~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/kernel/cgroup.c

Version: ~ [ linux-5.5-rc7 ] ~ [ linux-5.4.13 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.97 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.166 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.210 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.210 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.140 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.81 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-3.9.11 ] ~ [ linux-3.8.13 ] ~ [ linux-3.7.10 ] ~ [ linux-3.6.11 ] ~ [ linux-3.5.7 ] ~ [ linux-3.4.113 ] ~ [ linux-3.3.8 ] ~ [ linux-3.2.102 ] ~ [ linux-3.1.10 ] ~ [ linux-3.0.101 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

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

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp