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/hashtable.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
 379static DEFINE_HASHTABLE(css_set_table, CSS_SET_HASH_BITS);
 380
 381static unsigned long css_set_hash(struct cgroup_subsys_state *css[])
 382{
 383        int i;
 384        unsigned long key = 0UL;
 385
 386        for (i = 0; i < CGROUP_SUBSYS_COUNT; i++)
 387                key += (unsigned long)css[i];
 388        key = (key >> 16) ^ key;
 389
 390        return key;
 391}
 392
 393/* We don't maintain the lists running through each css_set to its
 394 * task until after the first call to cgroup_iter_start(). This
 395 * reduces the fork()/exit() overhead for people who have cgroups
 396 * compiled into their kernel but not actually in use */
 397static int use_task_css_set_links __read_mostly;
 398
 399static void __put_css_set(struct css_set *cg, int taskexit)
 400{
 401        struct cg_cgroup_link *link;
 402        struct cg_cgroup_link *saved_link;
 403        /*
 404         * Ensure that the refcount doesn't hit zero while any readers
 405         * can see it. Similar to atomic_dec_and_lock(), but for an
 406         * rwlock
 407         */
 408        if (atomic_add_unless(&cg->refcount, -1, 1))
 409                return;
 410        write_lock(&css_set_lock);
 411        if (!atomic_dec_and_test(&cg->refcount)) {
 412                write_unlock(&css_set_lock);
 413                return;
 414        }
 415
 416        /* This css_set is dead. unlink it and release cgroup refcounts */
 417        hash_del(&cg->hlist);
 418        css_set_count--;
 419
 420        list_for_each_entry_safe(link, saved_link, &cg->cg_links,
 421                                 cg_link_list) {
 422                struct cgroup *cgrp = link->cgrp;
 423                list_del(&link->cg_link_list);
 424                list_del(&link->cgrp_link_list);
 425
 426                /*
 427                 * We may not be holding cgroup_mutex, and if cgrp->count is
 428                 * dropped to 0 the cgroup can be destroyed at any time, hence
 429                 * rcu_read_lock is used to keep it alive.
 430                 */
 431                rcu_read_lock();
 432                if (atomic_dec_and_test(&cgrp->count) &&
 433                    notify_on_release(cgrp)) {
 434                        if (taskexit)
 435                                set_bit(CGRP_RELEASABLE, &cgrp->flags);
 436                        check_for_release(cgrp);
 437                }
 438                rcu_read_unlock();
 439
 440                kfree(link);
 441        }
 442
 443        write_unlock(&css_set_lock);
 444        kfree_rcu(cg, rcu_head);
 445}
 446
 447/*
 448 * refcounted get/put for css_set objects
 449 */
 450static inline void get_css_set(struct css_set *cg)
 451{
 452        atomic_inc(&cg->refcount);
 453}
 454
 455static inline void put_css_set(struct css_set *cg)
 456{
 457        __put_css_set(cg, 0);
 458}
 459
 460static inline void put_css_set_taskexit(struct css_set *cg)
 461{
 462        __put_css_set(cg, 1);
 463}
 464
 465/*
 466 * compare_css_sets - helper function for find_existing_css_set().
 467 * @cg: candidate css_set being tested
 468 * @old_cg: existing css_set for a task
 469 * @new_cgrp: cgroup that's being entered by the task
 470 * @template: desired set of css pointers in css_set (pre-calculated)
 471 *
 472 * Returns true if "cg" matches "old_cg" except for the hierarchy
 473 * which "new_cgrp" belongs to, for which it should match "new_cgrp".
 474 */
 475static bool compare_css_sets(struct css_set *cg,
 476                             struct css_set *old_cg,
 477                             struct cgroup *new_cgrp,
 478                             struct cgroup_subsys_state *template[])
 479{
 480        struct list_head *l1, *l2;
 481
 482        if (memcmp(template, cg->subsys, sizeof(cg->subsys))) {
 483                /* Not all subsystems matched */
 484                return false;
 485        }
 486
 487        /*
 488         * Compare cgroup pointers in order to distinguish between
 489         * different cgroups in heirarchies with no subsystems. We
 490         * could get by with just this check alone (and skip the
 491         * memcmp above) but on most setups the memcmp check will
 492         * avoid the need for this more expensive check on almost all
 493         * candidates.
 494         */
 495
 496        l1 = &cg->cg_links;
 497        l2 = &old_cg->cg_links;
 498        while (1) {
 499                struct cg_cgroup_link *cgl1, *cgl2;
 500                struct cgroup *cg1, *cg2;
 501
 502                l1 = l1->next;
 503                l2 = l2->next;
 504                /* See if we reached the end - both lists are equal length. */
 505                if (l1 == &cg->cg_links) {
 506                        BUG_ON(l2 != &old_cg->cg_links);
 507                        break;
 508                } else {
 509                        BUG_ON(l2 == &old_cg->cg_links);
 510                }
 511                /* Locate the cgroups associated with these links. */
 512                cgl1 = list_entry(l1, struct cg_cgroup_link, cg_link_list);
 513                cgl2 = list_entry(l2, struct cg_cgroup_link, cg_link_list);
 514                cg1 = cgl1->cgrp;
 515                cg2 = cgl2->cgrp;
 516                /* Hierarchies should be linked in the same order. */
 517                BUG_ON(cg1->root != cg2->root);
 518
 519                /*
 520                 * If this hierarchy is the hierarchy of the cgroup
 521                 * that's changing, then we need to check that this
 522                 * css_set points to the new cgroup; if it's any other
 523                 * hierarchy, then this css_set should point to the
 524                 * same cgroup as the old css_set.
 525                 */
 526                if (cg1->root == new_cgrp->root) {
 527                        if (cg1 != new_cgrp)
 528                                return false;
 529                } else {
 530                        if (cg1 != cg2)
 531                                return false;
 532                }
 533        }
 534        return true;
 535}
 536
 537/*
 538 * find_existing_css_set() is a helper for
 539 * find_css_set(), and checks to see whether an existing
 540 * css_set is suitable.
 541 *
 542 * oldcg: the cgroup group that we're using before the cgroup
 543 * transition
 544 *
 545 * cgrp: the cgroup that we're moving into
 546 *
 547 * template: location in which to build the desired set of subsystem
 548 * state objects for the new cgroup group
 549 */
 550static struct css_set *find_existing_css_set(
 551        struct css_set *oldcg,
 552        struct cgroup *cgrp,
 553        struct cgroup_subsys_state *template[])
 554{
 555        int i;
 556        struct cgroupfs_root *root = cgrp->root;
 557        struct css_set *cg;
 558        unsigned long key;
 559
 560        /*
 561         * Build the set of subsystem state objects that we want to see in the
 562         * new css_set. while subsystems can change globally, the entries here
 563         * won't change, so no need for locking.
 564         */
 565        for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
 566                if (root->subsys_mask & (1UL << i)) {
 567                        /* Subsystem is in this hierarchy. So we want
 568                         * the subsystem state from the new
 569                         * cgroup */
 570                        template[i] = cgrp->subsys[i];
 571                } else {
 572                        /* Subsystem is not in this hierarchy, so we
 573                         * don't want to change the subsystem state */
 574                        template[i] = oldcg->subsys[i];
 575                }
 576        }
 577
 578        key = css_set_hash(template);
 579        hash_for_each_possible(css_set_table, cg, hlist, key) {
 580                if (!compare_css_sets(cg, oldcg, cgrp, template))
 581                        continue;
 582
 583                /* This css_set matches what we need */
 584                return cg;
 585        }
 586
 587        /* No existing cgroup group matched */
 588        return NULL;
 589}
 590
 591static void free_cg_links(struct list_head *tmp)
 592{
 593        struct cg_cgroup_link *link;
 594        struct cg_cgroup_link *saved_link;
 595
 596        list_for_each_entry_safe(link, saved_link, tmp, cgrp_link_list) {
 597                list_del(&link->cgrp_link_list);
 598                kfree(link);
 599        }
 600}
 601
 602/*
 603 * allocate_cg_links() allocates "count" cg_cgroup_link structures
 604 * and chains them on tmp through their cgrp_link_list fields. Returns 0 on
 605 * success or a negative error
 606 */
 607static int allocate_cg_links(int count, struct list_head *tmp)
 608{
 609        struct cg_cgroup_link *link;
 610        int i;
 611        INIT_LIST_HEAD(tmp);
 612        for (i = 0; i < count; i++) {
 613                link = kmalloc(sizeof(*link), GFP_KERNEL);
 614                if (!link) {
 615                        free_cg_links(tmp);
 616                        return -ENOMEM;
 617                }
 618                list_add(&link->cgrp_link_list, tmp);
 619        }
 620        return 0;
 621}
 622
 623/**
 624 * link_css_set - a helper function to link a css_set to a cgroup
 625 * @tmp_cg_links: cg_cgroup_link objects allocated by allocate_cg_links()
 626 * @cg: the css_set to be linked
 627 * @cgrp: the destination cgroup
 628 */
 629static void link_css_set(struct list_head *tmp_cg_links,
 630                         struct css_set *cg, struct cgroup *cgrp)
 631{
 632        struct cg_cgroup_link *link;
 633
 634        BUG_ON(list_empty(tmp_cg_links));
 635        link = list_first_entry(tmp_cg_links, struct cg_cgroup_link,
 636                                cgrp_link_list);
 637        link->cg = cg;
 638        link->cgrp = cgrp;
 639        atomic_inc(&cgrp->count);
 640        list_move(&link->cgrp_link_list, &cgrp->css_sets);
 641        /*
 642         * Always add links to the tail of the list so that the list
 643         * is sorted by order of hierarchy creation
 644         */
 645        list_add_tail(&link->cg_link_list, &cg->cg_links);
 646}
 647
 648/*
 649 * find_css_set() takes an existing cgroup group and a
 650 * cgroup object, and returns a css_set object that's
 651 * equivalent to the old group, but with the given cgroup
 652 * substituted into the appropriate hierarchy. Must be called with
 653 * cgroup_mutex held
 654 */
 655static struct css_set *find_css_set(
 656        struct css_set *oldcg, struct cgroup *cgrp)
 657{
 658        struct css_set *res;
 659        struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT];
 660
 661        struct list_head tmp_cg_links;
 662
 663        struct cg_cgroup_link *link;
 664        unsigned long key;
 665
 666        /* First see if we already have a cgroup group that matches
 667         * the desired set */
 668        read_lock(&css_set_lock);
 669        res = find_existing_css_set(oldcg, cgrp, template);
 670        if (res)
 671                get_css_set(res);
 672        read_unlock(&css_set_lock);
 673
 674        if (res)
 675                return res;
 676
 677        res = kmalloc(sizeof(*res), GFP_KERNEL);
 678        if (!res)
 679                return NULL;
 680
 681        /* Allocate all the cg_cgroup_link objects that we'll need */
 682        if (allocate_cg_links(root_count, &tmp_cg_links) < 0) {
 683                kfree(res);
 684                return NULL;
 685        }
 686
 687        atomic_set(&res->refcount, 1);
 688        INIT_LIST_HEAD(&res->cg_links);
 689        INIT_LIST_HEAD(&res->tasks);
 690        INIT_HLIST_NODE(&res->hlist);
 691
 692        /* Copy the set of subsystem state objects generated in
 693         * find_existing_css_set() */
 694        memcpy(res->subsys, template, sizeof(res->subsys));
 695
 696        write_lock(&css_set_lock);
 697        /* Add reference counts and links from the new css_set. */
 698        list_for_each_entry(link, &oldcg->cg_links, cg_link_list) {
 699                struct cgroup *c = link->cgrp;
 700                if (c->root == cgrp->root)
 701                        c = cgrp;
 702                link_css_set(&tmp_cg_links, res, c);
 703        }
 704
 705        BUG_ON(!list_empty(&tmp_cg_links));
 706
 707        css_set_count++;
 708
 709        /* Add this cgroup group to the hash table */
 710        key = css_set_hash(res->subsys);
 711        hash_add(css_set_table, &res->hlist, key);
 712
 713        write_unlock(&css_set_lock);
 714
 715        return res;
 716}
 717
 718/*
 719 * Return the cgroup for "task" from the given hierarchy. Must be
 720 * called with cgroup_mutex held.
 721 */
 722static struct cgroup *task_cgroup_from_root(struct task_struct *task,
 723                                            struct cgroupfs_root *root)
 724{
 725        struct css_set *css;
 726        struct cgroup *res = NULL;
 727
 728        BUG_ON(!mutex_is_locked(&cgroup_mutex));
 729        read_lock(&css_set_lock);
 730        /*
 731         * No need to lock the task - since we hold cgroup_mutex the
 732         * task can't change groups, so the only thing that can happen
 733         * is that it exits and its css is set back to init_css_set.
 734         */
 735        css = task->cgroups;
 736        if (css == &init_css_set) {
 737                res = &root->top_cgroup;
 738        } else {
 739                struct cg_cgroup_link *link;
 740                list_for_each_entry(link, &css->cg_links, cg_link_list) {
 741                        struct cgroup *c = link->cgrp;
 742                        if (c->root == root) {
 743                                res = c;
 744                                break;
 745                        }
 746                }
 747        }
 748        read_unlock(&css_set_lock);
 749        BUG_ON(!res);
 750        return res;
 751}
 752
 753/*
 754 * There is one global cgroup mutex. We also require taking
 755 * task_lock() when dereferencing a task's cgroup subsys pointers.
 756 * See "The task_lock() exception", at the end of this comment.
 757 *
 758 * A task must hold cgroup_mutex to modify cgroups.
 759 *
 760 * Any task can increment and decrement the count field without lock.
 761 * So in general, code holding cgroup_mutex can't rely on the count
 762 * field not changing.  However, if the count goes to zero, then only
 763 * cgroup_attach_task() can increment it again.  Because a count of zero
 764 * means that no tasks are currently attached, therefore there is no
 765 * way a task attached to that cgroup can fork (the other way to
 766 * increment the count).  So code holding cgroup_mutex can safely
 767 * assume that if the count is zero, it will stay zero. Similarly, if
 768 * a task holds cgroup_mutex on a cgroup with zero count, it
 769 * knows that the cgroup won't be removed, as cgroup_rmdir()
 770 * needs that mutex.
 771 *
 772 * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't
 773 * (usually) take cgroup_mutex.  These are the two most performance
 774 * critical pieces of code here.  The exception occurs on cgroup_exit(),
 775 * when a task in a notify_on_release cgroup exits.  Then cgroup_mutex
 776 * is taken, and if the cgroup count is zero, a usermode call made
 777 * to the release agent with the name of the cgroup (path relative to
 778 * the root of cgroup file system) as the argument.
 779 *
 780 * A cgroup can only be deleted if both its 'count' of using tasks
 781 * is zero, and its list of 'children' cgroups is empty.  Since all
 782 * tasks in the system use _some_ cgroup, and since there is always at
 783 * least one task in the system (init, pid == 1), therefore, top_cgroup
 784 * always has either children cgroups and/or using tasks.  So we don't
 785 * need a special hack to ensure that top_cgroup cannot be deleted.
 786 *
 787 *      The task_lock() exception
 788 *
 789 * The need for this exception arises from the action of
 790 * cgroup_attach_task(), which overwrites one task's cgroup pointer with
 791 * another.  It does so using cgroup_mutex, however there are
 792 * several performance critical places that need to reference
 793 * task->cgroup without the expense of grabbing a system global
 794 * mutex.  Therefore except as noted below, when dereferencing or, as
 795 * in cgroup_attach_task(), modifying a task's cgroup pointer we use
 796 * task_lock(), which acts on a spinlock (task->alloc_lock) already in
 797 * the task_struct routinely used for such matters.
 798 *
 799 * P.S.  One more locking exception.  RCU is used to guard the
 800 * update of a tasks cgroup pointer by cgroup_attach_task()
 801 */
 802
 803/**
 804 * cgroup_lock - lock out any changes to cgroup structures
 805 *
 806 */
 807void cgroup_lock(void)
 808{
 809        mutex_lock(&cgroup_mutex);
 810}
 811EXPORT_SYMBOL_GPL(cgroup_lock);
 812
 813/**
 814 * cgroup_unlock - release lock on cgroup changes
 815 *
 816 * Undo the lock taken in a previous cgroup_lock() call.
 817 */
 818void cgroup_unlock(void)
 819{
 820        mutex_unlock(&cgroup_mutex);
 821}
 822EXPORT_SYMBOL_GPL(cgroup_unlock);
 823
 824/*
 825 * A couple of forward declarations required, due to cyclic reference loop:
 826 * cgroup_mkdir -> cgroup_create -> cgroup_populate_dir ->
 827 * cgroup_add_file -> cgroup_create_file -> cgroup_dir_inode_operations
 828 * -> cgroup_mkdir.
 829 */
 830
 831static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode);
 832static struct dentry *cgroup_lookup(struct inode *, struct dentry *, unsigned int);
 833static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry);
 834static int cgroup_populate_dir(struct cgroup *cgrp, bool base_files,
 835                               unsigned long subsys_mask);
 836static const struct inode_operations cgroup_dir_inode_operations;
 837static const struct file_operations proc_cgroupstats_operations;
 838
 839static struct backing_dev_info cgroup_backing_dev_info = {
 840        .name           = "cgroup",
 841        .capabilities   = BDI_CAP_NO_ACCT_AND_WRITEBACK,
 842};
 843
 844static int alloc_css_id(struct cgroup_subsys *ss,
 845                        struct cgroup *parent, struct cgroup *child);
 846
 847static struct inode *cgroup_new_inode(umode_t mode, struct super_block *sb)
 848{
 849        struct inode *inode = new_inode(sb);
 850
 851        if (inode) {
 852                inode->i_ino = get_next_ino();
 853                inode->i_mode = mode;
 854                inode->i_uid = current_fsuid();
 855                inode->i_gid = current_fsgid();
 856                inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 857                inode->i_mapping->backing_dev_info = &cgroup_backing_dev_info;
 858        }
 859        return inode;
 860}
 861
 862static void cgroup_free_fn(struct work_struct *work)
 863{
 864        struct cgroup *cgrp = container_of(work, struct cgroup, free_work);
 865        struct cgroup_subsys *ss;
 866
 867        mutex_lock(&cgroup_mutex);
 868        /*
 869         * Release the subsystem state objects.
 870         */
 871        for_each_subsys(cgrp->root, ss)
 872                ss->css_free(cgrp);
 873
 874        cgrp->root->number_of_cgroups--;
 875        mutex_unlock(&cgroup_mutex);
 876
 877        /*
 878         * Drop the active superblock reference that we took when we
 879         * created the cgroup
 880         */
 881        deactivate_super(cgrp->root->sb);
 882
 883        /*
 884         * if we're getting rid of the cgroup, refcount should ensure
 885         * that there are no pidlists left.
 886         */
 887        BUG_ON(!list_empty(&cgrp->pidlists));
 888
 889        simple_xattrs_free(&cgrp->xattrs);
 890
 891        ida_simple_remove(&cgrp->root->cgroup_ida, cgrp->id);
 892        kfree(cgrp);
 893}
 894
 895static void cgroup_free_rcu(struct rcu_head *head)
 896{
 897        struct cgroup *cgrp = container_of(head, struct cgroup, rcu_head);
 898
 899        schedule_work(&cgrp->free_work);
 900}
 901
 902static void cgroup_diput(struct dentry *dentry, struct inode *inode)
 903{
 904        /* is dentry a directory ? if so, kfree() associated cgroup */
 905        if (S_ISDIR(inode->i_mode)) {
 906                struct cgroup *cgrp = dentry->d_fsdata;
 907
 908                BUG_ON(!(cgroup_is_removed(cgrp)));
 909                call_rcu(&cgrp->rcu_head, cgroup_free_rcu);
 910        } else {
 911                struct cfent *cfe = __d_cfe(dentry);
 912                struct cgroup *cgrp = dentry->d_parent->d_fsdata;
 913                struct cftype *cft = cfe->type;
 914
 915                WARN_ONCE(!list_empty(&cfe->node) &&
 916                          cgrp != &cgrp->root->top_cgroup,
 917                          "cfe still linked for %s\n", cfe->type->name);
 918                kfree(cfe);
 919                simple_xattrs_free(&cft->xattrs);
 920        }
 921        iput(inode);
 922}
 923
 924static int cgroup_delete(const struct dentry *d)
 925{
 926        return 1;
 927}
 928
 929static void remove_dir(struct dentry *d)
 930{
 931        struct dentry *parent = dget(d->d_parent);
 932
 933        d_delete(d);
 934        simple_rmdir(parent->d_inode, d);
 935        dput(parent);
 936}
 937
 938static void cgroup_rm_file(struct cgroup *cgrp, const struct cftype *cft)
 939{
 940        struct cfent *cfe;
 941
 942        lockdep_assert_held(&cgrp->dentry->d_inode->i_mutex);
 943        lockdep_assert_held(&cgroup_mutex);
 944
 945        /*
 946         * If we're doing cleanup due to failure of cgroup_create(),
 947         * the corresponding @cfe may not exist.
 948         */
 949        list_for_each_entry(cfe, &cgrp->files, node) {
 950                struct dentry *d = cfe->dentry;
 951
 952                if (cft && cfe->type != cft)
 953                        continue;
 954
 955                dget(d);
 956                d_delete(d);
 957                simple_unlink(cgrp->dentry->d_inode, d);
 958                list_del_init(&cfe->node);
 959                dput(d);
 960
 961                break;
 962        }
 963}
 964
 965/**
 966 * cgroup_clear_directory - selective removal of base and subsystem files
 967 * @dir: directory containing the files
 968 * @base_files: true if the base files should be removed
 969 * @subsys_mask: mask of the subsystem ids whose files should be removed
 970 */
 971static void cgroup_clear_directory(struct dentry *dir, bool base_files,
 972                                   unsigned long subsys_mask)
 973{
 974        struct cgroup *cgrp = __d_cgrp(dir);
 975        struct cgroup_subsys *ss;
 976
 977        for_each_subsys(cgrp->root, ss) {
 978                struct cftype_set *set;
 979                if (!test_bit(ss->subsys_id, &subsys_mask))
 980                        continue;
 981                list_for_each_entry(set, &ss->cftsets, node)
 982                        cgroup_addrm_files(cgrp, NULL, set->cfts, false);
 983        }
 984        if (base_files) {
 985                while (!list_empty(&cgrp->files))
 986                        cgroup_rm_file(cgrp, NULL);
 987        }
 988}
 989
 990/*
 991 * NOTE : the dentry must have been dget()'ed
 992 */
 993static void cgroup_d_remove_dir(struct dentry *dentry)
 994{
 995        struct dentry *parent;
 996        struct cgroupfs_root *root = dentry->d_sb->s_fs_info;
 997
 998        cgroup_clear_directory(dentry, true, root->subsys_mask);
 999
1000        parent = dentry->d_parent;
1001        spin_lock(&parent->d_lock);
1002        spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1003        list_del_init(&dentry->d_u.d_child);
1004        spin_unlock(&dentry->d_lock);
1005        spin_unlock(&parent->d_lock);
1006        remove_dir(dentry);
1007}
1008
1009/*
1010 * Call with cgroup_mutex held. Drops reference counts on modules, including
1011 * any duplicate ones that parse_cgroupfs_options took. If this function
1012 * returns an error, no reference counts are touched.
