linux/kernel/cgroup.c
<<
>>
Prefs
   1/*
   2 *  Generic process-grouping system.
   3 *
   4 *  Based originally on the cpuset system, extracted by Paul Menage
   5 *  Copyright (C) 2006 Google, Inc
   6 *
   7 *  Notifications support
   8 *  Copyright (C) 2009 Nokia Corporation
   9 *  Author: Kirill A. Shutemov
  10 *
  11 *  Copyright notices from the original cpuset code:
  12 *  --------------------------------------------------
  13 *  Copyright (C) 2003 BULL SA.
  14 *  Copyright (C) 2004-2006 Silicon Graphics, Inc.
  15 *
  16 *  Portions derived from Patrick Mochel's sysfs code.
  17 *  sysfs is Copyright (c) 2001-3 Patrick Mochel
  18 *
  19 *  2003-10-10 Written by Simon Derr.
  20 *  2003-10-22 Updates by Stephen Hemminger.
  21 *  2004 May-July Rework by Paul Jackson.
  22 *  ---------------------------------------------------
  23 *
  24 *  This file is subject to the terms and conditions of the GNU General Public
  25 *  License.  See the file COPYING in the main directory of the Linux
  26 *  distribution for more details.
  27 */
  28
  29#include <linux/cgroup.h>
  30#include <linux/cred.h>
  31#include <linux/ctype.h>
  32#include <linux/errno.h>
  33#include <linux/fs.h>
  34#include <linux/init_task.h>
  35#include <linux/kernel.h>
  36#include <linux/list.h>
  37#include <linux/mm.h>
  38#include <linux/mutex.h>
  39#include <linux/mount.h>
  40#include <linux/pagemap.h>
  41#include <linux/proc_fs.h>
  42#include <linux/rcupdate.h>
  43#include <linux/sched.h>
  44#include <linux/backing-dev.h>
  45#include <linux/seq_file.h>
  46#include <linux/slab.h>
  47#include <linux/magic.h>
  48#include <linux/spinlock.h>
  49#include <linux/string.h>
  50#include <linux/sort.h>
  51#include <linux/kmod.h>
  52#include <linux/module.h>
  53#include <linux/delayacct.h>
  54#include <linux/cgroupstats.h>
  55#include <linux/hash.h>
  56#include <linux/namei.h>
  57#include <linux/pid_namespace.h>
  58#include <linux/idr.h>
  59#include <linux/vmalloc.h> /* TODO: replace with more sophisticated array */
  60#include <linux/eventfd.h>
  61#include <linux/poll.h>
  62#include <linux/flex_array.h> /* used in cgroup_attach_proc */
  63#include <linux/kthread.h>
  64
  65#include <linux/atomic.h>
  66
  67/* css deactivation bias, makes css->refcnt negative to deny new trygets */
  68#define CSS_DEACT_BIAS          INT_MIN
  69
  70/*
  71 * cgroup_mutex is the master lock.  Any modification to cgroup or its
  72 * hierarchy must be performed while holding it.
  73 *
  74 * cgroup_root_mutex nests inside cgroup_mutex and should be held to modify
  75 * cgroupfs_root of any cgroup hierarchy - subsys list, flags,
  76 * release_agent_path and so on.  Modifying requires both cgroup_mutex and
  77 * cgroup_root_mutex.  Readers can acquire either of the two.  This is to
  78 * break the following locking order cycle.
  79 *
  80 *  A. cgroup_mutex -> cred_guard_mutex -> s_type->i_mutex_key -> namespace_sem
  81 *  B. namespace_sem -> cgroup_mutex
  82 *
  83 * B happens only through cgroup_show_options() and using cgroup_root_mutex
  84 * breaks it.
  85 */
  86static DEFINE_MUTEX(cgroup_mutex);
  87static DEFINE_MUTEX(cgroup_root_mutex);
  88
  89/*
  90 * Generate an array of cgroup subsystem pointers. At boot time, this is
  91 * populated with the built in subsystems, and modular subsystems are
  92 * registered after that. The mutable section of this array is protected by
  93 * cgroup_mutex.
  94 */
  95#define SUBSYS(_x) [_x ## _subsys_id] = &_x ## _subsys,
  96#define IS_SUBSYS_ENABLED(option) IS_BUILTIN(option)
  97static struct cgroup_subsys *subsys[CGROUP_SUBSYS_COUNT] = {
  98#include <linux/cgroup_subsys.h>
  99};
 100
 101#define MAX_CGROUP_ROOT_NAMELEN 64
 102
 103/*
 104 * A cgroupfs_root represents the root of a cgroup hierarchy,
 105 * and may be associated with a superblock to form an active
 106 * hierarchy
 107 */
 108struct cgroupfs_root {
 109        struct super_block *sb;
 110
 111        /*
 112         * The bitmask of subsystems intended to be attached to this
 113         * hierarchy
 114         */
 115        unsigned long subsys_mask;
 116
 117        /* Unique id for this hierarchy. */
 118        int hierarchy_id;
 119
 120        /* The bitmask of subsystems currently attached to this hierarchy */
 121        unsigned long actual_subsys_mask;
 122
 123        /* A list running through the attached subsystems */
 124        struct list_head subsys_list;
 125
 126        /* The root cgroup for this hierarchy */
 127        struct cgroup top_cgroup;
 128
 129        /* Tracks how many cgroups are currently defined in hierarchy.*/
 130        int number_of_cgroups;
 131
 132        /* A list running through the active hierarchies */
 133        struct list_head root_list;
 134
 135        /* All cgroups on this root, cgroup_mutex protected */
 136        struct list_head allcg_list;
 137
 138        /* Hierarchy-specific flags */
 139        unsigned long flags;
 140
 141        /* IDs for cgroups in this hierarchy */
 142        struct ida cgroup_ida;
 143
 144        /* The path to use for release notifications. */
 145        char release_agent_path[PATH_MAX];
 146
 147        /* The name for this hierarchy - may be empty */
 148        char name[MAX_CGROUP_ROOT_NAMELEN];
 149};
 150
 151/*
 152 * The "rootnode" hierarchy is the "dummy hierarchy", reserved for the
 153 * subsystems that are otherwise unattached - it never has more than a
 154 * single cgroup, and all tasks are part of that cgroup.
 155 */
 156static struct cgroupfs_root rootnode;
 157
 158/*
 159 * cgroupfs file entry, pointed to from leaf dentry->d_fsdata.
 160 */
 161struct cfent {
 162        struct list_head                node;
 163        struct dentry                   *dentry;
 164        struct cftype                   *type;
 165};
 166
 167/*
 168 * CSS ID -- ID per subsys's Cgroup Subsys State(CSS). used only when
 169 * cgroup_subsys->use_id != 0.
 170 */
 171#define CSS_ID_MAX      (65535)
 172struct css_id {
 173        /*
 174         * The css to which this ID points. This pointer is set to valid value
 175         * after cgroup is populated. If cgroup is removed, this will be NULL.
 176         * This pointer is expected to be RCU-safe because destroy()
 177         * is called after synchronize_rcu(). But for safe use, css_tryget()
 178         * should be used for avoiding race.
 179         */
 180        struct cgroup_subsys_state __rcu *css;
 181        /*
 182         * ID of this css.
 183         */
 184        unsigned short id;
 185        /*
 186         * Depth in hierarchy which this ID belongs to.
 187         */
 188        unsigned short depth;
 189        /*
 190         * ID is freed by RCU. (and lookup routine is RCU safe.)
 191         */
 192        struct rcu_head rcu_head;
 193        /*
 194         * Hierarchy of CSS ID belongs to.
 195         */
 196        unsigned short stack[0]; /* Array of Length (depth+1) */
 197};
 198
 199/*
 200 * cgroup_event represents events which userspace want to receive.
 201 */
 202struct cgroup_event {
 203        /*
 204         * Cgroup which the event belongs to.
 205         */
 206        struct cgroup *cgrp;
 207        /*
 208         * Control file which the event associated.
 209         */
 210        struct cftype *cft;
 211        /*
 212         * eventfd to signal userspace about the event.
 213         */
 214        struct eventfd_ctx *eventfd;
 215        /*
 216         * Each of these stored in a list by the cgroup.
 217         */
 218        struct list_head list;
 219        /*
 220         * All fields below needed to unregister event when
 221         * userspace closes eventfd.
 222         */
 223        poll_table pt;
 224        wait_queue_head_t *wqh;
 225        wait_queue_t wait;
 226        struct work_struct remove;
 227};
 228
 229/* The list of hierarchy roots */
 230
 231static LIST_HEAD(roots);
 232static int root_count;
 233
 234static DEFINE_IDA(hierarchy_ida);
 235static int next_hierarchy_id;
 236static DEFINE_SPINLOCK(hierarchy_id_lock);
 237
 238/* dummytop is a shorthand for the dummy hierarchy's top cgroup */
 239#define dummytop (&rootnode.top_cgroup)
 240
 241/* This flag indicates whether tasks in the fork and exit paths should
 242 * check for fork/exit handlers to call. This avoids us having to do
 243 * extra work in the fork/exit path if none of the subsystems need to
 244 * be called.
 245 */
 246static int need_forkexit_callback __read_mostly;
 247
 248static int cgroup_destroy_locked(struct cgroup *cgrp);
 249static int cgroup_addrm_files(struct cgroup *cgrp, struct cgroup_subsys *subsys,
 250                              struct cftype cfts[], bool is_add);
 251
 252#ifdef CONFIG_PROVE_LOCKING
 253int cgroup_lock_is_held(void)
 254{
 255        return lockdep_is_held(&cgroup_mutex);
 256}
 257#else /* #ifdef CONFIG_PROVE_LOCKING */
 258int cgroup_lock_is_held(void)
 259{
 260        return mutex_is_locked(&cgroup_mutex);
 261}
 262#endif /* #else #ifdef CONFIG_PROVE_LOCKING */
 263
 264EXPORT_SYMBOL_GPL(cgroup_lock_is_held);
 265
 266static int css_unbias_refcnt(int refcnt)
 267{
 268        return refcnt >= 0 ? refcnt : refcnt - CSS_DEACT_BIAS;
 269}
 270
 271/* the current nr of refs, always >= 0 whether @css is deactivated or not */
 272static int css_refcnt(struct cgroup_subsys_state *css)
 273{
 274        int v = atomic_read(&css->refcnt);
 275
 276        return css_unbias_refcnt(v);
 277}
 278
 279/* convenient tests for these bits */
 280inline int cgroup_is_removed(const struct cgroup *cgrp)
 281{
 282        return test_bit(CGRP_REMOVED, &cgrp->flags);
 283}
 284
 285/* bits in struct cgroupfs_root flags field */
 286enum {
 287        ROOT_NOPREFIX,  /* mounted subsystems have no named prefix */
 288        ROOT_XATTR,     /* supports extended attributes */
 289};
 290
 291static int cgroup_is_releasable(const struct cgroup *cgrp)
 292{
 293        const int bits =
 294                (1 << CGRP_RELEASABLE) |
 295                (1 << CGRP_NOTIFY_ON_RELEASE);
 296        return (cgrp->flags & bits) == bits;
 297}
 298
 299static int notify_on_release(const struct cgroup *cgrp)
 300{
 301        return test_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
 302}
 303
 304/*
 305 * for_each_subsys() allows you to iterate on each subsystem attached to
 306 * an active hierarchy
 307 */
 308#define for_each_subsys(_root, _ss) \
 309list_for_each_entry(_ss, &_root->subsys_list, sibling)
 310
 311/* for_each_active_root() allows you to iterate across the active hierarchies */
 312#define for_each_active_root(_root) \
 313list_for_each_entry(_root, &roots, root_list)
 314
 315static inline struct cgroup *__d_cgrp(struct dentry *dentry)
 316{
 317        return dentry->d_fsdata;
 318}
 319
 320static inline struct cfent *__d_cfe(struct dentry *dentry)
 321{
 322        return dentry->d_fsdata;
 323}
 324
 325static inline struct cftype *__d_cft(struct dentry *dentry)
 326{
 327        return __d_cfe(dentry)->type;
 328}
 329
 330/* the list of cgroups eligible for automatic release. Protected by
 331 * release_list_lock */
 332static LIST_HEAD(release_list);
 333static DEFINE_RAW_SPINLOCK(release_list_lock);
 334static void cgroup_release_agent(struct work_struct *work);
 335static DECLARE_WORK(release_agent_work, cgroup_release_agent);
 336static void check_for_release(struct cgroup *cgrp);
 337
 338/* Link structure for associating css_set objects with cgroups */
 339struct cg_cgroup_link {
 340        /*
 341         * List running through cg_cgroup_links associated with a
 342         * cgroup, anchored on cgroup->css_sets
 343         */
 344        struct list_head cgrp_link_list;
 345        struct cgroup *cgrp;
 346        /*
 347         * List running through cg_cgroup_links pointing at a
 348         * single css_set object, anchored on css_set->cg_links
 349         */
 350        struct list_head cg_link_list;
 351        struct css_set *cg;
 352};
 353
 354/* The default css_set - used by init and its children prior to any
 355 * hierarchies being mounted. It contains a pointer to the root state
 356 * for each subsystem. Also used to anchor the list of css_sets. Not
 357 * reference-counted, to improve performance when child cgroups
 358 * haven't been created.
 359 */
 360
 361static struct css_set init_css_set;
 362static struct cg_cgroup_link init_css_set_link;
 363
 364static int cgroup_init_idr(struct cgroup_subsys *ss,
 365                           struct cgroup_subsys_state *css);
 366
 367/* css_set_lock protects the list of css_set objects, and the
 368 * chain of tasks off each css_set.  Nests outside task->alloc_lock
 369 * due to cgroup_iter_start() */
 370static DEFINE_RWLOCK(css_set_lock);
 371static int css_set_count;
 372
 373/*
 374 * hash table for cgroup groups. This improves the performance to find
 375 * an existing css_set. This hash doesn't (currently) take into
 376 * account cgroups in empty hierarchies.
 377 */
 378#define CSS_SET_HASH_BITS       7
 379#define CSS_SET_TABLE_SIZE      (1 << CSS_SET_HASH_BITS)
 380static struct hlist_head css_set_table[CSS_SET_TABLE_SIZE];
 381
 382static struct hlist_head *css_set_hash(struct cgroup_subsys_state *css[])
 383{
 384        int i;
 385        int index;
 386        unsigned long tmp = 0UL;
 387
 388        for (i = 0; i < CGROUP_SUBSYS_COUNT; i++)
 389                tmp += (unsigned long)css[i];
 390        tmp = (tmp >> 16) ^ tmp;
 391
 392        index = hash_long(tmp, CSS_SET_HASH_BITS);
 393
 394        return &css_set_table[index];
 395}
 396
 397/* We don't maintain the lists running through each css_set to its
 398 * task until after the first call to cgroup_iter_start(). This
 399 * reduces the fork()/exit() overhead for people who have cgroups
 400 * compiled into their kernel but not actually in use */
 401static int use_task_css_set_links __read_mostly;
 402
 403static void __put_css_set(struct css_set *cg, int taskexit)
 404{
 405        struct cg_cgroup_link *link;
 406        struct cg_cgroup_link *saved_link;
 407        /*
 408         * Ensure that the refcount doesn't hit zero while any readers
 409         * can see it. Similar to atomic_dec_and_lock(), but for an
 410         * rwlock
 411         */
 412        if (atomic_add_unless(&cg->refcount, -1, 1))
 413                return;
 414        write_lock(&css_set_lock);
 415        if (!atomic_dec_and_test(&cg->refcount)) {
 416                write_unlock(&css_set_lock);
 417                return;
 418        }
 419
 420        /* This css_set is dead. unlink it and release cgroup refcounts */
 421        hlist_del(&cg->hlist);
 422        css_set_count--;
 423
 424        list_for_each_entry_safe(link, saved_link, &cg->cg_links,
 425                                 cg_link_list) {
 426                struct cgroup *cgrp = link->cgrp;
 427                list_del(&link->cg_link_list);
 428                list_del(&link->cgrp_link_list);
 429
 430                /*
 431                 * We may not be holding cgroup_mutex, and if cgrp->count is
 432                 * dropped to 0 the cgroup can be destroyed at any time, hence
 433                 * rcu_read_lock is used to keep it alive.
 434                 */
 435                rcu_read_lock();
 436                if (atomic_dec_and_test(&cgrp->count) &&
 437                    notify_on_release(cgrp)) {
 438                        if (taskexit)
 439                                set_bit(CGRP_RELEASABLE, &cgrp->flags);
 440                        check_for_release(cgrp);
 441                }
 442                rcu_read_unlock();
 443
 444                kfree(link);
 445        }
 446
 447        write_unlock(&css_set_lock);
 448        kfree_rcu(cg, rcu_head);
 449}
 450
 451/*
 452 * refcounted get/put for css_set objects
 453 */
 454static inline void get_css_set(struct css_set *cg)
 455{
 456        atomic_inc(&cg->refcount);
 457}
 458
 459static inline void put_css_set(struct css_set *cg)
 460{
 461        __put_css_set(cg, 0);
 462}
 463
 464static inline void put_css_set_taskexit(struct css_set *cg)
 465{
 466        __put_css_set(cg, 1);
 467}
 468
 469/*
 470 * compare_css_sets - helper function for find_existing_css_set().
 471 * @cg: candidate css_set being tested
 472 * @old_cg: existing css_set for a task
 473 * @new_cgrp: cgroup that's being entered by the task
 474 * @template: desired set of css pointers in css_set (pre-calculated)
 475 *
 476 * Returns true if "cg" matches "old_cg" except for the hierarchy
 477 * which "new_cgrp" belongs to, for which it should match "new_cgrp".
 478 */
 479static bool compare_css_sets(struct css_set *cg,
 480                             struct css_set *old_cg,
 481                             struct cgroup *new_cgrp,
 482                             struct cgroup_subsys_state *template[])
 483{
 484        struct list_head *l1, *l2;
 485
 486        if (memcmp(template, cg->subsys, sizeof(cg->subsys))) {
 487                /* Not all subsystems matched */
 488                return false;
 489        }
 490
 491        /*
 492         * Compare cgroup pointers in order to distinguish between
 493         * different cgroups in heirarchies with no subsystems. We
 494         * could get by with just this check alone (and skip the
 495         * memcmp above) but on most setups the memcmp check will
 496         * avoid the need for this more expensive check on almost all
 497         * candidates.
 498         */
 499
 500        l1 = &cg->cg_links;
 501        l2 = &old_cg->cg_links;
 502        while (1) {
 503                struct cg_cgroup_link *cgl1, *cgl2;
 504                struct cgroup *cg1, *cg2;
 505
 506                l1 = l1->next;
 507                l2 = l2->next;
 508                /* See if we reached the end - both lists are equal length. */
 509                if (l1 == &cg->cg_links) {
 510                        BUG_ON(l2 != &old_cg->cg_links);
 511                        break;
 512                } else {
 513                        BUG_ON(l2 == &old_cg->cg_links);
 514                }
 515                /* Locate the cgroups associated with these links. */
 516                cgl1 = list_entry(l1, struct cg_cgroup_link, cg_link_list);
 517                cgl2 = list_entry(l2, struct cg_cgroup_link, cg_link_list);
 518                cg1 = cgl1->cgrp;
 519                cg2 = cgl2->cgrp;
 520                /* Hierarchies should be linked in the same order. */
 521                BUG_ON(cg1->root != cg2->root);
 522
 523                /*
 524                 * If this hierarchy is the hierarchy of the cgroup
 525                 * that's changing, then we need to check that this
 526                 * css_set points to the new cgroup; if it's any other
 527                 * hierarchy, then this css_set should point to the
 528                 * same cgroup as the old css_set.
 529                 */
 530                if (cg1->root == new_cgrp->root) {
 531                        if (cg1 != new_cgrp)
 532                                return false;
 533                } else {
 534                        if (cg1 != cg2)
 535                                return false;
 536                }
 537        }
 538        return true;
 539}
 540
 541/*
 542 * find_existing_css_set() is a helper for
 543 * find_css_set(), and checks to see whether an existing
 544 * css_set is suitable.
 545 *
 546 * oldcg: the cgroup group that we're using before the cgroup
 547 * transition
 548 *
 549 * cgrp: the cgroup that we're moving into
 550 *
 551 * template: location in which to build the desired set of subsystem
 552 * state objects for the new cgroup group
 553 */
 554static struct css_set *find_existing_css_set(
 555        struct css_set *oldcg,
 556        struct cgroup *cgrp,
 557        struct cgroup_subsys_state *template[])
 558{
 559        int i;
 560        struct cgroupfs_root *root = cgrp->root;
 561        struct hlist_head *hhead;
 562        struct hlist_node *node;
 563        struct css_set *cg;
 564
 565        /*
 566         * Build the set of subsystem state objects that we want to see in the
 567         * new css_set. while subsystems can change globally, the entries here
 568         * won't change, so no need for locking.
 569         */
 570        for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
 571                if (root->subsys_mask & (1UL << i)) {
 572                        /* Subsystem is in this hierarchy. So we want
 573                         * the subsystem state from the new
 574                         * cgroup */
 575                        template[i] = cgrp->subsys[i];
 576                } else {
 577                        /* Subsystem is not in this hierarchy, so we
 578                         * don't want to change the subsystem state */
 579                        template[i] = oldcg->subsys[i];
 580                }
 581        }
 582
 583        hhead = css_set_hash(template);
 584        hlist_for_each_entry(cg, node, hhead, hlist) {
 585                if (!compare_css_sets(cg, oldcg, cgrp, template))
 586                        continue;
 587
 588                /* This css_set matches what we need */
 589                return cg;
 590        }
 591
 592        /* No existing cgroup group matched */
 593        return NULL;
 594}
 595
 596static void free_cg_links(struct list_head *tmp)
 597{
 598        struct cg_cgroup_link *link;
 599        struct cg_cgroup_link *saved_link;
 600
 601        list_for_each_entry_safe(link, saved_link, tmp, cgrp_link_list) {
 602                list_del(&link->cgrp_link_list);
 603                kfree(link);
 604        }
 605}
 606
 607/*
 608 * allocate_cg_links() allocates "count" cg_cgroup_link structures
 609 * and chains them on tmp through their cgrp_link_list fields. Returns 0 on
 610 * success or a negative error
 611 */
 612static int allocate_cg_links(int count, struct list_head *tmp)
 613{
 614        struct cg_cgroup_link *link;
 615        int i;
 616        INIT_LIST_HEAD(tmp);
 617        for (i = 0; i < count; i++) {
 618                link = kmalloc(sizeof(*link), GFP_KERNEL);
 619                if (!link) {
 620                        free_cg_links(tmp);
 621                        return -ENOMEM;
 622                }
 623                list_add(&link->cgrp_link_list, tmp);
 624        }
 625        return 0;
 626}
 627
 628/**
 629 * link_css_set - a helper function to link a css_set to a cgroup
 630 * @tmp_cg_links: cg_cgroup_link objects allocated by allocate_cg_links()
 631 * @cg: the css_set to be linked
 632 * @cgrp: the destination cgroup
 633 */
 634static void link_css_set(struct list_head *tmp_cg_links,
 635                         struct css_set *cg, struct cgroup *cgrp)
 636{
 637        struct cg_cgroup_link *link;
 638
 639        BUG_ON(list_empty(tmp_cg_links));
 640        link = list_first_entry(tmp_cg_links, struct cg_cgroup_link,
 641                                cgrp_link_list);
 642        link->cg = cg;
 643        link->cgrp = cgrp;
 644        atomic_inc(&cgrp->count);
 645        list_move(&link->cgrp_link_list, &cgrp->css_sets);
 646        /*
 647         * Always add links to the tail of the list so that the list
 648         * is sorted by order of hierarchy creation
 649         */
 650        list_add_tail(&link->cg_link_list, &cg->cg_links);
 651}
 652
 653/*
 654 * find_css_set() takes an existing cgroup group and a
 655 * cgroup object, and returns a css_set object that's
 656 * equivalent to the old group, but with the given cgroup
 657 * substituted into the appropriate hierarchy. Must be called with
 658 * cgroup_mutex held
 659 */
 660static struct css_set *find_css_set(
 661        struct css_set *oldcg, struct cgroup *cgrp)
 662{
 663        struct css_set *res;
 664        struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT];
 665
 666        struct list_head tmp_cg_links;
 667
 668        struct hlist_head *hhead;
 669        struct cg_cgroup_link *link;
 670
 671        /* First see if we already have a cgroup group that matches
 672         * the desired set */
 673        read_lock(&css_set_lock);
 674        res = find_existing_css_set(oldcg, cgrp, template);
 675        if (res)
 676                get_css_set(res);
 677        read_unlock(&css_set_lock);
 678
 679        if (res)
 680                return res;
 681
 682        res = kmalloc(sizeof(*res), GFP_KERNEL);
 683        if (!res)
 684                return NULL;
 685
 686        /* Allocate all the cg_cgroup_link objects that we'll need */
 687        if (allocate_cg_links(root_count, &tmp_cg_links) < 0) {
 688                kfree(res);
 689                return NULL;
 690        }
 691
 692        atomic_set(&res->refcount, 1);
 693        INIT_LIST_HEAD(&res->cg_links);
 694        INIT_LIST_HEAD(&res->tasks);
 695        INIT_HLIST_NODE(&res->hlist);
 696
 697        /* Copy the set of subsystem state objects generated in
 698         * find_existing_css_set() */
 699        memcpy(res->subsys, template, sizeof(res->subsys));
 700
 701        write_lock(&css_set_lock);
 702        /* Add reference counts and links from the new css_set. */
 703        list_for_each_entry(link, &oldcg->cg_links, cg_link_list) {
 704                struct cgroup *c = link->cgrp;
 705                if (c->root == cgrp->root)
 706                        c = cgrp;
 707                link_css_set(&tmp_cg_links, res, c);
 708        }
 709
 710        BUG_ON(!list_empty(&tmp_cg_links));
 711
 712        css_set_count++;
 713
 714        /* Add this cgroup group to the hash table */
 715        hhead = css_set_hash(res->subsys);
 716        hlist_add_head(&res->hlist, hhead);
 717
 718        write_unlock(&css_set_lock);
 719
 720        return res;
 721}
 722
 723/*
 724 * Return the cgroup for "task" from the given hierarchy. Must be
 725 * called with cgroup_mutex held.
