Version:  2.0.40 2.2.26 2.4.37 3.13 3.14 3.15 3.16 3.17 3.18 3.19 4.0 4.1 4.2 4.3 4.4 4.5 4.6 4.7 4.8 4.9 4.10

Linux/kernel/cgroup.c

  1 /*
  2  *  Generic process-grouping system.
  3  *
  4  *  Based originally on the cpuset system, extracted by Paul Menage
  5  *  Copyright (C) 2006 Google, Inc
  6  *
  7  *  Notifications support
  8  *  Copyright (C) 2009 Nokia Corporation
  9  *  Author: Kirill A. Shutemov
 10  *
 11  *  Copyright notices from the original cpuset code:
 12  *  --------------------------------------------------
 13  *  Copyright (C) 2003 BULL SA.
 14  *  Copyright (C) 2004-2006 Silicon Graphics, Inc.
 15  *
 16  *  Portions derived from Patrick Mochel's sysfs code.
 17  *  sysfs is Copyright (c) 2001-3 Patrick Mochel
 18  *
 19  *  2003-10-10 Written by Simon Derr.
 20  *  2003-10-22 Updates by Stephen Hemminger.
 21  *  2004 May-July Rework by Paul Jackson.
 22  *  ---------------------------------------------------
 23  *
 24  *  This file is subject to the terms and conditions of the GNU General Public
 25  *  License.  See the file COPYING in the main directory of the Linux
 26  *  distribution for more details.
 27  */
 28 
 29 #include <linux/cgroup.h>
 30 #include <linux/cred.h>
 31 #include <linux/ctype.h>
 32 #include <linux/errno.h>
 33 #include <linux/init_task.h>
 34 #include <linux/kernel.h>
 35 #include <linux/list.h>
 36 #include <linux/mm.h>
 37 #include <linux/mutex.h>
 38 #include <linux/mount.h>
 39 #include <linux/pagemap.h>
 40 #include <linux/proc_fs.h>
 41 #include <linux/rcupdate.h>
 42 #include <linux/sched.h>
 43 #include <linux/backing-dev.h>
 44 #include <linux/seq_file.h>
 45 #include <linux/slab.h>
 46 #include <linux/magic.h>
 47 #include <linux/spinlock.h>
 48 #include <linux/string.h>
 49 #include <linux/sort.h>
 50 #include <linux/kmod.h>
 51 #include <linux/module.h>
 52 #include <linux/delayacct.h>
 53 #include <linux/cgroupstats.h>
 54 #include <linux/hashtable.h>
 55 #include <linux/namei.h>
 56 #include <linux/pid_namespace.h>
 57 #include <linux/idr.h>
 58 #include <linux/vmalloc.h> /* TODO: replace with more sophisticated array */
 59 #include <linux/eventfd.h>
 60 #include <linux/poll.h>
 61 #include <linux/flex_array.h> /* used in cgroup_attach_task */
 62 #include <linux/kthread.h>
 63 #include <linux/file.h>
 64 
 65 #include <linux/atomic.h>
 66 
 67 /*
 68  * cgroup_mutex is the master lock.  Any modification to cgroup or its
 69  * hierarchy must be performed while holding it.
 70  *
 71  * cgroup_root_mutex nests inside cgroup_mutex and should be held to modify
 72  * cgroupfs_root of any cgroup hierarchy - subsys list, flags,
 73  * release_agent_path and so on.  Modifying requires both cgroup_mutex and
 74  * cgroup_root_mutex.  Readers can acquire either of the two.  This is to
 75  * break the following locking order cycle.
 76  *
 77  *  A. cgroup_mutex -> cred_guard_mutex -> s_type->i_mutex_key -> namespace_sem
 78  *  B. namespace_sem -> cgroup_mutex
 79  *
 80  * B happens only through cgroup_show_options() and using cgroup_root_mutex
 81  * breaks it.
 82  */
 83 #ifdef CONFIG_PROVE_RCU
 84 DEFINE_MUTEX(cgroup_mutex);
 85 EXPORT_SYMBOL_GPL(cgroup_mutex);        /* only for lockdep */
 86 #else
 87 static DEFINE_MUTEX(cgroup_mutex);
 88 #endif
 89 
 90 static DEFINE_MUTEX(cgroup_root_mutex);
 91 
 92 /*
 93  * cgroup destruction makes heavy use of work items and there can be a lot
 94  * of concurrent destructions.  Use a separate workqueue so that cgroup
 95  * destruction work items don't end up filling up max_active of system_wq
 96  * which may lead to deadlock.
 97  */
 98 static struct workqueue_struct *cgroup_destroy_wq;
 99 
100 /*
101  * Generate an array of cgroup subsystem pointers. At boot time, this is
102  * populated with the built in subsystems, and modular subsystems are
103  * registered after that. The mutable section of this array is protected by
104  * cgroup_mutex.
105  */
106 #define SUBSYS(_x) [_x ## _subsys_id] = &_x ## _subsys,
107 #define IS_SUBSYS_ENABLED(option) IS_BUILTIN(option)
108 static struct cgroup_subsys *cgroup_subsys[CGROUP_SUBSYS_COUNT] = {
109 #include <linux/cgroup_subsys.h>
110 };
111 
112 /*
113  * The dummy hierarchy, reserved for the subsystems that are otherwise
114  * unattached - it never has more than a single cgroup, and all tasks are
115  * part of that cgroup.
116  */
117 static struct cgroupfs_root cgroup_dummy_root;
118 
119 /* dummy_top is a shorthand for the dummy hierarchy's top cgroup */
120 static struct cgroup * const cgroup_dummy_top = &cgroup_dummy_root.top_cgroup;
121 
122 /*
123  * cgroupfs file entry, pointed to from leaf dentry->d_fsdata.
124  */
125 struct cfent {
126         struct list_head                node;
127         struct dentry                   *dentry;
128         struct cftype                   *type;
129         struct cgroup_subsys_state      *css;
130 
131         /* file xattrs */
132         struct simple_xattrs            xattrs;
133 };
134 
135 /*
136  * cgroup_event represents events which userspace want to receive.
137  */
138 struct cgroup_event {
139         /*
140          * css which the event belongs to.
141          */
142         struct cgroup_subsys_state *css;
143         /*
144          * Control file which the event associated.
145          */
146         struct cftype *cft;
147         /*
148          * eventfd to signal userspace about the event.
149          */
150         struct eventfd_ctx *eventfd;
151         /*
152          * Each of these stored in a list by the cgroup.
153          */
154         struct list_head list;
155         /*
156          * All fields below needed to unregister event when
157          * userspace closes eventfd.
158          */
159         poll_table pt;
160         wait_queue_head_t *wqh;
161         wait_queue_t wait;
162         struct work_struct remove;
163 };
164 
165 /* The list of hierarchy roots */
166 
167 static LIST_HEAD(cgroup_roots);
168 static int cgroup_root_count;
169 
170 /*
171  * Hierarchy ID allocation and mapping.  It follows the same exclusion
172  * rules as other root ops - both cgroup_mutex and cgroup_root_mutex for
173  * writes, either for reads.
174  */
175 static DEFINE_IDR(cgroup_hierarchy_idr);
176 
177 static struct cgroup_name root_cgroup_name = { .name = "/" };
178 
179 /*
180  * Assign a monotonically increasing serial number to cgroups.  It
181  * guarantees cgroups with bigger numbers are newer than those with smaller
182  * numbers.  Also, as cgroups are always appended to the parent's
183  * ->children list, it guarantees that sibling cgroups are always sorted in
184  * the ascending serial number order on the list.  Protected by
185  * cgroup_mutex.
186  */
187 static u64 cgroup_serial_nr_next = 1;
188 
189 /* This flag indicates whether tasks in the fork and exit paths should
190  * check for fork/exit handlers to call. This avoids us having to do
191  * extra work in the fork/exit path if none of the subsystems need to
192  * be called.
193  */
194 static int need_forkexit_callback __read_mostly;
195 
196 static struct cftype cgroup_base_files[];
197 
198 static void cgroup_destroy_css_killed(struct cgroup *cgrp);
199 static int cgroup_destroy_locked(struct cgroup *cgrp);
200 static int cgroup_addrm_files(struct cgroup *cgrp, struct cftype cfts[],
201                               bool is_add);
202 static int cgroup_file_release(struct inode *inode, struct file *file);
203 
204 /**
205  * cgroup_css - obtain a cgroup's css for the specified subsystem
206  * @cgrp: the cgroup of interest
207  * @ss: the subsystem of interest (%NULL returns the dummy_css)
208  *
209  * Return @cgrp's css (cgroup_subsys_state) associated with @ss.  This
210  * function must be called either under cgroup_mutex or rcu_read_lock() and
211  * the caller is responsible for pinning the returned css if it wants to
212  * keep accessing it outside the said locks.  This function may return
213  * %NULL if @cgrp doesn't have @subsys_id enabled.
214  */
215 static struct cgroup_subsys_state *cgroup_css(struct cgroup *cgrp,
216                                               struct cgroup_subsys *ss)
217 {
218         if (ss)
219                 return rcu_dereference_check(cgrp->subsys[ss->subsys_id],
220                                              lockdep_is_held(&cgroup_mutex));
221         else
222                 return &cgrp->dummy_css;
223 }
224 
225 /* convenient tests for these bits */
226 static inline bool cgroup_is_dead(const struct cgroup *cgrp)
227 {
228         return test_bit(CGRP_DEAD, &cgrp->flags);
229 }
230 
231 /**
232  * cgroup_is_descendant - test ancestry
233  * @cgrp: the cgroup to be tested
234  * @ancestor: possible ancestor of @cgrp
235  *
236  * Test whether @cgrp is a descendant of @ancestor.  It also returns %true
237  * if @cgrp == @ancestor.  This function is safe to call as long as @cgrp
238  * and @ancestor are accessible.
239  */
240 bool cgroup_is_descendant(struct cgroup *cgrp, struct cgroup *ancestor)
241 {
242         while (cgrp) {
243                 if (cgrp == ancestor)
244                         return true;
245                 cgrp = cgrp->parent;
246         }
247         return false;
248 }
249 EXPORT_SYMBOL_GPL(cgroup_is_descendant);
250 
251 static int cgroup_is_releasable(const struct cgroup *cgrp)
252 {
253         const int bits =
254                 (1 << CGRP_RELEASABLE) |
255                 (1 << CGRP_NOTIFY_ON_RELEASE);
256         return (cgrp->flags & bits) == bits;
257 }
258 
259 static int notify_on_release(const struct cgroup *cgrp)
260 {
261         return test_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
262 }
263 
264 /**
265  * for_each_subsys - iterate all loaded cgroup subsystems
266  * @ss: the iteration cursor
267  * @i: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end
268  *
269  * Should be called under cgroup_mutex.
270  */
271 #define for_each_subsys(ss, i)                                          \
272         for ((i) = 0; (i) < CGROUP_SUBSYS_COUNT; (i)++)                 \
273                 if (({ lockdep_assert_held(&cgroup_mutex);              \
274                        !((ss) = cgroup_subsys[i]); })) { }              \
275                 else
276 
277 /**
278  * for_each_builtin_subsys - iterate all built-in cgroup subsystems
279  * @ss: the iteration cursor
280  * @i: the index of @ss, CGROUP_BUILTIN_SUBSYS_COUNT after reaching the end
281  *
282  * Bulit-in subsystems are always present and iteration itself doesn't
283  * require any synchronization.
284  */
285 #define for_each_builtin_subsys(ss, i)                                  \
286         for ((i) = 0; (i) < CGROUP_BUILTIN_SUBSYS_COUNT &&              \
287              (((ss) = cgroup_subsys[i]) || true); (i)++)
288 
289 /* iterate each subsystem attached to a hierarchy */
290 #define for_each_root_subsys(root, ss)                                  \
291         list_for_each_entry((ss), &(root)->subsys_list, sibling)
292 
293 /* iterate across the active hierarchies */
294 #define for_each_active_root(root)                                      \
295         list_for_each_entry((root), &cgroup_roots, root_list)
296 
297 static inline struct cgroup *__d_cgrp(struct dentry *dentry)
298 {
299         return dentry->d_fsdata;
300 }
301 
302 static inline struct cfent *__d_cfe(struct dentry *dentry)
303 {
304         return dentry->d_fsdata;
305 }
306 
307 static inline struct cftype *__d_cft(struct dentry *dentry)
308 {
309         return __d_cfe(dentry)->type;
310 }
311 
312 /**
313  * cgroup_lock_live_group - take cgroup_mutex and check that cgrp is alive.
314  * @cgrp: the cgroup to be checked for liveness
315  *
316  * On success, returns true; the mutex should be later unlocked.  On
317  * failure returns false with no lock held.
318  */
319 static bool cgroup_lock_live_group(struct cgroup *cgrp)
320 {
321         mutex_lock(&cgroup_mutex);
322         if (cgroup_is_dead(cgrp)) {
323                 mutex_unlock(&cgroup_mutex);
324                 return false;
325         }
326         return true;
327 }
328 
329 /* the list of cgroups eligible for automatic release. Protected by
330  * release_list_lock */
331 static LIST_HEAD(release_list);
332 static DEFINE_RAW_SPINLOCK(release_list_lock);
333 static void cgroup_release_agent(struct work_struct *work);
334 static DECLARE_WORK(release_agent_work, cgroup_release_agent);
335 static void check_for_release(struct cgroup *cgrp);
336 
337 /*
338  * A cgroup can be associated with multiple css_sets as different tasks may
339  * belong to different cgroups on different hierarchies.  In the other
340  * direction, a css_set is naturally associated with multiple cgroups.
341  * This M:N relationship is represented by the following link structure
342  * which exists for each association and allows traversing the associations
343  * from both sides.
344  */
345 struct cgrp_cset_link {
346         /* the cgroup and css_set this link associates */
347         struct cgroup           *cgrp;
348         struct css_set          *cset;
349 
350         /* list of cgrp_cset_links anchored at cgrp->cset_links */
351         struct list_head        cset_link;
352 
353         /* list of cgrp_cset_links anchored at css_set->cgrp_links */
354         struct list_head        cgrp_link;
355 };
356 
357 /* The default css_set - used by init and its children prior to any
358  * hierarchies being mounted. It contains a pointer to the root state
359  * for each subsystem. Also used to anchor the list of css_sets. Not
360  * reference-counted, to improve performance when child cgroups
361  * haven't been created.
362  */
363 
364 static struct css_set init_css_set;
365 static struct cgrp_cset_link init_cgrp_cset_link;
366 
367 /*
368  * css_set_lock protects the list of css_set objects, and the chain of
369  * tasks off each css_set.  Nests outside task->alloc_lock due to
370  * css_task_iter_start().
371  */
372 static DEFINE_RWLOCK(css_set_lock);
373 static int css_set_count;
374 
375 /*
376  * hash table for cgroup groups. This improves the performance to find
377  * an existing css_set. This hash doesn't (currently) take into
378  * account cgroups in empty hierarchies.
379  */
380 #define CSS_SET_HASH_BITS       7
381 static DEFINE_HASHTABLE(css_set_table, CSS_SET_HASH_BITS);
382 
383 static unsigned long css_set_hash(struct cgroup_subsys_state *css[])
384 {
385         unsigned long key = 0UL;
386         struct cgroup_subsys *ss;
387         int i;
388 
389         for_each_subsys(ss, i)
390                 key += (unsigned long)css[i];
391         key = (key >> 16) ^ key;
392 
393         return key;
394 }
395 
396 /*
397  * We don't maintain the lists running through each css_set to its task
398  * until after the first call to css_task_iter_start().  This reduces the
399  * fork()/exit() overhead for people who have cgroups compiled into their
400  * kernel but not actually in use.
401  */
402 static int use_task_css_set_links __read_mostly;
403 
404 static void __put_css_set(struct css_set *cset, int taskexit)
405 {
406         struct cgrp_cset_link *link, *tmp_link;
407 
408         /*
409          * Ensure that the refcount doesn't hit zero while any readers
410          * can see it. Similar to atomic_dec_and_lock(), but for an
411          * rwlock
412          */
413         if (atomic_add_unless(&cset->refcount, -1, 1))
414                 return;
415         write_lock(&css_set_lock);
416         if (!atomic_dec_and_test(&cset->refcount)) {
417                 write_unlock(&css_set_lock);
418                 return;
419         }
420 
421         /* This css_set is dead. unlink it and release cgroup refcounts */
422         hash_del(&cset->hlist);
423         css_set_count--;
424 
425         list_for_each_entry_safe(link, tmp_link, &cset->cgrp_links, cgrp_link) {
426                 struct cgroup *cgrp = link->cgrp;
427 
428                 list_del(&link->cset_link);
429                 list_del(&link->cgrp_link);
430 
431                 /* @cgrp can't go away while we're holding css_set_lock */
432                 if (list_empty(&cgrp->cset_links) && notify_on_release(cgrp)) {
433                         if (taskexit)
434                                 set_bit(CGRP_RELEASABLE, &cgrp->flags);
435                         check_for_release(cgrp);
436                 }
437 
438                 kfree(link);
439         }
440 
441         write_unlock(&css_set_lock);
442         kfree_rcu(cset, rcu_head);
443 }
444 
445 /*
446  * refcounted get/put for css_set objects
447  */
448 static inline void get_css_set(struct css_set *cset)
449 {
450         atomic_inc(&cset->refcount);
451 }
452 
453 static inline void put_css_set(struct css_set *cset)
454 {
455         __put_css_set(cset, 0);
456 }
457 
458 static inline void put_css_set_taskexit(struct css_set *cset)
459 {
460         __put_css_set(cset, 1);
461 }
462 
463 /**
464  * compare_css_sets - helper function for find_existing_css_set().
465  * @cset: candidate css_set being tested
466  * @old_cset: existing css_set for a task
467  * @new_cgrp: cgroup that's being entered by the task
468  * @template: desired set of css pointers in css_set (pre-calculated)
469  *
470  * Returns true if "cset" matches "old_cset" except for the hierarchy
471  * which "new_cgrp" belongs to, for which it should match "new_cgrp".
472  */
473 static bool compare_css_sets(struct css_set *cset,
474                              struct css_set *old_cset,
475                              struct cgroup *new_cgrp,
476                              struct cgroup_subsys_state *template[])
477 {
478         struct list_head *l1, *l2;
479 
480         if (memcmp(template, cset->subsys, sizeof(cset->subsys))) {
481                 /* Not all subsystems matched */
482                 return false;
483         }
484 
485         /*
486          * Compare cgroup pointers in order to distinguish between
487          * different cgroups in heirarchies with no subsystems. We
488          * could get by with just this check alone (and skip the
489          * memcmp above) but on most setups the memcmp check will
490          * avoid the need for this more expensive check on almost all
491          * candidates.
492          */
493 
494         l1 = &cset->cgrp_links;
495         l2 = &old_cset->cgrp_links;
496         while (1) {
497                 struct cgrp_cset_link *link1, *link2;
498                 struct cgroup *cgrp1, *cgrp2;
499 
500                 l1 = l1->next;
501                 l2 = l2->next;
502                 /* See if we reached the end - both lists are equal length. */
503                 if (l1 == &cset->cgrp_links) {
504                         BUG_ON(l2 != &old_cset->cgrp_links);
505                         break;
506                 } else {
507                         BUG_ON(l2 == &old_cset->cgrp_links);
508                 }
509                 /* Locate the cgroups associated with these links. */
510                 link1 = list_entry(l1, struct cgrp_cset_link, cgrp_link);
511                 link2 = list_entry(l2, struct cgrp_cset_link, cgrp_link);
512                 cgrp1 = link1->cgrp;
513                 cgrp2 = link2->cgrp;
514                 /* Hierarchies should be linked in the same order. */
515                 BUG_ON(cgrp1->root != cgrp2->root);
516 
517                 /*
518                  * If this hierarchy is the hierarchy of the cgroup
519                  * that's changing, then we need to check that this
520                  * css_set points to the new cgroup; if it's any other
521                  * hierarchy, then this css_set should point to the
522                  * same cgroup as the old css_set.