1013 */
1014static int rebind_subsystems(struct cgroupfs_root *root,
1015                              unsigned long final_subsys_mask)
1016{
1017        unsigned long added_mask, removed_mask;
1018        struct cgroup *cgrp = &root->top_cgroup;
1019        int i;
1020
1021        BUG_ON(!mutex_is_locked(&cgroup_mutex));
1022        BUG_ON(!mutex_is_locked(&cgroup_root_mutex));
1023
1024        removed_mask = root->actual_subsys_mask & ~final_subsys_mask;
1025        added_mask = final_subsys_mask & ~root->actual_subsys_mask;
1026        /* Check that any added subsystems are currently free */
1027        for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1028                unsigned long bit = 1UL << i;
1029                struct cgroup_subsys *ss = subsys[i];
1030                if (!(bit & added_mask))
1031                        continue;
1032                /*
1033                 * Nobody should tell us to do a subsys that doesn't exist:
1034                 * parse_cgroupfs_options should catch that case and refcounts
1035                 * ensure that subsystems won't disappear once selected.
1036                 */
1037                BUG_ON(ss == NULL);
1038                if (ss->root != &rootnode) {
1039                        /* Subsystem isn't free */
1040                        return -EBUSY;
1041                }
1042        }
1043
1044        /* Currently we don't handle adding/removing subsystems when
1045         * any child cgroups exist. This is theoretically supportable
1046         * but involves complex error handling, so it's being left until
1047         * later */
1048        if (root->number_of_cgroups > 1)
1049                return -EBUSY;
1050
1051        /* Process each subsystem */
1052        for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1053                struct cgroup_subsys *ss = subsys[i];
1054                unsigned long bit = 1UL << i;
1055                if (bit & added_mask) {
1056                        /* We're binding this subsystem to this hierarchy */
1057                        BUG_ON(ss == NULL);
1058                        BUG_ON(cgrp->subsys[i]);
1059                        BUG_ON(!dummytop->subsys[i]);
1060                        BUG_ON(dummytop->subsys[i]->cgroup != dummytop);
1061                        cgrp->subsys[i] = dummytop->subsys[i];
1062                        cgrp->subsys[i]->cgroup = cgrp;
1063                        list_move(&ss->sibling, &root->subsys_list);
1064                        ss->root = root;
1065                        if (ss->bind)
1066                                ss->bind(cgrp);
1067                        /* refcount was already taken, and we're keeping it */
1068                } else if (bit & removed_mask) {
1069                        /* We're removing this subsystem */
1070                        BUG_ON(ss == NULL);
1071                        BUG_ON(cgrp->subsys[i] != dummytop->subsys[i]);
1072                        BUG_ON(cgrp->subsys[i]->cgroup != cgrp);
1073                        if (ss->bind)
1074                                ss->bind(dummytop);
1075                        dummytop->subsys[i]->cgroup = dummytop;
1076                        cgrp->subsys[i] = NULL;
1077                        subsys[i]->root = &rootnode;
1078                        list_move(&ss->sibling, &rootnode.subsys_list);
1079                        /* subsystem is now free - drop reference on module */
1080                        module_put(ss->module);
1081                } else if (bit & final_subsys_mask) {
1082                        /* Subsystem state should already exist */
1083                        BUG_ON(ss == NULL);
1084                        BUG_ON(!cgrp->subsys[i]);
1085                        /*
1086                         * a refcount was taken, but we already had one, so
1087                         * drop the extra reference.
1088                         */
1089                        module_put(ss->module);
1090#ifdef CONFIG_MODULE_UNLOAD
1091                        BUG_ON(ss->module && !module_refcount(ss->module));
1092#endif
1093                } else {
1094                        /* Subsystem state shouldn't exist */
1095                        BUG_ON(cgrp->subsys[i]);
1096                }
1097        }
1098        root->subsys_mask = root->actual_subsys_mask = final_subsys_mask;
1099
1100        return 0;
1101}
1102
1103static int cgroup_show_options(struct seq_file *seq, struct dentry *dentry)
1104{
1105        struct cgroupfs_root *root = dentry->d_sb->s_fs_info;
1106        struct cgroup_subsys *ss;
1107
1108        mutex_lock(&cgroup_root_mutex);
1109        for_each_subsys(root, ss)
1110                seq_printf(seq, ",%s", ss->name);
1111        if (test_bit(ROOT_NOPREFIX, &root->flags))
1112                seq_puts(seq, ",noprefix");
1113        if (test_bit(ROOT_XATTR, &root->flags))
1114                seq_puts(seq, ",xattr");
1115        if (strlen(root->release_agent_path))
1116                seq_printf(seq, ",release_agent=%s", root->release_agent_path);
1117        if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->top_cgroup.flags))
1118                seq_puts(seq, ",clone_children");
1119        if (strlen(root->name))
1120                seq_printf(seq, ",name=%s", root->name);
1121        mutex_unlock(&cgroup_root_mutex);
1122        return 0;
1123}
1124
1125struct cgroup_sb_opts {
1126        unsigned long subsys_mask;
1127        unsigned long flags;
1128        char *release_agent;
1129        bool cpuset_clone_children;
1130        char *name;
1131        /* User explicitly requested empty subsystem */
1132        bool none;
1133
1134        struct cgroupfs_root *new_root;
1135
1136};
1137
1138/*
1139 * Convert a hierarchy specifier into a bitmask of subsystems and flags. Call
1140 * with cgroup_mutex held to protect the subsys[] array. This function takes
1141 * refcounts on subsystems to be used, unless it returns error, in which case
1142 * no refcounts are taken.
1143 */
1144static int parse_cgroupfs_options(char *data, struct cgroup_sb_opts *opts)
1145{
1146        char *token, *o = data;
1147        bool all_ss = false, one_ss = false;
1148        unsigned long mask = (unsigned long)-1;
1149        int i;
1150        bool module_pin_failed = false;
1151
1152        BUG_ON(!mutex_is_locked(&cgroup_mutex));
1153
1154#ifdef CONFIG_CPUSETS
1155        mask = ~(1UL << cpuset_subsys_id);
1156#endif
1157
1158        memset(opts, 0, sizeof(*opts));
1159
1160        while ((token = strsep(&o, ",")) != NULL) {
1161                if (!*token)
1162                        return -EINVAL;
1163                if (!strcmp(token, "none")) {
1164                        /* Explicitly have no subsystems */
1165                        opts->none = true;
1166                        continue;
1167                }
1168                if (!strcmp(token, "all")) {
1169                        /* Mutually exclusive option 'all' + subsystem name */
1170                        if (one_ss)
1171                                return -EINVAL;
1172                        all_ss = true;
1173                        continue;
1174                }
1175                if (!strcmp(token, "noprefix")) {
1176                        set_bit(ROOT_NOPREFIX, &opts->flags);
1177                        continue;
1178                }
1179                if (!strcmp(token, "clone_children")) {
1180                        opts->cpuset_clone_children = true;
1181                        continue;
1182                }
1183                if (!strcmp(token, "xattr")) {
1184                        set_bit(ROOT_XATTR, &opts->flags);
1185                        continue;
1186                }
1187                if (!strncmp(token, "release_agent=", 14)) {
1188                        /* Specifying two release agents is forbidden */
1189                        if (opts->release_agent)
1190                                return -EINVAL;
1191                        opts->release_agent =
1192                                kstrndup(token + 14, PATH_MAX - 1, GFP_KERNEL);
1193                        if (!opts->release_agent)
1194                                return -ENOMEM;
1195                        continue;
1196                }
1197                if (!strncmp(token, "name=", 5)) {
1198                        const char *name = token + 5;
1199                        /* Can't specify an empty name */
1200                        if (!strlen(name))
1201                                return -EINVAL;
1202                        /* Must match [\w.-]+ */
1203                        for (i = 0; i < strlen(name); i++) {
1204                                char c = name[i];
1205                                if (isalnum(c))
1206                                        continue;
1207                                if ((c == '.') || (c == '-') || (c == '_'))
1208                                        continue;
1209                                return -EINVAL;
1210                        }
1211                        /* Specifying two names is forbidden */
1212                        if (opts->name)
1213                                return -EINVAL;
1214                        opts->name = kstrndup(name,
1215                                              MAX_CGROUP_ROOT_NAMELEN - 1,
1216                                              GFP_KERNEL);
1217                        if (!opts->name)
1218                                return -ENOMEM;
1219
1220                        continue;
1221                }
1222
1223                for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1224                        struct cgroup_subsys *ss = subsys[i];
1225                        if (ss == NULL)
1226                                continue;
1227                        if (strcmp(token, ss->name))
1228                                continue;
1229                        if (ss->disabled)
1230                                continue;
1231
1232                        /* Mutually exclusive option 'all' + subsystem name */
1233                        if (all_ss)
1234                                return -EINVAL;
1235                        set_bit(i, &opts->subsys_mask);
1236                        one_ss = true;
1237
1238                        break;
1239                }
1240                if (i == CGROUP_SUBSYS_COUNT)
1241                        return -ENOENT;
1242        }
1243
1244        /*
1245         * If the 'all' option was specified select all the subsystems,
1246         * otherwise if 'none', 'name=' and a subsystem name options
1247         * were not specified, let's default to 'all'
1248         */
1249        if (all_ss || (!one_ss && !opts->none && !opts->name)) {
1250                for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1251                        struct cgroup_subsys *ss = subsys[i];
1252                        if (ss == NULL)
1253                                continue;
1254                        if (ss->disabled)
1255                                continue;
1256                        set_bit(i, &opts->subsys_mask);
1257                }
1258        }
1259
1260        /* Consistency checks */
1261
1262        /*
1263         * Option noprefix was introduced just for backward compatibility
1264         * with the old cpuset, so we allow noprefix only if mounting just
1265         * the cpuset subsystem.
1266         */
1267        if (test_bit(ROOT_NOPREFIX, &opts->flags) &&
1268            (opts->subsys_mask & mask))
1269                return -EINVAL;
1270
1271
1272        /* Can't specify "none" and some subsystems */
1273        if (opts->subsys_mask && opts->none)
1274                return -EINVAL;
1275
1276        /*
1277         * We either have to specify by name or by subsystems. (So all
1278         * empty hierarchies must have a name).
1279         */
1280        if (!opts->subsys_mask && !opts->name)
1281                return -EINVAL;
1282
1283        /*
1284         * Grab references on all the modules we'll need, so the subsystems
1285         * don't dance around before rebind_subsystems attaches them. This may
1286         * take duplicate reference counts on a subsystem that's already used,
1287         * but rebind_subsystems handles this case.
1288         */
1289        for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1290                unsigned long bit = 1UL << i;
1291
1292                if (!(bit & opts->subsys_mask))
1293                        continue;
1294                if (!try_module_get(subsys[i]->module)) {
1295                        module_pin_failed = true;
1296                        break;
1297                }
1298        }
1299        if (module_pin_failed) {
1300                /*
1301                 * oops, one of the modules was going away. this means that we
1302                 * raced with a module_delete call, and to the user this is
1303                 * essentially a "subsystem doesn't exist" case.
1304                 */
1305                for (i--; i >= 0; i--) {
1306                        /* drop refcounts only on the ones we took */
1307                        unsigned long bit = 1UL << i;
1308
1309                        if (!(bit & opts->subsys_mask))
1310                                continue;
1311                        module_put(subsys[i]->module);
1312                }
1313                return -ENOENT;
1314        }
1315
1316        return 0;
1317}
1318
1319static void drop_parsed_module_refcounts(unsigned long subsys_mask)
1320{
1321        int i;
1322        for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1323                unsigned long bit = 1UL << i;
1324
1325                if (!(bit & subsys_mask))
1326                        continue;
1327                module_put(subsys[i]->module);
1328        }
1329}
1330
1331static int cgroup_remount(struct super_block *sb, int *flags, char *data)
1332{
1333        int ret = 0;
1334        struct cgroupfs_root *root = sb->s_fs_info;
1335        struct cgroup *cgrp = &root->top_cgroup;
1336        struct cgroup_sb_opts opts;
1337        unsigned long added_mask, removed_mask;
1338
1339        mutex_lock(&cgrp->dentry->d_inode->i_mutex);
1340        mutex_lock(&cgroup_mutex);
1341        mutex_lock(&cgroup_root_mutex);
1342
1343        /* See what subsystems are wanted */
1344        ret = parse_cgroupfs_options(data, &opts);
1345        if (ret)
1346                goto out_unlock;
1347
1348        if (opts.subsys_mask != root->actual_subsys_mask || opts.release_agent)
1349                pr_warning("cgroup: option changes via remount are deprecated (pid=%d comm=%s)\n",
1350                           task_tgid_nr(current), current->comm);
1351
1352        added_mask = opts.subsys_mask & ~root->subsys_mask;
1353        removed_mask = root->subsys_mask & ~opts.subsys_mask;
1354
1355        /* Don't allow flags or name to change at remount */
1356        if (opts.flags != root->flags ||
1357            (opts.name && strcmp(opts.name, root->name))) {
1358                ret = -EINVAL;
1359                drop_parsed_module_refcounts(opts.subsys_mask);
1360                goto out_unlock;
1361        }
1362
1363        /*
1364         * Clear out the files of subsystems that should be removed, do
1365         * this before rebind_subsystems, since rebind_subsystems may
1366         * change this hierarchy's subsys_list.
1367         */
1368        cgroup_clear_directory(cgrp->dentry, false, removed_mask);
1369
1370        ret = rebind_subsystems(root, opts.subsys_mask);
1371        if (ret) {
1372                /* rebind_subsystems failed, re-populate the removed files */
1373                cgroup_populate_dir(cgrp, false, removed_mask);
1374                drop_parsed_module_refcounts(opts.subsys_mask);
1375                goto out_unlock;
1376        }
1377
1378        /* re-populate subsystem files */
1379        cgroup_populate_dir(cgrp, false, added_mask);
1380
1381        if (opts.release_agent)
1382                strcpy(root->release_agent_path, opts.release_agent);
1383 out_unlock:
1384        kfree(opts.release_agent);
1385        kfree(opts.name);
1386        mutex_unlock(&cgroup_root_mutex);
1387        mutex_unlock(&cgroup_mutex);
1388        mutex_unlock(&cgrp->dentry->d_inode->i_mutex);
1389        return ret;
1390}
1391
1392static const struct super_operations cgroup_ops = {
1393        .statfs = simple_statfs,
1394        .drop_inode = generic_delete_inode,
1395        .show_options = cgroup_show_options,
1396        .remount_fs = cgroup_remount,
1397};
1398
1399static void init_cgroup_housekeeping(struct cgroup *cgrp)
1400{
1401        INIT_LIST_HEAD(&cgrp->sibling);
1402        INIT_LIST_HEAD(&cgrp->children);
1403        INIT_LIST_HEAD(&cgrp->files);
1404        INIT_LIST_HEAD(&cgrp->css_sets);
1405        INIT_LIST_HEAD(&cgrp->allcg_node);
1406        INIT_LIST_HEAD(&cgrp->release_list);
1407        INIT_LIST_HEAD(&cgrp->pidlists);
1408        INIT_WORK(&cgrp->free_work, cgroup_free_fn);
1409        mutex_init(&cgrp->pidlist_mutex);
1410        INIT_LIST_HEAD(&cgrp->event_list);
1411        spin_lock_init(&cgrp->event_list_lock);
1412        simple_xattrs_init(&cgrp->xattrs);
1413}
1414
1415static void init_cgroup_root(struct cgroupfs_root *root)
1416{
1417        struct cgroup *cgrp = &root->top_cgroup;
1418
1419        INIT_LIST_HEAD(&root->subsys_list);
1420        INIT_LIST_HEAD(&root->root_list);
1421        INIT_LIST_HEAD(&root->allcg_list);
1422        root->number_of_cgroups = 1;
1423        cgrp->root = root;
1424        cgrp->top_cgroup = cgrp;
1425        init_cgroup_housekeeping(cgrp);
1426        list_add_tail(&cgrp->allcg_node, &root->allcg_list);
1427}
1428
1429static bool init_root_id(struct cgroupfs_root *root)
1430{
1431        int ret = 0;
1432
1433        do {
1434                if (!ida_pre_get(&hierarchy_ida, GFP_KERNEL))
1435                        return false;
1436                spin_lock(&hierarchy_id_lock);
1437                /* Try to allocate the next unused ID */
1438                ret = ida_get_new_above(&hierarchy_ida, next_hierarchy_id,
1439                                        &root->hierarchy_id);
1440                if (ret == -ENOSPC)
1441                        /* Try again starting from 0 */
1442                        ret = ida_get_new(&hierarchy_ida, &root->hierarchy_id);
1443                if (!ret) {
1444                        next_hierarchy_id = root->hierarchy_id + 1;
1445                } else if (ret != -EAGAIN) {
1446                        /* Can only get here if the 31-bit IDR is full ... */
1447                        BUG_ON(ret);
1448                }
1449                spin_unlock(&hierarchy_id_lock);
1450        } while (ret);
1451        return true;
1452}
1453
1454static int cgroup_test_super(struct super_block *sb, void *data)
1455{
1456        struct cgroup_sb_opts *opts = data;
1457        struct cgroupfs_root *root = sb->s_fs_info;
1458
1459        /* If we asked for a name then it must match */
1460        if (opts->name && strcmp(opts->name, root->name))
1461                return 0;
1462
1463        /*
1464         * If we asked for subsystems (or explicitly for no
1465         * subsystems) then they must match
1466         */
1467        if ((opts->subsys_mask || opts->none)
1468            && (opts->subsys_mask != root->subsys_mask))
1469                return 0;
1470
1471        return 1;
1472}
1473
1474static struct cgroupfs_root *cgroup_root_from_opts(struct cgroup_sb_opts *opts)
1475{
1476        struct cgroupfs_root *root;
1477
1478        if (!opts->subsys_mask && !opts->none)
1479                return NULL;
1480
1481        root = kzalloc(sizeof(*root), GFP_KERNEL);
1482        if (!root)
1483                return ERR_PTR(-ENOMEM);
1484
1485        if (!init_root_id(root)) {
1486                kfree(root);
1487                return ERR_PTR(-ENOMEM);
1488        }
1489        init_cgroup_root(root);
1490
1491        root->subsys_mask = opts->subsys_mask;
1492        root->flags = opts->flags;
1493        ida_init(&root->cgroup_ida);
1494        if (opts->release_agent)
1495                strcpy(root->release_agent_path, opts->release_agent);
1496        if (opts->name)
1497                strcpy(root->name, opts->name);
1498        if (opts->cpuset_clone_children)
1499                set_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->top_cgroup.flags);
1500        return root;
1501}
1502
1503static void cgroup_drop_root(struct cgroupfs_root *root)
1504{
1505        if (!root)
1506                return;
1507
1508        BUG_ON(!root->hierarchy_id);
1509        spin_lock(&hierarchy_id_lock);
1510        ida_remove(&hierarchy_ida, root->hierarchy_id);
1511        spin_unlock(&hierarchy_id_lock);
1512        ida_destroy(&root->cgroup_ida);
1513        kfree(root);
1514}
1515
1516static int cgroup_set_super(struct super_block *sb, void *data)
1517{
1518        int ret;
1519        struct cgroup_sb_opts *opts = data;
1520
1521        /* If we don't have a new root, we can't set up a new sb */
1522        if (!opts->new_root)
1523                return -EINVAL;
1524
1525        BUG_ON(!opts->subsys_mask && !opts->none);
1526
1527        ret = set_anon_super(sb, NULL);
1528        if (ret)
1529                return ret;
1530
1531        sb->s_fs_info = opts->new_root;
1532        opts->new_root->sb = sb;
1533
1534        sb->s_blocksize = PAGE_CACHE_SIZE;
1535        sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
1536        sb->s_magic = CGROUP_SUPER_MAGIC;
1537        sb->s_op = &cgroup_ops;
1538
1539        return 0;
1540}
1541
1542static int cgroup_get_rootdir(struct super_block *sb)
1543{
1544        static const struct dentry_operations cgroup_dops = {
1545                .d_iput = cgroup_diput,
1546                .d_delete = cgroup_delete,
1547        };
1548
1549        struct inode *inode =
1550                cgroup_new_inode(S_IFDIR | S_IRUGO | S_IXUGO | S_IWUSR, sb);
1551
1552        if (!inode)
1553                return -ENOMEM;
1554
1555        inode->i_fop = &simple_dir_operations;
1556        inode->i_op = &cgroup_dir_inode_operations;
1557        /* directories start off with i_nlink == 2 (for "." entry) */
1558        inc_nlink(inode);
1559        sb->s_root = d_make_root(inode);
1560        if (!sb->s_root)
1561                return -ENOMEM;
1562        /* for everything else we want ->d_op set */
1563        sb->s_d_op = &cgroup_dops;
1564        return 0;
1565}
1566
1567static struct dentry *cgroup_mount(struct file_system_type *fs_type,
1568                         int flags, const char *unused_dev_name,
1569                         void *data)
1570{
1571        struct cgroup_sb_opts opts;
1572        struct cgroupfs_root *root;
1573        int ret = 0;
1574        struct super_block *sb;
1575        struct cgroupfs_root *new_root;
1576        struct inode *inode;
1577
1578        /* First find the desired set of subsystems */
1579        mutex_lock(&cgroup_mutex);
1580        ret = parse_cgroupfs_options(data, &opts);
1581        mutex_unlock(&cgroup_mutex);
1582        if (ret)
1583                goto out_err;
1584
1585        /*
1586         * Allocate a new cgroup root. We may not need it if we're
1587         * reusing an existing hierarchy.
1588         */
1589        new_root = cgroup_root_from_opts(&opts);
1590        if (IS_ERR(new_root)) {
1591                ret = PTR_ERR(new_root);
1592                goto drop_modules;
1593        }
1594        opts.new_root = new_root;
1595
1596        /* Locate an existing or new sb for this hierarchy */
1597        sb = sget(fs_type, cgroup_test_super, cgroup_set_super, 0, &opts);
1598        if (IS_ERR(sb)) {
1599                ret = PTR_ERR(sb);
1600                cgroup_drop_root(opts.new_root);
1601                goto drop_modules;
1602        }
1603
1604        root = sb->s_fs_info;
1605        BUG_ON(!root);
1606        if (root == opts.new_root) {
1607                /* We used the new root structure, so this is a new hierarchy */
1608                struct list_head tmp_cg_links;
1609                struct cgroup *root_cgrp = &root->top_cgroup;
1610                struct cgroupfs_root *existing_root;
1611                const struct cred *cred;
1612                int i;
1613                struct css_set *cg;
1614
1615                BUG_ON(sb->s_root != NULL);
1616
1617                ret = cgroup_get_rootdir(sb);
1618                if (ret)
1619                        goto drop_new_super;
1620                inode = sb->s_root->d_inode;
1621
1622                mutex_lock(&inode->i_mutex);
1623                mutex_lock(&cgroup_mutex);
1624                mutex_lock(&cgroup_root_mutex);
1625
1626                /* Check for name clashes with existing mounts */
1627                ret = -EBUSY;
1628                if (strlen(root->name))
1629                        for_each_active_root(existing_root)
1630                                if (!strcmp(existing_root->name, root->name))
1631                                        goto unlock_drop;
1632
1633                /*
1634                 * We're accessing css_set_count without locking
1635                 * css_set_lock here, but that's OK - it can only be
1636                 * increased by someone holding cgroup_lock, and
1637                 * that's us. The worst that can happen is that we
1638                 * have some link structures left over
1639                 */
1640                ret = allocate_cg_links(css_set_count, &tmp_cg_links);
1641                if (ret)
1642                        goto unlock_drop;
1643
1644                ret = rebind_subsystems(root, root->subsys_mask);
1645                if (ret == -EBUSY) {
1646                        free_cg_links(&tmp_cg_links);
1647                        goto unlock_drop;
1648                }
1649                /*
1650                 * There must be no failure case after here, since rebinding
1651                 * takes care of subsystems' refcounts, which are explicitly
1652                 * dropped in the failure exit path.
