Root/kernel/cgroup.c

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/ctype.h>
31#include <linux/errno.h>
32#include <linux/fs.h>
33#include <linux/kernel.h>
34#include <linux/list.h>
35#include <linux/mm.h>
36#include <linux/mutex.h>
37#include <linux/mount.h>
38#include <linux/pagemap.h>
39#include <linux/proc_fs.h>
40#include <linux/rcupdate.h>
41#include <linux/sched.h>
42#include <linux/backing-dev.h>
43#include <linux/seq_file.h>
44#include <linux/slab.h>
45#include <linux/magic.h>
46#include <linux/spinlock.h>
47#include <linux/string.h>
48#include <linux/sort.h>
49#include <linux/kmod.h>
50#include <linux/module.h>
51#include <linux/delayacct.h>
52#include <linux/cgroupstats.h>
53#include <linux/hash.h>
54#include <linux/namei.h>
55#include <linux/pid_namespace.h>
56#include <linux/idr.h>
57#include <linux/vmalloc.h> /* TODO: replace with more sophisticated array */
58#include <linux/eventfd.h>
59#include <linux/poll.h>
60
61#include <asm/atomic.h>
62
63static DEFINE_MUTEX(cgroup_mutex);
64
65/*
66 * Generate an array of cgroup subsystem pointers. At boot time, this is
67 * populated up to CGROUP_BUILTIN_SUBSYS_COUNT, and modular subsystems are
68 * registered after that. The mutable section of this array is protected by
69 * cgroup_mutex.
70 */
71#define SUBSYS(_x) &_x ## _subsys,
72static struct cgroup_subsys *subsys[CGROUP_SUBSYS_COUNT] = {
73#include <linux/cgroup_subsys.h>
74};
75
76#define MAX_CGROUP_ROOT_NAMELEN 64
77
78/*
79 * A cgroupfs_root represents the root of a cgroup hierarchy,
80 * and may be associated with a superblock to form an active
81 * hierarchy
82 */
83struct cgroupfs_root {
84    struct super_block *sb;
85
86    /*
87     * The bitmask of subsystems intended to be attached to this
88     * hierarchy
89     */
90    unsigned long subsys_bits;
91
92    /* Unique id for this hierarchy. */
93    int hierarchy_id;
94
95    /* The bitmask of subsystems currently attached to this hierarchy */
96    unsigned long actual_subsys_bits;
97
98    /* A list running through the attached subsystems */
99    struct list_head subsys_list;
100
101    /* The root cgroup for this hierarchy */
102    struct cgroup top_cgroup;
103
104    /* Tracks how many cgroups are currently defined in hierarchy.*/
105    int number_of_cgroups;
106
107    /* A list running through the active hierarchies */
108    struct list_head root_list;
109
110    /* Hierarchy-specific flags */
111    unsigned long flags;
112
113    /* The path to use for release notifications. */
114    char release_agent_path[PATH_MAX];
115
116    /* The name for this hierarchy - may be empty */
117    char name[MAX_CGROUP_ROOT_NAMELEN];
118};
119
120/*
121 * The "rootnode" hierarchy is the "dummy hierarchy", reserved for the
122 * subsystems that are otherwise unattached - it never has more than a
123 * single cgroup, and all tasks are part of that cgroup.
124 */
125static struct cgroupfs_root rootnode;
126
127/*
128 * CSS ID -- ID per subsys's Cgroup Subsys State(CSS). used only when
129 * cgroup_subsys->use_id != 0.
130 */
131#define CSS_ID_MAX (65535)
132struct css_id {
133    /*
134     * The css to which this ID points. This pointer is set to valid value
135     * after cgroup is populated. If cgroup is removed, this will be NULL.
136     * This pointer is expected to be RCU-safe because destroy()
137     * is called after synchronize_rcu(). But for safe use, css_is_removed()
138     * css_tryget() should be used for avoiding race.
139     */
140    struct cgroup_subsys_state __rcu *css;
141    /*
142     * ID of this css.
143     */
144    unsigned short id;
145    /*
146     * Depth in hierarchy which this ID belongs to.
147     */
148    unsigned short depth;
149    /*
150     * ID is freed by RCU. (and lookup routine is RCU safe.)
151     */
152    struct rcu_head rcu_head;
153    /*
154     * Hierarchy of CSS ID belongs to.
155     */
156    unsigned short stack[0]; /* Array of Length (depth+1) */
157};
158
159/*
160 * cgroup_event represents events which userspace want to recieve.
161 */
162struct cgroup_event {
163    /*
164     * Cgroup which the event belongs to.
165     */
166    struct cgroup *cgrp;
167    /*
168     * Control file which the event associated.
169     */
170    struct cftype *cft;
171    /*
172     * eventfd to signal userspace about the event.
173     */
174    struct eventfd_ctx *eventfd;
175    /*
176     * Each of these stored in a list by the cgroup.
177     */
178    struct list_head list;
179    /*
180     * All fields below needed to unregister event when
181     * userspace closes eventfd.
182     */
183    poll_table pt;
184    wait_queue_head_t *wqh;
185    wait_queue_t wait;
186    struct work_struct remove;
187};
188
189/* The list of hierarchy roots */
190
191static LIST_HEAD(roots);
192static int root_count;
193
194static DEFINE_IDA(hierarchy_ida);
195static int next_hierarchy_id;
196static DEFINE_SPINLOCK(hierarchy_id_lock);
197
198/* dummytop is a shorthand for the dummy hierarchy's top cgroup */
199#define dummytop (&rootnode.top_cgroup)
200
201/* This flag indicates whether tasks in the fork and exit paths should
202 * check for fork/exit handlers to call. This avoids us having to do
203 * extra work in the fork/exit path if none of the subsystems need to
204 * be called.
205 */
206static int need_forkexit_callback __read_mostly;
207
208#ifdef CONFIG_PROVE_LOCKING
209int cgroup_lock_is_held(void)
210{
211    return lockdep_is_held(&cgroup_mutex);
212}
213#else /* #ifdef CONFIG_PROVE_LOCKING */
214int cgroup_lock_is_held(void)
215{
216    return mutex_is_locked(&cgroup_mutex);
217}
218#endif /* #else #ifdef CONFIG_PROVE_LOCKING */
219
220EXPORT_SYMBOL_GPL(cgroup_lock_is_held);
221
222/* convenient tests for these bits */
223inline int cgroup_is_removed(const struct cgroup *cgrp)
224{
225    return test_bit(CGRP_REMOVED, &cgrp->flags);
226}
227
228/* bits in struct cgroupfs_root flags field */
229enum {
230    ROOT_NOPREFIX, /* mounted subsystems have no named prefix */
231};
232
233static int cgroup_is_releasable(const struct cgroup *cgrp)
234{
235    const int bits =
236        (1 << CGRP_RELEASABLE) |
237        (1 << CGRP_NOTIFY_ON_RELEASE);
238    return (cgrp->flags & bits) == bits;
239}
240
241static int notify_on_release(const struct cgroup *cgrp)
242{
243    return test_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
244}
245
246static int clone_children(const struct cgroup *cgrp)
247{
248    return test_bit(CGRP_CLONE_CHILDREN, &cgrp->flags);
249}
250
251/*
252 * for_each_subsys() allows you to iterate on each subsystem attached to
253 * an active hierarchy
254 */
255#define for_each_subsys(_root, _ss) \
256list_for_each_entry(_ss, &_root->subsys_list, sibling)
257
258/* for_each_active_root() allows you to iterate across the active hierarchies */
259#define for_each_active_root(_root) \
260list_for_each_entry(_root, &roots, root_list)
261
262/* the list of cgroups eligible for automatic release. Protected by
263 * release_list_lock */
264static LIST_HEAD(release_list);
265static DEFINE_SPINLOCK(release_list_lock);
266static void cgroup_release_agent(struct work_struct *work);
267static DECLARE_WORK(release_agent_work, cgroup_release_agent);
268static void check_for_release(struct cgroup *cgrp);
269
270/* Link structure for associating css_set objects with cgroups */
271struct cg_cgroup_link {
272    /*
273     * List running through cg_cgroup_links associated with a
274     * cgroup, anchored on cgroup->css_sets
275     */
276    struct list_head cgrp_link_list;
277    struct cgroup *cgrp;
278    /*
279     * List running through cg_cgroup_links pointing at a
280     * single css_set object, anchored on css_set->cg_links
281     */
282    struct list_head cg_link_list;
283    struct css_set *cg;
284};
285
286/* The default css_set - used by init and its children prior to any
287 * hierarchies being mounted. It contains a pointer to the root state
288 * for each subsystem. Also used to anchor the list of css_sets. Not
289 * reference-counted, to improve performance when child cgroups
290 * haven't been created.
291 */
292
293static struct css_set init_css_set;
294static struct cg_cgroup_link init_css_set_link;
295
296static int cgroup_init_idr(struct cgroup_subsys *ss,
297               struct cgroup_subsys_state *css);
298
299/* css_set_lock protects the list of css_set objects, and the
300 * chain of tasks off each css_set. Nests outside task->alloc_lock
301 * due to cgroup_iter_start() */
302static DEFINE_RWLOCK(css_set_lock);
303static int css_set_count;
304
305/*
306 * hash table for cgroup groups. This improves the performance to find
307 * an existing css_set. This hash doesn't (currently) take into
308 * account cgroups in empty hierarchies.
309 */
310#define CSS_SET_HASH_BITS 7
311#define CSS_SET_TABLE_SIZE (1 << CSS_SET_HASH_BITS)
312static struct hlist_head css_set_table[CSS_SET_TABLE_SIZE];
313
314static struct hlist_head *css_set_hash(struct cgroup_subsys_state *css[])
315{
316    int i;
317    int index;
318    unsigned long tmp = 0UL;
319
320    for (i = 0; i < CGROUP_SUBSYS_COUNT; i++)
321        tmp += (unsigned long)css[i];
322    tmp = (tmp >> 16) ^ tmp;
323
324    index = hash_long(tmp, CSS_SET_HASH_BITS);
325
326    return &css_set_table[index];
327}
328
329static void free_css_set_rcu(struct rcu_head *obj)
330{
331    struct css_set *cg = container_of(obj, struct css_set, rcu_head);
332    kfree(cg);
333}
334
335/* We don't maintain the lists running through each css_set to its
336 * task until after the first call to cgroup_iter_start(). This
337 * reduces the fork()/exit() overhead for people who have cgroups
338 * compiled into their kernel but not actually in use */
339static int use_task_css_set_links __read_mostly;
340
341static void __put_css_set(struct css_set *cg, int taskexit)
342{
343    struct cg_cgroup_link *link;
344    struct cg_cgroup_link *saved_link;
345    /*
346     * Ensure that the refcount doesn't hit zero while any readers
347     * can see it. Similar to atomic_dec_and_lock(), but for an
348     * rwlock
349     */
350    if (atomic_add_unless(&cg->refcount, -1, 1))
351        return;
352    write_lock(&css_set_lock);
353    if (!atomic_dec_and_test(&cg->refcount)) {
354        write_unlock(&css_set_lock);
355        return;
356    }
357
358    /* This css_set is dead. unlink it and release cgroup refcounts */
359    hlist_del(&cg->hlist);
360    css_set_count--;
361
362    list_for_each_entry_safe(link, saved_link, &cg->cg_links,
363                 cg_link_list) {
364        struct cgroup *cgrp = link->cgrp;
365        list_del(&link->cg_link_list);
366        list_del(&link->cgrp_link_list);
367        if (atomic_dec_and_test(&cgrp->count) &&
368            notify_on_release(cgrp)) {
369            if (taskexit)
370                set_bit(CGRP_RELEASABLE, &cgrp->flags);
371            check_for_release(cgrp);
372        }
373
374        kfree(link);
375    }
376
377    write_unlock(&css_set_lock);
378    call_rcu(&cg->rcu_head, free_css_set_rcu);
379}
380
381/*
382 * refcounted get/put for css_set objects
383 */
384static inline void get_css_set(struct css_set *cg)
385{
386    atomic_inc(&cg->refcount);
387}
388
389static inline void put_css_set(struct css_set *cg)
390{
391    __put_css_set(cg, 0);
392}
393
394static inline void put_css_set_taskexit(struct css_set *cg)
395{
396    __put_css_set(cg, 1);
397}
398
399/*
400 * compare_css_sets - helper function for find_existing_css_set().
401 * @cg: candidate css_set being tested
402 * @old_cg: existing css_set for a task
403 * @new_cgrp: cgroup that's being entered by the task
404 * @template: desired set of css pointers in css_set (pre-calculated)
405 *
406 * Returns true if "cg" matches "old_cg" except for the hierarchy
407 * which "new_cgrp" belongs to, for which it should match "new_cgrp".
408 */
409static bool compare_css_sets(struct css_set *cg,
410                 struct css_set *old_cg,
411                 struct cgroup *new_cgrp,
412                 struct cgroup_subsys_state *template[])
413{
414    struct list_head *l1, *l2;
415
416    if (memcmp(template, cg->subsys, sizeof(cg->subsys))) {
417        /* Not all subsystems matched */
418        return false;
419    }
420
421    /*
422     * Compare cgroup pointers in order to distinguish between
423     * different cgroups in heirarchies with no subsystems. We
424     * could get by with just this check alone (and skip the
425     * memcmp above) but on most setups the memcmp check will
426     * avoid the need for this more expensive check on almost all
427     * candidates.
428     */
429
430    l1 = &cg->cg_links;
431    l2 = &old_cg->cg_links;
432    while (1) {
433        struct cg_cgroup_link *cgl1, *cgl2;
434        struct cgroup *cg1, *cg2;
435
436        l1 = l1->next;
437        l2 = l2->next;
438        /* See if we reached the end - both lists are equal length. */
439        if (l1 == &cg->cg_links) {
440            BUG_ON(l2 != &old_cg->cg_links);
441            break;
442        } else {
443            BUG_ON(l2 == &old_cg->cg_links);
444        }
445        /* Locate the cgroups associated with these links. */
446        cgl1 = list_entry(l1, struct cg_cgroup_link, cg_link_list);
447        cgl2 = list_entry(l2, struct cg_cgroup_link, cg_link_list);
448        cg1 = cgl1->cgrp;
449        cg2 = cgl2->cgrp;
450        /* Hierarchies should be linked in the same order. */
451        BUG_ON(cg1->root != cg2->root);
452
453        /*
454         * If this hierarchy is the hierarchy of the cgroup
455         * that's changing, then we need to check that this
456         * css_set points to the new cgroup; if it's any other
457         * hierarchy, then this css_set should point to the
458         * same cgroup as the old css_set.
459         */
460        if (cg1->root == new_cgrp->root) {
461            if (cg1 != new_cgrp)
462                return false;
463        } else {
464            if (cg1 != cg2)
465                return false;
466        }
467    }
468    return true;
469}
470
471/*
472 * find_existing_css_set() is a helper for
473 * find_css_set(), and checks to see whether an existing
474 * css_set is suitable.
475 *
476 * oldcg: the cgroup group that we're using before the cgroup
477 * transition
478 *
479 * cgrp: the cgroup that we're moving into
480 *
481 * template: location in which to build the desired set of subsystem
482 * state objects for the new cgroup group
483 */
484static struct css_set *find_existing_css_set(
485    struct css_set *oldcg,
486    struct cgroup *cgrp,
487    struct cgroup_subsys_state *template[])
488{
489    int i;
490    struct cgroupfs_root *root = cgrp->root;
491    struct hlist_head *hhead;
492    struct hlist_node *node;
493    struct css_set *cg;
494
495    /*
496     * Build the set of subsystem state objects that we want to see in the
497     * new css_set. while subsystems can change globally, the entries here
498     * won't change, so no need for locking.
499     */
500    for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
501        if (root->subsys_bits & (1UL << i)) {
502            /* Subsystem is in this hierarchy. So we want
503             * the subsystem state from the new
504             * cgroup */
505            template[i] = cgrp->subsys[i];
506        } else {
507            /* Subsystem is not in this hierarchy, so we
508             * don't want to change the subsystem state */
509            template[i] = oldcg->subsys[i];
510        }
511    }
512
513    hhead = css_set_hash(template);
514    hlist_for_each_entry(cg, node, hhead, hlist) {
515        if (!compare_css_sets(cg, oldcg, cgrp, template))
516            continue;
517
518        /* This css_set matches what we need */
519        return cg;
520    }
521
522    /* No existing cgroup group matched */
523    return NULL;
524}
525
526static void free_cg_links(struct list_head *tmp)
527{
528    struct cg_cgroup_link *link;
529    struct cg_cgroup_link *saved_link;
530
531    list_for_each_entry_safe(link, saved_link, tmp, cgrp_link_list) {
532        list_del(&link->cgrp_link_list);
533        kfree(link);
534    }
535}
536
537/*
538 * allocate_cg_links() allocates "count" cg_cgroup_link structures
539 * and chains them on tmp through their cgrp_link_list fields. Returns 0 on
540 * success or a negative error
541 */
542static int allocate_cg_links(int count, struct list_head *tmp)
543{
544    struct cg_cgroup_link *link;
545    int i;
546    INIT_LIST_HEAD(tmp);
547    for (i = 0; i < count; i++) {
548        link = kmalloc(sizeof(*link), GFP_KERNEL);
549        if (!link) {
550            free_cg_links(tmp);
551            return -ENOMEM;
552        }
553        list_add(&link->cgrp_link_list, tmp);
554    }
555    return 0;
556}
557
558/**
559 * link_css_set - a helper function to link a css_set to a cgroup
560 * @tmp_cg_links: cg_cgroup_link objects allocated by allocate_cg_links()
561 * @cg: the css_set to be linked
562 * @cgrp: the destination cgroup
563 */
564static void link_css_set(struct list_head *tmp_cg_links,
565             struct css_set *cg, struct cgroup *cgrp)
566{
567    struct cg_cgroup_link *link;
568
569    BUG_ON(list_empty(tmp_cg_links));
570    link = list_first_entry(tmp_cg_links, struct cg_cgroup_link,
571                cgrp_link_list);
572    link->cg = cg;
573    link->cgrp = cgrp;
574    atomic_inc(&cgrp->count);
575    list_move(&link->cgrp_link_list, &cgrp->css_sets);
576    /*
577     * Always add links to the tail of the list so that the list
578     * is sorted by order of hierarchy creation
579     */
580    list_add_tail(&link->cg_link_list, &cg->cg_links);
581}
582
583/*
584 * find_css_set() takes an existing cgroup group and a
585 * cgroup object, and returns a css_set object that's
586 * equivalent to the old group, but with the given cgroup
587 * substituted into the appropriate hierarchy. Must be called with
588 * cgroup_mutex held
589 */
590static struct css_set *find_css_set(
591    struct css_set *oldcg, struct cgroup *cgrp)
592{
593    struct css_set *res;
594    struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT];
595
596    struct list_head tmp_cg_links;
597
598    struct hlist_head *hhead;
599    struct cg_cgroup_link *link;
600
601    /* First see if we already have a cgroup group that matches
602     * the desired set */
603    read_lock(&css_set_lock);
604    res = find_existing_css_set(oldcg, cgrp, template);
605    if (res)
606        get_css_set(res);
607    read_unlock(&css_set_lock);
608
609    if (res)
610        return res;
611
612    res = kmalloc(sizeof(*res), GFP_KERNEL);
613    if (!res)
614        return NULL;
615
616    /* Allocate all the cg_cgroup_link objects that we'll need */
617    if (allocate_cg_links(root_count, &tmp_cg_links) < 0) {
618        kfree(res);
619        return NULL;
620    }
621
622    atomic_set(&res->refcount, 1);
623    INIT_LIST_HEAD(&res->cg_links);
624    INIT_LIST_HEAD(&res->tasks);
625    INIT_HLIST_NODE(&res->hlist);
626
627    /* Copy the set of subsystem state objects generated in
628     * find_existing_css_set() */
629    memcpy(res->subsys, template, sizeof(res->subsys));
630
631    write_lock(&css_set_lock);
632    /* Add reference counts and links from the new css_set. */
633    list_for_each_entry(link, &oldcg->cg_links, cg_link_list) {
634        struct cgroup *c = link->cgrp;
635        if (c->root == cgrp->root)
636            c = cgrp;
637        link_css_set(&tmp_cg_links, res, c);
638    }
639
640    BUG_ON(!list_empty(&tmp_cg_links));
641
642    css_set_count++;
643
644    /* Add this cgroup group to the hash table */
645    hhead = css_set_hash(res->subsys);
646    hlist_add_head(&res->hlist, hhead);
647
648    write_unlock(&css_set_lock);
649
650    return res;
651}
652
653/*
654 * Return the cgroup for "task" from the given hierarchy. Must be
655 * called with cgroup_mutex held.
656 */
657static struct cgroup *task_cgroup_from_root(struct task_struct *task,
658                        struct cgroupfs_root *root)
659{
660    struct css_set *css;
661    struct cgroup *res = NULL;
662
663    BUG_ON(!mutex_is_locked(&cgroup_mutex));
664    read_lock(&css_set_lock);
665    /*
666     * No need to lock the task - since we hold cgroup_mutex the
667     * task can't change groups, so the only thing that can happen
668     * is that it exits and its css is set back to init_css_set.
669     */
670    css = task->cgroups;
671    if (css == &init_css_set) {
672        res = &root->top_cgroup;
673    } else {
674        struct cg_cgroup_link *link;
675        list_for_each_entry(link, &css->cg_links, cg_link_list) {
676            struct cgroup *c = link->cgrp;
677            if (c->root == root) {
678                res = c;
679                break;
680            }
681        }
682    }
683    read_unlock(&css_set_lock);
684    BUG_ON(!res);
685    return res;
686}
687
688/*
689 * There is one global cgroup mutex. We also require taking
690 * task_lock() when dereferencing a task's cgroup subsys pointers.
