_inode)
2209                return -EEXIST;
2210
2211        inode = cgroup_new_inode(mode, sb);
2212        if (!inode)
2213                return -ENOMEM;
2214
2215        if (S_ISDIR(mode)) {
2216                inode->i_op = &cgroup_dir_inode_operations;
2217                inode->i_fop = &simple_dir_operations;
2218
2219                /* start off with i_nlink == 2 (for "." entry) */
2220                inc_nlink(inode);
2221
2222                /* start with the directory inode held, so that we can
2223                 * populate it without racing with another mkdir */
2224                mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
2225        } else if (S_ISREG(mode)) {
2226                inode->i_size = 0;
2227                inode->i_fop = &cgroup_file_operations;
2228        }
2229        dentry->d_op = &cgroup_dops;
2230        d_instantiate(dentry, inode);
2231        dget(dentry);   /* Extra count - pin the dentry in core */
2232        return 0;
2233}
2234
2235/*
2236 * cgroup_create_dir - create a directory for an object.
2237 * @cgrp: the cgroup we create the directory for. It must have a valid
2238 *        ->parent field. And we are going to fill its ->dentry field.
2239 * @dentry: dentry of the new cgroup
2240 * @mode: mode to set on new directory.
2241 */
2242static int cgroup_create_dir(struct cgroup *cgrp, struct dentry *dentry,
2243                                mode_t mode)
2244{
2245        struct dentry *parent;
2246        int error = 0;
2247
2248        parent = cgrp->parent->dentry;
2249        error = cgroup_create_file(dentry, S_IFDIR | mode, cgrp->root->sb);
2250        if (!error) {
2251                dentry->d_fsdata = cgrp;
2252                inc_nlink(parent->d_inode);
2253                rcu_assign_pointer(cgrp->dentry, dentry);
2254                dget(dentry);
2255        }
2256        dput(dentry);
2257
2258        return error;
2259}
2260
2261/**
2262 * cgroup_file_mode - deduce file mode of a control file
2263 * @cft: the control file in question
2264 *
2265 * returns cft->mode if ->mode is not 0
2266 * returns S_IRUGO|S_IWUSR if it has both a read and a write handler
2267 * returns S_IRUGO if it has only a read handler
2268 * returns S_IWUSR if it has only a write hander
2269 */
2270static mode_t cgroup_file_mode(const struct cftype *cft)
2271{
2272        mode_t mode = 0;
2273
2274        if (cft->mode)
2275                return cft->mode;
2276
2277        if (cft->read || cft->read_u64 || cft->read_s64 ||
2278            cft->read_map || cft->read_seq_string)
2279                mode |= S_IRUGO;
2280
2281        if (cft->write || cft->write_u64 || cft->write_s64 ||
2282            cft->write_string || cft->trigger)
2283                mode |= S_IWUSR;
2284
2285        return mode;
2286}
2287
2288int cgroup_add_file(struct cgroup *cgrp,
2289                       struct cgroup_subsys *subsys,
2290                       const struct cftype *cft)
2291{
2292        struct dentry *dir = cgrp->dentry;
2293        struct dentry *dentry;
2294        int error;
2295        mode_t mode;
2296
2297        char name[MAX_CGROUP_TYPE_NAMELEN + MAX_CFTYPE_NAME + 2] = { 0 };
2298        if (subsys && !test_bit(ROOT_NOPREFIX, &cgrp->root->flags)) {
2299                strcpy(name, subsys->name);
2300                strcat(name, ".");
2301        }
2302        strcat(name, cft->name);
2303        BUG_ON(!mutex_is_locked(&dir->d_inode->i_mutex));
2304        dentry = lookup_one_len(name, dir, strlen(name));
2305        if (!IS_ERR(dentry)) {
2306                mode = cgroup_file_mode(cft);
2307                error = cgroup_create_file(dentry, mode | S_IFREG,
2308                                                cgrp->root->sb);
2309                if (!error)
2310                        dentry->d_fsdata = (void *)cft;
2311                dput(dentry);
2312        } else
2313                error = PTR_ERR(dentry);
2314        return error;
2315}
2316EXPORT_SYMBOL_GPL(cgroup_add_file);
2317
2318int cgroup_add_files(struct cgroup *cgrp,
2319                        struct cgroup_subsys *subsys,
2320                        const struct cftype cft[],
2321                        int count)
2322{
2323        int i, err;
2324        for (i = 0; i < count; i++) {
2325                err = cgroup_add_file(cgrp, subsys, &cft[i]);
2326                if (err)
2327                        return err;
2328        }
2329        return 0;
2330}
2331EXPORT_SYMBOL_GPL(cgroup_add_files);
2332
2333/**
2334 * cgroup_task_count - count the number of tasks in a cgroup.
2335 * @cgrp: the cgroup in question
2336 *
2337 * Return the number of tasks in the cgroup.
2338 */
2339int cgroup_task_count(const struct cgroup *cgrp)
2340{
2341        int count = 0;
2342        struct cg_cgroup_link *link;
2343
2344        read_lock(&css_set_lock);
2345        list_for_each_entry(link, &cgrp->css_sets, cgrp_link_list) {
2346                count += atomic_read(&link->cg->refcount);
2347        }
2348        read_unlock(&css_set_lock);
2349        return count;
2350}
2351
2352/*
2353 * Advance a list_head iterator.  The iterator should be positioned at
2354 * the start of a css_set
2355 */
2356static void cgroup_advance_iter(struct cgroup *cgrp,
2357                                struct cgroup_iter *it)
2358{
2359        struct list_head *l = it->cg_link;
2360        struct cg_cgroup_link *link;
2361        struct css_set *cg;
2362
2363        /* Advance to the next non-empty css_set */
2364        do {
2365                l = l->next;
2366                if (l == &cgrp->css_sets) {
2367                        it->cg_link = NULL;
2368                        return;
2369                }
2370                link = list_entry(l, struct cg_cgroup_link, cgrp_link_list);
2371                cg = link->cg;
2372        } while (list_empty(&cg->tasks));
2373        it->cg_link = l;
2374        it->task = cg->tasks.next;
2375}
2376
2377/*
2378 * To reduce the fork() overhead for systems that are not actually
2379 * using their cgroups capability, we don't maintain the lists running
2380 * through each css_set to its tasks until we see the list actually
2381 * used - in other words after the first call to cgroup_iter_start().
2382 *
2383 * The tasklist_lock is not held here, as do_each_thread() and
2384 * while_each_thread() are protected by RCU.
2385 */
2386static void cgroup_enable_task_cg_lists(void)
2387{
2388        struct task_struct *p, *g;
2389        write_lock(&css_set_lock);
2390        use_task_css_set_links = 1;
2391        do_each_thread(g, p) {
2392                task_lock(p);
2393                /*
2394                 * We should check if the process is exiting, otherwise
2395                 * it will race with cgroup_exit() in that the list
2396                 * entry won't be deleted though the process has exited.
2397                 */
2398                if (!(p->flags & PF_EXITING) && list_empty(&p->cg_list))
2399                        list_add(&p->cg_list, &p->cgroups->tasks);
2400                task_unlock(p);
2401        } while_each_thread(g, p);
2402        write_unlock(&css_set_lock);
2403}
2404
2405void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it)
2406{
2407        /*
2408         * The first time anyone tries to iterate across a cgroup,
2409         * we need to enable the list linking each css_set to its
2410         * tasks, and fix up all existing tasks.
2411         */
2412        if (!use_task_css_set_links)
2413                cgroup_enable_task_cg_lists();
2414
2415        read_lock(&css_set_lock);
2416        it->cg_link = &cgrp->css_sets;
2417        cgroup_advance_iter(cgrp, it);
2418}
2419
2420struct task_struct *cgroup_iter_next(struct cgroup *cgrp,
2421                                        struct cgroup_iter *it)
2422{
2423        struct task_struct *res;
2424        struct list_head *l = it->task;
2425        struct cg_cgroup_link *link;
2426
2427        /* If the iterator cg is NULL, we have no tasks */
2428        if (!it->cg_link)
2429                return NULL;
2430        res = list_entry(l, struct task_struct, cg_list);
2431        /* Advance iterator to find next entry */
2432        l = l->next;
2433        link = list_entry(it->cg_link, struct cg_cgroup_link, cgrp_link_list);
2434        if (l == &link->cg->tasks) {
2435                /* We reached the end of this task list - move on to
2436                 * the next cg_cgroup_link */
2437                cgroup_advance_iter(cgrp, it);
2438        } else {
2439                it->task = l;
2440        }
2441        return res;
2442}
2443
2444void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it)
2445{
2446        read_unlock(&css_set_lock);
2447}
2448
2449static inline int started_after_time(struct task_struct *t1,
2450                                     struct timespec *time,
2451                                     struct task_struct *t2)
2452{
2453        int start_diff = timespec_compare(&t1->start_time, time);
2454        if (start_diff > 0) {
2455                return 1;
2456        } else if (start_diff < 0) {
2457                return 0;
2458        } else {
2459                /*
2460                 * Arbitrarily, if two processes started at the same
2461                 * time, we'll say that the lower pointer value
2462                 * started first. Note that t2 may have exited by now
2463                 * so this may not be a valid pointer any longer, but
2464                 * that's fine - it still serves to distinguish
2465                 * between two tasks started (effectively) simultaneously.
2466                 */
2467                return t1 > t2;
2468        }
2469}
2470
2471/*
2472 * This function is a callback from heap_insert() and is used to order
2473 * the heap.
2474 * In this case we order the heap in descending task start time.
2475 */
2476static inline int started_after(void *p1, void *p2)
2477{
2478        struct task_struct *t1 = p1;
2479        struct task_struct *t2 = p2;
2480        return started_after_time(t1, &t2->start_time, t2);
2481}
2482
2483/**
2484 * cgroup_scan_tasks - iterate though all the tasks in a cgroup
2485 * @scan: struct cgroup_scanner containing arguments for the scan
2486 *
2487 * Arguments include pointers to callback functions test_task() and
2488 * process_task().
2489 * Iterate through all the tasks in a cgroup, calling test_task() for each,
2490 * and if it returns true, call process_task() for it also.
2491 * The test_task pointer may be NULL, meaning always true (select all tasks).
2492 * Effectively duplicates cgroup_iter_{start,next,end}()
2493 * but does not lock css_set_lock for the call to process_task().
2494 * The struct cgroup_scanner may be embedded in any structure of the caller's
2495 * creation.
2496 * It is guaranteed that process_task() will act on every task that
2497 * is a member of the cgroup for the duration of this call. This
2498 * function may or may not call process_task() for tasks that exit
2499 * or move to a different cgroup during the call, or are forked or
2500 * move into the cgroup during the call.
2501 *
2502 * Note that test_task() may be called with locks held, and may in some
2503 * situations be called multiple times for the same task, so it should
2504 * be cheap.
2505 * If the heap pointer in the struct cgroup_scanner is non-NULL, a heap has been
2506 * pre-allocated and will be used for heap operations (and its "gt" member will
2507 * be overwritten), else a temporary heap will be used (allocation of which
2508 * may cause this function to fail).
2509 */
2510int cgroup_scan_tasks(struct cgroup_scanner *scan)
2511{
2512        int retval, i;
2513        struct cgroup_iter it;
2514        struct task_struct *p, *dropped;
2515        /* Never dereference latest_task, since it's not refcounted */
2516        struct task_struct *latest_task = NULL;
2517        struct ptr_heap tmp_heap;
2518        struct ptr_heap *heap;
2519        struct timespec latest_time = { 0, 0 };
2520
2521        if (scan->heap) {
2522                /* The caller supplied our heap and pre-allocated its memory */
2523                heap = scan->heap;
2524                heap->gt = &started_after;
2525        } else {
2526                /* We need to allocate our own heap memory */
2527                heap = &tmp_heap;
2528                retval = heap_init(heap, PAGE_SIZE, GFP_KERNEL, &started_after);
2529                if (retval)
2530                        /* cannot allocate the heap */
2531                        return retval;
2532        }
2533
2534 again:
2535        /*
2536         * Scan tasks in the cgroup, using the scanner's "test_task" callback
2537         * to determine which are of interest, and using the scanner's
2538         * "process_task" callback to process any of them that need an update.