 726 */
 727static struct cgroup *task_cgroup_from_root(struct task_struct *task,
 728                                            struct cgroupfs_root *root)
 729{
 730        struct css_set *css;
 731        struct cgroup *res = NULL;
 732
 733        BUG_ON(!mutex_is_locked(&cgroup_mutex));
 734        read_lock(&css_set_lock);
 735        /*
 736         * No need to lock the task - since we hold cgroup_mutex the
 737         * task can't change groups, so the only thing that can happen
 738         * is that it exits and its css is set back to init_css_set.
 739         */
 740        css = task->cgroups;
 741        if (css == &init_css_set) {
 742                res = &root->top_cgroup;
 743        } else {
 744                struct cg_cgroup_link *link;
 745                list_for_each_entry(link, &css->cg_links, cg_link_list) {
 746                        struct cgroup *c = link->cgrp;
 747                        if (c->root == root) {
 748                                res = c;
 749                                break;
 750                        }
 751                }
 752        }
 753        read_unlock(&css_set_lock);
 754        BUG_ON(!res);
 755        return res;
 756}
 757
 758/*
 759 * There is one global cgroup mutex. We also require taking
 760 * task_lock() when dereferencing a task's cgroup subsys pointers.
 761 * See "The task_lock() exception", at the end of this comment.
 762 *
 763 * A task must hold cgroup_mutex to modify cgroups.
 764 *
 765 * Any task can increment and decrement the count field without lock.
 766 * So in general, code holding cgroup_mutex can't rely on the count
 767 * field not changing.  However, if the count goes to zero, then only
 768 * cgroup_attach_task() can increment it again.  Because a count of zero
 769 * means that no tasks are currently attached, therefore there is no
 770 * way a task attached to that cgroup can fork (the other way to
 771 * increment the count).  So code holding cgroup_mutex can safely
 772 * assume that if the count is zero, it will stay zero. Similarly, if
 773 * a task holds cgroup_mutex on a cgroup with zero count, it
 774 * knows that the cgroup won't be removed, as cgroup_rmdir()
 775 * needs that mutex.
 776 *
 777 * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't
 778 * (usually) take cgroup_mutex.  These are the two most performance
 779 * critical pieces of code here.  The exception occurs on cgroup_exit(),
 780 * when a task in a notify_on_release cgroup exits.  Then cgroup_mutex
 781 * is taken, and if the cgroup count is zero, a usermode call made
 782 * to the release agent with the name of the cgroup (path relative to
 783 * the root of cgroup file system) as the argument.
 784 *
 785 * A cgroup can only be deleted if both its 'count' of using tasks
 786 * is zero, and its list of 'children' cgroups is empty.  Since all
 787 * tasks in the system use _some_ cgroup, and since there is always at
 788 * least one task in the system (init, pid == 1), therefore, top_cgroup
 789 * always has either children cgroups and/or using tasks.  So we don't
 790 * need a special hack to ensure that top_cgroup cannot be deleted.
 791 *
 792 *      The task_lock() exception
 793 *
 794 * The need for this exception arises from the action of
 795 * cgroup_attach_task(), which overwrites one task's cgroup pointer with
 796 * another.  It does so using cgroup_mutex, however there are
 797 * several performance critical places that need to reference
 798 * task->cgroup without the expense of grabbing a system global
 799 * mutex.  Therefore except as noted below, when dereferencing or, as
 800 * in cgroup_attach_task(), modifying a task's cgroup pointer we use
 801 * task_lock(), which acts on a spinlock (task->alloc_lock) already in
 802 * the task_struct routinely used for such matters.
 803 *
 804 * P.S.  One more locking exception.  RCU is used to guard the
 805 * update of a tasks cgroup pointer by cgroup_attach_task()
 806 */
 807
 808/**
 809 * cgroup_lock - lock out any changes to cgroup structures
 810 *
 811 */
 812void cgroup_lock(void)
 813{
 814        mutex_lock(&cgroup_mutex);
 815}
 816EXPORT_SYMBOL_GPL(cgroup_lock);
 817
 818/**
 819 * cgroup_unlock - release lock on cgroup changes
 820 *
 821 * Undo the lock taken in a previous cgroup_lock() call.
 822 */
 823void cgroup_unlock(void)
 824{
 825        mutex_unlock(&cgroup_mutex);
 826}
 827EXPORT_SYMBOL_GPL(cgroup_unlock);
 828
 829/*
 830 * A couple of forward declarations required, due to cyclic reference loop:
 831 * cgroup_mkdir -> cgroup_create -> cgroup_populate_dir ->
 832 * cgroup_add_file -> cgroup_create_file -> cgroup_dir_inode_operations
 833 * -> cgroup_mkdir.
 834 */
 835
 836static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode);
 837static struct dentry *cgroup_lookup(struct inode *, struct dentry *, unsigned int);
 838static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry);
 839static int cgroup_populate_dir(struct cgroup *cgrp, bool base_files,
 840                               unsigned long subsys_mask);
 841static const struct inode_operations cgroup_dir_inode_operations;
 842static const struct file_operations proc_cgroupstats_operations;
 843
 844static struct backing_dev_info cgroup_backing_dev_info = {
 845        .name           = "cgroup",
 846        .capabilities   = BDI_CAP_NO_ACCT_AND_WRITEBACK,
 847};
 848
 849static int alloc_css_id(struct cgroup_subsys *ss,
 850                        struct cgroup *parent, struct cgroup *child);
 851
 852static struct inode *cgroup_new_inode(umode_t mode, struct super_block *sb)
 853{
 854        struct inode *inode = new_inode(sb);
 855
 856        if (inode) {
 857                inode->i_ino = get_next_ino();
 858                inode->i_mode = mode;
 859                inode->i_uid = current_fsuid();
 860                inode->i_gid = current_fsgid();
 861                inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 862                inode->i_mapping->backing_dev_info = &cgroup_backing_dev_info;
 863        }
 864        return inode;
 865}
 866
 867static void cgroup_diput(struct dentry *dentry, struct inode *inode)
 868{
 869        /* is dentry a directory ? if so, kfree() associated cgroup */
 870        if (S_ISDIR(inode->i_mode)) {
 871                struct cgroup *cgrp = dentry->d_fsdata;
 872                struct cgroup_subsys *ss;
 873                BUG_ON(!(cgroup_is_removed(cgrp)));
 874                /* It's possible for external users to be holding css
 875                 * reference counts on a cgroup; css_put() needs to
 876                 * be able to access the cgroup after decrementing
 877                 * the reference count in order to know if it needs to
 878                 * queue the cgroup to be handled by the release
 879                 * agent */
 880                synchronize_rcu();
 881
 882                mutex_lock(&cgroup_mutex);
 883                /*
 884                 * Release the subsystem state objects.
 885                 */
 886                for_each_subsys(cgrp->root, ss)
 887                        ss->css_free(cgrp);
 888
 889                cgrp->root->number_of_cgroups--;
 890                mutex_unlock(&cgroup_mutex);
 891
 892                /*
 893                 * Drop the active superblock reference that we took when we
 894                 * created the cgroup
 895                 */
 896                deactivate_super(cgrp->root->sb);
 897
 898                /*
 899                 * if we're getting rid of the cgroup, refcount should ensure
 900                 * that there are no pidlists left.
 901                 */
 902                BUG_ON(!list_empty(&cgrp->pidlists));
 903
 904                simple_xattrs_free(&cgrp->xattrs);
 905
 906                ida_simple_remove(&cgrp->root->cgroup_ida, cgrp->id);
 907                kfree_rcu(cgrp, rcu_head);
 908        } else {
 909                struct cfent *cfe = __d_cfe(dentry);
 910                struct cgroup *cgrp = dentry->d_parent->d_fsdata;
 911                struct cftype *cft = cfe->type;
 912
 913                WARN_ONCE(!list_empty(&cfe->node) &&
 914                          cgrp != &cgrp->root->top_cgroup,
 915                          "cfe still linked for %s\n", cfe->type->name);
 916                kfree(cfe);
 917                simple_xattrs_free(&cft->xattrs);
 918        }
 919        iput(inode);
 920}
 921
 922static int cgroup_delete(const struct dentry *d)
 923{
 924        return 1;
 925}
 926
 927static void remove_dir(struct dentry *d)
 928{
 929        struct dentry *parent = dget(d->d_parent);
 930
 931        d_delete(d);
 932        simple_rmdir(parent->d_inode, d);
 933        dput(parent);
 934}
 935
 936static int cgroup_rm_file(struct cgroup *cgrp, const struct cftype *cft)
 937{
 938        struct cfent *cfe;
 939
 940        lockdep_assert_held(&cgrp->dentry->d_inode->i_mutex);
 941        lockdep_assert_held(&cgroup_mutex);
 942
 943        list_for_each_entry(cfe, &cgrp->files, node) {
 944                struct dentry *d = cfe->dentry;
 945
 946                if (cft && cfe->type != cft)
 947                        continue;
 948
 949                dget(d);
 950                d_delete(d);
 951                simple_unlink(cgrp->dentry->d_inode, d);
 952                list_del_init(&cfe->node);
 953                dput(d);
 954
 955                return 0;
 956        }
 957        return -ENOENT;
 958}
 959
 960/**
 961 * cgroup_clear_directory - selective removal of base and subsystem files
 962 * @dir: directory containing the files
 963 * @base_files: true if the base files should be removed
 964 * @subsys_mask: mask of the subsystem ids whose files should be removed
 965 */
 966static void cgroup_clear_directory(struct dentry *dir, bool base_files,
 967                                   unsigned long subsys_mask)
 968{
 969        struct cgroup *cgrp = __d_cgrp(dir);
 970        struct cgroup_subsys *ss;
 971
 972        for_each_subsys(cgrp->root, ss) {
 973                struct cftype_set *set;
 974                if (!test_bit(ss->subsys_id, &subsys_mask))
 975                        continue;
 976                list_for_each_entry(set, &ss->cftsets, node)
 977                        cgroup_addrm_files(cgrp, NULL, set->cfts, false);
 978        }
 979        if (base_files) {
 980                while (!list_empty(&cgrp->files))
 981                        cgroup_rm_file(cgrp, NULL);
 982        }
 983}
 984
 985/*
 986 * NOTE : the dentry must have been dget()'ed
 987 */
 988static void cgroup_d_remove_dir(struct dentry *dentry)
 989{
 990        struct dentry *parent;
 991        struct cgroupfs_root *root = dentry->d_sb->s_fs_info;
 992
 993        cgroup_clear_directory(dentry, true, root->subsys_mask);
 994
 995        parent = dentry->d_parent;
 996        spin_lock(&parent->d_lock);
 997        spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
 998        list_del_init(&dentry->d_u.d_child);
 999        spin_unlock(&dentry->d_lock);
1000        spin_unlock(&parent->d_lock);
1001        remove_dir(dentry);
1002}
1003
1004/*
1005 * Call with cgroup_mutex held. Drops reference counts on modules, including
1006 * any duplicate ones that parse_cgroupfs_options took. If this function
1007 * returns an error, no reference counts are touched.
1008 */
1009static int rebind_subsystems(struct cgroupfs_root *root,
1010                              unsigned long final_subsys_mask)
1011{
1012        unsigned long added_mask, removed_mask;
1013        struct cgroup *cgrp = &root->top_cgroup;
1014        int i;
1015
1016        BUG_ON(!mutex_is_locked(&cgroup_mutex));
1017        BUG_ON(!mutex_is_locked(&cgroup_root_mutex));
1018
1019        removed_mask = root->actual_subsys_mask & ~final_subsys_mask;
1020        added_mask = final_subsys_mask & ~root->actual_subsys_mask;
1021        /* Check that any added subsystems are currently free */
1022        for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1023                unsigned long bit = 1UL << i;
1024                struct cgroup_subsys *ss = subsys[i];
1025                if (!(bit & added_mask))
1026                        continue;
1027                /*
1028                 * Nobody should tell us to do a subsys that doesn't exist:
1029                 * parse_cgroupfs_options should catch that case and refcounts
1030                 * ensure that subsystems won't disappear once selected.
1031                 */
1032                BUG_ON(ss == NULL);
1033                if (ss->root != &rootnode) {
1034                        /* Subsystem isn't free */
1035                        return -EBUSY;
1036                }
1037        }
1038
1039        /* Currently we don't handle adding/removing subsystems when
1040         * any child cgroups exist. This is theoretically supportable
1041         * but involves complex error handling, so it's being left until
1042         * later */
1043        if (root->number_of_cgroups > 1)
1044                return -EBUSY;
1045
1046        /* Process each subsystem */
1047        for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1048                struct cgroup_subsys *ss = subsys[i];
1049                unsigned long bit = 1UL << i;
1050                if (bit & added_mask) {
1051                        /* We're binding this subsystem to this hierarchy */
1052                        BUG_ON(ss == NULL);
1053                        BUG_ON(cgrp->subsys[i]);
1054                        BUG_ON(!dummytop->subsys[i]);
1055                        BUG_ON(dummytop->subsys[i]->cgroup != dummytop);
1056                        cgrp->subsys[i] = dummytop->subsys[i];
1057                        cgrp->subsys[i]->cgroup = cgrp;
1058                        list_move(&ss->sibling, &root->subsys_list);
1059                        ss->root = root;
1060                        if (ss->bind)
1061                                ss->bind(cgrp);
1062                        /* refcount was already taken, and we're keeping it */
1063                } else if (bit & removed_mask) {
1064                        /* We're removing this subsystem */
1065                        BUG_ON(ss == NULL);
1066                        BUG_ON(cgrp->subsys[i] != dummytop->subsys[i]);
1067                        BUG_ON(cgrp->subsys[i]->cgroup != cgrp);
1068                        if (ss->bind)
1069                                ss->bind(dummytop);
1070                        dummytop->subsys[i]->cgroup = dummytop;
1071                        cgrp->subsys[i] = NULL;
1072                        subsys[i]->root = &rootnode;
1073                        list_move(&ss->sibling, &rootnode.subsys_list);
1074                        /* subsystem is now free - drop reference on module */
1075                        module_put(ss->module);
1076                } else if (bit & final_subsys_mask) {
1077                        /* Subsystem state should already exist */
1078                        BUG_ON(ss == NULL);
1079                        BUG_ON(!cgrp->subsys[i]);
1080                        /*
1081                         * a refcount was taken, but we already had one, so
1082                         * drop the extra reference.
1083                         */
1084                        module_put(ss->module);
1085#ifdef CONFIG_MODULE_UNLOAD
1086                        BUG_ON(ss->module && !module_refcount(ss->module));
1087#endif
1088                } else {
1089                        /* Subsystem state shouldn't exist */
1090                        BUG_ON(cgrp->subsys[i]);
1091                }
1092        }
1093        root->subsys_mask = root->actual_subsys_mask = final_subsys_mask;
1094        synchronize_rcu();
1095
1096        return 0;
1097}
1098
1099static int cgroup_show_options(struct seq_file *seq, struct dentry *dentry)
1100{
1101        struct cgroupfs_root *root = dentry->d_sb->s_fs_info;
1102        struct cgroup_subsys *ss;
1103
1104        mutex_lock(&cgroup_root_mutex);
1105        for_each_subsys(root, ss)
1106                seq_printf(seq, ",%s", ss->name);
1107        if (test_bit(ROOT_NOPREFIX, &root->flags))
1108                seq_puts(seq, ",noprefix");
1109        if (test_bit(ROOT_XATTR, &root->flags))
1110                seq_puts(seq, ",xattr");
1111        if (strlen(root->release_agent_path))
1112                seq_printf(seq, ",release_agent=%s", root->release_agent_path);
1113        if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->top_cgroup.flags))
1114                seq_puts(seq, ",clone_children");
1115        if (strlen(root->name))
1116                seq_printf(seq, ",name=%s", root->name);
1117        mutex_unlock(&cgroup_root_mutex);
1118        return 0;
1119}
1120
1121struct cgroup_sb_opts {
1122        unsigned long subsys_mask;
1123        unsigned long flags;
1124        char *release_agent;
1125        bool cpuset_clone_children;
1126        char *name;
1127        /* User explicitly requested empty subsystem */
1128        bool none;
1129
1130        struct cgroupfs_root *new_root;
1131
1132};
1133
1134/*
1135 * Convert a hierarchy specifier into a bitmask of subsystems and flags. Call
1136 * with cgroup_mutex held to protect the subsys[] array. This function takes
1137 * refcounts on subsystems to be used, unless it returns error, in which case
1138 * no refcounts are taken.
1139 */
1140static int parse_cgroupfs_options(char *data, struct cgroup_sb_opts *opts)
1141{
1142        char *token, *o = data;
1143        bool all_ss = false, one_ss = false;
1144        unsigned long mask = (unsigned long)-1;
1145        int i;
1146        bool module_pin_failed = false;
1147
1148        BUG_ON(!mutex_is_locked(&cgroup_mutex));
1149
1150#ifdef CONFIG_CPUSETS
1151        mask = ~(1UL << cpuset_subsys_id);
1152#endif
1153
1154        memset(opts, 0, sizeof(*opts));
1155
1156        while ((token = strsep(&o, ",")) != NULL) {
1157                if (!*token)
1158                        return -EINVAL;
1159                if (!strcmp(token, "none")) {
1160                        /* Explicitly have no subsystems */
1161                        opts->none = true;
1162                        continue;
1163                }
1164                if (!strcmp(token, "all")) {
1165                        /* Mutually exclusive option 'all' + subsystem name */
1166                        if (one_ss)
1167                                return -EINVAL;
1168                        all_ss = true;
1169                        continue;
1170                }
1171                if (!strcmp(token, "noprefix")) {
1172                        set_bit(ROOT_NOPREFIX, &opts->flags);
1173                        continue;
1174                }
1175                if (!strcmp(token, "clone_children")) {
1176                        opts->cpuset_clone_children = true;
1177                        continue;
1178                }
1179                if (!strcmp(token, "xattr")) {
1180                        set_bit(ROOT_XATTR, &opts->flags);
1181                        continue;
1182                }
1183                if (!strncmp(token, "release_agent=", 14)) {
1184                        /* Specifying two release agents is forbidden */
1185                        if (opts->release_agent)
1186                                return -EINVAL;
1187                        opts->release_agent =
1188                                kstrndup(token + 14, PATH_MAX - 1, GFP_KERNEL);
1189                        if (!opts->release_agent)
1190                                return -ENOMEM;
1191                        continue;
1192                }
1193                if (!strncmp(token, "name=", 5)) {
1194                        const char *name = token + 5;
1195                        /* Can't specify an empty name */
1196                        if (!strlen(name))
1197                                return -EINVAL;
1198                        /* Must match [\w.-]+ */
1199                        for (i = 0; i < strlen(name); i++) {
1200                                char c = name[i];
1201                                if (isalnum(c))
1202                                        continue;
1203                                if ((c == '.') || (c == '-') || (c == '_'))
1204                                        continue;
1205                                return -EINVAL;
1206                        }
1207                        /* Specifying two names is forbidden */
1208                        if (opts->name)
1209                                return -EINVAL;
1210                        opts->name = kstrndup(name,
1211                                              MAX_CGROUP_ROOT_NAMELEN - 1,
1212                                              GFP_KERNEL);
1213                        if (!opts->name)
1214                                return -ENOMEM;
1215
1216                        continue;
1217                }
1218
1219                for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1220                        struct cgroup_subsys *ss = subsys[i];
1221                        if (ss == NULL)
1222                                continue;
1223                        if (strcmp(token, ss->name))
1224                                continue;
1225                        if (ss->disabled)
1226                                continue;
1227
1228                        /* Mutually exclusive option 'all' + subsystem name */
1229                        if (all_ss)
1230                                return -EINVAL;
1231                        set_bit(i, &opts->subsys_mask);
1232                        one_ss = true;
1233
1234                        break;
1235                }
1236                if (i == CGROUP_SUBSYS_COUNT)
1237                        return -ENOENT;
1238        }
1239
1240        /*
1241         * If the 'all' option was specified select all the subsystems,
1242         * otherwise if 'none', 'name=' and a subsystem name options
1243         * were not specified, let's default to 'all'
1244         */
1245        if (all_ss || (!one_ss && !opts->none && !opts->name)) {
1246                for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1247                        struct cgroup_subsys *ss = subsys[i];
1248                        if (ss == NULL)
1249                                continue;
1250                        if (ss->disabled)
1251                                continue;
1252                        set_bit(i, &opts->subsys_mask);
1253                }
1254        }
1255
1256        /* Consistency checks */
1257
1258        /*
1259         * Option noprefix was introduced just for backward compatibility
1260         * with the old cpuset, so we allow noprefix only if mounting just
1261         * the cpuset subsystem.
1262         */
1263        if (test_bit(ROOT_NOPREFIX, &opts->flags) &&
1264            (opts->subsys_mask & mask))
1265                return -EINVAL;
1266
1267
1268        /* Can't specify "none" and some subsystems */
1269        if (opts->subsys_mask && opts->none)
1270                return -EINVAL;
1271
1272        /*
1273         * We either have to specify by name or by subsystems. (So all
1274         * empty hierarchies must have a name).
1275         */
1276        if (!opts->subsys_mask && !opts->name)
1277                return -EINVAL;
1278
1279        /*
1280         * Grab references on all the modules we'll need, so the subsystems
1281         * don't dance around before rebind_subsystems attaches them. This may
1282         * take duplicate reference counts on a subsystem that's already used,
1283         * but rebind_subsystems handles this case.
1284         */
1285        for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1286                unsigned long bit = 1UL << i;
1287
1288                if (!(bit & opts->subsys_mask))
1289                        continue;
1290                if (!try_module_get(subsys[i]->module)) {
1291                        module_pin_failed = true;
1292                        break;
1293                }
1294        }
1295        if (module_pin_failed) {
1296                /*
1297                 * oops, one of the modules was going away. this means that we
1298                 * raced with a module_delete call, and to the user this is
1299                 * essentially a "subsystem doesn't exist" case.
1300                 */
1301                for (i--; i >= 0; i--) {
1302                        /* drop refcounts only on the ones we took */
1303                        unsigned long bit = 1UL << i;
1304
1305                        if (!(bit & opts->subsys_mask))
1306                                continue;
1307                        module_put(subsys[i]->module);
1308                }
1309                return -ENOENT;
1310        }
1311
1312        return 0;
1313}
1314
1315static void drop_parsed_module_refcounts(unsigned long subsys_mask)
1316{
1317        int i;
1318        for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1319                unsigned long bit = 1UL << i;
1320
1321                if (!(bit & subsys_mask))
1322                        continue;
1323                module_put(subsys[i]->module);
1324        }
1325}
1326
1327static int cgroup_remount(struct super_block *sb, int *flags, char *data)
1328{
1329        int ret = 0;
1330        struct cgroupfs_root *root = sb->s_fs_info;
1331        struct cgroup *cgrp = &root->top_cgroup;
1332        struct cgroup_sb_opts opts;
1333        unsigned long added_mask, removed_mask;
1334
1335        mutex_lock(&cgrp->dentry->d_inode->i_mutex);
1336        mutex_lock(&cgroup_mutex);
1337        mutex_lock(&cgroup_root_mutex);
1338
1339        /* See what subsystems are wanted */
1340        ret = parse_cgroupfs_options(data, &opts);
1341        if (ret)
1342                goto out_unlock;
1343
1344        if (opts.subsys_mask != root->actual_subsys_mask || opts.release_agent)
1345                pr_warning("cgroup: option changes via remount are deprecated (pid=%d comm=%s)\n",
1346                           task_tgid_nr(current), current->comm);
1347
1348        added_mask = opts.subsys_mask & ~root->subsys_mask;
1349        removed_mask = root->subsys_mask & ~opts.subsys_mask;
1350
1351        /* Don't allow flags or name to change at remount */
1352        if (opts.flags != root->flags ||
1353            (opts.name && strcmp(opts.name, root->name))) {
1354                ret = -EINVAL;
1355                drop_parsed_module_refcounts(opts.subsys_mask);
1356                goto out_unlock;
1357        }
1358
1359        /*
1360         * Clear out the files of subsystems that should be removed, do
1361         * this before rebind_subsystems, since rebind_subsystems may
1362         * change this hierarchy's subsys_list.