691 * See "The task_lock() exception", at the end of this comment.
692 *
693 * A task must hold cgroup_mutex to modify cgroups.
694 *
695 * Any task can increment and decrement the count field without lock.
696 * So in general, code holding cgroup_mutex can't rely on the count
697 * field not changing. However, if the count goes to zero, then only
698 * cgroup_attach_task() can increment it again. Because a count of zero
699 * means that no tasks are currently attached, therefore there is no
700 * way a task attached to that cgroup can fork (the other way to
701 * increment the count). So code holding cgroup_mutex can safely
702 * assume that if the count is zero, it will stay zero. Similarly, if
703 * a task holds cgroup_mutex on a cgroup with zero count, it
704 * knows that the cgroup won't be removed, as cgroup_rmdir()
705 * needs that mutex.
706 *
707 * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't
708 * (usually) take cgroup_mutex. These are the two most performance
709 * critical pieces of code here. The exception occurs on cgroup_exit(),
710 * when a task in a notify_on_release cgroup exits. Then cgroup_mutex
711 * is taken, and if the cgroup count is zero, a usermode call made
712 * to the release agent with the name of the cgroup (path relative to
713 * the root of cgroup file system) as the argument.
714 *
715 * A cgroup can only be deleted if both its 'count' of using tasks
716 * is zero, and its list of 'children' cgroups is empty. Since all
717 * tasks in the system use _some_ cgroup, and since there is always at
718 * least one task in the system (init, pid == 1), therefore, top_cgroup
719 * always has either children cgroups and/or using tasks. So we don't
720 * need a special hack to ensure that top_cgroup cannot be deleted.
721 *
722 * The task_lock() exception
723 *
724 * The need for this exception arises from the action of
725 * cgroup_attach_task(), which overwrites one tasks cgroup pointer with
726 * another. It does so using cgroup_mutex, however there are
727 * several performance critical places that need to reference
728 * task->cgroup without the expense of grabbing a system global
729 * mutex. Therefore except as noted below, when dereferencing or, as
730 * in cgroup_attach_task(), modifying a task'ss cgroup pointer we use
731 * task_lock(), which acts on a spinlock (task->alloc_lock) already in
732 * the task_struct routinely used for such matters.
733 *
734 * P.S. One more locking exception. RCU is used to guard the
735 * update of a tasks cgroup pointer by cgroup_attach_task()
736 */
737
738/**
739 * cgroup_lock - lock out any changes to cgroup structures
740 *
741 */
742void cgroup_lock(void)
743{
744    mutex_lock(&cgroup_mutex);
745}
746EXPORT_SYMBOL_GPL(cgroup_lock);
747
748/**
749 * cgroup_unlock - release lock on cgroup changes
750 *
751 * Undo the lock taken in a previous cgroup_lock() call.
752 */
753void cgroup_unlock(void)
754{
755    mutex_unlock(&cgroup_mutex);
756}
757EXPORT_SYMBOL_GPL(cgroup_unlock);
758
759/*
760 * A couple of forward declarations required, due to cyclic reference loop:
761 * cgroup_mkdir -> cgroup_create -> cgroup_populate_dir ->
762 * cgroup_add_file -> cgroup_create_file -> cgroup_dir_inode_operations
763 * -> cgroup_mkdir.
764 */
765
766static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, int mode);
767static struct dentry *cgroup_lookup(struct inode *, struct dentry *, struct nameidata *);
768static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry);
769static int cgroup_populate_dir(struct cgroup *cgrp);
770static const struct inode_operations cgroup_dir_inode_operations;
771static const struct file_operations proc_cgroupstats_operations;
772
773static struct backing_dev_info cgroup_backing_dev_info = {
774    .name = "cgroup",
775    .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK,
776};
777
778static int alloc_css_id(struct cgroup_subsys *ss,
779            struct cgroup *parent, struct cgroup *child);
780
781static struct inode *cgroup_new_inode(mode_t mode, struct super_block *sb)
782{
783    struct inode *inode = new_inode(sb);
784
785    if (inode) {
786        inode->i_ino = get_next_ino();
787        inode->i_mode = mode;
788        inode->i_uid = current_fsuid();
789        inode->i_gid = current_fsgid();
790        inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
791        inode->i_mapping->backing_dev_info = &cgroup_backing_dev_info;
792    }
793    return inode;
794}
795
796/*
797 * Call subsys's pre_destroy handler.
798 * This is called before css refcnt check.
799 */
800static int cgroup_call_pre_destroy(struct cgroup *cgrp)
801{
802    struct cgroup_subsys *ss;
803    int ret = 0;
804
805    for_each_subsys(cgrp->root, ss)
806        if (ss->pre_destroy) {
807            ret = ss->pre_destroy(ss, cgrp);
808            if (ret)
809                break;
810        }
811
812    return ret;
813}
814
815static void free_cgroup_rcu(struct rcu_head *obj)
816{
817    struct cgroup *cgrp = container_of(obj, struct cgroup, rcu_head);
818
819    kfree(cgrp);
820}
821
822static void cgroup_diput(struct dentry *dentry, struct inode *inode)
823{
824    /* is dentry a directory ? if so, kfree() associated cgroup */
825    if (S_ISDIR(inode->i_mode)) {
826        struct cgroup *cgrp = dentry->d_fsdata;
827        struct cgroup_subsys *ss;
828        BUG_ON(!(cgroup_is_removed(cgrp)));
829        /* It's possible for external users to be holding css
830         * reference counts on a cgroup; css_put() needs to
831         * be able to access the cgroup after decrementing
832         * the reference count in order to know if it needs to
833         * queue the cgroup to be handled by the release
834         * agent */
835        synchronize_rcu();
836
837        mutex_lock(&cgroup_mutex);
838        /*
839         * Release the subsystem state objects.
840         */
841        for_each_subsys(cgrp->root, ss)
842            ss->destroy(ss, cgrp);
843
844        cgrp->root->number_of_cgroups--;
845        mutex_unlock(&cgroup_mutex);
846
847        /*
848         * Drop the active superblock reference that we took when we
849         * created the cgroup
850         */
851        deactivate_super(cgrp->root->sb);
852
853        /*
854         * if we're getting rid of the cgroup, refcount should ensure
855         * that there are no pidlists left.
856         */
857        BUG_ON(!list_empty(&cgrp->pidlists));
858
859        call_rcu(&cgrp->rcu_head, free_cgroup_rcu);
860    }
861    iput(inode);
862}
863
864static int cgroup_delete(const struct dentry *d)
865{
866    return 1;
867}
868
869static void remove_dir(struct dentry *d)
870{
871    struct dentry *parent = dget(d->d_parent);
872
873    d_delete(d);
874    simple_rmdir(parent->d_inode, d);
875    dput(parent);
876}
877
878static void cgroup_clear_directory(struct dentry *dentry)
879{
880    struct list_head *node;
881
882    BUG_ON(!mutex_is_locked(&dentry->d_inode->i_mutex));
883    spin_lock(&dentry->d_lock);
884    node = dentry->d_subdirs.next;
885    while (node != &dentry->d_subdirs) {
886        struct dentry *d = list_entry(node, struct dentry, d_u.d_child);
887
888        spin_lock_nested(&d->d_lock, DENTRY_D_LOCK_NESTED);
889        list_del_init(node);
890        if (d->d_inode) {
891            /* This should never be called on a cgroup
892             * directory with child cgroups */
893            BUG_ON(d->d_inode->i_mode & S_IFDIR);
894            dget_dlock(d);
895            spin_unlock(&d->d_lock);
896            spin_unlock(&dentry->d_lock);
897            d_delete(d);
898            simple_unlink(dentry->d_inode, d);
899            dput(d);
900            spin_lock(&dentry->d_lock);
901        } else
902            spin_unlock(&d->d_lock);
903        node = dentry->d_subdirs.next;
904    }
905    spin_unlock(&dentry->d_lock);
906}
907
908/*
909 * NOTE : the dentry must have been dget()'ed
910 */
911static void cgroup_d_remove_dir(struct dentry *dentry)
912{
913    struct dentry *parent;
914
915    cgroup_clear_directory(dentry);
916
917    parent = dentry->d_parent;
918    spin_lock(&parent->d_lock);
919    spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
920    list_del_init(&dentry->d_u.d_child);
921    spin_unlock(&dentry->d_lock);
922    spin_unlock(&parent->d_lock);
923    remove_dir(dentry);
924}
925
926/*
927 * A queue for waiters to do rmdir() cgroup. A tasks will sleep when
928 * cgroup->count == 0 && list_empty(&cgroup->children) && subsys has some
929 * reference to css->refcnt. In general, this refcnt is expected to goes down
930 * to zero, soon.
931 *
932 * CGRP_WAIT_ON_RMDIR flag is set under cgroup's inode->i_mutex;
933 */
934DECLARE_WAIT_QUEUE_HEAD(cgroup_rmdir_waitq);
935
936static void cgroup_wakeup_rmdir_waiter(struct cgroup *cgrp)
937{
938    if (unlikely(test_and_clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags)))
939        wake_up_all(&cgroup_rmdir_waitq);
940}
941
942void cgroup_exclude_rmdir(struct cgroup_subsys_state *css)
943{
944    css_get(css);
945}
946
947void cgroup_release_and_wakeup_rmdir(struct cgroup_subsys_state *css)
948{
949    cgroup_wakeup_rmdir_waiter(css->cgroup);
950    css_put(css);
951}
952
953/*
954 * Call with cgroup_mutex held. Drops reference counts on modules, including
955 * any duplicate ones that parse_cgroupfs_options took. If this function
956 * returns an error, no reference counts are touched.
957 */
958static int rebind_subsystems(struct cgroupfs_root *root,
959                  unsigned long final_bits)
960{
961    unsigned long added_bits, removed_bits;
962    struct cgroup *cgrp = &root->top_cgroup;
963    int i;
964
965    BUG_ON(!mutex_is_locked(&cgroup_mutex));
966
967    removed_bits = root->actual_subsys_bits & ~final_bits;
968    added_bits = final_bits & ~root->actual_subsys_bits;
969    /* Check that any added subsystems are currently free */
970    for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
971        unsigned long bit = 1UL << i;
972        struct cgroup_subsys *ss = subsys[i];
973        if (!(bit & added_bits))
974            continue;
975        /*
976         * Nobody should tell us to do a subsys that doesn't exist:
977         * parse_cgroupfs_options should catch that case and refcounts
978         * ensure that subsystems won't disappear once selected.
979         */
980        BUG_ON(ss == NULL);
981        if (ss->root != &rootnode) {
982            /* Subsystem isn't free */
983            return -EBUSY;
984        }
985    }
986
987    /* Currently we don't handle adding/removing subsystems when
988     * any child cgroups exist. This is theoretically supportable
989     * but involves complex error handling, so it's being left until
990     * later */
991    if (root->number_of_cgroups > 1)
992        return -EBUSY;
993
994    /* Process each subsystem */
995    for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
996        struct cgroup_subsys *ss = subsys[i];
997        unsigned long bit = 1UL << i;
998        if (bit & added_bits) {
999            /* We're binding this subsystem to this hierarchy */
1000            BUG_ON(ss == NULL);
1001            BUG_ON(cgrp->subsys[i]);
1002            BUG_ON(!dummytop->subsys[i]);
1003            BUG_ON(dummytop->subsys[i]->cgroup != dummytop);
1004            mutex_lock(&ss->hierarchy_mutex);
1005            cgrp->subsys[i] = dummytop->subsys[i];
1006            cgrp->subsys[i]->cgroup = cgrp;
1007            list_move(&ss->sibling, &root->subsys_list);
1008            ss->root = root;
1009            if (ss->bind)
1010                ss->bind(ss, cgrp);
1011            mutex_unlock(&ss->hierarchy_mutex);
1012            /* refcount was already taken, and we're keeping it */
1013        } else if (bit & removed_bits) {
1014            /* We're removing this subsystem */
1015            BUG_ON(ss == NULL);
1016            BUG_ON(cgrp->subsys[i] != dummytop->subsys[i]);
1017            BUG_ON(cgrp->subsys[i]->cgroup != cgrp);
1018            mutex_lock(&ss->hierarchy_mutex);
1019            if (ss->bind)
1020                ss->bind(ss, dummytop);
1021            dummytop->subsys[i]->cgroup = dummytop;
1022            cgrp->subsys[i] = NULL;
1023            subsys[i]->root = &rootnode;
1024            list_move(&ss->sibling, &rootnode.subsys_list);
1025            mutex_unlock(&ss->hierarchy_mutex);
1026            /* subsystem is now free - drop reference on module */
1027            module_put(ss->module);
1028        } else if (bit & final_bits) {
1029            /* Subsystem state should already exist */
1030            BUG_ON(ss == NULL);
1031            BUG_ON(!cgrp->subsys[i]);
1032            /*
1033             * a refcount was taken, but we already had one, so
1034             * drop the extra reference.
1035             */
1036            module_put(ss->module);
1037#ifdef CONFIG_MODULE_UNLOAD
1038            BUG_ON(ss->module && !module_refcount(ss->module));
1039#endif
1040        } else {
1041            /* Subsystem state shouldn't exist */
1042            BUG_ON(cgrp->subsys[i]);
1043        }
1044    }
1045    root->subsys_bits = root->actual_subsys_bits = final_bits;
1046    synchronize_rcu();
1047
1048    return 0;
1049}
1050
1051static int cgroup_show_options(struct seq_file *seq, struct vfsmount *vfs)
1052{
1053    struct cgroupfs_root *root = vfs->mnt_sb->s_fs_info;
1054    struct cgroup_subsys *ss;
1055
1056    mutex_lock(&cgroup_mutex);
1057    for_each_subsys(root, ss)
1058        seq_printf(seq, ",%s", ss->name);
1059    if (test_bit(ROOT_NOPREFIX, &root->flags))
1060        seq_puts(seq, ",noprefix");
1061    if (strlen(root->release_agent_path))
1062        seq_printf(seq, ",release_agent=%s", root->release_agent_path);
1063    if (clone_children(&root->top_cgroup))
1064        seq_puts(seq, ",clone_children");
1065    if (strlen(root->name))
1066        seq_printf(seq, ",name=%s", root->name);
1067    mutex_unlock(&cgroup_mutex);
1068    return 0;
1069}
1070
1071struct cgroup_sb_opts {
1072    unsigned long subsys_bits;
1073    unsigned long flags;
1074    char *release_agent;
1075    bool clone_children;
1076    char *name;
1077    /* User explicitly requested empty subsystem */
1078    bool none;
1079
1080    struct cgroupfs_root *new_root;
1081
1082};
1083
1084/*
1085 * Convert a hierarchy specifier into a bitmask of subsystems and flags. Call
1086 * with cgroup_mutex held to protect the subsys[] array. This function takes
1087 * refcounts on subsystems to be used, unless it returns error, in which case
1088 * no refcounts are taken.
1089 */
1090static int parse_cgroupfs_options(char *data, struct cgroup_sb_opts *opts)
1091{
1092    char *token, *o = data;
1093    bool all_ss = false, one_ss = false;
1094    unsigned long mask = (unsigned long)-1;
1095    int i;
1096    bool module_pin_failed = false;
1097
1098    BUG_ON(!mutex_is_locked(&cgroup_mutex));
1099
1100#ifdef CONFIG_CPUSETS
1101    mask = ~(1UL << cpuset_subsys_id);
1102#endif
1103
1104    memset(opts, 0, sizeof(*opts));
1105
1106    while ((token = strsep(&o, ",")) != NULL) {
1107        if (!*token)
1108            return -EINVAL;
1109        if (!strcmp(token, "none")) {
1110            /* Explicitly have no subsystems */
1111            opts->none = true;
1112            continue;
1113        }
1114        if (!strcmp(token, "all")) {
1115            /* Mutually exclusive option 'all' + subsystem name */
1116            if (one_ss)
1117                return -EINVAL;
1118            all_ss = true;
1119            continue;
1120        }
1121        if (!strcmp(token, "noprefix")) {
1122            set_bit(ROOT_NOPREFIX, &opts->flags);
1123            continue;
1124        }
1125        if (!strcmp(token, "clone_children")) {
1126            opts->clone_children = true;
1127            continue;
1128        }
1129        if (!strncmp(token, "release_agent=", 14)) {
1130            /* Specifying two release agents is forbidden */
1131            if (opts->release_agent)
1132                return -EINVAL;
1133            opts->release_agent =
1134                kstrndup(token + 14, PATH_MAX - 1, GFP_KERNEL);
1135            if (!opts->release_agent)
1136                return -ENOMEM;
1137            continue;
1138        }
1139        if (!strncmp(token, "name=", 5)) {
1140            const char *name = token + 5;
1141            /* Can't specify an empty name */
1142            if (!strlen(name))
1143                return -EINVAL;
1144            /* Must match [\w.-]+ */
1145            for (i = 0; i < strlen(name); i++) {
1146                char c = name[i];
1147                if (isalnum(c))
1148                    continue;
1149                if ((c == '.') || (c == '-') || (c == '_'))
1150                    continue;
1151                return -EINVAL;
1152            }
1153            /* Specifying two names is forbidden */
1154            if (opts->name)
1155                return -EINVAL;
1156            opts->name = kstrndup(name,
1157                          MAX_CGROUP_ROOT_NAMELEN - 1,
1158                          GFP_KERNEL);
1159            if (!opts->name)
1160                return -ENOMEM;
1161
1162            continue;
1163        }
1164
1165        for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1166            struct cgroup_subsys *ss = subsys[i];
1167            if (ss == NULL)
1168                continue;
1169            if (strcmp(token, ss->name))
1170                continue;
1171            if (ss->disabled)
1172                continue;
1173
1174            /* Mutually exclusive option 'all' + subsystem name */
1175            if (all_ss)
1176                return -EINVAL;
1177            set_bit(i, &opts->subsys_bits);
1178            one_ss = true;
1179
1180            break;
1181        }
1182        if (i == CGROUP_SUBSYS_COUNT)
1183            return -ENOENT;
1184    }
1185
1186    /*
1187     * If the 'all' option was specified select all the subsystems,
1188     * otherwise 'all, 'none' and a subsystem name options were not
1189     * specified, let's default to 'all'
1190     */
1191    if (all_ss || (!all_ss && !one_ss && !opts->none)) {
1192        for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1193            struct cgroup_subsys *ss = subsys[i];
1194            if (ss == NULL)
1195                continue;
1196            if (ss->disabled)
1197                continue;
1198            set_bit(i, &opts->subsys_bits);
1199        }
1200    }
1201
1202    /* Consistency checks */
1203
1204    /*
1205     * Option noprefix was introduced just for backward compatibility
1206     * with the old cpuset, so we allow noprefix only if mounting just
1207     * the cpuset subsystem.
1208     */
1209    if (test_bit(ROOT_NOPREFIX, &opts->flags) &&
1210        (opts->subsys_bits & mask))
1211        return -EINVAL;
1212
1213
1214    /* Can't specify "none" and some subsystems */
1215    if (opts->subsys_bits && opts->none)
1216        return -EINVAL;
1217
1218    /*
1219     * We either have to specify by name or by subsystems. (So all
1220     * empty hierarchies must have a name).
1221     */
1222    if (!opts->subsys_bits && !opts->name)
1223        return -EINVAL;
1224
1225    /*
1226     * Grab references on all the modules we'll need, so the subsystems
1227     * don't dance around before rebind_subsystems attaches them. This may
1228     * take duplicate reference counts on a subsystem that's already used,
1229     * but rebind_subsystems handles this case.
1230     */
1231    for (i = CGROUP_BUILTIN_SUBSYS_COUNT; i < CGROUP_SUBSYS_COUNT; i++) {
1232        unsigned long bit = 1UL << i;
1233
1234        if (!(bit & opts->subsys_bits))
1235            continue;
1236        if (!try_module_get(subsys[i]->module)) {
1237            module_pin_failed = true;
1238            break;
1239        }
1240    }
1241    if (module_pin_failed) {
1242        /*
1243         * oops, one of the modules was going away. this means that we
1244         * raced with a module_delete call, and to the user this is
1245         * essentially a "subsystem doesn't exist" case.