1653                 */
1654
1655                /* EBUSY should be the only error here */
1656                BUG_ON(ret);
1657
1658                list_add(&root->root_list, &roots);
1659                root_count++;
1660
1661                sb->s_root->d_fsdata = root_cgrp;
1662                root->top_cgroup.dentry = sb->s_root;
1663
1664                /* Link the top cgroup in this hierarchy into all
1665                 * the css_set objects */
1666                write_lock(&css_set_lock);
1667                hash_for_each(css_set_table, i, cg, hlist)
1668                        link_css_set(&tmp_cg_links, cg, root_cgrp);
1669                write_unlock(&css_set_lock);
1670
1671                free_cg_links(&tmp_cg_links);
1672
1673                BUG_ON(!list_empty(&root_cgrp->children));
1674                BUG_ON(root->number_of_cgroups != 1);
1675
1676                cred = override_creds(&init_cred);
1677                cgroup_populate_dir(root_cgrp, true, root->subsys_mask);
1678                revert_creds(cred);
1679                mutex_unlock(&cgroup_root_mutex);
1680                mutex_unlock(&cgroup_mutex);
1681                mutex_unlock(&inode->i_mutex);
1682        } else {
1683                /*
1684                 * We re-used an existing hierarchy - the new root (if
1685                 * any) is not needed
1686                 */
1687                cgroup_drop_root(opts.new_root);
1688                /* no subsys rebinding, so refcounts don't change */
1689                drop_parsed_module_refcounts(opts.subsys_mask);
1690        }
1691
1692        kfree(opts.release_agent);
1693        kfree(opts.name);
1694        return dget(sb->s_root);
1695
1696 unlock_drop:
1697        mutex_unlock(&cgroup_root_mutex);
1698        mutex_unlock(&cgroup_mutex);
1699        mutex_unlock(&inode->i_mutex);
1700 drop_new_super:
1701        deactivate_locked_super(sb);
1702 drop_modules:
1703        drop_parsed_module_refcounts(opts.subsys_mask);
1704 out_err:
1705        kfree(opts.release_agent);
1706        kfree(opts.name);
1707        return ERR_PTR(ret);
1708}
1709
1710static void cgroup_kill_sb(struct super_block *sb) {
1711        struct cgroupfs_root *root = sb->s_fs_info;
1712        struct cgroup *cgrp = &root->top_cgroup;
1713        int ret;
1714        struct cg_cgroup_link *link;
1715        struct cg_cgroup_link *saved_link;
1716
1717        BUG_ON(!root);
1718
1719        BUG_ON(root->number_of_cgroups != 1);
1720        BUG_ON(!list_empty(&cgrp->children));
1721
1722        mutex_lock(&cgroup_mutex);
1723        mutex_lock(&cgroup_root_mutex);
1724
1725        /* Rebind all subsystems back to the default hierarchy */
1726        ret = rebind_subsystems(root, 0);
1727        /* Shouldn't be able to fail ... */
1728        BUG_ON(ret);
1729
1730        /*
1731         * Release all the links from css_sets to this hierarchy's
1732         * root cgroup
1733         */
1734        write_lock(&css_set_lock);
1735
1736        list_for_each_entry_safe(link, saved_link, &cgrp->css_sets,
1737                                 cgrp_link_list) {
1738                list_del(&link->cg_link_list);
1739                list_del(&link->cgrp_link_list);
1740                kfree(link);
1741        }
1742        write_unlock(&css_set_lock);
1743
1744        if (!list_empty(&root->root_list)) {
1745                list_del(&root->root_list);
1746                root_count--;
1747        }
1748
1749        mutex_unlock(&cgroup_root_mutex);
1750        mutex_unlock(&cgroup_mutex);
1751
1752        simple_xattrs_free(&cgrp->xattrs);
1753
1754        kill_litter_super(sb);
1755        cgroup_drop_root(root);
1756}
1757
1758static struct file_system_type cgroup_fs_type = {
1759        .name = "cgroup",
1760        .mount = cgroup_mount,
1761        .kill_sb = cgroup_kill_sb,
1762};
1763
1764static struct kobject *cgroup_kobj;
1765
1766/**
1767 * cgroup_path - generate the path of a cgroup
1768 * @cgrp: the cgroup in question
1769 * @buf: the buffer to write the path into
1770 * @buflen: the length of the buffer
1771 *
1772 * Called with cgroup_mutex held or else with an RCU-protected cgroup
1773 * reference.  Writes path of cgroup into buf.  Returns 0 on success,
1774 * -errno on error.
1775 */
1776int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen)
1777{
1778        struct dentry *dentry = cgrp->dentry;
1779        char *start;
1780
1781        rcu_lockdep_assert(rcu_read_lock_held() || cgroup_lock_is_held(),
1782                           "cgroup_path() called without proper locking");
1783
1784        if (cgrp == dummytop) {
1785                /*
1786                 * Inactive subsystems have no dentry for their root
1787                 * cgroup
1788                 */
1789                strcpy(buf, "/");
1790                return 0;
1791        }
1792
1793        start = buf + buflen - 1;
1794
1795        *start = '\0';
1796        for (;;) {
1797                int len = dentry->d_name.len;
1798
1799                if ((start -= len) < buf)
1800                        return -ENAMETOOLONG;
1801                memcpy(start, dentry->d_name.name, len);
1802                cgrp = cgrp->parent;
1803                if (!cgrp)
1804                        break;
1805
1806                dentry = cgrp->dentry;
1807                if (!cgrp->parent)
1808                        continue;
1809                if (--start < buf)
1810                        return -ENAMETOOLONG;
1811                *start = '/';
1812        }
1813        memmove(buf, start, buf + buflen - start);
1814        return 0;
1815}
1816EXPORT_SYMBOL_GPL(cgroup_path);
1817
1818/*
1819 * Control Group taskset
1820 */
1821struct task_and_cgroup {
1822        struct task_struct      *task;
1823        struct cgroup           *cgrp;
1824        struct css_set          *cg;
1825};
1826
1827struct cgroup_taskset {
1828        struct task_and_cgroup  single;
1829        struct flex_array       *tc_array;
1830        int                     tc_array_len;
1831        int                     idx;
1832        struct cgroup           *cur_cgrp;
1833};
1834
1835/**
1836 * cgroup_taskset_first - reset taskset and return the first task
1837 * @tset: taskset of interest
1838 *
1839 * @tset iteration is initialized and the first task is returned.
1840 */
1841struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset)
1842{
1843        if (tset->tc_array) {
1844                tset->idx = 0;
1845                return cgroup_taskset_next(tset);
1846        } else {
1847                tset->cur_cgrp = tset->single.cgrp;
1848                return tset->single.task;
1849        }
1850}
1851EXPORT_SYMBOL_GPL(cgroup_taskset_first);
1852
1853/**
1854 * cgroup_taskset_next - iterate to the next task in taskset
1855 * @tset: taskset of interest
1856 *
1857 * Return the next task in @tset.  Iteration must have been initialized
1858 * with cgroup_taskset_first().
1859 */
1860struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset)
1861{
1862        struct task_and_cgroup *tc;
1863
1864        if (!tset->tc_array || tset->idx >= tset->tc_array_len)
1865                return NULL;
1866
1867        tc = flex_array_get(tset->tc_array, tset->idx++);
1868        tset->cur_cgrp = tc->cgrp;
1869        return tc->task;
1870}
1871EXPORT_SYMBOL_GPL(cgroup_taskset_next);
1872
1873/**
1874 * cgroup_taskset_cur_cgroup - return the matching cgroup for the current task
1875 * @tset: taskset of interest
1876 *
1877 * Return the cgroup for the current (last returned) task of @tset.  This
1878 * function must be preceded by either cgroup_taskset_first() or
1879 * cgroup_taskset_next().
1880 */
1881struct cgroup *cgroup_taskset_cur_cgroup(struct cgroup_taskset *tset)
1882{
1883        return tset->cur_cgrp;
1884}
1885EXPORT_SYMBOL_GPL(cgroup_taskset_cur_cgroup);
1886
1887/**
1888 * cgroup_taskset_size - return the number of tasks in taskset
1889 * @tset: taskset of interest
1890 */
1891int cgroup_taskset_size(struct cgroup_taskset *tset)
1892{
1893        return tset->tc_array ? tset->tc_array_len : 1;
1894}
1895EXPORT_SYMBOL_GPL(cgroup_taskset_size);
1896
1897
1898/*
1899 * cgroup_task_migrate - move a task from one cgroup to another.
1900 *
1901 * Must be called with cgroup_mutex and threadgroup locked.
1902 */
1903static void cgroup_task_migrate(struct cgroup *cgrp, struct cgroup *oldcgrp,
1904                                struct task_struct *tsk, struct css_set *newcg)
1905{
1906        struct css_set *oldcg;
1907
1908        /*
1909         * We are synchronized through threadgroup_lock() against PF_EXITING
1910         * setting such that we can't race against cgroup_exit() changing the
1911         * css_set to init_css_set and dropping the old one.
1912         */
1913        WARN_ON_ONCE(tsk->flags & PF_EXITING);
1914        oldcg = tsk->cgroups;
1915
1916        task_lock(tsk);
1917        rcu_assign_pointer(tsk->cgroups, newcg);
1918        task_unlock(tsk);
1919
1920        /* Update the css_set linked lists if we're using them */
1921        write_lock(&css_set_lock);
1922        if (!list_empty(&tsk->cg_list))
1923                list_move(&tsk->cg_list, &newcg->tasks);
1924        write_unlock(&css_set_lock);
1925
1926        /*
1927         * We just gained a reference on oldcg by taking it from the task. As
1928         * trading it for newcg is protected by cgroup_mutex, we're safe to drop
1929         * it here; it will be freed under RCU.
1930         */
1931        set_bit(CGRP_RELEASABLE, &oldcgrp->flags);
1932        put_css_set(oldcg);
1933}
1934
1935/**
1936 * cgroup_attach_task - attach task 'tsk' to cgroup 'cgrp'
1937 * @cgrp: the cgroup the task is attaching to
1938 * @tsk: the task to be attached
1939 *
1940 * Call with cgroup_mutex and threadgroup locked. May take task_lock of
1941 * @tsk during call.
1942 */
1943int cgroup_attach_task(struct cgroup *cgrp, struct task_struct *tsk)
1944{
1945        int retval = 0;
1946        struct cgroup_subsys *ss, *failed_ss = NULL;
1947        struct cgroup *oldcgrp;
1948        struct cgroupfs_root *root = cgrp->root;
1949        struct cgroup_taskset tset = { };
1950        struct css_set *newcg;
1951
1952        /* @tsk either already exited or can't exit until the end */
1953        if (tsk->flags & PF_EXITING)
1954                return -ESRCH;
1955
1956        /* Nothing to do if the task is already in that cgroup */
1957        oldcgrp = task_cgroup_from_root(tsk, root);
1958        if (cgrp == oldcgrp)
1959                return 0;
1960
1961        tset.single.task = tsk;
1962        tset.single.cgrp = oldcgrp;
1963
1964        for_each_subsys(root, ss) {
1965                if (ss->can_attach) {
1966                        retval = ss->can_attach(cgrp, &tset);
1967                        if (retval) {
1968                                /*
1969                                 * Remember on which subsystem the can_attach()
1970                                 * failed, so that we only call cancel_attach()
1971                                 * against the subsystems whose can_attach()
1972                                 * succeeded. (See below)
1973                                 */
1974                                failed_ss = ss;
1975                                goto out;
1976                        }
1977                }
1978        }
1979
1980        newcg = find_css_set(tsk->cgroups, cgrp);
1981        if (!newcg) {
1982                retval = -ENOMEM;
1983                goto out;
1984        }
1985
1986        cgroup_task_migrate(cgrp, oldcgrp, tsk, newcg);
1987
1988        for_each_subsys(root, ss) {
1989                if (ss->attach)
1990                        ss->attach(cgrp, &tset);
1991        }
1992
1993out:
1994        if (retval) {
1995                for_each_subsys(root, ss) {
1996                        if (ss == failed_ss)
1997                                /*
1998                                 * This subsystem was the one that failed the
1999                                 * can_attach() check earlier, so we don't need
2000                                 * to call cancel_attach() against it or any
2001                                 * remaining subsystems.
2002                                 */
2003                                break;
2004                        if (ss->cancel_attach)
2005                                ss->cancel_attach(cgrp, &tset);
2006                }
2007        }
2008        return retval;
2009}
2010
2011/**
2012 * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from'
2013 * @from: attach to all cgroups of a given task
2014 * @tsk: the task to be attached
2015 */
2016int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk)
2017{
2018        struct cgroupfs_root *root;
2019        int retval = 0;
2020
2021        cgroup_lock();
2022        for_each_active_root(root) {
2023                struct cgroup *from_cg = task_cgroup_from_root(from, root);
2024
2025                retval = cgroup_attach_task(from_cg, tsk);
2026                if (retval)
2027                        break;
2028        }
2029        cgroup_unlock();
2030
2031        return retval;
2032}
2033EXPORT_SYMBOL_GPL(cgroup_attach_task_all);
2034
2035/**
2036 * cgroup_attach_proc - attach all threads in a threadgroup to a cgroup
2037 * @cgrp: the cgroup to attach to
2038 * @leader: the threadgroup leader task_struct of the group to be attached
2039 *
2040 * Call holding cgroup_mutex and the group_rwsem of the leader. Will take
2041 * task_lock of each thread in leader's threadgroup individually in turn.
2042 */
2043static int cgroup_attach_proc(struct cgroup *cgrp, struct task_struct *leader)
2044{
2045        int retval, i, group_size;
2046        struct cgroup_subsys *ss, *failed_ss = NULL;
2047        /* guaranteed to be initialized later, but the compiler needs this */
2048        struct cgroupfs_root *root = cgrp->root;
2049        /* threadgroup list cursor and array */
2050        struct task_struct *tsk;
2051        struct task_and_cgroup *tc;
2052        struct flex_array *group;
2053        struct cgroup_taskset tset = { };
2054
2055        /*
2056         * step 0: in order to do expensive, possibly blocking operations for
2057         * every thread, we cannot iterate the thread group list, since it needs
2058         * rcu or tasklist locked. instead, build an array of all threads in the
2059         * group - group_rwsem prevents new threads from appearing, and if
2060         * threads exit, this will just be an over-estimate.
2061         */
2062        group_size = get_nr_threads(leader);
2063        /* flex_array supports very large thread-groups better than kmalloc. */
2064        group = flex_array_alloc(sizeof(*tc), group_size, GFP_KERNEL);
2065        if (!group)
2066                return -ENOMEM;
2067        /* pre-allocate to guarantee space while iterating in rcu read-side. */
2068        retval = flex_array_prealloc(group, 0, group_size - 1, GFP_KERNEL);
2069        if (retval)
2070                goto out_free_group_list;
2071
2072        tsk = leader;
2073        i = 0;
2074        /*
2075         * Prevent freeing of tasks while we take a snapshot. Tasks that are
2076         * already PF_EXITING could be freed from underneath us unless we
2077         * take an rcu_read_lock.
2078         */
2079        rcu_read_lock();
2080        do {
2081                struct task_and_cgroup ent;
2082
2083                /* @tsk either already exited or can't exit until the end */
2084                if (tsk->flags & PF_EXITING)
2085                        continue;
2086
2087                /* as per above, nr_threads may decrease, but not increase. */
2088                BUG_ON(i >= group_size);
2089                ent.task = tsk;
2090                ent.cgrp = task_cgroup_from_root(tsk, root);
2091                /* nothing to do if this task is already in the cgroup */
2092                if (ent.cgrp == cgrp)
2093                        continue;
2094                /*
2095                 * saying GFP_ATOMIC has no effect here because we did prealloc
2096                 * earlier, but it's good form to communicate our expectations.
2097                 */
2098                retval = flex_array_put(group, i, &ent, GFP_ATOMIC);
2099                BUG_ON(retval != 0);
2100                i++;
2101        } while_each_thread(leader, tsk);
2102        rcu_read_unlock();
2103        /* remember the number of threads in the array for later. */
2104        group_size = i;
2105        tset.tc_array = group;
2106        tset.tc_array_len = group_size;
2107
2108        /* methods shouldn't be called if no task is actually migrating */
2109        retval = 0;
2110        if (!group_size)
2111                goto out_free_group_list;
2112
2113        /*
2114         * step 1: check that we can legitimately attach to the cgroup.
2115         */
2116        for_each_subsys(root, ss) {
2117                if (ss->can_attach) {
2118                        retval = ss->can_attach(cgrp, &tset);
2119                        if (retval) {
2120                                failed_ss = ss;
2121                                goto out_cancel_attach;
2122                        }
2123                }
2124        }
2125
2126        /*
2127         * step 2: make sure css_sets exist for all threads to be migrated.
2128         * we use find_css_set, which allocates a new one if necessary.
2129         */
2130        for (i = 0; i < group_size; i++) {
2131                tc = flex_array_get(group, i);
2132                tc->cg = find_css_set(tc->task->cgroups, cgrp);
2133                if (!tc->cg) {
2134                        retval = -ENOMEM;
2135                        goto out_put_css_set_refs;
2136                }
2137        }
2138
2139        /*
2140         * step 3: now that we're guaranteed success wrt the css_sets,
2141         * proceed to move all tasks to the new cgroup.  There are no
2142         * failure cases after here, so this is the commit point.
2143         */
2144        for (i = 0; i < group_size; i++) {
2145                tc = flex_array_get(group, i);
2146                cgroup_task_migrate(cgrp, tc->cgrp, tc->task, tc->cg);
2147        }
2148        /* nothing is sensitive to fork() after this point. */
2149
2150        /*
2151         * step 4: do subsystem attach callbacks.
2152         */
2153        for_each_subsys(root, ss) {
2154                if (ss->attach)
2155                        ss->attach(cgrp, &tset);
2156        }
2157
2158        /*
2159         * step 5: success! and cleanup
2160         */
2161        retval = 0;
2162out_put_css_set_refs:
2163        if (retval) {
2164                for (i = 0; i < group_size; i++) {
2165                        tc = flex_array_get(group, i);
2166                        if (!tc->cg)
2167                                break;
2168                        put_css_set(tc->cg);
2169                }
2170        }
2171out_cancel_attach:
2172        if (retval) {
2173                for_each_subsys(root, ss) {
2174                        if (ss == failed_ss)
2175                                break;
2176                        if (ss->cancel_attach)
2177                                ss->cancel_attach(cgrp, &tset);
2178                }
2179        }
2180out_free_group_list:
2181        flex_array_free(group);
2182        return retval;
2183}
2184
2185/*
2186 * Find the task_struct of the task to attach by vpid and pass it along to the
2187 * function to attach either it or all tasks in its threadgroup. Will lock
2188 * cgroup_mutex and threadgroup; may take task_lock of task.
2189 */
2190static int attach_task_by_pid(struct cgroup *cgrp, u64 pid, bool threadgroup)
2191{
2192        struct task_struct *tsk;
2193        const struct cred *cred = current_cred(), *tcred;
2194        int ret;
2195
2196        if (!cgroup_lock_live_group(cgrp))
2197                return -ENODEV;
2198
2199retry_find_task:
2200        rcu_read_lock();
2201        if (pid) {
2202                tsk = find_task_by_vpid(pid);
2203                if (!tsk) {
2204                        rcu_read_unlock();
2205                        ret= -ESRCH;
2206                        goto out_unlock_cgroup;
2207                }
2208                /*
2209                 * even if we're attaching all tasks in the thread group, we
2210                 * only need to check permissions on one of them.
2211                 */
2212                tcred = __task_cred(tsk);
2213                if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) &&
2214                    !uid_eq(cred->euid, tcred->uid) &&
2215                    !uid_eq(cred->euid, tcred->suid)) {
2216                        rcu_read_unlock();
2217                        ret = -EACCES;
2218                        goto out_unlock_cgroup;
2219                }
2220        } else
2221                tsk = current;
2222
2223        if (threadgroup)
2224                tsk = tsk->group_leader;
2225
2226        /*
2227         * Workqueue threads may acquire PF_THREAD_BOUND and become
2228         * trapped in a cpuset, or RT worker may be born in a cgroup
2229         * with no rt_runtime allocated.  Just say no.
2230         */
2231        if (tsk == kthreadd_task || (tsk->flags & PF_THREAD_BOUND)) {
2232                ret = -EINVAL;
2233                rcu_read_unlock();
2234                goto out_unlock_cgroup;
2235        }
2236
2237        get_task_struct(tsk);
2238        rcu_read_unlock();
2239
2240        threadgroup_lock(tsk);
2241        if (threadgroup) {
2242                if (!thread_group_leader(tsk)) {
2243                        /*
2244                         * a race with de_thread from another thread's exec()
2245                         * may strip us of our leadership, if this happens,
2246                         * there is no choice but to throw this task away and
2247                         * try again; this is
2248                         * "double-double-toil-and-trouble-check locking".
2249                         */
2250                        threadgroup_unlock(tsk);
2251                        put_task_struct(tsk);
2252                        goto retry_find_task;
2253                }
2254                ret = cgroup_attach_proc(cgrp, tsk);
2255        } else
2256                ret = cgroup_attach_task(cgrp, tsk);
2257        threadgroup_unlock(tsk);
2258
2259        put_task_struct(tsk);
2260out_unlock_cgroup:
2261        cgroup_unlock();
2262        return ret;
2263}
2264
2265static int cgroup_tasks_write(struct cgroup *cgrp, struct cftype *cft, u64 pid)
2266{
2267        return attach_task_by_pid(cgrp, pid, false);
2268}
2269
2270static int cgroup_procs_write(struct cgroup *cgrp, struct cftype *cft, u64 tgid)
2271{
2272        return attach_task_by_pid(cgrp, tgid, true);
2273}
2274
2275/**
2276 * cgroup_lock_live_group - take cgroup_mutex and check that cgrp is alive.
2277 * @cgrp: the cgroup to be checked for liveness
2278 *
2279 * On success, returns true; the lock should be later released with
2280 * cgroup_unlock(). On failure returns false with no lock held.