523                  */
524                 if (cgrp1->root == new_cgrp->root) {
525                         if (cgrp1 != new_cgrp)
526                                 return false;
527                 } else {
528                         if (cgrp1 != cgrp2)
529                                 return false;
530                 }
531         }
532         return true;
533 }
534 
535 /**
536  * find_existing_css_set - init css array and find the matching css_set
537  * @old_cset: the css_set that we're using before the cgroup transition
538  * @cgrp: the cgroup that we're moving into
539  * @template: out param for the new set of csses, should be clear on entry
540  */
541 static struct css_set *find_existing_css_set(struct css_set *old_cset,
542                                         struct cgroup *cgrp,
543                                         struct cgroup_subsys_state *template[])
544 {
545         struct cgroupfs_root *root = cgrp->root;
546         struct cgroup_subsys *ss;
547         struct css_set *cset;
548         unsigned long key;
549         int i;
550 
551         /*
552          * Build the set of subsystem state objects that we want to see in the
553          * new css_set. while subsystems can change globally, the entries here
554          * won't change, so no need for locking.
555          */
556         for_each_subsys(ss, i) {
557                 if (root->subsys_mask & (1UL << i)) {
558                         /* Subsystem is in this hierarchy. So we want
559                          * the subsystem state from the new
560                          * cgroup */
561                         template[i] = cgroup_css(cgrp, ss);
562                 } else {
563                         /* Subsystem is not in this hierarchy, so we
564                          * don't want to change the subsystem state */
565                         template[i] = old_cset->subsys[i];
566                 }
567         }
568 
569         key = css_set_hash(template);
570         hash_for_each_possible(css_set_table, cset, hlist, key) {
571                 if (!compare_css_sets(cset, old_cset, cgrp, template))
572                         continue;
573 
574                 /* This css_set matches what we need */
575                 return cset;
576         }
577 
578         /* No existing cgroup group matched */
579         return NULL;
580 }
581 
582 static void free_cgrp_cset_links(struct list_head *links_to_free)
583 {
584         struct cgrp_cset_link *link, *tmp_link;
585 
586         list_for_each_entry_safe(link, tmp_link, links_to_free, cset_link) {
587                 list_del(&link->cset_link);
588                 kfree(link);
589         }
590 }
591 
592 /**
593  * allocate_cgrp_cset_links - allocate cgrp_cset_links
594  * @count: the number of links to allocate
595  * @tmp_links: list_head the allocated links are put on
596  *
597  * Allocate @count cgrp_cset_link structures and chain them on @tmp_links
598  * through ->cset_link.  Returns 0 on success or -errno.
599  */
600 static int allocate_cgrp_cset_links(int count, struct list_head *tmp_links)
601 {
602         struct cgrp_cset_link *link;
603         int i;
604 
605         INIT_LIST_HEAD(tmp_links);
606 
607         for (i = 0; i < count; i++) {
608                 link = kzalloc(sizeof(*link), GFP_KERNEL);
609                 if (!link) {
610                         free_cgrp_cset_links(tmp_links);
611                         return -ENOMEM;
612                 }
613                 list_add(&link->cset_link, tmp_links);
614         }
615         return 0;
616 }
617 
618 /**
619  * link_css_set - a helper function to link a css_set to a cgroup
620  * @tmp_links: cgrp_cset_link objects allocated by allocate_cgrp_cset_links()
621  * @cset: the css_set to be linked
622  * @cgrp: the destination cgroup
623  */
624 static void link_css_set(struct list_head *tmp_links, struct css_set *cset,
625                          struct cgroup *cgrp)
626 {
627         struct cgrp_cset_link *link;
628 
629         BUG_ON(list_empty(tmp_links));
630         link = list_first_entry(tmp_links, struct cgrp_cset_link, cset_link);
631         link->cset = cset;
632         link->cgrp = cgrp;
633         list_move(&link->cset_link, &cgrp->cset_links);
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->cgrp_link, &cset->cgrp_links);
639 }
640 
641 /**
642  * find_css_set - return a new css_set with one cgroup updated
643  * @old_cset: the baseline css_set
644  * @cgrp: the cgroup to be updated
645  *
646  * Return a new css_set that's equivalent to @old_cset, but with @cgrp
647  * substituted into the appropriate hierarchy.
648  */
649 static struct css_set *find_css_set(struct css_set *old_cset,
650                                     struct cgroup *cgrp)
651 {
652         struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT] = { };
653         struct css_set *cset;
654         struct list_head tmp_links;
655         struct cgrp_cset_link *link;
656         unsigned long key;
657 
658         lockdep_assert_held(&cgroup_mutex);
659 
660         /* First see if we already have a cgroup group that matches
661          * the desired set */
662         read_lock(&css_set_lock);
663         cset = find_existing_css_set(old_cset, cgrp, template);
664         if (cset)
665                 get_css_set(cset);
666         read_unlock(&css_set_lock);
667 
668         if (cset)
669                 return cset;
670 
671         cset = kzalloc(sizeof(*cset), GFP_KERNEL);
672         if (!cset)
673                 return NULL;
674 
675         /* Allocate all the cgrp_cset_link objects that we'll need */
676         if (allocate_cgrp_cset_links(cgroup_root_count, &tmp_links) < 0) {
677                 kfree(cset);
678                 return NULL;
679         }
680 
681         atomic_set(&cset->refcount, 1);
682         INIT_LIST_HEAD(&cset->cgrp_links);
683         INIT_LIST_HEAD(&cset->tasks);
684         INIT_HLIST_NODE(&cset->hlist);
685 
686         /* Copy the set of subsystem state objects generated in
687          * find_existing_css_set() */
688         memcpy(cset->subsys, template, sizeof(cset->subsys));
689 
690         write_lock(&css_set_lock);
691         /* Add reference counts and links from the new css_set. */
692         list_for_each_entry(link, &old_cset->cgrp_links, cgrp_link) {
693                 struct cgroup *c = link->cgrp;
694 
695                 if (c->root == cgrp->root)
696                         c = cgrp;
697                 link_css_set(&tmp_links, cset, c);
698         }
699 
700         BUG_ON(!list_empty(&tmp_links));
701 
702         css_set_count++;
703 
704         /* Add this cgroup group to the hash table */
705         key = css_set_hash(cset->subsys);
706         hash_add(css_set_table, &cset->hlist, key);
707 
708         write_unlock(&css_set_lock);
709 
710         return cset;
711 }
712 
713 /*
714  * Return the cgroup for "task" from the given hierarchy. Must be
715  * called with cgroup_mutex held.
716  */
717 static struct cgroup *task_cgroup_from_root(struct task_struct *task,
718                                             struct cgroupfs_root *root)
719 {
720         struct css_set *cset;
721         struct cgroup *res = NULL;
722 
723         BUG_ON(!mutex_is_locked(&cgroup_mutex));
724         read_lock(&css_set_lock);
725         /*
726          * No need to lock the task - since we hold cgroup_mutex the
727          * task can't change groups, so the only thing that can happen
728          * is that it exits and its css is set back to init_css_set.
729          */
730         cset = task_css_set(task);
731         if (cset == &init_css_set) {
732                 res = &root->top_cgroup;
733         } else {
734                 struct cgrp_cset_link *link;
735 
736                 list_for_each_entry(link, &cset->cgrp_links, cgrp_link) {
737                         struct cgroup *c = link->cgrp;
738 
739                         if (c->root == root) {
740                                 res = c;
741                                 break;
742                         }
743                 }
744         }
745         read_unlock(&css_set_lock);
746         BUG_ON(!res);
747         return res;
748 }
749 
750 /*
751  * There is one global cgroup mutex. We also require taking
752  * task_lock() when dereferencing a task's cgroup subsys pointers.
753  * See "The task_lock() exception", at the end of this comment.
754  *
755  * A task must hold cgroup_mutex to modify cgroups.
756  *
757  * Any task can increment and decrement the count field without lock.
758  * So in general, code holding cgroup_mutex can't rely on the count
759  * field not changing.  However, if the count goes to zero, then only
760  * cgroup_attach_task() can increment it again.  Because a count of zero
761  * means that no tasks are currently attached, therefore there is no
762  * way a task attached to that cgroup can fork (the other way to
763  * increment the count).  So code holding cgroup_mutex can safely
764  * assume that if the count is zero, it will stay zero. Similarly, if
765  * a task holds cgroup_mutex on a cgroup with zero count, it
766  * knows that the cgroup won't be removed, as cgroup_rmdir()
767  * needs that mutex.
768  *
769  * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't
770  * (usually) take cgroup_mutex.  These are the two most performance
771  * critical pieces of code here.  The exception occurs on cgroup_exit(),
772  * when a task in a notify_on_release cgroup exits.  Then cgroup_mutex
773  * is taken, and if the cgroup count is zero, a usermode call made
774  * to the release agent with the name of the cgroup (path relative to
775  * the root of cgroup file system) as the argument.
776  *
777  * A cgroup can only be deleted if both its 'count' of using tasks
778  * is zero, and its list of 'children' cgroups is empty.  Since all
779  * tasks in the system use _some_ cgroup, and since there is always at
780  * least one task in the system (init, pid == 1), therefore, top_cgroup
781  * always has either children cgroups and/or using tasks.  So we don't
782  * need a special hack to ensure that top_cgroup cannot be deleted.
783  *
784  *      The task_lock() exception
785  *
786  * The need for this exception arises from the action of
787  * cgroup_attach_task(), which overwrites one task's cgroup pointer with
788  * another.  It does so using cgroup_mutex, however there are
789  * several performance critical places that need to reference
790  * task->cgroup without the expense of grabbing a system global
791  * mutex.  Therefore except as noted below, when dereferencing or, as
792  * in cgroup_attach_task(), modifying a task's cgroup pointer we use
793  * task_lock(), which acts on a spinlock (task->alloc_lock) already in
794  * the task_struct routinely used for such matters.
795  *
796  * P.S.  One more locking exception.  RCU is used to guard the
797  * update of a tasks cgroup pointer by cgroup_attach_task()
798  */
799 
800 /*
801  * A couple of forward declarations required, due to cyclic reference loop:
802  * cgroup_mkdir -> cgroup_create -> cgroup_populate_dir ->
803  * cgroup_add_file -> cgroup_create_file -> cgroup_dir_inode_operations
804  * -> cgroup_mkdir.
805  */
806 
807 static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode);
808 static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry);
809 static int cgroup_populate_dir(struct cgroup *cgrp, unsigned long subsys_mask);
810 static const struct inode_operations cgroup_dir_inode_operations;
811 static const struct file_operations proc_cgroupstats_operations;
812 
813 static struct backing_dev_info cgroup_backing_dev_info = {
814         .name           = "cgroup",
815         .capabilities   = BDI_CAP_NO_ACCT_AND_WRITEBACK,
816 };
817 
818 static struct inode *cgroup_new_inode(umode_t mode, struct super_block *sb)
819 {
820         struct inode *inode = new_inode(sb);
821 
822         if (inode) {
823                 inode->i_ino = get_next_ino();
824                 inode->i_mode = mode;
825                 inode->i_uid = current_fsuid();
826                 inode->i_gid = current_fsgid();
827                 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
828                 inode->i_mapping->backing_dev_info = &cgroup_backing_dev_info;
829         }
830         return inode;
831 }
832 
833 static struct cgroup_name *cgroup_alloc_name(struct dentry *dentry)
834 {
835         struct cgroup_name *name;
836 
837         name = kmalloc(sizeof(*name) + dentry->d_name.len + 1, GFP_KERNEL);
838         if (!name)
839                 return NULL;
840         strcpy(name->name, dentry->d_name.name);
841         return name;
842 }
843 
844 static void cgroup_free_fn(struct work_struct *work)
845 {
846         struct cgroup *cgrp = container_of(work, struct cgroup, destroy_work);
847 
848         mutex_lock(&cgroup_mutex);
849         cgrp->root->number_of_cgroups--;
850         mutex_unlock(&cgroup_mutex);
851 
852         /*
853          * We get a ref to the parent's dentry, and put the ref when
854          * this cgroup is being freed, so it's guaranteed that the
855          * parent won't be destroyed before its children.
856          */
857         dput(cgrp->parent->dentry);
858 
859         /*
860          * Drop the active superblock reference that we took when we
861          * created the cgroup. This will free cgrp->root, if we are
862          * holding the last reference to @sb.
863          */
864         deactivate_super(cgrp->root->sb);
865 
866         /*
867          * if we're getting rid of the cgroup, refcount should ensure
868          * that there are no pidlists left.
869          */
870         BUG_ON(!list_empty(&cgrp->pidlists));
871 
872         simple_xattrs_free(&cgrp->xattrs);
873 
874         kfree(rcu_dereference_raw(cgrp->name));
875         kfree(cgrp);
876 }
877 
878 static void cgroup_free_rcu(struct rcu_head *head)
879 {
880         struct cgroup *cgrp = container_of(head, struct cgroup, rcu_head);
881 
882         INIT_WORK(&cgrp->destroy_work, cgroup_free_fn);
883         queue_work(cgroup_destroy_wq, &cgrp->destroy_work);
884 }
885 
886 static void cgroup_diput(struct dentry *dentry, struct inode *inode)
887 {
888         /* is dentry a directory ? if so, kfree() associated cgroup */
889         if (S_ISDIR(inode->i_mode)) {
890                 struct cgroup *cgrp = dentry->d_fsdata;
891 
892                 BUG_ON(!(cgroup_is_dead(cgrp)));
893 
894                 /*
895                  * XXX: cgrp->id is only used to look up css's.  As cgroup
896                  * and css's lifetimes will be decoupled, it should be made
897                  * per-subsystem and moved to css->id so that lookups are
898                  * successful until the target css is released.
899                  */
900                 idr_remove(&cgrp->root->cgroup_idr, cgrp->id);
901                 cgrp->id = -1;
902 
903                 call_rcu(&cgrp->rcu_head, cgroup_free_rcu);
904         } else {
905                 struct cfent *cfe = __d_cfe(dentry);
906                 struct cgroup *cgrp = dentry->d_parent->d_fsdata;
907 
908                 WARN_ONCE(!list_empty(&cfe->node) &&
909                           cgrp != &cgrp->root->top_cgroup,
910                           "cfe still linked for %s\n", cfe->type->name);
911                 simple_xattrs_free(&cfe->xattrs);
912                 kfree(cfe);
913         }
914         iput(inode);
915 }
916 
917 static void remove_dir(struct dentry *d)
918 {
919         struct dentry *parent = dget(d->d_parent);
920 
921         d_delete(d);
922         simple_rmdir(parent->d_inode, d);
923         dput(parent);
924 }
925 
926 static void cgroup_rm_file(struct cgroup *cgrp, const struct cftype *cft)
927 {
928         struct cfent *cfe;
929 
930         lockdep_assert_held(&cgrp->dentry->d_inode->i_mutex);
931         lockdep_assert_held(&cgroup_mutex);
932 
933         /*
934          * If we're doing cleanup due to failure of cgroup_create(),
935          * the corresponding @cfe may not exist.
936          */
937         list_for_each_entry(cfe, &cgrp->files, node) {
938                 struct dentry *d = cfe->dentry;
939 
940                 if (cft && cfe->type != cft)
941                         continue;
942 
943                 dget(d);
944                 d_delete(d);
945                 simple_unlink(cgrp->dentry->d_inode, d);
946                 list_del_init(&cfe->node);
947                 dput(d);
948 
949                 break;
950         }
951 }
952 
953 /**
954  * cgroup_clear_dir - remove subsys files in a cgroup directory
955  * @cgrp: target cgroup
956  * @subsys_mask: mask of the subsystem ids whose files should be removed
957  */
958 static void cgroup_clear_dir(struct cgroup *cgrp, unsigned long subsys_mask)
959 {
960         struct cgroup_subsys *ss;
961         int i;
962 
963         for_each_subsys(ss, i) {
964                 struct cftype_set *set;
965 
966                 if (!test_bit(i, &subsys_mask))
967                         continue;
968                 list_for_each_entry(set, &ss->cftsets, node)
969                         cgroup_addrm_files(cgrp, set->cfts, false);
970         }
971 }
972 
973 /*
974  * NOTE : the dentry must have been dget()'ed
975  */
976 static void cgroup_d_remove_dir(struct dentry *dentry)
977 {
978         struct dentry *parent;
979 
980         parent = dentry->d_parent;
981         spin_lock(&parent->d_lock);
982         spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
983         list_del_init(&dentry->d_u.d_child);
984         spin_unlock(&dentry->d_lock);
985         spin_unlock(&parent->d_lock);
986         remove_dir(dentry);
987 }
988 
989 /*
990  * Call with cgroup_mutex held. Drops reference counts on modules, including
991  * any duplicate ones that parse_cgroupfs_options took. If this function
992  * returns an error, no reference counts are touched.
993  */
994 static int rebind_subsystems(struct cgroupfs_root *root,
995                              unsigned long added_mask, unsigned removed_mask)
996 {
997         struct cgroup *cgrp = &root->top_cgroup;
998         struct cgroup_subsys *ss;
999         unsigned long pinned = 0;
1000         int i, ret;
1001 
1002         BUG_ON(!mutex_is_locked(&cgroup_mutex));
1003         BUG_ON(!mutex_is_locked(&cgroup_root_mutex));
1004 
1005         /* Check that any added subsystems are currently free */
1006         for_each_subsys(ss, i) {
1007                 if (!(added_mask & (1 << i)))
1008                         continue;
1009 
1010                 /* is the subsystem mounted elsewhere? */
1011                 if (ss->root != &cgroup_dummy_root) {
1012                         ret = -EBUSY;
1013                         goto out_put;
1014                 }
1015 
1016                 /* pin the module */
1017                 if (!try_module_get(ss->module)) {
1018                         ret = -ENOENT;
1019                         goto out_put;
1020                 }
1021                 pinned |= 1 << i;
1022         }
1023 
1024         /* subsys could be missing if unloaded between parsing and here */
1025         if (added_mask != pinned) {
1026                 ret = -ENOENT;
1027                 goto out_put;
1028         }
1029 
1030         ret = cgroup_populate_dir(cgrp, added_mask);
1031         if (ret)
1032                 goto out_put;
1033 
1034         /*
1035          * Nothing can fail from this point on.  Remove files for the
1036          * removed subsystems and rebind each subsystem.
1037          */
1038         cgroup_clear_dir(cgrp, removed_mask);
1039 
1040         for_each_subsys(ss, i) {
1041                 unsigned long bit = 1UL << i;
1042 
1043                 if (bit & added_mask) {
1044                         /* We're binding this subsystem to this hierarchy */
1045                         BUG_ON(cgroup_css(cgrp, ss));
1046                         BUG_ON(!cgroup_css(cgroup_dummy_top, ss));
1047                         BUG_ON(cgroup_css(cgroup_dummy_top, ss)->cgroup != cgroup_dummy_top);
1048 
1049                         rcu_assign_pointer(cgrp->subsys[i],
1050                                            cgroup_css(cgroup_dummy_top, ss));
1051                         cgroup_css(cgrp, ss)->cgroup = cgrp;
1052 
1053                         list_move(&ss->sibling, &root->subsys_list);
1054                         ss->root = root;
1055                         if (ss->bind)
1056                                 ss->bind(cgroup_css(cgrp, ss));
1057 
1058                         /* refcount was already taken, and we're keeping it */
1059                         root->subsys_mask |= bit;
1060                 } else if (bit & removed_mask) {
1061                         /* We're removing this subsystem */
1062                         BUG_ON(cgroup_css(cgrp, ss) != cgroup_css(cgroup_dummy_top, ss));
1063                         BUG_ON(cgroup_css(cgrp, ss)->cgroup != cgrp);
1064 
1065                         if (ss->bind)
1066                                 ss->bind(cgroup_css(cgroup_dummy_top, ss));
1067 
1068                         cgroup_css(cgroup_dummy_top, ss)->cgroup = cgroup_dummy_top;
1069                         RCU_INIT_POINTER(cgrp->subsys[i], NULL);
1070 
1071                         cgroup_subsys[i]->root = &cgroup_dummy_root;
1072                         list_move(&ss->sibling, &cgroup_dummy_root.subsys_list);
1073 
1074                         /* subsystem is now free - drop reference on module */
1075                         module_put(ss->module);
1076                         root->subsys_mask &= ~bit;
1077                 }
1078         }
1079 
1080         /*
1081          * Mark @root has finished binding subsystems.  @root->subsys_mask
1082          * now matches the bound subsystems.