1363         */
1364        cgroup_clear_directory(cgrp->dentry, false, removed_mask);
1365
1366        ret = rebind_subsystems(root, opts.subsys_mask);
1367        if (ret) {
1368                /* rebind_subsystems failed, re-populate the removed files */
1369                cgroup_populate_dir(cgrp, false, removed_mask);
1370                drop_parsed_module_refcounts(opts.subsys_mask);
1371                goto out_unlock;
1372        }
1373
1374        /* re-populate subsystem files */
1375        cgroup_populate_dir(cgrp, false, added_mask);
1376
1377        if (opts.release_agent)
1378                strcpy(root->release_agent_path, opts.release_agent);
1379 out_unlock:
1380        kfree(opts.release_agent);
1381        kfree(opts.name);
1382        mutex_unlock(&cgroup_root_mutex);
1383        mutex_unlock(&cgroup_mutex);
1384        mutex_unlock(&cgrp->dentry->d_inode->i_mutex);
1385        return ret;
1386}
1387
1388static const struct super_operations cgroup_ops = {
1389        .statfs = simple_statfs,
1390        .drop_inode = generic_delete_inode,
1391        .show_options = cgroup_show_options,
1392        .remount_fs = cgroup_remount,
1393};
1394
1395static void init_cgroup_housekeeping(struct cgroup *cgrp)
1396{
1397        INIT_LIST_HEAD(&cgrp->sibling);
1398        INIT_LIST_HEAD(&cgrp->children);
1399        INIT_LIST_HEAD(&cgrp->files);
1400        INIT_LIST_HEAD(&cgrp->css_sets);
1401        INIT_LIST_HEAD(&cgrp->allcg_node);
1402        INIT_LIST_HEAD(&cgrp->release_list);
1403        INIT_LIST_HEAD(&cgrp->pidlists);
1404        mutex_init(&cgrp->pidlist_mutex);
1405        INIT_LIST_HEAD(&cgrp->event_list);
1406        spin_lock_init(&cgrp->event_list_lock);
1407        simple_xattrs_init(&cgrp->xattrs);
1408}
1409
1410static void init_cgroup_root(struct cgroupfs_root *root)
1411{
1412        struct cgroup *cgrp = &root->top_cgroup;
1413
1414        INIT_LIST_HEAD(&root->subsys_list);
1415        INIT_LIST_HEAD(&root->root_list);
1416        INIT_LIST_HEAD(&root->allcg_list);
1417        root->number_of_cgroups = 1;
1418        cgrp->root = root;
1419        cgrp->top_cgroup = cgrp;
1420        init_cgroup_housekeeping(cgrp);
1421        list_add_tail(&cgrp->allcg_node, &root->allcg_list);
1422}
1423
1424static bool init_root_id(struct cgroupfs_root *root)
1425{
1426        int ret = 0;
1427
1428        do {
1429                if (!ida_pre_get(&hierarchy_ida, GFP_KERNEL))
1430                        return false;
1431                spin_lock(&hierarchy_id_lock);
1432                /* Try to allocate the next unused ID */
1433                ret = ida_get_new_above(&hierarchy_ida, next_hierarchy_id,
1434                                        &root->hierarchy_id);
1435                if (ret == -ENOSPC)
1436                        /* Try again starting from 0 */
1437                        ret = ida_get_new(&hierarchy_ida, &root->hierarchy_id);
1438                if (!ret) {
1439                        next_hierarchy_id = root->hierarchy_id + 1;
1440                } else if (ret != -EAGAIN) {
1441                        /* Can only get here if the 31-bit IDR is full ... */
1442                        BUG_ON(ret);
1443                }
1444                spin_unlock(&hierarchy_id_lock);
1445        } while (ret);
1446        return true;
1447}
1448
1449static int cgroup_test_super(struct super_block *sb, void *data)
1450{
1451        struct cgroup_sb_opts *opts = data;
1452        struct cgroupfs_root *root = sb->s_fs_info;
1453
1454        /* If we asked for a name then it must match */
1455        if (opts->name && strcmp(opts->name, root->name))
1456                return 0;
1457
1458        /*
1459         * If we asked for subsystems (or explicitly for no
1460         * subsystems) then they must match
1461         */
1462        if ((opts->subsys_mask || opts->none)
1463            && (opts->subsys_mask != root->subsys_mask))
1464                return 0;
1465
1466        return 1;
1467}
1468
1469static struct cgroupfs_root *cgroup_root_from_opts(struct cgroup_sb_opts *opts)
1470{
1471        struct cgroupfs_root *root;
1472
1473        if (!opts->subsys_mask && !opts->none)
1474                return NULL;
1475
1476        root = kzalloc(sizeof(*root), GFP_KERNEL);
1477        if (!root)
1478                return ERR_PTR(-ENOMEM);
1479
1480        if (!init_root_id(root)) {
1481                kfree(root);
1482                return ERR_PTR(-ENOMEM);
1483        }
1484        init_cgroup_root(root);
1485
1486        root->subsys_mask = opts->subsys_mask;
1487        root->flags = opts->flags;
1488        ida_init(&root->cgroup_ida);
1489        if (opts->release_agent)
1490                strcpy(root->release_agent_path, opts->release_agent);
1491        if (opts->name)
1492                strcpy(root->name, opts->name);
1493        if (opts->cpuset_clone_children)
1494                set_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->top_cgroup.flags);
1495        return root;
1496}
1497
1498static void cgroup_drop_root(struct cgroupfs_root *root)
1499{
1500        if (!root)
1501                return;
1502
1503        BUG_ON(!root->hierarchy_id);
1504        spin_lock(&hierarchy_id_lock);
1505        ida_remove(&hierarchy_ida, root->hierarchy_id);
1506        spin_unlock(&hierarchy_id_lock);
1507        ida_destroy(&root->cgroup_ida);
1508        kfree(root);
1509}
1510
1511static int cgroup_set_super(struct super_block *sb, void *data)
1512{
1513        int ret;
1514        struct cgroup_sb_opts *opts = data;
1515
1516        /* If we don't have a new root, we can't set up a new sb */
1517        if (!opts->new_root)
1518                return -EINVAL;
1519
1520        BUG_ON(!opts->subsys_mask && !opts->none);
1521
1522        ret = set_anon_super(sb, NULL);
1523        if (ret)
1524                return ret;
1525
1526        sb->s_fs_info = opts->new_root;
1527        opts->new_root->sb = sb;
1528
1529        sb->s_blocksize = PAGE_CACHE_SIZE;
1530        sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
1531        sb->s_magic = CGROUP_SUPER_MAGIC;
1532        sb->s_op = &cgroup_ops;
1533
1534        return 0;
1535}
1536
1537static int cgroup_get_rootdir(struct super_block *sb)
1538{
1539        static const struct dentry_operations cgroup_dops = {
1540                .d_iput = cgroup_diput,
1541                .d_delete = cgroup_delete,
1542        };
1543
1544        struct inode *inode =
1545                cgroup_new_inode(S_IFDIR | S_IRUGO | S_IXUGO | S_IWUSR, sb);
1546
1547        if (!inode)
1548                return -ENOMEM;
1549
1550        inode->i_fop = &simple_dir_operations;
1551        inode->i_op = &cgroup_dir_inode_operations;
1552        /* directories start off with i_nlink == 2 (for "." entry) */
1553        inc_nlink(inode);
1554        sb->s_root = d_make_root(inode);
1555        if (!sb->s_root)
1556                return -ENOMEM;
1557        /* for everything else we want ->d_op set */
1558        sb->s_d_op = &cgroup_dops;
1559        return 0;
1560}
1561
1562static struct dentry *cgroup_mount(struct file_system_type *fs_type,
1563                         int flags, const char *unused_dev_name,
1564                         void *data)
1565{
1566        struct cgroup_sb_opts opts;
1567        struct cgroupfs_root *root;
1568        int ret = 0;
1569        struct super_block *sb;
1570        struct cgroupfs_root *new_root;
1571        struct inode *inode;
1572
1573        /* First find the desired set of subsystems */
1574        mutex_lock(&cgroup_mutex);
1575        ret = parse_cgroupfs_options(data, &opts);
1576        mutex_unlock(&cgroup_mutex);
1577        if (ret)
1578                goto out_err;
1579
1580        /*
1581         * Allocate a new cgroup root. We may not need it if we're
1582         * reusing an existing hierarchy.
1583         */
1584        new_root = cgroup_root_from_opts(&opts);
1585        if (IS_ERR(new_root)) {
1586                ret = PTR_ERR(new_root);
1587                goto drop_modules;
1588        }
1589        opts.new_root = new_root;
1590
1591        /* Locate an existing or new sb for this hierarchy */
1592        sb = sget(fs_type, cgroup_test_super, cgroup_set_super, 0, &opts);
1593        if (IS_ERR(sb)) {
1594                ret = PTR_ERR(sb);
1595                cgroup_drop_root(opts.new_root);
1596                goto drop_modules;
1597        }
1598
1599        root = sb->s_fs_info;
1600        BUG_ON(!root);
1601        if (root == opts.new_root) {
1602                /* We used the new root structure, so this is a new hierarchy */
1603                struct list_head tmp_cg_links;
1604                struct cgroup *root_cgrp = &root->top_cgroup;
1605                struct cgroupfs_root *existing_root;
1606                const struct cred *cred;
1607                int i;
1608
1609                BUG_ON(sb->s_root != NULL);
1610
1611                ret = cgroup_get_rootdir(sb);
1612                if (ret)
1613                        goto drop_new_super;
1614                inode = sb->s_root->d_inode;
1615
1616                mutex_lock(&inode->i_mutex);
1617                mutex_lock(&cgroup_mutex);
1618                mutex_lock(&cgroup_root_mutex);
1619
1620                /* Check for name clashes with existing mounts */
1621                ret = -EBUSY;
1622                if (strlen(root->name))
1623                        for_each_active_root(existing_root)
1624                                if (!strcmp(existing_root->name, root->name))
1625                                        goto unlock_drop;
1626
1627                /*
1628                 * We're accessing css_set_count without locking
1629                 * css_set_lock here, but that's OK - it can only be
1630                 * increased by someone holding cgroup_lock, and
1631                 * that's us. The worst that can happen is that we
1632                 * have some link structures left over
1633                 */
1634                ret = allocate_cg_links(css_set_count, &tmp_cg_links);
1635                if (ret)
1636                        goto unlock_drop;
1637
1638                ret = rebind_subsystems(root, root->subsys_mask);
1639                if (ret == -EBUSY) {
1640                        free_cg_links(&tmp_cg_links);
1641                        goto unlock_drop;
1642                }
1643                /*
1644                 * There must be no failure case after here, since rebinding
1645                 * takes care of subsystems' refcounts, which are explicitly
1646                 * dropped in the failure exit path.
1647                 */
1648
1649                /* EBUSY should be the only error here */
1650                BUG_ON(ret);
1651
1652                list_add(&root->root_list, &roots);
1653                root_count++;
1654
1655                sb->s_root->d_fsdata = root_cgrp;
1656                root->top_cgroup.dentry = sb->s_root;
1657
1658                /* Link the top cgroup in this hierarchy into all
1659                 * the css_set objects */
1660                write_lock(&css_set_lock);
1661                for (i = 0; i < CSS_SET_TABLE_SIZE; i++) {
1662                        struct hlist_head *hhead = &css_set_table[i];
1663                        struct hlist_node *node;
1664                        struct css_set *cg;
1665
1666                        hlist_for_each_entry(cg, node, hhead, hlist)
1667                                link_css_set(&tmp_cg_links, cg, root_cgrp);
1668                }
1669                write_unlock(&css_set_lock);
1670
1671                free_cg_links(&tmp_cg_links);
1672
1673                BUG_ON(!list_empty(&root_cgrp->children));
1674                BUG_ON(root->number_of_cgroups != 1);
1675
1676                cred = override_creds(&init_cred);
1677                cgroup_populate_dir(root_cgrp, true, root->subsys_mask);
1678                revert_creds(cred);
1679                mutex_unlock(&cgroup_root_mutex);
1680                mutex_unlock(&cgroup_mutex);
1681                mutex_unlock(&inode->i_mutex);
1682        } else {
1683                /*
1684                 * We re-used an existing hierarchy - the new root (if
1685                 * any) is not needed
1686                 */
1687                cgroup_drop_root(opts.new_root);
1688                /* no subsys rebinding, so refcounts don't change */
1689                drop_parsed_module_refcounts(opts.subsys_mask);
1690        }
1691
1692        kfree(opts.release_agent);
1693        kfree(opts.name);
1694        return dget(sb->s_root);
1695
1696 unlock_drop:
1697        mutex_unlock(&cgroup_root_mutex);
1698        mutex_unlock(&cgroup_mutex);
1699        mutex_unlock(&inode->i_mutex);
1700 drop_new_super:
1701        deactivate_locked_super(sb);
1702 drop_modules:
1703        drop_parsed_module_refcounts(opts.subsys_mask);
1704 out_err:
1705        kfree(opts.release_agent);
1706        kfree(opts.name);
1707        return ERR_PTR(ret);
1708}
1709
1710static void cgroup_kill_sb(struct super_block *sb) {
1711        struct cgroupfs_root *root = sb->s_fs_info;
1712        struct cgroup *cgrp = &root->top_cgroup;
1713        int ret;
1714        struct cg_cgroup_link *link;
1715        struct cg_cgroup_link *saved_link;
1716
1717        BUG_ON(!root);
1718
1719        BUG_ON(root->number_of_cgroups != 1);
1720        BUG_ON(!list_empty(&cgrp->children));
1721
1722        mutex_lock(&cgroup_mutex);
1723        mutex_lock(&cgroup_root_mutex);
1724
1725        /* Rebind all subsystems back to the default hierarchy */
1726        ret = rebind_subsystems(root, 0);
1727        /* Shouldn't be able to fail ... */
1728        BUG_ON(ret);
1729
1730        /*
1731         * Release all the links from css_sets to this hierarchy's
1732         * root cgroup
1733         */
1734        write_lock(&css_set_lock);
1735
1736        list_for_each_entry_safe(link, saved_link, &cgrp->css_sets,
1737                                 cgrp_link_list) {
1738                list_del(&link->cg_link_list);
1739                list_del(&link->cgrp_link_list);
1740                kfree(link);
1741        }
1742        write_unlock(&css_set_lock);
1743
1744        if (!list_empty(&root->root_list)) {
1745                list_del(&root->root_list);
1746                root_count--;
1747        }
1748
1749        mutex_unlock(&cgroup_root_mutex);
1750        mutex_unlock(&cgroup_mutex);
1751
1752        simple_xattrs_free(&cgrp->xattrs);
1753
1754        kill_litter_super(sb);
1755        cgroup_drop_root(root);
1756}
1757
1758static struct file_system_type cgroup_fs_type = {
1759        .name = "cgroup",
1760        .mount = cgroup_mount,
1761        .kill_sb = cgroup_kill_sb,
1762};
1763
1764static struct kobject *cgroup_kobj;
1765
1766/**
1767 * cgroup_path - generate the path of a cgroup
1768 * @cgrp: the cgroup in question
1769 * @buf: the buffer to write the path into
1770 * @buflen: the length of the buffer
1771 *
1772 * Called with cgroup_mutex held or else with an RCU-protected cgroup
1773 * reference.  Writes path of cgroup into buf.  Returns 0 on success,
1774 * -errno on error.
1775 */
1776int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen)
1777{
1778        struct dentry *dentry = cgrp->dentry;
1779        char *start;
1780
1781        rcu_lockdep_assert(rcu_read_lock_held() || cgroup_lock_is_held(),
1782                           "cgroup_path() called without proper locking");
1783
1784        if (!dentry || cgrp == dummytop) {
1785                /*
1786                 * Inactive subsystems have no dentry for their root
1787                 * cgroup
1788                 */
1789                strcpy(buf, "/");
1790                return 0;
1791        }
1792
1793        start = buf + buflen - 1;
1794
1795        *start = '\0';
1796        for (;;) {
1797                int len = dentry->d_name.len;
1798
1799                if ((start -= len) < buf)
1800                        return -ENAMETOOLONG;
1801                memcpy(start, dentry->d_name.name, len);
1802                cgrp = cgrp->parent;
1803                if (!cgrp)
1804                        break;
1805
1806                dentry = cgrp->dentry;
1807                if (!cgrp->parent)
1808                        continue;
1809                if (--start < buf)
1810                        return -ENAMETOOLONG;
1811                *start = '/';
1812        }
1813        memmove(buf, start, buf + buflen - start);
1814        return 0;
1815}
1816EXPORT_SYMBOL_GPL(cgroup_path);
1817
1818/*
1819 * Control Group taskset
1820 */
1821struct task_and_cgroup {
1822        struct task_struct      *task;
1823        struct cgroup           *cgrp;
1824        struct css_set          *cg;
1825};
1826
1827struct cgroup_taskset {
1828        struct task_and_cgroup  single;
1829        struct flex_array       *tc_array;
1830        int                     tc_array_len;
1831        int                     idx;
1832        struct cgroup           *cur_cgrp;
1833};
1834
1835/**
1836 * cgroup_taskset_first - reset taskset and return the first task
1837 * @tset: taskset of interest
1838 *
1839 * @tset iteration is initialized and the first task is returned.
1840 */
1841struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset)
1842{
1843        if (tset->tc_array) {
1844                tset->idx = 0;
1845                return cgroup_taskset_next(tset);
1846        } else {
1847                tset->cur_cgrp = tset->single.cgrp;
1848                return tset->single.task;
1849        }
1850}
1851EXPORT_SYMBOL_GPL(cgroup_taskset_first);
1852
1853/**
1854 * cgroup_taskset_next - iterate to the next task in taskset
1855 * @tset: taskset of interest
1856 *
1857 * Return the next task in @tset.  Iteration must have been initialized
1858 * with cgroup_taskset_first().
1859 */
1860struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset)
1861{
1862        struct task_and_cgroup *tc;
1863
1864        if (!tset->tc_array || tset->idx >= tset->tc_array_len)
1865                return NULL;
1866
1867        tc = flex_array_get(tset->tc_array, tset->idx++);
1868        tset->cur_cgrp = tc->cgrp;
1869        return tc->task;
1870}
1871EXPORT_SYMBOL_GPL(cgroup_taskset_next);
1872
1873/**
1874 * cgroup_taskset_cur_cgroup - return the matching cgroup for the current task
1875 * @tset: taskset of interest
1876 *
1877 * Return the cgroup for the current (last returned) task of @tset.  This
1878 * function must be preceded by either cgroup_taskset_first() or
1879 * cgroup_taskset_next().
1880 */
1881struct cgroup *cgroup_taskset_cur_cgroup(struct cgroup_taskset *tset)
1882{
1883        return tset->cur_cgrp;
1884}
1885EXPORT_SYMBOL_GPL(cgroup_taskset_cur_cgroup);
1886
1887/**
1888 * cgroup_taskset_size - return the number of tasks in taskset
1889 * @tset: taskset of interest
1890 */
1891int cgroup_taskset_size(struct cgroup_taskset *tset)
1892{
1893        return tset->tc_array ? tset->tc_array_len : 1;
1894}
1895EXPORT_SYMBOL_GPL(cgroup_taskset_size);
1896
1897
1898/*
1899 * cgroup_task_migrate - move a task from one cgroup to another.
1900 *
1901 * Must be called with cgroup_mutex and threadgroup locked.
1902 */
1903static void cgroup_task_migrate(struct cgroup *cgrp, struct cgroup *oldcgrp,
1904                                struct task_struct *tsk, struct css_set *newcg)
1905{
1906        struct css_set *oldcg;
1907
1908        /*
1909         * We are synchronized through threadgroup_lock() against PF_EXITING
1910         * setting such that we can't race against cgroup_exit() changing the
1911         * css_set to init_css_set and dropping the old one.
1912         */
1913        WARN_ON_ONCE(tsk->flags & PF_EXITING);
1914        oldcg = tsk->cgroups;
1915
1916        task_lock(tsk);
1917        rcu_assign_pointer(tsk->cgroups, newcg);
1918        task_unlock(tsk);
1919
1920        /* Update the css_set linked lists if we're using them */
1921        write_lock(&css_set_lock);
1922        if (!list_empty(&tsk->cg_list))
1923                list_move(&tsk->cg_list, &newcg->tasks);
1924        write_unlock(&css_set_lock);
1925
1926        /*
1927         * We just gained a reference on oldcg by taking it from the task. As
1928         * trading it for newcg is protected by cgroup_mutex, we're safe to drop
1929         * it here; it will be freed under RCU.
1930         */
1931        set_bit(CGRP_RELEASABLE, &oldcgrp->flags);
1932        put_css_set(oldcg);
1933}
1934
1935/**
1936 * cgroup_attach_task - attach task 'tsk' to cgroup 'cgrp'
1937 * @cgrp: the cgroup the task is attaching to
1938 * @tsk: the task to be attached
1939 *
1940 * Call with cgroup_mutex and threadgroup locked. May take task_lock of
1941 * @tsk during call.
1942 */
1943int cgroup_attach_task(struct cgroup *cgrp, struct task_struct *tsk)
1944{
1945        int retval = 0;
1946        struct cgroup_subsys *ss, *failed_ss = NULL;
1947        struct cgroup *oldcgrp;
1948        struct cgroupfs_root *root = cgrp->root;
1949        struct cgroup_taskset tset = { };
1950        struct css_set *newcg;
1951
1952        /* @tsk either already exited or can't exit until the end */
1953        if (tsk->flags & PF_EXITING)
1954                return -ESRCH;
1955
1956        /* Nothing to do if the task is already in that cgroup */
1957        oldcgrp = task_cgroup_from_root(tsk, root);
1958        if (cgrp == oldcgrp)
1959                return 0;
1960
1961        tset.single.task = tsk;
1962        tset.single.cgrp = oldcgrp;
1963
1964        for_each_subsys(root, ss) {
1965                if (ss->can_attach) {
1966                        retval = ss->can_attach(cgrp, &tset);
1967                        if (retval) {
1968                                /*
1969                                 * Remember on which subsystem the can_attach()
1970                                 * failed, so that we only call cancel_attach()
1971                                 * against the subsystems whose can_attach()
1972                                 * succeeded. (See below)
1973                                 */
1974                                failed_ss = ss;
1975                                goto out;
1976                        }
1977                }
1978        }
1979
1980        newcg = find_css_set(tsk->cgroups, cgrp);
1981        if (!newcg) {
1982                retval = -ENOMEM;
1983                goto out;
1984        }
1985
1986        cgroup_task_migrate(cgrp, oldcgrp, tsk, newcg);
1987
1988        for_each_subsys(root, ss) {
1989                if (ss->attach)
1990                        ss->attach(cgrp, &tset);
1991        }
1992
1993        synchronize_rcu();
1994out:
1995        if (retval) {
1996                for_each_subsys(root, ss) {
1997                        if (ss == failed_ss)
1998                                /*
1999                                 * This subsystem was the one that failed the
2000                                 * can_attach() check earlier, so we don't need
2001                                 * to call cancel_attach() against it or any
2002                                 * remaining subsystems.
2003                                 */
2004                                break;
2005                        if (ss->cancel_attach)
2006                                ss->cancel_attach(cgrp, &tset);
2007                }
2008        }
2009        return retval;
2010}
2011
2012/**
2013 * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from'
2014 * @from: attach to all cgroups of a given task
2015 * @tsk: the task to be attached
2016 */
2017int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk)
2018{
2019        struct cgroupfs_root *root;
2020        int retval = 0;
2021
2022        cgroup_lock();
2023        for_each_active_root(root) {
2024                struct cgroup *from_cg = task_cgroup_from_root(from, root);
2025
2026                retval = cgroup_attach_task(from_cg, tsk);
2027                if (retval)
2028                        break;
2029        }
2030        cgroup_unlock();
2031
2032        return retval;
2033}
2034EXPORT_SYMBOL_GPL(cgroup_attach_task_all);
2035
2036/**
2037 * cgroup_attach_proc - attach all threads in a threadgroup to a cgroup
2038 * @cgrp: the cgroup to attach to
2039 * @leader: the threadgroup leader task_struct of the group to be attached
2040 *
2041 * Call holding cgroup_mutex and the group_rwsem of the leader. Will take
2042 * task_lock of each thread in leader's threadgroup individually in turn.
2043 */
2044static int cgroup_attach_proc(struct cgroup *cgrp, struct task_struct *leader)
2045{
2046        int retval, i, group_size;
2047        struct cgroup_subsys *ss, *failed_ss = NULL;
2048        /* guaranteed to be initialized later, but the compiler needs this */
2049        struct cgroupfs_root *root = cgrp->root;
2050        /* threadgroup list cursor and array */
2051        struct task_struct *tsk;
2052        struct task_and_cgroup *tc;
2053        struct flex_array *group;
2054        struct cgroup_taskset tset = { };
2055
2056        /*
2057         * step 0: in order to do expensive, possibly blocking operations for
2058         * every thread, we cannot iterate the thread group list, since it needs
2059         * rcu or tasklist locked. instead, build an array of all threads in the
2060         * group - group_rwsem prevents new threads from appearing, and if
2061         * threads exit, this will just be an over-estimate.
2062         */
2063        group_size = get_nr_threads(leader);
2064        /* flex_array supports very large thread-groups better than kmalloc. */
2065        group = flex_array_alloc(sizeof(*tc), group_size, GFP_KERNEL);
2066        if (!group)
2067                return -ENOMEM;
2068        /* pre-allocate to guarantee space while iterating in rcu read-side. */
2069        retval = flex_array_prealloc(group, 0, group_size - 1, GFP_KERNEL);
2070        if (retval)
2071                goto out_free_group_list;
2072
2073        tsk = leader;
2074        i = 0;
2075        /*
2076         * Prevent freeing of tasks while we take a snapshot. Tasks that are
2077         * already PF_EXITING could be freed from underneath us unless we
2078         * take an rcu_read_lock.