2281 */
2282bool cgroup_lock_live_group(struct cgroup *cgrp)
2283{
2284        mutex_lock(&cgroup_mutex);
2285        if (cgroup_is_removed(cgrp)) {
2286                mutex_unlock(&cgroup_mutex);
2287                return false;
2288        }
2289        return true;
2290}
2291EXPORT_SYMBOL_GPL(cgroup_lock_live_group);
2292
2293static int cgroup_release_agent_write(struct cgroup *cgrp, struct cftype *cft,
2294                                      const char *buffer)
2295{
2296        BUILD_BUG_ON(sizeof(cgrp->root->release_agent_path) < PATH_MAX);
2297        if (strlen(buffer) >= PATH_MAX)
2298                return -EINVAL;
2299        if (!cgroup_lock_live_group(cgrp))
2300                return -ENODEV;
2301        mutex_lock(&cgroup_root_mutex);
2302        strcpy(cgrp->root->release_agent_path, buffer);
2303        mutex_unlock(&cgroup_root_mutex);
2304        cgroup_unlock();
2305        return 0;
2306}
2307
2308static int cgroup_release_agent_show(struct cgroup *cgrp, struct cftype *cft,
2309                                     struct seq_file *seq)
2310{
2311        if (!cgroup_lock_live_group(cgrp))
2312                return -ENODEV;
2313        seq_puts(seq, cgrp->root->release_agent_path);
2314        seq_putc(seq, '\n');
2315        cgroup_unlock();
2316        return 0;
2317}
2318
2319/* A buffer size big enough for numbers or short strings */
2320#define CGROUP_LOCAL_BUFFER_SIZE 64
2321
2322static ssize_t cgroup_write_X64(struct cgroup *cgrp, struct cftype *cft,
2323                                struct file *file,
2324                                const char __user *userbuf,
2325                                size_t nbytes, loff_t *unused_ppos)
2326{
2327        char buffer[CGROUP_LOCAL_BUFFER_SIZE];
2328        int retval = 0;
2329        char *end;
2330
2331        if (!nbytes)
2332                return -EINVAL;
2333        if (nbytes >= sizeof(buffer))
2334                return -E2BIG;
2335        if (copy_from_user(buffer, userbuf, nbytes))
2336                return -EFAULT;
2337
2338        buffer[nbytes] = 0;     /* nul-terminate */
2339        if (cft->write_u64) {
2340                u64 val = simple_strtoull(strstrip(buffer), &end, 0);
2341                if (*end)
2342                        return -EINVAL;
2343                retval = cft->write_u64(cgrp, cft, val);
2344        } else {
2345                s64 val = simple_strtoll(strstrip(buffer), &end, 0);
2346                if (*end)
2347                        return -EINVAL;
2348                retval = cft->write_s64(cgrp, cft, val);
2349        }
2350        if (!retval)
2351                retval = nbytes;
2352        return retval;
2353}
2354
2355static ssize_t cgroup_write_string(struct cgroup *cgrp, struct cftype *cft,
2356                                   struct file *file,
2357                                   const char __user *userbuf,
2358                                   size_t nbytes, loff_t *unused_ppos)
2359{
2360        char local_buffer[CGROUP_LOCAL_BUFFER_SIZE];
2361        int retval = 0;
2362        size_t max_bytes = cft->max_write_len;
2363        char *buffer = local_buffer;
2364
2365        if (!max_bytes)
2366                max_bytes = sizeof(local_buffer) - 1;
2367        if (nbytes >= max_bytes)
2368                return -E2BIG;
2369        /* Allocate a dynamic buffer if we need one */
2370        if (nbytes >= sizeof(local_buffer)) {
2371                buffer = kmalloc(nbytes + 1, GFP_KERNEL);
2372                if (buffer == NULL)
2373                        return -ENOMEM;
2374        }
2375        if (nbytes && copy_from_user(buffer, userbuf, nbytes)) {
2376                retval = -EFAULT;
2377                goto out;
2378        }
2379
2380        buffer[nbytes] = 0;     /* nul-terminate */
2381        retval = cft->write_string(cgrp, cft, strstrip(buffer));
2382        if (!retval)
2383                retval = nbytes;
2384out:
2385        if (buffer != local_buffer)
2386                kfree(buffer);
2387        return retval;
2388}
2389
2390static ssize_t cgroup_file_write(struct file *file, const char __user *buf,
2391                                                size_t nbytes, loff_t *ppos)
2392{
2393        struct cftype *cft = __d_cft(file->f_dentry);
2394        struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
2395
2396        if (cgroup_is_removed(cgrp))
2397                return -ENODEV;
2398        if (cft->write)
2399                return cft->write(cgrp, cft, file, buf, nbytes, ppos);
2400        if (cft->write_u64 || cft->write_s64)
2401                return cgroup_write_X64(cgrp, cft, file, buf, nbytes, ppos);
2402        if (cft->write_string)
2403                return cgroup_write_string(cgrp, cft, file, buf, nbytes, ppos);
2404        if (cft->trigger) {
2405                int ret = cft->trigger(cgrp, (unsigned int)cft->private);
2406                return ret ? ret : nbytes;
2407        }
2408        return -EINVAL;
2409}
2410
2411static ssize_t cgroup_read_u64(struct cgroup *cgrp, struct cftype *cft,
2412                               struct file *file,
2413                               char __user *buf, size_t nbytes,
2414                               loff_t *ppos)
2415{
2416        char tmp[CGROUP_LOCAL_BUFFER_SIZE];
2417        u64 val = cft->read_u64(cgrp, cft);
2418        int len = sprintf(tmp, "%llu\n", (unsigned long long) val);
2419
2420        return simple_read_from_buffer(buf, nbytes, ppos, tmp, len);
2421}
2422
2423static ssize_t cgroup_read_s64(struct cgroup *cgrp, struct cftype *cft,
2424                               struct file *file,
2425                               char __user *buf, size_t nbytes,
2426                               loff_t *ppos)
2427{
2428        char tmp[CGROUP_LOCAL_BUFFER_SIZE];
2429        s64 val = cft->read_s64(cgrp, cft);
2430        int len = sprintf(tmp, "%lld\n", (long long) val);
2431
2432        return simple_read_from_buffer(buf, nbytes, ppos, tmp, len);
2433}
2434
2435static ssize_t cgroup_file_read(struct file *file, char __user *buf,
2436                                   size_t nbytes, loff_t *ppos)
2437{
2438        struct cftype *cft = __d_cft(file->f_dentry);
2439        struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
2440
2441        if (cgroup_is_removed(cgrp))
2442                return -ENODEV;
2443
2444        if (cft->read)
2445                return cft->read(cgrp, cft, file, buf, nbytes, ppos);
2446        if (cft->read_u64)
2447                return cgroup_read_u64(cgrp, cft, file, buf, nbytes, ppos);
2448        if (cft->read_s64)
2449                return cgroup_read_s64(cgrp, cft, file, buf, nbytes, ppos);
2450        return -EINVAL;
2451}
2452
2453/*
2454 * seqfile ops/methods for returning structured data. Currently just
2455 * supports string->u64 maps, but can be extended in future.
2456 */
2457
2458struct cgroup_seqfile_state {
2459        struct cftype *cft;
2460        struct cgroup *cgroup;
2461};
2462
2463static int cgroup_map_add(struct cgroup_map_cb *cb, const char *key, u64 value)
2464{
2465        struct seq_file *sf = cb->state;
2466        return seq_printf(sf, "%s %llu\n", key, (unsigned long long)value);
2467}
2468
2469static int cgroup_seqfile_show(struct seq_file *m, void *arg)
2470{
2471        struct cgroup_seqfile_state *state = m->private;
2472        struct cftype *cft = state->cft;
2473        if (cft->read_map) {
2474                struct cgroup_map_cb cb = {
2475                        .fill = cgroup_map_add,
2476                        .state = m,
2477                };
2478                return cft->read_map(state->cgroup, cft, &cb);
2479        }
2480        return cft->read_seq_string(state->cgroup, cft, m);
2481}
2482
2483static int cgroup_seqfile_release(struct inode *inode, struct file *file)
2484{
2485        struct seq_file *seq = file->private_data;
2486        kfree(seq->private);
2487        return single_release(inode, file);
2488}
2489
2490static const struct file_operations cgroup_seqfile_operations = {
2491        .read = seq_read,
2492        .write = cgroup_file_write,
2493        .llseek = seq_lseek,
2494        .release = cgroup_seqfile_release,
2495};
2496
2497static int cgroup_file_open(struct inode *inode, struct file *file)
2498{
2499        int err;
2500        struct cftype *cft;
2501
2502        err = generic_file_open(inode, file);
2503        if (err)
2504                return err;
2505        cft = __d_cft(file->f_dentry);
2506
2507        if (cft->read_map || cft->read_seq_string) {
2508                struct cgroup_seqfile_state *state =
2509                        kzalloc(sizeof(*state), GFP_USER);
2510                if (!state)
2511                        return -ENOMEM;
2512                state->cft = cft;
2513                state->cgroup = __d_cgrp(file->f_dentry->d_parent);
2514                file->f_op = &cgroup_seqfile_operations;
2515                err = single_open(file, cgroup_seqfile_show, state);
2516                if (err < 0)
2517                        kfree(state);
2518        } else if (cft->open)
2519                err = cft->open(inode, file);
2520        else
2521                err = 0;
2522
2523        return err;
2524}
2525
2526static int cgroup_file_release(struct inode *inode, struct file *file)
2527{
2528        struct cftype *cft = __d_cft(file->f_dentry);
2529        if (cft->release)
2530                return cft->release(inode, file);
2531        return 0;
2532}
2533
2534/*
2535 * cgroup_rename - Only allow simple rename of directories in place.
2536 */
2537static int cgroup_rename(struct inode *old_dir, struct dentry *old_dentry,
2538                            struct inode *new_dir, struct dentry *new_dentry)
2539{
2540        if (!S_ISDIR(old_dentry->d_inode->i_mode))
2541                return -ENOTDIR;
2542        if (new_dentry->d_inode)
2543                return -EEXIST;
2544        if (old_dir != new_dir)
2545                return -EIO;
2546        return simple_rename(old_dir, old_dentry, new_dir, new_dentry);
2547}
2548
2549static struct simple_xattrs *__d_xattrs(struct dentry *dentry)
2550{
2551        if (S_ISDIR(dentry->d_inode->i_mode))
2552                return &__d_cgrp(dentry)->xattrs;
2553        else
2554                return &__d_cft(dentry)->xattrs;
2555}
2556
2557static inline int xattr_enabled(struct dentry *dentry)
2558{
2559        struct cgroupfs_root *root = dentry->d_sb->s_fs_info;
2560        return test_bit(ROOT_XATTR, &root->flags);
2561}
2562
2563static bool is_valid_xattr(const char *name)
2564{
2565        if (!strncmp(name, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN) ||
2566            !strncmp(name, XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN))
2567                return true;
2568        return false;
2569}
2570
2571static int cgroup_setxattr(struct dentry *dentry, const char *name,
2572                           const void *val, size_t size, int flags)
2573{
2574        if (!xattr_enabled(dentry))
2575                return -EOPNOTSUPP;
2576        if (!is_valid_xattr(name))
2577                return -EINVAL;
2578        return simple_xattr_set(__d_xattrs(dentry), name, val, size, flags);
2579}
2580
2581static int cgroup_removexattr(struct dentry *dentry, const char *name)
2582{
2583        if (!xattr_enabled(dentry))
2584                return -EOPNOTSUPP;
2585        if (!is_valid_xattr(name))
2586                return -EINVAL;
2587        return simple_xattr_remove(__d_xattrs(dentry), name);
2588}
2589
2590static ssize_t cgroup_getxattr(struct dentry *dentry, const char *name,
2591                               void *buf, size_t size)
2592{
2593        if (!xattr_enabled(dentry))
2594                return -EOPNOTSUPP;
2595        if (!is_valid_xattr(name))
2596                return -EINVAL;
2597        return simple_xattr_get(__d_xattrs(dentry), name, buf, size);
2598}
2599
2600static ssize_t cgroup_listxattr(struct dentry *dentry, char *buf, size_t size)
2601{
2602        if (!xattr_enabled(dentry))
2603                return -EOPNOTSUPP;
2604        return simple_xattr_list(__d_xattrs(dentry), buf, size);
2605}
2606
2607static const struct file_operations cgroup_file_operations = {
2608        .read = cgroup_file_read,
2609        .write = cgroup_file_write,
2610        .llseek = generic_file_llseek,
2611        .open = cgroup_file_open,
2612        .release = cgroup_file_release,
2613};
2614
2615static const struct inode_operations cgroup_file_inode_operations = {
2616        .setxattr = cgroup_setxattr,
2617        .getxattr = cgroup_getxattr,
2618        .listxattr = cgroup_listxattr,
2619        .removexattr = cgroup_removexattr,
2620};
2621
2622static const struct inode_operations cgroup_dir_inode_operations = {
2623        .lookup = cgroup_lookup,
2624        .mkdir = cgroup_mkdir,
2625        .rmdir = cgroup_rmdir,
2626        .rename = cgroup_rename,
2627        .setxattr = cgroup_setxattr,
2628        .getxattr = cgroup_getxattr,
2629        .listxattr = cgroup_listxattr,
2630        .removexattr = cgroup_removexattr,
2631};
2632
2633static struct dentry *cgroup_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
2634{
2635        if (dentry->d_name.len > NAME_MAX)
2636                return ERR_PTR(-ENAMETOOLONG);
2637        d_add(dentry, NULL);
2638        return NULL;
2639}
2640
2641/*
2642 * Check if a file is a control file
2643 */
2644static inline struct cftype *__file_cft(struct file *file)
2645{
2646        if (file_inode(file)->i_fop != &cgroup_file_operations)
2647                return ERR_PTR(-EINVAL);
2648        return __d_cft(file->f_dentry);
2649}
2650
2651static int cgroup_create_file(struct dentry *dentry, umode_t mode,
2652                                struct super_block *sb)
2653{
2654        struct inode *inode;
2655
2656        if (!dentry)
2657                return -ENOENT;
2658        if (dentry->d_inode)
2659                return -EEXIST;
2660
2661        inode = cgroup_new_inode(mode, sb);
2662        if (!inode)
2663                return -ENOMEM;
2664
2665        if (S_ISDIR(mode)) {
2666                inode->i_op = &cgroup_dir_inode_operations;
2667                inode->i_fop = &simple_dir_operations;
2668
2669                /* start off with i_nlink == 2 (for "." entry) */
2670                inc_nlink(inode);
2671                inc_nlink(dentry->d_parent->d_inode);
2672
2673                /*
2674                 * Control reaches here with cgroup_mutex held.
2675                 * @inode->i_mutex should nest outside cgroup_mutex but we
2676                 * want to populate it immediately without releasing
2677                 * cgroup_mutex.  As @inode isn't visible to anyone else
2678                 * yet, trylock will always succeed without affecting
2679                 * lockdep checks.
2680                 */
2681                WARN_ON_ONCE(!mutex_trylock(&inode->i_mutex));
2682        } else if (S_ISREG(mode)) {
2683                inode->i_size = 0;
2684                inode->i_fop = &cgroup_file_operations;
2685                inode->i_op = &cgroup_file_inode_operations;
2686        }
2687        d_instantiate(dentry, inode);
2688        dget(dentry);   /* Extra count - pin the dentry in core */
2689        return 0;
2690}
2691
2692/**
2693 * cgroup_file_mode - deduce file mode of a control file
2694 * @cft: the control file in question
2695 *
2696 * returns cft->mode if ->mode is not 0
2697 * returns S_IRUGO|S_IWUSR if it has both a read and a write handler
2698 * returns S_IRUGO if it has only a read handler
2699 * returns S_IWUSR if it has only a write hander
2700 */
2701static umode_t cgroup_file_mode(const struct cftype *cft)
2702{
2703        umode_t mode = 0;
2704
2705        if (cft->mode)
2706                return cft->mode;
2707
2708        if (cft->read || cft->read_u64 || cft->read_s64 ||
2709            cft->read_map || cft->read_seq_string)
2710                mode |= S_IRUGO;
2711
2712        if (cft->write || cft->write_u64 || cft->write_s64 ||
2713            cft->write_string || cft->trigger)
2714                mode |= S_IWUSR;
2715
2716        return mode;
2717}
2718
2719static int cgroup_add_file(struct cgroup *cgrp, struct cgroup_subsys *subsys,
2720                           struct cftype *cft)
2721{
2722        struct dentry *dir = cgrp->dentry;
2723        struct cgroup *parent = __d_cgrp(dir);
2724        struct dentry *dentry;
2725        struct cfent *cfe;
2726        int error;
2727        umode_t mode;
2728        char name[MAX_CGROUP_TYPE_NAMELEN + MAX_CFTYPE_NAME + 2] = { 0 };
2729
2730        simple_xattrs_init(&cft->xattrs);
2731
2732        if (subsys && !test_bit(ROOT_NOPREFIX, &cgrp->root->flags)) {
2733                strcpy(name, subsys->name);
2734                strcat(name, ".");
2735        }
2736        strcat(name, cft->name);
2737
2738        BUG_ON(!mutex_is_locked(&dir->d_inode->i_mutex));
2739
2740        cfe = kzalloc(sizeof(*cfe), GFP_KERNEL);
2741        if (!cfe)
2742                return -ENOMEM;
2743
2744        dentry = lookup_one_len(name, dir, strlen(name));
2745        if (IS_ERR(dentry)) {
2746                error = PTR_ERR(dentry);
2747                goto out;
2748        }
2749
2750        mode = cgroup_file_mode(cft);
2751        error = cgroup_create_file(dentry, mode | S_IFREG, cgrp->root->sb);
2752        if (!error) {
2753                cfe->type = (void *)cft;
2754                cfe->dentry = dentry;
2755                dentry->d_fsdata = cfe;
2756                list_add_tail(&cfe->node, &parent->files);
2757                cfe = NULL;
2758        }
2759        dput(dentry);
2760out:
2761        kfree(cfe);
2762        return error;
2763}
2764
2765static int cgroup_addrm_files(struct cgroup *cgrp, struct cgroup_subsys *subsys,
2766                              struct cftype cfts[], bool is_add)
2767{
2768        struct cftype *cft;
2769        int err, ret = 0;
2770
2771        for (cft = cfts; cft->name[0] != '\0'; cft++) {
2772                /* does cft->flags tell us to skip this file on @cgrp? */
2773                if ((cft->flags & CFTYPE_NOT_ON_ROOT) && !cgrp->parent)
2774                        continue;
2775                if ((cft->flags & CFTYPE_ONLY_ON_ROOT) && cgrp->parent)
2776                        continue;
2777
2778                if (is_add) {
2779                        err = cgroup_add_file(cgrp, subsys, cft);
2780                        if (err)
2781                                pr_warn("cgroup_addrm_files: failed to add %s, err=%d\n",
2782                                        cft->name, err);
2783                        ret = err;
2784                } else {
2785                        cgroup_rm_file(cgrp, cft);
2786                }
2787        }
2788        return ret;
2789}
2790
2791static DEFINE_MUTEX(cgroup_cft_mutex);
2792
2793static void cgroup_cfts_prepare(void)
2794        __acquires(&cgroup_cft_mutex) __acquires(&cgroup_mutex)
2795{
2796        /*
2797         * Thanks to the entanglement with vfs inode locking, we can't walk
2798         * the existing cgroups under cgroup_mutex and create files.
2799         * Instead, we increment reference on all cgroups and build list of
2800         * them using @cgrp->cft_q_node.  Grab cgroup_cft_mutex to ensure
2801         * exclusive access to the field.
2802         */
2803        mutex_lock(&cgroup_cft_mutex);
2804        mutex_lock(&cgroup_mutex);
2805}
2806
2807static void cgroup_cfts_commit(struct cgroup_subsys *ss,
2808                               struct cftype *cfts, bool is_add)
2809        __releases(&cgroup_mutex) __releases(&cgroup_cft_mutex)
2810{
2811        LIST_HEAD(pending);
2812        struct cgroup *cgrp, *n;
2813
2814        /* %NULL @cfts indicates abort and don't bother if @ss isn't attached */
2815        if (cfts && ss->root != &rootnode) {
2816                list_for_each_entry(cgrp, &ss->root->allcg_list, allcg_node) {
2817                        dget(cgrp->dentry);
2818                        list_add_tail(&cgrp->cft_q_node, &pending);
2819                }
2820        }
2821
2822        mutex_unlock(&cgroup_mutex);
2823
2824        /*
2825         * All new cgroups will see @cfts update on @ss->cftsets.  Add/rm
2826         * files for all cgroups which were created before.
2827         */
2828        list_for_each_entry_safe(cgrp, n, &pending, cft_q_node) {
2829                struct inode *inode = cgrp->dentry->d_inode;
2830
2831                mutex_lock(&inode->i_mutex);
2832                mutex_lock(&cgroup_mutex);
2833                if (!cgroup_is_removed(cgrp))
2834                        cgroup_addrm_files(cgrp, ss, cfts, is_add);
2835                mutex_unlock(&cgroup_mutex);
2836                mutex_unlock(&inode->i_mutex);
2837
2838                list_del_init(&cgrp->cft_q_node);
2839                dput(cgrp->dentry);
2840        }
2841
2842        mutex_unlock(&cgroup_cft_mutex);
2843}
2844
2845/**
2846 * cgroup_add_cftypes - add an array of cftypes to a subsystem
2847 * @ss: target cgroup subsystem
2848 * @cfts: zero-length name terminated array of cftypes
2849 *
2850 * Register @cfts to @ss.  Files described by @cfts are created for all
2851 * existing cgroups to which @ss is attached and all future cgroups will
2852 * have them too.  This function can be called anytime whether @ss is
2853 * attached or not.
2854 *
2855 * Returns 0 on successful registration, -errno on failure.  Note that this
2856 * function currently returns 0 as long as @cfts registration is successful
2857 * even if some file creation attempts on existing cgroups fail.
2858 */
2859int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
2860{
2861        struct cftype_set *set;
2862
2863        set = kzalloc(sizeof(*set), GFP_KERNEL);
2864        if (!set)
2865                return -ENOMEM;
2866
2867        cgroup_cfts_prepare();
2868        set->cfts = cfts;
2869        list_add_tail(&set->node, &ss->cftsets);
2870        cgroup_cfts_commit(ss, cfts, true);
2871
2872        return 0;
2873}
2874EXPORT_SYMBOL_GPL(cgroup_add_cftypes);
2875
2876/**
2877 * cgroup_rm_cftypes - remove an array of cftypes from a subsystem
2878 * @ss: target cgroup subsystem
2879 * @cfts: zero-length name terminated array of cftypes
2880 *
2881 * Unregister @cfts from @ss.  Files described by @cfts are removed from
2882 * all existing cgroups to which @ss is attached and all future cgroups
2883 * won't have them either.  This function can be called anytime whether @ss
2884 * is attached or not.
2885 *
2886 * Returns 0 on successful unregistration, -ENOENT if @cfts is not
2887 * registered with @ss.
2888 */
2889int cgroup_rm_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
2890{
2891        struct cftype_set *set;
2892
2893        cgroup_cfts_prepare();
2894
2895        list_for_each_entry(set, &ss->cftsets, node) {
2896                if (set->cfts == cfts) {
2897                        list_del_init(&set->node);
2898                        cgroup_cfts_commit(ss, cfts, false);
2899                        return 0;
2900                }
2901        }
2902
2903        cgroup_cfts_commit(ss, NULL, false);
2904        return -ENOENT;
2905}
2906
2907/**
2908 * cgroup_task_count - count the number of tasks in a cgroup.
2909 * @cgrp: the cgroup in question
2910 *
2911 * Return the number of tasks in the cgroup.
2912 */
2913int cgroup_task_count(const struct cgroup *cgrp)
2914{
2915        int count = 0;
2916        struct cg_cgroup_link *link;
2917
2918        read_lock(&css_set_lock);
2919        list_for_each_entry(link, &cgrp->css_sets, cgrp_link_list) {
2920                count += atomic_read(&link->cg->refcount);
2921        }
2922        read_unlock(&css_set_lock);
2923        return count;
2924}
2925
2926/*
2927 * Advance a list_head iterator.  The iterator should be positioned at
2928 * the start of a css_set
2929 */
2930static void cgroup_advance_iter(struct cgroup *cgrp,
2931                                struct cgroup_iter *it)
2932{
2933        struct list_head *l = it->cg_link;
2934        struct cg_cgroup_link *link;
2935        struct css_set *cg;
2936
2937        /* Advance to the next non-empty css_set */
2938        do {
2939                l = l->next;
2940                if (l == &cgrp->css_sets) {
2941                        it->cg_link = NULL;
2942                        return;
2943                }
2944                link = list_entry(l, struct cg_cgroup_link, cgrp_link_list);
2945                cg = link->cg;
2946        } while (list_empty(&cg->tasks));
2947        it->cg_link = l;
2948        it->task = cg->tasks.next;
2949}
2950
2951/*
2952 * To reduce the fork() overhead for systems that are not actually
2953 * using their cgroups capability, we don't maintain the lists running
2954 * through each css_set to its tasks until we see the list actually
2955 * used - in other words after the first call to cgroup_iter_start().