2539         * Since we don't want to hold any locks during the task updates,
2540         * gather tasks to be processed in a heap structure.
2541         * The heap is sorted by descending task start time.
2542         * If the statically-sized heap fills up, we overflow tasks that
2543         * started later, and in future iterations only consider tasks that
2544         * started after the latest task in the previous pass. This
2545         * guarantees forward progress and that we don't miss any tasks.
2546         */
2547        heap->size = 0;
2548        cgroup_iter_start(scan->cg, &it);
2549        while ((p = cgroup_iter_next(scan->cg, &it))) {
2550                /*
2551                 * Only affect tasks that qualify per the caller's callback,
2552                 * if he provided one
2553                 */
2554                if (scan->test_task && !scan->test_task(p, scan))
2555                        continue;
2556                /*
2557                 * Only process tasks that started after the last task
2558                 * we processed
2559                 */
2560                if (!started_after_time(p, &latest_time, latest_task))
2561                        continue;
2562                dropped = heap_insert(heap, p);
2563                if (dropped == NULL) {
2564                        /*
2565                         * The new task was inserted; the heap wasn't
2566                         * previously full
2567                         */
2568                        get_task_struct(p);
2569                } else if (dropped != p) {
2570                        /*
2571                         * The new task was inserted, and pushed out a
2572                         * different task
2573                         */
2574                        get_task_struct(p);
2575                        put_task_struct(dropped);
2576                }
2577                /*
2578                 * Else the new task was newer than anything already in
2579                 * the heap and wasn't inserted
2580                 */
2581        }
2582        cgroup_iter_end(scan->cg, &it);
2583
2584        if (heap->size) {
2585                for (i = 0; i < heap->size; i++) {
2586                        struct task_struct *q = heap->ptrs[i];
2587                        if (i == 0) {
2588                                latest_time = q->start_time;
2589                                latest_task = q;
2590                        }
2591                        /* Process the task per the caller's callback */
2592                        scan->process_task(q, scan);
2593                        put_task_struct(q);
2594                }
2595                /*
2596                 * If we had to process any tasks at all, scan again
2597                 * in case some of them were in the middle of forking
2598                 * children that didn't get processed.
2599                 * Not the most efficient way to do it, but it avoids
2600                 * having to take callback_mutex in the fork path
2601                 */
2602                goto again;
2603        }
2604        if (heap == &tmp_heap)
2605                heap_free(&tmp_heap);
2606        return 0;
2607}
2608
2609/*
2610 * Stuff for reading the 'tasks'/'procs' files.
2611 *
2612 * Reading this file can return large amounts of data if a cgroup has
2613 * *lots* of attached tasks. So it may need several calls to read(),
2614 * but we cannot guarantee that the information we produce is correct
2615 * unless we produce it entirely atomically.
2616 *
2617 */
2618
2619/*
2620 * The following two functions "fix" the issue where there are more pids
2621 * than kmalloc will give memory for; in such cases, we use vmalloc/vfree.
2622 * TODO: replace with a kernel-wide solution to this problem
2623 */
2624#define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2))
2625static void *pidlist_allocate(int count)
2626{
2627        if (PIDLIST_TOO_LARGE(count))
2628                return vmalloc(count * sizeof(pid_t));
2629        else
2630                return kmalloc(count * sizeof(pid_t), GFP_KERNEL);
2631}
2632static void pidlist_free(void *p)
2633{
2634        if (is_vmalloc_addr(p))
2635                vfree(p);
2636        else
2637                kfree(p);
2638}
2639static void *pidlist_resize(void *p, int newcount)
2640{
2641        void *newlist;
2642        /* note: if new alloc fails, old p will still be valid either way */
2643        if (is_vmalloc_addr(p)) {
2644                newlist = vmalloc(newcount * sizeof(pid_t));
2645                if (!newlist)
2646                        return NULL;
2647                memcpy(newlist, p, newcount * sizeof(pid_t));
2648                vfree(p);
2649        } else {
2650                newlist = krealloc(p, newcount * sizeof(pid_t), GFP_KERNEL);
2651        }
2652        return newlist;
2653}
2654
2655/*
2656 * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
2657 * If the new stripped list is sufficiently smaller and there's enough memory
2658 * to allocate a new buffer, will let go of the unneeded memory. Returns the
2659 * number of unique elements.
2660 */
2661/* is the size difference enough that we should re-allocate the array? */
2662#define PIDLIST_REALLOC_DIFFERENCE(old, new) ((old) - PAGE_SIZE >= (new))
2663static int pidlist_uniq(pid_t **p, int length)
2664{
2665        int src, dest = 1;
2666        pid_t *list = *p;
2667        pid_t *newlist;
2668
2669        /*
2670         * we presume the 0th element is unique, so i starts at 1. trivial
2671         * edge cases first; no work needs to be done for either
2672         */
2673        if (length == 0 || length == 1)
2674                return length;
2675        /* src and dest walk down the list; dest counts unique elements */
2676        for (src = 1; src < length; src++) {
2677                /* find next unique element */
2678                while (list[src] == list[src-1]) {
2679                        src++;
2680                        if (src == length)
2681                                goto after;
2682                }
2683                /* dest always points to where the next unique element goes */
2684                list[dest] = list[src];
2685                dest++;
2686        }
2687after:
2688        /*
2689         * if the length difference is large enough, we want to allocate a
2690         * smaller buffer to save memory. if this fails due to out of memory,
2691         * we'll just stay with what we've got.
2692         */
2693        if (PIDLIST_REALLOC_DIFFERENCE(length, dest)) {
2694                newlist = pidlist_resize(list, dest);
2695                if (newlist)
2696                        *p = newlist;
2697        }
2698        return dest;
2699}
2700
2701static int cmppid(const void *a, const void *b)
2702{
2703        return *(pid_t *)a - *(pid_t *)b;
2704}
2705
2706/*
2707 * find the appropriate pidlist for our purpose (given procs vs tasks)
2708 * returns with the lock on that pidlist already held, and takes care
2709 * of the use count, or returns NULL with no locks held if we're out of
2710 * memory.
2711 */
2712static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp,
2713                                                  enum cgroup_filetype type)
2714{
2715        struct cgroup_pidlist *l;
2716        /* don't need task_nsproxy() if we're looking at ourself */
2717        struct pid_namespace *ns = current->nsproxy->pid_ns;
2718
2719        /*
2720         * We can't drop the pidlist_mutex before taking the l->mutex in case
2721         * the last ref-holder is trying to remove l from the list at the same
2722         * time. Holding the pidlist_mutex precludes somebody taking whichever
2723         * list we find out from under us - compare release_pid_array().
2724         */
2725        mutex_lock(&cgrp->pidlist_mutex);
2726        list_for_each_entry(l, &cgrp->pidlists, links) {
2727                if (l->key.type == type && l->key.ns == ns) {
2728                        /* make sure l doesn't vanish out from under us */
2729                        down_write(&l->mutex);
2730                        mutex_unlock(&cgrp->pidlist_mutex);
2731                        return l;
2732                }
2733        }
2734        /* entry not found; create a new one */
2735        l = kmalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL);
2736        if (!l) {
2737                mutex_unlock(&cgrp->pidlist_mutex);
2738                return l;
2739        }
2740        init_rwsem(&l->mutex);
2741        down_write(&l->mutex);
2742        l->key.type = type;
2743        l->key.ns = get_pid_ns(ns);
2744        l->use_count = 0; /* don't increment here */
2745        l->list = NULL;
2746        l->owner = cgrp;
2747        list_add(&l->links, &cgrp->pidlists);
2748        mutex_unlock(&cgrp->pidlist_mutex);
2749        return l;
2750}
2751
2752/*
2753 * Load a cgroup's pidarray with either procs' tgids or tasks' pids
2754 */
2755static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type,
2756                              struct cgroup_pidlist **lp)
2757{
2758        pid_t *array;
2759        int length;
2760        int pid, n = 0; /* used for populating the array */
2761        struct cgroup_iter it;
2762        struct task_struct *tsk;
2763        struct cgroup_pidlist *l;
2764
2765        /*
2766         * If cgroup gets more users after we read count, we won't have
2767         * enough space - tough.  This race is indistinguishable to the
2768         * caller from the case that the additional cgroup users didn't
2769         * show up until sometime later on.
2770         */
2771        length = cgroup_task_count(cgrp);
2772        array = pidlist_allocate(length);
2773        if (!array)
2774                return -ENOMEM;
2775        /* now, populate the array */
2776        cgroup_iter_start(cgrp, &it);
2777        while ((tsk = cgroup_iter_next(cgrp, &it))) {
2778                if (unlikely(n == length))
2779                        break;
2780                /* get tgid or pid for procs or tasks file respectively */
2781                if (type == CGROUP_FILE_PROCS)
2782                        pid = task_tgid_vnr(tsk);
2783                else
2784                        pid = task_pid_vnr(tsk);
2785                if (pid > 0) /* make sure to only use valid results */
2786                        array[n++] = pid;
2787        }
2788        cgroup_iter_end(cgrp, &it);
2789        length = n;
2790        /* now sort & (if procs) strip out duplicates */
2791        sort(array, length, sizeof(pid_t), cmppid, NULL);
2792        if (type == CGROUP_FILE_PROCS)
2793                length = pidlist_uniq(&array, length);
2794        l = cgroup_pidlist_find(cgrp, type);
2795        if (!l) {
2796                pidlist_free(array);
2797                return -ENOMEM;
2798        }
2799        /* store array, freeing old if necessary - lock already held */
2800        pidlist_free(l->list);
2801        l->list = array;
2802        l->length = length;
2803        l->use_count++;
2804        up_write(&l->mutex);
2805        *lp = l;
2806        return 0;
2807}
2808
2809/**
2810 * cgroupstats_build - build and fill cgroupstats
2811 * @stats: cgroupstats to fill information into
2812 * @dentry: A dentry entry belonging to the cgroup for which stats have
2813 * been requested.
2814 *
2815 * Build and fill cgroupstats so that taskstats can export it to user
2816 * space.
2817 */
2818int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry)
2819{
2820        int ret = -EINVAL;
2821        struct cgroup *cgrp;
2822        struct cgroup_iter it;
2823        struct task_struct *tsk;
2824
2825        /*
2826         * Validate dentry by checking the superblock operations,
2827         * and make sure it's a directory.
2828         */
2829        if (dentry->d_sb->s_op != &cgroup_ops ||
2830            !S_ISDIR(dentry->d_inode->i_mode))
2831                 goto err;
2832
2833        ret = 0;
2834        cgrp = dentry->d_fsdata;
2835
2836        cgroup_iter_start(cgrp, &it);
2837        while ((tsk = cgroup_iter_next(cgrp, &it))) {
2838                switch (tsk->state) {
2839                case TASK_RUNNING:
2840                        stats->nr_running++;
2841                        break;
2842                case TASK_INTERRUPTIBLE:
2843                        stats->nr_sleeping++;
2844                        break;
2845                case TASK_UNINTERRUPTIBLE:
2846                        stats->nr_uninterruptible++;
2847                        break;
2848                case TASK_STOPPED:
2849                        stats->nr_stopped++;
2850                        break;
2851                default:
2852                        if (delayacct_is_task_waiting_on_io(tsk))
2853                                stats->nr_io_wait++;
2854                        break;
2855                }
2856        }
2857        cgroup_iter_end(cgrp, &it);
2858
2859err:
2860        return ret;
2861}
2862
2863
2864/*
2865 * seq_file methods for the tasks/procs files. The seq_file position is the
2866 * next pid to display; the seq_file iterator is a pointer to the pid
2867 * in the cgroup->l->list array.
