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