1083          */
1084         root->flags |= CGRP_ROOT_SUBSYS_BOUND;
1085 
1086         return 0;
1087 
1088 out_put:
1089         for_each_subsys(ss, i)
1090                 if (pinned & (1 << i))
1091                         module_put(ss->module);
1092         return ret;
1093 }
1094 
1095 static int cgroup_show_options(struct seq_file *seq, struct dentry *dentry)
1096 {
1097         struct cgroupfs_root *root = dentry->d_sb->s_fs_info;
1098         struct cgroup_subsys *ss;
1099 
1100         mutex_lock(&cgroup_root_mutex);
1101         for_each_root_subsys(root, ss)
1102                 seq_printf(seq, ",%s", ss->name);
1103         if (root->flags & CGRP_ROOT_SANE_BEHAVIOR)
1104                 seq_puts(seq, ",sane_behavior");
1105         if (root->flags & CGRP_ROOT_NOPREFIX)
1106                 seq_puts(seq, ",noprefix");
1107         if (root->flags & CGRP_ROOT_XATTR)
1108                 seq_puts(seq, ",xattr");
1109         if (strlen(root->release_agent_path))
1110                 seq_printf(seq, ",release_agent=%s", root->release_agent_path);
1111         if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->top_cgroup.flags))
1112                 seq_puts(seq, ",clone_children");
1113         if (strlen(root->name))
1114                 seq_printf(seq, ",name=%s", root->name);
1115         mutex_unlock(&cgroup_root_mutex);
1116         return 0;
1117 }
1118 
1119 struct cgroup_sb_opts {
1120         unsigned long subsys_mask;
1121         unsigned long flags;
1122         char *release_agent;
1123         bool cpuset_clone_children;
1124         char *name;
1125         /* User explicitly requested empty subsystem */
1126         bool none;
1127 
1128         struct cgroupfs_root *new_root;
1129 
1130 };
1131 
1132 /*
1133  * Convert a hierarchy specifier into a bitmask of subsystems and
1134  * flags. Call with cgroup_mutex held to protect the cgroup_subsys[]
1135  * array. This function takes refcounts on subsystems to be used, unless it
1136  * returns error, in which case no refcounts are taken.
1137  */
1138 static int parse_cgroupfs_options(char *data, struct cgroup_sb_opts *opts)
1139 {
1140         char *token, *o = data;
1141         bool all_ss = false, one_ss = false;
1142         unsigned long mask = (unsigned long)-1;
1143         struct cgroup_subsys *ss;
1144         int i;
1145 
1146         BUG_ON(!mutex_is_locked(&cgroup_mutex));
1147 
1148 #ifdef CONFIG_CPUSETS
1149         mask = ~(1UL << cpuset_subsys_id);
1150 #endif
1151 
1152         memset(opts, 0, sizeof(*opts));
1153 
1154         while ((token = strsep(&o, ",")) != NULL) {
1155                 if (!*token)
1156                         return -EINVAL;
1157                 if (!strcmp(token, "none")) {
1158                         /* Explicitly have no subsystems */
1159                         opts->none = true;
1160                         continue;
1161                 }
1162                 if (!strcmp(token, "all")) {
1163                         /* Mutually exclusive option 'all' + subsystem name */
1164                         if (one_ss)
1165                                 return -EINVAL;
1166                         all_ss = true;
1167                         continue;
1168                 }
1169                 if (!strcmp(token, "__DEVEL__sane_behavior")) {
1170                         opts->flags |= CGRP_ROOT_SANE_BEHAVIOR;
1171                         continue;
1172                 }
1173                 if (!strcmp(token, "noprefix")) {
1174                         opts->flags |= CGRP_ROOT_NOPREFIX;
1175                         continue;
1176                 }
1177                 if (!strcmp(token, "clone_children")) {
1178                         opts->cpuset_clone_children = true;
1179                         continue;
1180                 }
1181                 if (!strcmp(token, "xattr")) {
1182                         opts->flags |= CGRP_ROOT_XATTR;
1183                         continue;
1184                 }
1185                 if (!strncmp(token, "release_agent=", 14)) {
1186                         /* Specifying two release agents is forbidden */
1187                         if (opts->release_agent)
1188                                 return -EINVAL;
1189                         opts->release_agent =
1190                                 kstrndup(token + 14, PATH_MAX - 1, GFP_KERNEL);
1191                         if (!opts->release_agent)
1192                                 return -ENOMEM;
1193                         continue;
1194                 }
1195                 if (!strncmp(token, "name=", 5)) {
1196                         const char *name = token + 5;
1197                         /* Can't specify an empty name */
1198                         if (!strlen(name))
1199                                 return -EINVAL;
1200                         /* Must match [\w.-]+ */
1201                         for (i = 0; i < strlen(name); i++) {
1202                                 char c = name[i];
1203                                 if (isalnum(c))
1204                                         continue;
1205                                 if ((c == '.') || (c == '-') || (c == '_'))
1206                                         continue;
1207                                 return -EINVAL;
1208                         }
1209                         /* Specifying two names is forbidden */
1210                         if (opts->name)
1211                                 return -EINVAL;
1212                         opts->name = kstrndup(name,
1213                                               MAX_CGROUP_ROOT_NAMELEN - 1,
1214                                               GFP_KERNEL);
1215                         if (!opts->name)
1216                                 return -ENOMEM;
1217 
1218                         continue;
1219                 }
1220 
1221                 for_each_subsys(ss, i) {
1222                         if (strcmp(token, ss->name))
1223                                 continue;
1224                         if (ss->disabled)
1225                                 continue;
1226 
1227                         /* Mutually exclusive option 'all' + subsystem name */
1228                         if (all_ss)
1229                                 return -EINVAL;
1230                         set_bit(i, &opts->subsys_mask);
1231                         one_ss = true;
1232 
1233                         break;
1234                 }
1235                 if (i == CGROUP_SUBSYS_COUNT)
1236                         return -ENOENT;
1237         }
1238 
1239         /*
1240          * If the 'all' option was specified select all the subsystems,
1241          * otherwise if 'none', 'name=' and a subsystem name options
1242          * were not specified, let's default to 'all'
1243          */
1244         if (all_ss || (!one_ss && !opts->none && !opts->name))
1245                 for_each_subsys(ss, i)
1246                         if (!ss->disabled)
1247                                 set_bit(i, &opts->subsys_mask);
1248 
1249         /* Consistency checks */
1250 
1251         if (opts->flags & CGRP_ROOT_SANE_BEHAVIOR) {
1252                 pr_warning("cgroup: sane_behavior: this is still under development and its behaviors will change, proceed at your own risk\n");
1253 
1254                 if (opts->flags & CGRP_ROOT_NOPREFIX) {
1255                         pr_err("cgroup: sane_behavior: noprefix is not allowed\n");
1256                         return -EINVAL;
1257                 }
1258 
1259                 if (opts->cpuset_clone_children) {
1260                         pr_err("cgroup: sane_behavior: clone_children is not allowed\n");
1261                         return -EINVAL;
1262                 }
1263         }
1264 
1265         /*
1266          * Option noprefix was introduced just for backward compatibility
1267          * with the old cpuset, so we allow noprefix only if mounting just
1268          * the cpuset subsystem.
1269          */
1270         if ((opts->flags & CGRP_ROOT_NOPREFIX) && (opts->subsys_mask & mask))
1271                 return -EINVAL;
1272 
1273 
1274         /* Can't specify "none" and some subsystems */
1275         if (opts->subsys_mask && opts->none)
1276                 return -EINVAL;
1277 
1278         /*
1279          * We either have to specify by name or by subsystems. (So all
1280          * empty hierarchies must have a name).
1281          */
1282         if (!opts->subsys_mask && !opts->name)
1283                 return -EINVAL;
1284 
1285         return 0;
1286 }
1287 
1288 static int cgroup_remount(struct super_block *sb, int *flags, char *data)
1289 {
1290         int ret = 0;
1291         struct cgroupfs_root *root = sb->s_fs_info;
1292         struct cgroup *cgrp = &root->top_cgroup;
1293         struct cgroup_sb_opts opts;
1294         unsigned long added_mask, removed_mask;
1295 
1296         if (root->flags & CGRP_ROOT_SANE_BEHAVIOR) {
1297                 pr_err("cgroup: sane_behavior: remount is not allowed\n");
1298                 return -EINVAL;
1299         }
1300 
1301         mutex_lock(&cgrp->dentry->d_inode->i_mutex);
1302         mutex_lock(&cgroup_mutex);
1303         mutex_lock(&cgroup_root_mutex);
1304 
1305         /* See what subsystems are wanted */
1306         ret = parse_cgroupfs_options(data, &opts);
1307         if (ret)
1308                 goto out_unlock;
1309 
1310         if (opts.subsys_mask != root->subsys_mask || opts.release_agent)
1311                 pr_warning("cgroup: option changes via remount are deprecated (pid=%d comm=%s)\n",
1312                            task_tgid_nr(current), current->comm);
1313 
1314         added_mask = opts.subsys_mask & ~root->subsys_mask;
1315         removed_mask = root->subsys_mask & ~opts.subsys_mask;
1316 
1317         /* Don't allow flags or name to change at remount */
1318         if (((opts.flags ^ root->flags) & CGRP_ROOT_OPTION_MASK) ||
1319             (opts.name && strcmp(opts.name, root->name))) {
1320                 pr_err("cgroup: option or name mismatch, new: 0x%lx \"%s\", old: 0x%lx \"%s\"\n",
1321                        opts.flags & CGRP_ROOT_OPTION_MASK, opts.name ?: "",
1322                        root->flags & CGRP_ROOT_OPTION_MASK, root->name);
1323                 ret = -EINVAL;
1324                 goto out_unlock;
1325         }
1326 
1327         /* remounting is not allowed for populated hierarchies */
1328         if (root->number_of_cgroups > 1) {
1329                 ret = -EBUSY;
1330                 goto out_unlock;
1331         }
1332 
1333         ret = rebind_subsystems(root, added_mask, removed_mask);
1334         if (ret)
1335                 goto out_unlock;
1336 
1337         if (opts.release_agent)
1338                 strcpy(root->release_agent_path, opts.release_agent);
1339  out_unlock:
1340         kfree(opts.release_agent);
1341         kfree(opts.name);
1342         mutex_unlock(&cgroup_root_mutex);
1343         mutex_unlock(&cgroup_mutex);
1344         mutex_unlock(&cgrp->dentry->d_inode->i_mutex);
1345         return ret;
1346 }
1347 
1348 static const struct super_operations cgroup_ops = {
1349         .statfs = simple_statfs,
1350         .drop_inode = generic_delete_inode,
1351         .show_options = cgroup_show_options,
1352         .remount_fs = cgroup_remount,
1353 };
1354 
1355 static void init_cgroup_housekeeping(struct cgroup *cgrp)
1356 {
1357         INIT_LIST_HEAD(&cgrp->sibling);
1358         INIT_LIST_HEAD(&cgrp->children);
1359         INIT_LIST_HEAD(&cgrp->files);
1360         INIT_LIST_HEAD(&cgrp->cset_links);
1361         INIT_LIST_HEAD(&cgrp->release_list);
1362         INIT_LIST_HEAD(&cgrp->pidlists);
1363         mutex_init(&cgrp->pidlist_mutex);
1364         cgrp->dummy_css.cgroup = cgrp;
1365         INIT_LIST_HEAD(&cgrp->event_list);
1366         spin_lock_init(&cgrp->event_list_lock);
1367         simple_xattrs_init(&cgrp->xattrs);
1368 }
1369 
1370 static void init_cgroup_root(struct cgroupfs_root *root)
1371 {
1372         struct cgroup *cgrp = &root->top_cgroup;
1373 
1374         INIT_LIST_HEAD(&root->subsys_list);
1375         INIT_LIST_HEAD(&root->root_list);
1376         root->number_of_cgroups = 1;
1377         cgrp->root = root;
1378         RCU_INIT_POINTER(cgrp->name, &root_cgroup_name);
1379         init_cgroup_housekeeping(cgrp);
1380         idr_init(&root->cgroup_idr);
1381 }
1382 
1383 static int cgroup_init_root_id(struct cgroupfs_root *root, int start, int end)
1384 {
1385         int id;
1386 
1387         lockdep_assert_held(&cgroup_mutex);
1388         lockdep_assert_held(&cgroup_root_mutex);
1389 
1390         id = idr_alloc_cyclic(&cgroup_hierarchy_idr, root, start, end,
1391                               GFP_KERNEL);
1392         if (id < 0)
1393                 return id;
1394 
1395         root->hierarchy_id = id;
1396         return 0;
1397 }
1398 
1399 static void cgroup_exit_root_id(struct cgroupfs_root *root)
1400 {
1401         lockdep_assert_held(&cgroup_mutex);
1402         lockdep_assert_held(&cgroup_root_mutex);
1403 
1404         if (root->hierarchy_id) {
1405                 idr_remove(&cgroup_hierarchy_idr, root->hierarchy_id);
1406                 root->hierarchy_id = 0;
1407         }
1408 }
1409 
1410 static int cgroup_test_super(struct super_block *sb, void *data)
1411 {
1412         struct cgroup_sb_opts *opts = data;
1413         struct cgroupfs_root *root = sb->s_fs_info;
1414 
1415         /* If we asked for a name then it must match */
1416         if (opts->name && strcmp(opts->name, root->name))
1417                 return 0;
1418 
1419         /*
1420          * If we asked for subsystems (or explicitly for no
1421          * subsystems) then they must match
1422          */
1423         if ((opts->subsys_mask || opts->none)
1424             && (opts->subsys_mask != root->subsys_mask))
1425                 return 0;
1426 
1427         return 1;
1428 }
1429 
1430 static struct cgroupfs_root *cgroup_root_from_opts(struct cgroup_sb_opts *opts)
1431 {
1432         struct cgroupfs_root *root;
1433 
1434         if (!opts->subsys_mask && !opts->none)
1435                 return NULL;
1436 
1437         root = kzalloc(sizeof(*root), GFP_KERNEL);
1438         if (!root)
1439                 return ERR_PTR(-ENOMEM);
1440 
1441         init_cgroup_root(root);
1442 
1443         /*
1444          * We need to set @root->subsys_mask now so that @root can be
1445          * matched by cgroup_test_super() before it finishes
1446          * initialization; otherwise, competing mounts with the same
1447          * options may try to bind the same subsystems instead of waiting
1448          * for the first one leading to unexpected mount errors.
1449          * SUBSYS_BOUND will be set once actual binding is complete.
1450          */
1451         root->subsys_mask = opts->subsys_mask;
1452         root->flags = opts->flags;
1453         if (opts->release_agent)
1454                 strcpy(root->release_agent_path, opts->release_agent);
1455         if (opts->name)
1456                 strcpy(root->name, opts->name);
1457         if (opts->cpuset_clone_children)
1458                 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->top_cgroup.flags);
1459         return root;
1460 }
1461 
1462 static void cgroup_free_root(struct cgroupfs_root *root)
1463 {
1464         if (root) {
1465                 /* hierarhcy ID shoulid already have been released */
1466                 WARN_ON_ONCE(root->hierarchy_id);
1467 
1468                 idr_destroy(&root->cgroup_idr);
1469                 kfree(root);
1470         }
1471 }
1472 
1473 static int cgroup_set_super(struct super_block *sb, void *data)
1474 {
1475         int ret;
1476         struct cgroup_sb_opts *opts = data;
1477 
1478         /* If we don't have a new root, we can't set up a new sb */
1479         if (!opts->new_root)
1480                 return -EINVAL;
1481 
1482         BUG_ON(!opts->subsys_mask && !opts->none);
1483 
1484         ret = set_anon_super(sb, NULL);
1485         if (ret)
1486                 return ret;
1487 
1488         sb->s_fs_info = opts->new_root;
1489         opts->new_root->sb = sb;
1490 
1491         sb->s_blocksize = PAGE_CACHE_SIZE;
1492         sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
1493         sb->s_magic = CGROUP_SUPER_MAGIC;
1494         sb->s_op = &cgroup_ops;
1495 
1496         return 0;
1497 }
1498 
1499 static int cgroup_get_rootdir(struct super_block *sb)
1500 {
1501         static const struct dentry_operations cgroup_dops = {
1502                 .d_iput = cgroup_diput,
1503                 .d_delete = always_delete_dentry,
1504         };
1505 
1506         struct inode *inode =
1507                 cgroup_new_inode(S_IFDIR | S_IRUGO | S_IXUGO | S_IWUSR, sb);
1508 
1509         if (!inode)
1510                 return -ENOMEM;
1511 
1512         inode->i_fop = &simple_dir_operations;
1513         inode->i_op = &cgroup_dir_inode_operations;
1514         /* directories start off with i_nlink == 2 (for "." entry) */
1515         inc_nlink(inode);
1516         sb->s_root = d_make_root(inode);
1517         if (!sb->s_root)
1518                 return -ENOMEM;
1519         /* for everything else we want ->d_op set */
1520         sb->s_d_op = &cgroup_dops;
1521         return 0;
1522 }
1523 
1524 static struct dentry *cgroup_mount(struct file_system_type *fs_type,
1525                          int flags, const char *unused_dev_name,
1526                          void *data)
1527 {
1528         struct cgroup_sb_opts opts;
1529         struct cgroupfs_root *root;
1530         int ret = 0;
1531         struct super_block *sb;
1532         struct cgroupfs_root *new_root;
1533         struct list_head tmp_links;
1534         struct inode *inode;
1535         const struct cred *cred;
1536 
1537         /* First find the desired set of subsystems */
1538         mutex_lock(&cgroup_mutex);
1539         ret = parse_cgroupfs_options(data, &opts);
1540         mutex_unlock(&cgroup_mutex);
1541         if (ret)
1542                 goto out_err;
1543 
1544         /*
1545          * Allocate a new cgroup root. We may not need it if we're
1546          * reusing an existing hierarchy.
1547          */
1548         new_root = cgroup_root_from_opts(&opts);
1549         if (IS_ERR(new_root)) {
1550                 ret = PTR_ERR(new_root);
1551                 goto out_err;
1552         }
1553         opts.new_root = new_root;
1554 
1555         /* Locate an existing or new sb for this hierarchy */
1556         sb = sget(fs_type, cgroup_test_super, cgroup_set_super, 0, &opts);
1557         if (IS_ERR(sb)) {
1558                 ret = PTR_ERR(sb);
1559                 cgroup_free_root(opts.new_root);
1560                 goto out_err;
1561         }
1562 
1563         root = sb->s_fs_info;
1564         BUG_ON(!root);
1565         if (root == opts.new_root) {
1566                 /* We used the new root structure, so this is a new hierarchy */
1567                 struct cgroup *root_cgrp = &root->top_cgroup;
1568                 struct cgroupfs_root *existing_root;
1569                 int i;
1570                 struct css_set *cset;
1571 
1572                 BUG_ON(sb->s_root != NULL);
1573 
1574                 ret = cgroup_get_rootdir(sb);
1575                 if (ret)
1576                         goto drop_new_super;
1577                 inode = sb->s_root->d_inode;
1578 
1579                 mutex_lock(&inode->i_mutex);
1580                 mutex_lock(&cgroup_mutex);
1581                 mutex_lock(&cgroup_root_mutex);
1582 
1583                 root_cgrp->id = idr_alloc(&root->cgroup_idr, root_cgrp,
1584                                            0, 1, GFP_KERNEL);
1585                 if (root_cgrp->id < 0)
1586                         goto unlock_drop;
1587 
1588                 /* Check for name clashes with existing mounts */
1589                 ret = -EBUSY;
1590                 if (strlen(root->name))
1591                         for_each_active_root(existing_root)
1592                                 if (!strcmp(existing_root->name, root->name))
1593                                         goto unlock_drop;
1594 
1595                 /*
1596                  * We're accessing css_set_count without locking
1597                  * css_set_lock here, but that's OK - it can only be
1598                  * increased by someone holding cgroup_lock, and
1599                  * that's us. The worst that can happen is that we
1600                  * have some link structures left over
1601                  */
1602                 ret = allocate_cgrp_cset_links(css_set_count, &tmp_links);
1603                 if (ret)
1604                         goto unlock_drop;
1605 
1606                 /* ID 0 is reserved for dummy root, 1 for unified hierarchy */
1607                 ret = cgroup_init_root_id(root, 2, 0);
1608                 if (ret)
1609                         goto unlock_drop;
1610 
1611                 sb->s_root->d_fsdata = root_cgrp;
1612                 root_cgrp->dentry = sb->s_root;
1613 
1614                 /*
1615                  * We're inside get_sb() and will call lookup_one_len() to
1616                  * create the root files, which doesn't work if SELinux is
1617                  * in use.  The following cred dancing somehow works around
1618                  * it.  See 2ce9738ba ("cgroupfs: use init_cred when
1619                  * populating new cgroupfs mount") for more details.