2079         */
2080        rcu_read_lock();
2081        do {
2082                struct task_and_cgroup ent;
2083
2084                /* @tsk either already exited or can't exit until the end */
2085                if (tsk->flags & PF_EXITING)
2086                        continue;
2087
2088                /* as per above, nr_threads may decrease, but not increase. */
2089                BUG_ON(i >= group_size);
2090                ent.task = tsk;
2091                ent.cgrp = task_cgroup_from_root(tsk, root);
2092                /* nothing to do if this task is already in the cgroup */
2093                if (ent.cgrp == cgrp)
2094                        continue;
2095                /*
2096                 * saying GFP_ATOMIC has no effect here because we did prealloc
2097                 * earlier, but it's good form to communicate our expectations.
2098                 */
2099                retval = flex_array_put(group, i, &ent, GFP_ATOMIC);
2100                BUG_ON(retval != 0);
2101                i++;
2102        } while_each_thread(leader, tsk);
2103        rcu_read_unlock();
2104        /* remember the number of threads in the array for later. */
2105        group_size = i;
2106        tset.tc_array = group;
2107        tset.tc_array_len = group_size;
2108
2109        /* methods shouldn't be called if no task is actually migrating */
2110        retval = 0;
2111        if (!group_size)
2112                goto out_free_group_list;
2113
2114        /*
2115         * step 1: check that we can legitimately attach to the cgroup.
2116         */
2117        for_each_subsys(root, ss) {
2118                if (ss->can_attach) {
2119                        retval = ss->can_attach(cgrp, &tset);
2120                        if (retval) {
2121                                failed_ss = ss;
2122                                goto out_cancel_attach;
2123                        }
2124                }
2125        }
2126
2127        /*
2128         * step 2: make sure css_sets exist for all threads to be migrated.
2129         * we use find_css_set, which allocates a new one if necessary.
2130         */
2131        for (i = 0; i < group_size; i++) {
2132                tc = flex_array_get(group, i);
2133                tc->cg = find_css_set(tc->task->cgroups, cgrp);
2134                if (!tc->cg) {
2135                        retval = -ENOMEM;
2136                        goto out_put_css_set_refs;
2137                }
2138        }
2139
2140        /*
2141         * step 3: now that we're guaranteed success wrt the css_sets,
2142         * proceed to move all tasks to the new cgroup.  There are no
2143         * failure cases after here, so this is the commit point.
2144         */
2145        for (i = 0; i < group_size; i++) {
2146                tc = flex_array_get(group, i);
2147                cgroup_task_migrate(cgrp, tc->cgrp, tc->task, tc->cg);
2148        }
2149        /* nothing is sensitive to fork() after this point. */
2150
2151        /*
2152         * step 4: do subsystem attach callbacks.
2153         */
2154        for_each_subsys(root, ss) {
2155                if (ss->attach)
2156                        ss->attach(cgrp, &tset);
2157        }
2158
2159        /*
2160         * step 5: success! and cleanup
2161         */
2162        synchronize_rcu();
2163        retval = 0;
2164out_put_css_set_refs:
2165        if (retval) {
2166                for (i = 0; i < group_size; i++) {
2167                        tc = flex_array_get(group, i);
2168                        if (!tc->cg)
2169                                break;
2170                        put_css_set(tc->cg);
2171                }
2172        }
2173out_cancel_attach:
2174        if (retval) {
2175                for_each_subsys(root, ss) {
2176                        if (ss == failed_ss)
2177                                break;
2178                        if (ss->cancel_attach)
2179                                ss->cancel_attach(cgrp, &tset);
2180                }
2181        }
2182out_free_group_list:
2183        flex_array_free(group);
2184        return retval;
2185}
2186
2187/*
2188 * Find the task_struct of the task to attach by vpid and pass it along to the
2189 * function to attach either it or all tasks in its threadgroup. Will lock
2190 * cgroup_mutex and threadgroup; may take task_lock of task.
2191 */
2192static int attach_task_by_pid(struct cgroup *cgrp, u64 pid, bool threadgroup)
2193{
2194        struct task_struct *tsk;
2195        const struct cred *cred = current_cred(), *tcred;
2196        int ret;
2197
2198        if (!cgroup_lock_live_group(cgrp))
2199                return -ENODEV;
2200
2201retry_find_task:
2202        rcu_read_lock();
2203        if (pid) {
2204                tsk = find_task_by_vpid(pid);
2205                if (!tsk) {
2206                        rcu_read_unlock();
2207                        ret= -ESRCH;
2208                        goto out_unlock_cgroup;
2209                }
2210                /*
2211                 * even if we're attaching all tasks in the thread group, we
2212                 * only need to check permissions on one of them.
2213                 */
2214                tcred = __task_cred(tsk);
2215                if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) &&
2216                    !uid_eq(cred->euid, tcred->uid) &&
2217                    !uid_eq(cred->euid, tcred->suid)) {
2218                        rcu_read_unlock();
2219                        ret = -EACCES;
2220                        goto out_unlock_cgroup;
2221                }
2222        } else
2223                tsk = current;
2224
2225        if (threadgroup)
2226                tsk = tsk->group_leader;
2227
2228        /*
2229         * Workqueue threads may acquire PF_THREAD_BOUND and become
2230         * trapped in a cpuset, or RT worker may be born in a cgroup
2231         * with no rt_runtime allocated.  Just say no.
2232         */
2233        if (tsk == kthreadd_task || (tsk->flags & PF_THREAD_BOUND)) {
2234                ret = -EINVAL;
2235                rcu_read_unlock();
2236                goto out_unlock_cgroup;
2237        }
2238
2239        get_task_struct(tsk);
2240        rcu_read_unlock();
2241
2242        threadgroup_lock(tsk);
2243        if (threadgroup) {
2244                if (!thread_group_leader(tsk)) {
2245                        /*
2246                         * a race with de_thread from another thread's exec()
2247                         * may strip us of our leadership, if this happens,
2248                         * there is no choice but to throw this task away and
2249                         * try again; this is
2250                         * "double-double-toil-and-trouble-check locking".
2251                         */
2252                        threadgroup_unlock(tsk);
2253                        put_task_struct(tsk);
2254                        goto retry_find_task;
2255                }
2256                ret = cgroup_attach_proc(cgrp, tsk);
2257        } else
2258                ret = cgroup_attach_task(cgrp, tsk);
2259        threadgroup_unlock(tsk);
2260
2261        put_task_struct(tsk);
2262out_unlock_cgroup:
2263        cgroup_unlock();
2264        return ret;
2265}
2266
2267static int cgroup_tasks_write(struct cgroup *cgrp, struct cftype *cft, u64 pid)
2268{
2269        return attach_task_by_pid(cgrp, pid, false);
2270}
2271
2272static int cgroup_procs_write(struct cgroup *cgrp, struct cftype *cft, u64 tgid)
2273{
2274        return attach_task_by_pid(cgrp, tgid, true);
2275}
2276
2277/**
2278 * cgroup_lock_live_group - take cgroup_mutex and check that cgrp is alive.
2279 * @cgrp: the cgroup to be checked for liveness
2280 *
2281 * On success, returns true; the lock should be later released with
2282 * cgroup_unlock(). On failure returns false with no lock held.
2283 */
2284bool cgroup_lock_live_group(struct cgroup *cgrp)
2285{
2286        mutex_lock(&cgroup_mutex);
2287        if (cgroup_is_removed(cgrp)) {
2288                mutex_unlock(&cgroup_mutex);
2289                return false;
2290        }
2291        return true;
2292}
2293EXPORT_SYMBOL_GPL(cgroup_lock_live_group);
2294
2295static int cgroup_release_agent_write(struct cgroup *cgrp, struct cftype *cft,
2296                                      const char *buffer)
2297{
2298        BUILD_BUG_ON(sizeof(cgrp->root->release_agent_path) < PATH_MAX);
2299        if (strlen(buffer) >= PATH_MAX)
2300                return -EINVAL;
2301        if (!cgroup_lock_live_group(cgrp))
2302                return -ENODEV;
2303        mutex_lock(&cgroup_root_mutex);
2304        strcpy(cgrp->root->release_agent_path, buffer);
2305        mutex_unlock(&cgroup_root_mutex);
2306        cgroup_unlock();
2307        return 0;
2308}
2309
2310static int cgroup_release_agent_show(struct cgroup *cgrp, struct cftype *cft,
2311                                     struct seq_file *seq)
2312{
2313        if (!cgroup_lock_live_group(cgrp))
2314                return -ENODEV;
2315        seq_puts(seq, cgrp->root->release_agent_path);
2316        seq_putc(seq, '\n');
2317        cgroup_unlock();
2318        return 0;
2319}
2320
2321/* A buffer size big enough for numbers or short strings */
2322#define CGROUP_LOCAL_BUFFER_SIZE 64
2323
2324static ssize_t cgroup_write_X64(struct cgroup *cgrp, struct cftype *cft,
2325                                struct file *file,
2326                                const char __user *userbuf,
2327                                size_t nbytes, loff_t *unused_ppos)
2328{
2329        char buffer[CGROUP_LOCAL_BUFFER_SIZE];
2330        int retval = 0;
2331        char *end;
2332
2333        if (!nbytes)
2334                return -EINVAL;
2335        if (nbytes >= sizeof(buffer))
2336                return -E2BIG;
2337        if (copy_from_user(buffer, userbuf, nbytes))
2338                return -EFAULT;
2339
2340        buffer[nbytes] = 0;     /* nul-terminate */
2341        if (cft->write_u64) {
2342                u64 val = simple_strtoull(strstrip(buffer), &end, 0);
2343                if (*end)
2344                        return -EINVAL;
2345                retval = cft->write_u64(cgrp, cft, val);
2346        } else {
2347                s64 val = simple_strtoll(strstrip(buffer), &end, 0);
2348                if (*end)
2349                        return -EINVAL;
2350                retval = cft->write_s64(cgrp, cft, val);
2351        }
2352        if (!retval)
2353                retval = nbytes;
2354        return retval;
2355}
2356
2357static ssize_t cgroup_write_string(struct cgroup *cgrp, struct cftype *cft,
2358                                   struct file *file,
2359                                   const char __user *userbuf,
2360                                   size_t nbytes, loff_t *unused_ppos)
2361{
2362        char local_buffer[CGROUP_LOCAL_BUFFER_SIZE];
2363        int retval = 0;
2364        size_t max_bytes = cft->max_write_len;
2365        char *buffer = local_buffer;
2366
2367        if (!max_bytes)
2368                max_bytes = sizeof(local_buffer) - 1;
2369        if (nbytes >= max_bytes)
2370                return -E2BIG;
2371        /* Allocate a dynamic buffer if we need one */
2372        if (nbytes >= sizeof(local_buffer)) {
2373                buffer = kmalloc(nbytes + 1, GFP_KERNEL);
2374                if (buffer == NULL)
2375                        return -ENOMEM;
2376        }
2377        if (nbytes && copy_from_user(buffer, userbuf, nbytes)) {
2378                retval = -EFAULT;
2379                goto out;
2380        }
2381
2382        buffer[nbytes] = 0;     /* nul-terminate */
2383        retval = cft->write_string(cgrp, cft, strstrip(buffer));
2384        if (!retval)
2385                retval = nbytes;
2386out:
2387        if (buffer != local_buffer)
2388                kfree(buffer);
2389        return retval;
2390}
2391
2392static ssize_t cgroup_file_write(struct file *file, const char __user *buf,
2393                                                size_t nbytes, loff_t *ppos)
2394{
2395        struct cftype *cft = __d_cft(file->f_dentry);
2396        struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
2397
2398        if (cgroup_is_removed(cgrp))
2399                return -ENODEV;
2400        if (cft->write)
2401                return cft->write(cgrp, cft, file, buf, nbytes, ppos);
2402        if (cft->write_u64 || cft->write_s64)
2403                return cgroup_write_X64(cgrp, cft, file, buf, nbytes, ppos);
2404        if (cft->write_string)
2405                return cgroup_write_string(cgrp, cft, file, buf, nbytes, ppos);
2406        if (cft->trigger) {
2407                int ret = cft->trigger(cgrp, (unsigned int)cft->private);
2408                return ret ? ret : nbytes;
2409        }
2410        return -EINVAL;
2411}
2412
2413static ssize_t cgroup_read_u64(struct cgroup *cgrp, struct cftype *cft,
2414                               struct file *file,
2415                               char __user *buf, size_t nbytes,
2416                               loff_t *ppos)
2417{
2418        char tmp[CGROUP_LOCAL_BUFFER_SIZE];
2419        u64 val = cft->read_u64(cgrp, cft);
2420        int len = sprintf(tmp, "%llu\n", (unsigned long long) val);
2421
2422        return simple_read_from_buffer(buf, nbytes, ppos, tmp, len);
2423}
2424
2425static ssize_t cgroup_read_s64(struct cgroup *cgrp, struct cftype *cft,
2426                               struct file *file,
2427                               char __user *buf, size_t nbytes,
2428                               loff_t *ppos)
2429{
2430        char tmp[CGROUP_LOCAL_BUFFER_SIZE];
2431        s64 val = cft->read_s64(cgrp, cft);
2432        int len = sprintf(tmp, "%lld\n", (long long) val);
2433
2434        return simple_read_from_buffer(buf, nbytes, ppos, tmp, len);
2435}
2436
2437static ssize_t cgroup_file_read(struct file *file, char __user *buf,
2438                                   size_t nbytes, loff_t *ppos)
2439{
2440        struct cftype *cft = __d_cft(file->f_dentry);
2441        struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
2442
2443        if (cgroup_is_removed(cgrp))
2444                return -ENODEV;
2445
2446        if (cft->read)
2447                return cft->read(cgrp, cft, file, buf, nbytes, ppos);
2448        if (cft->read_u64)
2449                return cgroup_read_u64(cgrp, cft, file, buf, nbytes, ppos);
2450        if (cft->read_s64)
2451                return cgroup_read_s64(cgrp, cft, file, buf, nbytes, ppos);
2452        return -EINVAL;
2453}
2454
2455/*
2456 * seqfile ops/methods for returning structured data. Currently just
2457 * supports string->u64 maps, but can be extended in future.
2458 */
2459
2460struct cgroup_seqfile_state {
2461        struct cftype *cft;
2462        struct cgroup *cgroup;
2463};
2464
2465static int cgroup_map_add(struct cgroup_map_cb *cb, const char *key, u64 value)
2466{
2467        struct seq_file *sf = cb->state;
2468        return seq_printf(sf, "%s %llu\n", key, (unsigned long long)value);
2469}
2470
2471static int cgroup_seqfile_show(struct seq_file *m, void *arg)
2472{
2473        struct cgroup_seqfile_state *state = m->private;
2474        struct cftype *cft = state->cft;
2475        if (cft->read_map) {
2476                struct cgroup_map_cb cb = {
2477                        .fill = cgroup_map_add,
2478                        .state = m,
2479                };
2480                return cft->read_map(state->cgroup, cft, &cb);
2481        }
2482        return cft->read_seq_string(state->cgroup, cft, m);
2483}
2484
2485static int cgroup_seqfile_release(struct inode *inode, struct file *file)
2486{
2487        struct seq_file *seq = file->private_data;
2488        kfree(seq->private);
2489        return single_release(inode, file);
2490}
2491
2492static const struct file_operations cgroup_seqfile_operations = {
2493        .read = seq_read,
2494        .write = cgroup_file_write,
2495        .llseek = seq_lseek,
2496        .release = cgroup_seqfile_release,
2497};
2498
2499static int cgroup_file_open(struct inode *inode, struct file *file)
2500{
2501        int err;
2502        struct cftype *cft;
2503
2504        err = generic_file_open(inode, file);
2505        if (err)
2506                return err;
2507        cft = __d_cft(file->f_dentry);
2508
2509        if (cft->read_map || cft->read_seq_string) {
2510                struct cgroup_seqfile_state *state =
2511                        kzalloc(sizeof(*state), GFP_USER);
2512                if (!state)
2513                        return -ENOMEM;
2514                state->cft = cft;
2515                state->cgroup = __d_cgrp(file->f_dentry->d_parent);
2516                file->f_op = &cgroup_seqfile_operations;
2517                err = single_open(file, cgroup_seqfile_show, state);
2518                if (err < 0)
2519                        kfree(state);
2520        } else if (cft->open)
2521                err = cft->open(inode, file);
2522        else
2523                err = 0;
2524
2525        return err;
2526}
2527
2528static int cgroup_file_release(struct inode *inode, struct file *file)
2529{
2530        struct cftype *cft = __d_cft(file->f_dentry);
2531        if (cft->release)
2532                return cft->release(inode, file);
2533        return 0;
2534}
2535
2536/*
2537 * cgroup_rename - Only allow simple rename of directories in place.
2538 */
2539static int cgroup_rename(struct inode *old_dir, struct dentry *old_dentry,
2540                            struct inode *new_dir, struct dentry *new_dentry)
2541{
2542        if (!S_ISDIR(old_dentry->d_inode->i_mode))
2543                return -ENOTDIR;
2544        if (new_dentry->d_inode)
2545                return -EEXIST;
2546        if (old_dir != new_dir)
2547                return -EIO;
2548        return simple_rename(old_dir, old_dentry, new_dir, new_dentry);
2549}
2550
2551static struct simple_xattrs *__d_xattrs(struct dentry *dentry)
2552{
2553        if (S_ISDIR(dentry->d_inode->i_mode))
2554                return &__d_cgrp(dentry)->xattrs;
2555        else
2556                return &__d_cft(dentry)->xattrs;
2557}
2558
2559static inline int xattr_enabled(struct dentry *dentry)
2560{
2561        struct cgroupfs_root *root = dentry->d_sb->s_fs_info;
2562        return test_bit(ROOT_XATTR, &root->flags);
2563}
2564
2565static bool is_valid_xattr(const char *name)
2566{
2567        if (!strncmp(name, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN) ||
2568            !strncmp(name, XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN))
2569                return true;
2570        return false;
2571}
2572
2573static int cgroup_setxattr(struct dentry *dentry, const char *name,
2574                           const void *val, size_t size, int flags)
2575{
2576        if (!xattr_enabled(dentry))
2577                return -EOPNOTSUPP;
2578        if (!is_valid_xattr(name))
2579                return -EINVAL;
2580        return simple_xattr_set(__d_xattrs(dentry), name, val, size, flags);
2581}
2582
2583static int cgroup_removexattr(struct dentry *dentry, const char *name)
2584{
2585        if (!xattr_enabled(dentry))
2586                return -EOPNOTSUPP;
2587        if (!is_valid_xattr(name))
2588                return -EINVAL;
2589        return simple_xattr_remove(__d_xattrs(dentry), name);
2590}
2591
2592static ssize_t cgroup_getxattr(struct dentry *dentry, const char *name,
2593                               void *buf, size_t size)
2594{
2595        if (!xattr_enabled(dentry))
2596                return -EOPNOTSUPP;
2597        if (!is_valid_xattr(name))
2598                return -EINVAL;
2599        return simple_xattr_get(__d_xattrs(dentry), name, buf, size);
2600}
2601
2602static ssize_t cgroup_listxattr(struct dentry *dentry, char *buf, size_t size)
2603{
2604        if (!xattr_enabled(dentry))
2605                return -EOPNOTSUPP;
2606        return simple_xattr_list(__d_xattrs(dentry), buf, size);
2607}
2608
2609static const struct file_operations cgroup_file_operations = {
2610        .read = cgroup_file_read,
2611        .write = cgroup_file_write,
2612        .llseek = generic_file_llseek,
2613        .open = cgroup_file_open,
2614        .release = cgroup_file_release,
2615};
2616
2617static const struct inode_operations cgroup_file_inode_operations = {
2618        .setxattr = cgroup_setxattr,
2619        .getxattr = cgroup_getxattr,
2620        .listxattr = cgroup_listxattr,
2621        .removexattr = cgroup_removexattr,
2622};
2623
2624static const struct inode_operations cgroup_dir_inode_operations = {
2625        .lookup = cgroup_lookup,
2626        .mkdir = cgroup_mkdir,
2627        .rmdir = cgroup_rmdir,
2628        .rename = cgroup_rename,
2629        .setxattr = cgroup_setxattr,
2630        .getxattr = cgroup_getxattr,
2631        .listxattr = cgroup_listxattr,
2632        .removexattr = cgroup_removexattr,
2633};
2634
2635static struct dentry *cgroup_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
2636{
2637        if (dentry->d_name.len > NAME_MAX)
2638                return ERR_PTR(-ENAMETOOLONG);
2639        d_add(dentry, NULL);
2640        return NULL;
2641}
2642
2643/*
2644 * Check if a file is a control file
2645 */
2646static inline struct cftype *__file_cft(struct file *file)
2647{
2648        if (file->f_dentry->d_inode->i_fop != &cgroup_file_operations)
2649                return ERR_PTR(-EINVAL);
2650        return __d_cft(file->f_dentry);
2651}
2652
2653static int cgroup_create_file(struct dentry *dentry, umode_t mode,
2654                                struct super_block *sb)
2655{
2656        struct inode *inode;
2657
2658        if (!dentry)
2659                return -ENOENT;
2660        if (dentry->d_inode)
2661                return -EEXIST;
2662
2663        inode = cgroup_new_inode(mode, sb);
2664        if (!inode)
2665                return -ENOMEM;
2666
2667        if (S_ISDIR(mode)) {
2668                inode->i_op = &cgroup_dir_inode_operations;
2669                inode->i_fop = &simple_dir_operations;
2670
2671                /* start off with i_nlink == 2 (for "." entry) */
2672                inc_nlink(inode);
2673                inc_nlink(dentry->d_parent->d_inode);
2674
2675                /*
2676                 * Control reaches here with cgroup_mutex held.
2677                 * @inode->i_mutex should nest outside cgroup_mutex but we
2678                 * want to populate it immediately without releasing
2679                 * cgroup_mutex.  As @inode isn't visible to anyone else
2680                 * yet, trylock will always succeed without affecting
2681                 * lockdep checks.
2682                 */
2683                WARN_ON_ONCE(!mutex_trylock(&inode->i_mutex));
2684        } else if (S_ISREG(mode)) {
2685                inode->i_size = 0;
2686                inode->i_fop = &cgroup_file_operations;
2687                inode->i_op = &cgroup_file_inode_operations;
2688        }
2689        d_instantiate(dentry, inode);
2690        dget(dentry);   /* Extra count - pin the dentry in core */
2691        return 0;
2692}
2693
2694/**
2695 * cgroup_file_mode - deduce file mode of a control file
2696 * @cft: the control file in question
2697 *
2698 * returns cft->mode if ->mode is not 0
2699 * returns S_IRUGO|S_IWUSR if it has both a read and a write handler
2700 * returns S_IRUGO if it has only a read handler
2701 * returns S_IWUSR if it has only a write hander
2702 */
2703static umode_t cgroup_file_mode(const struct cftype *cft)
2704{
2705        umode_t mode = 0;
2706
2707        if (cft->mode)
2708                return cft->mode;
2709
2710        if (cft->read || cft->read_u64 || cft->read_s64 ||
2711            cft->read_map || cft->read_seq_string)
2712                mode |= S_IRUGO;
2713
2714        if (cft->write || cft->write_u64 || cft->write_s64 ||
2715            cft->write_string || cft->trigger)
2716                mode |= S_IWUSR;
2717
2718        return mode;
2719}
2720
2721static int cgroup_add_file(struct cgroup *cgrp, struct cgroup_subsys *subsys,
2722                           struct cftype *cft)
2723{
2724        struct dentry *dir = cgrp->dentry;
2725        struct cgroup *parent = __d_cgrp(dir);
2726        struct dentry *dentry;
2727        struct cfent *cfe;
2728        int error;
2729        umode_t mode;
2730        char name[MAX_CGROUP_TYPE_NAMELEN + MAX_CFTYPE_NAME + 2] = { 0 };
2731
2732        simple_xattrs_init(&cft->xattrs);
2733
2734        if (subsys && !test_bit(ROOT_NOPREFIX, &cgrp->root->flags)) {
2735                strcpy(name, subsys->name);
2736                strcat(name, ".");
2737        }
2738        strcat(name, cft->name);
2739
2740        BUG_ON(!mutex_is_locked(&dir->d_inode->i_mutex));
2741
2742        cfe = kzalloc(sizeof(*cfe), GFP_KERNEL);
2743        if (!cfe)
2744                return -ENOMEM;
2745
2746        dentry = lookup_one_len(name, dir, strlen(name));
2747        if (IS_ERR(dentry)) {
2748                error = PTR_ERR(dentry);
2749                goto out;
2750        }
2751
2752        mode = cgroup_file_mode(cft);
2753        error = cgroup_create_file(dentry, mode | S_IFREG, cgrp->root->sb);
2754        if (!error) {
2755                cfe->type = (void *)cft;
2756                cfe->dentry = dentry;
2757                dentry->d_fsdata = cfe;
2758                list_add_tail(&cfe->node, &parent->files);
2759                cfe = NULL;
2760        }
2761        dput(dentry);
2762out:
2763        kfree(cfe);
2764        return error;
2765}
2766
2767static int cgroup_addrm_files(struct cgroup *cgrp, struct cgroup_subsys *subsys,
2768                              struct cftype cfts[], bool is_add)
2769{
2770        struct cftype *cft;
2771        int err, ret = 0;
2772
2773        for (cft = cfts; cft->name[0] != '\0'; cft++) {
2774                /* does cft->flags tell us to skip this file on @cgrp? */
2775                if ((cft->flags & CFTYPE_NOT_ON_ROOT) && !cgrp->parent)
2776                        continue;
2777                if ((cft->flags & CFTYPE_ONLY_ON_ROOT) && cgrp->parent)
2778                        continue;
2779
2780                if (is_add)
2781                        err = cgroup_add_file(cgrp, subsys, cft);
2782                else
2783                        err = cgroup_rm_file(cgrp, cft);
2784                if (err) {
2785                        pr_warning("cgroup_addrm_files: failed to %s %s, err=%d\n",
2786                                   is_add ? "add" : "remove", cft->name, err);
2787                        ret = err;
2788                }
2789        }
2790        return ret;
2791}
2792
2793static DEFINE_MUTEX(cgroup_cft_mutex);
2794
2795static void cgroup_cfts_prepare(void)
2796        __acquires(&cgroup_cft_mutex) __acquires(&cgroup_mutex)
2797{
2798        /*
2799         * Thanks to the entanglement with vfs inode locking, we can't walk
2800         * the existing cgroups under cgroup_mutex and create files.
