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