2956 */
2957static void cgroup_enable_task_cg_lists(void)
2958{
2959        struct task_struct *p, *g;
2960        write_lock(&css_set_lock);
2961        use_task_css_set_links = 1;
2962        /*
2963         * We need tasklist_lock because RCU is not safe against
2964         * while_each_thread(). Besides, a forking task that has passed
2965         * cgroup_post_fork() without seeing use_task_css_set_links = 1
2966         * is not guaranteed to have its child immediately visible in the
2967         * tasklist if we walk through it with RCU.
2968         */
2969        read_lock(&tasklist_lock);
2970        do_each_thread(g, p) {
2971                task_lock(p);
2972                /*
2973                 * We should check if the process is exiting, otherwise
2974                 * it will race with cgroup_exit() in that the list
2975                 * entry won't be deleted though the process has exited.
2976                 */
2977                if (!(p->flags & PF_EXITING) && list_empty(&p->cg_list))
2978                        list_add(&p->cg_list, &p->cgroups->tasks);
2979                task_unlock(p);
2980        } while_each_thread(g, p);
2981        read_unlock(&tasklist_lock);
2982        write_unlock(&css_set_lock);
2983}
2984
2985/**
2986 * cgroup_next_descendant_pre - find the next descendant for pre-order walk
2987 * @pos: the current position (%NULL to initiate traversal)
2988 * @cgroup: cgroup whose descendants to walk
2989 *
2990 * To be used by cgroup_for_each_descendant_pre().  Find the next
2991 * descendant to visit for pre-order traversal of @cgroup's descendants.
2992 */
2993struct cgroup *cgroup_next_descendant_pre(struct cgroup *pos,
2994                                          struct cgroup *cgroup)
2995{
2996        struct cgroup *next;
2997
2998        WARN_ON_ONCE(!rcu_read_lock_held());
2999
3000        /* if first iteration, pretend we just visited @cgroup */
3001        if (!pos) {
3002                if (list_empty(&cgroup->children))
3003                        return NULL;
3004                pos = cgroup;
3005        }
3006
3007        /* visit the first child if exists */
3008        next = list_first_or_null_rcu(&pos->children, struct cgroup, sibling);
3009        if (next)
3010                return next;
3011
3012        /* no child, visit my or the closest ancestor's next sibling */
3013        do {
3014                next = list_entry_rcu(pos->sibling.next, struct cgroup,
3015                                      sibling);
3016                if (&next->sibling != &pos->parent->children)
3017                        return next;
3018
3019                pos = pos->parent;
3020        } while (pos != cgroup);
3021
3022        return NULL;
3023}
3024EXPORT_SYMBOL_GPL(cgroup_next_descendant_pre);
3025
3026/**
3027 * cgroup_rightmost_descendant - return the rightmost descendant of a cgroup
3028 * @pos: cgroup of interest
3029 *
3030 * Return the rightmost descendant of @pos.  If there's no descendant,
3031 * @pos is returned.  This can be used during pre-order traversal to skip
3032 * subtree of @pos.
3033 */
3034struct cgroup *cgroup_rightmost_descendant(struct cgroup *pos)
3035{
3036        struct cgroup *last, *tmp;
3037
3038        WARN_ON_ONCE(!rcu_read_lock_held());
3039
3040        do {
3041                last = pos;
3042                /* ->prev isn't RCU safe, walk ->next till the end */
3043                pos = NULL;
3044                list_for_each_entry_rcu(tmp, &last->children, sibling)
3045                        pos = tmp;
3046        } while (pos);
3047
3048        return last;
3049}
3050EXPORT_SYMBOL_GPL(cgroup_rightmost_descendant);
3051
3052static struct cgroup *cgroup_leftmost_descendant(struct cgroup *pos)
3053{
3054        struct cgroup *last;
3055
3056        do {
3057                last = pos;
3058                pos = list_first_or_null_rcu(&pos->children, struct cgroup,
3059                                             sibling);
3060        } while (pos);
3061
3062        return last;
3063}
3064
3065/**
3066 * cgroup_next_descendant_post - find the next descendant for post-order walk
3067 * @pos: the current position (%NULL to initiate traversal)
3068 * @cgroup: cgroup whose descendants to walk
3069 *
3070 * To be used by cgroup_for_each_descendant_post().  Find the next
3071 * descendant to visit for post-order traversal of @cgroup's descendants.
3072 */
3073struct cgroup *cgroup_next_descendant_post(struct cgroup *pos,
3074                                           struct cgroup *cgroup)
3075{
3076        struct cgroup *next;
3077
3078        WARN_ON_ONCE(!rcu_read_lock_held());
3079
3080        /* if first iteration, visit the leftmost descendant */
3081        if (!pos) {
3082                next = cgroup_leftmost_descendant(cgroup);
3083                return next != cgroup ? next : NULL;
3084        }
3085
3086        /* if there's an unvisited sibling, visit its leftmost descendant */
3087        next = list_entry_rcu(pos->sibling.next, struct cgroup, sibling);
3088        if (&next->sibling != &pos->parent->children)
3089                return cgroup_leftmost_descendant(next);
3090
3091        /* no sibling left, visit parent */
3092        next = pos->parent;
3093        return next != cgroup ? next : NULL;
3094}
3095EXPORT_SYMBOL_GPL(cgroup_next_descendant_post);
3096
3097void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it)
3098        __acquires(css_set_lock)
3099{
3100        /*
3101         * The first time anyone tries to iterate across a cgroup,
3102         * we need to enable the list linking each css_set to its
3103         * tasks, and fix up all existing tasks.
3104         */
3105        if (!use_task_css_set_links)
3106                cgroup_enable_task_cg_lists();
3107
3108        read_lock(&css_set_lock);
3109        it->cg_link = &cgrp->css_sets;
3110        cgroup_advance_iter(cgrp, it);
3111}
3112
3113struct task_struct *cgroup_iter_next(struct cgroup *cgrp,
3114                                        struct cgroup_iter *it)
3115{
3116        struct task_struct *res;
3117        struct list_head *l = it->task;
3118        struct cg_cgroup_link *link;
3119
3120        /* If the iterator cg is NULL, we have no tasks */
3121        if (!it->cg_link)
3122                return NULL;
3123        res = list_entry(l, struct task_struct, cg_list);
3124        /* Advance iterator to find next entry */
3125        l = l->next;
3126        link = list_entry(it->cg_link, struct cg_cgroup_link, cgrp_link_list);
3127        if (l == &link->cg->tasks) {
3128                /* We reached the end of this task list - move on to
3129                 * the next cg_cgroup_link */
3130                cgroup_advance_iter(cgrp, it);
3131        } else {
3132                it->task = l;
3133        }
3134        return res;
3135}
3136
3137void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it)
3138        __releases(css_set_lock)
3139{
3140        read_unlock(&css_set_lock);
3141}
3142
3143static inline int started_after_time(struct task_struct *t1,
3144                                     struct timespec *time,
3145                                     struct task_struct *t2)
3146{
3147        int start_diff = timespec_compare(&t1->start_time, time);
3148        if (start_diff > 0) {
3149                return 1;
3150        } else if (start_diff < 0) {
3151                return 0;
3152        } else {
3153                /*
3154                 * Arbitrarily, if two processes started at the same
3155                 * time, we'll say that the lower pointer value
3156                 * started first. Note that t2 may have exited by now
3157                 * so this may not be a valid pointer any longer, but
3158                 * that's fine - it still serves to distinguish
3159                 * between two tasks started (effectively) simultaneously.
3160                 */
3161                return t1 > t2;
3162        }
3163}
3164
3165/*
3166 * This function is a callback from heap_insert() and is used to order
3167 * the heap.
3168 * In this case we order the heap in descending task start time.
3169 */
3170static inline int started_after(void *p1, void *p2)
3171{
3172        struct task_struct *t1 = p1;
3173        struct task_struct *t2 = p2;
3174        return started_after_time(t1, &t2->start_time, t2);
3175}
3176
3177/**
3178 * cgroup_scan_tasks - iterate though all the tasks in a cgroup
3179 * @scan: struct cgroup_scanner containing arguments for the scan
3180 *
3181 * Arguments include pointers to callback functions test_task() and
3182 * process_task().
3183 * Iterate through all the tasks in a cgroup, calling test_task() for each,
3184 * and if it returns true, call process_task() for it also.
3185 * The test_task pointer may be NULL, meaning always true (select all tasks).
3186 * Effectively duplicates cgroup_iter_{start,next,end}()
3187 * but does not lock css_set_lock for the call to process_task().
3188 * The struct cgroup_scanner may be embedded in any structure of the caller's
3189 * creation.
3190 * It is guaranteed that process_task() will act on every task that
3191 * is a member of the cgroup for the duration of this call. This
3192 * function may or may not call process_task() for tasks that exit
3193 * or move to a different cgroup during the call, or are forked or
3194 * move into the cgroup during the call.
3195 *
3196 * Note that test_task() may be called with locks held, and may in some
3197 * situations be called multiple times for the same task, so it should
3198 * be cheap.
3199 * If the heap pointer in the struct cgroup_scanner is non-NULL, a heap has been
3200 * pre-allocated and will be used for heap operations (and its "gt" member will
3201 * be overwritten), else a temporary heap will be used (allocation of which
3202 * may cause this function to fail).
3203 */
3204int cgroup_scan_tasks(struct cgroup_scanner *scan)
3205{
3206        int retval, i;
3207        struct cgroup_iter it;
3208        struct task_struct *p, *dropped;
3209        /* Never dereference latest_task, since it's not refcounted */
3210        struct task_struct *latest_task = NULL;
3211        struct ptr_heap tmp_heap;
3212        struct ptr_heap *heap;
3213        struct timespec latest_time = { 0, 0 };
3214
3215        if (scan->heap) {
3216                /* The caller supplied our heap and pre-allocated its memory */
3217                heap = scan->heap;
3218                heap->gt = &started_after;
3219        } else {
3220                /* We need to allocate our own heap memory */
3221                heap = &tmp_heap;
3222                retval = heap_init(heap, PAGE_SIZE, GFP_KERNEL, &started_after);
3223                if (retval)
3224                        /* cannot allocate the heap */
3225                        return retval;
3226        }
3227
3228 again:
3229        /*
3230         * Scan tasks in the cgroup, using the scanner's "test_task" callback
3231         * to determine which are of interest, and using the scanner's
3232         * "process_task" callback to process any of them that need an update.
3233         * Since we don't want to hold any locks during the task updates,
3234         * gather tasks to be processed in a heap structure.
3235         * The heap is sorted by descending task start time.
3236         * If the statically-sized heap fills up, we overflow tasks that
3237         * started later, and in future iterations only consider tasks that
3238         * started after the latest task in the previous pass. This
3239         * guarantees forward progress and that we don't miss any tasks.
3240         */
3241        heap->size = 0;
3242        cgroup_iter_start(scan->cg, &it);
3243        while ((p = cgroup_iter_next(scan->cg, &it))) {
3244                /*
3245                 * Only affect tasks that qualify per the caller's callback,
3246                 * if he provided one
3247                 */
3248                if (scan->test_task && !scan->test_task(p, scan))
3249                        continue;
3250                /*
3251                 * Only process tasks that started after the last task
3252                 * we processed
3253                 */
3254                if (!started_after_time(p, &latest_time, latest_task))
3255                        continue;
3256                dropped = heap_insert(heap, p);
3257                if (dropped == NULL) {
3258                        /*
3259                         * The new task was inserted; the heap wasn't
3260                         * previously full
3261                         */
3262                        get_task_struct(p);
3263                } else if (dropped != p) {
3264                        /*
3265                         * The new task was inserted, and pushed out a
3266                         * different task
3267                         */
3268                        get_task_struct(p);
3269                        put_task_struct(dropped);
3270                }
3271                /*
3272                 * Else the new task was newer than anything already in
3273                 * the heap and wasn't inserted
3274                 */
3275        }
3276        cgroup_iter_end(scan->cg, &it);
3277
3278        if (heap->size) {
3279                for (i = 0; i < heap->size; i++) {
3280                        struct task_struct *q = heap->ptrs[i];
3281                        if (i == 0) {
3282                                latest_time = q->start_time;
3283                                latest_task = q;
3284                        }
3285                        /* Process the task per the caller's callback */
3286                        scan->process_task(q, scan);
3287                        put_task_struct(q);
3288                }
3289                /*
3290                 * If we had to process any tasks at all, scan again
3291                 * in case some of them were in the middle of forking
3292                 * children that didn't get processed.
3293                 * Not the most efficient way to do it, but it avoids
3294                 * having to take callback_mutex in the fork path
3295                 */
3296                goto again;
3297        }
3298        if (heap == &tmp_heap)
3299                heap_free(&tmp_heap);
3300        return 0;
3301}
3302
3303/*
3304 * Stuff for reading the 'tasks'/'procs' files.
3305 *
3306 * Reading this file can return large amounts of data if a cgroup has
3307 * *lots* of attached tasks. So it may need several calls to read(),
3308 * but we cannot guarantee that the information we produce is correct
3309 * unless we produce it entirely atomically.
3310 *
3311 */
3312
3313/* which pidlist file are we talking about? */
3314enum cgroup_filetype {
3315        CGROUP_FILE_PROCS,
3316        CGROUP_FILE_TASKS,
3317};
3318
3319/*
3320 * A pidlist is a list of pids that virtually represents the contents of one
3321 * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists,
3322 * a pair (one each for procs, tasks) for each pid namespace that's relevant
3323 * to the cgroup.
3324 */
3325struct cgroup_pidlist {
3326        /*
3327         * used to find which pidlist is wanted. doesn't change as long as
3328         * this particular list stays in the list.
3329        */
3330        struct { enum cgroup_filetype type; struct pid_namespace *ns; } key;
3331        /* array of xids */
3332        pid_t *list;
3333        /* how many elements the above list has */
3334        int length;
3335        /* how many files are using the current array */
3336        int use_count;
3337        /* each of these stored in a list by its cgroup */
3338        struct list_head links;
3339        /* pointer to the cgroup we belong to, for list removal purposes */
3340        struct cgroup *owner;
3341        /* protects the other fields */
3342        struct rw_semaphore mutex;
3343};
3344
3345/*
3346 * The following two functions "fix" the issue where there are more pids
3347 * than kmalloc will give memory for; in such cases, we use vmalloc/vfree.
3348 * TODO: replace with a kernel-wide solution to this problem
3349 */
3350#define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2))
3351static void *pidlist_allocate(int count)
3352{
3353        if (PIDLIST_TOO_LARGE(count))
3354                return vmalloc(count * sizeof(pid_t));
3355        else
3356                return kmalloc(count * sizeof(pid_t), GFP_KERNEL);
3357}
3358static void pidlist_free(void *p)
3359{
3360        if (is_vmalloc_addr(p))
3361                vfree(p);
3362        else
3363                kfree(p);
3364}
3365static void *pidlist_resize(void *p, int newcount)
3366{
3367        void *newlist;
3368        /* note: if new alloc fails, old p will still be valid either way */
3369        if (is_vmalloc_addr(p)) {
3370                newlist = vmalloc(newcount * sizeof(pid_t));
3371                if (!newlist)
3372                        return NULL;
3373                memcpy(newlist, p, newcount * sizeof(pid_t));
3374                vfree(p);
3375        } else {
3376                newlist = krealloc(p, newcount * sizeof(pid_t), GFP_KERNEL);
3377        }
3378        return newlist;
3379}
3380
3381/*
3382 * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
3383 * If the new stripped list is sufficiently smaller and there's enough memory
3384 * to allocate a new buffer, will let go of the unneeded memory. Returns the
3385 * number of unique elements.
3386 */
3387/* is the size difference enough that we should re-allocate the array? */
3388#define PIDLIST_REALLOC_DIFFERENCE(old, new) ((old) - PAGE_SIZE >= (new))
3389static int pidlist_uniq(pid_t **p, int length)
3390{
3391        int src, dest = 1;
3392        pid_t *list = *p;
3393        pid_t *newlist;
3394
3395        /*
3396         * we presume the 0th element is unique, so i starts at 1. trivial
3397         * edge cases first; no work needs to be done for either
3398         */
3399        if (length == 0 || length == 1)
3400                return length;
3401        /* src and dest walk down the list; dest counts unique elements */
3402        for (src = 1; src < length; src++) {
3403                /* find next unique element */
3404                while (list[src] == list[src-1]) {
3405                        src++;
3406                        if (src == length)
3407                                goto after;
3408                }
3409                /* dest always points to where the next unique element goes */
3410                list[dest] = list[src];
3411                dest++;
3412        }
3413after:
3414        /*
3415         * if the length difference is large enough, we want to allocate a
3416         * smaller buffer to save memory. if this fails due to out of memory,
3417         * we'll just stay with what we've got.
3418         */
3419        if (PIDLIST_REALLOC_DIFFERENCE(length, dest)) {
3420                newlist = pidlist_resize(list, dest);
3421                if (newlist)
3422                        *p = newlist;
3423        }
3424        return dest;
3425}
3426
3427static int cmppid(const void *a, const void *b)
3428{
3429        return *(pid_t *)a - *(pid_t *)b;
3430}
3431
3432/*
3433 * find the appropriate pidlist for our purpose (given procs vs tasks)
3434 * returns with the lock on that pidlist already held, and takes care
3435 * of the use count, or returns NULL with no locks held if we're out of
3436 * memory.
3437 */
3438static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp,
3439                                                  enum cgroup_filetype type)
3440{
3441        struct cgroup_pidlist *l;
3442        /* don't need task_nsproxy() if we're looking at ourself */
3443        struct pid_namespace *ns = task_active_pid_ns(current);
3444
3445        /*
3446         * We can't drop the pidlist_mutex before taking the l->mutex in case
3447         * the last ref-holder is trying to remove l from the list at the same
3448         * time. Holding the pidlist_mutex precludes somebody taking whichever
3449         * list we find out from under us - compare release_pid_array().
3450         */
3451        mutex_lock(&cgrp->pidlist_mutex);
3452        list_for_each_entry(l, &cgrp->pidlists, links) {
3453                if (l->key.type == type && l->key.ns == ns) {
3454                        /* make sure l doesn't vanish out from under us */
3455                        down_write(&l->mutex);
3456                        mutex_unlock(&cgrp->pidlist_mutex);
3457                        return l;
3458                }
3459        }
3460        /* entry not found; create a new one */
3461        l = kmalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL);
3462        if (!l) {
3463                mutex_unlock(&cgrp->pidlist_mutex);
3464                return l;
3465        }
3466        init_rwsem(&l->mutex);
3467        down_write(&l->mutex);
3468        l->key.type = type;
3469        l->key.ns = get_pid_ns(ns);
3470        l->use_count = 0; /* don't increment here */
3471        l->list = NULL;
3472        l->owner = cgrp;
3473        list_add(&l->links, &cgrp->pidlists);
3474        mutex_unlock(&cgrp->pidlist_mutex);
3475        return l;
3476}
3477
3478/*
3479 * Load a cgroup's pidarray with either procs' tgids or tasks' pids
3480 */
3481static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type,
3482                              struct cgroup_pidlist **lp)
3483{
3484        pid_t *array;
3485        int length;
3486        int pid, n = 0; /* used for populating the array */
3487        struct cgroup_iter it;
3488        struct task_struct *tsk;
3489        struct cgroup_pidlist *l;
3490
3491        /*
3492         * If cgroup gets more users after we read count, we won't have
3493         * enough space - tough.  This race is indistinguishable to the
3494         * caller from the case that the additional cgroup users didn't
3495         * show up until sometime later on.
3496         */
3497        length = cgroup_task_count(cgrp);
3498        array = pidlist_allocate(length);
3499        if (!array)
3500                return -ENOMEM;
3501        /* now, populate the array */
3502        cgroup_iter_start(cgrp, &it);
3503        while ((tsk = cgroup_iter_next(cgrp, &it))) {
3504                if (unlikely(n == length))
3505                        break;
3506                /* get tgid or pid for procs or tasks file respectively */
3507                if (type == CGROUP_FILE_PROCS)
3508                        pid = task_tgid_vnr(tsk);
3509                else
3510                        pid = task_pid_vnr(tsk);
3511                if (pid > 0) /* make sure to only use valid results */
3512                        array[n++] = pid;
3513        }
3514        cgroup_iter_end(cgrp, &it);
3515        length = n;
3516        /* now sort & (if procs) strip out duplicates */
3517        sort(array, length, sizeof(pid_t), cmppid, NULL);
3518        if (type == CGROUP_FILE_PROCS)
3519                length = pidlist_uniq(&array, length);
3520        l = cgroup_pidlist_find(cgrp, type);
3521        if (!l) {
3522                pidlist_free(array);
3523                return -ENOMEM;
3524        }
3525        /* store array, freeing old if necessary - lock already held */
3526        pidlist_free(l->list);
3527        l->list = array;
3528        l->length = length;
3529        l->use_count++;
3530        up_write(&l->mutex);
3531        *lp = l;
3532        return 0;
3533}
3534
3535/**
3536 * cgroupstats_build - build and fill cgroupstats
3537 * @stats: cgroupstats to fill information into
3538 * @dentry: A dentry entry belonging to the cgroup for which stats have
3539 * been requested.
3540 *
3541 * Build and fill cgroupstats so that taskstats can export it to user
3542 * space.
3543 */
3544int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry)
3545{
3546        int ret = -EINVAL;
3547        struct cgroup *cgrp;
3548        struct cgroup_iter it;
3549        struct task_struct *tsk;
3550
3551        /*
3552         * Validate dentry by checking the superblock operations,
3553         * and make sure it's a directory.
3554         */
3555        if (dentry->d_sb->s_op != &cgroup_ops ||
3556            !S_ISDIR(dentry->d_inode->i_mode))
3557                 goto err;
3558
3559        ret = 0;
3560        cgrp = dentry->d_fsdata;
3561
3562        cgroup_iter_start(cgrp, &it);
3563        while ((tsk = cgroup_iter_next(cgrp, &it))) {
3564                switch (tsk->state) {
3565                case TASK_RUNNING:
3566                        stats->nr_running++;
3567                        break;
3568                case TASK_INTERRUPTIBLE:
3569                        stats->nr_sleeping++;
3570                        break;
3571                case TASK_UNINTERRUPTIBLE:
3572                        stats->nr_uninterruptible++;
3573                        break;
3574                case TASK_STOPPED:
3575                        stats->nr_stopped++;
3576                        break;
3577                default:
3578                        if (delayacct_is_task_waiting_on_io(tsk))
3579                                stats->nr_io_wait++;
3580                        break;
3581                }
3582        }
3583        cgroup_iter_end(cgrp, &it);
3584
3585err:
3586        return ret;
3587}
3588
3589
3590/*
3591 * seq_file methods for the tasks/procs files. The seq_file position is the
3592 * next pid to display; the seq_file iterator is a pointer to the pid
3593 * in the cgroup->l->list array.