2801         * Instead, we increment reference on all cgroups and build list of
2802         * them using @cgrp->cft_q_node.  Grab cgroup_cft_mutex to ensure
2803         * exclusive access to the field.
2804         */
2805        mutex_lock(&cgroup_cft_mutex);
2806        mutex_lock(&cgroup_mutex);
2807}
2808
2809static void cgroup_cfts_commit(struct cgroup_subsys *ss,
2810                               struct cftype *cfts, bool is_add)
2811        __releases(&cgroup_mutex) __releases(&cgroup_cft_mutex)
2812{
2813        LIST_HEAD(pending);
2814        struct cgroup *cgrp, *n;
2815
2816        /* %NULL @cfts indicates abort and don't bother if @ss isn't attached */
2817        if (cfts && ss->root != &rootnode) {
2818                list_for_each_entry(cgrp, &ss->root->allcg_list, allcg_node) {
2819                        dget(cgrp->dentry);
2820                        list_add_tail(&cgrp->cft_q_node, &pending);
2821                }
2822        }
2823
2824        mutex_unlock(&cgroup_mutex);
2825
2826        /*
2827         * All new cgroups will see @cfts update on @ss->cftsets.  Add/rm
2828         * files for all cgroups which were created before.
2829         */
2830        list_for_each_entry_safe(cgrp, n, &pending, cft_q_node) {
2831                struct inode *inode = cgrp->dentry->d_inode;
2832
2833                mutex_lock(&inode->i_mutex);
2834                mutex_lock(&cgroup_mutex);
2835                if (!cgroup_is_removed(cgrp))
2836                        cgroup_addrm_files(cgrp, ss, cfts, is_add);
2837                mutex_unlock(&cgroup_mutex);
2838                mutex_unlock(&inode->i_mutex);
2839
2840                list_del_init(&cgrp->cft_q_node);
2841                dput(cgrp->dentry);
2842        }
2843
2844        mutex_unlock(&cgroup_cft_mutex);
2845}
2846
2847/**
2848 * cgroup_add_cftypes - add an array of cftypes to a subsystem
2849 * @ss: target cgroup subsystem
2850 * @cfts: zero-length name terminated array of cftypes
2851 *
2852 * Register @cfts to @ss.  Files described by @cfts are created for all
2853 * existing cgroups to which @ss is attached and all future cgroups will
2854 * have them too.  This function can be called anytime whether @ss is
2855 * attached or not.
2856 *
2857 * Returns 0 on successful registration, -errno on failure.  Note that this
2858 * function currently returns 0 as long as @cfts registration is successful
2859 * even if some file creation attempts on existing cgroups fail.
2860 */
2861int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
2862{
2863        struct cftype_set *set;
2864
2865        set = kzalloc(sizeof(*set), GFP_KERNEL);
2866        if (!set)
2867                return -ENOMEM;
2868
2869        cgroup_cfts_prepare();
2870        set->cfts = cfts;
2871        list_add_tail(&set->node, &ss->cftsets);
2872        cgroup_cfts_commit(ss, cfts, true);
2873
2874        return 0;
2875}
2876EXPORT_SYMBOL_GPL(cgroup_add_cftypes);
2877
2878/**
2879 * cgroup_rm_cftypes - remove an array of cftypes from a subsystem
2880 * @ss: target cgroup subsystem
2881 * @cfts: zero-length name terminated array of cftypes
2882 *
2883 * Unregister @cfts from @ss.  Files described by @cfts are removed from
2884 * all existing cgroups to which @ss is attached and all future cgroups
2885 * won't have them either.  This function can be called anytime whether @ss
2886 * is attached or not.
2887 *
2888 * Returns 0 on successful unregistration, -ENOENT if @cfts is not
2889 * registered with @ss.
2890 */
2891int cgroup_rm_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
2892{
2893        struct cftype_set *set;
2894
2895        cgroup_cfts_prepare();
2896
2897        list_for_each_entry(set, &ss->cftsets, node) {
2898                if (set->cfts == cfts) {
2899                        list_del_init(&set->node);
2900                        cgroup_cfts_commit(ss, cfts, false);
2901                        return 0;
2902                }
2903        }
2904
2905        cgroup_cfts_commit(ss, NULL, false);
2906        return -ENOENT;
2907}
2908
2909/**
2910 * cgroup_task_count - count the number of tasks in a cgroup.
2911 * @cgrp: the cgroup in question
2912 *
2913 * Return the number of tasks in the cgroup.
2914 */
2915int cgroup_task_count(const struct cgroup *cgrp)
2916{
2917        int count = 0;
2918        struct cg_cgroup_link *link;
2919
2920        read_lock(&css_set_lock);
2921        list_for_each_entry(link, &cgrp->css_sets, cgrp_link_list) {
2922                count += atomic_read(&link->cg->refcount);
2923        }
2924        read_unlock(&css_set_lock);
2925        return count;
2926}
2927
2928/*
2929 * Advance a list_head iterator.  The iterator should be positioned at
2930 * the start of a css_set
2931 */
2932static void cgroup_advance_iter(struct cgroup *cgrp,
2933                                struct cgroup_iter *it)
2934{
2935        struct list_head *l = it->cg_link;
2936        struct cg_cgroup_link *link;
2937        struct css_set *cg;
2938
2939        /* Advance to the next non-empty css_set */
2940        do {
2941                l = l->next;
2942                if (l == &cgrp->css_sets) {
2943                        it->cg_link = NULL;
2944                        return;
2945                }
2946                link = list_entry(l, struct cg_cgroup_link, cgrp_link_list);
2947                cg = link->cg;
2948        } while (list_empty(&cg->tasks));
2949        it->cg_link = l;
2950        it->task = cg->tasks.next;
2951}
2952
2953/*
2954 * To reduce the fork() overhead for systems that are not actually
2955 * using their cgroups capability, we don't maintain the lists running
2956 * through each css_set to its tasks until we see the list actually
2957 * used - in other words after the first call to cgroup_iter_start().
2958 */
2959static void cgroup_enable_task_cg_lists(void)
2960{
2961        struct task_struct *p, *g;
2962        write_lock(&css_set_lock);
2963        use_task_css_set_links = 1;
2964        /*
2965         * We need tasklist_lock because RCU is not safe against
2966         * while_each_thread(). Besides, a forking task that has passed
2967         * cgroup_post_fork() without seeing use_task_css_set_links = 1
2968         * is not guaranteed to have its child immediately visible in the
2969         * tasklist if we walk through it with RCU.
2970         */
2971        read_lock(&tasklist_lock);
2972        do_each_thread(g, p) {
2973                task_lock(p);
2974                /*
2975                 * We should check if the process is exiting, otherwise
2976                 * it will race with cgroup_exit() in that the list
2977                 * entry won't be deleted though the process has exited.
2978                 */
2979                if (!(p->flags & PF_EXITING) && list_empty(&p->cg_list))
2980                        list_add(&p->cg_list, &p->cgroups->tasks);
2981                task_unlock(p);
2982        } while_each_thread(g, p);
2983        read_unlock(&tasklist_lock);
2984        write_unlock(&css_set_lock);
2985}
2986
2987/**
2988 * cgroup_next_descendant_pre - find the next descendant for pre-order walk
2989 * @pos: the current position (%NULL to initiate traversal)
2990 * @cgroup: cgroup whose descendants to walk
2991 *
2992 * To be used by cgroup_for_each_descendant_pre().  Find the next
2993 * descendant to visit for pre-order traversal of @cgroup's descendants.
2994 */
2995struct cgroup *cgroup_next_descendant_pre(struct cgroup *pos,
2996                                          struct cgroup *cgroup)
2997{
2998        struct cgroup *next;
2999
3000        WARN_ON_ONCE(!rcu_read_lock_held());
3001
3002        /* if first iteration, pretend we just visited @cgroup */
3003        if (!pos) {
3004                if (list_empty(&cgroup->children))
3005                        return NULL;
3006                pos = cgroup;
3007        }
3008
3009        /* visit the first child if exists */
3010        next = list_first_or_null_rcu(&pos->children, struct cgroup, sibling);
3011        if (next)
3012                return next;
3013
3014        /* no child, visit my or the closest ancestor's next sibling */
3015        do {
3016                next = list_entry_rcu(pos->sibling.next, struct cgroup,
3017                                      sibling);
3018                if (&next->sibling != &pos->parent->children)
3019                        return next;
3020
3021                pos = pos->parent;
3022        } while (pos != cgroup);
3023
3024        return NULL;
3025}
3026EXPORT_SYMBOL_GPL(cgroup_next_descendant_pre);
3027
3028static struct cgroup *cgroup_leftmost_descendant(struct cgroup *pos)
3029{
3030        struct cgroup *last;
3031
3032        do {
3033                last = pos;
3034                pos = list_first_or_null_rcu(&pos->children, struct cgroup,
3035                                             sibling);
3036        } while (pos);
3037
3038        return last;
3039}
3040
3041/**
3042 * cgroup_next_descendant_post - find the next descendant for post-order walk
3043 * @pos: the current position (%NULL to initiate traversal)
3044 * @cgroup: cgroup whose descendants to walk
3045 *
3046 * To be used by cgroup_for_each_descendant_post().  Find the next
3047 * descendant to visit for post-order traversal of @cgroup's descendants.
3048 */
3049struct cgroup *cgroup_next_descendant_post(struct cgroup *pos,
3050                                           struct cgroup *cgroup)
3051{
3052        struct cgroup *next;
3053
3054        WARN_ON_ONCE(!rcu_read_lock_held());
3055
3056        /* if first iteration, visit the leftmost descendant */
3057        if (!pos) {
3058                next = cgroup_leftmost_descendant(cgroup);
3059                return next != cgroup ? next : NULL;
3060        }
3061
3062        /* if there's an unvisited sibling, visit its leftmost descendant */
3063        next = list_entry_rcu(pos->sibling.next, struct cgroup, sibling);
3064        if (&next->sibling != &pos->parent->children)
3065                return cgroup_leftmost_descendant(next);
3066
3067        /* no sibling left, visit parent */
3068        next = pos->parent;
3069        return next != cgroup ? next : NULL;
3070}
3071EXPORT_SYMBOL_GPL(cgroup_next_descendant_post);
3072
3073void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it)
3074        __acquires(css_set_lock)
3075{
3076        /*
3077         * The first time anyone tries to iterate across a cgroup,
3078         * we need to enable the list linking each css_set to its
3079         * tasks, and fix up all existing tasks.
3080         */
3081        if (!use_task_css_set_links)
3082                cgroup_enable_task_cg_lists();
3083
3084        read_lock(&css_set_lock);
3085        it->cg_link = &cgrp->css_sets;
3086        cgroup_advance_iter(cgrp, it);
3087}
3088
3089struct task_struct *cgroup_iter_next(struct cgroup *cgrp,
3090                                        struct cgroup_iter *it)
3091{
3092        struct task_struct *res;
3093        struct list_head *l = it->task;
3094        struct cg_cgroup_link *link;
3095
3096        /* If the iterator cg is NULL, we have no tasks */
3097        if (!it->cg_link)
3098                return NULL;
3099        res = list_entry(l, struct task_struct, cg_list);
3100        /* Advance iterator to find next entry */
3101        l = l->next;
3102        link = list_entry(it->cg_link, struct cg_cgroup_link, cgrp_link_list);
3103        if (l == &link->cg->tasks) {
3104                /* We reached the end of this task list - move on to
3105                 * the next cg_cgroup_link */
3106                cgroup_advance_iter(cgrp, it);
3107        } else {
3108                it->task = l;
3109        }
3110        return res;
3111}
3112
3113void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it)
3114        __releases(css_set_lock)
3115{
3116        read_unlock(&css_set_lock);
3117}
3118
3119static inline int started_after_time(struct task_struct *t1,
3120                                     struct timespec *time,
3121                                     struct task_struct *t2)
3122{
3123        int start_diff = timespec_compare(&t1->start_time, time);
3124        if (start_diff > 0) {
3125                return 1;
3126        } else if (start_diff < 0) {
3127                return 0;
3128        } else {
3129                /*
3130                 * Arbitrarily, if two processes started at the same
3131                 * time, we'll say that the lower pointer value
3132                 * started first. Note that t2 may have exited by now
3133                 * so this may not be a valid pointer any longer, but
3134                 * that's fine - it still serves to distinguish
3135                 * between two tasks started (effectively) simultaneously.
3136                 */
3137                return t1 > t2;
3138        }
3139}
3140
3141/*
3142 * This function is a callback from heap_insert() and is used to order
3143 * the heap.
3144 * In this case we order the heap in descending task start time.
3145 */
3146static inline int started_after(void *p1, void *p2)
3147{
3148        struct task_struct *t1 = p1;
3149        struct task_struct *t2 = p2;
3150        return started_after_time(t1, &t2->start_time, t2);
3151}
3152
3153/**
3154 * cgroup_scan_tasks - iterate though all the tasks in a cgroup
3155 * @scan: struct cgroup_scanner containing arguments for the scan
3156 *
3157 * Arguments include pointers to callback functions test_task() and
3158 * process_task().
3159 * Iterate through all the tasks in a cgroup, calling test_task() for each,
3160 * and if it returns true, call process_task() for it also.
3161 * The test_task pointer may be NULL, meaning always true (select all tasks).
3162 * Effectively duplicates cgroup_iter_{start,next,end}()
3163 * but does not lock css_set_lock for the call to process_task().
3164 * The struct cgroup_scanner may be embedded in any structure of the caller's
3165 * creation.
3166 * It is guaranteed that process_task() will act on every task that
3167 * is a member of the cgroup for the duration of this call. This
3168 * function may or may not call process_task() for tasks that exit
3169 * or move to a different cgroup during the call, or are forked or
3170 * move into the cgroup during the call.
3171 *
3172 * Note that test_task() may be called with locks held, and may in some
3173 * situations be called multiple times for the same task, so it should
3174 * be cheap.
3175 * If the heap pointer in the struct cgroup_scanner is non-NULL, a heap has been
3176 * pre-allocated and will be used for heap operations (and its "gt" member will
3177 * be overwritten), else a temporary heap will be used (allocation of which
3178 * may cause this function to fail).
3179 */
3180int cgroup_scan_tasks(struct cgroup_scanner *scan)
3181{
3182        int retval, i;
3183        struct cgroup_iter it;
3184        struct task_struct *p, *dropped;
3185        /* Never dereference latest_task, since it's not refcounted */
3186        struct task_struct *latest_task = NULL;
3187        struct ptr_heap tmp_heap;
3188        struct ptr_heap *heap;
3189        struct timespec latest_time = { 0, 0 };
3190
3191        if (scan->heap) {
3192                /* The caller supplied our heap and pre-allocated its memory */
3193                heap = scan->heap;
3194                heap->gt = &started_after;
3195        } else {
3196                /* We need to allocate our own heap memory */
3197                heap = &tmp_heap;
3198                retval = heap_init(heap, PAGE_SIZE, GFP_KERNEL, &started_after);
3199                if (retval)
3200                        /* cannot allocate the heap */
3201                        return retval;
3202        }
3203
3204 again:
3205        /*
3206         * Scan tasks in the cgroup, using the scanner's "test_task" callback
3207         * to determine which are of interest, and using the scanner's
3208         * "process_task" callback to process any of them that need an update.
3209         * Since we don't want to hold any locks during the task updates,
3210         * gather tasks to be processed in a heap structure.
3211         * The heap is sorted by descending task start time.
3212         * If the statically-sized heap fills up, we overflow tasks that
3213         * started later, and in future iterations only consider tasks that
3214         * started after the latest task in the previous pass. This
3215         * guarantees forward progress and that we don't miss any tasks.
3216         */
3217        heap->size = 0;
3218        cgroup_iter_start(scan->cg, &it);
3219        while ((p = cgroup_iter_next(scan->cg, &it))) {
3220                /*
3221                 * Only affect tasks that qualify per the caller's callback,
3222                 * if he provided one
3223                 */
3224                if (scan->test_task && !scan->test_task(p, scan))
3225                        continue;
3226                /*
3227                 * Only process tasks that started after the last task
3228                 * we processed
3229                 */
3230                if (!started_after_time(p, &latest_time, latest_task))
3231                        continue;
3232                dropped = heap_insert(heap, p);
3233                if (dropped == NULL) {
3234                        /*
3235                         * The new task was inserted; the heap wasn't
3236                         * previously full
3237                         */
3238                        get_task_struct(p);
3239                } else if (dropped != p) {
3240                        /*
3241                         * The new task was inserted, and pushed out a
3242                         * different task
3243                         */
3244                        get_task_struct(p);
3245                        put_task_struct(dropped);
3246                }
3247                /*
3248                 * Else the new task was newer than anything already in
3249                 * the heap and wasn't inserted
3250                 */
3251        }
3252        cgroup_iter_end(scan->cg, &it);
3253
3254        if (heap->size) {
3255                for (i = 0; i < heap->size; i++) {
3256                        struct task_struct *q = heap->ptrs[i];
3257                        if (i == 0) {
3258                                latest_time = q->start_time;
3259                                latest_task = q;
3260                        }
3261                        /* Process the task per the caller's callback */
3262                        scan->process_task(q, scan);
3263                        put_task_struct(q);
3264                }
3265                /*
3266                 * If we had to process any tasks at all, scan again
3267                 * in case some of them were in the middle of forking
3268                 * children that didn't get processed.
3269                 * Not the most efficient way to do it, but it avoids
3270                 * having to take callback_mutex in the fork path
3271                 */
3272                goto again;
3273        }
3274        if (heap == &tmp_heap)
3275                heap_free(&tmp_heap);
3276        return 0;
3277}
3278
3279/*
3280 * Stuff for reading the 'tasks'/'procs' files.
3281 *
3282 * Reading this file can return large amounts of data if a cgroup has
3283 * *lots* of attached tasks. So it may need several calls to read(),
3284 * but we cannot guarantee that the information we produce is correct
3285 * unless we produce it entirely atomically.
3286 *
3287 */
3288
3289/* which pidlist file are we talking about? */
3290enum cgroup_filetype {
3291        CGROUP_FILE_PROCS,
3292        CGROUP_FILE_TASKS,
3293};
3294
3295/*
3296 * A pidlist is a list of pids that virtually represents the contents of one
3297 * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists,
3298 * a pair (one each for procs, tasks) for each pid namespace that's relevant
3299 * to the cgroup.
3300 */
3301struct cgroup_pidlist {
3302        /*
3303         * used to find which pidlist is wanted. doesn't change as long as
3304         * this particular list stays in the list.
3305        */
3306        struct { enum cgroup_filetype type; struct pid_namespace *ns; } key;
3307        /* array of xids */
3308        pid_t *list;
3309        /* how many elements the above list has */
3310        int length;
3311        /* how many files are using the current array */
3312        int use_count;
3313        /* each of these stored in a list by its cgroup */
3314        struct list_head links;
3315        /* pointer to the cgroup we belong to, for list removal purposes */
3316        struct cgroup *owner;
3317        /* protects the other fields */
3318        struct rw_semaphore mutex;
3319};
3320
3321/*
3322 * The following two functions "fix" the issue where there are more pids
3323 * than kmalloc will give memory for; in such cases, we use vmalloc/vfree.
3324 * TODO: replace with a kernel-wide solution to this problem
3325 */
3326#define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2))
3327static void *pidlist_allocate(int count)
3328{
3329        if (PIDLIST_TOO_LARGE(count))
3330                return vmalloc(count * sizeof(pid_t));
3331        else
3332                return kmalloc(count * sizeof(pid_t), GFP_KERNEL);
3333}
3334static void pidlist_free(void *p)
3335{
3336        if (is_vmalloc_addr(p))
3337                vfree(p);
3338        else
3339                kfree(p);
3340}
3341static void *pidlist_resize(void *p, int newcount)
3342{
3343        void *newlist;
3344        /* note: if new alloc fails, old p will still be valid either way */
3345        if (is_vmalloc_addr(p)) {
3346                newlist = vmalloc(newcount * sizeof(pid_t));
3347                if (!newlist)
3348                        return NULL;
3349                memcpy(newlist, p, newcount * sizeof(pid_t));
3350                vfree(p);
3351        } else {
3352                newlist = krealloc(p, newcount * sizeof(pid_t), GFP_KERNEL);
3353        }
3354        return newlist;
3355}
3356
3357/*
3358 * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
3359 * If the new stripped list is sufficiently smaller and there's enough memory
3360 * to allocate a new buffer, will let go of the unneeded memory. Returns the
3361 * number of unique elements.
3362 */
3363/* is the size difference enough that we should re-allocate the array? */
3364#define PIDLIST_REALLOC_DIFFERENCE(old, new) ((old) - PAGE_SIZE >= (new))
3365static int pidlist_uniq(pid_t **p, int length)
3366{
3367        int src, dest = 1;
3368        pid_t *list = *p;
3369        pid_t *newlist;
3370
3371        /*
3372         * we presume the 0th element is unique, so i starts at 1. trivial
3373         * edge cases first; no work needs to be done for either
3374         */
3375        if (length == 0 || length == 1)
3376                return length;
3377        /* src and dest walk down the list; dest counts unique elements */
3378        for (src = 1; src < length; src++) {
3379                /* find next unique element */
3380                while (list[src] == list[src-1]) {
3381                        src++;
3382                        if (src == length)
3383                                goto after;
3384                }
3385                /* dest always points to where the next unique element goes */
3386                list[dest] = list[src];
3387                dest++;
3388        }
3389after:
3390        /*
3391         * if the length difference is large enough, we want to allocate a
3392         * smaller buffer to save memory. if this fails due to out of memory,
3393         * we'll just stay with what we've got.
3394         */
3395        if (PIDLIST_REALLOC_DIFFERENCE(length, dest)) {
3396                newlist = pidlist_resize(list, dest);
3397                if (newlist)
3398                        *p = newlist;
3399        }
3400        return dest;
3401}
3402
3403static int cmppid(const void *a, const void *b)
3404{
3405        return *(pid_t *)a - *(pid_t *)b;
3406}
3407
3408/*
3409 * find the appropriate pidlist for our purpose (given procs vs tasks)
3410 * returns with the lock on that pidlist already held, and takes care
3411 * of the use count, or returns NULL with no locks held if we're out of
3412 * memory.
3413 */
3414static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp,
3415                                                  enum cgroup_filetype type)
3416{
3417        struct cgroup_pidlist *l;
3418        /* don't need task_nsproxy() if we're looking at ourself */
3419        struct pid_namespace *ns = task_active_pid_ns(current);
3420
3421        /*
3422         * We can't drop the pidlist_mutex before taking the l->mutex in case
3423         * the last ref-holder is trying to remove l from the list at the same
3424         * time. Holding the pidlist_mutex precludes somebody taking whichever
3425         * list we find out from under us - compare release_pid_array().
3426         */
3427        mutex_lock(&cgrp->pidlist_mutex);
3428        list_for_each_entry(l, &cgrp->pidlists, links) {
3429                if (l->key.type == type && l->key.ns == ns) {
3430                        /* make sure l doesn't vanish out from under us */
3431                        down_write(&l->mutex);
3432                        mutex_unlock(&cgrp->pidlist_mutex);
3433                        return l;
3434                }
3435        }
3436        /* entry not found; create a new one */
3437        l = kmalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL);
3438        if (!l) {
3439                mutex_unlock(&cgrp->pidlist_mutex);
3440                return l;
3441        }
3442        init_rwsem(&l->mutex);
3443        down_write(&l->mutex);
3444        l->key.type = type;
3445        l->key.ns = get_pid_ns(ns);
3446        l->use_count = 0; /* don't increment here */
3447        l->list = NULL;
3448        l->owner = cgrp;
3449        list_add(&l->links, &cgrp->pidlists);
3450        mutex_unlock(&cgrp->pidlist_mutex);
3451        return l;
3452}
3453
3454/*
3455 * Load a cgroup's pidarray with either procs' tgids or tasks' pids
3456 */
3457static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type,
3458                              struct cgroup_pidlist **lp)
3459{
3460        pid_t *array;
3461        int length;
3462        int pid, n = 0; /* used for populating the array */
3463        struct cgroup_iter it;
3464        struct task_struct *tsk;
3465        struct cgroup_pidlist *l;
3466
3467        /*
3468         * If cgroup gets more users after we read count, we won't have
3469         * enough space - tough.  This race is indistinguishable to the
3470         * caller from the case that the additional cgroup users didn't
3471         * show up until sometime later on.