2868 */
2869
2870static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos)
2871{
2872        /*
2873         * Initially we receive a position value that corresponds to
2874         * one more than the last pid shown (or 0 on the first call or
2875         * after a seek to the start). Use a binary-search to find the
2876         * next pid to display, if any
2877         */
2878        struct cgroup_pidlist *l = s->private;
2879        int index = 0, pid = *pos;
2880        int *iter;
2881
2882        down_read(&l->mutex);
2883        if (pid) {
2884                int end = l->length;
2885
2886                while (index < end) {
2887                        int mid = (index + end) / 2;
2888                        if (l->list[mid] == pid) {
2889                                index = mid;
2890                                break;
2891                        } else if (l->list[mid] <= pid)
2892                                index = mid + 1;
2893                        else
2894                                end = mid;
2895                }
2896        }
2897        /* If we're off the end of the array, we're done */
2898        if (index >= l->length)
2899                return NULL;
2900        /* Update the abstract position to be the actual pid that we found */
2901        iter = l->list + index;
2902        *pos = *iter;
2903        return iter;
2904}
2905
2906static void cgroup_pidlist_stop(struct seq_file *s, void *v)
2907{
2908        struct cgroup_pidlist *l = s->private;
2909        up_read(&l->mutex);
2910}
2911
2912static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos)
2913{
2914        struct cgroup_pidlist *l = s->private;
2915        pid_t *p = v;
2916        pid_t *end = l->list + l->length;
2917        /*
2918         * Advance to the next pid in the array. If this goes off the
2919         * end, we're done
2920         */
2921        p++;
2922        if (p >= end) {
2923                return NULL;
2924        } else {
2925                *pos = *p;
2926                return p;
2927        }
2928}
2929
2930static int cgroup_pidlist_show(struct seq_file *s, void *v)
2931{
2932        return seq_printf(s, "%d\n", *(int *)v);
2933}
2934
2935/*
2936 * seq_operations functions for iterating on pidlists through seq_file -
2937 * independent of whether it's tasks or procs
2938 */
2939static const struct seq_operations cgroup_pidlist_seq_operations = {
2940        .start = cgroup_pidlist_start,
2941        .stop = cgroup_pidlist_stop,
2942        .next = cgroup_pidlist_next,
2943        .show = cgroup_pidlist_show,
2944};
2945
2946static void cgroup_release_pid_array(struct cgroup_pidlist *l)
2947{
2948        /*
2949         * the case where we're the last user of this particular pidlist will
2950         * have us remove it from the cgroup's list, which entails taking the
2951         * mutex. since in pidlist_find the pidlist->lock depends on cgroup->
2952         * pidlist_mutex, we have to take pidlist_mutex first.
2953         */
2954        mutex_lock(&l->owner->pidlist_mutex);
2955        down_write(&l->mutex);
2956        BUG_ON(!l->use_count);
2957        if (!--l->use_count) {
2958                /* we're the last user if refcount is 0; remove and free */
2959                list_del(&l->links);
2960                mutex_unlock(&l->owner->pidlist_mutex);
2961                pidlist_free(l->list);
2962                put_pid_ns(l->key.ns);
2963                up_write(&l->mutex);
2964                kfree(l);
2965                return;
2966        }
2967        mutex_unlock(&l->owner->pidlist_mutex);
2968        up_write(&l->mutex);
2969}
2970
2971static int cgroup_pidlist_release(struct inode *inode, struct file *file)
2972{
2973        struct cgroup_pidlist *l;
2974        if (!(file->f_mode & FMODE_READ))
2975                return 0;
2976        /*
2977         * the seq_file will only be initialized if the file was opened for
2978         * reading; hence we check if it's not null only in that case.
2979         */
2980        l = ((struct seq_file *)file->private_data)->private;
2981        cgroup_release_pid_array(l);
2982        return seq_release(inode, file);
2983}
2984
2985static const struct file_operations cgroup_pidlist_operations = {
2986        .read = seq_read,
2987        .llseek = seq_lseek,
2988        .write = cgroup_file_write,
2989        .release = cgroup_pidlist_release,
2990};
2991
2992/*
2993 * The following functions handle opens on a file that displays a pidlist
2994 * (tasks or procs). Prepare an array of the process/thread IDs of whoever's
2995 * in the cgroup.
2996 */
2997/* helper function for the two below it */
2998static int cgroup_pidlist_open(struct file *file, enum cgroup_filetype type)
2999{
3000        struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
3001        struct cgroup_pidlist *l;
3002        int retval;
3003
3004        /* Nothing to do for write-only files */
3005        if (!(file->f_mode & FMODE_READ))
3006                return 0;
3007
3008        /* have the array populated */
3009        retval = pidlist_array_load(cgrp, type, &l);
3010        if (retval)
3011                return retval;
3012        /* configure file information */
3013        file->f_op = &cgroup_pidlist_operations;
3014
3015        retval = seq_open(file, &cgroup_pidlist_seq_operations);
3016        if (retval) {
3017                cgroup_release_pid_array(l);
3018                return retval;
3019        }
3020        ((struct seq_file *)file->private_data)->private = l;
3021        return 0;
3022}
3023static int cgroup_tasks_open(struct inode *unused, struct file *file)
3024{
3025        return cgroup_pidlist_open(file, CGROUP_FILE_TASKS);
3026}
3027static int cgroup_procs_open(struct inode *unused, struct file *file)
3028{
3029        return cgroup_pidlist_open(file, CGROUP_FILE_PROCS);
3030}
3031
3032static u64 cgroup_read_notify_on_release(struct cgroup *cgrp,
3033                                            struct cftype *cft)
3034{
3035        return notify_on_release(cgrp);
3036}
3037
3038static int cgroup_write_notify_on_release(struct cgroup *cgrp,
3039                                          struct cftype *cft,
3040                                          u64 val)
3041{
3042        clear_bit(CGRP_RELEASABLE, &cgrp->flags);
3043        if (val)
3044                set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
3045        else
3046                clear_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
3047        return 0;
3048}
3049
3050/*
3051 * Unregister event and free resources.
3052 *
3053 * Gets called from workqueue.
3054 */
3055static void cgroup_event_remove(struct work_struct *work)
3056{
3057        struct cgroup_event *event = container_of(work, struct cgroup_event,
3058                        remove);
3059        struct cgroup *cgrp = event->cgrp;
3060
3061        event->cft->unregister_event(cgrp, event->cft, event->eventfd);
3062
3063        eventfd_ctx_put(event->eventfd);
3064        kfree(event);
3065        dput(cgrp->dentry);
3066}
3067
3068/*
3069 * Gets called on POLLHUP on eventfd when user closes it.
3070 *
3071 * Called with wqh->lock held and interrupts disabled.
3072 */
3073static int cgroup_event_wake(wait_queue_t *wait, unsigned mode,
3074                int sync, void *key)
3075{
3076        struct cgroup_event *event = container_of(wait,
3077                        struct cgroup_event, wait);
3078        struct cgroup *cgrp = event->cgrp;
3079        unsigned long flags = (unsigned long)key;
3080
3081        if (flags & POLLHUP) {
3082                __remove_wait_queue(event->wqh, &event->wait);
3083                spin_lock(&cgrp->event_list_lock);
3084                list_del(&event->list);
3085                spin_unlock(&cgrp->event_list_lock);
3086                /*
3087                 * We are in atomic context, but cgroup_event_remove() may
3088                 * sleep, so we have to call it in workqueue.
3089                 */
3090                schedule_work(&event->remove);
3091        }
3092
3093        return 0;
3094}
3095
3096static void cgroup_event_ptable_queue_proc(struct file *file,
3097                wait_queue_head_t *wqh, poll_table *pt)
3098{
3099        struct cgroup_event *event = container_of(pt,
3100                        struct cgroup_event, pt);
3101
3102        event->wqh = wqh;
3103        add_wait_queue(wqh, &event->wait);
3104}
3105
3106/*
3107 * Parse input and register new cgroup event handler.
3108 *
3109 * Input must be in format '<event_fd> <control_fd> <args>'.
3110 * Interpretation of args is defined by control file implementation.
3111 */
3112static int cgroup_write_event_control(struct cgroup *cgrp, struct cftype *cft,
3113                                      const char *buffer)
3114{
3115        struct cgroup_event *event = NULL;
3116        unsigned int efd, cfd;
3117        struct file *efile = NULL;
3118        struct file *cfile = NULL;
3119        char *endp;
3120        int ret;
3121
3122        efd = simple_strtoul(buffer, &endp, 10);
3123        if (*endp != ' ')
3124                return -EINVAL;
3125        buffer = endp + 1;
3126
3127        cfd = simple_strtoul(buffer, &endp, 10);
3128        if ((*endp != ' ') && (*endp != '\0'))
3129                return -EINVAL;
3130        buffer = endp + 1;
3131
3132        event = kzalloc(sizeof(*event), GFP_KERNEL);
3133        if (!event)
3134                return -ENOMEM;
3135        event->cgrp = cgrp;
3136        INIT_LIST_HEAD(&event->list);
3137        init_poll_funcptr(&event->pt, cgroup_event_ptable_queue_proc);
3138        init_waitqueue_func_entry(&event->wait, cgroup_event_wake);
3139        INIT_WORK(&event->remove, cgroup_event_remove);
3140
3141        efile = eventfd_fget(efd);
3142        if (IS_ERR(efile)) {
3143                ret = PTR_ERR(efile);
3144                goto fail;
3145        }
3146
3147        event->eventfd = eventfd_ctx_fileget(efile);
3148        if (IS_ERR(event->eventfd)) {
3149                ret = PTR_ERR(event->eventfd);
3150                goto fail;
3151        }
3152
3153        cfile = fget(cfd);
3154        if (!cfile) {
3155                ret = -EBADF;
3156                goto fail;
3157        }
3158
3159        /* the process need read permission on control file */
3160        ret = file_permission(cfile, MAY_READ);
3161        if (ret < 0)
3162                goto fail;
3163
3164        event->cft = __file_cft(cfile);
3165        if (IS_ERR(event->cft)) {
3166                ret = PTR_ERR(event->cft);
3167                goto fail;
3168        }
3169
3170        if (!event->cft->register_event || !event->cft->unregister_event) {
3171                ret = -EINVAL;
3172                goto fail;
3173        }
3174
3175        ret = event->cft->register_event(cgrp, event->cft,
3176                        event->eventfd, buffer);
3177        if (ret)
3178                goto fail;
3179
3180        if (efile->f_op->poll(efile, &event->pt) & POLLHUP) {
3181                event->cft->unregister_event(cgrp, event->cft, event->eventfd);
3182                ret = 0;
3183                goto fail;
3184        }
3185
3186        /*
3187         * Events should be removed after rmdir of cgroup directory, but before
3188         * destroying subsystem state objects. Let's take reference to cgroup
3189         * directory dentry to do that.
3190         */
3191        dget(cgrp->dentry);
3192
3193        spin_lock(&cgrp->event_list_lock);
3194        list_add(&event->list, &cgrp->event_list);
3195        spin_unlock(&cgrp->event_list_lock);
3196
3197        fput(cfile);
3198        fput(efile);
3199
3200        return 0;
3201
3202fail:
3203        if (cfile)
3204                fput(cfile);
3205
3206        if (event && event->eventfd && !IS_ERR(event->eventfd))
3207                eventfd_ctx_put(event->eventfd);
3208
3209        if (!IS_ERR_OR_NULL(efile))
3210                fput(efile);
3211
3212        kfree(event);
3213
3214        return ret;
3215}
3216
3217static u64 cgroup_clone_children_read(struct cgroup *cgrp,
3218                                    struct cftype *cft)
3219{
3220        return clone_children(cgrp);
3221}
3222
3223static int cgroup_clone_children_write(struct cgroup *cgrp,
3224                                     struct cftype *cft,
3225                                     u64 val)
3226{
3227        if (val)
3228                set_bit(CGRP_CLONE_CHILDREN, &cgrp->flags);
3229        else
3230                clear_bit(CGRP_CLONE_CHILDREN, &cgrp->flags);
3231        return 0;
3232}
3233
3234/*
3235 * for the common functions, 'private' gives the type of file
3236 */
3237/* for hysterical raisins, we can't put this on the older files */
3238#define CGROUP_FILE_GENERIC_PREFIX "cgroup."