1246         */
1247        for (i--; i >= CGROUP_BUILTIN_SUBSYS_COUNT; i--) {
1248            /* drop refcounts only on the ones we took */
1249            unsigned long bit = 1UL << i;
1250
1251            if (!(bit & opts->subsys_bits))
1252                continue;
1253            module_put(subsys[i]->module);
1254        }
1255        return -ENOENT;
1256    }
1257
1258    return 0;
1259}
1260
1261static void drop_parsed_module_refcounts(unsigned long subsys_bits)
1262{
1263    int i;
1264    for (i = CGROUP_BUILTIN_SUBSYS_COUNT; i < CGROUP_SUBSYS_COUNT; i++) {
1265        unsigned long bit = 1UL << i;
1266
1267        if (!(bit & subsys_bits))
1268            continue;
1269        module_put(subsys[i]->module);
1270    }
1271}
1272
1273static int cgroup_remount(struct super_block *sb, int *flags, char *data)
1274{
1275    int ret = 0;
1276    struct cgroupfs_root *root = sb->s_fs_info;
1277    struct cgroup *cgrp = &root->top_cgroup;
1278    struct cgroup_sb_opts opts;
1279
1280    mutex_lock(&cgrp->dentry->d_inode->i_mutex);
1281    mutex_lock(&cgroup_mutex);
1282
1283    /* See what subsystems are wanted */
1284    ret = parse_cgroupfs_options(data, &opts);
1285    if (ret)
1286        goto out_unlock;
1287
1288    /* Don't allow flags or name to change at remount */
1289    if (opts.flags != root->flags ||
1290        (opts.name && strcmp(opts.name, root->name))) {
1291        ret = -EINVAL;
1292        drop_parsed_module_refcounts(opts.subsys_bits);
1293        goto out_unlock;
1294    }
1295
1296    ret = rebind_subsystems(root, opts.subsys_bits);
1297    if (ret) {
1298        drop_parsed_module_refcounts(opts.subsys_bits);
1299        goto out_unlock;
1300    }
1301
1302    /* (re)populate subsystem files */
1303    cgroup_populate_dir(cgrp);
1304
1305    if (opts.release_agent)
1306        strcpy(root->release_agent_path, opts.release_agent);
1307 out_unlock:
1308    kfree(opts.release_agent);
1309    kfree(opts.name);
1310    mutex_unlock(&cgroup_mutex);
1311    mutex_unlock(&cgrp->dentry->d_inode->i_mutex);
1312    return ret;
1313}
1314
1315static const struct super_operations cgroup_ops = {
1316    .statfs = simple_statfs,
1317    .drop_inode = generic_delete_inode,
1318    .show_options = cgroup_show_options,
1319    .remount_fs = cgroup_remount,
1320};
1321
1322static void init_cgroup_housekeeping(struct cgroup *cgrp)
1323{
1324    INIT_LIST_HEAD(&cgrp->sibling);
1325    INIT_LIST_HEAD(&cgrp->children);
1326    INIT_LIST_HEAD(&cgrp->css_sets);
1327    INIT_LIST_HEAD(&cgrp->release_list);
1328    INIT_LIST_HEAD(&cgrp->pidlists);
1329    mutex_init(&cgrp->pidlist_mutex);
1330    INIT_LIST_HEAD(&cgrp->event_list);
1331    spin_lock_init(&cgrp->event_list_lock);
1332}
1333
1334static void init_cgroup_root(struct cgroupfs_root *root)
1335{
1336    struct cgroup *cgrp = &root->top_cgroup;
1337    INIT_LIST_HEAD(&root->subsys_list);
1338    INIT_LIST_HEAD(&root->root_list);
1339    root->number_of_cgroups = 1;
1340    cgrp->root = root;
1341    cgrp->top_cgroup = cgrp;
1342    init_cgroup_housekeeping(cgrp);
1343}
1344
1345static bool init_root_id(struct cgroupfs_root *root)
1346{
1347    int ret = 0;
1348
1349    do {
1350        if (!ida_pre_get(&hierarchy_ida, GFP_KERNEL))
1351            return false;
1352        spin_lock(&hierarchy_id_lock);
1353        /* Try to allocate the next unused ID */
1354        ret = ida_get_new_above(&hierarchy_ida, next_hierarchy_id,
1355                    &root->hierarchy_id);
1356        if (ret == -ENOSPC)
1357            /* Try again starting from 0 */
1358            ret = ida_get_new(&hierarchy_ida, &root->hierarchy_id);
1359        if (!ret) {
1360            next_hierarchy_id = root->hierarchy_id + 1;
1361        } else if (ret != -EAGAIN) {
1362            /* Can only get here if the 31-bit IDR is full ... */
1363            BUG_ON(ret);
1364        }
1365        spin_unlock(&hierarchy_id_lock);
1366    } while (ret);
1367    return true;
1368}
1369
1370static int cgroup_test_super(struct super_block *sb, void *data)
1371{
1372    struct cgroup_sb_opts *opts = data;
1373    struct cgroupfs_root *root = sb->s_fs_info;
1374
1375    /* If we asked for a name then it must match */
1376    if (opts->name && strcmp(opts->name, root->name))
1377        return 0;
1378
1379    /*
1380     * If we asked for subsystems (or explicitly for no
1381     * subsystems) then they must match
1382     */
1383    if ((opts->subsys_bits || opts->none)
1384        && (opts->subsys_bits != root->subsys_bits))
1385        return 0;
1386
1387    return 1;
1388}
1389
1390static struct cgroupfs_root *cgroup_root_from_opts(struct cgroup_sb_opts *opts)
1391{
1392    struct cgroupfs_root *root;
1393
1394    if (!opts->subsys_bits && !opts->none)
1395        return NULL;
1396
1397    root = kzalloc(sizeof(*root), GFP_KERNEL);
1398    if (!root)
1399        return ERR_PTR(-ENOMEM);
1400
1401    if (!init_root_id(root)) {
1402        kfree(root);
1403        return ERR_PTR(-ENOMEM);
1404    }
1405    init_cgroup_root(root);
1406
1407    root->subsys_bits = opts->subsys_bits;
1408    root->flags = opts->flags;
1409    if (opts->release_agent)
1410        strcpy(root->release_agent_path, opts->release_agent);
1411    if (opts->name)
1412        strcpy(root->name, opts->name);
1413    if (opts->clone_children)
1414        set_bit(CGRP_CLONE_CHILDREN, &root->top_cgroup.flags);
1415    return root;
1416}
1417
1418static void cgroup_drop_root(struct cgroupfs_root *root)
1419{
1420    if (!root)
1421        return;
1422
1423    BUG_ON(!root->hierarchy_id);
1424    spin_lock(&hierarchy_id_lock);
1425    ida_remove(&hierarchy_ida, root->hierarchy_id);
1426    spin_unlock(&hierarchy_id_lock);
1427    kfree(root);
1428}
1429
1430static int cgroup_set_super(struct super_block *sb, void *data)
1431{
1432    int ret;
1433    struct cgroup_sb_opts *opts = data;
1434
1435    /* If we don't have a new root, we can't set up a new sb */
1436    if (!opts->new_root)
1437        return -EINVAL;
1438
1439    BUG_ON(!opts->subsys_bits && !opts->none);
1440
1441    ret = set_anon_super(sb, NULL);
1442    if (ret)
1443        return ret;
1444
1445    sb->s_fs_info = opts->new_root;
1446    opts->new_root->sb = sb;
1447
1448    sb->s_blocksize = PAGE_CACHE_SIZE;
1449    sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
1450    sb->s_magic = CGROUP_SUPER_MAGIC;
1451    sb->s_op = &cgroup_ops;
1452
1453    return 0;
1454}
1455
1456static int cgroup_get_rootdir(struct super_block *sb)
1457{
1458    static const struct dentry_operations cgroup_dops = {
1459        .d_iput = cgroup_diput,
1460        .d_delete = cgroup_delete,
1461    };
1462
1463    struct inode *inode =
1464        cgroup_new_inode(S_IFDIR | S_IRUGO | S_IXUGO | S_IWUSR, sb);
1465    struct dentry *dentry;
1466
1467    if (!inode)
1468        return -ENOMEM;
1469
1470    inode->i_fop = &simple_dir_operations;
1471    inode->i_op = &cgroup_dir_inode_operations;
1472    /* directories start off with i_nlink == 2 (for "." entry) */
1473    inc_nlink(inode);
1474    dentry = d_alloc_root(inode);
1475    if (!dentry) {
1476        iput(inode);
1477        return -ENOMEM;
1478    }
1479    sb->s_root = dentry;
1480    /* for everything else we want ->d_op set */
1481    sb->s_d_op = &cgroup_dops;
1482    return 0;
1483}
1484
1485static struct dentry *cgroup_mount(struct file_system_type *fs_type,
1486             int flags, const char *unused_dev_name,
1487             void *data)
1488{
1489    struct cgroup_sb_opts opts;
1490    struct cgroupfs_root *root;
1491    int ret = 0;
1492    struct super_block *sb;
1493    struct cgroupfs_root *new_root;
1494
1495    /* First find the desired set of subsystems */
1496    mutex_lock(&cgroup_mutex);
1497    ret = parse_cgroupfs_options(data, &opts);
1498    mutex_unlock(&cgroup_mutex);
1499    if (ret)
1500        goto out_err;
1501
1502    /*
1503     * Allocate a new cgroup root. We may not need it if we're
1504     * reusing an existing hierarchy.
1505     */
1506    new_root = cgroup_root_from_opts(&opts);
1507    if (IS_ERR(new_root)) {
1508        ret = PTR_ERR(new_root);
1509        goto drop_modules;
1510    }
1511    opts.new_root = new_root;
1512
1513    /* Locate an existing or new sb for this hierarchy */
1514    sb = sget(fs_type, cgroup_test_super, cgroup_set_super, &opts);
1515    if (IS_ERR(sb)) {
1516        ret = PTR_ERR(sb);
1517        cgroup_drop_root(opts.new_root);
1518        goto drop_modules;
1519    }
1520
1521    root = sb->s_fs_info;
1522    BUG_ON(!root);
1523    if (root == opts.new_root) {
1524        /* We used the new root structure, so this is a new hierarchy */
1525        struct list_head tmp_cg_links;
1526        struct cgroup *root_cgrp = &root->top_cgroup;
1527        struct inode *inode;
1528        struct cgroupfs_root *existing_root;
1529        int i;
1530
1531        BUG_ON(sb->s_root != NULL);
1532
1533        ret = cgroup_get_rootdir(sb);
1534        if (ret)
1535            goto drop_new_super;
1536        inode = sb->s_root->d_inode;
1537
1538        mutex_lock(&inode->i_mutex);
1539        mutex_lock(&cgroup_mutex);
1540
1541        if (strlen(root->name)) {
1542            /* Check for name clashes with existing mounts */
1543            for_each_active_root(existing_root) {
1544                if (!strcmp(existing_root->name, root->name)) {
1545                    ret = -EBUSY;
1546                    mutex_unlock(&cgroup_mutex);
1547                    mutex_unlock(&inode->i_mutex);
1548                    goto drop_new_super;
1549                }
1550            }
1551        }
1552
1553        /*
1554         * We're accessing css_set_count without locking
1555         * css_set_lock here, but that's OK - it can only be
1556         * increased by someone holding cgroup_lock, and
1557         * that's us. The worst that can happen is that we
1558         * have some link structures left over
1559         */
1560        ret = allocate_cg_links(css_set_count, &tmp_cg_links);
1561        if (ret) {
1562            mutex_unlock(&cgroup_mutex);
1563            mutex_unlock(&inode->i_mutex);
1564            goto drop_new_super;
1565        }
1566
1567        ret = rebind_subsystems(root, root->subsys_bits);
1568        if (ret == -EBUSY) {
1569            mutex_unlock(&cgroup_mutex);
1570            mutex_unlock(&inode->i_mutex);
1571            free_cg_links(&tmp_cg_links);
1572            goto drop_new_super;
1573        }
1574        /*
1575         * There must be no failure case after here, since rebinding
1576         * takes care of subsystems' refcounts, which are explicitly
1577         * dropped in the failure exit path.
1578         */
1579
1580        /* EBUSY should be the only error here */
1581        BUG_ON(ret);
1582
1583        list_add(&root->root_list, &roots);
1584        root_count++;
1585
1586        sb->s_root->d_fsdata = root_cgrp;
1587        root->top_cgroup.dentry = sb->s_root;
1588
1589        /* Link the top cgroup in this hierarchy into all
1590         * the css_set objects */
1591        write_lock(&css_set_lock);
1592        for (i = 0; i < CSS_SET_TABLE_SIZE; i++) {
1593            struct hlist_head *hhead = &css_set_table[i];
1594            struct hlist_node *node;
1595            struct css_set *cg;
1596
1597            hlist_for_each_entry(cg, node, hhead, hlist)
1598                link_css_set(&tmp_cg_links, cg, root_cgrp);
1599        }
1600        write_unlock(&css_set_lock);
1601
1602        free_cg_links(&tmp_cg_links);
1603
1604        BUG_ON(!list_empty(&root_cgrp->sibling));
1605        BUG_ON(!list_empty(&root_cgrp->children));
1606        BUG_ON(root->number_of_cgroups != 1);
1607
1608        cgroup_populate_dir(root_cgrp);
1609        mutex_unlock(&cgroup_mutex);
1610        mutex_unlock(&inode->i_mutex);
1611    } else {
1612        /*
1613         * We re-used an existing hierarchy - the new root (if
1614         * any) is not needed
1615         */
1616        cgroup_drop_root(opts.new_root);
1617        /* no subsys rebinding, so refcounts don't change */
1618        drop_parsed_module_refcounts(opts.subsys_bits);
1619    }
1620
1621    kfree(opts.release_agent);
1622    kfree(opts.name);
1623    return dget(sb->s_root);
1624
1625 drop_new_super:
1626    deactivate_locked_super(sb);
1627 drop_modules:
1628    drop_parsed_module_refcounts(opts.subsys_bits);
1629 out_err:
1630    kfree(opts.release_agent);
1631    kfree(opts.name);
1632    return ERR_PTR(ret);
1633}
1634
1635static void cgroup_kill_sb(struct super_block *sb) {
1636    struct cgroupfs_root *root = sb->s_fs_info;
1637    struct cgroup *cgrp = &root->top_cgroup;
1638    int ret;
1639    struct cg_cgroup_link *link;
1640    struct cg_cgroup_link *saved_link;
1641
1642    BUG_ON(!root);
1643
1644    BUG_ON(root->number_of_cgroups != 1);
1645    BUG_ON(!list_empty(&cgrp->children));
1646    BUG_ON(!list_empty(&cgrp->sibling));
1647
1648    mutex_lock(&cgroup_mutex);
1649
1650    /* Rebind all subsystems back to the default hierarchy */
1651    ret = rebind_subsystems(root, 0);
1652    /* Shouldn't be able to fail ... */
1653    BUG_ON(ret);
1654
1655    /*
1656     * Release all the links from css_sets to this hierarchy's
1657     * root cgroup
1658     */
1659    write_lock(&css_set_lock);
1660
1661    list_for_each_entry_safe(link, saved_link, &cgrp->css_sets,
1662                 cgrp_link_list) {
1663        list_del(&link->cg_link_list);
1664        list_del(&link->cgrp_link_list);
1665        kfree(link);
1666    }
1667    write_unlock(&css_set_lock);
1668
1669    if (!list_empty(&root->root_list)) {
1670        list_del(&root->root_list);
1671        root_count--;
1672    }
1673
1674    mutex_unlock(&cgroup_mutex);
1675
1676    kill_litter_super(sb);
1677    cgroup_drop_root(root);
1678}
1679
1680static struct file_system_type cgroup_fs_type = {
1681    .name = "cgroup",
1682    .mount = cgroup_mount,
1683    .kill_sb = cgroup_kill_sb,
1684};
1685
1686static struct kobject *cgroup_kobj;
1687
1688static inline struct cgroup *__d_cgrp(struct dentry *dentry)
1689{
1690    return dentry->d_fsdata;
1691}
1692
1693static inline struct cftype *__d_cft(struct dentry *dentry)
1694{
1695    return dentry->d_fsdata;
1696}
1697
1698/**
1699 * cgroup_path - generate the path of a cgroup
1700 * @cgrp: the cgroup in question
1701 * @buf: the buffer to write the path into
1702 * @buflen: the length of the buffer
1703 *
1704 * Called with cgroup_mutex held or else with an RCU-protected cgroup
1705 * reference. Writes path of cgroup into buf. Returns 0 on success,
1706 * -errno on error.
1707 */
1708int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen)
1709{
1710    char *start;
1711    struct dentry *dentry = rcu_dereference_check(cgrp->dentry,
1712                              rcu_read_lock_held() ||
1713                              cgroup_lock_is_held());
1714
1715    if (!dentry || cgrp == dummytop) {
1716        /*
1717         * Inactive subsystems have no dentry for their root
1718         * cgroup
1719         */
1720        strcpy(buf, "/");
1721        return 0;
1722    }
1723
1724    start = buf + buflen;
1725
1726    *--start = '\0';
1727    for (;;) {
1728        int len = dentry->d_name.len;
1729
1730        if ((start -= len) < buf)
1731            return -ENAMETOOLONG;
1732        memcpy(start, dentry->d_name.name, len);
1733        cgrp = cgrp->parent;
1734        if (!cgrp)
1735            break;
1736
1737        dentry = rcu_dereference_check(cgrp->dentry,
1738                           rcu_read_lock_held() ||
1739                           cgroup_lock_is_held());
1740        if (!cgrp->parent)
1741            continue;
1742        if (--start < buf)
1743            return -ENAMETOOLONG;
1744        *start = '/';
1745    }
1746    memmove(buf, start, buf + buflen - start);
1747    return 0;
1748}
1749EXPORT_SYMBOL_GPL(cgroup_path);
1750
1751/**
1752 * cgroup_attach_task - attach task 'tsk' to cgroup 'cgrp'
1753 * @cgrp: the cgroup the task is attaching to
1754 * @tsk: the task to be attached
1755 *
1756 * Call holding cgroup_mutex. May take task_lock of
1757 * the task 'tsk' during call.
1758 */
1759int cgroup_attach_task(struct cgroup *cgrp, struct task_struct *tsk)
1760{
1761    int retval = 0;
1762    struct cgroup_subsys *ss, *failed_ss = NULL;
1763    struct cgroup *oldcgrp;
1764    struct css_set *cg;
1765    struct css_set *newcg;
1766    struct cgroupfs_root *root = cgrp->root;
1767
1768    /* Nothing to do if the task is already in that cgroup */
1769    oldcgrp = task_cgroup_from_root(tsk, root);
1770    if (cgrp == oldcgrp)
1771        return 0;
1772
1773    for_each_subsys(root, ss) {
1774        if (ss->can_attach) {
1775            retval = ss->can_attach(ss, cgrp, tsk, false);
1776            if (retval) {
1777                /*
1778                 * Remember on which subsystem the can_attach()
1779                 * failed, so that we only call cancel_attach()
1780                 * against the subsystems whose can_attach()
1781                 * succeeded. (See below)
1782                 */
1783                failed_ss = ss;
1784                goto out;
1785            }
1786        }
1787    }
1788
1789    task_lock(tsk);
1790    cg = tsk->cgroups;
1791    get_css_set(cg);
1792    task_unlock(tsk);
1793    /*
1794     * Locate or allocate a new css_set for this task,
1795     * based on its final set of cgroups
1796     */
1797    newcg = find_css_set(cg, cgrp);
1798    put_css_set(cg);
1799    if (!newcg) {
1800        retval = -ENOMEM;
1801        goto out;
1802    }
1803
1804    task_lock(tsk);
1805    if (tsk->flags & PF_EXITING) {
1806        task_unlock(tsk);
1807        put_css_set(newcg);
1808        retval = -ESRCH;
1809        goto out;
1810    }
1811    rcu_assign_pointer(tsk->cgroups, newcg);
1812    task_unlock(tsk);
1813
1814    /* Update the css_set linked lists if we're using them */
1815    write_lock(&css_set_lock);
1816    if (!list_empty(&tsk->cg_list)) {
1817        list_del(&tsk->cg_list);
1818        list_add(&tsk->cg_list, &newcg->tasks);
1819    }
1820    write_unlock(&css_set_lock);
1821
1822    for_each_subsys(root, ss) {
1823        if (ss->attach)
1824            ss->attach(ss, cgrp, oldcgrp, tsk, false);
1825    }
1826    set_bit(CGRP_RELEASABLE, &oldcgrp->flags);
1827    synchronize_rcu();
1828    put_css_set(cg);
1829
1830    /*
1831     * wake up rmdir() waiter. the rmdir should fail since the cgroup
1832     * is no longer empty.
1833     */
1834    cgroup_wakeup_rmdir_waiter(cgrp);
1835out:
1836    if (retval) {
1837        for_each_subsys(root, ss) {
1838            if (ss == failed_ss)
1839                /*
1840                 * This subsystem was the one that failed the
1841                 * can_attach() check earlier, so we don't need
1842                 * to call cancel_attach() against it or any
1843                 * remaining subsystems.
1844                 */
1845                break;
1846            if (ss->cancel_attach)
1847                ss->cancel_attach(ss, cgrp, tsk, false);
1848        }
1849    }
1850    return retval;
1851}
1852
1853/**
1854 * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from'
1855 * @from: attach to all cgroups of a given task
1856 * @tsk: the task to be attached
1857 */
1858int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk)
1859{
1860    struct cgroupfs_root *root;
1861    int retval = 0;
1862
1863    cgroup_lock();
1864    for_each_active_root(root) {
1865        struct cgroup *from_cg = task_cgroup_from_root(from, root);
1866
1867        retval = cgroup_attach_task(from_cg, tsk);
1868        if (retval)
1869            break;
1870    }
1871    cgroup_unlock();
1872
1873    return retval;
1874}
1875EXPORT_SYMBOL_GPL(cgroup_attach_task_all);
1876
1877/*
1878 * Attach task with pid 'pid' to cgroup 'cgrp'. Call with cgroup_mutex
1879 * held. May take task_lock of task
1880 */
1881static int attach_task_by_pid(struct cgroup *cgrp, u64 pid)
1882{
1883    struct task_struct *tsk;
1884    const struct cred *cred = current_cred(), *tcred;
1885    int ret;
1886
1887    if (pid) {
1888        rcu_read_lock();
1889        tsk = find_task_by_vpid(pid);
1890        if (!tsk || tsk->flags & PF_EXITING) {
1891            rcu_read_unlock();
1892            return -ESRCH;
1893        }
1894
1895        tcred = __task_cred(tsk);
1896        if (cred->euid &&
1897            cred->euid != tcred->uid &&
1898            cred->euid != tcred->suid) {
1899            rcu_read_unlock();
1900            return -EACCES;
1901        }
1902        get_task_struct(tsk);
1903        rcu_read_unlock();
1904    } else {
1905        tsk = current;
1906        get_task_struct(tsk);
1907    }
1908
1909    ret = cgroup_attach_task(cgrp, tsk);
1910    put_task_struct(tsk);
1911    return ret;
1912}
1913
1914static int cgroup_tasks_write(struct cgroup *cgrp, struct cftype *cft, u64 pid)
1915{
1916    int ret;
1917    if (!cgroup_lock_live_group(cgrp))
1918        return -ENODEV;
1919    ret = attach_task_by_pid(cgrp, pid);
1920    cgroup_unlock();
1921    return ret;
1922}
1923
1924/**
1925 * cgroup_lock_live_group - take cgroup_mutex and check that cgrp is alive.
1926 * @cgrp: the cgroup to be checked for liveness
1927 *
1928 * On success, returns true; the lock should be later released with
1929 * cgroup_unlock(). On failure returns false with no lock held.
1930 */
1931bool cgroup_lock_live_group(struct cgroup *cgrp)
1932{
1933    mutex_lock(&cgroup_mutex);
1934    if (cgroup_is_removed(cgrp)) {
1935        mutex_unlock(&cgroup_mutex);
1936        return false;
1937    }
1938    return true;
1939}
1940EXPORT_SYMBOL_GPL(cgroup_lock_live_group);
1941
1942static int cgroup_release_agent_write(struct cgroup *cgrp, struct cftype *cft,
1943                      const char *buffer)
1944{
1945    BUILD_BUG_ON(sizeof(cgrp->root->release_agent_path) < PATH_MAX);
1946    if (strlen(buffer) >= PATH_MAX)
1947        return -EINVAL;
1948    if (!cgroup_lock_live_group(cgrp))
1949        return -ENODEV;
1950    strcpy(cgrp->root->release_agent_path, buffer);
1951    cgroup_unlock();
1952    return 0;
1953}
1954
1955static int cgroup_release_agent_show(struct cgroup *cgrp, struct cftype *cft,
1956                     struct seq_file *seq)
1957{
1958    if (!cgroup_lock_live_group(cgrp))
1959        return -ENODEV;
1960    seq_puts(seq, cgrp->root->release_agent_path);
1961    seq_putc(seq, '\n');
1962    cgroup_unlock();
1963    return 0;
1964}
1965
1966/* A buffer size big enough for numbers or short strings */
1967#define CGROUP_LOCAL_BUFFER_SIZE 64
1968
1969static ssize_t cgroup_write_X64(struct cgroup *cgrp, struct cftype *cft,
1970                struct file *file,
1971                const char __user *userbuf,
1972                size_t nbytes, loff_t *unused_ppos)
1973{
1974    char buffer[CGROUP_LOCAL_BUFFER_SIZE];
1975    int retval = 0;
1976    char *end;
1977
1978    if (!nbytes)
1979        return -EINVAL;
1980    if (nbytes >= sizeof(buffer))
1981        return -E2BIG;
1982    if (copy_from_user(buffer, userbuf, nbytes))
1983        return -EFAULT;
1984
1985    buffer[nbytes] = 0; /* nul-terminate */
1986    if (cft->write_u64) {
1987        u64 val = simple_strtoull(strstrip(buffer), &end, 0);
1988        if (*end)
1989            return -EINVAL;
1990        retval = cft->write_u64(cgrp, cft, val);
1991    } else {
1992        s64 val = simple_strtoll(strstrip(buffer), &end, 0);
1993        if (*end)
1994            return -EINVAL;
1995        retval = cft->write_s64(cgrp, cft, val);
1996    }
1997    if (!retval)
1998        retval = nbytes;
1999    return retval;
2000}
2001
2002static ssize_t cgroup_write_string(struct cgroup *cgrp, struct cftype *cft,
2003                   struct file *file,
2004                   const char __user *userbuf,
2005                   size_t nbytes, loff_t *unused_ppos)
2006{
2007    char local_buffer[CGROUP_LOCAL_BUFFER_SIZE];
2008    int retval = 0;
2009    size_t max_bytes = cft->max_write_len;
2010    char *buffer = local_buffer;
2011
2012    if (!max_bytes)
2013        max_bytes = sizeof(local_buffer) - 1;
2014    if (nbytes >= max_bytes)
2015        return -E2BIG;
2016    /* Allocate a dynamic buffer if we need one */
2017    if (nbytes >= sizeof(local_buffer)) {
2018        buffer = kmalloc(nbytes + 1, GFP_KERNEL);
2019        if (buffer == NULL)
2020            return -ENOMEM;
2021    }
2022    if (nbytes && copy_from_user(buffer, userbuf, nbytes)) {
2023        retval = -EFAULT;
2024        goto out;
2025    }
2026
2027    buffer[nbytes] = 0; /* nul-terminate */
2028    retval = cft->write_string(cgrp, cft, strstrip(buffer));
2029    if (!retval)
2030        retval = nbytes;
2031out:
2032    if (buffer != local_buffer)
2033        kfree(buffer);
2034    return retval;
2035}
2036
2037static ssize_t cgroup_file_write(struct file *file, const char __user *buf,
2038                        size_t nbytes, loff_t *ppos)
2039{
2040    struct cftype *cft = __d_cft(file->f_dentry);
2041    struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
2042
2043    if (cgroup_is_removed(cgrp))
2044        return -ENODEV;
2045    if (cft->write)
2046        return cft->write(cgrp, cft, file, buf, nbytes, ppos);
2047    if (cft->write_u64 || cft->write_s64)
2048        return cgroup_write_X64(cgrp, cft, file, buf, nbytes, ppos);
2049    if (cft->write_string)
2050        return cgroup_write_string(cgrp, cft, file, buf, nbytes, ppos);
2051    if (cft->trigger) {
2052        int ret = cft->trigger(cgrp, (unsigned int)cft->private);
2053        return ret ? ret : nbytes;
2054    }
2055    return -EINVAL;
2056}
2057
2058static ssize_t cgroup_read_u64(struct cgroup *cgrp, struct cftype *cft,
2059                   struct file *file,
2060                   char __user *buf, size_t nbytes,
2061                   loff_t *ppos)
2062{
2063    char tmp[CGROUP_LOCAL_BUFFER_SIZE];
2064    u64 val = cft->read_u64(cgrp, cft);
2065    int len = sprintf(tmp, "%llu\n", (unsigned long long) val);
2066
2067    return simple_read_from_buffer(buf, nbytes, ppos, tmp, len);
2068}
2069
2070static ssize_t cgroup_read_s64(struct cgroup *cgrp, struct cftype *cft,
2071                   struct file *file,
2072                   char __user *buf, size_t nbytes,
2073                   loff_t *ppos)
2074{
2075    char tmp[CGROUP_LOCAL_BUFFER_SIZE];
2076    s64 val = cft->read_s64(cgrp, cft);
2077    int len = sprintf(tmp, "%lld\n", (long long) val);
2078
2079    return simple_read_from_buffer(buf, nbytes, ppos, tmp, len);
2080}
2081
2082static ssize_t cgroup_file_read(struct file *file, char __user *buf,
2083                   size_t nbytes, loff_t *ppos)
2084{
2085    struct cftype *cft = __d_cft(file->f_dentry);
2086    struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
2087
2088    if (cgroup_is_removed(cgrp))
2089        return -ENODEV;
2090
2091    if (cft->read)
2092        return cft->read(cgrp, cft, file, buf, nbytes, ppos);
2093    if (cft->read_u64)
2094        return cgroup_read_u64(cgrp, cft, file, buf, nbytes, ppos);
2095    if (cft->read_s64)
2096        return cgroup_read_s64(cgrp, cft, file, buf, nbytes, ppos);
2097    return -EINVAL;
2098}
2099
2100/*
2101 * seqfile ops/methods for returning structured data. Currently just
2102 * supports string->u64 maps, but can be extended in future.
2103 */
2104
2105struct cgroup_seqfile_state {
2106    struct cftype *cft;
2107    struct cgroup *cgroup;
2108};
2109
2110static int cgroup_map_add(struct cgroup_map_cb *cb, const char *key, u64 value)
2111{
2112    struct seq_file *sf = cb->state;
2113    return seq_printf(sf, "%s %llu\n", key, (unsigned long long)value);
2114}
2115
2116static int cgroup_seqfile_show(struct seq_file *m, void *arg)
2117{
2118    struct cgroup_seqfile_state *state = m->private;
2119    struct cftype *cft = state->cft;
2120    if (cft->read_map) {
2121        struct cgroup_map_cb cb = {
2122            .fill = cgroup_map_add,
2123            .state = m,
2124        };
2125        return cft->read_map(state->cgroup, cft, &cb);
2126    }
2127    return cft->read_seq_string(state->cgroup, cft, m);
2128}
2129
2130static int cgroup_seqfile_release(struct inode *inode, struct file *file)
2131{
2132    struct seq_file *seq = file->private_data;
2133    kfree(seq->private);
2134    return single_release(inode, file);
2135}
2136
2137static const struct file_operations cgroup_seqfile_operations = {
2138    .read = seq_read,
2139    .write = cgroup_file_write,
2140    .llseek = seq_lseek,
2141    .release = cgroup_seqfile_release,
2142};
2143
2144static int cgroup_file_open(struct inode *inode, struct file *file)
2145{
2146    int err;
2147    struct cftype *cft;
2148
2149    err = generic_file_open(inode, file);
2150    if (err)
2151        return err;
2152    cft = __d_cft(file->f_dentry);
2153
2154    if (cft->read_map || cft->read_seq_string) {
2155        struct cgroup_seqfile_state *state =
2156            kzalloc(sizeof(*state), GFP_USER);
2157        if (!state)
2158            return -ENOMEM;
2159        state->cft = cft;
2160        state->cgroup = __d_cgrp(file->f_dentry->d_parent);
2161        file->f_op = &cgroup_seqfile_operations;
2162        err = single_open(file, cgroup_seqfile_show, state);
2163        if (err < 0)
2164            kfree(state);
2165    } else if (cft->open)
2166        err = cft->open(inode, file);
2167    else
2168        err = 0;
2169
2170    return err;
2171}
2172
2173static int cgroup_file_release(struct inode *inode, struct file *file)
2174{
2175    struct cftype *cft = __d_cft(file->f_dentry);
2176    if (cft->release)
2177        return cft->release(inode, file);
2178    return 0;
2179}
2180
2181/*
2182 * cgroup_rename - Only allow simple rename of directories in place.
2183 */
2184static int cgroup_rename(struct inode *old_dir, struct dentry *old_dentry,
2185                struct inode *new_dir, struct dentry *new_dentry)
2186{
2187    if (!S_ISDIR(old_dentry->d_inode->i_mode))
2188        return -ENOTDIR;
2189    if (new_dentry->d_inode)
2190        return -EEXIST;
2191    if (old_dir != new_dir)
2192        return -EIO;
2193    return simple_rename(old_dir, old_dentry, new_dir, new_dentry);
2194}
2195
2196static const struct file_operations cgroup_file_operations = {
2197    .read = cgroup_file_read,
2198    .write = cgroup_file_write,
2199    .llseek = generic_file_llseek,
2200    .open = cgroup_file_open,
2201    .release = cgroup_file_release,
2202};
2203
2204static const struct inode_operations cgroup_dir_inode_operations = {
2205    .lookup = cgroup_lookup,
2206    .mkdir = cgroup_mkdir,
2207    .rmdir = cgroup_rmdir,
2208    .rename = cgroup_rename,
2209};
2210
2211static struct dentry *cgroup_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
2212{
2213    if (dentry->d_name.len > NAME_MAX)
2214        return ERR_PTR(-ENAMETOOLONG);
2215    d_add(dentry, NULL);
2216    return NULL;
2217}
2218
2219/*
2220 * Check if a file is a control file
2221 */
2222static inline struct cftype *__file_cft(struct file *file)
2223{
2224    if (file->f_dentry->d_inode->i_fop != &cgroup_file_operations)
2225        return ERR_PTR(-EINVAL);
2226    return __d_cft(file->f_dentry);
2227}
2228
2229static int cgroup_create_file(struct dentry *dentry, mode_t mode,
2230                struct super_block *sb)
2231{
2232    struct inode *inode;
2233
2234    if (!dentry)
2235        return -ENOENT;
2236    if (dentry->d_inode)
2237        return -EEXIST;
2238
2239    inode = cgroup_new_inode(mode, sb);
2240    if (!inode)
2241        return -ENOMEM;
2242
2243    if (S_ISDIR(mode)) {
2244        inode->i_op = &cgroup_dir_inode_operations;
2245        inode->i_fop = &simple_dir_operations;
2246
2247        /* start off with i_nlink == 2 (for "." entry) */
2248        inc_nlink(inode);
2249
2250        /* start with the directory inode held, so that we can
2251         * populate it without racing with another mkdir */
2252        mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
2253    } else if (S_ISREG(mode)) {
2254        inode->i_size = 0;
2255        inode->i_fop = &cgroup_file_operations;
2256    }
2257    d_instantiate(dentry, inode);
2258    dget(dentry); /* Extra count - pin the dentry in core */
2259    return 0;
2260}
2261
2262/*
2263 * cgroup_create_dir - create a directory for an object.
2264 * @cgrp: the cgroup we create the directory for. It must have a valid
2265 * ->parent field. And we are going to fill its ->dentry field.
2266 * @dentry: dentry of the new cgroup
2267 * @mode: mode to set on new directory.
2268 */
2269static int cgroup_create_dir(struct cgroup *cgrp, struct dentry *dentry,
2270                mode_t mode)
2271{
2272    struct dentry *parent;
2273    int error = 0;
2274
2275    parent = cgrp->parent->dentry;
2276    error = cgroup_create_file(dentry, S_IFDIR | mode, cgrp->root->sb);
2277    if (!error) {
2278        dentry->d_fsdata = cgrp;
2279        inc_nlink(parent->d_inode);
2280        rcu_assign_pointer(cgrp->dentry, dentry);
2281        dget(dentry);
2282    }
2283    dput(dentry);
2284
2285    return error;
2286}
2287
2288/**
2289 * cgroup_file_mode - deduce file mode of a control file
2290 * @cft: the control file in question
2291 *
2292 * returns cft->mode if ->mode is not 0
2293 * returns S_IRUGO|S_IWUSR if it has both a read and a write handler
2294 * returns S_IRUGO if it has only a read handler
2295 * returns S_IWUSR if it has only a write hander
2296 */
2297static mode_t cgroup_file_mode(const struct cftype *cft)
2298{
2299    mode_t mode = 0;
2300
2301    if (cft->mode)
2302        return cft->mode;
2303
2304    if (cft->read || cft->read_u64 || cft->read_s64 ||
2305        cft->read_map || cft->read_seq_string)
2306        mode |= S_IRUGO;
2307
2308    if (cft->write || cft->write_u64 || cft->write_s64 ||
2309        cft->write_string || cft->trigger)
2310        mode |= S_IWUSR;
2311
2312    return mode;
2313}
2314
2315int cgroup_add_file(struct cgroup *cgrp,
2316               struct cgroup_subsys *subsys,
2317               const struct cftype *cft)
2318{
2319    struct dentry *dir = cgrp->dentry;
2320    struct dentry *dentry;
2321    int error;
2322    mode_t mode;
2323
2324    char name[MAX_CGROUP_TYPE_NAMELEN + MAX_CFTYPE_NAME + 2] = { 0 };
2325    if (subsys && !test_bit(ROOT_NOPREFIX, &cgrp->root->flags)) {
2326        strcpy(name, subsys->name);
2327        strcat(name, ".");
2328    }
2329    strcat(name, cft->name);
2330    BUG_ON(!mutex_is_locked(&dir->d_inode->i_mutex));
2331    dentry = lookup_one_len(name, dir, strlen(name));
2332    if (!IS_ERR(dentry)) {
2333        mode = cgroup_file_mode(cft);
2334        error = cgroup_create_file(dentry, mode | S_IFREG,
2335                        cgrp->root->sb);
2336        if (!error)
2337            dentry->d_fsdata = (void *)cft;
2338        dput(dentry);
2339    } else
2340        error = PTR_ERR(dentry);
2341    return error;
2342}
2343EXPORT_SYMBOL_GPL(cgroup_add_file);
2344
2345int cgroup_add_files(struct cgroup *cgrp,
2346            struct cgroup_subsys *subsys,
2347            const struct cftype cft[],
2348            int count)
2349{
2350    int i, err;
2351    for (i = 0; i < count; i++) {
2352        err = cgroup_add_file(cgrp, subsys, &cft[i]);
2353        if (err)
2354            return err;
2355    }
2356    return 0;
2357}
2358EXPORT_SYMBOL_GPL(cgroup_add_files);
2359
2360/**
2361 * cgroup_task_count - count the number of tasks in a cgroup.
2362 * @cgrp: the cgroup in question
2363 *
2364 * Return the number of tasks in the cgroup.
2365 */
2366int cgroup_task_count(const struct cgroup *cgrp)
2367{
2368    int count = 0;
2369    struct cg_cgroup_link *link;
2370
2371    read_lock(&css_set_lock);
2372    list_for_each_entry(link, &cgrp->css_sets, cgrp_link_list) {
2373        count += atomic_read(&link->cg->refcount);
2374    }
2375    read_unlock(&css_set_lock);
2376    return count;
2377}
2378
2379/*
2380 * Advance a list_head iterator. The iterator should be positioned at
2381 * the start of a css_set
2382 */
2383static void cgroup_advance_iter(struct cgroup *cgrp,
2384                struct cgroup_iter *it)
2385{
2386    struct list_head *l = it->cg_link;
2387    struct cg_cgroup_link *link;
2388    struct css_set *cg;
2389
2390    /* Advance to the next non-empty css_set */
2391    do {
2392        l = l->next;
2393        if (l == &cgrp->css_sets) {
2394            it->cg_link = NULL;
2395            return;
2396        }
2397        link = list_entry(l, struct cg_cgroup_link, cgrp_link_list);
2398        cg = link->cg;
2399    } while (list_empty(&cg->tasks));
2400    it->cg_link = l;
2401    it->task = cg->tasks.next;
2402}
2403
2404/*
2405 * To reduce the fork() overhead for systems that are not actually
2406 * using their cgroups capability, we don't maintain the lists running
2407 * through each css_set to its tasks until we see the list actually
2408 * used - in other words after the first call to cgroup_iter_start().
2409 *
2410 * The tasklist_lock is not held here, as do_each_thread() and
2411 * while_each_thread() are protected by RCU.
2412 */
2413static void cgroup_enable_task_cg_lists(void)
2414{
2415    struct task_struct *p, *g;
2416    write_lock(&css_set_lock);
2417    use_task_css_set_links = 1;
2418    do_each_thread(g, p) {
2419        task_lock(p);
2420        /*
2421         * We should check if the process is exiting, otherwise
2422         * it will race with cgroup_exit() in that the list
2423         * entry won't be deleted though the process has exited.
2424         */
2425        if (!(p->flags & PF_EXITING) && list_empty(&p->cg_list))
2426            list_add(&p->cg_list, &p->cgroups->tasks);
2427        task_unlock(p);
2428    } while_each_thread(g, p);
2429    write_unlock(&css_set_lock);
2430}
2431
2432void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it)
2433{
2434    /*
2435     * The first time anyone tries to iterate across a cgroup,
2436     * we need to enable the list linking each css_set to its
2437     * tasks, and fix up all existing tasks.
2438     */
2439    if (!use_task_css_set_links)
2440        cgroup_enable_task_cg_lists();
2441
2442    read_lock(&css_set_lock);
2443    it->cg_link = &cgrp->css_sets;
2444    cgroup_advance_iter(cgrp, it);
2445}
2446
2447struct task_struct *cgroup_iter_next(struct cgroup *cgrp,
2448                    struct cgroup_iter *it)
2449{
2450    struct task_struct *res;
2451    struct list_head *l = it->task;
2452    struct cg_cgroup_link *link;
2453
2454    /* If the iterator cg is NULL, we have no tasks */
2455    if (!it->cg_link)
2456        return NULL;
2457    res = list_entry(l, struct task_struct, cg_list);
2458    /* Advance iterator to find next entry */
2459    l = l->next;
2460    link = list_entry(it->cg_link, struct cg_cgroup_link, cgrp_link_list);
2461    if (l == &link->cg->tasks) {
2462        /* We reached the end of this task list - move on to
2463         * the next cg_cgroup_link */
2464        cgroup_advance_iter(cgrp, it);
2465    } else {
2466        it->task = l;
2467    }
2468    return res;
2469}
2470
2471void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it)
2472{
2473    read_unlock(&css_set_lock);
2474}
2475
2476static inline int started_after_time(struct task_struct *t1,
2477                     struct timespec *time,
2478                     struct task_struct *t2)
2479{
2480    int start_diff = timespec_compare(&t1->start_time, time);
2481    if (start_diff > 0) {
2482        return 1;
2483    } else if (start_diff < 0) {
2484        return 0;
2485    } else {
2486        /*
2487         * Arbitrarily, if two processes started at the same
2488         * time, we'll say that the lower pointer value
2489         * started first. Note that t2 may have exited by now
2490         * so this may not be a valid pointer any longer, but
2491         * that's fine - it still serves to distinguish
2492         * between two tasks started (effectively) simultaneously.
2493         */
2494        return t1 > t2;
2495    }
2496}
2497
2498/*
2499 * This function is a callback from heap_insert() and is used to order
2500 * the heap.
2501 * In this case we order the heap in descending task start time.
2502 */
2503static inline int started_after(void *p1, void *p2)
2504{
2505    struct task_struct *t1 = p1;
2506    struct task_struct *t2 = p2;
2507    return started_after_time(t1, &t2->start_time, t2);
2508}
2509
2510/**
2511 * cgroup_scan_tasks - iterate though all the tasks in a cgroup
2512 * @scan: struct cgroup_scanner containing arguments for the scan
2513 *
2514 * Arguments include pointers to callback functions test_task() and
2515 * process_task().