1620                  */
1621                 cred = override_creds(&init_cred);
1622 
1623                 ret = cgroup_addrm_files(root_cgrp, cgroup_base_files, true);
1624                 if (ret)
1625                         goto rm_base_files;
1626 
1627                 ret = rebind_subsystems(root, root->subsys_mask, 0);
1628                 if (ret)
1629                         goto rm_base_files;
1630 
1631                 revert_creds(cred);
1632 
1633                 /*
1634                  * There must be no failure case after here, since rebinding
1635                  * takes care of subsystems' refcounts, which are explicitly
1636                  * dropped in the failure exit path.
1637                  */
1638 
1639                 list_add(&root->root_list, &cgroup_roots);
1640                 cgroup_root_count++;
1641 
1642                 /* Link the top cgroup in this hierarchy into all
1643                  * the css_set objects */
1644                 write_lock(&css_set_lock);
1645                 hash_for_each(css_set_table, i, cset, hlist)
1646                         link_css_set(&tmp_links, cset, root_cgrp);
1647                 write_unlock(&css_set_lock);
1648 
1649                 free_cgrp_cset_links(&tmp_links);
1650 
1651                 BUG_ON(!list_empty(&root_cgrp->children));
1652                 BUG_ON(root->number_of_cgroups != 1);
1653 
1654                 mutex_unlock(&cgroup_root_mutex);
1655                 mutex_unlock(&cgroup_mutex);
1656                 mutex_unlock(&inode->i_mutex);
1657         } else {
1658                 /*
1659                  * We re-used an existing hierarchy - the new root (if
1660                  * any) is not needed
1661                  */
1662                 cgroup_free_root(opts.new_root);
1663 
1664                 if ((root->flags ^ opts.flags) & CGRP_ROOT_OPTION_MASK) {
1665                         if ((root->flags | opts.flags) & CGRP_ROOT_SANE_BEHAVIOR) {
1666                                 pr_err("cgroup: sane_behavior: new mount options should match the existing superblock\n");
1667                                 ret = -EINVAL;
1668                                 goto drop_new_super;
1669                         } else {
1670                                 pr_warning("cgroup: new mount options do not match the existing superblock, will be ignored\n");
1671                         }
1672                 }
1673         }
1674 
1675         kfree(opts.release_agent);
1676         kfree(opts.name);
1677         return dget(sb->s_root);
1678 
1679  rm_base_files:
1680         free_cgrp_cset_links(&tmp_links);
1681         cgroup_addrm_files(&root->top_cgroup, cgroup_base_files, false);
1682         revert_creds(cred);
1683  unlock_drop:
1684         cgroup_exit_root_id(root);
1685         mutex_unlock(&cgroup_root_mutex);
1686         mutex_unlock(&cgroup_mutex);
1687         mutex_unlock(&inode->i_mutex);
1688  drop_new_super:
1689         deactivate_locked_super(sb);
1690  out_err:
1691         kfree(opts.release_agent);
1692         kfree(opts.name);
1693         return ERR_PTR(ret);
1694 }
1695 
1696 static void cgroup_kill_sb(struct super_block *sb) {
1697         struct cgroupfs_root *root = sb->s_fs_info;
1698         struct cgroup *cgrp = &root->top_cgroup;
1699         struct cgrp_cset_link *link, *tmp_link;
1700         int ret;
1701 
1702         BUG_ON(!root);
1703 
1704         BUG_ON(root->number_of_cgroups != 1);
1705         BUG_ON(!list_empty(&cgrp->children));
1706 
1707         mutex_lock(&cgrp->dentry->d_inode->i_mutex);
1708         mutex_lock(&cgroup_mutex);
1709         mutex_lock(&cgroup_root_mutex);
1710 
1711         /* Rebind all subsystems back to the default hierarchy */
1712         if (root->flags & CGRP_ROOT_SUBSYS_BOUND) {
1713                 ret = rebind_subsystems(root, 0, root->subsys_mask);
1714                 /* Shouldn't be able to fail ... */
1715                 BUG_ON(ret);
1716         }
1717 
1718         /*
1719          * Release all the links from cset_links to this hierarchy's
1720          * root cgroup
1721          */
1722         write_lock(&css_set_lock);
1723 
1724         list_for_each_entry_safe(link, tmp_link, &cgrp->cset_links, cset_link) {
1725                 list_del(&link->cset_link);
1726                 list_del(&link->cgrp_link);
1727                 kfree(link);
1728         }
1729         write_unlock(&css_set_lock);
1730 
1731         if (!list_empty(&root->root_list)) {
1732                 list_del(&root->root_list);
1733                 cgroup_root_count--;
1734         }
1735 
1736         cgroup_exit_root_id(root);
1737 
1738         mutex_unlock(&cgroup_root_mutex);
1739         mutex_unlock(&cgroup_mutex);
1740         mutex_unlock(&cgrp->dentry->d_inode->i_mutex);
1741 
1742         simple_xattrs_free(&cgrp->xattrs);
1743 
1744         kill_litter_super(sb);
1745         cgroup_free_root(root);
1746 }
1747 
1748 static struct file_system_type cgroup_fs_type = {
1749         .name = "cgroup",
1750         .mount = cgroup_mount,
1751         .kill_sb = cgroup_kill_sb,
1752 };
1753 
1754 static struct kobject *cgroup_kobj;
1755 
1756 /**
1757  * cgroup_path - generate the path of a cgroup
1758  * @cgrp: the cgroup in question
1759  * @buf: the buffer to write the path into
1760  * @buflen: the length of the buffer
1761  *
1762  * Writes path of cgroup into buf.  Returns 0 on success, -errno on error.
1763  *
1764  * We can't generate cgroup path using dentry->d_name, as accessing
1765  * dentry->name must be protected by irq-unsafe dentry->d_lock or parent
1766  * inode's i_mutex, while on the other hand cgroup_path() can be called
1767  * with some irq-safe spinlocks held.
1768  */
1769 int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen)
1770 {
1771         int ret = -ENAMETOOLONG;
1772         char *start;
1773 
1774         if (!cgrp->parent) {
1775                 if (strlcpy(buf, "/", buflen) >= buflen)
1776                         return -ENAMETOOLONG;
1777                 return 0;
1778         }
1779 
1780         start = buf + buflen - 1;
1781         *start = '\0';
1782 
1783         rcu_read_lock();
1784         do {
1785                 const char *name = cgroup_name(cgrp);
1786                 int len;
1787 
1788                 len = strlen(name);
1789                 if ((start -= len) < buf)
1790                         goto out;
1791                 memcpy(start, name, len);
1792 
1793                 if (--start < buf)
1794                         goto out;
1795                 *start = '/';
1796 
1797                 cgrp = cgrp->parent;
1798         } while (cgrp->parent);
1799         ret = 0;
1800         memmove(buf, start, buf + buflen - start);
1801 out:
1802         rcu_read_unlock();
1803         return ret;
1804 }
1805 EXPORT_SYMBOL_GPL(cgroup_path);
1806 
1807 /**
1808  * task_cgroup_path - cgroup path of a task in the first cgroup hierarchy
1809  * @task: target task
1810  * @buf: the buffer to write the path into
1811  * @buflen: the length of the buffer
1812  *
1813  * Determine @task's cgroup on the first (the one with the lowest non-zero
1814  * hierarchy_id) cgroup hierarchy and copy its path into @buf.  This
1815  * function grabs cgroup_mutex and shouldn't be used inside locks used by
1816  * cgroup controller callbacks.
1817  *
1818  * Returns 0 on success, fails with -%ENAMETOOLONG if @buflen is too short.
1819  */
1820 int task_cgroup_path(struct task_struct *task, char *buf, size_t buflen)
1821 {
1822         struct cgroupfs_root *root;
1823         struct cgroup *cgrp;
1824         int hierarchy_id = 1, ret = 0;
1825 
1826         if (buflen < 2)
1827                 return -ENAMETOOLONG;
1828 
1829         mutex_lock(&cgroup_mutex);
1830 
1831         root = idr_get_next(&cgroup_hierarchy_idr, &hierarchy_id);
1832 
1833         if (root) {
1834                 cgrp = task_cgroup_from_root(task, root);
1835                 ret = cgroup_path(cgrp, buf, buflen);
1836         } else {
1837                 /* if no hierarchy exists, everyone is in "/" */
1838                 memcpy(buf, "/", 2);
1839         }
1840 
1841         mutex_unlock(&cgroup_mutex);
1842         return ret;
1843 }
1844 EXPORT_SYMBOL_GPL(task_cgroup_path);
1845 
1846 /*
1847  * Control Group taskset
1848  */
1849 struct task_and_cgroup {
1850         struct task_struct      *task;
1851         struct cgroup           *cgrp;
1852         struct css_set          *cset;
1853 };
1854 
1855 struct cgroup_taskset {
1856         struct task_and_cgroup  single;
1857         struct flex_array       *tc_array;
1858         int                     tc_array_len;
1859         int                     idx;
1860         struct cgroup           *cur_cgrp;
1861 };
1862 
1863 /**
1864  * cgroup_taskset_first - reset taskset and return the first task
1865  * @tset: taskset of interest
1866  *
1867  * @tset iteration is initialized and the first task is returned.
1868  */
1869 struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset)
1870 {
1871         if (tset->tc_array) {
1872                 tset->idx = 0;
1873                 return cgroup_taskset_next(tset);
1874         } else {
1875                 tset->cur_cgrp = tset->single.cgrp;
1876                 return tset->single.task;
1877         }
1878 }
1879 EXPORT_SYMBOL_GPL(cgroup_taskset_first);
1880 
1881 /**
1882  * cgroup_taskset_next - iterate to the next task in taskset
1883  * @tset: taskset of interest
1884  *
1885  * Return the next task in @tset.  Iteration must have been initialized
1886  * with cgroup_taskset_first().
1887  */
1888 struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset)
1889 {
1890         struct task_and_cgroup *tc;
1891 
1892         if (!tset->tc_array || tset->idx >= tset->tc_array_len)
1893                 return NULL;
1894 
1895         tc = flex_array_get(tset->tc_array, tset->idx++);
1896         tset->cur_cgrp = tc->cgrp;
1897         return tc->task;
1898 }
1899 EXPORT_SYMBOL_GPL(cgroup_taskset_next);
1900 
1901 /**
1902  * cgroup_taskset_cur_css - return the matching css for the current task
1903  * @tset: taskset of interest
1904  * @subsys_id: the ID of the target subsystem
1905  *
1906  * Return the css for the current (last returned) task of @tset for
1907  * subsystem specified by @subsys_id.  This function must be preceded by
1908  * either cgroup_taskset_first() or cgroup_taskset_next().
1909  */
1910 struct cgroup_subsys_state *cgroup_taskset_cur_css(struct cgroup_taskset *tset,
1911                                                    int subsys_id)
1912 {
1913         return cgroup_css(tset->cur_cgrp, cgroup_subsys[subsys_id]);
1914 }
1915 EXPORT_SYMBOL_GPL(cgroup_taskset_cur_css);
1916 
1917 /**
1918  * cgroup_taskset_size - return the number of tasks in taskset
1919  * @tset: taskset of interest
1920  */
1921 int cgroup_taskset_size(struct cgroup_taskset *tset)
1922 {
1923         return tset->tc_array ? tset->tc_array_len : 1;
1924 }
1925 EXPORT_SYMBOL_GPL(cgroup_taskset_size);
1926 
1927 
1928 /*
1929  * cgroup_task_migrate - move a task from one cgroup to another.
1930  *
1931  * Must be called with cgroup_mutex and threadgroup locked.
1932  */
1933 static void cgroup_task_migrate(struct cgroup *old_cgrp,
1934                                 struct task_struct *tsk,
1935                                 struct css_set *new_cset)
1936 {
1937         struct css_set *old_cset;
1938 
1939         /*
1940          * We are synchronized through threadgroup_lock() against PF_EXITING
1941          * setting such that we can't race against cgroup_exit() changing the
1942          * css_set to init_css_set and dropping the old one.
1943          */
1944         WARN_ON_ONCE(tsk->flags & PF_EXITING);
1945         old_cset = task_css_set(tsk);
1946 
1947         task_lock(tsk);
1948         rcu_assign_pointer(tsk->cgroups, new_cset);
1949         task_unlock(tsk);
1950 
1951         /* Update the css_set linked lists if we're using them */
1952         write_lock(&css_set_lock);
1953         if (!list_empty(&tsk->cg_list))
1954                 list_move(&tsk->cg_list, &new_cset->tasks);
1955         write_unlock(&css_set_lock);
1956 
1957         /*
1958          * We just gained a reference on old_cset by taking it from the
1959          * task. As trading it for new_cset is protected by cgroup_mutex,
1960          * we're safe to drop it here; it will be freed under RCU.
1961          */
1962         set_bit(CGRP_RELEASABLE, &old_cgrp->flags);
1963         put_css_set(old_cset);
1964 }
1965 
1966 /**
1967  * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup
1968  * @cgrp: the cgroup to attach to
1969  * @tsk: the task or the leader of the threadgroup to be attached
1970  * @threadgroup: attach the whole threadgroup?
1971  *
1972  * Call holding cgroup_mutex and the group_rwsem of the leader. Will take
1973  * task_lock of @tsk or each thread in the threadgroup individually in turn.
1974  */
1975 static int cgroup_attach_task(struct cgroup *cgrp, struct task_struct *tsk,
1976                               bool threadgroup)
1977 {
1978         int retval, i, group_size;
1979         struct cgroup_subsys *ss, *failed_ss = NULL;
1980         struct cgroupfs_root *root = cgrp->root;
1981         /* threadgroup list cursor and array */
1982         struct task_struct *leader = tsk;
1983         struct task_and_cgroup *tc;
1984         struct flex_array *group;
1985         struct cgroup_taskset tset = { };
1986 
1987         /*
1988          * step 0: in order to do expensive, possibly blocking operations for
1989          * every thread, we cannot iterate the thread group list, since it needs
1990          * rcu or tasklist locked. instead, build an array of all threads in the
1991          * group - group_rwsem prevents new threads from appearing, and if
1992          * threads exit, this will just be an over-estimate.
1993          */
1994         if (threadgroup)
1995                 group_size = get_nr_threads(tsk);
1996         else
1997                 group_size = 1;
1998         /* flex_array supports very large thread-groups better than kmalloc. */
1999         group = flex_array_alloc(sizeof(*tc), group_size, GFP_KERNEL);
2000         if (!group)
2001                 return -ENOMEM;
2002         /* pre-allocate to guarantee space while iterating in rcu read-side. */
2003         retval = flex_array_prealloc(group, 0, group_size, GFP_KERNEL);
2004         if (retval)
2005                 goto out_free_group_list;
2006 
2007         i = 0;
2008         /*
2009          * Prevent freeing of tasks while we take a snapshot. Tasks that are
2010          * already PF_EXITING could be freed from underneath us unless we
2011          * take an rcu_read_lock.
2012          */
2013         rcu_read_lock();
2014         do {
2015                 struct task_and_cgroup ent;
2016 
2017                 /* @tsk either already exited or can't exit until the end */
2018                 if (tsk->flags & PF_EXITING)
2019                         goto next;
2020 
2021                 /* as per above, nr_threads may decrease, but not increase. */
2022                 BUG_ON(i >= group_size);
2023                 ent.task = tsk;
2024                 ent.cgrp = task_cgroup_from_root(tsk, root);
2025                 /* nothing to do if this task is already in the cgroup */
2026                 if (ent.cgrp == cgrp)
2027                         goto next;
2028                 /*
2029                  * saying GFP_ATOMIC has no effect here because we did prealloc
2030                  * earlier, but it's good form to communicate our expectations.
2031                  */
2032                 retval = flex_array_put(group, i, &ent, GFP_ATOMIC);
2033                 BUG_ON(retval != 0);
2034                 i++;
2035         next:
2036                 if (!threadgroup)
2037                         break;
2038         } while_each_thread(leader, tsk);
2039         rcu_read_unlock();
2040         /* remember the number of threads in the array for later. */
2041         group_size = i;
2042         tset.tc_array = group;
2043         tset.tc_array_len = group_size;
2044 
2045         /* methods shouldn't be called if no task is actually migrating */
2046         retval = 0;
2047         if (!group_size)
2048                 goto out_free_group_list;
2049 
2050         /*
2051          * step 1: check that we can legitimately attach to the cgroup.
2052          */
2053         for_each_root_subsys(root, ss) {
2054                 struct cgroup_subsys_state *css = cgroup_css(cgrp, ss);
2055 
2056                 if (ss->can_attach) {
2057                         retval = ss->can_attach(css, &tset);
2058                         if (retval) {
2059                                 failed_ss = ss;
2060                                 goto out_cancel_attach;
2061                         }
2062                 }
2063         }
2064 
2065         /*
2066          * step 2: make sure css_sets exist for all threads to be migrated.
2067          * we use find_css_set, which allocates a new one if necessary.
2068          */
2069         for (i = 0; i < group_size; i++) {
2070                 struct css_set *old_cset;
2071 
2072                 tc = flex_array_get(group, i);
2073                 old_cset = task_css_set(tc->task);
2074                 tc->cset = find_css_set(old_cset, cgrp);
2075                 if (!tc->cset) {
2076                         retval = -ENOMEM;
2077                         goto out_put_css_set_refs;
2078                 }
2079         }
2080 
2081         /*
2082          * step 3: now that we're guaranteed success wrt the css_sets,
2083          * proceed to move all tasks to the new cgroup.  There are no
2084          * failure cases after here, so this is the commit point.
2085          */
2086         for (i = 0; i < group_size; i++) {
2087                 tc = flex_array_get(group, i);
2088                 cgroup_task_migrate(tc->cgrp, tc->task, tc->cset);
2089         }
2090         /* nothing is sensitive to fork() after this point. */
2091 
2092         /*
2093          * step 4: do subsystem attach callbacks.
2094          */
2095         for_each_root_subsys(root, ss) {
2096                 struct cgroup_subsys_state *css = cgroup_css(cgrp, ss);
2097 
2098                 if (ss->attach)
2099                         ss->attach(css, &tset);
2100         }
2101 
2102         /*
2103          * step 5: success! and cleanup
2104          */
2105         retval = 0;
2106 out_put_css_set_refs:
2107         if (retval) {
2108                 for (i = 0; i < group_size; i++) {
2109                         tc = flex_array_get(group, i);
2110                         if (!tc->cset)
2111                                 break;
2112                         put_css_set(tc->cset);
2113                 }
2114         }
2115 out_cancel_attach:
2116         if (retval) {
2117                 for_each_root_subsys(root, ss) {
2118                         struct cgroup_subsys_state *css = cgroup_css(cgrp, ss);
2119 
2120                         if (ss == failed_ss)
2121                                 break;
2122                         if (ss->cancel_attach)
2123                                 ss->cancel_attach(css, &tset);
2124                 }
2125         }
2126 out_free_group_list:
2127         flex_array_free(group);
2128         return retval;
2129 }
2130 
2131 /*
2132  * Find the task_struct of the task to attach by vpid and pass it along to the
2133  * function to attach either it or all tasks in its threadgroup. Will lock
2134  * cgroup_mutex and threadgroup; may take task_lock of task.