3239static struct cftype files[] = {
3240        {
3241                .name = "tasks",
3242                .open = cgroup_tasks_open,
3243                .write_u64 = cgroup_tasks_write,
3244                .release = cgroup_pidlist_release,
3245                .mode = S_IRUGO | S_IWUSR,
3246        },
3247        {
3248                .name = CGROUP_FILE_GENERIC_PREFIX "procs",
3249                .open = cgroup_procs_open,
3250                /* .write_u64 = cgroup_procs_write, TODO */
3251                .release = cgroup_pidlist_release,
3252                .mode = S_IRUGO,
3253        },
3254        {
3255                .name = "notify_on_release",
3256                .read_u64 = cgroup_read_notify_on_release,
3257                .write_u64 = cgroup_write_notify_on_release,
3258        },
3259        {
3260                .name = CGROUP_FILE_GENERIC_PREFIX "event_control",
3261                .write_string = cgroup_write_event_control,
3262                .mode = S_IWUGO,
3263        },
3264        {
3265                .name = "cgroup.clone_children",
3266                .read_u64 = cgroup_clone_children_read,
3267                .write_u64 = cgroup_clone_children_write,
3268        },
3269};
3270
3271static struct cftype cft_release_agent = {
3272        .name = "release_agent",
3273        .read_seq_string = cgroup_release_agent_show,
3274        .write_string = cgroup_release_agent_write,
3275        .max_write_len = PATH_MAX,
3276};
3277
3278static int cgroup_populate_dir(struct cgroup *cgrp)
3279{
3280        int err;
3281        struct cgroup_subsys *ss;
3282
3283        /* First clear out any existing files */
3284        cgroup_clear_directory(cgrp->dentry);
3285
3286        err = cgroup_add_files(cgrp, NULL, files, ARRAY_SIZE(files));
3287        if (err < 0)
3288                return err;
3289
3290        if (cgrp == cgrp->top_cgroup) {
3291                if ((err = cgroup_add_file(cgrp, NULL, &cft_release_agent)) < 0)
3292                        return err;
3293        }
3294
3295        for_each_subsys(cgrp->root, ss) {
3296                if (ss->populate && (err = ss->populate(ss, cgrp)) < 0)
3297                        return err;
3298        }
3299        /* This cgroup is ready now */
3300        for_each_subsys(cgrp->root, ss) {
3301                struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
3302                /*
3303                 * Update id->css pointer and make this css visible from
3304                 * CSS ID functions. This pointer will be dereferened
3305                 * from RCU-read-side without locks.
3306                 */
3307                if (css->id)
3308                        rcu_assign_pointer(css->id->css, css);
3309        }
3310
3311        return 0;
3312}
3313
3314static void init_cgroup_css(struct cgroup_subsys_state *css,
3315                               struct cgroup_subsys *ss,
3316                               struct cgroup *cgrp)
3317{
3318        css->cgroup = cgrp;
3319        atomic_set(&css->refcnt, 1);
3320        css->flags = 0;
3321        css->id = NULL;
3322        if (cgrp == dummytop)
3323                set_bit(CSS_ROOT, &css->flags);
3324        BUG_ON(cgrp->subsys[ss->subsys_id]);
3325        cgrp->subsys[ss->subsys_id] = css;
3326}
3327
3328static void cgroup_lock_hierarchy(struct cgroupfs_root *root)
3329{
3330        /* We need to take each hierarchy_mutex in a consistent order */
3331        int i;
3332
3333        /*
3334         * No worry about a race with rebind_subsystems that might mess up the
3335         * locking order, since both parties are under cgroup_mutex.
3336         */
3337        for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
3338                struct cgroup_subsys *ss = subsys[i];
3339                if (ss == NULL)
3340                        continue;
3341                if (ss->root == root)
3342                        mutex_lock(&ss->hierarchy_mutex);
3343        }
3344}
3345
3346static void cgroup_unlock_hierarchy(struct cgroupfs_root *root)
3347{
3348        int i;
3349
3350        for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
3351                struct cgroup_subsys *ss = subsys[i];
3352                if (ss == NULL)
3353                        continue;
3354                if (ss->root == root)
3355                        mutex_unlock(&ss->hierarchy_mutex);
3356        }
3357}
3358
3359/*
3360 * cgroup_create - create a cgroup
3361 * @parent: cgroup that will be parent of the new cgroup
3362 * @dentry: dentry of the new cgroup
3363 * @mode: mode to set on new inode
3364 *
3365 * Must be called with the mutex on the parent inode held
3366 */
3367static long cgroup_create(struct cgroup *parent, struct dentry *dentry,
3368                             mode_t mode)
3369{
3370        struct cgroup *cgrp;
3371        struct cgroupfs_root *root = parent->root;
3372        int err = 0;
3373        struct cgroup_subsys *ss;
3374        struct super_block *sb = root->sb;
3375
3376        cgrp = kzalloc(sizeof(*cgrp), GFP_KERNEL);
3377        if (!cgrp)
3378                return -ENOMEM;
3379
3380        /* Grab a reference on the superblock so the hierarchy doesn't
3381         * get deleted on unmount if there are child cgroups.  This
3382         * can be done outside cgroup_mutex, since the sb can't
3383         * disappear while someone has an open control file on the
3384         * fs */
3385        atomic_inc(&sb->s_active);
3386
3387        mutex_lock(&cgroup_mutex);
3388
3389        init_cgroup_housekeeping(cgrp);
3390
3391        cgrp->parent = parent;
3392        cgrp->root = parent->root;
3393        cgrp->top_cgroup = parent->top_cgroup;
3394
3395        if (notify_on_release(parent))
3396                set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
3397
3398        if (clone_children(parent))
3399                set_bit(CGRP_CLONE_CHILDREN, &cgrp->flags);
3400
3401        for_each_subsys(root, ss) {
3402                struct cgroup_subsys_state *css = ss->create(ss, cgrp);
3403
3404                if (IS_ERR(css)) {
3405                        err = PTR_ERR(css);
3406                        goto err_destroy;
3407                }
3408                init_cgroup_css(css, ss, cgrp);
3409                if (ss->use_id) {
3410                        err = alloc_css_id(ss, parent, cgrp);
3411                        if (err)
3412                                goto err_destroy;
3413                }
3414                /* At error, ->destroy() callback has to free assigned ID. */
3415                if (clone_children(parent) && ss->post_clone)
3416                        ss->post_clone(ss, cgrp);
3417        }
3418
3419        cgroup_lock_hierarchy(root);
3420        list_add(&cgrp->sibling, &cgrp->parent->children);
3421        cgroup_unlock_hierarchy(root);
3422        root->number_of_cgroups++;
3423
3424        err = cgroup_create_dir(cgrp, dentry, mode);
3425        if (err < 0)
3426                goto err_remove;
3427
3428        /* The cgroup directory was pre-locked for us */
3429        BUG_ON(!mutex_is_locked(&cgrp->dentry->d_inode->i_mutex));
3430
3431        err = cgroup_populate_dir(cgrp);
3432        /* If err < 0, we have a half-filled directory - oh well ;) */
3433
3434        mutex_unlock(&cgroup_mutex);
3435        mutex_unlock(&cgrp->dentry->d_inode->i_mutex);
3436
3437        return 0;
3438
3439 err_remove:
3440
3441        cgroup_lock_hierarchy(root);
3442        list_del(&cgrp->sibling);
3443        cgroup_unlock_hierarchy(root);
3444        root->number_of_cgroups--;
3445
3446 err_destroy:
3447
3448        for_each_subsys(root, ss) {
3449                if (cgrp->subsys[ss->subsys_id])
3450                        ss->destroy(ss, cgrp);
3451        }
3452
3453        mutex_unlock(&cgroup_mutex);
3454
3455        /* Release the reference count that we took on the superblock */
3456        deactivate_super(sb);
3457
3458        kfree(cgrp);
3459        return err;
3460}
3461
3462static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, int mode)
3463{
3464        struct cgroup *c_parent = dentry->d_parent->d_fsdata;
3465
3466        /* the vfs holds inode->i_mutex already */
3467        return cgroup_create(c_parent, dentry, mode | S_IFDIR);
3468}
3469
3470static int cgroup_has_css_refs(struct cgroup *cgrp)
3471{
3472        /* Check the reference count on each subsystem. Since we
3473         * already established that there are no tasks in the
3474         * cgroup, if the css refcount is also 1, then there should
3475         * be no outstanding references, so the subsystem is safe to
3476         * destroy. We scan across all subsystems rather than using
3477         * the per-hierarchy linked list of mounted subsystems since
3478         * we can be called via check_for_release() with no
3479         * synchronization other than RCU, and the subsystem linked
3480         * list isn't RCU-safe */
3481        int i;
3482        /*
3483         * We won't need to lock the subsys array, because the subsystems
3484         * we're concerned about aren't going anywhere since our cgroup root
3485         * has a reference on them.
3486         */
3487        for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
3488                struct cgroup_subsys *ss = subsys[i];
3489                struct cgroup_subsys_state *css;
3490                /* Skip subsystems not present or not in this hierarchy */
3491                if (ss == NULL || ss->root != cgrp->root)
3492                        continue;
3493                css = cgrp->subsys[ss->subsys_id];
3494                /* When called from check_for_release() it's possible
3495                 * that by this point the cgroup has been removed
3496                 * and the css deleted. But a false-positive doesn't
3497                 * matter, since it can only happen if the cgroup
3498                 * has been deleted and hence no longer needs the
3499                 * release agent to be called anyway. */
3500                if (css && (atomic_read(&css->refcnt) > 1))
3501                        return 1;
3502        }
3503        return 0;
3504}
3505
3506/*
3507 * Atomically mark all (or else none) of the cgroup's CSS objects as
3508 * CSS_REMOVED. Return true on success, or false if the cgroup has
3509 * busy subsystems. Call with cgroup_mutex held
3510 */
3511
3512static int cgroup_clear_css_refs(struct cgroup *cgrp)
3513{
3514        struct cgroup_subsys *ss;
3515        unsigned long flags;
3516        bool failed = false;
3517        local_irq_save(flags);
3518        for_each_subsys(cgrp->root, ss) {
3519                struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
3520                int refcnt;
3521                while (1) {
3522                        /* We can only remove a CSS with a refcnt==1 */
3523                        refcnt = atomic_read(&css->refcnt);
3524                        if (refcnt > 1) {
3525                                failed = true;
3526                                goto done;
3527                        }
3528                        BUG_ON(!refcnt);
3529                        /*
3530                         * Drop the refcnt to 0 while we check other
3531                         * subsystems. This will cause any racing
3532                         * css_tryget() to spin until we set the
3533                         * CSS_REMOVED bits or abort
3534                         */
3535                        if (atomic_cmpxchg(&css->refcnt, refcnt, 0) == refcnt)
3536                                break;
3537                        cpu_relax();
3538                }
3539        }
3540 done:
3541        for_each_subsys(cgrp->root, ss) {
3542                struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
3543                if (failed) {
3544                        /*
3545                         * Restore old refcnt if we previously managed
3546                         * to clear it from 1 to 0
3547                         */
3548                        if (!atomic_read(&css->refcnt))
3549                                atomic_set(&css->refcnt, 1);
3550                } else {
3551                        /* Commit the fact that the CSS is removed */
3552                        set_bit(CSS_REMOVED, &css->flags);
3553                }
3554        }
3555        local_irq_restore(flags);
3556        return !failed;
3557}
3558
3559static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry)
3560{
3561        struct cgroup *cgrp = dentry->d_fsdata;
3562        struct dentry *d;
3563        struct cgroup *parent;
3564        DEFINE_WAIT(wait);
3565        struct cgroup_event *event, *tmp;
3566        int ret;
3567
3568        /* the vfs holds both inode->i_mutex already */
3569again:
3570        mutex_lock(&cgroup_mutex);
3571        if (atomic_read(&cgrp->count) != 0) {
3572                mutex_unlock(&cgroup_mutex);
3573                return -EBUSY;
3574        }
3575        if (!list_empty(&cgrp->children)) {
3576                mutex_unlock(&cgroup_mutex);
3577                return -EBUSY;
3578        }
3579        mutex_unlock(&cgroup_mutex);
3580
3581        /*
3582         * In general, subsystem has no css->refcnt after pre_destroy(). But
3583         * in racy cases, subsystem may have to get css->refcnt after
3584         * pre_destroy() and it makes rmdir return with -EBUSY. This sometimes
3585         * make rmdir return -EBUSY too often. To avoid that, we use waitqueue
3586         * for cgroup's rmdir. CGRP_WAIT_ON_RMDIR is for synchronizing rmdir
3587         * and subsystem's reference count handling. Please see css_get/put
3588         * and css_tryget() and cgroup_wakeup_rmdir_waiter() implementation.
3589         */
3590        set_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
3591
3592        /*
3593         * Call pre_destroy handlers of subsys. Notify subsystems
3594         * that rmdir() request comes.
3595         */
3596        ret = cgroup_call_pre_destroy(cgrp);
3597        if (ret) {
3598                clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
3599                return ret;
3600        }
3601
3602        mutex_lock(&cgroup_mutex);
3603        parent = cgrp->parent;
3604        if (atomic_read(&cgrp->count) || !list_empty(&cgrp->children)) {
3605                clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
3606                mutex_unlock(&cgroup_mutex);
3607                return -EBUSY;
3608        }
3609        prepare_to_wait(&cgroup_rmdir_waitq, &wait, TASK_INTERRUPTIBLE);
3610        if (!cgroup_clear_css_refs(cgrp)) {
3611                mutex_unlock(&cgroup_mutex);
3612                /*
3613                 * Because someone may call cgroup_wakeup_rmdir_waiter() before
3614                 * prepare_to_wait(), we need to check this flag.
3615                 */
3616                if (test_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags))
3617                        schedule();
3618                finish_wait(&cgroup_rmdir_waitq, &wait);
3619                clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
3620                if (signal_pending(current))
3621                        return -EINTR;
3622                goto again;
3623        }
3624        /* NO css_tryget() can success after here. */
3625        finish_wait(&cgroup_rmdir_waitq, &wait);
3626        clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
3627
3628        spin_lock(&release_list_lock);
3629        set_bit(CGRP_REMOVED, &cgrp->flags);
3630        if (!list_empty(&cgrp->release_list))
3631                list_del_init(&cgrp->release_list);
3632        spin_unlock(&release_list_lock);
3633
3634        cgroup_lock_hierarchy(cgrp->root);
3635        /* delete this cgroup from parent->children */
3636        list_del_init(&cgrp->sibling);
3637        cgroup_unlock_hierarchy(cgrp->root);
3638
3639        spin_lock(&cgrp->dentry->d_lock);
3640        d = dget(cgrp->dentry);
3641        spin_unlock(&d->d_lock);
3642
3643        cgroup_d_remove_dir(d);
3644        dput(d);
3645
3646        set_bit(CGRP_RELEASABLE, &parent->flags);
3647        check_for_release(parent);
3648
3649        /*
3650         * Unregister events and notify userspace.
3651         * Notify userspace about cgroup removing only after rmdir of cgroup
3652         * directory to avoid race between userspace and kernelspace
3653         */
3654        spin_lock(&cgrp->event_list_lock);
3655        list_for_each_entry_safe(event, tmp, &cgrp->event_list, list) {
3656                list_del(&event->list);
3657                remove_wait_queue(event->wqh, &event->wait);
3658                eventfd_signal(event->eventfd, 1);
3659                schedule_work(&event->remove);
3660        }
3661        spin_unlock(&cgrp->event_list_lock);
3662
3663        mutex_unlock(&cgroup_mutex);
3664        return 0;
3665}
3666
3667static void __init cgroup_init_subsys(struct cgroup_subsys *ss)
3668{
3669        struct cgroup_subsys_state *css;
3670
3671        printk(KERN_INFO "Initializing cgroup subsys %s\n", ss->name);
3672
3673        /* Create the top cgroup state for this subsystem */
3674        list_add(&ss->sibling, &rootnode.subsys_list);
3675        ss->root = &rootnode;
3676        css = ss->create(ss, dummytop);
3677        /* We don't handle early failures gracefully */
3678        BUG_ON(IS_ERR(css));
3679        init_cgroup_css(css, ss, dummytop);
3680
3681        /* Update the init_css_set to contain a subsys
3682         * pointer to this state - since the subsystem is
3683         * newly registered, all tasks and hence the
3684         * init_css_set is in the subsystem's top cgroup. */
3685        init_css_set.subsys[ss->subsys_id] = dummytop->subsys[ss->subsys_id];
3686
3687        need_forkexit_callback |= ss->fork || ss->exit;
3688
3689        /* At system boot, before all subsystems have been
3690         * registered, no tasks have been forked, so we don't
3691         * need to invoke fork callbacks here. */
3692        BUG_ON(!list_empty(&init_task.tasks));
3693
3694        mutex_init(&ss->hierarchy_mutex);
3695        lockdep_set_class(&ss->hierarchy_mutex, &ss->subsys_key);
3696        ss->active = 1;
3697
3698        /* this function shouldn't be used with modular subsystems, since they
3699         * need to register a subsys_id, among other things */
3700        BUG_ON(ss->module);
3701}
3702
3703/**
3704 * cgroup_load_subsys: load and register a modular subsystem at runtime
3705 * @ss: the subsystem to load
3706 *
3707 * This function should be called in a modular subsystem's initcall. If the
3708 * subsystem is built as a module, it will be assigned a new subsys_id and set
3709 * up for use. If the subsystem is built-in anyway, work is delegated to the
3710 * simpler cgroup_init_subsys.
3711 */
3712int __init_or_module cgroup_load_subsys(struct cgroup_subsys *ss)
3713{
3714        int i;
3715        struct cgroup_subsys_state *css;
3716
3717        /* check name and function validity */
3718        if (ss->name == NULL || strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN ||
3719            ss->create == NULL || ss->destroy == NULL)
3720                return -EINVAL;
3721
3722        /*
3723         * we don't support callbacks in modular subsystems. this check is
3724         * before the ss->module check for consistency; a subsystem that could
3725         * be a module should still have no callbacks even if the user isn't
3726         * compiling it as one.
3727         */
3728        if (ss->fork || ss->exit)
3729                return -EINVAL;
3730
3731        /*
3732         * an optionally modular subsystem is built-in: we want to do nothing,
3733         * since cgroup_init_subsys will have already taken care of it.
3734         */
3735        if (ss->module == NULL) {
3736                /* a few sanity checks */
3737                BUG_ON(ss->subsys_id >= CGROUP_BUILTIN_SUBSYS_COUNT);
3738                BUG_ON(subsys[ss->subsys_id] != ss);
3739                return 0;
3740        }
3741
3742        /*
3743         * need to register a subsys id before anything else - for example,
3744         * init_cgroup_css needs it.
3745         */
3746        mutex_lock(&cgroup_mutex);
3747        /* find the first empty slot in the array */
3748        for (i = CGROUP_BUILTIN_SUBSYS_COUNT; i < CGROUP_SUBSYS_COUNT; i++) {
3749                if (subsys[i] == NULL)
3750                        break;
3751        }
3752        if (i == CGROUP_SUBSYS_COUNT) {
3753                /* maximum number of subsystems already registered! */
3754                mutex_unlock(&cgroup_mutex);
3755                return -EBUSY;
3756        }
3757        /* assign ourselves the subsys_id */
3758        ss->subsys_id = i;
3759        subsys[i] = ss;
3760
3761        /*
3762         * no ss->create seems to need anything important in the ss struct, so
3763         * this can happen first (i.e. before the rootnode attachment).
3764         */
3765        css = ss->create(ss, dummytop);
3766        if (IS_ERR(css)) {
3767                /* failure case - need to deassign the subsys[] slot. */
3768                subsys[i] = NULL;
3769                mutex_unlock(&cgroup_mutex);
3770                return PTR_ERR(css);
3771        }
3772
3773        list_add(&ss->sibling, &rootnode.subsys_list);
3774        ss->root = &rootnode;
3775
3776        /* our new subsystem will be attached to the dummy hierarchy. */
3777        init_cgroup_css(css, ss, dummytop);
3778        /* init_idr must be after init_cgroup_css because it sets css->id. */
3779        if (ss->use_id) {
3780                int ret = cgroup_init_idr(ss, css);
3781                if (ret) {
3782                        dummytop->subsys[ss->subsys_id] = NULL;
3783                        ss->destroy(ss, dummytop);
3784                        subsys[i] = NULL;
3785                        mutex_unlock(&cgroup_mutex);
3786                        return ret;
3787                }
3788        }
3789
3790        /*
3791         * Now we need to entangle the css into the existing css_sets. unlike
3792         * in cgroup_init_subsys, there are now multiple css_sets, so each one
3793         * will need a new pointer to it; done by iterating the css_set_table.
3794         * furthermore, modifying the existing css_sets will corrupt the hash
3795         * table state, so each changed css_set will need its hash recomputed.
3796         * this is all done under the css_set_lock.
3797         */
3798        write_lock(&css_set_lock);
3799        for (i = 0; i < CSS_SET_TABLE_SIZE; i++) {
3800                struct css_set *cg;
3801                struct hlist_node *node, *tmp;
3802                struct hlist_head *bucket = &css_set_table[i], *new_bucket;
3803
3804                hlist_for_each_entry_safe(cg, node, tmp, bucket, hlist) {
3805                        /* skip entries that we already rehashed */
3806                        if (cg->subsys[ss->subsys_id])
3807                                continue;
3808                        /* remove existing entry */
3809                        hlist_del(&cg->hlist);
3810                        /* set new value */
3811                        cg->subsys[ss->subsys_id] = css;
3812                        /* recompute hash and restore entry */
3813                        new_bucket = css_set_hash(cg->subsys);
3814                        hlist_add_head(&cg->hlist, new_bucket);
3815                }
3816        }
3817        write_unlock(&css_set_lock);
3818
3819        mutex_init(&ss->hierarchy_mutex);
3820        lockdep_set_class(&ss->hierarchy_mutex, &ss->subsys_key);
3821        ss->active = 1;
3822
3823        /* success! */
3824        mutex_unlock(&cgroup_mutex);
3825        return 0;
3826}
3827EXPORT_SYMBOL_GPL(cgroup_load_subsys);
3828
3829/**
3830 * cgroup_unload_subsys: unload a modular subsystem
3831 * @ss: the subsystem to unload
3832 *
3833 * This function should be called in a modular subsystem's exitcall. When this
3834 * function is invoked, the refcount on the subsystem's module will be 0, so
3835 * the subsystem will not be attached to any hierarchy.
3836 */
3837void cgroup_unload_subsys(struct cgroup_subsys *ss)
3838{
3839        struct cg_cgroup_link *link;
3840        struct hlist_head *hhead;
3841
3842        BUG_ON(ss->module == NULL);
3843
3844        /*
3845         * we shouldn't be called if the subsystem is in use, and the use of
3846         * try_module_get in parse_cgroupfs_options should ensure that it
3847         * doesn't start being used while we're killing it off.
3848         */
3849        BUG_ON(ss->root != &rootnode);
3850
3851        mutex_lock(&cgroup_mutex);
3852        /* deassign the subsys_id */
3853        BUG_ON(ss->subsys_id < CGROUP_BUILTIN_SUBSYS_COUNT);
3854        subsys[ss->subsys_id] = NULL;
3855
3856        /* remove subsystem from rootnode's list of subsystems */
3857        list_del_init(&ss->sibling);
3858
3859        /*
3860         * disentangle the css from all css_sets attached to the dummytop. as
3861         * in loading, we need to pay our respects to the hashtable gods.