2516 * Iterate through all the tasks in a cgroup, calling test_task() for each,
2517 * and if it returns true, call process_task() for it also.
2518 * The test_task pointer may be NULL, meaning always true (select all tasks).
2519 * Effectively duplicates cgroup_iter_{start,next,end}()
2520 * but does not lock css_set_lock for the call to process_task().
2521 * The struct cgroup_scanner may be embedded in any structure of the caller's
2522 * creation.
2523 * It is guaranteed that process_task() will act on every task that
2524 * is a member of the cgroup for the duration of this call. This
2525 * function may or may not call process_task() for tasks that exit
2526 * or move to a different cgroup during the call, or are forked or
2527 * move into the cgroup during the call.
2528 *
2529 * Note that test_task() may be called with locks held, and may in some
2530 * situations be called multiple times for the same task, so it should
2531 * be cheap.
2532 * If the heap pointer in the struct cgroup_scanner is non-NULL, a heap has been
2533 * pre-allocated and will be used for heap operations (and its "gt" member will
2534 * be overwritten), else a temporary heap will be used (allocation of which
2535 * may cause this function to fail).
2536 */
2537int cgroup_scan_tasks(struct cgroup_scanner *scan)
2538{
2539    int retval, i;
2540    struct cgroup_iter it;
2541    struct task_struct *p, *dropped;
2542    /* Never dereference latest_task, since it's not refcounted */
2543    struct task_struct *latest_task = NULL;
2544    struct ptr_heap tmp_heap;
2545    struct ptr_heap *heap;
2546    struct timespec latest_time = { 0, 0 };
2547
2548    if (scan->heap) {
2549        /* The caller supplied our heap and pre-allocated its memory */
2550        heap = scan->heap;
2551        heap->gt = &started_after;
2552    } else {
2553        /* We need to allocate our own heap memory */
2554        heap = &tmp_heap;
2555        retval = heap_init(heap, PAGE_SIZE, GFP_KERNEL, &started_after);
2556        if (retval)
2557            /* cannot allocate the heap */
2558            return retval;
2559    }
2560
2561 again:
2562    /*
2563     * Scan tasks in the cgroup, using the scanner's "test_task" callback
2564     * to determine which are of interest, and using the scanner's
2565     * "process_task" callback to process any of them that need an update.
2566     * Since we don't want to hold any locks during the task updates,
2567     * gather tasks to be processed in a heap structure.
2568     * The heap is sorted by descending task start time.
2569     * If the statically-sized heap fills up, we overflow tasks that
2570     * started later, and in future iterations only consider tasks that
2571     * started after the latest task in the previous pass. This
2572     * guarantees forward progress and that we don't miss any tasks.
2573     */
2574    heap->size = 0;
2575    cgroup_iter_start(scan->cg, &it);
2576    while ((p = cgroup_iter_next(scan->cg, &it))) {
2577        /*
2578         * Only affect tasks that qualify per the caller's callback,
2579         * if he provided one
2580         */
2581        if (scan->test_task && !scan->test_task(p, scan))
2582            continue;
2583        /*
2584         * Only process tasks that started after the last task
2585         * we processed
2586         */
2587        if (!started_after_time(p, &latest_time, latest_task))
2588            continue;
2589        dropped = heap_insert(heap, p);
2590        if (dropped == NULL) {
2591            /*
2592             * The new task was inserted; the heap wasn't
2593             * previously full
2594             */
2595            get_task_struct(p);
2596        } else if (dropped != p) {
2597            /*
2598             * The new task was inserted, and pushed out a
2599             * different task
2600             */
2601            get_task_struct(p);
2602            put_task_struct(dropped);
2603        }
2604        /*
2605         * Else the new task was newer than anything already in
2606         * the heap and wasn't inserted
2607         */
2608    }
2609    cgroup_iter_end(scan->cg, &it);
2610
2611    if (heap->size) {
2612        for (i = 0; i < heap->size; i++) {
2613            struct task_struct *q = heap->ptrs[i];
2614            if (i == 0) {
2615                latest_time = q->start_time;
2616                latest_task = q;
2617            }
2618            /* Process the task per the caller's callback */
2619            scan->process_task(q, scan);
2620            put_task_struct(q);
2621        }
2622        /*
2623         * If we had to process any tasks at all, scan again
2624         * in case some of them were in the middle of forking
2625         * children that didn't get processed.
2626         * Not the most efficient way to do it, but it avoids
2627         * having to take callback_mutex in the fork path
2628         */
2629        goto again;
2630    }
2631    if (heap == &tmp_heap)
2632        heap_free(&tmp_heap);
2633    return 0;
2634}
2635
2636/*
2637 * Stuff for reading the 'tasks'/'procs' files.
2638 *
2639 * Reading this file can return large amounts of data if a cgroup has
2640 * *lots* of attached tasks. So it may need several calls to read(),
2641 * but we cannot guarantee that the information we produce is correct
2642 * unless we produce it entirely atomically.
2643 *
2644 */
2645
2646/*
2647 * The following two functions "fix" the issue where there are more pids
2648 * than kmalloc will give memory for; in such cases, we use vmalloc/vfree.
2649 * TODO: replace with a kernel-wide solution to this problem
2650 */
2651#define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2))
2652static void *pidlist_allocate(int count)
2653{
2654    if (PIDLIST_TOO_LARGE(count))
2655        return vmalloc(count * sizeof(pid_t));
2656    else
2657        return kmalloc(count * sizeof(pid_t), GFP_KERNEL);
2658}
2659static void pidlist_free(void *p)
2660{
2661    if (is_vmalloc_addr(p))
2662        vfree(p);
2663    else
2664        kfree(p);
2665}
2666static void *pidlist_resize(void *p, int newcount)
2667{
2668    void *newlist;
2669    /* note: if new alloc fails, old p will still be valid either way */
2670    if (is_vmalloc_addr(p)) {
2671        newlist = vmalloc(newcount * sizeof(pid_t));
2672        if (!newlist)
2673            return NULL;
2674        memcpy(newlist, p, newcount * sizeof(pid_t));
2675        vfree(p);
2676    } else {
2677        newlist = krealloc(p, newcount * sizeof(pid_t), GFP_KERNEL);
2678    }
2679    return newlist;
2680}
2681
2682/*
2683 * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
2684 * If the new stripped list is sufficiently smaller and there's enough memory
2685 * to allocate a new buffer, will let go of the unneeded memory. Returns the
2686 * number of unique elements.
2687 */
2688/* is the size difference enough that we should re-allocate the array? */
2689#define PIDLIST_REALLOC_DIFFERENCE(old, new) ((old) - PAGE_SIZE >= (new))
2690static int pidlist_uniq(pid_t **p, int length)
2691{
2692    int src, dest = 1;
2693    pid_t *list = *p;
2694    pid_t *newlist;
2695
2696    /*
2697     * we presume the 0th element is unique, so i starts at 1. trivial
2698     * edge cases first; no work needs to be done for either
2699     */
2700    if (length == 0 || length == 1)
2701        return length;
2702    /* src and dest walk down the list; dest counts unique elements */
2703    for (src = 1; src < length; src++) {
2704        /* find next unique element */
2705        while (list[src] == list[src-1]) {
2706            src++;
2707            if (src == length)
2708                goto after;
2709        }
2710        /* dest always points to where the next unique element goes */
2711        list[dest] = list[src];
2712        dest++;
2713    }
2714after:
2715    /*
2716     * if the length difference is large enough, we want to allocate a
2717     * smaller buffer to save memory. if this fails due to out of memory,
2718     * we'll just stay with what we've got.
2719     */
2720    if (PIDLIST_REALLOC_DIFFERENCE(length, dest)) {
2721        newlist = pidlist_resize(list, dest);
2722        if (newlist)
2723            *p = newlist;
2724    }
2725    return dest;
2726}
2727
2728static int cmppid(const void *a, const void *b)
2729{
2730    return *(pid_t *)a - *(pid_t *)b;
2731}
2732
2733/*
2734 * find the appropriate pidlist for our purpose (given procs vs tasks)
2735 * returns with the lock on that pidlist already held, and takes care
2736 * of the use count, or returns NULL with no locks held if we're out of
2737 * memory.
2738 */
2739static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp,
2740                          enum cgroup_filetype type)
2741{
2742    struct cgroup_pidlist *l;
2743    /* don't need task_nsproxy() if we're looking at ourself */
2744    struct pid_namespace *ns = current->nsproxy->pid_ns;
2745
2746    /*
2747     * We can't drop the pidlist_mutex before taking the l->mutex in case
2748     * the last ref-holder is trying to remove l from the list at the same
2749     * time. Holding the pidlist_mutex precludes somebody taking whichever
2750     * list we find out from under us - compare release_pid_array().
2751     */
2752    mutex_lock(&cgrp->pidlist_mutex);
2753    list_for_each_entry(l, &cgrp->pidlists, links) {
2754        if (l->key.type == type && l->key.ns == ns) {
2755            /* make sure l doesn't vanish out from under us */
2756            down_write(&l->mutex);
2757            mutex_unlock(&cgrp->pidlist_mutex);
2758            return l;
2759        }
2760    }
2761    /* entry not found; create a new one */
2762    l = kmalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL);
2763    if (!l) {
2764        mutex_unlock(&cgrp->pidlist_mutex);
2765        return l;
2766    }
2767    init_rwsem(&l->mutex);
2768    down_write(&l->mutex);
2769    l->key.type = type;
2770    l->key.ns = get_pid_ns(ns);
2771    l->use_count = 0; /* don't increment here */
2772    l->list = NULL;
2773    l->owner = cgrp;
2774    list_add(&l->links, &cgrp->pidlists);
2775    mutex_unlock(&cgrp->pidlist_mutex);
2776    return l;
2777}
2778
2779/*
2780 * Load a cgroup's pidarray with either procs' tgids or tasks' pids
2781 */
2782static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type,
2783                  struct cgroup_pidlist **lp)
2784{
2785    pid_t *array;
2786    int length;
2787    int pid, n = 0; /* used for populating the array */
2788    struct cgroup_iter it;
2789    struct task_struct *tsk;
2790    struct cgroup_pidlist *l;
2791
2792    /*
2793     * If cgroup gets more users after we read count, we won't have
2794     * enough space - tough. This race is indistinguishable to the
2795     * caller from the case that the additional cgroup users didn't
2796     * show up until sometime later on.
2797     */
2798    length = cgroup_task_count(cgrp);
2799    array = pidlist_allocate(length);
2800    if (!array)
2801        return -ENOMEM;
2802    /* now, populate the array */
2803    cgroup_iter_start(cgrp, &it);
2804    while ((tsk = cgroup_iter_next(cgrp, &it))) {
2805        if (unlikely(n == length))
2806            break;
2807        /* get tgid or pid for procs or tasks file respectively */
2808        if (type == CGROUP_FILE_PROCS)
2809            pid = task_tgid_vnr(tsk);
2810        else
2811            pid = task_pid_vnr(tsk);
2812        if (pid > 0) /* make sure to only use valid results */
2813            array[n++] = pid;
2814    }
2815    cgroup_iter_end(cgrp, &it);
2816    length = n;
2817    /* now sort & (if procs) strip out duplicates */
2818    sort(array, length, sizeof(pid_t), cmppid, NULL);
2819    if (type == CGROUP_FILE_PROCS)
2820        length = pidlist_uniq(&array, length);
2821    l = cgroup_pidlist_find(cgrp, type);
2822    if (!l) {
2823        pidlist_free(array);
2824        return -ENOMEM;
2825    }
2826    /* store array, freeing old if necessary - lock already held */
2827    pidlist_free(l->list);
2828    l->list = array;
2829    l->length = length;
2830    l->use_count++;
2831    up_write(&l->mutex);
2832    *lp = l;
2833    return 0;
2834}
2835
2836/**
2837 * cgroupstats_build - build and fill cgroupstats
2838 * @stats: cgroupstats to fill information into
2839 * @dentry: A dentry entry belonging to the cgroup for which stats have
2840 * been requested.
2841 *
2842 * Build and fill cgroupstats so that taskstats can export it to user
2843 * space.
2844 */
2845int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry)
2846{
2847    int ret = -EINVAL;
2848    struct cgroup *cgrp;
2849    struct cgroup_iter it;
2850    struct task_struct *tsk;
2851
2852    /*
2853     * Validate dentry by checking the superblock operations,
2854     * and make sure it's a directory.
2855     */
2856    if (dentry->d_sb->s_op != &cgroup_ops ||
2857        !S_ISDIR(dentry->d_inode->i_mode))
2858         goto err;
2859
2860    ret = 0;
2861    cgrp = dentry->d_fsdata;
2862
2863    cgroup_iter_start(cgrp, &it);
2864    while ((tsk = cgroup_iter_next(cgrp, &it))) {
2865        switch (tsk->state) {
2866        case TASK_RUNNING:
2867            stats->nr_running++;
2868            break;
2869        case TASK_INTERRUPTIBLE:
2870            stats->nr_sleeping++;
2871            break;
2872        case TASK_UNINTERRUPTIBLE:
2873            stats->nr_uninterruptible++;
2874            break;
2875        case TASK_STOPPED:
2876            stats->nr_stopped++;
2877            break;
2878        default:
2879            if (delayacct_is_task_waiting_on_io(tsk))
2880                stats->nr_io_wait++;
2881            break;
2882        }
2883    }
2884    cgroup_iter_end(cgrp, &it);
2885
2886err:
2887    return ret;
2888}
2889
2890
2891/*
2892 * seq_file methods for the tasks/procs files. The seq_file position is the
2893 * next pid to display; the seq_file iterator is a pointer to the pid
2894 * in the cgroup->l->list array.
2895 */
2896
2897static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos)
2898{
2899    /*
2900     * Initially we receive a position value that corresponds to
2901     * one more than the last pid shown (or 0 on the first call or
2902     * after a seek to the start). Use a binary-search to find the
2903     * next pid to display, if any
2904     */
2905    struct cgroup_pidlist *l = s->private;
2906    int index = 0, pid = *pos;
2907    int *iter;
2908
2909    down_read(&l->mutex);
2910    if (pid) {
2911        int end = l->length;
2912
2913        while (index < end) {
2914            int mid = (index + end) / 2;
2915            if (l->list[mid] == pid) {
2916                index = mid;
2917                break;
2918            } else if (l->list[mid] <= pid)
2919                index = mid + 1;
2920            else
2921                end = mid;
2922        }
2923    }
2924    /* If we're off the end of the array, we're done */
2925    if (index >= l->length)
2926        return NULL;
2927    /* Update the abstract position to be the actual pid that we found */
2928    iter = l->list + index;
2929    *pos = *iter;
2930    return iter;
2931}
2932
2933static void cgroup_pidlist_stop(struct seq_file *s, void *v)
2934{
2935    struct cgroup_pidlist *l = s->private;
2936    up_read(&l->mutex);
2937}
2938
2939static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos)
2940{
2941    struct cgroup_pidlist *l = s->private;
2942    pid_t *p = v;
2943    pid_t *end = l->list + l->length;
2944    /*
2945     * Advance to the next pid in the array. If this goes off the
2946     * end, we're done
2947     */
2948    p++;
2949    if (p >= end) {
2950        return NULL;
2951    } else {
2952        *pos = *p;
2953        return p;
2954    }
2955}
2956
2957static int cgroup_pidlist_show(struct seq_file *s, void *v)
2958{
2959    return seq_printf(s, "%d\n", *(int *)v);
2960}
2961
2962/*
2963 * seq_operations functions for iterating on pidlists through seq_file -
2964 * independent of whether it's tasks or procs
2965 */
2966static const struct seq_operations cgroup_pidlist_seq_operations = {
2967    .start = cgroup_pidlist_start,
2968    .stop = cgroup_pidlist_stop,
2969    .next = cgroup_pidlist_next,
2970    .show = cgroup_pidlist_show,
2971};
2972
2973static void cgroup_release_pid_array(struct cgroup_pidlist *l)
2974{
2975    /*
2976     * the case where we're the last user of this particular pidlist will
2977     * have us remove it from the cgroup's list, which entails taking the
2978     * mutex. since in pidlist_find the pidlist->lock depends on cgroup->
2979     * pidlist_mutex, we have to take pidlist_mutex first.
2980     */
2981    mutex_lock(&l->owner->pidlist_mutex);
2982    down_write(&l->mutex);
2983    BUG_ON(!l->use_count);
2984    if (!--l->use_count) {
2985        /* we're the last user if refcount is 0; remove and free */
2986        list_del(&l->links);
2987        mutex_unlock(&l->owner->pidlist_mutex);
2988        pidlist_free(l->list);
2989        put_pid_ns(l->key.ns);
2990        up_write(&l->mutex);
2991        kfree(l);
2992        return;
2993    }
2994    mutex_unlock(&l->owner->pidlist_mutex);
2995    up_write(&l->mutex);
2996}
2997
2998static int cgroup_pidlist_release(struct inode *inode, struct file *file)
2999{
3000    struct cgroup_pidlist *l;
3001    if (!(file->f_mode & FMODE_READ))
3002        return 0;
3003    /*
3004     * the seq_file will only be initialized if the file was opened for
3005     * reading; hence we check if it's not null only in that case.
3006     */
3007    l = ((struct seq_file *)file->private_data)->private;
3008    cgroup_release_pid_array(l);
3009    return seq_release(inode, file);
3010}
3011
3012static const struct file_operations cgroup_pidlist_operations = {
3013    .read = seq_read,
3014    .llseek = seq_lseek,
3015    .write = cgroup_file_write,
3016    .release = cgroup_pidlist_release,
3017};
3018
3019/*
3020 * The following functions handle opens on a file that displays a pidlist
3021 * (tasks or procs). Prepare an array of the process/thread IDs of whoever's
3022 * in the cgroup.
3023 */
3024/* helper function for the two below it */
3025static int cgroup_pidlist_open(struct file *file, enum cgroup_filetype type)
3026{
3027    struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
3028    struct cgroup_pidlist *l;
3029    int retval;
3030
3031    /* Nothing to do for write-only files */
3032    if (!(file->f_mode & FMODE_READ))
3033        return 0;
3034
3035    /* have the array populated */
3036    retval = pidlist_array_load(cgrp, type, &l);
3037    if (retval)
3038        return retval;
3039    /* configure file information */
3040    file->f_op = &cgroup_pidlist_operations;
3041
3042    retval = seq_open(file, &cgroup_pidlist_seq_operations);
3043    if (retval) {
3044        cgroup_release_pid_array(l);
3045        return retval;
3046    }
3047    ((struct seq_file *)file->private_data)->private = l;
3048    return 0;
3049}
3050static int cgroup_tasks_open(struct inode *unused, struct file *file)
3051{
3052    return cgroup_pidlist_open(file, CGROUP_FILE_TASKS);
3053}
3054static int cgroup_procs_open(struct inode *unused, struct file *file)
3055{
3056    return cgroup_pidlist_open(file, CGROUP_FILE_PROCS);
3057}
3058
3059static u64 cgroup_read_notify_on_release(struct cgroup *cgrp,
3060                        struct cftype *cft)
3061{
3062    return notify_on_release(cgrp);
3063}
3064
3065static int cgroup_write_notify_on_release(struct cgroup *cgrp,
3066                      struct cftype *cft,
3067                      u64 val)
3068{
3069    clear_bit(CGRP_RELEASABLE, &cgrp->flags);
3070    if (val)
3071        set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
3072    else
3073        clear_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
3074    return 0;
3075}
3076
3077/*
3078 * Unregister event and free resources.
3079 *
3080 * Gets called from workqueue.
3081 */
3082static void cgroup_event_remove(struct work_struct *work)
3083{
3084    struct cgroup_event *event = container_of(work, struct cgroup_event,
3085            remove);
3086    struct cgroup *cgrp = event->cgrp;
3087
3088    event->cft->unregister_event(cgrp, event->cft, event->eventfd);
3089
3090    eventfd_ctx_put(event->eventfd);
3091    kfree(event);
3092    dput(cgrp->dentry);
3093}
3094
3095/*
3096 * Gets called on POLLHUP on eventfd when user closes it.
3097 *
3098 * Called with wqh->lock held and interrupts disabled.
3099 */
3100static int cgroup_event_wake(wait_queue_t *wait, unsigned mode,
3101        int sync, void *key)
3102{
3103    struct cgroup_event *event = container_of(wait,
3104            struct cgroup_event, wait);
3105    struct cgroup *cgrp = event->cgrp;
3106    unsigned long flags = (unsigned long)key;
3107
3108    if (flags & POLLHUP) {
3109        __remove_wait_queue(event->wqh, &event->wait);
3110        spin_lock(&cgrp->event_list_lock);
3111        list_del(&event->list);
3112        spin_unlock(&cgrp->event_list_lock);
3113        /*
3114         * We are in atomic context, but cgroup_event_remove() may
3115         * sleep, so we have to call it in workqueue.
3116         */
3117        schedule_work(&event->remove);
3118    }
3119
3120    return 0;
3121}
3122
3123static void cgroup_event_ptable_queue_proc(struct file *file,
3124        wait_queue_head_t *wqh, poll_table *pt)
3125{
3126    struct cgroup_event *event = container_of(pt,
3127            struct cgroup_event, pt);
3128
3129    event->wqh = wqh;
3130    add_wait_queue(wqh, &event->wait);
3131}
3132
3133/*
3134 * Parse input and register new cgroup event handler.
3135 *
3136 * Input must be in format '<event_fd> <control_fd> <args>'.
3137 * Interpretation of args is defined by control file implementation.