3594 */
3595
3596static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos)
3597{
3598        /*
3599         * Initially we receive a position value that corresponds to
3600         * one more than the last pid shown (or 0 on the first call or
3601         * after a seek to the start). Use a binary-search to find the
3602         * next pid to display, if any
3603         */
3604        struct cgroup_pidlist *l = s->private;
3605        int index = 0, pid = *pos;
3606        int *iter;
3607
3608        down_read(&l->mutex);
3609        if (pid) {
3610                int end = l->length;
3611
3612                while (index < end) {
3613                        int mid = (index + end) / 2;
3614                        if (l->list[mid] == pid) {
3615                                index = mid;
3616                                break;
3617                        } else if (l->list[mid] <= pid)
3618                                index = mid + 1;
3619                        else
3620                                end = mid;
3621                }
3622        }
3623        /* If we're off the end of the array, we're done */
3624        if (index >= l->length)
3625                return NULL;
3626        /* Update the abstract position to be the actual pid that we found */
3627        iter = l->list + index;
3628        *pos = *iter;
3629        return iter;
3630}
3631
3632static void cgroup_pidlist_stop(struct seq_file *s, void *v)
3633{
3634        struct cgroup_pidlist *l = s->private;
3635        up_read(&l->mutex);
3636}
3637
3638static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos)
3639{
3640        struct cgroup_pidlist *l = s->private;
3641        pid_t *p = v;
3642        pid_t *end = l->list + l->length;
3643        /*
3644         * Advance to the next pid in the array. If this goes off the
3645         * end, we're done
3646         */
3647        p++;
3648        if (p >= end) {
3649                return NULL;
3650        } else {
3651                *pos = *p;
3652                return p;
3653        }
3654}
3655
3656static int cgroup_pidlist_show(struct seq_file *s, void *v)
3657{
3658        return seq_printf(s, "%d\n", *(int *)v);
3659}
3660
3661/*
3662 * seq_operations functions for iterating on pidlists through seq_file -
3663 * independent of whether it's tasks or procs
3664 */
3665static const struct seq_operations cgroup_pidlist_seq_operations = {
3666        .start = cgroup_pidlist_start,
3667        .stop = cgroup_pidlist_stop,
3668        .next = cgroup_pidlist_next,
3669        .show = cgroup_pidlist_show,
3670};
3671
3672static void cgroup_release_pid_array(struct cgroup_pidlist *l)
3673{
3674        /*
3675         * the case where we're the last user of this particular pidlist will
3676         * have us remove it from the cgroup's list, which entails taking the
3677         * mutex. since in pidlist_find the pidlist->lock depends on cgroup->
3678         * pidlist_mutex, we have to take pidlist_mutex first.
3679         */
3680        mutex_lock(&l->owner->pidlist_mutex);
3681        down_write(&l->mutex);
3682        BUG_ON(!l->use_count);
3683        if (!--l->use_count) {
3684                /* we're the last user if refcount is 0; remove and free */
3685                list_del(&l->links);
3686                mutex_unlock(&l->owner->pidlist_mutex);
3687                pidlist_free(l->list);
3688                put_pid_ns(l->key.ns);
3689                up_write(&l->mutex);
3690                kfree(l);
3691                return;
3692        }
3693        mutex_unlock(&l->owner->pidlist_mutex);
3694        up_write(&l->mutex);
3695}
3696
3697static int cgroup_pidlist_release(struct inode *inode, struct file *file)
3698{
3699        struct cgroup_pidlist *l;
3700        if (!(file->f_mode & FMODE_READ))
3701                return 0;
3702        /*
3703         * the seq_file will only be initialized if the file was opened for
3704         * reading; hence we check if it's not null only in that case.
3705         */
3706        l = ((struct seq_file *)file->private_data)->private;
3707        cgroup_release_pid_array(l);
3708        return seq_release(inode, file);
3709}
3710
3711static const struct file_operations cgroup_pidlist_operations = {
3712        .read = seq_read,
3713        .llseek = seq_lseek,
3714        .write = cgroup_file_write,
3715        .release = cgroup_pidlist_release,
3716};
3717
3718/*
3719 * The following functions handle opens on a file that displays a pidlist
3720 * (tasks or procs). Prepare an array of the process/thread IDs of whoever's
3721 * in the cgroup.
3722 */
3723/* helper function for the two below it */
3724static int cgroup_pidlist_open(struct file *file, enum cgroup_filetype type)
3725{
3726        struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
3727        struct cgroup_pidlist *l;
3728        int retval;
3729
3730        /* Nothing to do for write-only files */
3731        if (!(file->f_mode & FMODE_READ))
3732                return 0;
3733
3734        /* have the array populated */
3735        retval = pidlist_array_load(cgrp, type, &l);
3736        if (retval)
3737                return retval;
3738        /* configure file information */
3739        file->f_op = &cgroup_pidlist_operations;
3740
3741        retval = seq_open(file, &cgroup_pidlist_seq_operations);
3742        if (retval) {
3743                cgroup_release_pid_array(l);
3744                return retval;
3745        }
3746        ((struct seq_file *)file->private_data)->private = l;
3747        return 0;
3748}
3749static int cgroup_tasks_open(struct inode *unused, struct file *file)
3750{
3751        return cgroup_pidlist_open(file, CGROUP_FILE_TASKS);
3752}
3753static int cgroup_procs_open(struct inode *unused, struct file *file)
3754{
3755        return cgroup_pidlist_open(file, CGROUP_FILE_PROCS);
3756}
3757
3758static u64 cgroup_read_notify_on_release(struct cgroup *cgrp,
3759                                            struct cftype *cft)
3760{
3761        return notify_on_release(cgrp);
3762}
3763
3764static int cgroup_write_notify_on_release(struct cgroup *cgrp,
3765                                          struct cftype *cft,
3766                                          u64 val)
3767{
3768        clear_bit(CGRP_RELEASABLE, &cgrp->flags);
3769        if (val)
3770                set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
3771        else
3772                clear_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
3773        return 0;
3774}
3775
3776/*
3777 * Unregister event and free resources.
3778 *
3779 * Gets called from workqueue.
3780 */
3781static void cgroup_event_remove(struct work_struct *work)
3782{
3783        struct cgroup_event *event = container_of(work, struct cgroup_event,
3784                        remove);
3785        struct cgroup *cgrp = event->cgrp;
3786
3787        remove_wait_queue(event->wqh, &event->wait);
3788
3789        event->cft->unregister_event(cgrp, event->cft, event->eventfd);
3790
3791        /* Notify userspace the event is going away. */
3792        eventfd_signal(event->eventfd, 1);
3793
3794        eventfd_ctx_put(event->eventfd);
3795        kfree(event);
3796        dput(cgrp->dentry);
3797}
3798
3799/*
3800 * Gets called on POLLHUP on eventfd when user closes it.
3801 *
3802 * Called with wqh->lock held and interrupts disabled.
3803 */
3804static int cgroup_event_wake(wait_queue_t *wait, unsigned mode,
3805                int sync, void *key)
3806{
3807        struct cgroup_event *event = container_of(wait,
3808                        struct cgroup_event, wait);
3809        struct cgroup *cgrp = event->cgrp;
3810        unsigned long flags = (unsigned long)key;
3811
3812        if (flags & POLLHUP) {
3813                /*
3814                 * If the event has been detached at cgroup removal, we
3815                 * can simply return knowing the other side will cleanup
3816                 * for us.
3817                 *
3818                 * We can't race against event freeing since the other
3819                 * side will require wqh->lock via remove_wait_queue(),
3820                 * which we hold.
3821                 */
3822                spin_lock(&cgrp->event_list_lock);
3823                if (!list_empty(&event->list)) {
3824                        list_del_init(&event->list);
3825                        /*
3826                         * We are in atomic context, but cgroup_event_remove()
3827                         * may sleep, so we have to call it in workqueue.
3828                         */
3829                        schedule_work(&event->remove);
3830                }
3831                spin_unlock(&cgrp->event_list_lock);
3832        }
3833
3834        return 0;
3835}
3836
3837static void cgroup_event_ptable_queue_proc(struct file *file,
3838                wait_queue_head_t *wqh, poll_table *pt)
3839{
3840        struct cgroup_event *event = container_of(pt,
3841                        struct cgroup_event, pt);
3842
3843        event->wqh = wqh;
3844        add_wait_queue(wqh, &event->wait);
3845}
3846
3847/*
3848 * Parse input and register new cgroup event handler.
3849 *
3850 * Input must be in format '<event_fd> <control_fd> <args>'.
3851 * Interpretation of args is defined by control file implementation.
3852 */
3853static int cgroup_write_event_control(struct cgroup *cgrp, struct cftype *cft,
3854                                      const char *buffer)
3855{
3856        struct cgroup_event *event = NULL;
3857        struct cgroup *cgrp_cfile;
3858        unsigned int efd, cfd;
3859        struct file *efile = NULL;
3860        struct file *cfile = NULL;
3861        char *endp;
3862        int ret;
3863
3864        efd = simple_strtoul(buffer, &endp, 10);
3865        if (*endp != ' ')
3866                return -EINVAL;
3867        buffer = endp + 1;
3868
3869        cfd = simple_strtoul(buffer, &endp, 10);
3870        if ((*endp != ' ') && (*endp != '\0'))
3871                return -EINVAL;
3872        buffer = endp + 1;
3873
3874        event = kzalloc(sizeof(*event), GFP_KERNEL);
3875        if (!event)
3876                return -ENOMEM;
3877        event->cgrp = cgrp;
3878        INIT_LIST_HEAD(&event->list);
3879        init_poll_funcptr(&event->pt, cgroup_event_ptable_queue_proc);
3880        init_waitqueue_func_entry(&event->wait, cgroup_event_wake);
3881        INIT_WORK(&event->remove, cgroup_event_remove);
3882
3883        efile = eventfd_fget(efd);
3884        if (IS_ERR(efile)) {
3885                ret = PTR_ERR(efile);
3886                goto fail;
3887        }
3888
3889        event->eventfd = eventfd_ctx_fileget(efile);
3890        if (IS_ERR(event->eventfd)) {
3891                ret = PTR_ERR(event->eventfd);
3892                goto fail;
3893        }
3894
3895        cfile = fget(cfd);
3896        if (!cfile) {
3897                ret = -EBADF;
3898                goto fail;
3899        }
3900
3901        /* the process need read permission on control file */
3902        /* AV: shouldn't we check that it's been opened for read instead? */
3903        ret = inode_permission(file_inode(cfile), MAY_READ);
3904        if (ret < 0)
3905                goto fail;
3906
3907        event->cft = __file_cft(cfile);
3908        if (IS_ERR(event->cft)) {
3909                ret = PTR_ERR(event->cft);
3910                goto fail;
3911        }
3912
3913        /*
3914         * The file to be monitored must be in the same cgroup as
3915         * cgroup.event_control is.
3916         */
3917        cgrp_cfile = __d_cgrp(cfile->f_dentry->d_parent);
3918        if (cgrp_cfile != cgrp) {
3919                ret = -EINVAL;
3920                goto fail;
3921        }
3922
3923        if (!event->cft->register_event || !event->cft->unregister_event) {
3924                ret = -EINVAL;
3925                goto fail;
3926        }
3927
3928        ret = event->cft->register_event(cgrp, event->cft,
3929                        event->eventfd, buffer);
3930        if (ret)
3931                goto fail;
3932
3933        if (efile->f_op->poll(efile, &event->pt) & POLLHUP) {
3934                event->cft->unregister_event(cgrp, event->cft, event->eventfd);
3935                ret = 0;
3936                goto fail;
3937        }
3938
3939        /*
3940         * Events should be removed after rmdir of cgroup directory, but before
3941         * destroying subsystem state objects. Let's take reference to cgroup
3942         * directory dentry to do that.
3943         */
3944        dget(cgrp->dentry);
3945
3946        spin_lock(&cgrp->event_list_lock);
3947        list_add(&event->list, &cgrp->event_list);
3948        spin_unlock(&cgrp->event_list_lock);
3949
3950        fput(cfile);
3951        fput(efile);
3952
3953        return 0;
3954
3955fail:
3956        if (cfile)
3957                fput(cfile);
3958
3959        if (event && event->eventfd && !IS_ERR(event->eventfd))
3960                eventfd_ctx_put(event->eventfd);
3961
3962        if (!IS_ERR_OR_NULL(efile))
3963                fput(efile);
3964
3965        kfree(event);
3966
3967        return ret;
3968}
3969
3970static u64 cgroup_clone_children_read(struct cgroup *cgrp,
3971                                    struct cftype *cft)
3972{
3973        return test_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags);
3974}
3975
3976static int cgroup_clone_children_write(struct cgroup *cgrp,
3977                                     struct cftype *cft,
3978                                     u64 val)
3979{
3980        if (val)
3981                set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags);
3982        else
3983                clear_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags);
3984        return 0;
3985}
3986
3987/*
3988 * for the common functions, 'private' gives the type of file
3989 */
3990/* for hysterical raisins, we can't put this on the older files */
3991#define CGROUP_FILE_GENERIC_PREFIX "cgroup."
3992static struct cftype files[] = {
3993        {
3994                .name = "tasks",
3995                .open = cgroup_tasks_open,
3996                .write_u64 = cgroup_tasks_write,
3997                .release = cgroup_pidlist_release,
3998                .mode = S_IRUGO | S_IWUSR,
3999        },
4000        {
4001                .name = CGROUP_FILE_GENERIC_PREFIX "procs",
4002                .open = cgroup_procs_open,
4003                .write_u64 = cgroup_procs_write,
4004                .release = cgroup_pidlist_release,
4005                .mode = S_IRUGO | S_IWUSR,
4006        },
4007        {
4008                .name = "notify_on_release",
4009                .read_u64 = cgroup_read_notify_on_release,
4010                .write_u64 = cgroup_write_notify_on_release,
4011        },
4012        {
4013                .name = CGROUP_FILE_GENERIC_PREFIX "event_control",
4014                .write_string = cgroup_write_event_control,
4015                .mode = S_IWUGO,
4016        },
4017        {
4018                .name = "cgroup.clone_children",
4019                .read_u64 = cgroup_clone_children_read,
4020                .write_u64 = cgroup_clone_children_write,
4021        },
4022        {
4023                .name = "release_agent",
4024                .flags = CFTYPE_ONLY_ON_ROOT,
4025                .read_seq_string = cgroup_release_agent_show,
4026                .write_string = cgroup_release_agent_write,
4027                .max_write_len = PATH_MAX,
4028        },
4029        { }     /* terminate */
4030};
4031
4032/**
4033 * cgroup_populate_dir - selectively creation of files in a directory
4034 * @cgrp: target cgroup
4035 * @base_files: true if the base files should be added
4036 * @subsys_mask: mask of the subsystem ids whose files should be added
4037 */
4038static int cgroup_populate_dir(struct cgroup *cgrp, bool base_files,
4039                               unsigned long subsys_mask)
4040{
4041        int err;
4042        struct cgroup_subsys *ss;
4043
4044        if (base_files) {
4045                err = cgroup_addrm_files(cgrp, NULL, files, true);
4046                if (err < 0)
4047                        return err;
4048        }
4049
4050        /* process cftsets of each subsystem */
4051        for_each_subsys(cgrp->root, ss) {
4052                struct cftype_set *set;
4053                if (!test_bit(ss->subsys_id, &subsys_mask))
4054                        continue;
4055
4056                list_for_each_entry(set, &ss->cftsets, node)
4057                        cgroup_addrm_files(cgrp, ss, set->cfts, true);
4058        }
4059
4060        /* This cgroup is ready now */
4061        for_each_subsys(cgrp->root, ss) {
4062                struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
4063                /*
4064                 * Update id->css pointer and make this css visible from
4065                 * CSS ID functions. This pointer will be dereferened
4066                 * from RCU-read-side without locks.
4067                 */
4068                if (css->id)
4069                        rcu_assign_pointer(css->id->css, css);
4070        }
4071
4072        return 0;
4073}
4074
4075static void css_dput_fn(struct work_struct *work)
4076{
4077        struct cgroup_subsys_state *css =
4078                container_of(work, struct cgroup_subsys_state, dput_work);
4079        struct dentry *dentry = css->cgroup->dentry;
4080        struct super_block *sb = dentry->d_sb;
4081
4082        atomic_inc(&sb->s_active);
4083        dput(dentry);
4084        deactivate_super(sb);
4085}
4086
4087static void init_cgroup_css(struct cgroup_subsys_state *css,
4088                               struct cgroup_subsys *ss,
4089                               struct cgroup *cgrp)
4090{
4091        css->cgroup = cgrp;
4092        atomic_set(&css->refcnt, 1);
4093        css->flags = 0;
4094        css->id = NULL;
4095        if (cgrp == dummytop)
4096                css->flags |= CSS_ROOT;
4097        BUG_ON(cgrp->subsys[ss->subsys_id]);
4098        cgrp->subsys[ss->subsys_id] = css;
4099
4100        /*
4101         * css holds an extra ref to @cgrp->dentry which is put on the last
4102         * css_put().  dput() requires process context, which css_put() may
4103         * be called without.  @css->dput_work will be used to invoke
4104         * dput() asynchronously from css_put().
4105         */
4106        INIT_WORK(&css->dput_work, css_dput_fn);
4107}
4108
4109/* invoke ->post_create() on a new CSS and mark it online if successful */
4110static int online_css(struct cgroup_subsys *ss, struct cgroup *cgrp)
4111{
4112        int ret = 0;
4113
4114        lockdep_assert_held(&cgroup_mutex);
4115
4116        if (ss->css_online)
4117                ret = ss->css_online(cgrp);
4118        if (!ret)
4119                cgrp->subsys[ss->subsys_id]->flags |= CSS_ONLINE;
4120        return ret;
4121}
4122
4123/* if the CSS is online, invoke ->pre_destory() on it and mark it offline */
4124static void offline_css(struct cgroup_subsys *ss, struct cgroup *cgrp)
4125        __releases(&cgroup_mutex) __acquires(&cgroup_mutex)
4126{
4127        struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
4128
4129        lockdep_assert_held(&cgroup_mutex);
4130
4131        if (!(css->flags & CSS_ONLINE))
4132                return;
4133
4134        /*
4135         * css_offline() should be called with cgroup_mutex unlocked.  See
4136         * 3fa59dfbc3 ("cgroup: fix potential deadlock in pre_destroy") for
4137         * details.  This temporary unlocking should go away once
4138         * cgroup_mutex is unexported from controllers.
4139         */
4140        if (ss->css_offline) {
4141                mutex_unlock(&cgroup_mutex);
4142                ss->css_offline(cgrp);
4143                mutex_lock(&cgroup_mutex);
4144        }
4145
4146        cgrp->subsys[ss->subsys_id]->flags &= ~CSS_ONLINE;
4147}
4148
4149/*
4150 * cgroup_create - create a cgroup
4151 * @parent: cgroup that will be parent of the new cgroup
4152 * @dentry: dentry of the new cgroup
4153 * @mode: mode to set on new inode
4154 *
4155 * Must be called with the mutex on the parent inode held
4156 */
4157static long cgroup_create(struct cgroup *parent, struct dentry *dentry,
4158                             umode_t mode)
4159{
4160        struct cgroup *cgrp;
4161        struct cgroupfs_root *root = parent->root;
4162        int err = 0;
4163        struct cgroup_subsys *ss;
4164        struct super_block *sb = root->sb;
4165
4166        /* allocate the cgroup and its ID, 0 is reserved for the root */
4167        cgrp = kzalloc(sizeof(*cgrp), GFP_KERNEL);
4168        if (!cgrp)
4169                return -ENOMEM;
4170
4171        cgrp->id = ida_simple_get(&root->cgroup_ida, 1, 0, GFP_KERNEL);
4172        if (cgrp->id < 0)
4173                goto err_free_cgrp;
4174
4175        /*
4176         * Only live parents can have children.  Note that the liveliness
4177         * check isn't strictly necessary because cgroup_mkdir() and
4178         * cgroup_rmdir() are fully synchronized by i_mutex; however, do it
4179         * anyway so that locking is contained inside cgroup proper and we
4180         * don't get nasty surprises if we ever grow another caller.
4181         */
4182        if (!cgroup_lock_live_group(parent)) {
4183                err = -ENODEV;
4184                goto err_free_id;
4185        }
4186
4187        /* Grab a reference on the superblock so the hierarchy doesn't
4188         * get deleted on unmount if there are child cgroups.  This
4189         * can be done outside cgroup_mutex, since the sb can't
4190         * disappear while someone has an open control file on the
4191         * fs */
4192        atomic_inc(&sb->s_active);
4193
4194        init_cgroup_housekeeping(cgrp);
4195
4196        dentry->d_fsdata = cgrp;
4197        cgrp->dentry = dentry;
4198
4199        cgrp->parent = parent;
4200        cgrp->root = parent->root;
4201        cgrp->top_cgroup = parent->top_cgroup;
4202
4203        if (notify_on_release(parent))
4204                set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
4205
4206        if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &parent->flags))
4207                set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags);
4208
4209        for_each_subsys(root, ss) {
4210                struct cgroup_subsys_state *css;
4211
4212                css = ss->css_alloc(cgrp);
4213                if (IS_ERR(css)) {
4214                        err = PTR_ERR(css);
4215                        goto err_free_all;
4216                }
4217                init_cgroup_css(css, ss, cgrp);
4218                if (ss->use_id) {
4219                        err = alloc_css_id(ss, parent, cgrp);
4220                        if (err)
4221                                goto err_free_all;
4222                }
4223        }
4224
4225        /*
4226         * Create directory.  cgroup_create_file() returns with the new
4227         * directory locked on success so that it can be populated without
4228         * dropping cgroup_mutex.
4229         */
4230        err = cgroup_create_file(dentry, S_IFDIR | mode, sb);
4231        if (err < 0)
4232                goto err_free_all;
4233        lockdep_assert_held(&dentry->d_inode->i_mutex);
4234
4235        /* allocation complete, commit to creation */
4236        list_add_tail(&cgrp->allcg_node, &root->allcg_list);
4237        list_add_tail_rcu(&cgrp->sibling, &cgrp->parent->children);
4238        root->number_of_cgroups++;
4239
4240        /* each css holds a ref to the cgroup's dentry */
4241        for_each_subsys(root, ss)
4242                dget(dentry);
4243
4244        /* creation succeeded, notify subsystems */
4245        for_each_subsys(root, ss) {
4246                err = online_css(ss, cgrp);
4247                if (err)
4248                        goto err_destroy;
4249
4250                if (ss->broken_hierarchy && !ss->warned_broken_hierarchy &&
4251                    parent->parent) {
4252                        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",
4253                                   current->comm, current->pid, ss->name);
4254                        if (!strcmp(ss->name, "memory"))
4255                                pr_warning("cgroup: \"memory\" requires setting use_hierarchy to 1 on the root.\n");
4256                        ss->warned_broken_hierarchy = true;
4257                }
4258        }
4259
4260        err = cgroup_populate_dir(cgrp, true, root->subsys_mask);
4261        if (err)
4262                goto err_destroy;
4263
4264        mutex_unlock(&cgroup_mutex);
4265        mutex_unlock(&cgrp->dentry->d_inode->i_mutex);
4266
4267        return 0;
4268
4269err_free_all:
4270        for_each_subsys(root, ss) {
4271                if (cgrp->subsys[ss->subsys_id])
4272                        ss->css_free(cgrp);
4273        }
4274        mutex_unlock(&cgroup_mutex);
4275        /* Release the reference count that we took on the superblock */
4276        deactivate_super(sb);
4277err_free_id:
4278        ida_simple_remove(&root->cgroup_ida, cgrp->id);
4279err_free_cgrp:
4280        kfree(cgrp);
4281        return err;
4282
4283err_destroy:
4284        cgroup_destroy_locked(cgrp);
4285        mutex_unlock(&cgroup_mutex);
4286        mutex_unlock(&dentry->d_inode->i_mutex);
4287        return err;
4288}
4289
4290static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
4291{
4292        struct cgroup *c_parent = dentry->d_parent->d_fsdata;
4293
4294        /* the vfs holds inode->i_mutex already */
4295        return cgroup_create(c_parent, dentry, mode | S_IFDIR);
4296}
4297
4298/*
4299 * Check the reference count on each subsystem. Since we already
4300 * established that there are no tasks in the cgroup, if the css refcount
4301 * is also 1, then there should be no outstanding references, so the
4302 * subsystem is safe to destroy. We scan across all subsystems rather than
4303 * using the per-hierarchy linked list of mounted subsystems since we can
4304 * be called via check_for_release() with no synchronization other than
4305 * RCU, and the subsystem linked list isn't RCU-safe.