3862         */
3863        write_lock(&css_set_lock);
3864        list_for_each_entry(link, &dummytop->css_sets, cgrp_link_list) {
3865                struct css_set *cg = link->cg;
3866
3867                hlist_del(&cg->hlist);
3868                BUG_ON(!cg->subsys[ss->subsys_id]);
3869                cg->subsys[ss->subsys_id] = NULL;
3870                hhead = css_set_hash(cg->subsys);
3871                hlist_add_head(&cg->hlist, hhead);
3872        }
3873        write_unlock(&css_set_lock);
3874
3875        /*
3876         * remove subsystem's css from the dummytop and free it - need to free
3877         * before marking as null because ss->destroy needs the cgrp->subsys
3878         * pointer to find their state. note that this also takes care of
3879         * freeing the css_id.
3880         */
3881        ss->destroy(ss, dummytop);
3882        dummytop->subsys[ss->subsys_id] = NULL;
3883
3884        mutex_unlock(&cgroup_mutex);
3885}
3886EXPORT_SYMBOL_GPL(cgroup_unload_subsys);
3887
3888/**
3889 * cgroup_init_early - cgroup initialization at system boot
3890 *
3891 * Initialize cgroups at system boot, and initialize any
3892 * subsystems that request early init.
3893 */
3894int __init cgroup_init_early(void)
3895{
3896        int i;
3897        atomic_set(&init_css_set.refcount, 1);
3898        INIT_LIST_HEAD(&init_css_set.cg_links);
3899        INIT_LIST_HEAD(&init_css_set.tasks);
3900        INIT_HLIST_NODE(&init_css_set.hlist);
3901        css_set_count = 1;
3902        init_cgroup_root(&rootnode);
3903        root_count = 1;
3904        init_task.cgroups = &init_css_set;
3905
3906        init_css_set_link.cg = &init_css_set;
3907        init_css_set_link.cgrp = dummytop;
3908        list_add(&init_css_set_link.cgrp_link_list,
3909                 &rootnode.top_cgroup.css_sets);
3910        list_add(&init_css_set_link.cg_link_list,
3911                 &init_css_set.cg_links);
3912
3913        for (i = 0; i < CSS_SET_TABLE_SIZE; i++)
3914                INIT_HLIST_HEAD(&css_set_table[i]);
3915
3916        /* at bootup time, we don't worry about modular subsystems */
3917        for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
3918                struct cgroup_subsys *ss = subsys[i];
3919
3920                BUG_ON(!ss->name);
3921                BUG_ON(strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN);
3922                BUG_ON(!ss->create);
3923                BUG_ON(!ss->destroy);
3924                if (ss->subsys_id != i) {
3925                        printk(KERN_ERR "cgroup: Subsys %s id == %d\n",
3926                               ss->name, ss->subsys_id);
3927                        BUG();
3928                }
3929
3930                if (ss->early_init)
3931                        cgroup_init_subsys(ss);
3932        }
3933        return 0;
3934}
3935
3936/**
3937 * cgroup_init - cgroup initialization
3938 *
3939 * Register cgroup filesystem and /proc file, and initialize
3940 * any subsystems that didn't request early init.
3941 */
3942int __init cgroup_init(void)
3943{
3944        int err;
3945        int i;
3946        struct hlist_head *hhead;
3947
3948        err = bdi_init(&cgroup_backing_dev_info);
3949        if (err)
3950                return err;
3951
3952        /* at bootup time, we don't worry about modular subsystems */
3953        for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
3954                struct cgroup_subsys *ss = subsys[i];
3955                if (!ss->early_init)
3956                        cgroup_init_subsys(ss);
3957                if (ss->use_id)
3958                        cgroup_init_idr(ss, init_css_set.subsys[ss->subsys_id]);
3959        }
3960
3961        /* Add init_css_set to the hash table */
3962        hhead = css_set_hash(init_css_set.subsys);
3963        hlist_add_head(&init_css_set.hlist, hhead);
3964        BUG_ON(!init_root_id(&rootnode));
3965
3966        cgroup_kobj = kobject_create_and_add("cgroup", fs_kobj);
3967        if (!cgroup_kobj) {
3968                err = -ENOMEM;
3969                goto out;
3970        }
3971
3972        err = register_filesystem(&cgroup_fs_type);
3973        if (err < 0) {
3974                kobject_put(cgroup_kobj);
3975                goto out;
3976        }
3977
3978        proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations);
3979
3980out:
3981        if (err)
3982                bdi_destroy(&cgroup_backing_dev_info);
3983
3984        return err;
3985}
3986
3987/*
3988 * proc_cgroup_show()
3989 *  - Print task's cgroup paths into seq_file, one line for each hierarchy
3990 *  - Used for /proc/<pid>/cgroup.
3991 *  - No need to task_lock(tsk) on this tsk->cgroup reference, as it
3992 *    doesn't really matter if tsk->cgroup changes after we read it,
3993 *    and we take cgroup_mutex, keeping cgroup_attach_task() from changing it
3994 *    anyway.  No need to check that tsk->cgroup != NULL, thanks to
3995 *    the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks
3996 *    cgroup to top_cgroup.
3997 */
3998
3999/* TODO: Use a proper seq_file iterator */
4000static int proc_cgroup_show(struct seq_file *m, void *v)
4001{
4002        struct pid *pid;
4003        struct task_struct *tsk;
4004        char *buf;
4005        int retval;
4006        struct cgroupfs_root *root;
4007
4008        retval = -ENOMEM;
4009        buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
4010        if (!buf)
4011                goto out;
4012
4013        retval = -ESRCH;
4014        pid = m->private;
4015        tsk = get_pid_task(pid, PIDTYPE_PID);
4016        if (!tsk)
4017                goto out_free;
4018
4019        retval = 0;
4020
4021        mutex_lock(&cgroup_mutex);
4022
4023        for_each_active_root(root) {
4024                struct cgroup_subsys *ss;
4025                struct cgroup *cgrp;
4026                int count = 0;
4027
4028                seq_printf(m, "%d:", root->hierarchy_id);
4029                for_each_subsys(root, ss)
4030                        seq_printf(m, "%s%s", count++ ? "," : "", ss->name);
4031                if (strlen(root->name))
4032                        seq_printf(m, "%sname=%s", count ? "," : "",
4033                                   root->name);
4034                seq_putc(m, ':');
4035                cgrp = task_cgroup_from_root(tsk, root);
4036                retval = cgroup_path(cgrp, buf, PAGE_SIZE);
4037                if (retval < 0)
4038                        goto out_unlock;
4039                seq_puts(m, buf);
4040                seq_putc(m, '\n');
4041        }
4042
4043out_unlock:
4044        mutex_unlock(&cgroup_mutex);
4045        put_task_struct(tsk);
4046out_free:
4047        kfree(buf);
4048out:
4049        return retval;
4050}
4051
4052static int cgroup_open(struct inode *inode, struct file *file)
4053{
4054        struct pid *pid = PROC_I(inode)->pid;
4055        return single_open(file, proc_cgroup_show, pid);
4056}
4057
4058const struct file_operations proc_cgroup_operations = {
4059        .open           = cgroup_open,
4060        .read           = seq_read,
4061        .llseek         = seq_lseek,
4062        .release        = single_release,
4063};
4064
4065/* Display information about each subsystem and each hierarchy */
4066static int proc_cgroupstats_show(struct seq_file *m, void *v)
4067{
4068        int i;
4069
4070        seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
4071        /*
4072         * ideally we don't want subsystems moving around while we do this.
4073         * cgroup_mutex is also necessary to guarantee an atomic snapshot of
4074         * subsys/hierarchy state.
4075         */
4076        mutex_lock(&cgroup_mutex);
4077        for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
4078                struct cgroup_subsys *ss = subsys[i];
4079                if (ss == NULL)
4080                        continue;
4081                seq_printf(m, "%s\t%d\t%d\t%d\n",
4082                           ss->name, ss->root->hierarchy_id,
4083                           ss->root->number_of_cgroups, !ss->disabled);
4084        }
4085        mutex_unlock(&cgroup_mutex);
4086        return 0;
4087}
4088
4089static int cgroupstats_open(struct inode *inode, struct file *file)
4090{
4091        return single_open(file, proc_cgroupstats_show, NULL);
4092}
4093
4094static const struct file_operations proc_cgroupstats_operations = {
4095        .open = cgroupstats_open,
4096        .read = seq_read,
4097        .llseek = seq_lseek,
4098        .release = single_release,
4099};
4100
4101/**
4102 * cgroup_fork - attach newly forked task to its parents cgroup.
4103 * @child: pointer to task_struct of forking parent process.
4104 *
4105 * Description: A task inherits its parent's cgroup at fork().
4106 *
4107 * A pointer to the shared css_set was automatically copied in
4108 * fork.c by dup_task_struct().  However, we ignore that copy, since
4109 * it was not made under the protection of RCU or cgroup_mutex, so
4110 * might no longer be a valid cgroup pointer.  cgroup_attach_task() might
4111 * have already changed current->cgroups, allowing the previously
4112 * referenced cgroup group to be removed and freed.
4113 *
4114 * At the point that cgroup_fork() is called, 'current' is the parent
4115 * task, and the passed argument 'child' points to the child task.
4116 */
4117void cgroup_fork(struct task_struct *child)
4118{
4119        task_lock(current);
4120        child->cgroups = current->cgroups;
4121        get_css_set(child->cgroups);
4122        task_unlock(current);
4123        INIT_LIST_HEAD(&child->cg_list);
4124}
4125
4126/**
4127 * cgroup_fork_callbacks - run fork callbacks
4128 * @child: the new task
4129 *
4130 * Called on a new task very soon before adding it to the
4131 * tasklist. No need to take any locks since no-one can
4132 * be operating on this task.
4133 */
4134void cgroup_fork_callbacks(struct task_struct *child)
4135{
4136        if (need_forkexit_callback) {
4137                int i;
4138                /*
4139                 * forkexit callbacks are only supported for builtin
4140                 * subsystems, and the builtin section of the subsys array is
4141                 * immutable, so we don't need to lock the subsys array here.
4142                 */
4143                for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
4144                        struct cgroup_subsys *ss = subsys[i];
4145                        if (ss->fork)
4146                                ss->fork(ss, child);
4147                }
4148        }
4149}
4150
4151/**
4152 * cgroup_post_fork - called on a new task after adding it to the task list
4153 * @child: the task in question
4154 *
4155 * Adds the task to the list running through its css_set if necessary.
4156 * Has to be after the task is visible on the task list in case we race
4157 * with the first call to cgroup_iter_start() - to guarantee that the
4158 * new task ends up on its list.
4159 */
4160void cgroup_post_fork(struct task_struct *child)
4161{
4162        if (use_task_css_set_links) {
4163                write_lock(&css_set_lock);
4164                task_lock(child);
4165                if (list_empty(&child->cg_list))
4166                        list_add(&child->cg_list, &child->cgroups->tasks);
4167                task_unlock(child);
4168                write_unlock(&css_set_lock);
4169        }
4170}
4171/**
4172 * cgroup_exit - detach cgroup from exiting task
4173 * @tsk: pointer to task_struct of exiting process
4174 * @run_callback: run exit callbacks?
4175 *
4176 * Description: Detach cgroup from @tsk and release it.
4177 *
4178 * Note that cgroups marked notify_on_release force every task in
4179 * them to take the global cgroup_mutex mutex when exiting.
4180 * This could impact scaling on very large systems.  Be reluctant to
4181 * use notify_on_release cgroups where very high task exit scaling
4182 * is required on large systems.
4183 *
4184 * the_top_cgroup_hack:
4185 *
4186 *    Set the exiting tasks cgroup to the root cgroup (top_cgroup).
4187 *
4188 *    We call cgroup_exit() while the task is still competent to
4189 *    handle notify_on_release(), then leave the task attached to the
4190 *    root cgroup in each hierarchy for the remainder of its exit.
4191 *
4192 *    To do this properly, we would increment the reference count on
4193 *    top_cgroup, and near the very end of the kernel/exit.c do_exit()
4194 *    code we would add a second cgroup function call, to drop that
4195 *    reference.  This would just create an unnecessary hot spot on
4196 *    the top_cgroup reference count, to no avail.
4197 *
4198 *    Normally, holding a reference to a cgroup without bumping its
4199 *    count is unsafe.   The cgroup could go away, or someone could
4200 *    attach us to a different cgroup, decrementing the count on
4201 *    the first cgroup that we never incremented.  But in this case,
4202 *    top_cgroup isn't going away, and either task has PF_EXITING set,
4203 *    which wards off any cgroup_attach_task() attempts, or task is a failed
4204 *    fork, never visible to cgroup_attach_task.
4205 */
4206void cgroup_exit(struct task_struct *tsk, int run_callbacks)
4207{
4208        int i;
4209        struct css_set *cg;
4210
4211        if (run_callbacks && need_forkexit_callback) {
4212                /*
4213                 * modular subsystems can't use callbacks, so no need to lock
4214                 * the subsys array
4215                 */
4216                for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
4217                        struct cgroup_subsys *ss = subsys[i];
4218                        if (ss->exit)
4219                                ss->exit(ss, tsk);
4220                }
4221        }
4222
4223        /*
4224         * Unlink from the css_set task list if necessary.
4225         * Optimistically check cg_list before taking
4226         * css_set_lock
4227         */
4228        if (!list_empty(&tsk->cg_list)) {
4229                write_lock(&css_set_lock);
4230                if (!list_empty(&tsk->cg_list))
4231                        list_del_init(&tsk->cg_list);
4232                write_unlock(&css_set_lock);
4233        }
4234
4235        /* Reassign the task to the init_css_set. */
4236        task_lock(tsk);
4237        cg = tsk->cgroups;
4238        tsk->cgroups = &init_css_set;
4239        task_unlock(tsk);
4240        if (cg)
4241                put_css_set_taskexit(cg);
4242}
4243
4244/**
4245 * cgroup_clone - clone the cgroup the given subsystem is attached to
4246 * @tsk: the task to be moved
4247 * @subsys: the given subsystem
4248 * @nodename: the name for the new cgroup
4249 *
4250 * Duplicate the current cgroup in the hierarchy that the given
4251 * subsystem is attached to, and move this task into the new
4252 * child.
4253 */
4254int cgroup_clone(struct task_struct *tsk, struct cgroup_subsys *subsys,
4255                                                        char *nodename)
4256{
4257        struct dentry *dentry;
4258        int ret = 0;
4259        struct cgroup *parent, *child;
4260        struct inode *inode;
4261        struct css_set *cg;
4262        struct cgroupfs_root *root;
4263        struct cgroup_subsys *ss;
4264
4265        /* We shouldn't be called by an unregistered subsystem */
4266        BUG_ON(!subsys->active);
4267
4268        /* First figure out what hierarchy and cgroup we're dealing
4269         * with, and pin them so we can drop cgroup_mutex */
4270        mutex_lock(&cgroup_mutex);
4271 again:
4272        root = subsys->root;
4273        if (root == &rootnode) {
4274                mutex_unlock(&cgroup_mutex);
4275                return 0;
4276        }
4277
4278        /* Pin the hierarchy */
4279        if (!atomic_inc_not_zero(&root->sb->s_active)) {
4280                /* We race with the final deactivate_super() */
4281                mutex_unlock(&cgroup_mutex);
4282                return 0;
4283        }
4284
4285        /* Keep the cgroup alive */
4286        task_lock(tsk);
4287        parent = task_cgroup(tsk, subsys->subsys_id);
4288        cg = tsk->cgroups;
4289        get_css_set(cg);
4290        task_unlock(tsk);
4291
4292        mutex_unlock(&cgroup_mutex);
4293
4294        /* Now do the VFS work to create a cgroup */
4295        inode = parent->dentry->d_inode;
4296
4297        /* Hold the parent directory mutex across this operation to
4298         * stop anyone else deleting the new cgroup */
4299        mutex_lock(&inode->i_mutex);
4300        dentry = lookup_one_len(nodename, parent->dentry, strlen(nodename));
4301        if (IS_ERR(dentry)) {
4302                printk(KERN_INFO
4303                       "cgroup: Couldn't allocate dentry for %s: %ld\n", nodename,
4304                       PTR_ERR(dentry));
4305                ret = PTR_ERR(dentry);
4306                goto out_release;
4307        }
4308
4309        /* Create the cgroup directory, which also creates the cgroup */
4310        ret = vfs_mkdir(inode, dentry, 0755);
4311        child = __d_cgrp(dentry);
4312        dput(dentry);
4313        if (ret) {
4314                printk(KERN_INFO
4315                       "Failed to create cgroup %s: %d\n", nodename,
4316                       ret);
4317                goto out_release;
4318        }
4319
4320        /* The cgroup now exists. Retake cgroup_mutex and check
4321         * that we're still in the same state that we thought we
4322         * were. */
4323        mutex_lock(&cgroup_mutex);
4324        if ((root != subsys->root) ||
4325            (parent != task_cgroup(tsk, subsys->subsys_id))) {
4326                /* Aargh, we raced ... */
4327                mutex_unlock(&inode->i_mutex);
4328                put_css_set(cg);
4329
4330                deactivate_super(root->sb);
4331                /* The cgroup is still accessible in the VFS, but
4332                 * we're not going to try to rmdir() it at this
4333                 * point. */
4334                printk(KERN_INFO
4335                       "Race in cgroup_clone() - leaking cgroup %s\n",
4336                       nodename);
4337                goto again;
4338        }
4339
4340        /* do any required auto-setup */
4341        for_each_subsys(root, ss) {
4342                if (ss->post_clone)
4343                        ss->post_clone(ss, child);
4344        }
4345
4346        /* All seems fine. Finish by moving the task into the new cgroup */
4347        ret = cgroup_attach_task(child, tsk);
4348        mutex_unlock(&cgroup_mutex);
4349
4350 out_release:
4351        mutex_unlock(&inode->i_mutex);
4352
4353        mutex_lock(&cgroup_mutex);
4354        put_css_set(cg);
4355        mutex_unlock(&cgroup_mutex);
4356        deactivate_super(root->sb);
4357        return ret;
4358}
4359
4360/**
4361 * cgroup_is_descendant - see if @cgrp is a descendant of @task's cgrp
4362 * @cgrp: the cgroup in question
4363 * @task: the task in question
4364 *
4365 * See if @cgrp is a descendant of @task's cgroup in the appropriate
4366 * hierarchy.
4367 *
4368 * If we are sending in dummytop, then presumably we are creating
4369 * the top cgroup in the subsystem.
4370 *
4371 * Called only by the ns (nsproxy) cgroup.
4372 */
4373int cgroup_is_descendant(const struct cgroup *cgrp, struct task_struct *task)
4374{
4375        int ret;
4376        struct cgroup *target;
4377
4378        if (cgrp == dummytop)
4379                return 1;
4380
4381        target = task_cgroup_from_root(task, cgrp->root);
4382        while (cgrp != target && cgrp!= cgrp->top_cgroup)
4383                cgrp = cgrp->parent;
4384        ret = (cgrp == target);
4385        return ret;
4386}
4387
4388static void check_for_release(struct cgroup *cgrp)
4389{
4390        /* All of these checks rely on RCU to keep the cgroup
4391         * structure alive */
4392        if (cgroup_is_releasable(cgrp) && !atomic_read(&cgrp->count)
4393            && list_empty(&cgrp->children) && !cgroup_has_css_refs(cgrp)) {
4394                /* Control Group is currently removeable. If it's not
4395                 * already queued for a userspace notification, queue
4396                 * it now */
4397                int need_schedule_work = 0;
4398                spin_lock(&release_list_lock);
4399                if (!cgroup_is_removed(cgrp) &&
4400                    list_empty(&cgrp->release_list)) {
4401                        list_add(&cgrp->release_list, &release_list);
4402                        need_schedule_work = 1;
4403                }
4404                spin_unlock(&release_list_lock);
4405                if (need_schedule_work)
4406                        schedule_work(&release_agent_work);
4407        }
4408}
4409
4410/* Caller must verify that the css is not for root cgroup */
4411void __css_put(struct cgroup_subsys_state *css, int count)
4412{
4413        struct cgroup *cgrp = css->cgroup;
4414        int val;
4415        rcu_read_lock();
4416        val = atomic_sub_return(count, &css->refcnt);
4417        if (val == 1) {
4418                if (notify_on_release(cgrp)) {
4419                        set_bit(CGRP_RELEASABLE, &cgrp->flags);
4420                        check_for_release(cgrp);
4421                }
4422                cgroup_wakeup_rmdir_waiter(cgrp);
4423        }
4424        rcu_read_unlock();
4425        WARN_ON_ONCE(val < 1);
4426}
4427EXPORT_SYMBOL_GPL(__css_put);
4428
4429/*
4430 * Notify userspace when a cgroup is released, by running the
4431 * configured release agent with the name of the cgroup (path
4432 * relative to the root of cgroup file system) as the argument.
4433 *
4434 * Most likely, this user command will try to rmdir this cgroup.
4435 *
4436 * This races with the possibility that some other task will be
4437 * attached to this cgroup before it is removed, or that some other
4438 * user task will 'mkdir' a child cgroup of this cgroup.  That's ok.
4439 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
4440 * unused, and this cgroup will be reprieved from its death sentence,
4441 * to continue to serve a useful existence.  Next time it's released,
4442 * we will get notified again, if it still has 'notify_on_release' set.
4443 *
4444 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
4445 * means only wait until the task is successfully execve()'d.  The
4446 * separate release agent task is forked by call_usermodehelper(),
4447 * then control in this thread returns here, without waiting for the
4448 * release agent task.  We don't bother to wait because the caller of
4449 * this routine has no use for the exit status of the release agent
4450 * task, so no sense holding our caller up for that.
4451 */
4452static void cgroup_release_agent(struct work_struct *work)
4453{
4454        BUG_ON(work != &release_agent_work);
4455        mutex_lock(&cgroup_mutex);
4456        spin_lock(&release_list_lock);
4457        while (!list_empty(&release_list)) {
4458                char *argv[3], *envp[3];
4459                int i;
4460                char *pathbuf = NULL, *agentbuf = NULL;
4461                struct cgroup *cgrp = list_entry(release_list.next,
4462                                                    struct cgroup,
4463                                                    release_list);
4464                list_del_init(&cgrp->release_list);
4465                spin_unlock(&release_list_lock);
4466                pathbuf = kmalloc(PAGE_SIZE, GFP_KERNEL);
4467                if (!pathbuf)
4468                        goto continue_free;
4469                if (cgroup_path(cgrp, pathbuf, PAGE_SIZE) < 0)
4470                        goto continue_free;
4471                agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL);
4472                if (!agentbuf)
4473                        goto continue_free;
4474
4475                i = 0;
4476                argv[i++] = agentbuf;
4477                argv[i++] = pathbuf;
4478                argv[i] = NULL;
4479
4480                i = 0;
4481                /* minimal command environment */
4482                envp[i++] = "HOME=/";
4483                envp[i++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
4484                envp[i] = NULL;
4485
4486                /* Drop the lock while we invoke the usermode helper,
4487                 * since the exec could involve hitting disk and hence
4488                 * be a slow process */
4489                mutex_unlock(&cgroup_mutex);
4490                call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC);
4491                mutex_lock(&cgroup_mutex);
4492 continue_free:
4493                kfree(pathbuf);
4494                kfree(agentbuf);
4495                spin_lock(&release_list_lock);
4496        }
4497        spin_unlock(&release_list_lock);
4498        mutex_unlock(&cgroup_mutex);
4499}
4500
4501static int __init cgroup_disable(char *str)
4502{
4503        int i;
4504        char *token;
4505
4506        while ((token = strsep(&str, ",")) != NULL) {
4507                if (!*token)
4508                        continue;
4509                /*
4510                 * cgroup_disable, being at boot time, can't know about module
4511                 * subsystems, so we don't worry about them.
4512                 */
4513                for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
4514                        struct cgroup_subsys *ss = subsys[i];
4515
4516                        if (!strcmp(token, ss->name)) {
4517                                ss->disabled = 1;
4518                                printk(KERN_INFO "Disabling %s control group"
4519                                        " subsystem\n", ss->name);
4520                                break;
4521                        }
4522                }
4523        }
4524        return 1;
4525}
4526__setup("cgroup_disable=", cgroup_disable);
4527
4528/*
4529 * Functons for CSS ID.