3138 */
3139static int cgroup_write_event_control(struct cgroup *cgrp, struct cftype *cft,
3140                      const char *buffer)
3141{
3142    struct cgroup_event *event = NULL;
3143    unsigned int efd, cfd;
3144    struct file *efile = NULL;
3145    struct file *cfile = NULL;
3146    char *endp;
3147    int ret;
3148
3149    efd = simple_strtoul(buffer, &endp, 10);
3150    if (*endp != ' ')
3151        return -EINVAL;
3152    buffer = endp + 1;
3153
3154    cfd = simple_strtoul(buffer, &endp, 10);
3155    if ((*endp != ' ') && (*endp != '\0'))
3156        return -EINVAL;
3157    buffer = endp + 1;
3158
3159    event = kzalloc(sizeof(*event), GFP_KERNEL);
3160    if (!event)
3161        return -ENOMEM;
3162    event->cgrp = cgrp;
3163    INIT_LIST_HEAD(&event->list);
3164    init_poll_funcptr(&event->pt, cgroup_event_ptable_queue_proc);
3165    init_waitqueue_func_entry(&event->wait, cgroup_event_wake);
3166    INIT_WORK(&event->remove, cgroup_event_remove);
3167
3168    efile = eventfd_fget(efd);
3169    if (IS_ERR(efile)) {
3170        ret = PTR_ERR(efile);
3171        goto fail;
3172    }
3173
3174    event->eventfd = eventfd_ctx_fileget(efile);
3175    if (IS_ERR(event->eventfd)) {
3176        ret = PTR_ERR(event->eventfd);
3177        goto fail;
3178    }
3179
3180    cfile = fget(cfd);
3181    if (!cfile) {
3182        ret = -EBADF;
3183        goto fail;
3184    }
3185
3186    /* the process need read permission on control file */
3187    ret = file_permission(cfile, MAY_READ);
3188    if (ret < 0)
3189        goto fail;
3190
3191    event->cft = __file_cft(cfile);
3192    if (IS_ERR(event->cft)) {
3193        ret = PTR_ERR(event->cft);
3194        goto fail;
3195    }
3196
3197    if (!event->cft->register_event || !event->cft->unregister_event) {
3198        ret = -EINVAL;
3199        goto fail;
3200    }
3201
3202    ret = event->cft->register_event(cgrp, event->cft,
3203            event->eventfd, buffer);
3204    if (ret)
3205        goto fail;
3206
3207    if (efile->f_op->poll(efile, &event->pt) & POLLHUP) {
3208        event->cft->unregister_event(cgrp, event->cft, event->eventfd);
3209        ret = 0;
3210        goto fail;
3211    }
3212
3213    /*
3214     * Events should be removed after rmdir of cgroup directory, but before
3215     * destroying subsystem state objects. Let's take reference to cgroup
3216     * directory dentry to do that.
3217     */
3218    dget(cgrp->dentry);
3219
3220    spin_lock(&cgrp->event_list_lock);
3221    list_add(&event->list, &cgrp->event_list);
3222    spin_unlock(&cgrp->event_list_lock);
3223
3224    fput(cfile);
3225    fput(efile);
3226
3227    return 0;
3228
3229fail:
3230    if (cfile)
3231        fput(cfile);
3232
3233    if (event && event->eventfd && !IS_ERR(event->eventfd))
3234        eventfd_ctx_put(event->eventfd);
3235
3236    if (!IS_ERR_OR_NULL(efile))
3237        fput(efile);
3238
3239    kfree(event);
3240
3241    return ret;
3242}
3243
3244static u64 cgroup_clone_children_read(struct cgroup *cgrp,
3245                    struct cftype *cft)
3246{
3247    return clone_children(cgrp);
3248}
3249
3250static int cgroup_clone_children_write(struct cgroup *cgrp,
3251                     struct cftype *cft,
3252                     u64 val)
3253{
3254    if (val)
3255        set_bit(CGRP_CLONE_CHILDREN, &cgrp->flags);
3256    else
3257        clear_bit(CGRP_CLONE_CHILDREN, &cgrp->flags);
3258    return 0;
3259}
3260
3261/*
3262 * for the common functions, 'private' gives the type of file
3263 */
3264/* for hysterical raisins, we can't put this on the older files */
3265#define CGROUP_FILE_GENERIC_PREFIX "cgroup."
3266static struct cftype files[] = {
3267    {
3268        .name = "tasks",
3269        .open = cgroup_tasks_open,
3270        .write_u64 = cgroup_tasks_write,
3271        .release = cgroup_pidlist_release,
3272        .mode = S_IRUGO | S_IWUSR,
3273    },
3274    {
3275        .name = CGROUP_FILE_GENERIC_PREFIX "procs",
3276        .open = cgroup_procs_open,
3277        /* .write_u64 = cgroup_procs_write, TODO */
3278        .release = cgroup_pidlist_release,
3279        .mode = S_IRUGO,
3280    },
3281    {
3282        .name = "notify_on_release",
3283        .read_u64 = cgroup_read_notify_on_release,
3284        .write_u64 = cgroup_write_notify_on_release,
3285    },
3286    {
3287        .name = CGROUP_FILE_GENERIC_PREFIX "event_control",
3288        .write_string = cgroup_write_event_control,
3289        .mode = S_IWUGO,
3290    },
3291    {
3292        .name = "cgroup.clone_children",
3293        .read_u64 = cgroup_clone_children_read,
3294        .write_u64 = cgroup_clone_children_write,
3295    },
3296};
3297
3298static struct cftype cft_release_agent = {
3299    .name = "release_agent",
3300    .read_seq_string = cgroup_release_agent_show,
3301    .write_string = cgroup_release_agent_write,
3302    .max_write_len = PATH_MAX,
3303};
3304
3305static int cgroup_populate_dir(struct cgroup *cgrp)
3306{
3307    int err;
3308    struct cgroup_subsys *ss;
3309
3310    /* First clear out any existing files */
3311    cgroup_clear_directory(cgrp->dentry);
3312
3313    err = cgroup_add_files(cgrp, NULL, files, ARRAY_SIZE(files));
3314    if (err < 0)
3315        return err;
3316
3317    if (cgrp == cgrp->top_cgroup) {
3318        if ((err = cgroup_add_file(cgrp, NULL, &cft_release_agent)) < 0)
3319            return err;
3320    }
3321
3322    for_each_subsys(cgrp->root, ss) {
3323        if (ss->populate && (err = ss->populate(ss, cgrp)) < 0)
3324            return err;
3325    }
3326    /* This cgroup is ready now */
3327    for_each_subsys(cgrp->root, ss) {
3328        struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
3329        /*
3330         * Update id->css pointer and make this css visible from
3331         * CSS ID functions. This pointer will be dereferened
3332         * from RCU-read-side without locks.
3333         */
3334        if (css->id)
3335            rcu_assign_pointer(css->id->css, css);
3336    }
3337
3338    return 0;
3339}
3340
3341static void init_cgroup_css(struct cgroup_subsys_state *css,
3342                   struct cgroup_subsys *ss,
3343                   struct cgroup *cgrp)
3344{
3345    css->cgroup = cgrp;
3346    atomic_set(&css->refcnt, 1);
3347    css->flags = 0;
3348    css->id = NULL;
3349    if (cgrp == dummytop)
3350        set_bit(CSS_ROOT, &css->flags);
3351    BUG_ON(cgrp->subsys[ss->subsys_id]);
3352    cgrp->subsys[ss->subsys_id] = css;
3353}
3354
3355static void cgroup_lock_hierarchy(struct cgroupfs_root *root)
3356{
3357    /* We need to take each hierarchy_mutex in a consistent order */
3358    int i;
3359
3360    /*
3361     * No worry about a race with rebind_subsystems that might mess up the
3362     * locking order, since both parties are under cgroup_mutex.
3363     */
3364    for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
3365        struct cgroup_subsys *ss = subsys[i];
3366        if (ss == NULL)
3367            continue;
3368        if (ss->root == root)
3369            mutex_lock(&ss->hierarchy_mutex);
3370    }
3371}
3372
3373static void cgroup_unlock_hierarchy(struct cgroupfs_root *root)
3374{
3375    int i;
3376
3377    for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
3378        struct cgroup_subsys *ss = subsys[i];
3379        if (ss == NULL)
3380            continue;
3381        if (ss->root == root)
3382            mutex_unlock(&ss->hierarchy_mutex);
3383    }
3384}
3385
3386/*
3387 * cgroup_create - create a cgroup
3388 * @parent: cgroup that will be parent of the new cgroup
3389 * @dentry: dentry of the new cgroup
3390 * @mode: mode to set on new inode
3391 *
3392 * Must be called with the mutex on the parent inode held
3393 */
3394static long cgroup_create(struct cgroup *parent, struct dentry *dentry,
3395                 mode_t mode)
3396{
3397    struct cgroup *cgrp;
3398    struct cgroupfs_root *root = parent->root;
3399    int err = 0;
3400    struct cgroup_subsys *ss;
3401    struct super_block *sb = root->sb;
3402
3403    cgrp = kzalloc(sizeof(*cgrp), GFP_KERNEL);
3404    if (!cgrp)
3405        return -ENOMEM;
3406
3407    /* Grab a reference on the superblock so the hierarchy doesn't
3408     * get deleted on unmount if there are child cgroups. This
3409     * can be done outside cgroup_mutex, since the sb can't
3410     * disappear while someone has an open control file on the
3411     * fs */
3412    atomic_inc(&sb->s_active);
3413
3414    mutex_lock(&cgroup_mutex);
3415
3416    init_cgroup_housekeeping(cgrp);
3417
3418    cgrp->parent = parent;
3419    cgrp->root = parent->root;
3420    cgrp->top_cgroup = parent->top_cgroup;
3421
3422    if (notify_on_release(parent))
3423        set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
3424
3425    if (clone_children(parent))
3426        set_bit(CGRP_CLONE_CHILDREN, &cgrp->flags);
3427
3428    for_each_subsys(root, ss) {
3429        struct cgroup_subsys_state *css = ss->create(ss, cgrp);
3430
3431        if (IS_ERR(css)) {
3432            err = PTR_ERR(css);
3433            goto err_destroy;
3434        }
3435        init_cgroup_css(css, ss, cgrp);
3436        if (ss->use_id) {
3437            err = alloc_css_id(ss, parent, cgrp);
3438            if (err)
3439                goto err_destroy;
3440        }
3441        /* At error, ->destroy() callback has to free assigned ID. */
3442        if (clone_children(parent) && ss->post_clone)
3443            ss->post_clone(ss, cgrp);
3444    }
3445
3446    cgroup_lock_hierarchy(root);
3447    list_add(&cgrp->sibling, &cgrp->parent->children);
3448    cgroup_unlock_hierarchy(root);
3449    root->number_of_cgroups++;
3450
3451    err = cgroup_create_dir(cgrp, dentry, mode);
3452    if (err < 0)
3453        goto err_remove;
3454
3455    /* The cgroup directory was pre-locked for us */
3456    BUG_ON(!mutex_is_locked(&cgrp->dentry->d_inode->i_mutex));
3457
3458    err = cgroup_populate_dir(cgrp);
3459    /* If err < 0, we have a half-filled directory - oh well ;) */
3460
3461    mutex_unlock(&cgroup_mutex);
3462    mutex_unlock(&cgrp->dentry->d_inode->i_mutex);
3463
3464    return 0;
3465
3466 err_remove:
3467
3468    cgroup_lock_hierarchy(root);
3469    list_del(&cgrp->sibling);
3470    cgroup_unlock_hierarchy(root);
3471    root->number_of_cgroups--;
3472
3473 err_destroy:
3474
3475    for_each_subsys(root, ss) {
3476        if (cgrp->subsys[ss->subsys_id])
3477            ss->destroy(ss, cgrp);
3478    }
3479
3480    mutex_unlock(&cgroup_mutex);
3481
3482    /* Release the reference count that we took on the superblock */
3483    deactivate_super(sb);
3484
3485    kfree(cgrp);
3486    return err;
3487}
3488
3489static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, int mode)
3490{
3491    struct cgroup *c_parent = dentry->d_parent->d_fsdata;
3492
3493    /* the vfs holds inode->i_mutex already */
3494    return cgroup_create(c_parent, dentry, mode | S_IFDIR);
3495}
3496
3497static int cgroup_has_css_refs(struct cgroup *cgrp)
3498{
3499    /* Check the reference count on each subsystem. Since we
3500     * already established that there are no tasks in the
3501     * cgroup, if the css refcount is also 1, then there should
3502     * be no outstanding references, so the subsystem is safe to
3503     * destroy. We scan across all subsystems rather than using
3504     * the per-hierarchy linked list of mounted subsystems since
3505     * we can be called via check_for_release() with no
3506     * synchronization other than RCU, and the subsystem linked
3507     * list isn't RCU-safe */
3508    int i;
3509    /*
3510     * We won't need to lock the subsys array, because the subsystems
3511     * we're concerned about aren't going anywhere since our cgroup root
3512     * has a reference on them.
3513     */
3514    for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
3515        struct cgroup_subsys *ss = subsys[i];
3516        struct cgroup_subsys_state *css;
3517        /* Skip subsystems not present or not in this hierarchy */
3518        if (ss == NULL || ss->root != cgrp->root)
3519            continue;
3520        css = cgrp->subsys[ss->subsys_id];
3521        /* When called from check_for_release() it's possible
3522         * that by this point the cgroup has been removed
3523         * and the css deleted. But a false-positive doesn't
3524         * matter, since it can only happen if the cgroup
3525         * has been deleted and hence no longer needs the
3526         * release agent to be called anyway. */
3527        if (css && (atomic_read(&css->refcnt) > 1))
3528            return 1;
3529    }
3530    return 0;
3531}
3532
3533/*
3534 * Atomically mark all (or else none) of the cgroup's CSS objects as
3535 * CSS_REMOVED. Return true on success, or false if the cgroup has
3536 * busy subsystems. Call with cgroup_mutex held
3537 */
3538
3539static int cgroup_clear_css_refs(struct cgroup *cgrp)
3540{
3541    struct cgroup_subsys *ss;
3542    unsigned long flags;
3543    bool failed = false;
3544    local_irq_save(flags);
3545    for_each_subsys(cgrp->root, ss) {
3546        struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
3547        int refcnt;
3548        while (1) {
3549            /* We can only remove a CSS with a refcnt==1 */
3550            refcnt = atomic_read(&css->refcnt);
3551            if (refcnt > 1) {
3552                failed = true;
3553                goto done;
3554            }
3555            BUG_ON(!refcnt);
3556            /*
3557             * Drop the refcnt to 0 while we check other
3558             * subsystems. This will cause any racing
3559             * css_tryget() to spin until we set the
3560             * CSS_REMOVED bits or abort
3561             */
3562            if (atomic_cmpxchg(&css->refcnt, refcnt, 0) == refcnt)
3563                break;
3564            cpu_relax();
3565        }
3566    }
3567 done:
3568    for_each_subsys(cgrp->root, ss) {
3569        struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
3570        if (failed) {
3571            /*
3572             * Restore old refcnt if we previously managed
3573             * to clear it from 1 to 0
3574             */
3575            if (!atomic_read(&css->refcnt))
3576                atomic_set(&css->refcnt, 1);
3577        } else {
3578            /* Commit the fact that the CSS is removed */
3579            set_bit(CSS_REMOVED, &css->flags);
3580        }
3581    }
3582    local_irq_restore(flags);
3583    return !failed;
3584}
3585
3586static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry)
3587{
3588    struct cgroup *cgrp = dentry->d_fsdata;
3589    struct dentry *d;
3590    struct cgroup *parent;
3591    DEFINE_WAIT(wait);
3592    struct cgroup_event *event, *tmp;
3593    int ret;
3594
3595    /* the vfs holds both inode->i_mutex already */
3596again:
3597    mutex_lock(&cgroup_mutex);
3598    if (atomic_read(&cgrp->count) != 0) {
3599        mutex_unlock(&cgroup_mutex);
3600        return -EBUSY;
3601    }
3602    if (!list_empty(&cgrp->children)) {
3603        mutex_unlock(&cgroup_mutex);
3604        return -EBUSY;
3605    }
3606    mutex_unlock(&cgroup_mutex);
3607
3608    /*
3609     * In general, subsystem has no css->refcnt after pre_destroy(). But
3610     * in racy cases, subsystem may have to get css->refcnt after
3611     * pre_destroy() and it makes rmdir return with -EBUSY. This sometimes
3612     * make rmdir return -EBUSY too often. To avoid that, we use waitqueue
3613     * for cgroup's rmdir. CGRP_WAIT_ON_RMDIR is for synchronizing rmdir
3614     * and subsystem's reference count handling. Please see css_get/put
3615     * and css_tryget() and cgroup_wakeup_rmdir_waiter() implementation.
3616     */
3617    set_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
3618
3619    /*
3620     * Call pre_destroy handlers of subsys. Notify subsystems
3621     * that rmdir() request comes.
3622     */
3623    ret = cgroup_call_pre_destroy(cgrp);
3624    if (ret) {
3625        clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
3626        return ret;
3627    }
3628
3629    mutex_lock(&cgroup_mutex);
3630    parent = cgrp->parent;
3631    if (atomic_read(&cgrp->count) || !list_empty(&cgrp->children)) {
3632        clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
3633        mutex_unlock(&cgroup_mutex);
3634        return -EBUSY;
3635    }
3636    prepare_to_wait(&cgroup_rmdir_waitq, &wait, TASK_INTERRUPTIBLE);
3637    if (!cgroup_clear_css_refs(cgrp)) {
3638        mutex_unlock(&cgroup_mutex);
3639        /*
3640         * Because someone may call cgroup_wakeup_rmdir_waiter() before
3641         * prepare_to_wait(), we need to check this flag.
3642         */
3643        if (test_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags))
3644            schedule();
3645        finish_wait(&cgroup_rmdir_waitq, &wait);
3646        clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
3647        if (signal_pending(current))
3648            return -EINTR;
3649        goto again;
3650    }
3651    /* NO css_tryget() can success after here. */
3652    finish_wait(&cgroup_rmdir_waitq, &wait);
3653    clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
3654
3655    spin_lock(&release_list_lock);
3656    set_bit(CGRP_REMOVED, &cgrp->flags);
3657    if (!list_empty(&cgrp->release_list))
3658        list_del(&cgrp->release_list);
3659    spin_unlock(&release_list_lock);
3660
3661    cgroup_lock_hierarchy(cgrp->root);
3662    /* delete this cgroup from parent->children */
3663    list_del(&cgrp->sibling);
3664    cgroup_unlock_hierarchy(cgrp->root);
3665
3666    d = dget(cgrp->dentry);
3667
3668    cgroup_d_remove_dir(d);
3669    dput(d);
3670
3671    set_bit(CGRP_RELEASABLE, &parent->flags);
3672    check_for_release(parent);
3673
3674    /*
3675     * Unregister events and notify userspace.
3676     * Notify userspace about cgroup removing only after rmdir of cgroup
3677     * directory to avoid race between userspace and kernelspace
3678     */
3679    spin_lock(&cgrp->event_list_lock);
3680    list_for_each_entry_safe(event, tmp, &cgrp->event_list, list) {
3681        list_del(&event->list);
3682        remove_wait_queue(event->wqh, &event->wait);
3683        eventfd_signal(event->eventfd, 1);
3684        schedule_work(&event->remove);
3685    }
3686    spin_unlock(&cgrp->event_list_lock);
3687
3688    mutex_unlock(&cgroup_mutex);
3689    return 0;
3690}
3691
3692static void __init cgroup_init_subsys(struct cgroup_subsys *ss)
3693{
3694    struct cgroup_subsys_state *css;
3695
3696    printk(KERN_INFO "Initializing cgroup subsys %s\n", ss->name);
3697
3698    /* Create the top cgroup state for this subsystem */
3699    list_add(&ss->sibling, &rootnode.subsys_list);
3700    ss->root = &rootnode;
3701    css = ss->create(ss, dummytop);
3702    /* We don't handle early failures gracefully */
3703    BUG_ON(IS_ERR(css));
3704    init_cgroup_css(css, ss, dummytop);
3705
3706    /* Update the init_css_set to contain a subsys
3707     * pointer to this state - since the subsystem is
3708     * newly registered, all tasks and hence the
3709     * init_css_set is in the subsystem's top cgroup. */
3710    init_css_set.subsys[ss->subsys_id] = dummytop->subsys[ss->subsys_id];
3711
3712    need_forkexit_callback |= ss->fork || ss->exit;
3713
3714    /* At system boot, before all subsystems have been
3715     * registered, no tasks have been forked, so we don't
3716     * need to invoke fork callbacks here. */
3717    BUG_ON(!list_empty(&init_task.tasks));
3718
3719    mutex_init(&ss->hierarchy_mutex);
3720    lockdep_set_class(&ss->hierarchy_mutex, &ss->subsys_key);
3721    ss->active = 1;
3722
3723    /* this function shouldn't be used with modular subsystems, since they
3724     * need to register a subsys_id, among other things */
3725    BUG_ON(ss->module);
3726}
3727
3728/**
3729 * cgroup_load_subsys: load and register a modular subsystem at runtime
3730 * @ss: the subsystem to load
3731 *
3732 * This function should be called in a modular subsystem's initcall. If the
3733 * subsystem is built as a module, it will be assigned a new subsys_id and set
3734 * up for use. If the subsystem is built-in anyway, work is delegated to the
3735 * simpler cgroup_init_subsys.
3736 */
3737int __init_or_module cgroup_load_subsys(struct cgroup_subsys *ss)
3738{
3739    int i;
3740    struct cgroup_subsys_state *css;
3741
3742    /* check name and function validity */
3743    if (ss->name == NULL || strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN ||
3744        ss->create == NULL || ss->destroy == NULL)
3745        return -EINVAL;
3746
3747    /*
3748     * we don't support callbacks in modular subsystems. this check is
3749     * before the ss->module check for consistency; a subsystem that could
3750     * be a module should still have no callbacks even if the user isn't
3751     * compiling it as one.