3472         */
3473        length = cgroup_task_count(cgrp);
3474        array = pidlist_allocate(length);
3475        if (!array)
3476                return -ENOMEM;
3477        /* now, populate the array */
3478        cgroup_iter_start(cgrp, &it);
3479        while ((tsk = cgroup_iter_next(cgrp, &it))) {
3480                if (unlikely(n == length))
3481                        break;
3482                /* get tgid or pid for procs or tasks file respectively */
3483                if (type == CGROUP_FILE_PROCS)
3484                        pid = task_tgid_vnr(tsk);
3485                else
3486                        pid = task_pid_vnr(tsk);
3487                if (pid > 0) /* make sure to only use valid results */
3488                        array[n++] = pid;
3489        }
3490        cgroup_iter_end(cgrp, &it);
3491        length = n;
3492        /* now sort & (if procs) strip out duplicates */
3493        sort(array, length, sizeof(pid_t), cmppid, NULL);
3494        if (type == CGROUP_FILE_PROCS)
3495                length = pidlist_uniq(&array, length);
3496        l = cgroup_pidlist_find(cgrp, type);
3497        if (!l) {
3498                pidlist_free(array);
3499                return -ENOMEM;
3500        }
3501        /* store array, freeing old if necessary - lock already held */
3502        pidlist_free(l->list);
3503        l->list = array;
3504        l->length = length;
3505        l->use_count++;
3506        up_write(&l->mutex);
3507        *lp = l;
3508        return 0;
3509}
3510
3511/**
3512 * cgroupstats_build - build and fill cgroupstats
3513 * @stats: cgroupstats to fill information into
3514 * @dentry: A dentry entry belonging to the cgroup for which stats have
3515 * been requested.
3516 *
3517 * Build and fill cgroupstats so that taskstats can export it to user
3518 * space.
3519 */
3520int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry)
3521{
3522        int ret = -EINVAL;
3523        struct cgroup *cgrp;
3524        struct cgroup_iter it;
3525        struct task_struct *tsk;
3526
3527        /*
3528         * Validate dentry by checking the superblock operations,
3529         * and make sure it's a directory.
3530         */
3531        if (dentry->d_sb->s_op != &cgroup_ops ||
3532            !S_ISDIR(dentry->d_inode->i_mode))
3533                 goto err;
3534
3535        ret = 0;
3536        cgrp = dentry->d_fsdata;
3537
3538        cgroup_iter_start(cgrp, &it);
3539        while ((tsk = cgroup_iter_next(cgrp, &it))) {
3540                switch (tsk->state) {
3541                case TASK_RUNNING:
3542                        stats->nr_running++;
3543                        break;
3544                case TASK_INTERRUPTIBLE:
3545                        stats->nr_sleeping++;
3546                        break;
3547                case TASK_UNINTERRUPTIBLE:
3548                        stats->nr_uninterruptible++;
3549                        break;
3550                case TASK_STOPPED:
3551                        stats->nr_stopped++;
3552                        break;
3553                default:
3554                        if (delayacct_is_task_waiting_on_io(tsk))
3555                                stats->nr_io_wait++;
3556                        break;
3557                }
3558        }
3559        cgroup_iter_end(cgrp, &it);
3560
3561err:
3562        return ret;
3563}
3564
3565
3566/*
3567 * seq_file methods for the tasks/procs files. The seq_file position is the
3568 * next pid to display; the seq_file iterator is a pointer to the pid
3569 * in the cgroup->l->list array.
3570 */
3571
3572static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos)
3573{
3574        /*
3575         * Initially we receive a position value that corresponds to
3576         * one more than the last pid shown (or 0 on the first call or
3577         * after a seek to the start). Use a binary-search to find the
3578         * next pid to display, if any
3579         */
3580        struct cgroup_pidlist *l = s->private;
3581        int index = 0, pid = *pos;
3582        int *iter;
3583
3584        down_read(&l->mutex);
3585        if (pid) {
3586                int end = l->length;
3587
3588                while (index < end) {
3589                        int mid = (index + end) / 2;
3590                        if (l->list[mid] == pid) {
3591                                index = mid;
3592                                break;
3593                        } else if (l->list[mid] <= pid)
3594                                index = mid + 1;
3595                        else
3596                                end = mid;
3597                }
3598        }
3599        /* If we're off the end of the array, we're done */
3600        if (index >= l->length)
3601                return NULL;
3602        /* Update the abstract position to be the actual pid that we found */
3603        iter = l->list + index;
3604        *pos = *iter;
3605        return iter;
3606}
3607
3608static void cgroup_pidlist_stop(struct seq_file *s, void *v)
3609{
3610        struct cgroup_pidlist *l = s->private;
3611        up_read(&l->mutex);
3612}
3613
3614static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos)
3615{
3616        struct cgroup_pidlist *l = s->private;
3617        pid_t *p = v;
3618        pid_t *end = l->list + l->length;
3619        /*
3620         * Advance to the next pid in the array. If this goes off the
3621         * end, we're done
3622         */
3623        p++;
3624        if (p >= end) {
3625                return NULL;
3626        } else {
3627                *pos = *p;
3628                return p;
3629        }
3630}
3631
3632static int cgroup_pidlist_show(struct seq_file *s, void *v)
3633{
3634        return seq_printf(s, "%d\n", *(int *)v);
3635}
3636
3637/*
3638 * seq_operations functions for iterating on pidlists through seq_file -
3639 * independent of whether it's tasks or procs
3640 */
3641static const struct seq_operations cgroup_pidlist_seq_operations = {
3642        .start = cgroup_pidlist_start,
3643        .stop = cgroup_pidlist_stop,
3644        .next = cgroup_pidlist_next,
3645        .show = cgroup_pidlist_show,
3646};
3647
3648static void cgroup_release_pid_array(struct cgroup_pidlist *l)
3649{
3650        /*
3651         * the case where we're the last user of this particular pidlist will
3652         * have us remove it from the cgroup's list, which entails taking the
3653         * mutex. since in pidlist_find the pidlist->lock depends on cgroup->
3654         * pidlist_mutex, we have to take pidlist_mutex first.
3655         */
3656        mutex_lock(&l->owner->pidlist_mutex);
3657        down_write(&l->mutex);
3658        BUG_ON(!l->use_count);
3659        if (!--l->use_count) {
3660                /* we're the last user if refcount is 0; remove and free */
3661                list_del(&l->links);
3662                mutex_unlock(&l->owner->pidlist_mutex);
3663                pidlist_free(l->list);
3664                put_pid_ns(l->key.ns);
3665                up_write(&l->mutex);
3666                kfree(l);
3667                return;
3668        }
3669        mutex_unlock(&l->owner->pidlist_mutex);
3670        up_write(&l->mutex);
3671}
3672
3673static int cgroup_pidlist_release(struct inode *inode, struct file *file)
3674{
3675        struct cgroup_pidlist *l;
3676        if (!(file->f_mode & FMODE_READ))
3677                return 0;
3678        /*
3679         * the seq_file will only be initialized if the file was opened for
3680         * reading; hence we check if it's not null only in that case.
3681         */
3682        l = ((struct seq_file *)file->private_data)->private;
3683        cgroup_release_pid_array(l);
3684        return seq_release(inode, file);
3685}
3686
3687static const struct file_operations cgroup_pidlist_operations = {
3688        .read = seq_read,
3689        .llseek = seq_lseek,
3690        .write = cgroup_file_write,
3691        .release = cgroup_pidlist_release,
3692};
3693
3694/*
3695 * The following functions handle opens on a file that displays a pidlist
3696 * (tasks or procs). Prepare an array of the process/thread IDs of whoever's
3697 * in the cgroup.
3698 */
3699/* helper function for the two below it */
3700static int cgroup_pidlist_open(struct file *file, enum cgroup_filetype type)
3701{
3702        struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
3703        struct cgroup_pidlist *l;
3704        int retval;
3705
3706        /* Nothing to do for write-only files */
3707        if (!(file->f_mode & FMODE_READ))
3708                return 0;
3709
3710        /* have the array populated */
3711        retval = pidlist_array_load(cgrp, type, &l);
3712        if (retval)
3713                return retval;
3714        /* configure file information */
3715        file->f_op = &cgroup_pidlist_operations;
3716
3717        retval = seq_open(file, &cgroup_pidlist_seq_operations);
3718        if (retval) {
3719                cgroup_release_pid_array(l);
3720                return retval;
3721        }
3722        ((struct seq_file *)file->private_data)->private = l;
3723        return 0;
3724}
3725static int cgroup_tasks_open(struct inode *unused, struct file *file)
3726{
3727        return cgroup_pidlist_open(file, CGROUP_FILE_TASKS);
3728}
3729static int cgroup_procs_open(struct inode *unused, struct file *file)
3730{
3731        return cgroup_pidlist_open(file, CGROUP_FILE_PROCS);
3732}
3733
3734static u64 cgroup_read_notify_on_release(struct cgroup *cgrp,
3735                                            struct cftype *cft)
3736{
3737        return notify_on_release(cgrp);
3738}
3739
3740static int cgroup_write_notify_on_release(struct cgroup *cgrp,
3741                                          struct cftype *cft,
3742                                          u64 val)
3743{
3744        clear_bit(CGRP_RELEASABLE, &cgrp->flags);
3745        if (val)
3746                set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
3747        else
3748                clear_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
3749        return 0;
3750}
3751
3752/*
3753 * Unregister event and free resources.
3754 *
3755 * Gets called from workqueue.
3756 */
3757static void cgroup_event_remove(struct work_struct *work)
3758{
3759        struct cgroup_event *event = container_of(work, struct cgroup_event,
3760                        remove);
3761        struct cgroup *cgrp = event->cgrp;
3762
3763        event->cft->unregister_event(cgrp, event->cft, event->eventfd);
3764
3765        eventfd_ctx_put(event->eventfd);
3766        kfree(event);
3767        dput(cgrp->dentry);
3768}
3769
3770/*
3771 * Gets called on POLLHUP on eventfd when user closes it.
3772 *
3773 * Called with wqh->lock held and interrupts disabled.
3774 */
3775static int cgroup_event_wake(wait_queue_t *wait, unsigned mode,
3776                int sync, void *key)
3777{
3778        struct cgroup_event *event = container_of(wait,
3779                        struct cgroup_event, wait);
3780        struct cgroup *cgrp = event->cgrp;
3781        unsigned long flags = (unsigned long)key;
3782
3783        if (flags & POLLHUP) {
3784                __remove_wait_queue(event->wqh, &event->wait);
3785                spin_lock(&cgrp->event_list_lock);
3786                list_del_init(&event->list);
3787                spin_unlock(&cgrp->event_list_lock);
3788                /*
3789                 * We are in atomic context, but cgroup_event_remove() may
3790                 * sleep, so we have to call it in workqueue.
3791                 */
3792                schedule_work(&event->remove);
3793        }
3794
3795        return 0;
3796}
3797
3798static void cgroup_event_ptable_queue_proc(struct file *file,
3799                wait_queue_head_t *wqh, poll_table *pt)
3800{
3801        struct cgroup_event *event = container_of(pt,
3802                        struct cgroup_event, pt);
3803
3804        event->wqh = wqh;
3805        add_wait_queue(wqh, &event->wait);
3806}
3807
3808/*
3809 * Parse input and register new cgroup event handler.
3810 *
3811 * Input must be in format '<event_fd> <control_fd> <args>'.
3812 * Interpretation of args is defined by control file implementation.
3813 */
3814static int cgroup_write_event_control(struct cgroup *cgrp, struct cftype *cft,
3815                                      const char *buffer)
3816{
3817        struct cgroup_event *event = NULL;
3818        unsigned int efd, cfd;
3819        struct file *efile = NULL;
3820        struct file *cfile = NULL;
3821        char *endp;
3822        int ret;
3823
3824        efd = simple_strtoul(buffer, &endp, 10);
3825        if (*endp != ' ')
3826                return -EINVAL;
3827        buffer = endp + 1;
3828
3829        cfd = simple_strtoul(buffer, &endp, 10);
3830        if ((*endp != ' ') && (*endp != '\0'))
3831                return -EINVAL;
3832        buffer = endp + 1;
3833
3834        event = kzalloc(sizeof(*event), GFP_KERNEL);
3835        if (!event)
3836                return -ENOMEM;
3837        event->cgrp = cgrp;
3838        INIT_LIST_HEAD(&event->list);
3839        init_poll_funcptr(&event->pt, cgroup_event_ptable_queue_proc);
3840        init_waitqueue_func_entry(&event->wait, cgroup_event_wake);
3841        INIT_WORK(&event->remove, cgroup_event_remove);
3842
3843        efile = eventfd_fget(efd);
3844        if (IS_ERR(efile)) {
3845                ret = PTR_ERR(efile);
3846                goto fail;
3847        }
3848
3849        event->eventfd = eventfd_ctx_fileget(efile);
3850        if (IS_ERR(event->eventfd)) {
3851                ret = PTR_ERR(event->eventfd);
3852                goto fail;
3853        }
3854
3855        cfile = fget(cfd);
3856        if (!cfile) {
3857                ret = -EBADF;
3858                goto fail;
3859        }
3860
3861        /* the process need read permission on control file */
3862        /* AV: shouldn't we check that it's been opened for read instead? */
3863        ret = inode_permission(cfile->f_path.dentry->d_inode, MAY_READ);
3864        if (ret < 0)
3865                goto fail;
3866
3867        event->cft = __file_cft(cfile);
3868        if (IS_ERR(event->cft)) {
3869                ret = PTR_ERR(event->cft);
3870                goto fail;
3871        }
3872
3873        if (!event->cft->register_event || !event->cft->unregister_event) {
3874                ret = -EINVAL;
3875                goto fail;
3876        }
3877
3878        ret = event->cft->register_event(cgrp, event->cft,
3879                        event->eventfd, buffer);
3880        if (ret)
3881                goto fail;
3882
3883        if (efile->f_op->poll(efile, &event->pt) & POLLHUP) {
3884                event->cft->unregister_event(cgrp, event->cft, event->eventfd);
3885                ret = 0;
3886                goto fail;
3887        }
3888
3889        /*
3890         * Events should be removed after rmdir of cgroup directory, but before
3891         * destroying subsystem state objects. Let's take reference to cgroup
3892         * directory dentry to do that.
3893         */
3894        dget(cgrp->dentry);
3895
3896        spin_lock(&cgrp->event_list_lock);
3897        list_add(&event->list, &cgrp->event_list);
3898        spin_unlock(&cgrp->event_list_lock);
3899
3900        fput(cfile);
3901        fput(efile);
3902
3903        return 0;
3904
3905fail:
3906        if (cfile)
3907                fput(cfile);
3908
3909        if (event && event->eventfd && !IS_ERR(event->eventfd))
3910                eventfd_ctx_put(event->eventfd);
3911
3912        if (!IS_ERR_OR_NULL(efile))
3913                fput(efile);
3914
3915        kfree(event);
3916
3917        return ret;
3918}
3919
3920static u64 cgroup_clone_children_read(struct cgroup *cgrp,
3921                                    struct cftype *cft)
3922{
3923        return test_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags);
3924}
3925
3926static int cgroup_clone_children_write(struct cgroup *cgrp,
3927                                     struct cftype *cft,
3928                                     u64 val)
3929{
3930        if (val)
3931                set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags);
3932        else
3933                clear_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags);
3934        return 0;
3935}
3936
3937/*
3938 * for the common functions, 'private' gives the type of file
3939 */
3940/* for hysterical raisins, we can't put this on the older files */
3941#define CGROUP_FILE_GENERIC_PREFIX "cgroup."
3942static struct cftype files[] = {
3943        {
3944                .name = "tasks",
3945                .open = cgroup_tasks_open,
3946                .write_u64 = cgroup_tasks_write,
3947                .release = cgroup_pidlist_release,
3948                .mode = S_IRUGO | S_IWUSR,
3949        },
3950        {
3951                .name = CGROUP_FILE_GENERIC_PREFIX "procs",
3952                .open = cgroup_procs_open,
3953                .write_u64 = cgroup_procs_write,
3954                .release = cgroup_pidlist_release,
3955                .mode = S_IRUGO | S_IWUSR,
3956        },
3957        {
3958                .name = "notify_on_release",
3959                .read_u64 = cgroup_read_notify_on_release,
3960                .write_u64 = cgroup_write_notify_on_release,
3961        },
3962        {
3963                .name = CGROUP_FILE_GENERIC_PREFIX "event_control",
3964                .write_string = cgroup_write_event_control,
3965                .mode = S_IWUGO,
3966        },
3967        {
3968                .name = "cgroup.clone_children",
3969                .read_u64 = cgroup_clone_children_read,
3970                .write_u64 = cgroup_clone_children_write,
3971        },
3972        {
3973                .name = "release_agent",
3974                .flags = CFTYPE_ONLY_ON_ROOT,
3975                .read_seq_string = cgroup_release_agent_show,
3976                .write_string = cgroup_release_agent_write,
3977                .max_write_len = PATH_MAX,
3978        },
3979        { }     /* terminate */
3980};
3981
3982/**
3983 * cgroup_populate_dir - selectively creation of files in a directory
3984 * @cgrp: target cgroup
3985 * @base_files: true if the base files should be added
3986 * @subsys_mask: mask of the subsystem ids whose files should be added
3987 */
3988static int cgroup_populate_dir(struct cgroup *cgrp, bool base_files,
3989                               unsigned long subsys_mask)
3990{
3991        int err;
3992        struct cgroup_subsys *ss;
3993
3994        if (base_files) {
3995                err = cgroup_addrm_files(cgrp, NULL, files, true);
3996                if (err < 0)
3997                        return err;
3998        }
3999
4000        /* process cftsets of each subsystem */
4001        for_each_subsys(cgrp->root, ss) {
4002                struct cftype_set *set;
4003                if (!test_bit(ss->subsys_id, &subsys_mask))
4004                        continue;
4005
4006                list_for_each_entry(set, &ss->cftsets, node)
4007                        cgroup_addrm_files(cgrp, ss, set->cfts, true);
4008        }
4009
4010        /* This cgroup is ready now */
4011        for_each_subsys(cgrp->root, ss) {
4012                struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
4013                /*
4014                 * Update id->css pointer and make this css visible from
4015                 * CSS ID functions. This pointer will be dereferened
4016                 * from RCU-read-side without locks.
4017                 */
4018                if (css->id)
4019                        rcu_assign_pointer(css->id->css, css);
4020        }
4021
4022        return 0;
4023}
4024
4025static void css_dput_fn(struct work_struct *work)
4026{
4027        struct cgroup_subsys_state *css =
4028                container_of(work, struct cgroup_subsys_state, dput_work);
4029        struct dentry *dentry = css->cgroup->dentry;
4030        struct super_block *sb = dentry->d_sb;
4031
4032        atomic_inc(&sb->s_active);
4033        dput(dentry);
4034        deactivate_super(sb);
4035}
4036
4037static void init_cgroup_css(struct cgroup_subsys_state *css,
4038                               struct cgroup_subsys *ss,
4039                               struct cgroup *cgrp)
4040{
4041        css->cgroup = cgrp;
4042        atomic_set(&css->refcnt, 1);
4043        css->flags = 0;
4044        css->id = NULL;
4045        if (cgrp == dummytop)
4046                css->flags |= CSS_ROOT;
4047        BUG_ON(cgrp->subsys[ss->subsys_id]);
4048        cgrp->subsys[ss->subsys_id] = css;
4049
4050        /*
4051         * css holds an extra ref to @cgrp->dentry which is put on the last
4052         * css_put().  dput() requires process context, which css_put() may
4053         * be called without.  @css->dput_work will be used to invoke
4054         * dput() asynchronously from css_put().
4055         */
4056        INIT_WORK(&css->dput_work, css_dput_fn);
4057}
4058
4059/* invoke ->post_create() on a new CSS and mark it online if successful */
4060static int online_css(struct cgroup_subsys *ss, struct cgroup *cgrp)
4061{
4062        int ret = 0;
4063
4064        lockdep_assert_held(&cgroup_mutex);
4065
4066        if (ss->css_online)
4067                ret = ss->css_online(cgrp);
4068        if (!ret)
4069                cgrp->subsys[ss->subsys_id]->flags |= CSS_ONLINE;
4070        return ret;
4071}
4072
4073/* if the CSS is online, invoke ->pre_destory() on it and mark it offline */
4074static void offline_css(struct cgroup_subsys *ss, struct cgroup *cgrp)
4075        __releases(&cgroup_mutex) __acquires(&cgroup_mutex)
4076{
4077        struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
4078
4079        lockdep_assert_held(&cgroup_mutex);
4080
4081        if (!(css->flags & CSS_ONLINE))
4082                return;
4083
4084        /*
4085         * css_offline() should be called with cgroup_mutex unlocked.  See
4086         * 3fa59dfbc3 ("cgroup: fix potential deadlock in pre_destroy") for
4087         * details.  This temporary unlocking should go away once
4088         * cgroup_mutex is unexported from controllers.
4089         */
4090        if (ss->css_offline) {
4091                mutex_unlock(&cgroup_mutex);
4092                ss->css_offline(cgrp);
4093                mutex_lock(&cgroup_mutex);
4094        }
4095
4096        cgrp->subsys[ss->subsys_id]->flags &= ~CSS_ONLINE;
4097}
4098
4099/*
4100 * cgroup_create - create a cgroup
4101 * @parent: cgroup that will be parent of the new cgroup
4102 * @dentry: dentry of the new cgroup
4103 * @mode: mode to set on new inode
4104 *
4105 * Must be called with the mutex on the parent inode held
4106 */
4107static long cgroup_create(struct cgroup *parent, struct dentry *dentry,
4108                             umode_t mode)
4109{
4110        struct cgroup *cgrp;
4111        struct cgroupfs_root *root = parent->root;
4112        int err = 0;
4113        struct cgroup_subsys *ss;
4114        struct super_block *sb = root->sb;
4115
4116        /* allocate the cgroup and its ID, 0 is reserved for the root */
4117        cgrp = kzalloc(sizeof(*cgrp), GFP_KERNEL);
4118        if (!cgrp)
4119                return -ENOMEM;
4120
4121        cgrp->id = ida_simple_get(&root->cgroup_ida, 1, 0, GFP_KERNEL);
4122        if (cgrp->id < 0)
4123                goto err_free_cgrp;
4124
4125        /*
4126         * Only live parents can have children.  Note that the liveliness
4127         * check isn't strictly necessary because cgroup_mkdir() and
4128         * cgroup_rmdir() are fully synchronized by i_mutex; however, do it
4129         * anyway so that locking is contained inside cgroup proper and we
4130         * don't get nasty surprises if we ever grow another caller.
4131         */
4132        if (!cgroup_lock_live_group(parent)) {
4133                err = -ENODEV;
4134                goto err_free_id;
4135        }
4136
4137        /* Grab a reference on the superblock so the hierarchy doesn't
4138         * get deleted on unmount if there are child cgroups.  This
4139         * can be done outside cgroup_mutex, since the sb can't
4140         * disappear while someone has an open control file on the
4141         * fs */
4142        atomic_inc(&sb->s_active);
4143
4144        init_cgroup_housekeeping(cgrp);
4145
4146        cgrp->parent = parent;
4147        cgrp->root = parent->root;
4148        cgrp->top_cgroup = parent->top_cgroup;
4149
4150        if (notify_on_release(parent))
4151                set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
4152
4153        if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &parent->flags))
4154                set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags);
4155
4156        for_each_subsys(root, ss) {
4157                struct cgroup_subsys_state *css;
4158
4159                css = ss->css_alloc(cgrp);
4160                if (IS_ERR(css)) {
4161                        err = PTR_ERR(css);
4162                        goto err_free_all;
4163                }
4164                init_cgroup_css(css, ss, cgrp);
4165                if (ss->use_id) {
4166                        err = alloc_css_id(ss, parent, cgrp);
4167                        if (err)
4168                                goto err_free_all;
4169                }
4170        }
4171
4172        /*
4173         * Create directory.  cgroup_create_file() returns with the new
4174         * directory locked on success so that it can be populated without
4175         * dropping cgroup_mutex.
