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