2135  */
2136 static int attach_task_by_pid(struct cgroup *cgrp, u64 pid, bool threadgroup)
2137 {
2138         struct task_struct *tsk;
2139         const struct cred *cred = current_cred(), *tcred;
2140         int ret;
2141 
2142         if (!cgroup_lock_live_group(cgrp))
2143                 return -ENODEV;
2144 
2145 retry_find_task:
2146         rcu_read_lock();
2147         if (pid) {
2148                 tsk = find_task_by_vpid(pid);
2149                 if (!tsk) {
2150                         rcu_read_unlock();
2151                         ret= -ESRCH;
2152                         goto out_unlock_cgroup;
2153                 }
2154                 /*
2155                  * even if we're attaching all tasks in the thread group, we
2156                  * only need to check permissions on one of them.
2157                  */
2158                 tcred = __task_cred(tsk);
2159                 if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) &&
2160                     !uid_eq(cred->euid, tcred->uid) &&
2161                     !uid_eq(cred->euid, tcred->suid)) {
2162                         rcu_read_unlock();
2163                         ret = -EACCES;
2164                         goto out_unlock_cgroup;
2165                 }
2166         } else
2167                 tsk = current;
2168 
2169         if (threadgroup)
2170                 tsk = tsk->group_leader;
2171 
2172         /*
2173          * Workqueue threads may acquire PF_NO_SETAFFINITY and become
2174          * trapped in a cpuset, or RT worker may be born in a cgroup
2175          * with no rt_runtime allocated.  Just say no.
2176          */
2177         if (tsk == kthreadd_task || (tsk->flags & PF_NO_SETAFFINITY)) {
2178                 ret = -EINVAL;
2179                 rcu_read_unlock();
2180                 goto out_unlock_cgroup;
2181         }
2182 
2183         get_task_struct(tsk);
2184         rcu_read_unlock();
2185 
2186         threadgroup_lock(tsk);
2187         if (threadgroup) {
2188                 if (!thread_group_leader(tsk)) {
2189                         /*
2190                          * a race with de_thread from another thread's exec()
2191                          * may strip us of our leadership, if this happens,
2192                          * there is no choice but to throw this task away and
2193                          * try again; this is
2194                          * "double-double-toil-and-trouble-check locking".
2195                          */
2196                         threadgroup_unlock(tsk);
2197                         put_task_struct(tsk);
2198                         goto retry_find_task;
2199                 }
2200         }
2201 
2202         ret = cgroup_attach_task(cgrp, tsk, threadgroup);
2203 
2204         threadgroup_unlock(tsk);
2205 
2206         put_task_struct(tsk);
2207 out_unlock_cgroup:
2208         mutex_unlock(&cgroup_mutex);
2209         return ret;
2210 }
2211 
2212 /**
2213  * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from'
2214  * @from: attach to all cgroups of a given task
2215  * @tsk: the task to be attached
2216  */
2217 int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk)
2218 {
2219         struct cgroupfs_root *root;
2220         int retval = 0;
2221 
2222         mutex_lock(&cgroup_mutex);
2223         for_each_active_root(root) {
2224                 struct cgroup *from_cgrp = task_cgroup_from_root(from, root);
2225 
2226                 retval = cgroup_attach_task(from_cgrp, tsk, false);
2227                 if (retval)
2228                         break;
2229         }
2230         mutex_unlock(&cgroup_mutex);
2231 
2232         return retval;
2233 }
2234 EXPORT_SYMBOL_GPL(cgroup_attach_task_all);
2235 
2236 static int cgroup_tasks_write(struct cgroup_subsys_state *css,
2237                               struct cftype *cft, u64 pid)
2238 {
2239         return attach_task_by_pid(css->cgroup, pid, false);
2240 }
2241 
2242 static int cgroup_procs_write(struct cgroup_subsys_state *css,
2243                               struct cftype *cft, u64 tgid)
2244 {
2245         return attach_task_by_pid(css->cgroup, tgid, true);
2246 }
2247 
2248 static int cgroup_release_agent_write(struct cgroup_subsys_state *css,
2249                                       struct cftype *cft, const char *buffer)
2250 {
2251         BUILD_BUG_ON(sizeof(css->cgroup->root->release_agent_path) < PATH_MAX);
2252         if (strlen(buffer) >= PATH_MAX)
2253                 return -EINVAL;
2254         if (!cgroup_lock_live_group(css->cgroup))
2255                 return -ENODEV;
2256         mutex_lock(&cgroup_root_mutex);
2257         strcpy(css->cgroup->root->release_agent_path, buffer);
2258         mutex_unlock(&cgroup_root_mutex);
2259         mutex_unlock(&cgroup_mutex);
2260         return 0;
2261 }
2262 
2263 static int cgroup_release_agent_show(struct cgroup_subsys_state *css,
2264                                      struct cftype *cft, struct seq_file *seq)
2265 {
2266         struct cgroup *cgrp = css->cgroup;
2267 
2268         if (!cgroup_lock_live_group(cgrp))
2269                 return -ENODEV;
2270         seq_puts(seq, cgrp->root->release_agent_path);
2271         seq_putc(seq, '\n');
2272         mutex_unlock(&cgroup_mutex);
2273         return 0;
2274 }
2275 
2276 static int cgroup_sane_behavior_show(struct cgroup_subsys_state *css,
2277                                      struct cftype *cft, struct seq_file *seq)
2278 {
2279         seq_printf(seq, "%d\n", cgroup_sane_behavior(css->cgroup));
2280         return 0;
2281 }
2282 
2283 /* A buffer size big enough for numbers or short strings */
2284 #define CGROUP_LOCAL_BUFFER_SIZE 64
2285 
2286 static ssize_t cgroup_write_X64(struct cgroup_subsys_state *css,
2287                                 struct cftype *cft, struct file *file,
2288                                 const char __user *userbuf, size_t nbytes,
2289                                 loff_t *unused_ppos)
2290 {
2291         char buffer[CGROUP_LOCAL_BUFFER_SIZE];
2292         int retval = 0;
2293         char *end;
2294 
2295         if (!nbytes)
2296                 return -EINVAL;
2297         if (nbytes >= sizeof(buffer))
2298                 return -E2BIG;
2299         if (copy_from_user(buffer, userbuf, nbytes))
2300                 return -EFAULT;
2301 
2302         buffer[nbytes] = 0;     /* nul-terminate */
2303         if (cft->write_u64) {
2304                 u64 val = simple_strtoull(strstrip(buffer), &end, 0);
2305                 if (*end)
2306                         return -EINVAL;
2307                 retval = cft->write_u64(css, cft, val);
2308         } else {
2309                 s64 val = simple_strtoll(strstrip(buffer), &end, 0);
2310                 if (*end)
2311                         return -EINVAL;
2312                 retval = cft->write_s64(css, cft, val);
2313         }
2314         if (!retval)
2315                 retval = nbytes;
2316         return retval;
2317 }
2318 
2319 static ssize_t cgroup_write_string(struct cgroup_subsys_state *css,
2320                                    struct cftype *cft, struct file *file,
2321                                    const char __user *userbuf, size_t nbytes,
2322                                    loff_t *unused_ppos)
2323 {
2324         char local_buffer[CGROUP_LOCAL_BUFFER_SIZE];
2325         int retval = 0;
2326         size_t max_bytes = cft->max_write_len;
2327         char *buffer = local_buffer;
2328 
2329         if (!max_bytes)
2330                 max_bytes = sizeof(local_buffer) - 1;
2331         if (nbytes >= max_bytes)
2332                 return -E2BIG;
2333         /* Allocate a dynamic buffer if we need one */
2334         if (nbytes >= sizeof(local_buffer)) {
2335                 buffer = kmalloc(nbytes + 1, GFP_KERNEL);
2336                 if (buffer == NULL)
2337                         return -ENOMEM;
2338         }
2339         if (nbytes && copy_from_user(buffer, userbuf, nbytes)) {
2340                 retval = -EFAULT;
2341                 goto out;
2342         }
2343 
2344         buffer[nbytes] = 0;     /* nul-terminate */
2345         retval = cft->write_string(css, cft, strstrip(buffer));
2346         if (!retval)
2347                 retval = nbytes;
2348 out:
2349         if (buffer != local_buffer)
2350                 kfree(buffer);
2351         return retval;
2352 }
2353 
2354 static ssize_t cgroup_file_write(struct file *file, const char __user *buf,
2355                                  size_t nbytes, loff_t *ppos)
2356 {
2357         struct cfent *cfe = __d_cfe(file->f_dentry);
2358         struct cftype *cft = __d_cft(file->f_dentry);
2359         struct cgroup_subsys_state *css = cfe->css;
2360 
2361         if (cft->write)
2362                 return cft->write(css, cft, file, buf, nbytes, ppos);
2363         if (cft->write_u64 || cft->write_s64)
2364                 return cgroup_write_X64(css, cft, file, buf, nbytes, ppos);
2365         if (cft->write_string)
2366                 return cgroup_write_string(css, cft, file, buf, nbytes, ppos);
2367         if (cft->trigger) {
2368                 int ret = cft->trigger(css, (unsigned int)cft->private);
2369                 return ret ? ret : nbytes;
2370         }
2371         return -EINVAL;
2372 }
2373 
2374 static ssize_t cgroup_read_u64(struct cgroup_subsys_state *css,
2375                                struct cftype *cft, struct file *file,
2376                                char __user *buf, size_t nbytes, loff_t *ppos)
2377 {
2378         char tmp[CGROUP_LOCAL_BUFFER_SIZE];
2379         u64 val = cft->read_u64(css, cft);
2380         int len = sprintf(tmp, "%llu\n", (unsigned long long) val);
2381 
2382         return simple_read_from_buffer(buf, nbytes, ppos, tmp, len);
2383 }
2384 
2385 static ssize_t cgroup_read_s64(struct cgroup_subsys_state *css,
2386                                struct cftype *cft, struct file *file,
2387                                char __user *buf, size_t nbytes, loff_t *ppos)
2388 {
2389         char tmp[CGROUP_LOCAL_BUFFER_SIZE];
2390         s64 val = cft->read_s64(css, cft);
2391         int len = sprintf(tmp, "%lld\n", (long long) val);
2392 
2393         return simple_read_from_buffer(buf, nbytes, ppos, tmp, len);
2394 }
2395 
2396 static ssize_t cgroup_file_read(struct file *file, char __user *buf,
2397                                 size_t nbytes, loff_t *ppos)
2398 {
2399         struct cfent *cfe = __d_cfe(file->f_dentry);
2400         struct cftype *cft = __d_cft(file->f_dentry);
2401         struct cgroup_subsys_state *css = cfe->css;
2402 
2403         if (cft->read)
2404                 return cft->read(css, cft, file, buf, nbytes, ppos);
2405         if (cft->read_u64)
2406                 return cgroup_read_u64(css, cft, file, buf, nbytes, ppos);
2407         if (cft->read_s64)
2408                 return cgroup_read_s64(css, cft, file, buf, nbytes, ppos);
2409         return -EINVAL;
2410 }
2411 
2412 /*
2413  * seqfile ops/methods for returning structured data. Currently just
2414  * supports string->u64 maps, but can be extended in future.
2415  */
2416 
2417 static int cgroup_map_add(struct cgroup_map_cb *cb, const char *key, u64 value)
2418 {
2419         struct seq_file *sf = cb->state;
2420         return seq_printf(sf, "%s %llu\n", key, (unsigned long long)value);
2421 }
2422 
2423 static int cgroup_seqfile_show(struct seq_file *m, void *arg)
2424 {
2425         struct cfent *cfe = m->private;
2426         struct cftype *cft = cfe->type;
2427         struct cgroup_subsys_state *css = cfe->css;
2428 
2429         if (cft->read_map) {
2430                 struct cgroup_map_cb cb = {
2431                         .fill = cgroup_map_add,
2432                         .state = m,
2433                 };
2434                 return cft->read_map(css, cft, &cb);
2435         }
2436         return cft->read_seq_string(css, cft, m);
2437 }
2438 
2439 static const struct file_operations cgroup_seqfile_operations = {
2440         .read = seq_read,
2441         .write = cgroup_file_write,
2442         .llseek = seq_lseek,
2443         .release = cgroup_file_release,
2444 };
2445 
2446 static int cgroup_file_open(struct inode *inode, struct file *file)
2447 {
2448         struct cfent *cfe = __d_cfe(file->f_dentry);
2449         struct cftype *cft = __d_cft(file->f_dentry);
2450         struct cgroup *cgrp = __d_cgrp(cfe->dentry->d_parent);
2451         struct cgroup_subsys_state *css;
2452         int err;
2453 
2454         err = generic_file_open(inode, file);
2455         if (err)
2456                 return err;
2457 
2458         /*
2459          * If the file belongs to a subsystem, pin the css.  Will be
2460          * unpinned either on open failure or release.  This ensures that
2461          * @css stays alive for all file operations.
2462          */
2463         rcu_read_lock();
2464         css = cgroup_css(cgrp, cft->ss);
2465         if (cft->ss && !css_tryget(css))
2466                 css = NULL;
2467         rcu_read_unlock();
2468 
2469         if (!css)
2470                 return -ENODEV;
2471 
2472         /*
2473          * @cfe->css is used by read/write/close to determine the
2474          * associated css.  @file->private_data would be a better place but
2475          * that's already used by seqfile.  Multiple accessors may use it
2476          * simultaneously which is okay as the association never changes.
2477          */
2478         WARN_ON_ONCE(cfe->css && cfe->css != css);
2479         cfe->css = css;
2480 
2481         if (cft->read_map || cft->read_seq_string) {
2482                 file->f_op = &cgroup_seqfile_operations;
2483                 err = single_open(file, cgroup_seqfile_show, cfe);
2484         } else if (cft->open) {
2485                 err = cft->open(inode, file);
2486         }
2487 
2488         if (css->ss && err)
2489                 css_put(css);
2490         return err;
2491 }
2492 
2493 static int cgroup_file_release(struct inode *inode, struct file *file)
2494 {
2495         struct cfent *cfe = __d_cfe(file->f_dentry);
2496         struct cftype *cft = __d_cft(file->f_dentry);
2497         struct cgroup_subsys_state *css = cfe->css;
2498         int ret = 0;
2499 
2500         if (cft->release)
2501                 ret = cft->release(inode, file);
2502         if (css->ss)
2503                 css_put(css);
2504         if (file->f_op == &cgroup_seqfile_operations)
2505                 single_release(inode, file);
2506         return ret;
2507 }
2508 
2509 /*
2510  * cgroup_rename - Only allow simple rename of directories in place.
2511  */
2512 static int cgroup_rename(struct inode *old_dir, struct dentry *old_dentry,
2513                             struct inode *new_dir, struct dentry *new_dentry)
2514 {
2515         int ret;
2516         struct cgroup_name *name, *old_name;
2517         struct cgroup *cgrp;
2518 
2519         /*
2520          * It's convinient to use parent dir's i_mutex to protected
2521          * cgrp->name.
2522          */
2523         lockdep_assert_held(&old_dir->i_mutex);
2524 
2525         if (!S_ISDIR(old_dentry->d_inode->i_mode))
2526                 return -ENOTDIR;
2527         if (new_dentry->d_inode)
2528                 return -EEXIST;
2529         if (old_dir != new_dir)
2530                 return -EIO;
2531 
2532         cgrp = __d_cgrp(old_dentry);
2533 
2534         /*
2535          * This isn't a proper migration and its usefulness is very
2536          * limited.  Disallow if sane_behavior.
2537          */
2538         if (cgroup_sane_behavior(cgrp))
2539                 return -EPERM;
2540 
2541         name = cgroup_alloc_name(new_dentry);
2542         if (!name)
2543                 return -ENOMEM;
2544 
2545         ret = simple_rename(old_dir, old_dentry, new_dir, new_dentry);
2546         if (ret) {
2547                 kfree(name);
2548                 return ret;
2549         }
2550 
2551         old_name = rcu_dereference_protected(cgrp->name, true);
2552         rcu_assign_pointer(cgrp->name, name);
2553 
2554         kfree_rcu(old_name, rcu_head);
2555         return 0;
2556 }
2557 
2558 static struct simple_xattrs *__d_xattrs(struct dentry *dentry)
2559 {
2560         if (S_ISDIR(dentry->d_inode->i_mode))
2561                 return &__d_cgrp(dentry)->xattrs;
2562         else
2563                 return &__d_cfe(dentry)->xattrs;
2564 }
2565 
2566 static inline int xattr_enabled(struct dentry *dentry)
2567 {
2568         struct cgroupfs_root *root = dentry->d_sb->s_fs_info;
2569         return root->flags & CGRP_ROOT_XATTR;
2570 }
2571 
2572 static bool is_valid_xattr(const char *name)
2573 {
2574         if (!strncmp(name, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN) ||
2575             !strncmp(name, XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN))
2576                 return true;
2577         return false;
2578 }
2579 
2580 static int cgroup_setxattr(struct dentry *dentry, const char *name,
2581                            const void *val, size_t size, int flags)
2582 {
2583         if (!xattr_enabled(dentry))
2584                 return -EOPNOTSUPP;
2585         if (!is_valid_xattr(name))
2586                 return -EINVAL;
2587         return simple_xattr_set(__d_xattrs(dentry), name, val, size, flags);
2588 }
2589 
2590 static int cgroup_removexattr(struct dentry *dentry, const char *name)
2591 {
2592         if (!xattr_enabled(dentry))
2593                 return -EOPNOTSUPP;
2594         if (!is_valid_xattr(name))
2595                 return -EINVAL;
2596         return simple_xattr_remove(__d_xattrs(dentry), name);
2597 }
2598 
2599 static ssize_t cgroup_getxattr(struct dentry *dentry, const char *name,
2600                                void *buf, size_t size)
2601 {
2602         if (!xattr_enabled(dentry))
2603                 return -EOPNOTSUPP;
2604         if (!is_valid_xattr(name))
2605                 return -EINVAL;
2606         return simple_xattr_get(__d_xattrs(dentry), name, buf, size);
2607 }
2608 
2609 static ssize_t cgroup_listxattr(struct dentry *dentry, char *buf, size_t size)
2610 {
2611         if (!xattr_enabled(dentry))
2612                 return -EOPNOTSUPP;
2613         return simple_xattr_list(__d_xattrs(dentry), buf, size);
2614 }
2615 
2616 static const struct file_operations cgroup_file_operations = {
2617         .read = cgroup_file_read,
2618         .write = cgroup_file_write,
2619         .llseek = generic_file_llseek,
2620         .open = cgroup_file_open,
2621         .release = cgroup_file_release,
2622 };
2623 
2624 static const struct inode_operations cgroup_file_inode_operations = {
2625         .setxattr = cgroup_setxattr,
2626         .getxattr = cgroup_getxattr,
2627         .listxattr = cgroup_listxattr,
2628         .removexattr = cgroup_removexattr,
2629 };
2630 
2631 static const struct inode_operations cgroup_dir_inode_operations = {
2632         .lookup = simple_lookup,
2633         .mkdir = cgroup_mkdir,
2634         .rmdir = cgroup_rmdir,
2635         .rename = cgroup_rename,
2636         .setxattr = cgroup_setxattr,
2637         .getxattr = cgroup_getxattr,
2638         .listxattr = cgroup_listxattr,
2639         .removexattr = cgroup_removexattr,
2640 };
2641 
2642 /*
2643  * Check if a file is a control file
2644  */
2645 static inline struct cftype *__file_cft(struct file *file)
2646 {
2647         if (file_inode(file)->i_fop != &cgroup_file_operations)
2648                 return ERR_PTR(-EINVAL);
2649         return __d_cft(file->f_dentry);
2650 }
2651 
2652 static int cgroup_create_file(struct dentry *dentry, umode_t mode,
2653                                 struct super_block *sb)
2654 {
2655         struct inode *inode;
2656 
2657         if (!dentry)
2658                 return -ENOENT;
2659         if (dentry->d_inode)
2660                 return -EEXIST;
2661 
2662         inode = cgroup_new_inode(mode, sb);
2663         if (!inode)
2664                 return -ENOMEM;
2665 
2666         if (S_ISDIR(mode)) {
2667                 inode->i_op = &cgroup_dir_inode_operations;
2668                 inode->i_fop = &simple_dir_operations;
2669 
2670                 /* start off with i_nlink == 2 (for "." entry) */
2671                 inc_nlink(inode);
2672                 inc_nlink(dentry->d_parent->d_inode);
2673 
2674                 /*
2675                  * Control reaches here with cgroup_mutex held.