3752     */
3753    if (ss->fork || ss->exit)
3754        return -EINVAL;
3755
3756    /*
3757     * an optionally modular subsystem is built-in: we want to do nothing,
3758     * since cgroup_init_subsys will have already taken care of it.
3759     */
3760    if (ss->module == NULL) {
3761        /* a few sanity checks */
3762        BUG_ON(ss->subsys_id >= CGROUP_BUILTIN_SUBSYS_COUNT);
3763        BUG_ON(subsys[ss->subsys_id] != ss);
3764        return 0;
3765    }
3766
3767    /*
3768     * need to register a subsys id before anything else - for example,
3769     * init_cgroup_css needs it.
3770     */
3771    mutex_lock(&cgroup_mutex);
3772    /* find the first empty slot in the array */
3773    for (i = CGROUP_BUILTIN_SUBSYS_COUNT; i < CGROUP_SUBSYS_COUNT; i++) {
3774        if (subsys[i] == NULL)
3775            break;
3776    }
3777    if (i == CGROUP_SUBSYS_COUNT) {
3778        /* maximum number of subsystems already registered! */
3779        mutex_unlock(&cgroup_mutex);
3780        return -EBUSY;
3781    }
3782    /* assign ourselves the subsys_id */
3783    ss->subsys_id = i;
3784    subsys[i] = ss;
3785
3786    /*
3787     * no ss->create seems to need anything important in the ss struct, so
3788     * this can happen first (i.e. before the rootnode attachment).
3789     */
3790    css = ss->create(ss, dummytop);
3791    if (IS_ERR(css)) {
3792        /* failure case - need to deassign the subsys[] slot. */
3793        subsys[i] = NULL;
3794        mutex_unlock(&cgroup_mutex);
3795        return PTR_ERR(css);
3796    }
3797
3798    list_add(&ss->sibling, &rootnode.subsys_list);
3799    ss->root = &rootnode;
3800
3801    /* our new subsystem will be attached to the dummy hierarchy. */
3802    init_cgroup_css(css, ss, dummytop);
3803    /* init_idr must be after init_cgroup_css because it sets css->id. */
3804    if (ss->use_id) {
3805        int ret = cgroup_init_idr(ss, css);
3806        if (ret) {
3807            dummytop->subsys[ss->subsys_id] = NULL;
3808            ss->destroy(ss, dummytop);
3809            subsys[i] = NULL;
3810            mutex_unlock(&cgroup_mutex);
3811            return ret;
3812        }
3813    }
3814
3815    /*
3816     * Now we need to entangle the css into the existing css_sets. unlike
3817     * in cgroup_init_subsys, there are now multiple css_sets, so each one
3818     * will need a new pointer to it; done by iterating the css_set_table.
3819     * furthermore, modifying the existing css_sets will corrupt the hash
3820     * table state, so each changed css_set will need its hash recomputed.
3821     * this is all done under the css_set_lock.
3822     */
3823    write_lock(&css_set_lock);
3824    for (i = 0; i < CSS_SET_TABLE_SIZE; i++) {
3825        struct css_set *cg;
3826        struct hlist_node *node, *tmp;
3827        struct hlist_head *bucket = &css_set_table[i], *new_bucket;
3828
3829        hlist_for_each_entry_safe(cg, node, tmp, bucket, hlist) {
3830            /* skip entries that we already rehashed */
3831            if (cg->subsys[ss->subsys_id])
3832                continue;
3833            /* remove existing entry */
3834            hlist_del(&cg->hlist);
3835            /* set new value */
3836            cg->subsys[ss->subsys_id] = css;
3837            /* recompute hash and restore entry */
3838            new_bucket = css_set_hash(cg->subsys);
3839            hlist_add_head(&cg->hlist, new_bucket);
3840        }
3841    }
3842    write_unlock(&css_set_lock);
3843
3844    mutex_init(&ss->hierarchy_mutex);
3845    lockdep_set_class(&ss->hierarchy_mutex, &ss->subsys_key);
3846    ss->active = 1;
3847
3848    /* success! */
3849    mutex_unlock(&cgroup_mutex);
3850    return 0;
3851}
3852EXPORT_SYMBOL_GPL(cgroup_load_subsys);
3853
3854/**
3855 * cgroup_unload_subsys: unload a modular subsystem
3856 * @ss: the subsystem to unload
3857 *
3858 * This function should be called in a modular subsystem's exitcall. When this
3859 * function is invoked, the refcount on the subsystem's module will be 0, so
3860 * the subsystem will not be attached to any hierarchy.
3861 */
3862void cgroup_unload_subsys(struct cgroup_subsys *ss)
3863{
3864    struct cg_cgroup_link *link;
3865    struct hlist_head *hhead;
3866
3867    BUG_ON(ss->module == NULL);
3868
3869    /*
3870     * we shouldn't be called if the subsystem is in use, and the use of
3871     * try_module_get in parse_cgroupfs_options should ensure that it
3872     * doesn't start being used while we're killing it off.
3873     */
3874    BUG_ON(ss->root != &rootnode);
3875
3876    mutex_lock(&cgroup_mutex);
3877    /* deassign the subsys_id */
3878    BUG_ON(ss->subsys_id < CGROUP_BUILTIN_SUBSYS_COUNT);
3879    subsys[ss->subsys_id] = NULL;
3880
3881    /* remove subsystem from rootnode's list of subsystems */
3882    list_del(&ss->sibling);
3883
3884    /*
3885     * disentangle the css from all css_sets attached to the dummytop. as
3886     * in loading, we need to pay our respects to the hashtable gods.
3887     */
3888    write_lock(&css_set_lock);
3889    list_for_each_entry(link, &dummytop->css_sets, cgrp_link_list) {
3890        struct css_set *cg = link->cg;
3891
3892        hlist_del(&cg->hlist);
3893        BUG_ON(!cg->subsys[ss->subsys_id]);
3894        cg->subsys[ss->subsys_id] = NULL;
3895        hhead = css_set_hash(cg->subsys);
3896        hlist_add_head(&cg->hlist, hhead);
3897    }
3898    write_unlock(&css_set_lock);
3899
3900    /*
3901     * remove subsystem's css from the dummytop and free it - need to free
3902     * before marking as null because ss->destroy needs the cgrp->subsys
3903     * pointer to find their state. note that this also takes care of
3904     * freeing the css_id.
3905     */
3906    ss->destroy(ss, dummytop);
3907    dummytop->subsys[ss->subsys_id] = NULL;
3908
3909    mutex_unlock(&cgroup_mutex);
3910}
3911EXPORT_SYMBOL_GPL(cgroup_unload_subsys);
3912
3913/**
3914 * cgroup_init_early - cgroup initialization at system boot
3915 *
3916 * Initialize cgroups at system boot, and initialize any
3917 * subsystems that request early init.
3918 */
3919int __init cgroup_init_early(void)
3920{
3921    int i;
3922    atomic_set(&init_css_set.refcount, 1);
3923    INIT_LIST_HEAD(&init_css_set.cg_links);
3924    INIT_LIST_HEAD(&init_css_set.tasks);
3925    INIT_HLIST_NODE(&init_css_set.hlist);
3926    css_set_count = 1;
3927    init_cgroup_root(&rootnode);
3928    root_count = 1;
3929    init_task.cgroups = &init_css_set;
3930
3931    init_css_set_link.cg = &init_css_set;
3932    init_css_set_link.cgrp = dummytop;
3933    list_add(&init_css_set_link.cgrp_link_list,
3934         &rootnode.top_cgroup.css_sets);
3935    list_add(&init_css_set_link.cg_link_list,
3936         &init_css_set.cg_links);
3937
3938    for (i = 0; i < CSS_SET_TABLE_SIZE; i++)
3939        INIT_HLIST_HEAD(&css_set_table[i]);
3940
3941    /* at bootup time, we don't worry about modular subsystems */
3942    for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
3943        struct cgroup_subsys *ss = subsys[i];
3944
3945        BUG_ON(!ss->name);
3946        BUG_ON(strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN);
3947        BUG_ON(!ss->create);
3948        BUG_ON(!ss->destroy);
3949        if (ss->subsys_id != i) {
3950            printk(KERN_ERR "cgroup: Subsys %s id == %d\n",
3951                   ss->name, ss->subsys_id);
3952            BUG();
3953        }
3954
3955        if (ss->early_init)
3956            cgroup_init_subsys(ss);
3957    }
3958    return 0;
3959}
3960
3961/**
3962 * cgroup_init - cgroup initialization
3963 *
3964 * Register cgroup filesystem and /proc file, and initialize
3965 * any subsystems that didn't request early init.
3966 */
3967int __init cgroup_init(void)
3968{
3969    int err;
3970    int i;
3971    struct hlist_head *hhead;
3972
3973    err = bdi_init(&cgroup_backing_dev_info);
3974    if (err)
3975        return err;
3976
3977    /* at bootup time, we don't worry about modular subsystems */
3978    for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
3979        struct cgroup_subsys *ss = subsys[i];
3980        if (!ss->early_init)
3981            cgroup_init_subsys(ss);
3982        if (ss->use_id)
3983            cgroup_init_idr(ss, init_css_set.subsys[ss->subsys_id]);
3984    }
3985
3986    /* Add init_css_set to the hash table */
3987    hhead = css_set_hash(init_css_set.subsys);
3988    hlist_add_head(&init_css_set.hlist, hhead);
3989    BUG_ON(!init_root_id(&rootnode));
3990
3991    cgroup_kobj = kobject_create_and_add("cgroup", fs_kobj);
3992    if (!cgroup_kobj) {
3993        err = -ENOMEM;
3994        goto out;
3995    }
3996
3997    err = register_filesystem(&cgroup_fs_type);
3998    if (err < 0) {
3999        kobject_put(cgroup_kobj);
4000        goto out;
4001    }
4002
4003    proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations);
4004
4005out:
4006    if (err)
4007        bdi_destroy(&cgroup_backing_dev_info);
4008
4009    return err;
4010}
4011
4012/*
4013 * proc_cgroup_show()
4014 * - Print task's cgroup paths into seq_file, one line for each hierarchy
4015 * - Used for /proc/<pid>/cgroup.
4016 * - No need to task_lock(tsk) on this tsk->cgroup reference, as it
4017 * doesn't really matter if tsk->cgroup changes after we read it,
4018 * and we take cgroup_mutex, keeping cgroup_attach_task() from changing it
4019 * anyway. No need to check that tsk->cgroup != NULL, thanks to
4020 * the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks
4021 * cgroup to top_cgroup.
4022 */
4023
4024/* TODO: Use a proper seq_file iterator */
4025static int proc_cgroup_show(struct seq_file *m, void *v)
4026{
4027    struct pid *pid;
4028    struct task_struct *tsk;
4029    char *buf;
4030    int retval;
4031    struct cgroupfs_root *root;
4032
4033    retval = -ENOMEM;
4034    buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
4035    if (!buf)
4036        goto out;
4037
4038    retval = -ESRCH;
4039    pid = m->private;
4040    tsk = get_pid_task(pid, PIDTYPE_PID);
4041    if (!tsk)
4042        goto out_free;
4043
4044    retval = 0;
4045
4046    mutex_lock(&cgroup_mutex);
4047
4048    for_each_active_root(root) {
4049        struct cgroup_subsys *ss;
4050        struct cgroup *cgrp;
4051        int count = 0;
4052
4053        seq_printf(m, "%d:", root->hierarchy_id);
4054        for_each_subsys(root, ss)
4055            seq_printf(m, "%s%s", count++ ? "," : "", ss->name);
4056        if (strlen(root->name))
4057            seq_printf(m, "%sname=%s", count ? "," : "",
4058                   root->name);
4059        seq_putc(m, ':');
4060        cgrp = task_cgroup_from_root(tsk, root);
4061        retval = cgroup_path(cgrp, buf, PAGE_SIZE);
4062        if (retval < 0)
4063            goto out_unlock;
4064        seq_puts(m, buf);
4065        seq_putc(m, '\n');
4066    }
4067
4068out_unlock:
4069    mutex_unlock(&cgroup_mutex);
4070    put_task_struct(tsk);
4071out_free:
4072    kfree(buf);
4073out:
4074    return retval;
4075}
4076
4077static int cgroup_open(struct inode *inode, struct file *file)
4078{
4079    struct pid *pid = PROC_I(inode)->pid;
4080    return single_open(file, proc_cgroup_show, pid);
4081}
4082
4083const struct file_operations proc_cgroup_operations = {
4084    .open = cgroup_open,
4085    .read = seq_read,
4086    .llseek = seq_lseek,
4087    .release = single_release,
4088};
4089
4090/* Display information about each subsystem and each hierarchy */
4091static int proc_cgroupstats_show(struct seq_file *m, void *v)
4092{
4093    int i;
4094
4095    seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
4096    /*
4097     * ideally we don't want subsystems moving around while we do this.
4098     * cgroup_mutex is also necessary to guarantee an atomic snapshot of
4099     * subsys/hierarchy state.
4100     */
4101    mutex_lock(&cgroup_mutex);
4102    for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
4103        struct cgroup_subsys *ss = subsys[i];
4104        if (ss == NULL)
4105            continue;
4106        seq_printf(m, "%s\t%d\t%d\t%d\n",
4107               ss->name, ss->root->hierarchy_id,
4108               ss->root->number_of_cgroups, !ss->disabled);
4109    }
4110    mutex_unlock(&cgroup_mutex);
4111    return 0;
4112}
4113
4114static int cgroupstats_open(struct inode *inode, struct file *file)
4115{
4116    return single_open(file, proc_cgroupstats_show, NULL);
4117}
4118
4119static const struct file_operations proc_cgroupstats_operations = {
4120    .open = cgroupstats_open,
4121    .read = seq_read,
4122    .llseek = seq_lseek,
4123    .release = single_release,
4124};
4125
4126/**
4127 * cgroup_fork - attach newly forked task to its parents cgroup.
4128 * @child: pointer to task_struct of forking parent process.
4129 *
4130 * Description: A task inherits its parent's cgroup at fork().
4131 *
4132 * A pointer to the shared css_set was automatically copied in
4133 * fork.c by dup_task_struct(). However, we ignore that copy, since
4134 * it was not made under the protection of RCU or cgroup_mutex, so
4135 * might no longer be a valid cgroup pointer. cgroup_attach_task() might
4136 * have already changed current->cgroups, allowing the previously
4137 * referenced cgroup group to be removed and freed.
4138 *
4139 * At the point that cgroup_fork() is called, 'current' is the parent
4140 * task, and the passed argument 'child' points to the child task.
4141 */
4142void cgroup_fork(struct task_struct *child)
4143{
4144    task_lock(current);
4145    child->cgroups = current->cgroups;
4146    get_css_set(child->cgroups);
4147    task_unlock(current);
4148    INIT_LIST_HEAD(&child->cg_list);
4149}
4150
4151/**
4152 * cgroup_fork_callbacks - run fork callbacks
4153 * @child: the new task
4154 *
4155 * Called on a new task very soon before adding it to the
4156 * tasklist. No need to take any locks since no-one can
4157 * be operating on this task.
4158 */
4159void cgroup_fork_callbacks(struct task_struct *child)
4160{
4161    if (need_forkexit_callback) {
4162        int i;
4163        /*
4164         * forkexit callbacks are only supported for builtin
4165         * subsystems, and the builtin section of the subsys array is
4166         * immutable, so we don't need to lock the subsys array here.
4167         */
4168        for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
4169            struct cgroup_subsys *ss = subsys[i];
4170            if (ss->fork)
4171                ss->fork(ss, child);
4172        }
4173    }
4174}
4175
4176/**
4177 * cgroup_post_fork - called on a new task after adding it to the task list
4178 * @child: the task in question
4179 *
4180 * Adds the task to the list running through its css_set if necessary.
4181 * Has to be after the task is visible on the task list in case we race
4182 * with the first call to cgroup_iter_start() - to guarantee that the
4183 * new task ends up on its list.
4184 */
4185void cgroup_post_fork(struct task_struct *child)
4186{
4187    if (use_task_css_set_links) {
4188        write_lock(&css_set_lock);
4189        task_lock(child);
4190        if (list_empty(&child->cg_list))
4191            list_add(&child->cg_list, &child->cgroups->tasks);
4192        task_unlock(child);
4193        write_unlock(&css_set_lock);
4194    }
4195}
4196/**
4197 * cgroup_exit - detach cgroup from exiting task
4198 * @tsk: pointer to task_struct of exiting process
4199 * @run_callback: run exit callbacks?
4200 *
4201 * Description: Detach cgroup from @tsk and release it.
4202 *
4203 * Note that cgroups marked notify_on_release force every task in
4204 * them to take the global cgroup_mutex mutex when exiting.
4205 * This could impact scaling on very large systems. Be reluctant to
4206 * use notify_on_release cgroups where very high task exit scaling
4207 * is required on large systems.
4208 *
4209 * the_top_cgroup_hack:
4210 *
4211 * Set the exiting tasks cgroup to the root cgroup (top_cgroup).
4212 *
4213 * We call cgroup_exit() while the task is still competent to
4214 * handle notify_on_release(), then leave the task attached to the
4215 * root cgroup in each hierarchy for the remainder of its exit.
4216 *
4217 * To do this properly, we would increment the reference count on
4218 * top_cgroup, and near the very end of the kernel/exit.c do_exit()
4219 * code we would add a second cgroup function call, to drop that
4220 * reference. This would just create an unnecessary hot spot on
4221 * the top_cgroup reference count, to no avail.
4222 *
4223 * Normally, holding a reference to a cgroup without bumping its
4224 * count is unsafe. The cgroup could go away, or someone could
4225 * attach us to a different cgroup, decrementing the count on
4226 * the first cgroup that we never incremented. But in this case,
4227 * top_cgroup isn't going away, and either task has PF_EXITING set,
4228 * which wards off any cgroup_attach_task() attempts, or task is a failed
4229 * fork, never visible to cgroup_attach_task.
4230 */
4231void cgroup_exit(struct task_struct *tsk, int run_callbacks)
4232{
4233    int i;
4234    struct css_set *cg;
4235
4236    if (run_callbacks && need_forkexit_callback) {
4237        /*
4238         * modular subsystems can't use callbacks, so no need to lock
4239         * the subsys array
4240         */
4241        for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
4242            struct cgroup_subsys *ss = subsys[i];
4243            if (ss->exit)
4244                ss->exit(ss, tsk);
4245        }
4246    }
4247
4248    /*
4249     * Unlink from the css_set task list if necessary.
4250     * Optimistically check cg_list before taking
4251     * css_set_lock
4252     */
4253    if (!list_empty(&tsk->cg_list)) {
4254        write_lock(&css_set_lock);
4255        if (!list_empty(&tsk->cg_list))
4256            list_del(&tsk->cg_list);
4257        write_unlock(&css_set_lock);
4258    }
4259
4260    /* Reassign the task to the init_css_set. */
4261    task_lock(tsk);
4262    cg = tsk->cgroups;
4263    tsk->cgroups = &init_css_set;
4264    task_unlock(tsk);
4265    if (cg)
4266        put_css_set_taskexit(cg);
4267}
4268
4269/**
4270 * cgroup_clone - clone the cgroup the given subsystem is attached to
4271 * @tsk: the task to be moved
4272 * @subsys: the given subsystem
4273 * @nodename: the name for the new cgroup
4274 *
4275 * Duplicate the current cgroup in the hierarchy that the given
4276 * subsystem is attached to, and move this task into the new
4277 * child.
4278 */
4279int cgroup_clone(struct task_struct *tsk, struct cgroup_subsys *subsys,
4280                            char *nodename)
4281{
4282    struct dentry *dentry;
4283    int ret = 0;
4284    struct cgroup *parent, *child;
4285    struct inode *inode;
4286    struct css_set *cg;
4287    struct cgroupfs_root *root;
4288    struct cgroup_subsys *ss;
4289
4290    /* We shouldn't be called by an unregistered subsystem */
4291    BUG_ON(!subsys->active);
4292
4293    /* First figure out what hierarchy and cgroup we're dealing
4294     * with, and pin them so we can drop cgroup_mutex */
4295    mutex_lock(&cgroup_mutex);
4296 again:
4297    root = subsys->root;
4298    if (root == &rootnode) {
4299        mutex_unlock(&cgroup_mutex);
4300        return 0;
4301    }
4302
4303    /* Pin the hierarchy */
4304    if (!atomic_inc_not_zero(&root->sb->s_active)) {
4305        /* We race with the final deactivate_super() */
4306        mutex_unlock(&cgroup_mutex);
4307        return 0;
4308    }
4309
4310    /* Keep the cgroup alive */
4311    task_lock(tsk);
4312    parent = task_cgroup(tsk, subsys->subsys_id);
4313    cg = tsk->cgroups;
4314    get_css_set(cg);
4315    task_unlock(tsk);
4316
4317    mutex_unlock(&cgroup_mutex);
4318
4319    /* Now do the VFS work to create a cgroup */
4320    inode = parent->dentry->d_inode;
4321
4322    /* Hold the parent directory mutex across this operation to
4323     * stop anyone else deleting the new cgroup */
4324    mutex_lock(&inode->i_mutex);
4325    dentry = lookup_one_len(nodename, parent->dentry, strlen(nodename));
4326    if (IS_ERR(dentry)) {
4327        printk(KERN_INFO
4328               "cgroup: Couldn't allocate dentry for %s: %ld\n", nodename,
4329               PTR_ERR(dentry));
4330        ret = PTR_ERR(dentry);
4331        goto out_release;
4332    }
4333
4334    /* Create the cgroup directory, which also creates the cgroup */
4335    ret = vfs_mkdir(inode, dentry, 0755);
4336    child = __d_cgrp(dentry);
4337    dput(dentry);
4338    if (ret) {
4339        printk(KERN_INFO
4340               "Failed to create cgroup %s: %d\n", nodename,
4341               ret);
4342        goto out_release;
4343    }
4344
4345    /* The cgroup now exists. Retake cgroup_mutex and check
4346     * that we're still in the same state that we thought we
4347     * were. */
4348    mutex_lock(&cgroup_mutex);
4349    if ((root != subsys->root) ||
4350        (parent != task_cgroup(tsk, subsys->subsys_id))) {
4351        /* Aargh, we raced ... */
4352        mutex_unlock(&inode->i_mutex);
4353        put_css_set(cg);
4354
4355        deactivate_super(root->sb);
4356        /* The cgroup is still accessible in the VFS, but
4357         * we're not going to try to rmdir() it at this
4358         * point. */
4359        printk(KERN_INFO
4360               "Race in cgroup_clone() - leaking cgroup %s\n",
4361               nodename);
4362        goto again;
4363    }
4364
4365    /* do any required auto-setup */
4366    for_each_subsys(root, ss) {
4367        if (ss->post_clone)
4368            ss->post_clone(ss, child);
4369    }
4370
4371    /* All seems fine. Finish by moving the task into the new cgroup */
4372    ret = cgroup_attach_task(child, tsk);
4373    mutex_unlock(&cgroup_mutex);
4374
4375 out_release:
4376    mutex_unlock(&inode->i_mutex);
4377
4378    mutex_lock(&cgroup_mutex);
4379    put_css_set(cg);
4380    mutex_unlock(&cgroup_mutex);
4381    deactivate_super(root->sb);
4382    return ret;
4383}
4384
4385/**
4386 * cgroup_is_descendant - see if @cgrp is a descendant of @task's cgrp
4387 * @cgrp: the cgroup in question
4388 * @task: the task in question
4389 *
4390 * See if @cgrp is a descendant of @task's cgroup in the appropriate
4391 * hierarchy.