4176         */
4177        err = cgroup_create_file(dentry, S_IFDIR | mode, sb);
4178        if (err < 0)
4179                goto err_free_all;
4180        lockdep_assert_held(&dentry->d_inode->i_mutex);
4181
4182        /* allocation complete, commit to creation */
4183        dentry->d_fsdata = cgrp;
4184        cgrp->dentry = dentry;
4185        list_add_tail(&cgrp->allcg_node, &root->allcg_list);
4186        list_add_tail_rcu(&cgrp->sibling, &cgrp->parent->children);
4187        root->number_of_cgroups++;
4188
4189        /* each css holds a ref to the cgroup's dentry */
4190        for_each_subsys(root, ss)
4191                dget(dentry);
4192
4193        /* creation succeeded, notify subsystems */
4194        for_each_subsys(root, ss) {
4195                err = online_css(ss, cgrp);
4196                if (err)
4197                        goto err_destroy;
4198
4199                if (ss->broken_hierarchy && !ss->warned_broken_hierarchy &&
4200                    parent->parent) {
4201                        pr_warning("cgroup: %s (%d) created nested cgroup for controller \"%s\" which has incomplete hierarchy support. Nested cgroups may change behavior in the future.\n",
4202                                   current->comm, current->pid, ss->name);
4203                        if (!strcmp(ss->name, "memory"))
4204                                pr_warning("cgroup: \"memory\" requires setting use_hierarchy to 1 on the root.\n");
4205                        ss->warned_broken_hierarchy = true;
4206                }
4207        }
4208
4209        err = cgroup_populate_dir(cgrp, true, root->subsys_mask);
4210        if (err)
4211                goto err_destroy;
4212
4213        mutex_unlock(&cgroup_mutex);
4214        mutex_unlock(&cgrp->dentry->d_inode->i_mutex);
4215
4216        return 0;
4217
4218err_free_all:
4219        for_each_subsys(root, ss) {
4220                if (cgrp->subsys[ss->subsys_id])
4221                        ss->css_free(cgrp);
4222        }
4223        mutex_unlock(&cgroup_mutex);
4224        /* Release the reference count that we took on the superblock */
4225        deactivate_super(sb);
4226err_free_id:
4227        ida_simple_remove(&root->cgroup_ida, cgrp->id);
4228err_free_cgrp:
4229        kfree(cgrp);
4230        return err;
4231
4232err_destroy:
4233        cgroup_destroy_locked(cgrp);
4234        mutex_unlock(&cgroup_mutex);
4235        mutex_unlock(&dentry->d_inode->i_mutex);
4236        return err;
4237}
4238
4239static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
4240{
4241        struct cgroup *c_parent = dentry->d_parent->d_fsdata;
4242
4243        /* the vfs holds inode->i_mutex already */
4244        return cgroup_create(c_parent, dentry, mode | S_IFDIR);
4245}
4246
4247/*
4248 * Check the reference count on each subsystem. Since we already
4249 * established that there are no tasks in the cgroup, if the css refcount
4250 * is also 1, then there should be no outstanding references, so the
4251 * subsystem is safe to destroy. We scan across all subsystems rather than
4252 * using the per-hierarchy linked list of mounted subsystems since we can
4253 * be called via check_for_release() with no synchronization other than
4254 * RCU, and the subsystem linked list isn't RCU-safe.
4255 */
4256static int cgroup_has_css_refs(struct cgroup *cgrp)
4257{
4258        int i;
4259
4260        /*
4261         * We won't need to lock the subsys array, because the subsystems
4262         * we're concerned about aren't going anywhere since our cgroup root
4263         * has a reference on them.
4264         */
4265        for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
4266                struct cgroup_subsys *ss = subsys[i];
4267                struct cgroup_subsys_state *css;
4268
4269                /* Skip subsystems not present or not in this hierarchy */
4270                if (ss == NULL || ss->root != cgrp->root)
4271                        continue;
4272
4273                css = cgrp->subsys[ss->subsys_id];
4274                /*
4275                 * When called from check_for_release() it's possible
4276                 * that by this point the cgroup has been removed
4277                 * and the css deleted. But a false-positive doesn't
4278                 * matter, since it can only happen if the cgroup
4279                 * has been deleted and hence no longer needs the
4280                 * release agent to be called anyway.
4281                 */
4282                if (css && css_refcnt(css) > 1)
4283                        return 1;
4284        }
4285        return 0;
4286}
4287
4288static int cgroup_destroy_locked(struct cgroup *cgrp)
4289        __releases(&cgroup_mutex) __acquires(&cgroup_mutex)
4290{
4291        struct dentry *d = cgrp->dentry;
4292        struct cgroup *parent = cgrp->parent;
4293        DEFINE_WAIT(wait);
4294        struct cgroup_event *event, *tmp;
4295        struct cgroup_subsys *ss;
4296        LIST_HEAD(tmp_list);
4297
4298        lockdep_assert_held(&d->d_inode->i_mutex);
4299        lockdep_assert_held(&cgroup_mutex);
4300
4301        if (atomic_read(&cgrp->count) || !list_empty(&cgrp->children))
4302                return -EBUSY;
4303
4304        /*
4305         * Block new css_tryget() by deactivating refcnt and mark @cgrp
4306         * removed.  This makes future css_tryget() and child creation
4307         * attempts fail thus maintaining the removal conditions verified
4308         * above.
4309         */
4310        for_each_subsys(cgrp->root, ss) {
4311                struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
4312
4313                WARN_ON(atomic_read(&css->refcnt) < 0);
4314                atomic_add(CSS_DEACT_BIAS, &css->refcnt);
4315        }
4316        set_bit(CGRP_REMOVED, &cgrp->flags);
4317
4318        /* tell subsystems to initate destruction */
4319        for_each_subsys(cgrp->root, ss)
4320                offline_css(ss, cgrp);
4321
4322        /*
4323         * Put all the base refs.  Each css holds an extra reference to the
4324         * cgroup's dentry and cgroup removal proceeds regardless of css
4325         * refs.  On the last put of each css, whenever that may be, the
4326         * extra dentry ref is put so that dentry destruction happens only
4327         * after all css's are released.
4328         */
4329        for_each_subsys(cgrp->root, ss)
4330                css_put(cgrp->subsys[ss->subsys_id]);
4331
4332        raw_spin_lock(&release_list_lock);
4333        if (!list_empty(&cgrp->release_list))
4334                list_del_init(&cgrp->release_list);
4335        raw_spin_unlock(&release_list_lock);
4336
4337        /* delete this cgroup from parent->children */
4338        list_del_rcu(&cgrp->sibling);
4339        list_del_init(&cgrp->allcg_node);
4340
4341        dget(d);
4342        cgroup_d_remove_dir(d);
4343        dput(d);
4344
4345        set_bit(CGRP_RELEASABLE, &parent->flags);
4346        check_for_release(parent);
4347
4348        /*
4349         * Unregister events and notify userspace.
4350         * Notify userspace about cgroup removing only after rmdir of cgroup
4351         * directory to avoid race between userspace and kernelspace. Use
4352         * a temporary list to avoid a deadlock with cgroup_event_wake(). Since
4353         * cgroup_event_wake() is called with the wait queue head locked,
4354         * remove_wait_queue() cannot be called while holding event_list_lock.
4355         */
4356        spin_lock(&cgrp->event_list_lock);
4357        list_splice_init(&cgrp->event_list, &tmp_list);
4358        spin_unlock(&cgrp->event_list_lock);
4359        list_for_each_entry_safe(event, tmp, &tmp_list, list) {
4360                list_del_init(&event->list);
4361                remove_wait_queue(event->wqh, &event->wait);
4362                eventfd_signal(event->eventfd, 1);
4363                schedule_work(&event->remove);
4364        }
4365
4366        return 0;
4367}
4368
4369static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry)
4370{
4371        int ret;
4372
4373        mutex_lock(&cgroup_mutex);
4374        ret = cgroup_destroy_locked(dentry->d_fsdata);
4375        mutex_unlock(&cgroup_mutex);
4376
4377        return ret;
4378}
4379
4380static void __init_or_module cgroup_init_cftsets(struct cgroup_subsys *ss)
4381{
4382        INIT_LIST_HEAD(&ss->cftsets);
4383
4384        /*
4385         * base_cftset is embedded in subsys itself, no need to worry about
4386         * deregistration.
4387         */
4388        if (ss->base_cftypes) {
4389                ss->base_cftset.cfts = ss->base_cftypes;
4390                list_add_tail(&ss->base_cftset.node, &ss->cftsets);
4391        }
4392}
4393
4394static void __init cgroup_init_subsys(struct cgroup_subsys *ss)
4395{
4396        struct cgroup_subsys_state *css;
4397
4398        printk(KERN_INFO "Initializing cgroup subsys %s\n", ss->name);
4399
4400        mutex_lock(&cgroup_mutex);
4401
4402        /* init base cftset */
4403        cgroup_init_cftsets(ss);
4404
4405        /* Create the top cgroup state for this subsystem */
4406        list_add(&ss->sibling, &rootnode.subsys_list);
4407        ss->root = &rootnode;
4408        css = ss->css_alloc(dummytop);
4409        /* We don't handle early failures gracefully */
4410        BUG_ON(IS_ERR(css));
4411        init_cgroup_css(css, ss, dummytop);
4412
4413        /* Update the init_css_set to contain a subsys
4414         * pointer to this state - since the subsystem is
4415         * newly registered, all tasks and hence the
4416         * init_css_set is in the subsystem's top cgroup. */
4417        init_css_set.subsys[ss->subsys_id] = css;
4418
4419        need_forkexit_callback |= ss->fork || ss->exit;
4420
4421        /* At system boot, before all subsystems have been
4422         * registered, no tasks have been forked, so we don't
4423         * need to invoke fork callbacks here. */
4424        BUG_ON(!list_empty(&init_task.tasks));
4425
4426        ss->active = 1;
4427        BUG_ON(online_css(ss, dummytop));
4428
4429        mutex_unlock(&cgroup_mutex);
4430
4431        /* this function shouldn't be used with modular subsystems, since they
4432         * need to register a subsys_id, among other things */
4433        BUG_ON(ss->module);
4434}
4435
4436/**
4437 * cgroup_load_subsys: load and register a modular subsystem at runtime
4438 * @ss: the subsystem to load
4439 *
4440 * This function should be called in a modular subsystem's initcall. If the
4441 * subsystem is built as a module, it will be assigned a new subsys_id and set
4442 * up for use. If the subsystem is built-in anyway, work is delegated to the
4443 * simpler cgroup_init_subsys.
4444 */
4445int __init_or_module cgroup_load_subsys(struct cgroup_subsys *ss)
4446{
4447        struct cgroup_subsys_state *css;
4448        int i, ret;
4449
4450        /* check name and function validity */
4451        if (ss->name == NULL || strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN ||
4452            ss->css_alloc == NULL || ss->css_free == NULL)
4453                return -EINVAL;
4454
4455        /*
4456         * we don't support callbacks in modular subsystems. this check is
4457         * before the ss->module check for consistency; a subsystem that could
4458         * be a module should still have no callbacks even if the user isn't
4459         * compiling it as one.
4460         */
4461        if (ss->fork || ss->exit)
4462                return -EINVAL;
4463
4464        /*
4465         * an optionally modular subsystem is built-in: we want to do nothing,
4466         * since cgroup_init_subsys will have already taken care of it.
4467         */
4468        if (ss->module == NULL) {
4469                /* a sanity check */
4470                BUG_ON(subsys[ss->subsys_id] != ss);
4471                return 0;
4472        }
4473
4474        /* init base cftset */
4475        cgroup_init_cftsets(ss);
4476
4477        mutex_lock(&cgroup_mutex);
4478        subsys[ss->subsys_id] = ss;
4479
4480        /*
4481         * no ss->css_alloc seems to need anything important in the ss
4482         * struct, so this can happen first (i.e. before the rootnode
4483         * attachment).
4484         */
4485        css = ss->css_alloc(dummytop);
4486        if (IS_ERR(css)) {
4487                /* failure case - need to deassign the subsys[] slot. */
4488                subsys[ss->subsys_id] = NULL;
4489                mutex_unlock(&cgroup_mutex);
4490                return PTR_ERR(css);
4491        }
4492
4493        list_add(&ss->sibling, &rootnode.subsys_list);
4494        ss->root = &rootnode;
4495
4496        /* our new subsystem will be attached to the dummy hierarchy. */
4497        init_cgroup_css(css, ss, dummytop);
4498        /* init_idr must be after init_cgroup_css because it sets css->id. */
4499        if (ss->use_id) {
4500                ret = cgroup_init_idr(ss, css);
4501                if (ret)
4502                        goto err_unload;
4503        }
4504
4505        /*
4506         * Now we need to entangle the css into the existing css_sets. unlike
4507         * in cgroup_init_subsys, there are now multiple css_sets, so each one
4508         * will need a new pointer to it; done by iterating the css_set_table.
4509         * furthermore, modifying the existing css_sets will corrupt the hash
4510         * table state, so each changed css_set will need its hash recomputed.
4511         * this is all done under the css_set_lock.
4512         */
4513        write_lock(&css_set_lock);
4514        for (i = 0; i < CSS_SET_TABLE_SIZE; i++) {
4515                struct css_set *cg;
4516                struct hlist_node *node, *tmp;
4517                struct hlist_head *bucket = &css_set_table[i], *new_bucket;
4518
4519                hlist_for_each_entry_safe(cg, node, tmp, bucket, hlist) {
4520                        /* skip entries that we already rehashed */
4521                        if (cg->subsys[ss->subsys_id])
4522                                continue;
4523                        /* remove existing entry */
4524                        hlist_del(&cg->hlist);
4525                        /* set new value */
4526                        cg->subsys[ss->subsys_id] = css;
4527                        /* recompute hash and restore entry */
4528                        new_bucket = css_set_hash(cg->subsys);
4529                        hlist_add_head(&cg->hlist, new_bucket);
4530                }
4531        }
4532        write_unlock(&css_set_lock);
4533
4534        ss->active = 1;
4535        ret = online_css(ss, dummytop);
4536        if (ret)
4537                goto err_unload;
4538
4539        /* success! */
4540        mutex_unlock(&cgroup_mutex);
4541        return 0;
4542
4543err_unload:
4544        mutex_unlock(&cgroup_mutex);
4545        /* @ss can't be mounted here as try_module_get() would fail */
4546        cgroup_unload_subsys(ss);
4547        return ret;
4548}
4549EXPORT_SYMBOL_GPL(cgroup_load_subsys);
4550
4551/**
4552 * cgroup_unload_subsys: unload a modular subsystem
4553 * @ss: the subsystem to unload
4554 *
4555 * This function should be called in a modular subsystem's exitcall. When this
4556 * function is invoked, the refcount on the subsystem's module will be 0, so
4557 * the subsystem will not be attached to any hierarchy.
4558 */
4559void cgroup_unload_subsys(struct cgroup_subsys *ss)
4560{
4561        struct cg_cgroup_link *link;
4562        struct hlist_head *hhead;
4563
4564        BUG_ON(ss->module == NULL);
4565
4566        /*
4567         * we shouldn't be called if the subsystem is in use, and the use of
4568         * try_module_get in parse_cgroupfs_options should ensure that it
4569         * doesn't start being used while we're killing it off.
4570         */
4571        BUG_ON(ss->root != &rootnode);
4572
4573        mutex_lock(&cgroup_mutex);
4574
4575        offline_css(ss, dummytop);
4576        ss->active = 0;
4577
4578        if (ss->use_id) {
4579                idr_remove_all(&ss->idr);
4580                idr_destroy(&ss->idr);
4581        }
4582
4583        /* deassign the subsys_id */
4584        subsys[ss->subsys_id] = NULL;
4585
4586        /* remove subsystem from rootnode's list of subsystems */
4587        list_del_init(&ss->sibling);
4588
4589        /*
4590         * disentangle the css from all css_sets attached to the dummytop. as
4591         * in loading, we need to pay our respects to the hashtable gods.
4592         */
4593        write_lock(&css_set_lock);
4594        list_for_each_entry(link, &dummytop->css_sets, cgrp_link_list) {
4595                struct css_set *cg = link->cg;
4596
4597                hlist_del(&cg->hlist);
4598                cg->subsys[ss->subsys_id] = NULL;
4599                hhead = css_set_hash(cg->subsys);
4600                hlist_add_head(&cg->hlist, hhead);
4601        }
4602        write_unlock(&css_set_lock);
4603
4604        /*
4605         * remove subsystem's css from the dummytop and free it - need to
4606         * free before marking as null because ss->css_free needs the
4607         * cgrp->subsys pointer to find their state. note that this also
4608         * takes care of freeing the css_id.
4609         */
4610        ss->css_free(dummytop);
4611        dummytop->subsys[ss->subsys_id] = NULL;
4612
4613        mutex_unlock(&cgroup_mutex);
4614}
4615EXPORT_SYMBOL_GPL(cgroup_unload_subsys);
4616
4617/**
4618 * cgroup_init_early - cgroup initialization at system boot
4619 *
4620 * Initialize cgroups at system boot, and initialize any
4621 * subsystems that request early init.
4622 */
4623int __init cgroup_init_early(void)
4624{
4625        int i;
4626        atomic_set(&init_css_set.refcount, 1);
4627        INIT_LIST_HEAD(&init_css_set.cg_links);
4628        INIT_LIST_HEAD(&init_css_set.tasks);
4629        INIT_HLIST_NODE(&init_css_set.hlist);
4630        css_set_count = 1;
4631        init_cgroup_root(&rootnode);
4632        root_count = 1;
4633        init_task.cgroups = &init_css_set;
4634
4635        init_css_set_link.cg = &init_css_set;
4636        init_css_set_link.cgrp = dummytop;
4637        list_add(&init_css_set_link.cgrp_link_list,
4638                 &rootnode.top_cgroup.css_sets);
4639        list_add(&init_css_set_link.cg_link_list,
4640                 &init_css_set.cg_links);
4641
4642        for (i = 0; i < CSS_SET_TABLE_SIZE; i++)
4643                INIT_HLIST_HEAD(&css_set_table[i]);
4644
4645        for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
4646                struct cgroup_subsys *ss = subsys[i];
4647
4648                /* at bootup time, we don't worry about modular subsystems */
4649                if (!ss || ss->module)
4650                        continue;
4651
4652                BUG_ON(!ss->name);
4653                BUG_ON(strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN);
4654                BUG_ON(!ss->css_alloc);
4655                BUG_ON(!ss->css_free);
4656                if (ss->subsys_id != i) {
4657                        printk(KERN_ERR "cgroup: Subsys %s id == %d\n",
4658                               ss->name, ss->subsys_id);
4659                        BUG();
4660                }
4661
4662                if (ss->early_init)
4663                        cgroup_init_subsys(ss);
4664        }
4665        return 0;
4666}
4667
4668/**
4669 * cgroup_init - cgroup initialization
4670 *
4671 * Register cgroup filesystem and /proc file, and initialize
4672 * any subsystems that didn't request early init.
4673 */
4674int __init cgroup_init(void)
4675{
4676        int err;
4677        int i;
4678        struct hlist_head *hhead;
4679
4680        err = bdi_init(&cgroup_backing_dev_info);
4681        if (err)
4682                return err;
4683
4684        for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
4685                struct cgroup_subsys *ss = subsys[i];
4686
4687                /* at bootup time, we don't worry about modular subsystems */
4688                if (!ss || ss->module)
4689                        continue;
4690                if (!ss->early_init)
4691                        cgroup_init_subsys(ss);
4692                if (ss->use_id)
4693                        cgroup_init_idr(ss, init_css_set.subsys[ss->subsys_id]);
4694        }
4695
4696        /* Add init_css_set to the hash table */
4697        hhead = css_set_hash(init_css_set.subsys);
4698        hlist_add_head(&init_css_set.hlist, hhead);
4699        BUG_ON(!init_root_id(&rootnode));
4700
4701        cgroup_kobj = kobject_create_and_add("cgroup", fs_kobj);
4702        if (!cgroup_kobj) {
4703                err = -ENOMEM;
4704                goto out;
4705        }
4706
4707        err = register_filesystem(&cgroup_fs_type);
4708        if (err < 0) {
4709                kobject_put(cgroup_kobj);
4710                goto out;
4711        }
4712
4713        proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations);
4714
4715out:
4716        if (err)
4717                bdi_destroy(&cgroup_backing_dev_info);
4718
4719        return err;
4720}
4721
4722/*
4723 * proc_cgroup_show()
4724 *  - Print task's cgroup paths into seq_file, one line for each hierarchy
4725 *  - Used for /proc/<pid>/cgroup.
4726 *  - No need to task_lock(tsk) on this tsk->cgroup reference, as it
4727 *    doesn't really matter if tsk->cgroup changes after we read it,
4728 *    and we take cgroup_mutex, keeping cgroup_attach_task() from changing it
4729 *    anyway.  No need to check that tsk->cgroup != NULL, thanks to
4730 *    the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks
4731 *    cgroup to top_cgroup.
4732 */
4733
4734/* TODO: Use a proper seq_file iterator */
4735static int proc_cgroup_show(struct seq_file *m, void *v)
4736{
4737        struct pid *pid;
4738        struct task_struct *tsk;
4739        char *buf;
4740        int retval;
4741        struct cgroupfs_root *root;
4742
4743        retval = -ENOMEM;
4744        buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
4745        if (!buf)
4746                goto out;
4747
4748        retval = -ESRCH;
4749        pid = m->private;
4750        tsk = get_pid_task(pid, PIDTYPE_PID);
4751        if (!tsk)
4752                goto out_free;
4753
4754        retval = 0;
4755
4756        mutex_lock(&cgroup_mutex);
4757
4758        for_each_active_root(root) {
4759                struct cgroup_subsys *ss;
4760                struct cgroup *cgrp;
4761                int count = 0;
4762
4763                seq_printf(m, "%d:", root->hierarchy_id);
4764                for_each_subsys(root, ss)
4765                        seq_printf(m, "%s%s", count++ ? "," : "", ss->name);
4766                if (strlen(root->name))
4767                        seq_printf(m, "%sname=%s", count ? "," : "",
4768                                   root->name);
4769                seq_putc(m, ':');
4770                cgrp = task_cgroup_from_root(tsk, root);
4771                retval = cgroup_path(cgrp, buf, PAGE_SIZE);
4772                if (retval < 0)
4773                        goto out_unlock;
4774                seq_puts(m, buf);
4775                seq_putc(m, '\n');
4776        }
4777
4778out_unlock:
4779        mutex_unlock(&cgroup_mutex);
4780        put_task_struct(tsk);
4781out_free:
4782        kfree(buf);
4783out:
4784        return retval;
4785}
4786
4787static int cgroup_open(struct inode *inode, struct file *file)
4788{
4789        struct pid *pid = PROC_I(inode)->pid;
4790        return single_open(file, proc_cgroup_show, pid);
4791}
4792
4793const struct file_operations proc_cgroup_operations = {
4794        .open           = cgroup_open,
4795        .read           = seq_read,
4796        .llseek         = seq_lseek,
4797        .release        = single_release,
4798};
4799
4800/* Display information about each subsystem and each hierarchy */
4801static int proc_cgroupstats_show(struct seq_file *m, void *v)
4802{
4803        int i;
4804
4805        seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
4806        /*
4807         * ideally we don't want subsystems moving around while we do this.
4808         * cgroup_mutex is also necessary to guarantee an atomic snapshot of
4809         * subsys/hierarchy state.
4810         */
4811        mutex_lock(&cgroup_mutex);
4812        for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
4813                struct cgroup_subsys *ss = subsys[i];
4814                if (ss == NULL)
4815                        continue;
4816                seq_printf(m, "%s\t%d\t%d\t%d\n",
4817                           ss->name, ss->root->hierarchy_id,
4818                           ss->root->number_of_cgroups, !ss->disabled);
4819        }
4820        mutex_unlock(&cgroup_mutex);
4821        return 0;
4822}
4823
4824static int cgroupstats_open(struct inode *inode, struct file *file)
4825{
4826        return single_open(file, proc_cgroupstats_show, NULL);
4827}
4828
4829static const struct file_operations proc_cgroupstats_operations = {
4830        .open = cgroupstats_open,
4831        .read = seq_read,
4832        .llseek = seq_lseek,
4833        .release = single_release,
4834};
4835
4836/**
4837 * cgroup_fork - attach newly forked task to its parents cgroup.
4838 * @child: pointer to task_struct of forking parent process.
4839 *
4840 * Description: A task inherits its parent's cgroup at fork().
4841 *
4842 * A pointer to the shared css_set was automatically copied in
4843 * fork.c by dup_task_struct().  However, we ignore that copy, since
4844 * it was not made under the protection of RCU or cgroup_mutex, so
4845 * might no longer be a valid cgroup pointer.  cgroup_attach_task() might
4846 * have already changed current->cgroups, allowing the previously
4847 * referenced cgroup group to be removed and freed.
4848 *
4849 * At the point that cgroup_fork() is called, 'current' is the parent
4850 * task, and the passed argument 'child' points to the child task.
4851 */
4852void cgroup_fork(struct task_struct *child)
4853{
4854        task_lock(current);
4855        child->cgroups = current->cgroups;
4856        get_css_set(child->cgroups);
4857        task_unlock(current);
4858        INIT_LIST_HEAD(&child->cg_list);
4859}
4860
4861/**
4862 * cgroup_post_fork - called on a new task after adding it to the task list
4863 * @child: the task in question
4864 *
4865 * Adds the task to the list running through its css_set if necessary and
4866 * call the subsystem fork() callbacks.  Has to be after the task is
4867 * visible on the task list in case we race with the first call to
4868 * cgroup_iter_start() - to guarantee that the new task ends up on its
4869 * list.
4870 */
4871void cgroup_post_fork(struct task_struct *child)
4872{
4873        int i;
4874
4875        /*
4876         * use_task_css_set_links is set to 1 before we walk the tasklist
4877         * under the tasklist_lock and we read it here after we added the child
4878         * to the tasklist under the tasklist_lock as well. If the child wasn't
4879         * yet in the tasklist when we walked through it from
4880         * cgroup_enable_task_cg_lists(), then use_task_css_set_links value
4881         * should be visible now due to the paired locking and barriers implied
4882         * by LOCK/UNLOCK: it is written before the tasklist_lock unlock
4883         * in cgroup_enable_task_cg_lists() and read here after the tasklist_lock
4884         * lock on fork.