2676                  * @inode->i_mutex should nest outside cgroup_mutex but we
2677                  * want to populate it immediately without releasing
2678                  * cgroup_mutex.  As @inode isn't visible to anyone else
2679                  * yet, trylock will always succeed without affecting
2680                  * lockdep checks.
2681                  */
2682                 WARN_ON_ONCE(!mutex_trylock(&inode->i_mutex));
2683         } else if (S_ISREG(mode)) {
2684                 inode->i_size = 0;
2685                 inode->i_fop = &cgroup_file_operations;
2686                 inode->i_op = &cgroup_file_inode_operations;
2687         }
2688         d_instantiate(dentry, inode);
2689         dget(dentry);   /* Extra count - pin the dentry in core */
2690         return 0;
2691 }
2692 
2693 /**
2694  * cgroup_file_mode - deduce file mode of a control file
2695  * @cft: the control file in question
2696  *
2697  * returns cft->mode if ->mode is not 0
2698  * returns S_IRUGO|S_IWUSR if it has both a read and a write handler
2699  * returns S_IRUGO if it has only a read handler
2700  * returns S_IWUSR if it has only a write hander
2701  */
2702 static umode_t cgroup_file_mode(const struct cftype *cft)
2703 {
2704         umode_t mode = 0;
2705 
2706         if (cft->mode)
2707                 return cft->mode;
2708 
2709         if (cft->read || cft->read_u64 || cft->read_s64 ||
2710             cft->read_map || cft->read_seq_string)
2711                 mode |= S_IRUGO;
2712 
2713         if (cft->write || cft->write_u64 || cft->write_s64 ||
2714             cft->write_string || cft->trigger)
2715                 mode |= S_IWUSR;
2716 
2717         return mode;
2718 }
2719 
2720 static int cgroup_add_file(struct cgroup *cgrp, struct cftype *cft)
2721 {
2722         struct dentry *dir = cgrp->dentry;
2723         struct cgroup *parent = __d_cgrp(dir);
2724         struct dentry *dentry;
2725         struct cfent *cfe;
2726         int error;
2727         umode_t mode;
2728         char name[MAX_CGROUP_TYPE_NAMELEN + MAX_CFTYPE_NAME + 2] = { 0 };
2729 
2730         if (cft->ss && !(cft->flags & CFTYPE_NO_PREFIX) &&
2731             !(cgrp->root->flags & CGRP_ROOT_NOPREFIX)) {
2732                 strcpy(name, cft->ss->name);
2733                 strcat(name, ".");
2734         }
2735         strcat(name, cft->name);
2736 
2737         BUG_ON(!mutex_is_locked(&dir->d_inode->i_mutex));
2738 
2739         cfe = kzalloc(sizeof(*cfe), GFP_KERNEL);
2740         if (!cfe)
2741                 return -ENOMEM;
2742 
2743         dentry = lookup_one_len(name, dir, strlen(name));
2744         if (IS_ERR(dentry)) {
2745                 error = PTR_ERR(dentry);
2746                 goto out;
2747         }
2748 
2749         cfe->type = (void *)cft;
2750         cfe->dentry = dentry;
2751         dentry->d_fsdata = cfe;
2752         simple_xattrs_init(&cfe->xattrs);
2753 
2754         mode = cgroup_file_mode(cft);
2755         error = cgroup_create_file(dentry, mode | S_IFREG, cgrp->root->sb);
2756         if (!error) {
2757                 list_add_tail(&cfe->node, &parent->files);
2758                 cfe = NULL;
2759         }
2760         dput(dentry);
2761 out:
2762         kfree(cfe);
2763         return error;
2764 }
2765 
2766 /**
2767  * cgroup_addrm_files - add or remove files to a cgroup directory
2768  * @cgrp: the target cgroup
2769  * @cfts: array of cftypes to be added
2770  * @is_add: whether to add or remove
2771  *
2772  * Depending on @is_add, add or remove files defined by @cfts on @cgrp.
2773  * For removals, this function never fails.  If addition fails, this
2774  * function doesn't remove files already added.  The caller is responsible
2775  * for cleaning up.
2776  */
2777 static int cgroup_addrm_files(struct cgroup *cgrp, struct cftype cfts[],
2778                               bool is_add)
2779 {
2780         struct cftype *cft;
2781         int ret;
2782 
2783         lockdep_assert_held(&cgrp->dentry->d_inode->i_mutex);
2784         lockdep_assert_held(&cgroup_mutex);
2785 
2786         for (cft = cfts; cft->name[0] != '\0'; cft++) {
2787                 /* does cft->flags tell us to skip this file on @cgrp? */
2788                 if ((cft->flags & CFTYPE_INSANE) && cgroup_sane_behavior(cgrp))
2789                         continue;
2790                 if ((cft->flags & CFTYPE_NOT_ON_ROOT) && !cgrp->parent)
2791                         continue;
2792                 if ((cft->flags & CFTYPE_ONLY_ON_ROOT) && cgrp->parent)
2793                         continue;
2794 
2795                 if (is_add) {
2796                         ret = cgroup_add_file(cgrp, cft);
2797                         if (ret) {
2798                                 pr_warn("cgroup_addrm_files: failed to add %s, err=%d\n",
2799                                         cft->name, ret);
2800                                 return ret;
2801                         }
2802                 } else {
2803                         cgroup_rm_file(cgrp, cft);
2804                 }
2805         }
2806         return 0;
2807 }
2808 
2809 static void cgroup_cfts_prepare(void)
2810         __acquires(&cgroup_mutex)
2811 {
2812         /*
2813          * Thanks to the entanglement with vfs inode locking, we can't walk
2814          * the existing cgroups under cgroup_mutex and create files.
2815          * Instead, we use css_for_each_descendant_pre() and drop RCU read
2816          * lock before calling cgroup_addrm_files().
2817          */
2818         mutex_lock(&cgroup_mutex);
2819 }
2820 
2821 static int cgroup_cfts_commit(struct cftype *cfts, bool is_add)
2822         __releases(&cgroup_mutex)
2823 {
2824         LIST_HEAD(pending);
2825         struct cgroup_subsys *ss = cfts[0].ss;
2826         struct cgroup *root = &ss->root->top_cgroup;
2827         struct super_block *sb = ss->root->sb;
2828         struct dentry *prev = NULL;
2829         struct inode *inode;
2830         struct cgroup_subsys_state *css;
2831         u64 update_before;
2832         int ret = 0;
2833 
2834         /* %NULL @cfts indicates abort and don't bother if @ss isn't attached */
2835         if (!cfts || ss->root == &cgroup_dummy_root ||
2836             !atomic_inc_not_zero(&sb->s_active)) {
2837                 mutex_unlock(&cgroup_mutex);
2838                 return 0;
2839         }
2840 
2841         /*
2842          * All cgroups which are created after we drop cgroup_mutex will
2843          * have the updated set of files, so we only need to update the
2844          * cgroups created before the current @cgroup_serial_nr_next.
2845          */
2846         update_before = cgroup_serial_nr_next;
2847 
2848         mutex_unlock(&cgroup_mutex);
2849 
2850         /* add/rm files for all cgroups created before */
2851         rcu_read_lock();
2852         css_for_each_descendant_pre(css, cgroup_css(root, ss)) {
2853                 struct cgroup *cgrp = css->cgroup;
2854 
2855                 if (cgroup_is_dead(cgrp))
2856                         continue;
2857 
2858                 inode = cgrp->dentry->d_inode;
2859                 dget(cgrp->dentry);
2860                 rcu_read_unlock();
2861 
2862                 dput(prev);
2863                 prev = cgrp->dentry;
2864 
2865                 mutex_lock(&inode->i_mutex);
2866                 mutex_lock(&cgroup_mutex);
2867                 if (cgrp->serial_nr < update_before && !cgroup_is_dead(cgrp))
2868                         ret = cgroup_addrm_files(cgrp, cfts, is_add);
2869                 mutex_unlock(&cgroup_mutex);
2870                 mutex_unlock(&inode->i_mutex);
2871 
2872                 rcu_read_lock();
2873                 if (ret)
2874                         break;
2875         }
2876         rcu_read_unlock();
2877         dput(prev);
2878         deactivate_super(sb);
2879         return ret;
2880 }
2881 
2882 /**
2883  * cgroup_add_cftypes - add an array of cftypes to a subsystem
2884  * @ss: target cgroup subsystem
2885  * @cfts: zero-length name terminated array of cftypes
2886  *
2887  * Register @cfts to @ss.  Files described by @cfts are created for all
2888  * existing cgroups to which @ss is attached and all future cgroups will
2889  * have them too.  This function can be called anytime whether @ss is
2890  * attached or not.
2891  *
2892  * Returns 0 on successful registration, -errno on failure.  Note that this
2893  * function currently returns 0 as long as @cfts registration is successful
2894  * even if some file creation attempts on existing cgroups fail.
2895  */
2896 int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
2897 {
2898         struct cftype_set *set;
2899         struct cftype *cft;
2900         int ret;
2901 
2902         set = kzalloc(sizeof(*set), GFP_KERNEL);
2903         if (!set)
2904                 return -ENOMEM;
2905 
2906         for (cft = cfts; cft->name[0] != '\0'; cft++)
2907                 cft->ss = ss;
2908 
2909         cgroup_cfts_prepare();
2910         set->cfts = cfts;
2911         list_add_tail(&set->node, &ss->cftsets);
2912         ret = cgroup_cfts_commit(cfts, true);
2913         if (ret)
2914                 cgroup_rm_cftypes(cfts);
2915         return ret;
2916 }
2917 EXPORT_SYMBOL_GPL(cgroup_add_cftypes);
2918 
2919 /**
2920  * cgroup_rm_cftypes - remove an array of cftypes from a subsystem
2921  * @cfts: zero-length name terminated array of cftypes
2922  *
2923  * Unregister @cfts.  Files described by @cfts are removed from all
2924  * existing cgroups and all future cgroups won't have them either.  This
2925  * function can be called anytime whether @cfts' subsys is attached or not.
2926  *
2927  * Returns 0 on successful unregistration, -ENOENT if @cfts is not
2928  * registered.
2929  */
2930 int cgroup_rm_cftypes(struct cftype *cfts)
2931 {
2932         struct cftype_set *set;
2933 
2934         if (!cfts || !cfts[0].ss)
2935                 return -ENOENT;
2936 
2937         cgroup_cfts_prepare();
2938 
2939         list_for_each_entry(set, &cfts[0].ss->cftsets, node) {
2940                 if (set->cfts == cfts) {
2941                         list_del(&set->node);
2942                         kfree(set);
2943                         cgroup_cfts_commit(cfts, false);
2944                         return 0;
2945                 }
2946         }
2947 
2948         cgroup_cfts_commit(NULL, false);
2949         return -ENOENT;
2950 }
2951 
2952 /**
2953  * cgroup_task_count - count the number of tasks in a cgroup.
2954  * @cgrp: the cgroup in question
2955  *
2956  * Return the number of tasks in the cgroup.
2957  */
2958 int cgroup_task_count(const struct cgroup *cgrp)
2959 {
2960         int count = 0;
2961         struct cgrp_cset_link *link;
2962 
2963         read_lock(&css_set_lock);
2964         list_for_each_entry(link, &cgrp->cset_links, cset_link)
2965                 count += atomic_read(&link->cset->refcount);
2966         read_unlock(&css_set_lock);
2967         return count;
2968 }
2969 
2970 /*
2971  * To reduce the fork() overhead for systems that are not actually using
2972  * their cgroups capability, we don't maintain the lists running through
2973  * each css_set to its tasks until we see the list actually used - in other
2974  * words after the first call to css_task_iter_start().
2975  */
2976 static void cgroup_enable_task_cg_lists(void)
2977 {
2978         struct task_struct *p, *g;
2979         write_lock(&css_set_lock);
2980         use_task_css_set_links = 1;
2981         /*
2982          * We need tasklist_lock because RCU is not safe against
2983          * while_each_thread(). Besides, a forking task that has passed
2984          * cgroup_post_fork() without seeing use_task_css_set_links = 1
2985          * is not guaranteed to have its child immediately visible in the
2986          * tasklist if we walk through it with RCU.
2987          */
2988         read_lock(&tasklist_lock);
2989         do_each_thread(g, p) {
2990                 task_lock(p);
2991                 /*
2992                  * We should check if the process is exiting, otherwise
2993                  * it will race with cgroup_exit() in that the list
2994                  * entry won't be deleted though the process has exited.
2995                  */
2996                 if (!(p->flags & PF_EXITING) && list_empty(&p->cg_list))
2997                         list_add(&p->cg_list, &task_css_set(p)->tasks);
2998                 task_unlock(p);
2999         } while_each_thread(g, p);
3000         read_unlock(&tasklist_lock);
3001         write_unlock(&css_set_lock);
3002 }
3003 
3004 /**
3005  * css_next_child - find the next child of a given css
3006  * @pos_css: the current position (%NULL to initiate traversal)
3007  * @parent_css: css whose children to walk
3008  *
3009  * This function returns the next child of @parent_css and should be called
3010  * under RCU read lock.  The only requirement is that @parent_css and
3011  * @pos_css are accessible.  The next sibling is guaranteed to be returned
3012  * regardless of their states.
3013  */
3014 struct cgroup_subsys_state *
3015 css_next_child(struct cgroup_subsys_state *pos_css,
3016                struct cgroup_subsys_state *parent_css)
3017 {
3018         struct cgroup *pos = pos_css ? pos_css->cgroup : NULL;
3019         struct cgroup *cgrp = parent_css->cgroup;
3020         struct cgroup *next;
3021 
3022         WARN_ON_ONCE(!rcu_read_lock_held());
3023 
3024         /*
3025          * @pos could already have been removed.  Once a cgroup is removed,
3026          * its ->sibling.next is no longer updated when its next sibling
3027          * changes.  As CGRP_DEAD assertion is serialized and happens
3028          * before the cgroup is taken off the ->sibling list, if we see it
3029          * unasserted, it's guaranteed that the next sibling hasn't
3030          * finished its grace period even if it's already removed, and thus
3031          * safe to dereference from this RCU critical section.  If
3032          * ->sibling.next is inaccessible, cgroup_is_dead() is guaranteed
3033          * to be visible as %true here.
3034          *
3035          * If @pos is dead, its next pointer can't be dereferenced;
3036          * however, as each cgroup is given a monotonically increasing
3037          * unique serial number and always appended to the sibling list,
3038          * the next one can be found by walking the parent's children until
3039          * we see a cgroup with higher serial number than @pos's.  While
3040          * this path can be slower, it's taken only when either the current
3041          * cgroup is removed or iteration and removal race.
3042          */
3043         if (!pos) {
3044                 next = list_entry_rcu(cgrp->children.next, struct cgroup, sibling);
3045         } else if (likely(!cgroup_is_dead(pos))) {
3046                 next = list_entry_rcu(pos->sibling.next, struct cgroup, sibling);
3047         } else {
3048                 list_for_each_entry_rcu(next, &cgrp->children, sibling)
3049                         if (next->serial_nr > pos->serial_nr)
3050                                 break;
3051         }
3052 
3053         if (&next->sibling == &cgrp->children)
3054                 return NULL;
3055 
3056         return cgroup_css(next, parent_css->ss);
3057 }
3058 EXPORT_SYMBOL_GPL(css_next_child);
3059 
3060 /**
3061  * css_next_descendant_pre - find the next descendant for pre-order walk
3062  * @pos: the current position (%NULL to initiate traversal)
3063  * @root: css whose descendants to walk
3064  *
3065  * To be used by css_for_each_descendant_pre().  Find the next descendant
3066  * to visit for pre-order traversal of @root's descendants.  @root is
3067  * included in the iteration and the first node to be visited.
3068  *
3069  * While this function requires RCU read locking, it doesn't require the
3070  * whole traversal to be contained in a single RCU critical section.  This
3071  * function will return the correct next descendant as long as both @pos
3072  * and @root are accessible and @pos is a descendant of @root.
3073  */
3074 struct cgroup_subsys_state *
3075 css_next_descendant_pre(struct cgroup_subsys_state *pos,
3076                         struct cgroup_subsys_state *root)
3077 {
3078         struct cgroup_subsys_state *next;
3079 
3080         WARN_ON_ONCE(!rcu_read_lock_held());
3081 
3082         /* if first iteration, visit @root */
3083         if (!pos)
3084                 return root;
3085 
3086         /* visit the first child if exists */
3087         next = css_next_child(NULL, pos);
3088         if (next)
3089                 return next;
3090 
3091         /* no child, visit my or the closest ancestor's next sibling */
3092         while (pos != root) {
3093                 next = css_next_child(pos, css_parent(pos));
3094                 if (next)
3095                         return next;
3096                 pos = css_parent(pos);
3097         }
3098 
3099         return NULL;
3100 }
3101 EXPORT_SYMBOL_GPL(css_next_descendant_pre);
3102 
3103 /**
3104  * css_rightmost_descendant - return the rightmost descendant of a css
3105  * @pos: css of interest
3106  *
3107  * Return the rightmost descendant of @pos.  If there's no descendant, @pos
3108  * is returned.  This can be used during pre-order traversal to skip
3109  * subtree of @pos.
3110  *
3111  * While this function requires RCU read locking, it doesn't require the
3112  * whole traversal to be contained in a single RCU critical section.  This
3113  * function will return the correct rightmost descendant as long as @pos is
3114  * accessible.
3115  */
3116 struct cgroup_subsys_state *
3117 css_rightmost_descendant(struct cgroup_subsys_state *pos)
3118 {
3119         struct cgroup_subsys_state *last, *tmp;
3120 
3121         WARN_ON_ONCE(!rcu_read_lock_held());
3122 
3123         do {
3124                 last = pos;
3125                 /* ->prev isn't RCU safe, walk ->next till the end */
3126                 pos = NULL;
3127                 css_for_each_child(tmp, last)
3128                         pos = tmp;
3129         } while (pos);
3130 
3131         return last;
3132 }
3133 EXPORT_SYMBOL_GPL(css_rightmost_descendant);
3134 
3135 static struct cgroup_subsys_state *
3136 css_leftmost_descendant(struct cgroup_subsys_state *pos)
3137 {
3138         struct cgroup_subsys_state *last;
3139 
3140         do {
3141                 last = pos;
3142                 pos = css_next_child(NULL, pos);
3143         } while (pos);
3144 
3145         return last;
3146 }
3147 
3148 /**
3149  * css_next_descendant_post - find the next descendant for post-order walk
3150  * @pos: the current position (%NULL to initiate traversal)
3151  * @root: css whose descendants to walk
3152  *
3153  * To be used by css_for_each_descendant_post().  Find the next descendant
3154  * to visit for post-order traversal of @root's descendants.  @root is
3155  * included in the iteration and the last node to be visited.
3156  *
3157  * While this function requires RCU read locking, it doesn't require the
3158  * whole traversal to be contained in a single RCU critical section.  This
3159  * function will return the correct next descendant as long as both @pos
3160  * and @cgroup are accessible and @pos is a descendant of @cgroup.