4306 */
4307static int cgroup_has_css_refs(struct cgroup *cgrp)
4308{
4309        int i;
4310
4311        /*
4312         * We won't need to lock the subsys array, because the subsystems
4313         * we're concerned about aren't going anywhere since our cgroup root
4314         * has a reference on them.
4315         */
4316        for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
4317                struct cgroup_subsys *ss = subsys[i];
4318                struct cgroup_subsys_state *css;
4319
4320                /* Skip subsystems not present or not in this hierarchy */
4321                if (ss == NULL || ss->root != cgrp->root)
4322                        continue;
4323
4324                css = cgrp->subsys[ss->subsys_id];
4325                /*
4326                 * When called from check_for_release() it's possible
4327                 * that by this point the cgroup has been removed
4328                 * and the css deleted. But a false-positive doesn't
4329                 * matter, since it can only happen if the cgroup
4330                 * has been deleted and hence no longer needs the
4331                 * release agent to be called anyway.
4332                 */
4333                if (css && css_refcnt(css) > 1)
4334                        return 1;
4335        }
4336        return 0;
4337}
4338
4339static int cgroup_destroy_locked(struct cgroup *cgrp)
4340        __releases(&cgroup_mutex) __acquires(&cgroup_mutex)
4341{
4342        struct dentry *d = cgrp->dentry;
4343        struct cgroup *parent = cgrp->parent;
4344        DEFINE_WAIT(wait);
4345        struct cgroup_event *event, *tmp;
4346        struct cgroup_subsys *ss;
4347        LIST_HEAD(tmp_list);
4348
4349        lockdep_assert_held(&d->d_inode->i_mutex);
4350        lockdep_assert_held(&cgroup_mutex);
4351
4352        if (atomic_read(&cgrp->count) || !list_empty(&cgrp->children))
4353                return -EBUSY;
4354
4355        /*
4356         * Block new css_tryget() by deactivating refcnt and mark @cgrp
4357         * removed.  This makes future css_tryget() and child creation
4358         * attempts fail thus maintaining the removal conditions verified
4359         * above.
4360         */
4361        for_each_subsys(cgrp->root, ss) {
4362                struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
4363
4364                WARN_ON(atomic_read(&css->refcnt) < 0);
4365                atomic_add(CSS_DEACT_BIAS, &css->refcnt);
4366        }
4367        set_bit(CGRP_REMOVED, &cgrp->flags);
4368
4369        /* tell subsystems to initate destruction */
4370        for_each_subsys(cgrp->root, ss)
4371                offline_css(ss, cgrp);
4372
4373        /*
4374         * Put all the base refs.  Each css holds an extra reference to the
4375         * cgroup's dentry and cgroup removal proceeds regardless of css
4376         * refs.  On the last put of each css, whenever that may be, the
4377         * extra dentry ref is put so that dentry destruction happens only
4378         * after all css's are released.
4379         */
4380        for_each_subsys(cgrp->root, ss)
4381                css_put(cgrp->subsys[ss->subsys_id]);
4382
4383        raw_spin_lock(&release_list_lock);
4384        if (!list_empty(&cgrp->release_list))
4385                list_del_init(&cgrp->release_list);
4386        raw_spin_unlock(&release_list_lock);
4387
4388        /* delete this cgroup from parent->children */
4389        list_del_rcu(&cgrp->sibling);
4390        list_del_init(&cgrp->allcg_node);
4391
4392        dget(d);
4393        cgroup_d_remove_dir(d);
4394        dput(d);
4395
4396        set_bit(CGRP_RELEASABLE, &parent->flags);
4397        check_for_release(parent);
4398
4399        /*
4400         * Unregister events and notify userspace.
4401         * Notify userspace about cgroup removing only after rmdir of cgroup
4402         * directory to avoid race between userspace and kernelspace.
4403         */
4404        spin_lock(&cgrp->event_list_lock);
4405        list_for_each_entry_safe(event, tmp, &cgrp->event_list, list) {
4406                list_del_init(&event->list);
4407                schedule_work(&event->remove);
4408        }
4409        spin_unlock(&cgrp->event_list_lock);
4410
4411        return 0;
4412}
4413
4414static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry)
4415{
4416        int ret;
4417
4418        mutex_lock(&cgroup_mutex);
4419        ret = cgroup_destroy_locked(dentry->d_fsdata);
4420        mutex_unlock(&cgroup_mutex);
4421
4422        return ret;
4423}
4424
4425static void __init_or_module cgroup_init_cftsets(struct cgroup_subsys *ss)
4426{
4427        INIT_LIST_HEAD(&ss->cftsets);
4428
4429        /*
4430         * base_cftset is embedded in subsys itself, no need to worry about
4431         * deregistration.
4432         */
4433        if (ss->base_cftypes) {
4434                ss->base_cftset.cfts = ss->base_cftypes;
4435                list_add_tail(&ss->base_cftset.node, &ss->cftsets);
4436        }
4437}
4438
4439static void __init cgroup_init_subsys(struct cgroup_subsys *ss)
4440{
4441        struct cgroup_subsys_state *css;
4442
4443        printk(KERN_INFO "Initializing cgroup subsys %s\n", ss->name);
4444
4445        mutex_lock(&cgroup_mutex);
4446
4447        /* init base cftset */
4448        cgroup_init_cftsets(ss);
4449
4450        /* Create the top cgroup state for this subsystem */
4451        list_add(&ss->sibling, &rootnode.subsys_list);
4452        ss->root = &rootnode;
4453        css = ss->css_alloc(dummytop);
4454        /* We don't handle early failures gracefully */
4455        BUG_ON(IS_ERR(css));
4456        init_cgroup_css(css, ss, dummytop);
4457
4458        /* Update the init_css_set to contain a subsys
4459         * pointer to this state - since the subsystem is
4460         * newly registered, all tasks and hence the
4461         * init_css_set is in the subsystem's top cgroup. */
4462        init_css_set.subsys[ss->subsys_id] = css;
4463
4464        need_forkexit_callback |= ss->fork || ss->exit;
4465
4466        /* At system boot, before all subsystems have been
4467         * registered, no tasks have been forked, so we don't
4468         * need to invoke fork callbacks here. */
4469        BUG_ON(!list_empty(&init_task.tasks));
4470
4471        ss->active = 1;
4472        BUG_ON(online_css(ss, dummytop));
4473
4474        mutex_unlock(&cgroup_mutex);
4475
4476        /* this function shouldn't be used with modular subsystems, since they
4477         * need to register a subsys_id, among other things */
4478        BUG_ON(ss->module);
4479}
4480
4481/**
4482 * cgroup_load_subsys: load and register a modular subsystem at runtime
4483 * @ss: the subsystem to load
4484 *
4485 * This function should be called in a modular subsystem's initcall. If the
4486 * subsystem is built as a module, it will be assigned a new subsys_id and set
4487 * up for use. If the subsystem is built-in anyway, work is delegated to the
4488 * simpler cgroup_init_subsys.
4489 */
4490int __init_or_module cgroup_load_subsys(struct cgroup_subsys *ss)
4491{
4492        struct cgroup_subsys_state *css;
4493        int i, ret;
4494        struct hlist_node *tmp;
4495        struct css_set *cg;
4496        unsigned long key;
4497
4498        /* check name and function validity */
4499        if (ss->name == NULL || strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN ||
4500            ss->css_alloc == NULL || ss->css_free == NULL)
4501                return -EINVAL;
4502
4503        /*
4504         * we don't support callbacks in modular subsystems. this check is
4505         * before the ss->module check for consistency; a subsystem that could
4506         * be a module should still have no callbacks even if the user isn't
4507         * compiling it as one.
4508         */
4509        if (ss->fork || ss->exit)
4510                return -EINVAL;
4511
4512        /*
4513         * an optionally modular subsystem is built-in: we want to do nothing,
4514         * since cgroup_init_subsys will have already taken care of it.
4515         */
4516        if (ss->module == NULL) {
4517                /* a sanity check */
4518                BUG_ON(subsys[ss->subsys_id] != ss);
4519                return 0;
4520        }
4521
4522        /* init base cftset */
4523        cgroup_init_cftsets(ss);
4524
4525        mutex_lock(&cgroup_mutex);
4526        subsys[ss->subsys_id] = ss;
4527
4528        /*
4529         * no ss->css_alloc seems to need anything important in the ss
4530         * struct, so this can happen first (i.e. before the rootnode
4531         * attachment).
4532         */
4533        css = ss->css_alloc(dummytop);
4534        if (IS_ERR(css)) {
4535                /* failure case - need to deassign the subsys[] slot. */
4536                subsys[ss->subsys_id] = NULL;
4537                mutex_unlock(&cgroup_mutex);
4538                return PTR_ERR(css);
4539        }
4540
4541        list_add(&ss->sibling, &rootnode.subsys_list);
4542        ss->root = &rootnode;
4543
4544        /* our new subsystem will be attached to the dummy hierarchy. */
4545        init_cgroup_css(css, ss, dummytop);
4546        /* init_idr must be after init_cgroup_css because it sets css->id. */
4547        if (ss->use_id) {
4548                ret = cgroup_init_idr(ss, css);
4549                if (ret)
4550                        goto err_unload;
4551        }
4552
4553        /*
4554         * Now we need to entangle the css into the existing css_sets. unlike
4555         * in cgroup_init_subsys, there are now multiple css_sets, so each one
4556         * will need a new pointer to it; done by iterating the css_set_table.
4557         * furthermore, modifying the existing css_sets will corrupt the hash
4558         * table state, so each changed css_set will need its hash recomputed.
4559         * this is all done under the css_set_lock.
4560         */
4561        write_lock(&css_set_lock);
4562        hash_for_each_safe(css_set_table, i, tmp, cg, hlist) {
4563                /* skip entries that we already rehashed */
4564                if (cg->subsys[ss->subsys_id])
4565                        continue;
4566                /* remove existing entry */
4567                hash_del(&cg->hlist);
4568                /* set new value */
4569                cg->subsys[ss->subsys_id] = css;
4570                /* recompute hash and restore entry */
4571                key = css_set_hash(cg->subsys);
4572                hash_add(css_set_table, &cg->hlist, key);
4573        }
4574        write_unlock(&css_set_lock);
4575
4576        ss->active = 1;
4577        ret = online_css(ss, dummytop);
4578        if (ret)
4579                goto err_unload;
4580
4581        /* success! */
4582        mutex_unlock(&cgroup_mutex);
4583        return 0;
4584
4585err_unload:
4586        mutex_unlock(&cgroup_mutex);
4587        /* @ss can't be mounted here as try_module_get() would fail */
4588        cgroup_unload_subsys(ss);
4589        return ret;
4590}
4591EXPORT_SYMBOL_GPL(cgroup_load_subsys);
4592
4593/**
4594 * cgroup_unload_subsys: unload a modular subsystem
4595 * @ss: the subsystem to unload
4596 *
4597 * This function should be called in a modular subsystem's exitcall. When this
4598 * function is invoked, the refcount on the subsystem's module will be 0, so
4599 * the subsystem will not be attached to any hierarchy.
4600 */
4601void cgroup_unload_subsys(struct cgroup_subsys *ss)
4602{
4603        struct cg_cgroup_link *link;
4604
4605        BUG_ON(ss->module == NULL);
4606
4607        /*
4608         * we shouldn't be called if the subsystem is in use, and the use of
4609         * try_module_get in parse_cgroupfs_options should ensure that it
4610         * doesn't start being used while we're killing it off.
4611         */
4612        BUG_ON(ss->root != &rootnode);
4613
4614        mutex_lock(&cgroup_mutex);
4615
4616        offline_css(ss, dummytop);
4617        ss->active = 0;
4618
4619        if (ss->use_id)
4620                idr_destroy(&ss->idr);
4621
4622        /* deassign the subsys_id */
4623        subsys[ss->subsys_id] = NULL;
4624
4625        /* remove subsystem from rootnode's list of subsystems */
4626        list_del_init(&ss->sibling);
4627
4628        /*
4629         * disentangle the css from all css_sets attached to the dummytop. as
4630         * in loading, we need to pay our respects to the hashtable gods.
4631         */
4632        write_lock(&css_set_lock);
4633        list_for_each_entry(link, &dummytop->css_sets, cgrp_link_list) {
4634                struct css_set *cg = link->cg;
4635                unsigned long key;
4636
4637                hash_del(&cg->hlist);
4638                cg->subsys[ss->subsys_id] = NULL;
4639                key = css_set_hash(cg->subsys);
4640                hash_add(css_set_table, &cg->hlist, key);
4641        }
4642        write_unlock(&css_set_lock);
4643
4644        /*
4645         * remove subsystem's css from the dummytop and free it - need to
4646         * free before marking as null because ss->css_free needs the
4647         * cgrp->subsys pointer to find their state. note that this also
4648         * takes care of freeing the css_id.
4649         */
4650        ss->css_free(dummytop);
4651        dummytop->subsys[ss->subsys_id] = NULL;
4652
4653        mutex_unlock(&cgroup_mutex);
4654}
4655EXPORT_SYMBOL_GPL(cgroup_unload_subsys);
4656
4657/**
4658 * cgroup_init_early - cgroup initialization at system boot
4659 *
4660 * Initialize cgroups at system boot, and initialize any
4661 * subsystems that request early init.
4662 */
4663int __init cgroup_init_early(void)
4664{
4665        int i;
4666        atomic_set(&init_css_set.refcount, 1);
4667        INIT_LIST_HEAD(&init_css_set.cg_links);
4668        INIT_LIST_HEAD(&init_css_set.tasks);
4669        INIT_HLIST_NODE(&init_css_set.hlist);
4670        css_set_count = 1;
4671        init_cgroup_root(&rootnode);
4672        root_count = 1;
4673        init_task.cgroups = &init_css_set;
4674
4675        init_css_set_link.cg = &init_css_set;
4676        init_css_set_link.cgrp = dummytop;
4677        list_add(&init_css_set_link.cgrp_link_list,
4678                 &rootnode.top_cgroup.css_sets);
4679        list_add(&init_css_set_link.cg_link_list,
4680                 &init_css_set.cg_links);
4681
4682        for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
4683                struct cgroup_subsys *ss = subsys[i];
4684
4685                /* at bootup time, we don't worry about modular subsystems */
4686                if (!ss || ss->module)
4687                        continue;
4688
4689                BUG_ON(!ss->name);
4690                BUG_ON(strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN);
4691                BUG_ON(!ss->css_alloc);
4692                BUG_ON(!ss->css_free);
4693                if (ss->subsys_id != i) {
4694                        printk(KERN_ERR "cgroup: Subsys %s id == %d\n",
4695                               ss->name, ss->subsys_id);
4696                        BUG();
4697                }
4698
4699                if (ss->early_init)
4700                        cgroup_init_subsys(ss);
4701        }
4702        return 0;
4703}
4704
4705/**
4706 * cgroup_init - cgroup initialization
4707 *
4708 * Register cgroup filesystem and /proc file, and initialize
4709 * any subsystems that didn't request early init.
4710 */
4711int __init cgroup_init(void)
4712{
4713        int err;
4714        int i;
4715        unsigned long key;
4716
4717        err = bdi_init(&cgroup_backing_dev_info);
4718        if (err)
4719                return err;
4720
4721        for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
4722                struct cgroup_subsys *ss = subsys[i];
4723
4724                /* at bootup time, we don't worry about modular subsystems */
4725                if (!ss || ss->module)
4726                        continue;
4727                if (!ss->early_init)
4728                        cgroup_init_subsys(ss);
4729                if (ss->use_id)
4730                        cgroup_init_idr(ss, init_css_set.subsys[ss->subsys_id]);
4731        }
4732
4733        /* Add init_css_set to the hash table */
4734        key = css_set_hash(init_css_set.subsys);
4735        hash_add(css_set_table, &init_css_set.hlist, key);
4736        BUG_ON(!init_root_id(&rootnode));
4737
4738        cgroup_kobj = kobject_create_and_add("cgroup", fs_kobj);
4739        if (!cgroup_kobj) {
4740                err = -ENOMEM;
4741                goto out;
4742        }
4743
4744        err = register_filesystem(&cgroup_fs_type);
4745        if (err < 0) {
4746                kobject_put(cgroup_kobj);
4747                goto out;
4748        }
4749
4750        proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations);
4751
4752out:
4753        if (err)
4754                bdi_destroy(&cgroup_backing_dev_info);
4755
4756        return err;
4757}
4758
4759/*
4760 * proc_cgroup_show()
4761 *  - Print task's cgroup paths into seq_file, one line for each hierarchy
4762 *  - Used for /proc/<pid>/cgroup.
4763 *  - No need to task_lock(tsk) on this tsk->cgroup reference, as it
4764 *    doesn't really matter if tsk->cgroup changes after we read it,
4765 *    and we take cgroup_mutex, keeping cgroup_attach_task() from changing it
4766 *    anyway.  No need to check that tsk->cgroup != NULL, thanks to
4767 *    the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks
4768 *    cgroup to top_cgroup.
4769 */
4770
4771/* TODO: Use a proper seq_file iterator */
4772static int proc_cgroup_show(struct seq_file *m, void *v)
4773{
4774        struct pid *pid;
4775        struct task_struct *tsk;
4776        char *buf;
4777        int retval;
4778        struct cgroupfs_root *root;
4779
4780        retval = -ENOMEM;
4781        buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
4782        if (!buf)
4783                goto out;
4784
4785        retval = -ESRCH;
4786        pid = m->private;
4787        tsk = get_pid_task(pid, PIDTYPE_PID);
4788        if (!tsk)
4789                goto out_free;
4790
4791        retval = 0;
4792
4793        mutex_lock(&cgroup_mutex);
4794
4795        for_each_active_root(root) {
4796                struct cgroup_subsys *ss;
4797                struct cgroup *cgrp;
4798                int count = 0;
4799
4800                seq_printf(m, "%d:", root->hierarchy_id);
4801                for_each_subsys(root, ss)
4802                        seq_printf(m, "%s%s", count++ ? "," : "", ss->name);
4803                if (strlen(root->name))
4804                        seq_printf(m, "%sname=%s", count ? "," : "",
4805                                   root->name);
4806                seq_putc(m, ':');
4807                cgrp = task_cgroup_from_root(tsk, root);
4808                retval = cgroup_path(cgrp, buf, PAGE_SIZE);
4809                if (retval < 0)
4810                        goto out_unlock;
4811                seq_puts(m, buf);
4812                seq_putc(m, '\n');
4813        }
4814
4815out_unlock:
4816        mutex_unlock(&cgroup_mutex);
4817        put_task_struct(tsk);
4818out_free:
4819        kfree(buf);
4820out:
4821        return retval;
4822}
4823
4824static int cgroup_open(struct inode *inode, struct file *file)
4825{
4826        struct pid *pid = PROC_I(inode)->pid;
4827        return single_open(file, proc_cgroup_show, pid);
4828}
4829
4830const struct file_operations proc_cgroup_operations = {
4831        .open           = cgroup_open,
4832        .read           = seq_read,
4833        .llseek         = seq_lseek,
4834        .release        = single_release,
4835};
4836
4837/* Display information about each subsystem and each hierarchy */
4838static int proc_cgroupstats_show(struct seq_file *m, void *v)
4839{
4840        int i;
4841
4842        seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
4843        /*
4844         * ideally we don't want subsystems moving around while we do this.
4845         * cgroup_mutex is also necessary to guarantee an atomic snapshot of
4846         * subsys/hierarchy state.
4847         */
4848        mutex_lock(&cgroup_mutex);
4849        for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
4850                struct cgroup_subsys *ss = subsys[i];
4851                if (ss == NULL)
4852                        continue;
4853                seq_printf(m, "%s\t%d\t%d\t%d\n",
4854                           ss->name, ss->root->hierarchy_id,
4855                           ss->root->number_of_cgroups, !ss->disabled);
4856        }
4857        mutex_unlock(&cgroup_mutex);
4858        return 0;
4859}
4860
4861static int cgroupstats_open(struct inode *inode, struct file *file)
4862{
4863        return single_open(file, proc_cgroupstats_show, NULL);
4864}
4865
4866static const struct file_operations proc_cgroupstats_operations = {
4867        .open = cgroupstats_open,
4868        .read = seq_read,
4869        .llseek = seq_lseek,
4870        .release = single_release,
4871};
4872
4873/**
4874 * cgroup_fork - attach newly forked task to its parents cgroup.
4875 * @child: pointer to task_struct of forking parent process.
4876 *
4877 * Description: A task inherits its parent's cgroup at fork().
4878 *
4879 * A pointer to the shared css_set was automatically copied in
4880 * fork.c by dup_task_struct().  However, we ignore that copy, since
4881 * it was not made under the protection of RCU or cgroup_mutex, so
4882 * might no longer be a valid cgroup pointer.  cgroup_attach_task() might
4883 * have already changed current->cgroups, allowing the previously
4884 * referenced cgroup group to be removed and freed.
4885 *
4886 * At the point that cgroup_fork() is called, 'current' is the parent
4887 * task, and the passed argument 'child' points to the child task.
4888 */
4889void cgroup_fork(struct task_struct *child)
4890{
4891        task_lock(current);
4892        child->cgroups = current->cgroups;
4893        get_css_set(child->cgroups);
4894        task_unlock(current);
4895        INIT_LIST_HEAD(&child->cg_list);
4896}
4897
4898/**
4899 * cgroup_post_fork - called on a new task after adding it to the task list
4900 * @child: the task in question
4901 *
4902 * Adds the task to the list running through its css_set if necessary and
4903 * call the subsystem fork() callbacks.  Has to be after the task is
4904 * visible on the task list in case we race with the first call to
4905 * cgroup_iter_start() - to guarantee that the new task ends up on its
4906 * list.
4907 */
4908void cgroup_post_fork(struct task_struct *child)
4909{
4910        int i;
4911
4912        /*
4913         * use_task_css_set_links is set to 1 before we walk the tasklist
4914         * under the tasklist_lock and we read it here after we added the child
4915         * to the tasklist under the tasklist_lock as well. If the child wasn't
4916         * yet in the tasklist when we walked through it from
4917         * cgroup_enable_task_cg_lists(), then use_task_css_set_links value
4918         * should be visible now due to the paired locking and barriers implied
4919         * by LOCK/UNLOCK: it is written before the tasklist_lock unlock
4920         * in cgroup_enable_task_cg_lists() and read here after the tasklist_lock
4921         * lock on fork.
4922         */
4923        if (use_task_css_set_links) {
4924                write_lock(&css_set_lock);
4925                task_lock(child);
4926                if (list_empty(&child->cg_list))
4927                        list_add(&child->cg_list, &child->cgroups->tasks);
4928                task_unlock(child);
4929                write_unlock(&css_set_lock);
4930        }
4931
4932        /*
4933         * Call ss->fork().  This must happen after @child is linked on
4934         * css_set; otherwise, @child might change state between ->fork()
4935         * and addition to css_set.
4936         */
4937        if (need_forkexit_callback) {
4938                for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
4939                        struct cgroup_subsys *ss = subsys[i];
4940
4941                        /*
4942                         * fork/exit callbacks are supported only for
4943                         * builtin subsystems and we don't need further
4944                         * synchronization as they never go away.
4945                         */
4946                        if (!ss || ss->module)
4947                                continue;
4948
4949                        if (ss->fork)
4950                                ss->fork(child);
4951                }
4952        }
4953}
4954
4955/**
4956 * cgroup_exit - detach cgroup from exiting task
4957 * @tsk: pointer to task_struct of exiting process
4958 * @run_callback: run exit callbacks?