4392 *
4393 * If we are sending in dummytop, then presumably we are creating
4394 * the top cgroup in the subsystem.
4395 *
4396 * Called only by the ns (nsproxy) cgroup.
4397 */
4398int cgroup_is_descendant(const struct cgroup *cgrp, struct task_struct *task)
4399{
4400    int ret;
4401    struct cgroup *target;
4402
4403    if (cgrp == dummytop)
4404        return 1;
4405
4406    target = task_cgroup_from_root(task, cgrp->root);
4407    while (cgrp != target && cgrp!= cgrp->top_cgroup)
4408        cgrp = cgrp->parent;
4409    ret = (cgrp == target);
4410    return ret;
4411}
4412
4413static void check_for_release(struct cgroup *cgrp)
4414{
4415    /* All of these checks rely on RCU to keep the cgroup
4416     * structure alive */
4417    if (cgroup_is_releasable(cgrp) && !atomic_read(&cgrp->count)
4418        && list_empty(&cgrp->children) && !cgroup_has_css_refs(cgrp)) {
4419        /* Control Group is currently removeable. If it's not
4420         * already queued for a userspace notification, queue
4421         * it now */
4422        int need_schedule_work = 0;
4423        spin_lock(&release_list_lock);
4424        if (!cgroup_is_removed(cgrp) &&
4425            list_empty(&cgrp->release_list)) {
4426            list_add(&cgrp->release_list, &release_list);
4427            need_schedule_work = 1;
4428        }
4429        spin_unlock(&release_list_lock);
4430        if (need_schedule_work)
4431            schedule_work(&release_agent_work);
4432    }
4433}
4434
4435/* Caller must verify that the css is not for root cgroup */
4436void __css_put(struct cgroup_subsys_state *css, int count)
4437{
4438    struct cgroup *cgrp = css->cgroup;
4439    int val;
4440    rcu_read_lock();
4441    val = atomic_sub_return(count, &css->refcnt);
4442    if (val == 1) {
4443        if (notify_on_release(cgrp)) {
4444            set_bit(CGRP_RELEASABLE, &cgrp->flags);
4445            check_for_release(cgrp);
4446        }
4447        cgroup_wakeup_rmdir_waiter(cgrp);
4448    }
4449    rcu_read_unlock();
4450    WARN_ON_ONCE(val < 1);
4451}
4452EXPORT_SYMBOL_GPL(__css_put);
4453
4454/*
4455 * Notify userspace when a cgroup is released, by running the
4456 * configured release agent with the name of the cgroup (path
4457 * relative to the root of cgroup file system) as the argument.
4458 *
4459 * Most likely, this user command will try to rmdir this cgroup.
4460 *
4461 * This races with the possibility that some other task will be
4462 * attached to this cgroup before it is removed, or that some other
4463 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
4464 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
4465 * unused, and this cgroup will be reprieved from its death sentence,
4466 * to continue to serve a useful existence. Next time it's released,
4467 * we will get notified again, if it still has 'notify_on_release' set.
4468 *
4469 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
4470 * means only wait until the task is successfully execve()'d. The
4471 * separate release agent task is forked by call_usermodehelper(),
4472 * then control in this thread returns here, without waiting for the
4473 * release agent task. We don't bother to wait because the caller of
4474 * this routine has no use for the exit status of the release agent
4475 * task, so no sense holding our caller up for that.
4476 */
4477static void cgroup_release_agent(struct work_struct *work)
4478{
4479    BUG_ON(work != &release_agent_work);
4480    mutex_lock(&cgroup_mutex);
4481    spin_lock(&release_list_lock);
4482    while (!list_empty(&release_list)) {
4483        char *argv[3], *envp[3];
4484        int i;
4485        char *pathbuf = NULL, *agentbuf = NULL;
4486        struct cgroup *cgrp = list_entry(release_list.next,
4487                            struct cgroup,
4488                            release_list);
4489        list_del_init(&cgrp->release_list);
4490        spin_unlock(&release_list_lock);
4491        pathbuf = kmalloc(PAGE_SIZE, GFP_KERNEL);
4492        if (!pathbuf)
4493            goto continue_free;
4494        if (cgroup_path(cgrp, pathbuf, PAGE_SIZE) < 0)
4495            goto continue_free;
4496        agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL);
4497        if (!agentbuf)
4498            goto continue_free;
4499
4500        i = 0;
4501        argv[i++] = agentbuf;
4502        argv[i++] = pathbuf;
4503        argv[i] = NULL;
4504
4505        i = 0;
4506        /* minimal command environment */
4507        envp[i++] = "HOME=/";
4508        envp[i++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
4509        envp[i] = NULL;
4510
4511        /* Drop the lock while we invoke the usermode helper,
4512         * since the exec could involve hitting disk and hence
4513         * be a slow process */
4514        mutex_unlock(&cgroup_mutex);
4515        call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC);
4516        mutex_lock(&cgroup_mutex);
4517 continue_free:
4518        kfree(pathbuf);
4519        kfree(agentbuf);
4520        spin_lock(&release_list_lock);
4521    }
4522    spin_unlock(&release_list_lock);
4523    mutex_unlock(&cgroup_mutex);
4524}
4525
4526static int __init cgroup_disable(char *str)
4527{
4528    int i;
4529    char *token;
4530
4531    while ((token = strsep(&str, ",")) != NULL) {
4532        if (!*token)
4533            continue;
4534        /*
4535         * cgroup_disable, being at boot time, can't know about module
4536         * subsystems, so we don't worry about them.
4537         */
4538        for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
4539            struct cgroup_subsys *ss = subsys[i];
4540
4541            if (!strcmp(token, ss->name)) {
4542                ss->disabled = 1;
4543                printk(KERN_INFO "Disabling %s control group"
4544                    " subsystem\n", ss->name);
4545                break;
4546            }
4547        }
4548    }
4549    return 1;
4550}
4551__setup("cgroup_disable=", cgroup_disable);
4552
4553/*
4554 * Functons for CSS ID.
4555 */
4556
4557/*
4558 *To get ID other than 0, this should be called when !cgroup_is_removed().
4559 */
4560unsigned short css_id(struct cgroup_subsys_state *css)
4561{
4562    struct css_id *cssid;
4563
4564    /*
4565     * This css_id() can return correct value when somone has refcnt
4566     * on this or this is under rcu_read_lock(). Once css->id is allocated,
4567     * it's unchanged until freed.
4568     */
4569    cssid = rcu_dereference_check(css->id,
4570            rcu_read_lock_held() || atomic_read(&css->refcnt));
4571
4572    if (cssid)
4573        return cssid->id;
4574    return 0;
4575}
4576EXPORT_SYMBOL_GPL(css_id);
4577
4578unsigned short css_depth(struct cgroup_subsys_state *css)
4579{
4580    struct css_id *cssid;
4581
4582    cssid = rcu_dereference_check(css->id,
4583            rcu_read_lock_held() || atomic_read(&css->refcnt));
4584
4585    if (cssid)
4586        return cssid->depth;
4587    return 0;
4588}
4589EXPORT_SYMBOL_GPL(css_depth);
4590
4591/**
4592 * css_is_ancestor - test "root" css is an ancestor of "child"
4593 * @child: the css to be tested.
4594 * @root: the css supporsed to be an ancestor of the child.
4595 *
4596 * Returns true if "root" is an ancestor of "child" in its hierarchy. Because
4597 * this function reads css->id, this use rcu_dereference() and rcu_read_lock().
4598 * But, considering usual usage, the csses should be valid objects after test.
4599 * Assuming that the caller will do some action to the child if this returns
4600 * returns true, the caller must take "child";s reference count.
4601 * If "child" is valid object and this returns true, "root" is valid, too.
4602 */
4603
4604bool css_is_ancestor(struct cgroup_subsys_state *child,
4605            const struct cgroup_subsys_state *root)
4606{
4607    struct css_id *child_id;
4608    struct css_id *root_id;
4609    bool ret = true;
4610
4611    rcu_read_lock();
4612    child_id = rcu_dereference(child->id);
4613    root_id = rcu_dereference(root->id);
4614    if (!child_id
4615        || !root_id
4616        || (child_id->depth < root_id->depth)
4617        || (child_id->stack[root_id->depth] != root_id->id))
4618        ret = false;
4619    rcu_read_unlock();
4620    return ret;
4621}
4622
4623static void __free_css_id_cb(struct rcu_head *head)
4624{
4625    struct css_id *id;
4626
4627    id = container_of(head, struct css_id, rcu_head);
4628    kfree(id);
4629}
4630
4631void free_css_id(struct cgroup_subsys *ss, struct cgroup_subsys_state *css)
4632{
4633    struct css_id *id = css->id;
4634    /* When this is called before css_id initialization, id can be NULL */
4635    if (!id)
4636        return;
4637
4638    BUG_ON(!ss->use_id);
4639
4640    rcu_assign_pointer(id->css, NULL);
4641    rcu_assign_pointer(css->id, NULL);
4642    spin_lock(&ss->id_lock);
4643    idr_remove(&ss->idr, id->id);
4644    spin_unlock(&ss->id_lock);
4645    call_rcu(&id->rcu_head, __free_css_id_cb);
4646}
4647EXPORT_SYMBOL_GPL(free_css_id);
4648
4649/*
4650 * This is called by init or create(). Then, calls to this function are
4651 * always serialized (By cgroup_mutex() at create()).
4652 */
4653
4654static struct css_id *get_new_cssid(struct cgroup_subsys *ss, int depth)
4655{
4656    struct css_id *newid;
4657    int myid, error, size;
4658
4659    BUG_ON(!ss->use_id);
4660
4661    size = sizeof(*newid) + sizeof(unsigned short) * (depth + 1);
4662    newid = kzalloc(size, GFP_KERNEL);
4663    if (!newid)
4664        return ERR_PTR(-ENOMEM);
4665    /* get id */
4666    if (unlikely(!idr_pre_get(&ss->idr, GFP_KERNEL))) {
4667        error = -ENOMEM;
4668        goto err_out;
4669    }
4670    spin_lock(&ss->id_lock);
4671    /* Don't use 0. allocates an ID of 1-65535 */
4672    error = idr_get_new_above(&ss->idr, newid, 1, &myid);
4673    spin_unlock(&ss->id_lock);
4674
4675    /* Returns error when there are no free spaces for new ID.*/
4676    if (error) {
4677        error = -ENOSPC;
4678        goto err_out;
4679    }
4680    if (myid > CSS_ID_MAX)
4681        goto remove_idr;
4682
4683    newid->id = myid;
4684    newid->depth = depth;
4685    return newid;
4686remove_idr:
4687    error = -ENOSPC;
4688    spin_lock(&ss->id_lock);
4689    idr_remove(&ss->idr, myid);
4690    spin_unlock(&ss->id_lock);
4691err_out:
4692    kfree(newid);
4693    return ERR_PTR(error);
4694
4695}
4696
4697static int __init_or_module cgroup_init_idr(struct cgroup_subsys *ss,
4698                        struct cgroup_subsys_state *rootcss)
4699{
4700    struct css_id *newid;
4701
4702    spin_lock_init(&ss->id_lock);
4703    idr_init(&ss->idr);
4704
4705    newid = get_new_cssid(ss, 0);
4706    if (IS_ERR(newid))
4707        return PTR_ERR(newid);
4708
4709    newid->stack[0] = newid->id;
4710    newid->css = rootcss;
4711    rootcss->id = newid;
4712    return 0;
4713}
4714
4715static int alloc_css_id(struct cgroup_subsys *ss, struct cgroup *parent,
4716            struct cgroup *child)
4717{
4718    int subsys_id, i, depth = 0;
4719    struct cgroup_subsys_state *parent_css, *child_css;
4720    struct css_id *child_id, *parent_id;
4721
4722    subsys_id = ss->subsys_id;
4723    parent_css = parent->subsys[subsys_id];
4724    child_css = child->subsys[subsys_id];
4725    parent_id = parent_css->id;
4726    depth = parent_id->depth + 1;
4727
4728    child_id = get_new_cssid(ss, depth);
4729    if (IS_ERR(child_id))
4730        return PTR_ERR(child_id);
4731
4732    for (i = 0; i < depth; i++)
4733        child_id->stack[i] = parent_id->stack[i];
4734    child_id->stack[depth] = child_id->id;
4735    /*
4736     * child_id->css pointer will be set after this cgroup is available
4737     * see cgroup_populate_dir()
4738     */
4739    rcu_assign_pointer(child_css->id, child_id);
4740
4741    return 0;
4742}
4743
4744/**
4745 * css_lookup - lookup css by id
4746 * @ss: cgroup subsys to be looked into.
4747 * @id: the id
4748 *
4749 * Returns pointer to cgroup_subsys_state if there is valid one with id.
4750 * NULL if not. Should be called under rcu_read_lock()
4751 */
4752struct cgroup_subsys_state *css_lookup(struct cgroup_subsys *ss, int id)
4753{
4754    struct css_id *cssid = NULL;
4755
4756    BUG_ON(!ss->use_id);
4757    cssid = idr_find(&ss->idr, id);
4758
4759    if (unlikely(!cssid))
4760        return NULL;
4761
4762    return rcu_dereference(cssid->css);
4763}
4764EXPORT_SYMBOL_GPL(css_lookup);
4765
4766/**
4767 * css_get_next - lookup next cgroup under specified hierarchy.
4768 * @ss: pointer to subsystem
4769 * @id: current position of iteration.
4770 * @root: pointer to css. search tree under this.
4771 * @foundid: position of found object.
4772 *
4773 * Search next css under the specified hierarchy of rootid. Calling under
4774 * rcu_read_lock() is necessary. Returns NULL if it reaches the end.
4775 */
4776struct cgroup_subsys_state *
4777css_get_next(struct cgroup_subsys *ss, int id,
4778         struct cgroup_subsys_state *root, int *foundid)
4779{
4780    struct cgroup_subsys_state *ret = NULL;
4781    struct css_id *tmp;
4782    int tmpid;
4783    int rootid = css_id(root);
4784    int depth = css_depth(root);
4785
4786    if (!rootid)
4787        return NULL;
4788
4789    BUG_ON(!ss->use_id);
4790    /* fill start point for scan */
4791    tmpid = id;
4792    while (1) {
4793        /*
4794         * scan next entry from bitmap(tree), tmpid is updated after
4795         * idr_get_next().
4796         */
4797        spin_lock(&ss->id_lock);
4798        tmp = idr_get_next(&ss->idr, &tmpid);
4799        spin_unlock(&ss->id_lock);
4800
4801        if (!tmp)
4802            break;
4803        if (tmp->depth >= depth && tmp->stack[depth] == rootid) {
4804            ret = rcu_dereference(tmp->css);
4805            if (ret) {
4806                *foundid = tmpid;
4807                break;
4808            }
4809        }
4810        /* continue to scan from next id */
4811        tmpid = tmpid + 1;
4812    }
4813    return ret;
4814}
4815
4816#ifdef CONFIG_CGROUP_DEBUG
4817static struct cgroup_subsys_state *debug_create(struct cgroup_subsys *ss,
4818                           struct cgroup *cont)
4819{
4820    struct cgroup_subsys_state *css = kzalloc(sizeof(*css), GFP_KERNEL);
4821
4822    if (!css)
4823        return ERR_PTR(-ENOMEM);
4824
4825    return css;
4826}
4827
4828static void debug_destroy(struct cgroup_subsys *ss, struct cgroup *cont)
4829{
4830    kfree(cont->subsys[debug_subsys_id]);
4831}
4832
4833static u64 cgroup_refcount_read(struct cgroup *cont, struct cftype *cft)
4834{
4835    return atomic_read(&cont->count);
4836}
4837
4838static u64 debug_taskcount_read(struct cgroup *cont, struct cftype *cft)
4839{
4840    return cgroup_task_count(cont);
4841}
4842
4843static u64 current_css_set_read(struct cgroup *cont, struct cftype *cft)
4844{
4845    return (u64)(unsigned long)current->cgroups;
4846}
4847
4848static u64 current_css_set_refcount_read(struct cgroup *cont,
4849                       struct cftype *cft)
4850{
4851    u64 count;
4852
4853    rcu_read_lock();
4854    count = atomic_read(&current->cgroups->refcount);
4855    rcu_read_unlock();
4856    return count;
4857}
4858
4859static int current_css_set_cg_links_read(struct cgroup *cont,
4860                     struct cftype *cft,
4861                     struct seq_file *seq)
4862{
4863    struct cg_cgroup_link *link;
4864    struct css_set *cg;
4865
4866    read_lock(&css_set_lock);
4867    rcu_read_lock();
4868    cg = rcu_dereference(current->cgroups);
4869    list_for_each_entry(link, &cg->cg_links, cg_link_list) {
4870        struct cgroup *c = link->cgrp;
4871        const char *name;
4872
4873        if (c->dentry)
4874            name = c->dentry->d_name.name;
4875        else
4876            name = "?";
4877        seq_printf(seq, "Root %d group %s\n",
4878               c->root->hierarchy_id, name);
4879    }
4880    rcu_read_unlock();
4881    read_unlock(&css_set_lock);
4882    return 0;
4883}
4884
4885#define MAX_TASKS_SHOWN_PER_CSS 25
4886static int cgroup_css_links_read(struct cgroup *cont,
4887                 struct cftype *cft,
4888                 struct seq_file *seq)
4889{
4890    struct cg_cgroup_link *link;
4891
4892    read_lock(&css_set_lock);
4893    list_for_each_entry(link, &cont->css_sets, cgrp_link_list) {
4894        struct css_set *cg = link->cg;
4895        struct task_struct *task;
4896        int count = 0;
4897        seq_printf(seq, "css_set %p\n", cg);
4898        list_for_each_entry(task, &cg->tasks, cg_list) {
4899            if (count++ > MAX_TASKS_SHOWN_PER_CSS) {
4900                seq_puts(seq, " ...\n");
4901                break;
4902            } else {
4903                seq_printf(seq, " task %d\n",
4904                       task_pid_vnr(task));
4905            }
4906        }
4907    }
4908    read_unlock(&css_set_lock);
4909    return 0;
4910}
4911
4912static u64 releasable_read(struct cgroup *cgrp, struct cftype *cft)
4913{
4914    return test_bit(CGRP_RELEASABLE, &cgrp->flags);
4915}
4916
4917static struct cftype debug_files[] = {
4918    {
4919        .name = "cgroup_refcount",
4920        .read_u64 = cgroup_refcount_read,
4921    },
4922    {
4923        .name = "taskcount",
4924        .read_u64 = debug_taskcount_read,
4925    },
4926
4927    {
4928        .name = "current_css_set",
4929        .read_u64 = current_css_set_read,
4930    },
4931
4932    {
4933        .name = "current_css_set_refcount",
4934        .read_u64 = current_css_set_refcount_read,
4935    },
4936
4937    {
4938        .name = "current_css_set_cg_links",
4939        .read_seq_string = current_css_set_cg_links_read,
4940    },
4941
4942    {
4943        .name = "cgroup_css_links",
4944        .read_seq_string = cgroup_css_links_read,
4945    },
4946
4947    {
4948        .name = "releasable",
4949        .read_u64 = releasable_read,
4950    },
4951};
4952
4953static int debug_populate(struct cgroup_subsys *ss, struct cgroup *cont)
4954{
4955    return cgroup_add_files(cont, ss, debug_files,
4956                ARRAY_SIZE(debug_files));
4957}
4958
4959struct cgroup_subsys debug_subsys = {
4960    .name = "debug",
4961    .create = debug_create,
4962    .destroy = debug_destroy,
4963    .populate = debug_populate,
4964    .subsys_id = debug_subsys_id,
4965};
4966#endif /* CONFIG_CGROUP_DEBUG */
4967

Archive Download this file



interactive