3161  */
3162 struct cgroup_subsys_state *
3163 css_next_descendant_post(struct cgroup_subsys_state *pos,
3164                          struct cgroup_subsys_state *root)
3165 {
3166         struct cgroup_subsys_state *next;
3167 
3168         WARN_ON_ONCE(!rcu_read_lock_held());
3169 
3170         /* if first iteration, visit leftmost descendant which may be @root */
3171         if (!pos)
3172                 return css_leftmost_descendant(root);
3173 
3174         /* if we visited @root, we're done */
3175         if (pos == root)
3176                 return NULL;
3177 
3178         /* if there's an unvisited sibling, visit its leftmost descendant */
3179         next = css_next_child(pos, css_parent(pos));
3180         if (next)
3181                 return css_leftmost_descendant(next);
3182 
3183         /* no sibling left, visit parent */
3184         return css_parent(pos);
3185 }
3186 EXPORT_SYMBOL_GPL(css_next_descendant_post);
3187 
3188 /**
3189  * css_advance_task_iter - advance a task itererator to the next css_set
3190  * @it: the iterator to advance
3191  *
3192  * Advance @it to the next css_set to walk.
3193  */
3194 static void css_advance_task_iter(struct css_task_iter *it)
3195 {
3196         struct list_head *l = it->cset_link;
3197         struct cgrp_cset_link *link;
3198         struct css_set *cset;
3199 
3200         /* Advance to the next non-empty css_set */
3201         do {
3202                 l = l->next;
3203                 if (l == &it->origin_css->cgroup->cset_links) {
3204                         it->cset_link = NULL;
3205                         return;
3206                 }
3207                 link = list_entry(l, struct cgrp_cset_link, cset_link);
3208                 cset = link->cset;
3209         } while (list_empty(&cset->tasks));
3210         it->cset_link = l;
3211         it->task = cset->tasks.next;
3212 }
3213 
3214 /**
3215  * css_task_iter_start - initiate task iteration
3216  * @css: the css to walk tasks of
3217  * @it: the task iterator to use
3218  *
3219  * Initiate iteration through the tasks of @css.  The caller can call
3220  * css_task_iter_next() to walk through the tasks until the function
3221  * returns NULL.  On completion of iteration, css_task_iter_end() must be
3222  * called.
3223  *
3224  * Note that this function acquires a lock which is released when the
3225  * iteration finishes.  The caller can't sleep while iteration is in
3226  * progress.
3227  */
3228 void css_task_iter_start(struct cgroup_subsys_state *css,
3229                          struct css_task_iter *it)
3230         __acquires(css_set_lock)
3231 {
3232         /*
3233          * The first time anyone tries to iterate across a css, we need to
3234          * enable the list linking each css_set to its tasks, and fix up
3235          * all existing tasks.
3236          */
3237         if (!use_task_css_set_links)
3238                 cgroup_enable_task_cg_lists();
3239 
3240         read_lock(&css_set_lock);
3241 
3242         it->origin_css = css;
3243         it->cset_link = &css->cgroup->cset_links;
3244 
3245         css_advance_task_iter(it);
3246 }
3247 
3248 /**
3249  * css_task_iter_next - return the next task for the iterator
3250  * @it: the task iterator being iterated
3251  *
3252  * The "next" function for task iteration.  @it should have been
3253  * initialized via css_task_iter_start().  Returns NULL when the iteration
3254  * reaches the end.
3255  */
3256 struct task_struct *css_task_iter_next(struct css_task_iter *it)
3257 {
3258         struct task_struct *res;
3259         struct list_head *l = it->task;
3260         struct cgrp_cset_link *link;
3261 
3262         /* If the iterator cg is NULL, we have no tasks */
3263         if (!it->cset_link)
3264                 return NULL;
3265         res = list_entry(l, struct task_struct, cg_list);
3266         /* Advance iterator to find next entry */
3267         l = l->next;
3268         link = list_entry(it->cset_link, struct cgrp_cset_link, cset_link);
3269         if (l == &link->cset->tasks) {
3270                 /*
3271                  * We reached the end of this task list - move on to the
3272                  * next cgrp_cset_link.
3273                  */
3274                 css_advance_task_iter(it);
3275         } else {
3276                 it->task = l;
3277         }
3278         return res;
3279 }
3280 
3281 /**
3282  * css_task_iter_end - finish task iteration
3283  * @it: the task iterator to finish
3284  *
3285  * Finish task iteration started by css_task_iter_start().
3286  */
3287 void css_task_iter_end(struct css_task_iter *it)
3288         __releases(css_set_lock)
3289 {
3290         read_unlock(&css_set_lock);
3291 }
3292 
3293 static inline int started_after_time(struct task_struct *t1,
3294                                      struct timespec *time,
3295                                      struct task_struct *t2)
3296 {
3297         int start_diff = timespec_compare(&t1->start_time, time);
3298         if (start_diff > 0) {
3299                 return 1;
3300         } else if (start_diff < 0) {
3301                 return 0;
3302         } else {
3303                 /*
3304                  * Arbitrarily, if two processes started at the same
3305                  * time, we'll say that the lower pointer value
3306                  * started first. Note that t2 may have exited by now
3307                  * so this may not be a valid pointer any longer, but
3308                  * that's fine - it still serves to distinguish
3309                  * between two tasks started (effectively) simultaneously.
3310                  */
3311                 return t1 > t2;
3312         }
3313 }
3314 
3315 /*
3316  * This function is a callback from heap_insert() and is used to order
3317  * the heap.
3318  * In this case we order the heap in descending task start time.
3319  */
3320 static inline int started_after(void *p1, void *p2)
3321 {
3322         struct task_struct *t1 = p1;
3323         struct task_struct *t2 = p2;
3324         return started_after_time(t1, &t2->start_time, t2);
3325 }
3326 
3327 /**
3328  * css_scan_tasks - iterate though all the tasks in a css
3329  * @css: the css to iterate tasks of
3330  * @test: optional test callback
3331  * @process: process callback
3332  * @data: data passed to @test and @process
3333  * @heap: optional pre-allocated heap used for task iteration
3334  *
3335  * Iterate through all the tasks in @css, calling @test for each, and if it
3336  * returns %true, call @process for it also.
3337  *
3338  * @test may be NULL, meaning always true (select all tasks), which
3339  * effectively duplicates css_task_iter_{start,next,end}() but does not
3340  * lock css_set_lock for the call to @process.
3341  *
3342  * It is guaranteed that @process will act on every task that is a member
3343  * of @css for the duration of this call.  This function may or may not
3344  * call @process for tasks that exit or move to a different css during the
3345  * call, or are forked or move into the css during the call.
3346  *
3347  * Note that @test may be called with locks held, and may in some
3348  * situations be called multiple times for the same task, so it should be
3349  * cheap.
3350  *
3351  * If @heap is non-NULL, a heap has been pre-allocated and will be used for
3352  * heap operations (and its "gt" member will be overwritten), else a
3353  * temporary heap will be used (allocation of which may cause this function
3354  * to fail).
3355  */
3356 int css_scan_tasks(struct cgroup_subsys_state *css,
3357                    bool (*test)(struct task_struct *, void *),
3358                    void (*process)(struct task_struct *, void *),
3359                    void *data, struct ptr_heap *heap)
3360 {
3361         int retval, i;
3362         struct css_task_iter it;
3363         struct task_struct *p, *dropped;
3364         /* Never dereference latest_task, since it's not refcounted */
3365         struct task_struct *latest_task = NULL;
3366         struct ptr_heap tmp_heap;
3367         struct timespec latest_time = { 0, 0 };
3368 
3369         if (heap) {
3370                 /* The caller supplied our heap and pre-allocated its memory */
3371                 heap->gt = &started_after;
3372         } else {
3373                 /* We need to allocate our own heap memory */
3374                 heap = &tmp_heap;
3375                 retval = heap_init(heap, PAGE_SIZE, GFP_KERNEL, &started_after);
3376                 if (retval)
3377                         /* cannot allocate the heap */
3378                         return retval;
3379         }
3380 
3381  again:
3382         /*
3383          * Scan tasks in the css, using the @test callback to determine
3384          * which are of interest, and invoking @process callback on the
3385          * ones which need an update.  Since we don't want to hold any
3386          * locks during the task updates, gather tasks to be processed in a
3387          * heap structure.  The heap is sorted by descending task start
3388          * time.  If the statically-sized heap fills up, we overflow tasks
3389          * that started later, and in future iterations only consider tasks
3390          * that started after the latest task in the previous pass. This
3391          * guarantees forward progress and that we don't miss any tasks.
3392          */
3393         heap->size = 0;
3394         css_task_iter_start(css, &it);
3395         while ((p = css_task_iter_next(&it))) {
3396                 /*
3397                  * Only affect tasks that qualify per the caller's callback,
3398                  * if he provided one
3399                  */
3400                 if (test && !test(p, data))
3401                         continue;
3402                 /*
3403                  * Only process tasks that started after the last task
3404                  * we processed
3405                  */
3406                 if (!started_after_time(p, &latest_time, latest_task))
3407                         continue;
3408                 dropped = heap_insert(heap, p);
3409                 if (dropped == NULL) {
3410                         /*
3411                          * The new task was inserted; the heap wasn't
3412                          * previously full
3413                          */
3414                         get_task_struct(p);
3415                 } else if (dropped != p) {
3416                         /*
3417                          * The new task was inserted, and pushed out a
3418                          * different task
3419                          */
3420                         get_task_struct(p);
3421                         put_task_struct(dropped);
3422                 }
3423                 /*
3424                  * Else the new task was newer than anything already in
3425                  * the heap and wasn't inserted
3426                  */
3427         }
3428         css_task_iter_end(&it);
3429 
3430         if (heap->size) {
3431                 for (i = 0; i < heap->size; i++) {
3432                         struct task_struct *q = heap->ptrs[i];
3433                         if (i == 0) {
3434                                 latest_time = q->start_time;
3435                                 latest_task = q;
3436                         }
3437                         /* Process the task per the caller's callback */
3438                         process(q, data);
3439                         put_task_struct(q);
3440                 }
3441                 /*
3442                  * If we had to process any tasks at all, scan again
3443                  * in case some of them were in the middle of forking
3444                  * children that didn't get processed.
3445                  * Not the most efficient way to do it, but it avoids
3446                  * having to take callback_mutex in the fork path
3447                  */
3448                 goto again;
3449         }
3450         if (heap == &tmp_heap)
3451                 heap_free(&tmp_heap);
3452         return 0;
3453 }
3454 
3455 static void cgroup_transfer_one_task(struct task_struct *task, void *data)
3456 {
3457         struct cgroup *new_cgroup = data;
3458 
3459         mutex_lock(&cgroup_mutex);
3460         cgroup_attach_task(new_cgroup, task, false);
3461         mutex_unlock(&cgroup_mutex);
3462 }
3463 
3464 /**
3465  * cgroup_trasnsfer_tasks - move tasks from one cgroup to another
3466  * @to: cgroup to which the tasks will be moved
3467  * @from: cgroup in which the tasks currently reside
3468  */
3469 int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from)
3470 {
3471         return css_scan_tasks(&from->dummy_css, NULL, cgroup_transfer_one_task,
3472                               to, NULL);
3473 }
3474 
3475 /*
3476  * Stuff for reading the 'tasks'/'procs' files.
3477  *
3478  * Reading this file can return large amounts of data if a cgroup has
3479  * *lots* of attached tasks. So it may need several calls to read(),
3480  * but we cannot guarantee that the information we produce is correct
3481  * unless we produce it entirely atomically.
3482  *
3483  */
3484 
3485 /* which pidlist file are we talking about? */
3486 enum cgroup_filetype {
3487         CGROUP_FILE_PROCS,
3488         CGROUP_FILE_TASKS,
3489 };
3490 
3491 /*
3492  * A pidlist is a list of pids that virtually represents the contents of one
3493  * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists,
3494  * a pair (one each for procs, tasks) for each pid namespace that's relevant
3495  * to the cgroup.
3496  */
3497 struct cgroup_pidlist {
3498         /*
3499          * used to find which pidlist is wanted. doesn't change as long as
3500          * this particular list stays in the list.
3501         */
3502         struct { enum cgroup_filetype type; struct pid_namespace *ns; } key;
3503         /* array of xids */
3504         pid_t *list;
3505         /* how many elements the above list has */
3506         int length;
3507         /* how many files are using the current array */
3508         int use_count;
3509         /* each of these stored in a list by its cgroup */
3510         struct list_head links;
3511         /* pointer to the cgroup we belong to, for list removal purposes */
3512         struct cgroup *owner;
3513         /* protects the other fields */
3514         struct rw_semaphore rwsem;
3515 };
3516 
3517 /*
3518  * The following two functions "fix" the issue where there are more pids
3519  * than kmalloc will give memory for; in such cases, we use vmalloc/vfree.
3520  * TODO: replace with a kernel-wide solution to this problem
3521  */
3522 #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2))
3523 static void *pidlist_allocate(int count)
3524 {
3525         if (PIDLIST_TOO_LARGE(count))
3526                 return vmalloc(count * sizeof(pid_t));
3527         else
3528                 return kmalloc(count * sizeof(pid_t), GFP_KERNEL);
3529 }
3530 static void pidlist_free(void *p)
3531 {
3532         if (is_vmalloc_addr(p))
3533                 vfree(p);
3534         else
3535                 kfree(p);
3536 }
3537 
3538 /*
3539  * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
3540  * Returns the number of unique elements.
3541  */
3542 static int pidlist_uniq(pid_t *list, int length)
3543 {
3544         int src, dest = 1;
3545 
3546         /*
3547          * we presume the 0th element is unique, so i starts at 1. trivial
3548          * edge cases first; no work needs to be done for either
3549          */
3550         if (length == 0 || length == 1)
3551                 return length;
3552         /* src and dest walk down the list; dest counts unique elements */
3553         for (src = 1; src < length; src++) {
3554                 /* find next unique element */
3555                 while (list[src] == list[src-1]) {
3556                         src++;
3557                         if (src == length)
3558                                 goto after;
3559                 }
3560                 /* dest always points to where the next unique element goes */
3561                 list[dest] = list[src];
3562                 dest++;
3563         }
3564 after:
3565         return dest;
3566 }
3567 
3568 static int cmppid(const void *a, const void *b)
3569 {
3570         return *(pid_t *)a - *(pid_t *)b;
3571 }
3572 
3573 /*
3574  * find the appropriate pidlist for our purpose (given procs vs tasks)
3575  * returns with the lock on that pidlist already held, and takes care
3576  * of the use count, or returns NULL with no locks held if we're out of
3577  * memory.
3578  */
3579 static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp,
3580                                                   enum cgroup_filetype type)
3581 {
3582         struct cgroup_pidlist *l;
3583         /* don't need task_nsproxy() if we're looking at ourself */
3584         struct pid_namespace *ns = task_active_pid_ns(current);
3585 
3586         /*
3587          * We can't drop the pidlist_mutex before taking the l->rwsem in case
3588          * the last ref-holder is trying to remove l from the list at the same
3589          * time. Holding the pidlist_mutex precludes somebody taking whichever
3590          * list we find out from under us - compare release_pid_array().
3591          */
3592         mutex_lock(&cgrp->pidlist_mutex);
3593         list_for_each_entry(l, &cgrp->pidlists, links) {
3594                 if (l->key.type == type && l->key.ns == ns) {
3595                         /* make sure l doesn't vanish out from under us */
3596                         down_write(&l->rwsem);
3597                         mutex_unlock(&cgrp->pidlist_mutex);
3598                         return l;
3599                 }
3600         }
3601         /* entry not found; create a new one */
3602         l = kzalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL);
3603         if (!l) {
3604                 mutex_unlock(&cgrp->pidlist_mutex);
3605                 return l;
3606         }
3607         init_rwsem(&l->rwsem);
3608         down_write(&l->rwsem);
3609         l->key.type = type;
3610         l->key.ns = get_pid_ns(ns);
3611         l->owner = cgrp;
3612         list_add(&l->links, &cgrp->pidlists);
3613         mutex_unlock(&cgrp->pidlist_mutex);
3614         return l;
3615 }
3616 
3617 /*
3618  * Load a cgroup's pidarray with either procs' tgids or tasks' pids
3619  */
3620 static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type,
3621                               struct cgroup_pidlist **lp)
3622 {
3623         pid_t *array;
3624         int length;
3625         int pid, n = 0; /* used for populating the array */
3626         struct css_task_iter it;
3627         struct task_struct *tsk;
3628         struct cgroup_pidlist *l;
3629 
3630         /*
3631          * If cgroup gets more users after we read count, we won't have
3632          * enough space - tough.  This race is indistinguishable to the
3633          * caller from the case that the additional cgroup users didn't
3634          * show up until sometime later on.
3635          */
3636         length = cgroup_task_count(cgrp);
3637         array = pidlist_allocate(length);
3638         if (!array)
3639                 return -ENOMEM;
3640         /* now, populate the array */
3641         css_task_iter_start(&cgrp->dummy_css, &it);
3642         while ((tsk = css_task_iter_next(&it))) {
3643                 if (unlikely(n == length))
3644                         break;
3645                 /* get tgid or pid for procs or tasks file respectively */
3646                 if (type == CGROUP_FILE_PROCS)
3647                         pid = task_tgid_vnr(tsk);
3648                 else
3649                         pid = task_pid_vnr(tsk);
3650                 if (pid > 0) /* make sure to only use valid results */
3651                         array[n++] = pid;
3652         }
3653         css_task_iter_end(&it);
3654         length = n;
3655         /* now sort & (if procs) strip out duplicates */
3656         sort(array, length, sizeof(pid_t), cmppid, NULL);
3657         if (type == CGROUP_FILE_PROCS)
3658                 length = pidlist_uniq(array, length);
3659         l = cgroup_pidlist_find(cgrp, type);
3660         if (!l) {
3661                 pidlist_free(array);
3662                 return -ENOMEM;
3663         }
3664         /* store array, freeing old if necessary - lock already held */
3665         pidlist_free(l->list);
3666         l->list = array;
3667         l->length = length;
3668         l->use_count++;
3669         up_write(&l->rwsem);
3670         *lp = l;
3671         return 0;
3672 }
3673 
3674 /**
3675  * cgroupstats_build - build and fill cgroupstats
3676  * @stats: cgroupstats to fill information into
3677  * @dentry: A dentry entry belonging to the cgroup for which stats have
3678  * been requested.
3679  *
3680  * Build and fill cgroupstats so that taskstats can export it to user
3681  * space.
3682  */
3683 int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry)
3684 {
3685         int ret = -EINVAL;
3686         struct cgroup *cgrp;
3687         struct css_task_iter it;
3688         struct task_struct *tsk;
3689 
3690         /*
3691          * Validate dentry by checking the superblock operations,
3692          * and make sure it's a directory.
3693          */
3694         if (dentry->d_sb->s_op != &cgroup_ops ||
3695             !S_ISDIR(dentry->d_inode->i_mode))
3696                  goto err;
3697 
3698         ret = 0;
3699         cgrp = dentry->d_fsdata;
3700 
3701         css_task_iter_start(&cgrp->dummy_css, &it);
3702         while ((tsk = css_task_iter_next(&it))) {
3703                 switch (tsk->state) {
3704                 case TASK_RUNNING:
3705                         stats->nr_running++;
3706                         break;
3707                 case TASK_INTERRUPTIBLE:
3708                         stats->nr_sleeping++;
3709                         break;
3710                 case TASK_UNINTERRUPTIBLE:
3711                         stats->nr_uninterruptible++;
3712                         break;
3713                 case TASK_STOPPED:
3714                         stats->nr_stopped++;
3715                         break;
3716                 default:
3717                         if (delayacct_is_task_waiting_on_io(tsk))
3718                                 stats->nr_io_wait++;
3719                         break;
3720                 }
3721         }
3722         css_task_iter_end(&it);
3723 
3724 err:
3725         return ret;
3726 }
3727 
3728 
3729 /*
3730  * seq_file methods for the tasks/procs files. The seq_file position is the
3731  * next pid to display; the seq_file iterator is a pointer to the pid
3732  * in the cgroup->l->list array.