4530 */
4531
4532/*
4533 *To get ID other than 0, this should be called when !cgroup_is_removed().
4534 */
4535unsigned short css_id(struct cgroup_subsys_state *css)
4536{
4537        struct css_id *cssid;
4538
4539        /*
4540         * This css_id() can return correct value when somone has refcnt
4541         * on this or this is under rcu_read_lock(). Once css->id is allocated,
4542         * it's unchanged until freed.
4543         */
4544        cssid = rcu_dereference_check(css->id,
4545                        rcu_read_lock_held() || atomic_read(&css->refcnt));
4546
4547        if (cssid)
4548                return cssid->id;
4549        return 0;
4550}
4551EXPORT_SYMBOL_GPL(css_id);
4552
4553unsigned short css_depth(struct cgroup_subsys_state *css)
4554{
4555        struct css_id *cssid;
4556
4557        cssid = rcu_dereference_check(css->id,
4558                        rcu_read_lock_held() || atomic_read(&css->refcnt));
4559
4560        if (cssid)
4561                return cssid->depth;
4562        return 0;
4563}
4564EXPORT_SYMBOL_GPL(css_depth);
4565
4566/**
4567 *  css_is_ancestor - test "root" css is an ancestor of "child"
4568 * @child: the css to be tested.
4569 * @root: the css supporsed to be an ancestor of the child.
4570 *
4571 * Returns true if "root" is an ancestor of "child" in its hierarchy. Because
4572 * this function reads css->id, this use rcu_dereference() and rcu_read_lock().
4573 * But, considering usual usage, the csses should be valid objects after test.
4574 * Assuming that the caller will do some action to the child if this returns
4575 * returns true, the caller must take "child";s reference count.
4576 * If "child" is valid object and this returns true, "root" is valid, too.
4577 */
4578
4579bool css_is_ancestor(struct cgroup_subsys_state *child,
4580                    const struct cgroup_subsys_state *root)
4581{
4582        struct css_id *child_id;
4583        struct css_id *root_id;
4584        bool ret = true;
4585
4586        rcu_read_lock();
4587        child_id  = rcu_dereference(child->id);
4588        root_id = rcu_dereference(root->id);
4589        if (!child_id
4590            || !root_id
4591            || (child_id->depth < root_id->depth)
4592            || (child_id->stack[root_id->depth] != root_id->id))
4593                ret = false;
4594        rcu_read_unlock();
4595        return ret;
4596}
4597
4598static void __free_css_id_cb(struct rcu_head *head)
4599{
4600        struct css_id *id;
4601
4602        id = container_of(head, struct css_id, rcu_head);
4603        kfree(id);
4604}
4605
4606void free_css_id(struct cgroup_subsys *ss, struct cgroup_subsys_state *css)
4607{
4608        struct css_id *id = css->id;
4609        /* When this is called before css_id initialization, id can be NULL */
4610        if (!id)
4611                return;
4612
4613        BUG_ON(!ss->use_id);
4614
4615        rcu_assign_pointer(id->css, NULL);
4616        rcu_assign_pointer(css->id, NULL);
4617        spin_lock(&ss->id_lock);
4618        idr_remove(&ss->idr, id->id);
4619        spin_unlock(&ss->id_lock);
4620        call_rcu(&id->rcu_head, __free_css_id_cb);
4621}
4622EXPORT_SYMBOL_GPL(free_css_id);
4623
4624/*
4625 * This is called by init or create(). Then, calls to this function are
4626 * always serialized (By cgroup_mutex() at create()).
4627 */
4628
4629static struct css_id *get_new_cssid(struct cgroup_subsys *ss, int depth)
4630{
4631        struct css_id *newid;
4632        int myid, error, size;
4633
4634        BUG_ON(!ss->use_id);
4635
4636        size = sizeof(*newid) + sizeof(unsigned short) * (depth + 1);
4637        newid = kzalloc(size, GFP_KERNEL);
4638        if (!newid)
4639                return ERR_PTR(-ENOMEM);
4640        /* get id */
4641        if (unlikely(!idr_pre_get(&ss->idr, GFP_KERNEL))) {
4642                error = -ENOMEM;
4643                goto err_out;
4644        }
4645        spin_lock(&ss->id_lock);
4646        /* Don't use 0. allocates an ID of 1-65535 */
4647        error = idr_get_new_above(&ss->idr, newid, 1, &myid);
4648        spin_unlock(&ss->id_lock);
4649
4650        /* Returns error when there are no free spaces for new ID.*/
4651        if (error) {
4652                error = -ENOSPC;
4653                goto err_out;
4654        }
4655        if (myid > CSS_ID_MAX)
4656                goto remove_idr;
4657
4658        newid->id = myid;
4659        newid->depth = depth;
4660        return newid;
4661remove_idr:
4662        error = -ENOSPC;
4663        spin_lock(&ss->id_lock);
4664        idr_remove(&ss->idr, myid);
4665        spin_unlock(&ss->id_lock);
4666err_out:
4667        kfree(newid);
4668        return ERR_PTR(error);
4669
4670}
4671
4672static int __init_or_module cgroup_init_idr(struct cgroup_subsys *ss,
4673                                            struct cgroup_subsys_state *rootcss)
4674{
4675        struct css_id *newid;
4676
4677        spin_lock_init(&ss->id_lock);
4678        idr_init(&ss->idr);
4679
4680        newid = get_new_cssid(ss, 0);
4681        if (IS_ERR(newid))
4682                return PTR_ERR(newid);
4683
4684        newid->stack[0] = newid->id;
4685        newid->css = rootcss;
4686        rootcss->id = newid;
4687        return 0;
4688}
4689
4690static int alloc_css_id(struct cgroup_subsys *ss, struct cgroup *parent,
4691                        struct cgroup *child)
4692{
4693        int subsys_id, i, depth = 0;
4694        struct cgroup_subsys_state *parent_css, *child_css;
4695        struct css_id *child_id, *parent_id;
4696
4697        subsys_id = ss->subsys_id;
4698        parent_css = parent->subsys[subsys_id];
4699        child_css = child->subsys[subsys_id];
4700        parent_id = parent_css->id;
4701        depth = parent_id->depth + 1;
4702
4703        child_id = get_new_cssid(ss, depth);
4704        if (IS_ERR(child_id))
4705                return PTR_ERR(child_id);
4706
4707        for (i = 0; i < depth; i++)
4708                child_id->stack[i] = parent_id->stack[i];
4709        child_id->stack[depth] = child_id->id;
4710        /*
4711         * child_id->css pointer will be set after this cgroup is available
4712         * see cgroup_populate_dir()
4713         */
4714        rcu_assign_pointer(child_css->id, child_id);
4715
4716        return 0;
4717}
4718
4719/**
4720 * css_lookup - lookup css by id
4721 * @ss: cgroup subsys to be looked into.
4722 * @id: the id
4723 *
4724 * Returns pointer to cgroup_subsys_state if there is valid one with id.
4725 * NULL if not. Should be called under rcu_read_lock()
4726 */
4727struct cgroup_subsys_state *css_lookup(struct cgroup_subsys *ss, int id)
4728{
4729        struct css_id *cssid = NULL;
4730
4731        BUG_ON(!ss->use_id);
4732        cssid = idr_find(&ss->idr, id);
4733
4734        if (unlikely(!cssid))
4735                return NULL;
4736
4737        return rcu_dereference(cssid->css);
4738}
4739EXPORT_SYMBOL_GPL(css_lookup);
4740
4741/**
4742 * css_get_next - lookup next cgroup under specified hierarchy.
4743 * @ss: pointer to subsystem
4744 * @id: current position of iteration.
4745 * @root: pointer to css. search tree under this.
4746 * @foundid: position of found object.
4747 *
4748 * Search next css under the specified hierarchy of rootid. Calling under
4749 * rcu_read_lock() is necessary. Returns NULL if it reaches the end.
4750 */
4751struct cgroup_subsys_state *
4752css_get_next(struct cgroup_subsys *ss, int id,
4753             struct cgroup_subsys_state *root, int *foundid)
4754{
4755        struct cgroup_subsys_state *ret = NULL;
4756        struct css_id *tmp;
4757        int tmpid;
4758        int rootid = css_id(root);
4759        int depth = css_depth(root);
4760
4761        if (!rootid)
4762                return NULL;
4763
4764        BUG_ON(!ss->use_id);
4765        /* fill start point for scan */
4766        tmpid = id;
4767        while (1) {
4768                /*
4769                 * scan next entry from bitmap(tree), tmpid is updated after
4770                 * idr_get_next().
4771                 */
4772                spin_lock(&ss->id_lock);
4773                tmp = idr_get_next(&ss->idr, &tmpid);
4774                spin_unlock(&ss->id_lock);
4775
4776                if (!tmp)
4777                        break;
4778                if (tmp->depth >= depth && tmp->stack[depth] == rootid) {
4779                        ret = rcu_dereference(tmp->css);
4780                        if (ret) {
4781                                *foundid = tmpid;
4782                                break;
4783                        }
4784                }
4785                /* continue to scan from next id */
4786                tmpid = tmpid + 1;
4787        }
4788        return ret;
4789}
4790
4791#ifdef CONFIG_CGROUP_DEBUG
4792static struct cgroup_subsys_state *debug_create(struct cgroup_subsys *ss,
4793                                                   struct cgroup *cont)
4794{
4795        struct cgroup_subsys_state *css = kzalloc(sizeof(*css), GFP_KERNEL);
4796
4797        if (!css)
4798                return ERR_PTR(-ENOMEM);
4799
4800        return css;
4801}
4802
4803static void debug_destroy(struct cgroup_subsys *ss, struct cgroup *cont)
4804{
4805        kfree(cont->subsys[debug_subsys_id]);
4806}
4807
4808static u64 cgroup_refcount_read(struct cgroup *cont, struct cftype *cft)
4809{
4810        return atomic_read(&cont->count);
4811}
4812
4813static u64 debug_taskcount_read(struct cgroup *cont, struct cftype *cft)
4814{
4815        return cgroup_task_count(cont);
4816}
4817
4818static u64 current_css_set_read(struct cgroup *cont, struct cftype *cft)
4819{
4820        return (u64)(unsigned long)current->cgroups;
4821}
4822
4823static u64 current_css_set_refcount_read(struct cgroup *cont,
4824                                           struct cftype *cft)
4825{
4826        u64 count;
4827
4828        rcu_read_lock();
4829        count = atomic_read(&current->cgroups->refcount);
4830        rcu_read_unlock();
4831        return count;
4832}
4833
4834static int current_css_set_cg_links_read(struct cgroup *cont,
4835                                         struct cftype *cft,
4836                                         struct seq_file *seq)
4837{
4838        struct cg_cgroup_link *link;
4839        struct css_set *cg;
4840
4841        read_lock(&css_set_lock);
4842        rcu_read_lock();
4843        cg = rcu_dereference(current->cgroups);
4844        list_for_each_entry(link, &cg->cg_links, cg_link_list) {
4845                struct cgroup *c = link->cgrp;
4846                const char *name;
4847
4848                if (c->dentry)
4849                        name = c->dentry->d_name.name;
4850                else
4851                        name = "?";
4852                seq_printf(seq, "Root %d group %s\n",
4853                           c->root->hierarchy_id, name);
4854        }
4855        rcu_read_unlock();
4856        read_unlock(&css_set_lock);
4857        return 0;
4858}
4859
4860#define MAX_TASKS_SHOWN_PER_CSS 25
4861static int cgroup_css_links_read(struct cgroup *cont,
4862                                 struct cftype *cft,
4863                                 struct seq_file *seq)
4864{
4865        struct cg_cgroup_link *link;
4866
4867        read_lock(&css_set_lock);
4868        list_for_each_entry(link, &cont->css_sets, cgrp_link_list) {
4869                struct css_set *cg = link->cg;
4870                struct task_struct *task<