4885         */
4886        if (use_task_css_set_links) {
4887                write_lock(&css_set_lock);
4888                task_lock(child);
4889                if (list_empty(&child->cg_list))
4890                        list_add(&child->cg_list, &child->cgroups->tasks);
4891                task_unlock(child);
4892                write_unlock(&css_set_lock);
4893        }
4894
4895        /*
4896         * Call ss->fork().  This must happen after @child is linked on
4897         * css_set; otherwise, @child might change state between ->fork()
4898         * and addition to css_set.
4899         */
4900        if (need_forkexit_callback) {
4901                for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
4902                        struct cgroup_subsys *ss = subsys[i];
4903
4904                        /*
4905                         * fork/exit callbacks are supported only for
4906                         * builtin subsystems and we don't need further
4907                         * synchronization as they never go away.
4908                         */
4909                        if (!ss || ss->module)
4910                                continue;
4911
4912                        if (ss->fork)
4913                                ss->fork(child);
4914                }
4915        }
4916}
4917
4918/**
4919 * cgroup_exit - detach cgroup from exiting task
4920 * @tsk: pointer to task_struct of exiting process
4921 * @run_callback: run exit callbacks?
4922 *
4923 * Description: Detach cgroup from @tsk and release it.
4924 *
4925 * Note that cgroups marked notify_on_release force every task in
4926 * them to take the global cgroup_mutex mutex when exiting.
4927 * This could impact scaling on very large systems.  Be reluctant to
4928 * use notify_on_release cgroups where very high task exit scaling
4929 * is required on large systems.
4930 *
4931 * the_top_cgroup_hack:
4932 *
4933 *    Set the exiting tasks cgroup to the root cgroup (top_cgroup).
4934 *
4935 *    We call cgroup_exit() while the task is still competent to
4936 *    handle notify_on_release(), then leave the task attached to the
4937 *    root cgroup in each hierarchy for the remainder of its exit.
4938 *
4939 *    To do this properly, we would increment the reference count on
4940 *    top_cgroup, and near the very end of the kernel/exit.c do_exit()
4941 *    code we would add a second cgroup function call, to drop that
4942 *    reference.  This would just create an unnecessary hot spot on
4943 *    the top_cgroup reference count, to no avail.
4944 *
4945 *    Normally, holding a reference to a cgroup without bumping its
4946 *    count is unsafe.   The cgroup could go away, or someone could
4947 *    attach us to a different cgroup, decrementing the count on
4948 *    the first cgroup that we never incremented.  But in this case,
4949 *    top_cgroup isn't going away, and either task has PF_EXITING set,
4950 *    which wards off any cgroup_attach_task() attempts, or task is a failed
4951 *    fork, never visible to cgroup_attach_task.
4952 */
4953void cgroup_exit(struct task_struct *tsk, int run_callbacks)
4954{
4955        struct css_set *cg;
4956        int i;
4957
4958        /*
4959         * Unlink from the css_set task list if necessary.
4960         * Optimistically check cg_list before taking
4961         * css_set_lock
4962         */
4963        if (!list_empty(&tsk->cg_list)) {
4964                write_lock(&css_set_lock);
4965                if (!list_empty(&tsk->cg_list))
4966                        list_del_init(&tsk->cg_list);
4967                write_unlock(&css_set_lock);
4968        }
4969
4970        /* Reassign the task to the init_css_set. */
4971        task_lock(tsk);
4972        cg = tsk->cgroups;
4973        tsk->cgroups = &init_css_set;
4974
4975        if (run_callbacks && need_forkexit_callback) {
4976                for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
4977                        struct cgroup_subsys *ss = subsys[i];
4978
4979                        /* modular subsystems can't use callbacks */
4980                        if (!ss || ss->module)
4981                                continue;
4982
4983                        if (ss->exit) {
4984                                struct cgroup *old_cgrp =
4985                                        rcu_dereference_raw(cg->subsys[i])->cgroup;
4986                                struct cgroup *cgrp = task_cgroup(tsk, i);
4987                                ss->exit(cgrp, old_cgrp, tsk);
4988                        }
4989                }
4990        }
4991        task_unlock(tsk);
4992
4993        if (cg)
4994                put_css_set_taskexit(cg);
4995}
4996
4997/**
4998 * cgroup_is_descendant - see if @cgrp is a descendant of @task's cgrp
4999 * @cgrp: the cgroup in question
5000 * @task: the task in question
5001 *
5002 * See if @cgrp is a descendant of @task's cgroup in the appropriate
5003 * hierarchy.
5004 *
5005 * If we are sending in dummytop, then presumably we are creating
5006 * the top cgroup in the subsystem.
5007 *
5008 * Called only by the ns (nsproxy) cgroup.
5009 */
5010int cgroup_is_descendant(const struct cgroup *cgrp, struct task_struct *task)
5011{
5012        int ret;
5013        struct cgroup *target;
5014
5015        if (cgrp == dummytop)
5016                return 1;
5017
5018        target = task_cgroup_from_root(task, cgrp->root);
5019        while (cgrp != target && cgrp!= cgrp->top_cgroup)
5020                cgrp = cgrp->parent;
5021        ret = (cgrp == target);
5022        return ret;
5023}
5024
5025static void check_for_release(struct cgroup *cgrp)
5026{
5027        /* All of these checks rely on RCU to keep the cgroup
5028         * structure alive */
5029        if (cgroup_is_releasable(cgrp) && !atomic_read(&cgrp->count)
5030            && list_empty(&cgrp->children) && !cgroup_has_css_refs(cgrp)) {
5031                /* Control Group is currently removeable. If it's not
5032                 * already queued for a userspace notification, queue
5033                 * it now */
5034                int need_schedule_work = 0;
5035                raw_spin_lock(&release_list_lock);
5036                if (!cgroup_is_removed(cgrp) &&
5037                    list_empty(&cgrp->release_list)) {
5038                        list_add(&cgrp->release_list, &release_list);
5039                        need_schedule_work = 1;
5040                }
5041                raw_spin_unlock(&release_list_lock);
5042                if (need_schedule_work)
5043                        schedule_work(&release_agent_work);
5044        }
5045}
5046
5047/* Caller must verify that the css is not for root cgroup */
5048bool __css_tryget(struct cgroup_subsys_state *css)
5049{
5050        while (true) {
5051                int t, v;
5052
5053                v = css_refcnt(css);
5054                t = atomic_cmpxchg(&css->refcnt, v, v + 1);
5055                if (likely(t == v))
5056                        return true;
5057                else if (t < 0)
5058                        return false;
5059                cpu_relax();
5060        }
5061}
5062EXPORT_SYMBOL_GPL(__css_tryget);
5063
5064/* Caller must verify that the css is not for root cgroup */
5065void __css_put(struct cgroup_subsys_state *css)
5066{
5067        struct cgroup *cgrp = css->cgroup;
5068        int v;
5069
5070        rcu_read_lock();
5071        v = css_unbias_refcnt(atomic_dec_return(&css->refcnt));
5072
5073        switch (v) {
5074        case 1:
5075                if (notify_on_release(cgrp)) {
5076                        set_bit(CGRP_RELEASABLE, &cgrp->flags);
5077                        check_for_release(cgrp);
5078                }
5079                break;
5080        case 0:
5081                schedule_work(&css->dput_work);
5082                break;
5083        }
5084        rcu_read_unlock();
5085}
5086EXPORT_SYMBOL_GPL(__css_put);
5087
5088/*
5089 * Notify userspace when a cgroup is released, by running the
5090 * configured release agent with the name of the cgroup (path
5091 * relative to the root of cgroup file system) as the argument.
5092 *
5093 * Most likely, this user command will try to rmdir this cgroup.
5094 *
5095 * This races with the possibility that some other task will be
5096 * attached to this cgroup before it is removed, or that some other
5097 * user task will 'mkdir' a child cgroup of this cgroup.  That's ok.
5098 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
5099 * unused, and this cgroup will be reprieved from its death sentence,
5100 * to continue to serve a useful existence.  Next time it's released,
5101 * we will get notified again, if it still has 'notify_on_release' set.
5102 *
5103 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
5104 * means only wait until the task is successfully execve()'d.  The
5105 * separate release agent task is forked by call_usermodehelper(),
5106 * then control in this thread returns here, without waiting for the
5107 * release agent task.  We don't bother to wait because the caller of
5108 * this routine has no use for the exit status of the release agent
5109 * task, so no sense holding our caller up for that.
5110 */
5111static void cgroup_release_agent(struct work_struct *work)
5112{
5113        BUG_ON(work != &release_agent_work);
5114        mutex_lock(&cgroup_mutex);
5115        raw_spin_lock(&release_list_lock);
5116        while (!list_empty(&release_list)) {
5117                char *argv[3], *envp[3];
5118                int i;
5119                char *pathbuf = NULL, *agentbuf = NULL;
5120                struct cgroup *cgrp = list_entry(release_list.next,
5121                                                    struct cgroup,
5122                                                    release_list);
5123                list_del_init(&cgrp->release_list);
5124                raw_spin_unlock(&release_list_lock);
5125                pathbuf = kmalloc(PAGE_SIZE, GFP_KERNEL);
5126                if (!pathbuf)
5127                        goto continue_free;
5128                if (cgroup_path(cgrp, pathbuf, PAGE_SIZE) < 0)
5129                        goto continue_free;
5130                agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL);
5131                if (!agentbuf)
5132                        goto continue_free;
5133
5134                i = 0;
5135                argv[i++] = agentbuf;
5136                argv[i++] = pathbuf;
5137                argv[i] = NULL;
5138
5139                i = 0;
5140                /* minimal command environment */
5141                envp[i++] = "HOME=/";
5142                envp[i++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
5143                envp[i] = NULL;
5144
5145                /* Drop the lock while we invoke the usermode helper,
5146                 * since the exec could involve hitting disk and hence
5147                 * be a slow process */
5148                mutex_unlock(&cgroup_mutex);
5149                call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC);
5150                mutex_lock(&cgroup_mutex);
5151 continue_free:
5152                kfree(pathbuf);
5153                kfree(agentbuf);
5154                raw_spin_lock(&release_list_lock);
5155        }
5156        raw_spin_unlock(&release_list_lock);
5157        mutex_unlock(&cgroup_mutex);
5158}
5159
5160static int __init cgroup_disable(char *str)
5161{
5162        int i;
5163        char *token;
5164
5165        while ((token = strsep(&str, ",")) != NULL) {
5166                if (!*token)
5167                        continue;
5168                for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
5169                        struct cgroup_subsys *ss = subsys[i];
5170
5171                        /*
5172                         * cgroup_disable, being at boot time, can't
5173                         * know about module subsystems, so we don't
5174                         * worry about them.
5175                         */
5176                        if (!ss || ss->module)
5177                                continue;
5178
5179                        if (!strcmp(token, ss->name)) {
5180                                ss->disabled = 1;
5181                                printk(KERN_INFO "Disabling %s control group"
5182                                        " subsystem\n", ss->name);
5183                                break;
5184                        }
5185                }
5186        }
5187        return 1;
5188}
5189__setup("cgroup_disable=", cgroup_disable);
5190
5191/*
5192 * Functons for CSS ID.
5193 */
5194
5195/*
5196 *To get ID other than 0, this should be called when !cgroup_is_removed().
5197 */
5198unsigned short css_id(struct cgroup_subsys_state *css)
5199{
5200        struct css_id *cssid;
5201
5202        /*
5203         * This css_id() can return correct value when somone has refcnt
5204         * on this or this is under rcu_read_lock(). Once css->id is allocated,
5205         * it's unchanged until freed.
5206         */
5207        cssid = rcu_dereference_check(css->id, css_refcnt(css));
5208
5209        if (cssid)
5210                return cssid->id;
5211        return 0;
5212}
5213EXPORT_SYMBOL_GPL(css_id);
5214
5215unsigned short css_depth(struct cgroup_subsys_state *css)
5216{
5217        struct css_id *cssid;
5218
5219        cssid = rcu_dereference_check(css->id, css_refcnt(css));
5220
5221        if (cssid)
5222                return cssid->depth;
5223        return 0;
5224}
5225EXPORT_SYMBOL_GPL(css_depth);
5226
5227/**
5228 *  css_is_ancestor - test "root" css is an ancestor of "child"
5229 * @child: the css to be tested.
5230 * @root: the css supporsed to be an ancestor of the child.
5231 *
5232 * Returns true if "root" is an ancestor of "child" in its hierarchy. Because
5233 * this function reads css->id, the caller must hold rcu_read_lock().
5234 * But, considering usual usage, the csses should be valid objects after test.
5235 * Assuming that the caller will do some action to the child if this returns
5236 * returns true, the caller must take "child";s reference count.
5237 * If "child" is valid object and this returns true, "root" is valid, too.
5238 */
5239
5240bool css_is_ancestor(struct cgroup_subsys_state *child,
5241                    const struct cgroup_subsys_state *root)
5242{
5243        struct css_id *child_id;
5244        struct css_id *root_id;
5245
5246        child_id  = rcu_dereference(child->id);
5247        if (!child_id)
5248                return false;
5249        root_id = rcu_dereference(root->id);
5250        if (!root_id)
5251                return false;
5252        if (child_id->depth < root_id->depth)
5253                return false;
5254        if (child_id->stack[root_id->depth] != root_id->id)
5255                return false;
5256        return true;
5257}
5258
5259void free_css_id(struct cgroup_subsys *ss, struct cgroup_subsys_state *css)
5260{
5261        struct css_id *id = css->id;
5262        /* When this is called before css_id initialization, id can be NULL */
5263        if (!id)
5264                return;
5265
5266        BUG_ON(!ss->use_id);
5267
5268        rcu_assign_pointer(id->css, NULL);
5269        rcu_assign_pointer(css->id, NULL);
5270        spin_lock(&ss->id_lock);
5271        idr_remove(&ss->idr, id->id);
5272        spin_unlock(&ss->id_lock);
5273        kfree_rcu(id, rcu_head);
5274}
5275EXPORT_SYMBOL_GPL(free_css_id);
5276
5277/*
5278 * This is called by init or create(). Then, calls to this function are
5279 * always serialized (By cgroup_mutex() at create()).
5280 */
5281
5282static struct css_id *get_new_cssid(struct cgroup_subsys *ss, int depth)
5283{
5284        struct css_id *newid;
5285        int myid, error, size;
5286
5287        BUG_ON(!ss->use_id);
5288
5289        size = sizeof(*newid) + sizeof(unsigned short) * (depth + 1);
5290        newid = kzalloc(size, GFP_KERNEL);
5291        if (!newid)
5292                return ERR_PTR(-ENOMEM);
5293        /* get id */
5294        if (unlikely(!idr_pre_get(&ss->idr, GFP_KERNEL))) {
5295                error = -ENOMEM;
5296                goto err_out;
5297        }
5298        spin_lock(&ss->id_lock);
5299        /* Don't use 0. allocates an ID of 1-65535 */
5300        error = idr_get_new_above(&ss->idr, newid, 1, &myid);
5301        spin_unlock(&ss->id_lock);
5302
5303        /* Returns error when there are no free spaces for new ID.*/
5304        if (error) {
5305                error = -ENOSPC;
5306                goto err_out;
5307        }
5308        if (myid > CSS_ID_MAX)
5309                goto remove_idr;
5310
5311        newid->id = myid;
5312        newid->depth = depth;
5313        return newid;
5314remove_idr:
5315        error = -ENOSPC;
5316        spin_lock(&ss->id_lock);
5317        idr_remove(&ss->idr, myid);
5318        spin_unlock(&ss->id_lock);
5319err_out:
5320        kfree(newid);
5321        return ERR_PTR(error);
5322
5323}
5324
5325static int __init_or_module cgroup_init_idr(struct cgroup_subsys *ss,
5326                                            struct cgroup_subsys_state *rootcss)
5327{
5328        struct css_id *newid;
5329
5330        spin_lock_init(&ss->id_lock);
5331        idr_init(&ss->idr);
5332
5333        newid = get_new_cssid(ss, 0);
5334        if (IS_ERR(newid))
5335                return PTR_ERR(newid);
5336
5337        newid->stack[0] = newid->id;
5338        newid->css = rootcss;
5339        rootcss->id = newid;
5340        return 0;
5341}
5342
5343static int alloc_css_id(struct cgroup_subsys *ss, struct cgroup *parent,
5344                        struct cgroup *child)
5345{
5346        int subsys_id, i, depth = 0;
5347        struct cgroup_subsys_state *parent_css, *child_css;
5348        struct css_id *child_id, *parent_id;
5349
5350        subsys_id = ss->subsys_id;
5351        parent_css = parent->subsys[subsys_id];
5352        child_css = child->subsys[subsys_id];
5353        parent_id = parent_css->id;
5354        depth = parent_id->depth + 1;
5355
5356        child_id = get_new_cssid(ss, depth);
5357        if (IS_ERR(child_id))
5358                return PTR_ERR(child_id);
5359
5360        for (i = 0; i < depth; i++)
5361                child_id->stack[i] = parent_id->stack[i];
5362        child_id->stack[depth] = child_id->id;
5363        /*
5364         * child_id->css pointer will be set after this cgroup is available
5365         * see cgroup_populate_dir()
5366         */
5367        rcu_assign_pointer(child_css->id, child_id);
5368
5369        return 0;
5370}
5371
5372/**
5373 * css_lookup - lookup css by id
5374 * @ss: cgroup subsys to be looked into.
5375 * @id: the id
5376 *
5377 * Returns pointer to cgroup_subsys_state if there is valid one with id.
5378 * NULL if not. Should be called under rcu_read_lock()
5379 */
5380struct cgroup_subsys_state *css_lookup(struct cgroup_subsys *ss, int id)
5381{
5382        struct css_id *cssid = NULL;
5383
5384        BUG_ON(!ss->use_id);
5385        cssid = idr_find(&ss->idr, id);
5386
5387        if (unlikely(!cssid))
5388                return NULL;
5389
5390        return rcu_dereference(cssid->css);
5391}
5392EXPORT_SYMBOL_GPL(css_lookup);
5393
5394/**
5395 * css_get_next - lookup next cgroup under specified hierarchy.
5396 * @ss: pointer to subsystem
5397 * @id: current position of iteration.
5398 * @root: pointer to css. search tree under this.
5399 * @foundid: position of found object.
5400 *
5401 * Search next css under the specified hierarchy of rootid. Calling under
5402 * rcu_read_lock() is necessary. Returns NULL if it reaches the end.
5403 */
5404struct cgroup_subsys_state *
5405css_get_next(struct cgroup_subsys *ss, int id,
5406             struct cgroup_subsys_state *root, int *foundid)
5407{
5408        struct cgroup_subsys_state *ret = NULL;
5409        struct css_id *tmp;
5410        int tmpid;
5411        int rootid = css_id(root);
5412        int depth = css_depth(root);
5413
5414        if (!rootid)
5415                return NULL;
5416
5417        BUG_ON(!ss->use_id);
5418        WARN_ON_ONCE(!rcu_read_lock_held());
5419
5420        /* fill start point for scan */
5421        tmpid = id;
5422        while (1) {
5423                /*
5424                 * scan next entry from bitmap(tree), tmpid is updated after
5425                 * idr_get_next().
5426                 */
5427                tmp = idr_get_next(&ss->idr, &tmpid);
5428                if (!tmp)
5429                        break;
5430                if (tmp->depth >= depth && tmp->stack[depth] == rootid) {
5431                        ret = rcu_dereference(tmp->css);
5432                        if (ret) {
5433                                *foundid = tmpid;
5434                                break;
5435                        }
5436                }
5437                /* continue to scan from next id */
5438                tmpid = tmpid + 1;
5439        }
5440        return ret;
5441}
5442
5443/*
5444 * get corresponding css from file open on cgroupfs directory
5445 */
5446struct cgroup_subsys_state *cgroup_css_from_dir(struct file *f, int id)
5447{
5448        struct cgroup *cgrp;
5449        struct inode *inode;
5450        struct cgroup_subsys_state *css;
5451
5452        inode = f->f_dentry->d_inode;
5453        /* check in cgroup filesystem dir */
5454        if (inode->i_op != &cgroup_dir_inode_operations)
5455                return ERR_PTR(-EBADF);
5456
5457        if (id < 0 || id >= CGROUP_SUBSYS_COUNT)
5458                return ERR_PTR(-EINVAL);
5459
5460        /* get cgroup */
5461        cgrp = __d_cgrp(f->f_dentry);
5462        css = cgrp->subsys[id];
5463        return css ? css : ERR_PTR(-ENOENT);
5464}
5465
5466#ifdef CONFIG_CGROUP_DEBUG
5467static struct cgroup_subsys_state *debug_css_alloc(struct cgroup *cont)
5468{
5469        struct cgroup_subsys_state *css = kzalloc(sizeof(*css), GFP_KERNEL);
5470
5471        if (!css)
5472                return ERR_PTR(-ENOMEM);
5473
5474        return css;
5475}
5476
5477static void debug_css_free(struct cgroup *cont)
5478{
5479        kfree(cont->subsys[debug_subsys_id]);
5480}
5481
5482static u64 cgroup_refcount_read(struct cgroup *cont, struct cftype *cft)
5483{
5484        return atomic_read(&cont->count);
5485}
5486
5487static u64 debug_taskcount_read(struct cgroup *cont, struct cftype *cft)
5488{
5489        return cgroup_task_count(cont);
5490}
5491
5492static u64 current_css_set_read(struct cgroup *cont, struct cftype *cft)
5493{
5494        return (u64)(unsigned long)current->cgroups;
5495}
5496
5497static u64 current_css_set_refcount_read(struct cgroup *cont,
5498                                           struct cftype *cft)
5499{
5500        u64 count;
5501
5502        rcu_read_lock();
5503        count = atomic_read(&current->cgroups->refcount);
5504        rcu_read_unlock();
5505        return count;
5506}
5507
5508static int current_css_set_cg_links_read(struct cgroup *cont,
5509                                         struct cftype *cft,
5510                                         struct seq_file *seq)
5511{
5512        struct cg_cgroup_link *link;
5513        struct css_set *cg;
5514
5515        read_lock(&css_set_lock);
5516        rcu_read_lock();
5517        cg = rcu_dereference(current->cgroups);
5518        list_for_each_entry(link, &cg->cg_links, cg_link_list) {
5519                struct cgroup *c = link->cgrp;
5520                const char *name;
5521
5522                if (c->dentry)
5523                        name = c->dentry->d_name.name;
5524                else
5525                        name = "?";
5526                seq_printf(seq, "Root %d group %s\n",
5527                           c->root->hierarchy_id, name);
5528        }
5529        rcu_read_unlock();
5530        read_unlock(&css_set_lock);
5531        return 0;
5532}
5533
5534#define MAX_TASKS_SHOWN_PER_CSS 25
5535static int cgroup_css_links_read(struct cgroup *cont,
5536                                 struct cftype *cft,
5537                                 struct seq_file *seq)
5538{
5539        struct cg_cgroup_link *link;
5540
5541        read_lock(&css_set_lock);
5542        list_for_each_entry(link, &cont->css_sets, cgrp_link_list) {
5543                struct css_set *cg = link->cg;
5544                struct task_struct *task;
5545                int count = 0;
5546                seq_printf(seq, "css_set %p\n", cg);
5547                list_for_each_entry(task, &cg->tasks, cg_list) {
5548                        if (count++ > MAX_TASKS_SHOWN_PER_CSS) {
5549                                seq_puts(seq, "  ...\n");
5550                                break;
5551                        } else {
5552                                seq_printf(seq, "  task %d\n",
5553                                           task_pid_vnr(task));
5554                        }
5555                }
5556        }
5557        read_unlock(&css_set_lock);
5558        return 0;
5559}
5560
5561static u64 releasable_read(struct cgroup *cgrp, struct cftype *cft)
5562{
5563        return test_bit(CGRP_RELEASABLE, &cgrp->flags);
5564}
5565
5566static struct cftype debug_files[] =  {
5567        {
5568                .name = "cgroup_refcount",
5569                .read_u64 = cgroup_refcount_read,
5570        },
5571        {
5572                .name = "taskcount",
5573                .read_u64 = debug_taskcount_read,
5574        },
5575
5576        {
5577                .name = "current_css_set",
5578                .read_u64 = current_css_set_read,
5579        },
5580
5581        {
5582                .name = "current_css_set_refcount",
5583                .read_u64 = current_css_set_refcount_read,
5584        },
5585
5586        {
5587                .name = "current_css_set_cg_links",
5588                .read_seq_string = current_css_set_cg_links_read,
5589        },
5590
5591        {
5592                .name = "cgroup_css_links",
5593                .read_seq_string = cgroup_css_links_read,
5594        },
5595
5596        {
5597                .name = "releasable",
5598                .read_u64 = releasable_read,
5599        },
5600
5601        { }     /* terminate */
5602};
5603
5604struct cgroup_subsys debug_subsys = {
5605        .name = "debug",
5606        .css_alloc = debug_css_alloc,
5607        .css_free = debug_css_free,
5608        .subsys_id = debug_subsys_id,
5609        .base_cftypes = debug_files,
5610};
5611#endif /* CONFIG_CGROUP_DEBUG */
5612
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.