4959 *
4960 * Description: Detach cgroup from @tsk and release it.
4961 *
4962 * Note that cgroups marked notify_on_release force every task in
4963 * them to take the global cgroup_mutex mutex when exiting.
4964 * This could impact scaling on very large systems.  Be reluctant to
4965 * use notify_on_release cgroups where very high task exit scaling
4966 * is required on large systems.
4967 *
4968 * the_top_cgroup_hack:
4969 *
4970 *    Set the exiting tasks cgroup to the root cgroup (top_cgroup).
4971 *
4972 *    We call cgroup_exit() while the task is still competent to
4973 *    handle notify_on_release(), then leave the task attached to the
4974 *    root cgroup in each hierarchy for the remainder of its exit.
4975 *
4976 *    To do this properly, we would increment the reference count on
4977 *    top_cgroup, and near the very end of the kernel/exit.c do_exit()
4978 *    code we would add a second cgroup function call, to drop that
4979 *    reference.  This would just create an unnecessary hot spot on
4980 *    the top_cgroup reference count, to no avail.
4981 *
4982 *    Normally, holding a reference to a cgroup without bumping its
4983 *    count is unsafe.   The cgroup could go away, or someone could
4984 *    attach us to a different cgroup, decrementing the count on
4985 *    the first cgroup that we never incremented.  But in this case,
4986 *    top_cgroup isn't going away, and either task has PF_EXITING set,
4987 *    which wards off any cgroup_attach_task() attempts, or task is a failed
4988 *    fork, never visible to cgroup_attach_task.
4989 */
4990void cgroup_exit(struct task_struct *tsk, int run_callbacks)
4991{
4992        struct css_set *cg;
4993        int i;
4994
4995        /*
4996         * Unlink from the css_set task list if necessary.
4997         * Optimistically check cg_list before taking
4998         * css_set_lock
4999         */
5000        if (!list_empty(&tsk->cg_list)) {
5001                write_lock(&css_set_lock);
5002                if (!list_empty(&tsk->cg_list))
5003                        list_del_init(&tsk->cg_list);
5004                write_unlock(&css_set_lock);
5005        }
5006
5007        /* Reassign the task to the init_css_set. */
5008        task_lock(tsk);
5009        cg = tsk->cgroups;
5010        tsk->cgroups = &init_css_set;
5011
5012        if (run_callbacks && need_forkexit_callback) {
5013                for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
5014                        struct cgroup_subsys *ss = subsys[i];
5015
5016                        /* modular subsystems can't use callbacks */
5017                        if (!ss || ss->module)
5018                                continue;
5019
5020                        if (ss->exit) {
5021                                struct cgroup *old_cgrp =
5022                                        rcu_dereference_raw(cg->subsys[i])->cgroup;
5023                                struct cgroup *cgrp = task_cgroup(tsk, i);
5024                                ss->exit(cgrp, old_cgrp, tsk);
5025                        }
5026                }
5027        }
5028        task_unlock(tsk);
5029
5030        put_css_set_taskexit(cg);
5031}
5032
5033/**
5034 * cgroup_is_descendant - see if @cgrp is a descendant of @task's cgrp
5035 * @cgrp: the cgroup in question
5036 * @task: the task in question
5037 *
5038 * See if @cgrp is a descendant of @task's cgroup in the appropriate
5039 * hierarchy.
5040 *
5041 * If we are sending in dummytop, then presumably we are creating
5042 * the top cgroup in the subsystem.
5043 *
5044 * Called only by the ns (nsproxy) cgroup.
5045 */
5046int cgroup_is_descendant(const struct cgroup *cgrp, struct task_struct *task)
5047{
5048        int ret;
5049        struct cgroup *target;
5050
5051        if (cgrp == dummytop)
5052                return 1;
5053
5054        target = task_cgroup_from_root(task, cgrp->root);
5055        while (cgrp != target && cgrp!= cgrp->top_cgroup)
5056                cgrp = cgrp->parent;
5057        ret = (cgrp == target);
5058        return ret;
5059}
5060
5061static void check_for_release(struct cgroup *cgrp)
5062{
5063        /* All of these checks rely on RCU to keep the cgroup
5064         * structure alive */
5065        if (cgroup_is_releasable(cgrp) && !atomic_read(&cgrp->count)
5066            && list_empty(&cgrp->children) && !cgroup_has_css_refs(cgrp)) {
5067                /* Control Group is currently removeable. If it's not
5068                 * already queued for a userspace notification, queue
5069                 * it now */
5070                int need_schedule_work = 0;
5071                raw_spin_lock(&release_list_lock);
5072                if (!cgroup_is_removed(cgrp) &&
5073                    list_empty(&cgrp->release_list)) {
5074                        list_add(&cgrp->release_list, &release_list);
5075                        need_schedule_work = 1;
5076                }
5077                raw_spin_unlock(&release_list_lock);
5078                if (need_schedule_work)
5079                        schedule_work(&release_agent_work);
5080        }
5081}
5082
5083/* Caller must verify that the css is not for root cgroup */
5084bool __css_tryget(struct cgroup_subsys_state *css)
5085{
5086        while (true) {
5087                int t, v;
5088
5089                v = css_refcnt(css);
5090                t = atomic_cmpxchg(&css->refcnt, v, v + 1);
5091                if (likely(t == v))
5092                        return true;
5093                else if (t < 0)
5094                        return false;
5095                cpu_relax();
5096        }
5097}
5098EXPORT_SYMBOL_GPL(__css_tryget);
5099
5100/* Caller must verify that the css is not for root cgroup */
5101void __css_put(struct cgroup_subsys_state *css)
5102{
5103        struct cgroup *cgrp = css->cgroup;
5104        int v;
5105
5106        rcu_read_lock();
5107        v = css_unbias_refcnt(atomic_dec_return(&css->refcnt));
5108
5109        switch (v) {
5110        case 1:
5111                if (notify_on_release(cgrp)) {
5112                        set_bit(CGRP_RELEASABLE, &cgrp->flags);
5113                        check_for_release(cgrp);
5114                }
5115                break;
5116        case 0:
5117                schedule_work(&css->dput_work);
5118                break;
5119        }
5120        rcu_read_unlock();
5121}
5122EXPORT_SYMBOL_GPL(__css_put);
5123
5124/*
5125 * Notify userspace when a cgroup is released, by running the
5126 * configured release agent with the name of the cgroup (path
5127 * relative to the root of cgroup file system) as the argument.
5128 *
5129 * Most likely, this user command will try to rmdir this cgroup.
5130 *
5131 * This races with the possibility that some other task will be
5132 * attached to this cgroup before it is removed, or that some other
5133 * user task will 'mkdir' a child cgroup of this cgroup.  That's ok.
5134 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
5135 * unused, and this cgroup will be reprieved from its death sentence,
5136 * to continue to serve a useful existence.  Next time it's released,
5137 * we will get notified again, if it still has 'notify_on_release' set.
5138 *
5139 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
5140 * means only wait until the task is successfully execve()'d.  The
5141 * separate release agent task is forked by call_usermodehelper(),
5142 * then control in this thread returns here, without waiting for the
5143 * release agent task.  We don't bother to wait because the caller of
5144 * this routine has no use for the exit status of the release agent
5145 * task, so no sense holding our caller up for that.
5146 */
5147static void cgroup_release_agent(struct work_struct *work)
5148{
5149        BUG_ON(work != &release_agent_work);
5150        mutex_lock(&cgroup_mutex);
5151        raw_spin_lock(&release_list_lock);
5152        while (!list_empty(&release_list)) {
5153                char *argv[3], *envp[3];
5154                int i;
5155                char *pathbuf = NULL, *agentbuf = NULL;
5156                struct cgroup *cgrp = list_entry(release_list.next,
5157                                                    struct cgroup,
5158                                                    release_list);
5159                list_del_init(&cgrp->release_list);
5160                raw_spin_unlock(&release_list_lock);
5161                pathbuf = kmalloc(PAGE_SIZE, GFP_KERNEL);
5162                if (!pathbuf)
5163                        goto continue_free;
5164                if (cgroup_path(cgrp, pathbuf, PAGE_SIZE) < 0)
5165                        goto continue_free;
5166                agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL);
5167                if (!agentbuf)
5168                        goto continue_free;
5169
5170                i = 0;
5171                argv[i++] = agentbuf;
5172                argv[i++] = pathbuf;
5173                argv[i] = NULL;
5174
5175                i = 0;
5176                /* minimal command environment */
5177                envp[i++] = "HOME=/";
5178                envp[i++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
5179                envp[i] = NULL;
5180
5181                /* Drop the lock while we invoke the usermode helper,
5182                 * since the exec could involve hitting disk and hence
5183                 * be a slow process */
5184                mutex_unlock(&cgroup_mutex);
5185                call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC);
5186                mutex_lock(&cgroup_mutex);
5187 continue_free:
5188                kfree(pathbuf);
5189                kfree(agentbuf);
5190                raw_spin_lock(&release_list_lock);
5191        }
5192        raw_spin_unlock(&release_list_lock);
5193        mutex_unlock(&cgroup_mutex);
5194}
5195
5196static int __init cgroup_disable(char *str)
5197{
5198        int i;
5199        char *token;
5200
5201        while ((token = strsep(&str, ",")) != NULL) {
5202                if (!*token)
5203                        continue;
5204                for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
5205                        struct cgroup_subsys *ss = subsys[i];
5206
5207                        /*
5208                         * cgroup_disable, being at boot time, can't
5209                         * know about module subsystems, so we don't
5210                         * worry about them.
5211                         */
5212                        if (!ss || ss->module)
5213                                continue;
5214
5215                        if (!strcmp(token, ss->name)) {
5216                                ss->disabled = 1;
5217                                printk(KERN_INFO "Disabling %s control group"
5218                                        " subsystem\n", ss->name);
5219                                break;
5220                        }
5221                }
5222        }
5223        return 1;
5224}
5225__setup("cgroup_disable=", cgroup_disable);
5226
5227/*
5228 * Functons for CSS ID.
5229 */
5230
5231/*
5232 *To get ID other than 0, this should be called when !cgroup_is_removed().
5233 */
5234unsigned short css_id(struct cgroup_subsys_state *css)
5235{
5236        struct css_id *cssid;
5237
5238        /*
5239         * This css_id() can return correct value when somone has refcnt
5240         * on this or this is under rcu_read_lock(). Once css->id is allocated,
5241         * it's unchanged until freed.
5242         */
5243        cssid = rcu_dereference_check(css->id, css_refcnt(css));
5244
5245        if (cssid)
5246                return cssid->id;
5247        return 0;
5248}
5249EXPORT_SYMBOL_GPL(css_id);
5250
5251unsigned short css_depth(struct cgroup_subsys_state *css)
5252{
5253        struct css_id *cssid;
5254
5255        cssid = rcu_dereference_check(css->id, css_refcnt(css));
5256
5257        if (cssid)
5258                return cssid->depth;
5259        return 0;
5260}
5261EXPORT_SYMBOL_GPL(css_depth);
5262
5263/**
5264 *  css_is_ancestor - test "root" css is an ancestor of "child"
5265 * @child: the css to be tested.
5266 * @root: the css supporsed to be an ancestor of the child.
5267 *
5268 * Returns true if "root" is an ancestor of "child" in its hierarchy. Because
5269 * this function reads css->id, the caller must hold rcu_read_lock().
5270 * But, considering usual usage, the csses should be valid objects after test.
5271 * Assuming that the caller will do some action to the child if this returns
5272 * returns true, the caller must take "child";s reference count.
5273 * If "child" is valid object and this returns true, "root" is valid, too.
5274 */
5275
5276bool css_is_ancestor(struct cgroup_subsys_state *child,
5277                    const struct cgroup_subsys_state *root)
5278{
5279        struct css_id *child_id;
5280        struct css_id *root_id;
5281
5282        child_id  = rcu_dereference(child->id);
5283        if (!child_id)
5284                return false;
5285        root_id = rcu_dereference(root->id);
5286        if (!root_id)
5287                return false;
5288        if (child_id->depth < root_id->depth)
5289                return false;
5290        if (child_id->stack[root_id->depth] != root_id->id)
5291                return false;
5292        return true;
5293}
5294
5295void free_css_id(struct cgroup_subsys *ss, struct cgroup_subsys_state *css)
5296{
5297        struct css_id *id = css->id;
5298        /* When this is called before css_id initialization, id can be NULL */
5299        if (!id)
5300                return;
5301
5302        BUG_ON(!ss->use_id);
5303
5304        rcu_assign_pointer(id->css, NULL);
5305        rcu_assign_pointer(css->id, NULL);
5306        spin_lock(&ss->id_lock);
5307        idr_remove(&ss->idr, id->id);
5308        spin_unlock(&ss->id_lock);
5309        kfree_rcu(id, rcu_head);
5310}
5311EXPORT_SYMBOL_GPL(free_css_id);
5312
5313/*
5314 * This is called by init or create(). Then, calls to this function are
5315 * always serialized (By cgroup_mutex() at create()).
5316 */
5317
5318static struct css_id *get_new_cssid(struct cgroup_subsys *ss, int depth)
5319{
5320        struct css_id *newid;
5321        int ret, size;
5322
5323        BUG_ON(!ss->use_id);
5324
5325        size = sizeof(*newid) + sizeof(unsigned short) * (depth + 1);
5326        newid = kzalloc(size, GFP_KERNEL);
5327        if (!newid)
5328                return ERR_PTR(-ENOMEM);
5329
5330        idr_preload(GFP_KERNEL);
5331        spin_lock(&ss->id_lock);
5332        /* Don't use 0. allocates an ID of 1-65535 */
5333        ret = idr_alloc(&ss->idr, newid, 1, CSS_ID_MAX + 1, GFP_NOWAIT);
5334        spin_unlock(&ss->id_lock);
5335        idr_preload_end();
5336
5337        /* Returns error when there are no free spaces for new ID.*/
5338        if (ret < 0)
5339                goto err_out;
5340
5341        newid->id = ret;
5342        newid->depth = depth;
5343        return newid;
5344err_out:
5345        kfree(newid);
5346        return ERR_PTR(ret);
5347
5348}
5349
5350static int __init_or_module cgroup_init_idr(struct cgroup_subsys *ss,
5351                                            struct cgroup_subsys_state *rootcss)
5352{
5353        struct css_id *newid;
5354
5355        spin_lock_init(&ss->id_lock);
5356        idr_init(&ss->idr);
5357
5358        newid = get_new_cssid(ss, 0);
5359        if (IS_ERR(newid))
5360                return PTR_ERR(newid);
5361
5362        newid->stack[0] = newid->id;
5363        newid->css = rootcss;
5364        rootcss->id = newid;
5365        return 0;
5366}
5367
5368static int alloc_css_id(struct cgroup_subsys *ss, struct cgroup *parent,
5369                        struct cgroup *child)
5370{
5371        int subsys_id, i, depth = 0;
5372        struct cgroup_subsys_state *parent_css, *child_css;
5373        struct css_id *child_id, *parent_id;
5374
5375        subsys_id = ss->subsys_id;
5376        parent_css = parent->subsys[subsys_id];
5377        child_css = child->subsys[subsys_id];
5378        parent_id = parent_css->id;
5379        depth = parent_id->depth + 1;
5380
5381        child_id = get_new_cssid(ss, depth);
5382        if (IS_ERR(child_id))
5383                return PTR_ERR(child_id);
5384
5385        for (i = 0; i < depth; i++)
5386                child_id->stack[i] = parent_id->stack[i];
5387        child_id->stack[depth] = child_id->id;
5388        /*
5389         * child_id->css pointer will be set after this cgroup is available
5390         * see cgroup_populate_dir()
5391         */
5392        rcu_assign_pointer(child_css->id, child_id);
5393
5394        return 0;
5395}
5396
5397/**
5398 * css_lookup - lookup css by id
5399 * @ss: cgroup subsys to be looked into.
5400 * @id: the id
5401 *
5402 * Returns pointer to cgroup_subsys_state if there is valid one with id.
5403 * NULL if not. Should be called under rcu_read_lock()
5404 */
5405struct cgroup_subsys_state *css_lookup(struct cgroup_subsys *ss, int id)
5406{
5407        struct css_id *cssid = NULL;
5408
5409        BUG_ON(!ss->use_id);
5410        cssid = idr_find(&ss->idr, id);
5411
5412        if (unlikely(!cssid))
5413                return NULL;
5414
5415        return rcu_dereference(cssid->css);
5416}
5417EXPORT_SYMBOL_GPL(css_lookup);
5418
5419/**
5420 * css_get_next - lookup next cgroup under specified hierarchy.
5421 * @ss: pointer to subsystem
5422 * @id: current position of iteration.
5423 * @root: pointer to css. search tree under this.
5424 * @foundid: position of found object.
5425 *
5426 * Search next css under the specified hierarchy of rootid. Calling under
5427 * rcu_read_lock() is necessary. Returns NULL if it reaches the end.
5428 */
5429struct cgroup_subsys_state *
5430css_get_next(struct cgroup_subsys *ss, int id,
5431             struct cgroup_subsys_state *root, int *foundid)
5432{
5433        struct cgroup_subsys_state *ret = NULL;
5434        struct css_id *tmp;
5435        int tmpid;
5436        int rootid = css_id(root);
5437        int depth = css_depth(root);
5438
5439        if (!rootid)
5440                return NULL;
5441
5442        BUG_ON(!ss->use_id);
5443        WARN_ON_ONCE(!rcu_read_lock_held());
5444
5445        /* fill start point for scan */
5446        tmpid = id;
5447        while (1) {
5448                /*
5449                 * scan next entry from bitmap(tree), tmpid is updated after
5450                 * idr_get_next().
5451                 */
5452                tmp = idr_get_next(&ss->idr, &tmpid);
5453                if (!tmp)
5454                        break;
5455                if (tmp->depth >= depth && tmp->stack[depth] == rootid) {
5456                        ret = rcu_dereference(tmp->css);
5457                        if (ret) {
5458                                *foundid = tmpid;
5459                                break;
5460                        }
5461                }
5462                /* continue to scan from next id */
5463                tmpid = tmpid + 1;
5464        }
5465        return ret;
5466}
5467
5468/*
5469 * get corresponding css from file open on cgroupfs directory
5470 */
5471struct cgroup_subsys_state *cgroup_css_from_dir(struct file *f, int id)
5472{
5473        struct cgroup *cgrp;
5474        struct inode *inode;
5475        struct cgroup_subsys_state *css;
5476
5477        inode = file_inode(f);
5478        /* check in cgroup filesystem dir */
5479        if (inode->i_op != &cgroup_dir_inode_operations)
5480                return ERR_PTR(-EBADF);
5481
5482        if (id < 0 || id >= CGROUP_SUBSYS_COUNT)
5483                return ERR_PTR(-EINVAL);
5484
5485        /* get cgroup */
5486        cgrp = __d_cgrp(f->f_dentry);
5487        css = cgrp->subsys[id];
5488        return css ? css : ERR_PTR(-ENOENT);
5489}
5490
5491#ifdef CONFIG_CGROUP_DEBUG
5492static struct cgroup_subsys_state *debug_css_alloc(struct cgroup *cont)
5493{
5494        struct cgroup_subsys_state *css = kzalloc(sizeof(*css), GFP_KERNEL);
5495
5496        if (!css)
5497                return ERR_PTR(-ENOMEM);
5498
5499        return css;
5500}
5501
5502static void debug_css_free(struct cgroup *cont)
5503{
5504        kfree(cont->subsys[debug_subsys_id]);
5505}
5506
5507static u64 cgroup_refcount_read(struct cgroup *cont, struct cftype *cft)
5508{
5509        return atomic_read(&cont->count);
5510}
5511
5512static u64 debug_taskcount_read(struct cgroup *cont, struct cftype *cft)
5513{
5514        return cgroup_task_count(cont);
5515}
5516
5517static u64 current_css_set_read(struct cgroup *cont, struct cftype *cft)
5518{
5519        return (u64)(unsigned long)current->cgroups;
5520}
5521
5522static u64 current_css_set_refcount_read(struct cgroup *cont,
5523                                           struct cftype *cft)
5524{
5525        u64 count;
5526
5527        rcu_read_lock();
5528        count = atomic_read(&current->cgroups->refcount);
5529        rcu_read_unlock();
5530        return count;
5531}
5532
5533static int current_css_set_cg_links_read(struct cgroup *cont,
5534                                         struct cftype *cft,
5535                                         struct seq_file *seq)
5536{
5537        struct cg_cgroup_link *link;
5538        struct css_set *cg;
5539
5540        read_lock(&css_set_lock);
5541        rcu_read_lock();
5542        cg = rcu_dereference(current->cgroups);
5543        list_for_each_entry(link, &cg->cg_links, cg_link_list) {
5544                struct cgroup *c = link->cgrp;
5545                const char *name;
5546
5547                if (c->dentry)
5548                        name = c->dentry->d_name.name;
5549                else
5550                        name = "?";
5551                seq_printf(seq, "Root %d group %s\n",
5552                           c->root->hierarchy_id, name);
5553        }
5554        rcu_read_unlock();
5555        read_unlock(&css_set_lock);
5556        return 0;
5557}
5558
5559#define MAX_TASKS_SHOWN_PER_CSS 25
5560static int cgroup_css_links_read(struct cgroup *cont,
5561                                 struct cftype *cft,
5562                                 struct seq_file *seq)
5563{
5564        struct cg_cgroup_link *link;
5565
5566        read_lock(&css_set_lock);
5567        list_for_each_entry(link, &cont->css_sets, cgrp_link_list) {
5568                struct css_set *cg = link->cg;
5569                struct task_struct *task;
5570                int count = 0;
5571                seq_printf(seq, "css_set %p\n", cg);
5572                list_for_each_entry(task, &cg->tasks, cg_list) {
5573                        if (count++ > MAX_TASKS_SHOWN_PER_CSS) {
5574                                seq_puts(seq, "  ...\n");
5575                                break;
5576                        } else {
5577                                seq_printf(seq, "  task %d\n",
5578                                           task_pid_vnr(task));
5579                        }
5580                }
5581        }
5582        read_unlock(&css_set_lock);
5583        return 0;
5584}
5585
5586static u64 releasable_read(struct cgroup *cgrp, struct cftype *cft)
5587{
5588        return test_bit(CGRP_RELEASABLE, &cgrp->flags);
5589}
5590
5591static struct cftype debug_files[] =  {
5592        {
5593                .name = "cgroup_refcount",
5594                .read_u64 = cgroup_refcount_read,
5595        },
5596        {
5597                .name = "taskcount",
5598                .read_u64 = debug_taskcount_read,
5599        },
5600
5601        {
5602                .name = "current_css_set",
5603                .read_u64 = current_css_set_read,
5604        },
5605
5606        {
5607                .name = "current_css_set_refcount",
5608                .read_u64 = current_css_set_refcount_read,
5609        },
5610
5611        {
5612                .name = "current_css_set_cg_links",
5613                .read_seq_string = current_css_set_cg_links_read,
5614        },
5615
5616        {
5617                .name = "cgroup_css_links",
5618                .read_seq_string = cgroup_css_links_read,
5619        },
5620
5621        {
5622                .name = "releasable",
5623                .read_u64 = releasable_read,
5624        },
5625
5626        { }     /* terminate */
5627};
5628
5629struct cgroup_subsys debug_subsys = {
5630        .name = "debug",
5631        .css_alloc = debug_css_alloc,
5632        .css_free = debug_css_free,
5633        .subsys_id = debug_subsys_id,
5634        .base_cftypes = debug_files,
5635};
5636#endif /* CONFIG_CGROUP_DEBUG */
5637
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.