3733  */
3734 
3735 static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos)
3736 {
3737         /*
3738          * Initially we receive a position value that corresponds to
3739          * one more than the last pid shown (or 0 on the first call or
3740          * after a seek to the start). Use a binary-search to find the
3741          * next pid to display, if any
3742          */
3743         struct cgroup_pidlist *l = s->private;
3744         int index = 0, pid = *pos;
3745         int *iter;
3746 
3747         down_read(&l->rwsem);
3748         if (pid) {
3749                 int end = l->length;
3750 
3751                 while (index < end) {
3752                         int mid = (index + end) / 2;
3753                         if (l->list[mid] == pid) {
3754                                 index = mid;
3755                                 break;
3756                         } else if (l->list[mid] <= pid)
3757                                 index = mid + 1;
3758                         else
3759                                 end = mid;
3760                 }
3761         }
3762         /* If we're off the end of the array, we're done */
3763         if (index >= l->length)
3764                 return NULL;
3765         /* Update the abstract position to be the actual pid that we found */
3766         iter = l->list + index;
3767         *pos = *iter;
3768         return iter;
3769 }
3770 
3771 static void cgroup_pidlist_stop(struct seq_file *s, void *v)
3772 {
3773         struct cgroup_pidlist *l = s->private;
3774         up_read(&l->rwsem);
3775 }
3776 
3777 static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos)
3778 {
3779         struct cgroup_pidlist *l = s->private;
3780         pid_t *p = v;
3781         pid_t *end = l->list + l->length;
3782         /*
3783          * Advance to the next pid in the array. If this goes off the
3784          * end, we're done
3785          */
3786         p++;
3787         if (p >= end) {
3788                 return NULL;
3789         } else {
3790                 *pos = *p;
3791                 return p;
3792         }
3793 }
3794 
3795 static int cgroup_pidlist_show(struct seq_file *s, void *v)
3796 {
3797         return seq_printf(s, "%d\n", *(int *)v);
3798 }
3799 
3800 /*
3801  * seq_operations functions for iterating on pidlists through seq_file -
3802  * independent of whether it's tasks or procs
3803  */
3804 static const struct seq_operations cgroup_pidlist_seq_operations = {
3805         .start = cgroup_pidlist_start,
3806         .stop = cgroup_pidlist_stop,
3807         .next = cgroup_pidlist_next,
3808         .show = cgroup_pidlist_show,
3809 };
3810 
3811 static void cgroup_release_pid_array(struct cgroup_pidlist *l)
3812 {
3813         /*
3814          * the case where we're the last user of this particular pidlist will
3815          * have us remove it from the cgroup's list, which entails taking the
3816          * mutex. since in pidlist_find the pidlist->lock depends on cgroup->
3817          * pidlist_mutex, we have to take pidlist_mutex first.
3818          */
3819         mutex_lock(&l->owner->pidlist_mutex);
3820         down_write(&l->rwsem);
3821         BUG_ON(!l->use_count);
3822         if (!--l->use_count) {
3823                 /* we're the last user if refcount is 0; remove and free */
3824                 list_del(&l->links);
3825                 mutex_unlock(&l->owner->pidlist_mutex);
3826                 pidlist_free(l->list);
3827                 put_pid_ns(l->key.ns);
3828                 up_write(&l->rwsem);
3829                 kfree(l);
3830                 return;
3831         }
3832         mutex_unlock(&l->owner->pidlist_mutex);
3833         up_write(&l->rwsem);
3834 }
3835 
3836 static int cgroup_pidlist_release(struct inode *inode, struct file *file)
3837 {
3838         struct cgroup_pidlist *l;
3839         if (!(file->f_mode & FMODE_READ))
3840                 return 0;
3841         /*
3842          * the seq_file will only be initialized if the file was opened for
3843          * reading; hence we check if it's not null only in that case.
3844          */
3845         l = ((struct seq_file *)file->private_data)->private;
3846         cgroup_release_pid_array(l);
3847         return seq_release(inode, file);
3848 }
3849 
3850 static const struct file_operations cgroup_pidlist_operations = {
3851         .read = seq_read,
3852         .llseek = seq_lseek,
3853         .write = cgroup_file_write,
3854         .release = cgroup_pidlist_release,
3855 };
3856 
3857 /*
3858  * The following functions handle opens on a file that displays a pidlist
3859  * (tasks or procs). Prepare an array of the process/thread IDs of whoever's
3860  * in the cgroup.
3861  */
3862 /* helper function for the two below it */
3863 static int cgroup_pidlist_open(struct file *file, enum cgroup_filetype type)
3864 {
3865         struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
3866         struct cgroup_pidlist *l;
3867         int retval;
3868 
3869         /* Nothing to do for write-only files */
3870         if (!(file->f_mode & FMODE_READ))
3871                 return 0;
3872 
3873         /* have the array populated */
3874         retval = pidlist_array_load(cgrp, type, &l);
3875         if (retval)
3876                 return retval;
3877         /* configure file information */
3878         file->f_op = &cgroup_pidlist_operations;
3879 
3880         retval = seq_open(file, &cgroup_pidlist_seq_operations);
3881         if (retval) {
3882                 cgroup_release_pid_array(l);
3883                 return retval;
3884         }
3885         ((struct seq_file *)file->private_data)->private = l;
3886         return 0;
3887 }
3888 static int cgroup_tasks_open(struct inode *unused, struct file *file)
3889 {
3890         return cgroup_pidlist_open(file, CGROUP_FILE_TASKS);
3891 }
3892 static int cgroup_procs_open(struct inode *unused, struct file *file)
3893 {
3894         return cgroup_pidlist_open(file, CGROUP_FILE_PROCS);
3895 }
3896 
3897 static u64 cgroup_read_notify_on_release(struct cgroup_subsys_state *css,
3898                                          struct cftype *cft)
3899 {
3900         return notify_on_release(css->cgroup);
3901 }
3902 
3903 static int cgroup_write_notify_on_release(struct cgroup_subsys_state *css,
3904                                           struct cftype *cft, u64 val)
3905 {
3906         clear_bit(CGRP_RELEASABLE, &css->cgroup->flags);
3907         if (val)
3908                 set_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags);
3909         else
3910                 clear_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags);
3911         return 0;
3912 }
3913 
3914 /*
3915  * When dput() is called asynchronously, if umount has been done and
3916  * then deactivate_super() in cgroup_free_fn() kills the superblock,
3917  * there's a small window that vfs will see the root dentry with non-zero
3918  * refcnt and trigger BUG().
3919  *
3920  * That's why we hold a reference before dput() and drop it right after.
3921  */
3922 static void cgroup_dput(struct cgroup *cgrp)
3923 {
3924         struct super_block *sb = cgrp->root->sb;
3925 
3926         atomic_inc(&sb->s_active);
3927         dput(cgrp->dentry);
3928         deactivate_super(sb);
3929 }
3930 
3931 /*
3932  * Unregister event and free resources.
3933  *
3934  * Gets called from workqueue.
3935  */
3936 static void cgroup_event_remove(struct work_struct *work)
3937 {
3938         struct cgroup_event *event = container_of(work, struct cgroup_event,
3939                         remove);
3940         struct cgroup_subsys_state *css = event->css;
3941 
3942         remove_wait_queue(event->wqh, &event->wait);
3943 
3944         event->cft->unregister_event(css, event->cft, event->eventfd);
3945 
3946         /* Notify userspace the event is going away. */
3947         eventfd_signal(event->eventfd, 1);
3948 
3949         eventfd_ctx_put(event->eventfd);
3950         kfree(event);
3951         css_put(css);
3952 }
3953 
3954 /*
3955  * Gets called on POLLHUP on eventfd when user closes it.
3956  *
3957  * Called with wqh->lock held and interrupts disabled.
3958  */
3959 static int cgroup_event_wake(wait_queue_t *wait, unsigned mode,
3960                 int sync, void *key)
3961 {
3962         struct cgroup_event *event = container_of(wait,
3963                         struct cgroup_event, wait);
3964         struct cgroup *cgrp = event->css->cgroup;
3965         unsigned long flags = (unsigned long)key;
3966 
3967         if (flags & POLLHUP) {
3968                 /*
3969                  * If the event has been detached at cgroup removal, we
3970                  * can simply return knowing the other side will cleanup
3971                  * for us.
3972                  *
3973                  * We can't race against event freeing since the other
3974                  * side will require wqh->lock via remove_wait_queue(),
3975                  * which we hold.
3976                  */
3977                 spin_lock(&cgrp->event_list_lock);
3978                 if (!list_empty(&event->list)) {
3979                         list_del_init(&event->list);
3980                         /*
3981                          * We are in atomic context, but cgroup_event_remove()
3982                          * may sleep, so we have to call it in workqueue.
3983                          */
3984                         schedule_work(&event->remove);
3985                 }
3986                 spin_unlock(&cgrp->event_list_lock);
3987         }
3988 
3989         return 0;
3990 }
3991 
3992 static void cgroup_event_ptable_queue_proc(struct file *file,
3993                 wait_queue_head_t *wqh, poll_table *pt)
3994 {
3995         struct cgroup_event *event = container_of(pt,
3996                         struct cgroup_event, pt);
3997 
3998         event->wqh = wqh;
3999         add_wait_queue(wqh, &event->wait);
4000 }
4001 
4002 /*
4003  * Parse input and register new cgroup event handler.
4004  *
4005  * Input must be in format '<event_fd> <control_fd> <args>'.
4006  * Interpretation of args is defined by control file implementation.
4007  */
4008 static int cgroup_write_event_control(struct cgroup_subsys_state *dummy_css,
4009                                       struct cftype *cft, const char *buffer)
4010 {
4011         struct cgroup *cgrp = dummy_css->cgroup;
4012         struct cgroup_event *event;
4013         struct cgroup_subsys_state *cfile_css;
4014         unsigned int efd, cfd;
4015         struct fd efile;
4016         struct fd cfile;
4017         char *endp;
4018         int ret;
4019 
4020         efd = simple_strtoul(buffer, &endp, 10);
4021         if (*endp != ' ')
4022                 return -EINVAL;
4023         buffer = endp + 1;
4024 
4025         cfd = simple_strtoul(buffer, &endp, 10);
4026         if ((*endp != ' ') && (*endp != '\0'))
4027                 return -EINVAL;
4028         buffer = endp + 1;
4029 
4030         event = kzalloc(sizeof(*event), GFP_KERNEL);
4031         if (!event)
4032                 return -ENOMEM;
4033 
4034         INIT_LIST_HEAD(&event->list);
4035         init_poll_funcptr(&event->pt, cgroup_event_ptable_queue_proc);
4036         init_waitqueue_func_entry(&event->wait, cgroup_event_wake);
4037         INIT_WORK(&event->remove, cgroup_event_remove);
4038 
4039         efile = fdget(efd);
4040         if (!efile.file) {
4041                 ret = -EBADF;
4042                 goto out_kfree;
4043         }
4044 
4045         event->eventfd = eventfd_ctx_fileget(efile.file);
4046         if (IS_ERR(event->eventfd)) {
4047                 ret = PTR_ERR(event->eventfd);
4048                 goto out_put_efile;
4049         }
4050 
4051         cfile = fdget(cfd);
4052         if (!cfile.file) {
4053                 ret = -EBADF;
4054                 goto out_put_eventfd;
4055         }
4056 
4057         /* the process need read permission on control file */
4058         /* AV: shouldn't we check that it's been opened for read instead? */
4059         ret = inode_permission(file_inode(cfile.file), MAY_READ);
4060         if (ret < 0)
4061                 goto out_put_cfile;
4062 
4063         event->cft = __file_cft(cfile.file);
4064         if (IS_ERR(event->cft)) {
4065                 ret = PTR_ERR(event->cft);
4066                 goto out_put_cfile;
4067         }
4068 
4069         if (!event->cft->ss) {
4070                 ret = -EBADF;
4071                 goto out_put_cfile;
4072         }
4073 
4074         /*
4075          * Determine the css of @cfile, verify it belongs to the same
4076          * cgroup as cgroup.event_control, and associate @event with it.
4077          * Remaining events are automatically removed on cgroup destruction
4078          * but the removal is asynchronous, so take an extra ref.
4079          */
4080         rcu_read_lock();
4081 
4082         ret = -EINVAL;
4083         event->css = cgroup_css(cgrp, event->cft->ss);
4084         cfile_css = css_from_dir(cfile.file->f_dentry->d_parent, event->cft->ss);
4085         if (event->css && event->css == cfile_css && css_tryget(event->css))
4086                 ret = 0;
4087 
4088         rcu_read_unlock();
4089         if (ret)
4090                 goto out_put_cfile;
4091 
4092         if (!event->cft->register_event || !event->cft->unregister_event) {
4093                 ret = -EINVAL;
4094                 goto out_put_css;
4095         }
4096 
4097         ret = event->cft->register_event(event->css, event->cft,
4098                         event->eventfd, buffer);
4099         if (ret)
4100                 goto out_put_css;
4101 
4102         efile.file->f_op->poll(efile.file, &event->pt);
4103 
4104         spin_lock(&cgrp->event_list_lock);
4105         list_add(&event->list, &cgrp->event_list);
4106         spin_unlock(&cgrp->event_list_lock);
4107 
4108         fdput(cfile);
4109         fdput(efile);
4110 
4111         return 0;
4112 
4113 out_put_css:
4114         css_put(event->css);
4115 out_put_cfile:
4116         fdput(cfile);
4117 out_put_eventfd:
4118         eventfd_ctx_put(event->eventfd);
4119 out_put_efile:
4120         fdput(efile);
4121 out_kfree:
4122         kfree(event);
4123 
4124         return ret;
4125 }
4126 
4127 static u64 cgroup_clone_children_read(struct cgroup_subsys_state *css,
4128                                       struct cftype *cft)
4129 {
4130         return test_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags);
4131 }
4132 
4133 static int cgroup_clone_children_write(struct cgroup_subsys_state *css,
4134                                        struct cftype *cft, u64 val)
4135 {
4136         if (val)
4137                 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags);
4138         else
4139                 clear_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags);
4140         return 0;
4141 }
4142 
4143 static struct cftype cgroup_base_files[] = {
4144         {
4145                 .name = "cgroup.procs",
4146                 .open = cgroup_procs_open,
4147                 .write_u64 = cgroup_procs_write,
4148                 .release = cgroup_pidlist_release,
4149                 .mode = S_IRUGO | S_IWUSR,
4150         },
4151         {
4152                 .name = "cgroup.event_control",
4153                 .write_string = cgroup_write_event_control,
4154                 .mode = S_IWUGO,
4155         },
4156         {
4157                 .name = "cgroup.clone_children",
4158                 .flags = CFTYPE_INSANE,
4159                 .read_u64 = cgroup_clone_children_read,
4160                 .write_u64 = cgroup_clone_children_write,
4161         },
4162         {
4163                 .name = "cgroup.sane_behavior",
4164                 .flags = CFTYPE_ONLY_ON_ROOT,
4165                 .read_seq_string = cgroup_sane_behavior_show,
4166         },
4167 
4168         /*
4169          * Historical crazy stuff.  These don't have "cgroup."  prefix and
4170          * don't exist if sane_behavior.  If you're depending on these, be
4171          * prepared to be burned.
4172          */
4173         {
4174                 .name = "tasks",
4175                 .flags = CFTYPE_INSANE,         /* use "procs" instead */
4176                 .open = cgroup_tasks_open,
4177                 .write_u64 = cgroup_tasks_write,
4178                 .release = cgroup_pidlist_release,
4179                 .mode = S_IRUGO | S_IWUSR,
4180         },
4181         {
4182                 .name = "notify_on_release",
4183                 .flags = CFTYPE_INSANE,
4184                 .read_u64 = cgroup_read_notify_on_release,
4185                 .write_u64 = cgroup_write_notify_on_release,
4186         },
4187         {
4188                 .name = "release_agent",
4189                 .flags = CFTYPE_INSANE | CFTYPE_ONLY_ON_ROOT,
4190                 .read_seq_string = cgroup_release_agent_show,
4191                 .write_string = cgroup_release_agent_write,
4192                 .max_write_len = PATH_MAX,
4193         },
4194         { }     /* terminate */
4195 };
4196 
4197 /**
4198  * cgroup_populate_dir - create subsys files in a cgroup directory
4199  * @cgrp: target cgroup
4200  * @subsys_mask: mask of the subsystem ids whose files should be added
4201  *
4202  * On failure, no file is added.
4203  */
4204 static int cgroup_populate_dir(struct cgroup *cgrp, unsigned long subsys_mask)
4205 {
4206         struct cgroup_subsys *ss;
4207         int i, ret = 0;
4208 
4209         /* process cftsets of each subsystem */
4210         for_each_subsys(ss, i) {
4211                 struct cftype_set *set;
4212 
4213                 if (!test_bit(i, &subsys_mask))
4214                         continue;
4215 
4216                 list_for_each_entry(set, &ss->cftsets, node) {
4217                         ret = cgroup_addrm_files(cgrp, set->cfts, true);
4218                         if (ret < 0)
4219                                 goto err;
4220                 }
4221         }
4222         return 0;
4223 err:
4224         cgroup_clear_dir(cgrp, subsys_mask);
4225         return ret;
4226 }
4227 
4228 /*
4229  * css destruction is four-stage process.
4230  *
4231  * 1. Destruction starts.  Killing of the percpu_ref is initiated.
4232  *    Implemented in kill_css().
4233  *
4234  * 2. When the percpu_ref is confirmed to be visible as killed on all CPUs
4235  *    and thus css_tryget() is guaranteed to fail, the css can be offlined
4236  *    by invoking offline_css().  After offlining, the base ref is put.
4237  *    Implemented in css_killed_work_fn().
4238  *
4239  * 3. When the percpu_ref reaches zero, the only possible remaining
4240  *    accessors are inside RCU read sections.  css_release() schedules the
4241  *    RCU callback.
4242  *
4243  * 4. After the grace period, the css can be freed.  Implemented in
4244  *    css_free_work_fn().
4245  *
4246  * It is actually hairier because both step 2 and 4 require process context
4247  * and thus involve punting to css->destroy_work adding two additional
4248  * steps to the already complex sequence.
4249  */
4250 static void css_free_work_fn(struct work_struct *work)
4251 {
4252         struct cgroup_subsys_state *css =
4253                 container_of(work, struct cgroup_subsys_state, destroy_work);
4254         struct cgroup *cgrp = css->cgroup;
4255 
4256         if (css->parent)
4257                 css_put(css->parent);
4258 
4259         css->ss->css_free(css);
4260         cgroup_dput(cgrp);
4261 }
4262 
4263 static void css_free_rcu_fn(struct rcu_head *rcu_head)
4264 {
4265         struct cgroup_subsys_state *css =
4266                 container_of(rcu_head, struct cgroup_subsys_state, rcu_head);
4267 
4268         /*
4269          * css holds an extra ref to @cgrp->dentry which is put on the last
4270          * css_put().  dput() requires process context which we don't have.
4271          */
4272         INIT_WORK(&css->destroy_work, css_free_work_fn);
4273         queue_work(cgroup_destroy_wq, &css->destroy_work);
4274 }
4275 
4276 static void css_release(struct percpu_ref *ref)
4277 {
4278         struct cgroup_subsys_state *css =
4279                 container_of(ref, struct cgroup_subsys_state, refcnt);
4280 
4281         rcu_assign_pointer(css->cgroup->subsys[css->ss->subsys_id], NULL);
4282         call_rcu(&css->rcu_head, css_free_rcu_fn);
4283 }
4284 
4285 static void init_css(struct cgroup_subsys_state *css, struct cgroup_subsys *ss,
4286                      struct cgroup *cgrp)
4287 {
4288         css->cgroup = cgrp;
4289         css->ss = ss;
4290         css->flags = 0;
4291 
4292         if (cgrp->parent)
4293                 css->parent = cgroup_css(cgrp->parent, ss);
4294         else
4295                 css->flags |= CSS_ROOT;
4296 
4297         BUG_ON(cgroup_css(cgrp, ss));
4298 }
4299 
4300 /* invoke ->css_online() on a new CSS and mark it online if successful */
4301 static int online_css(struct cgroup_subsys_state *css)
4302 {
4303         struct cgroup_subsys *ss = css->ss;
4304         int ret = 0;
4305 
4306         lockdep_assert_held(&cgroup_mutex);
4307 
4308         if (ss->css_online)
4309                 ret = ss->css_online(css);
4310         if (!ret) {
4311                 css->flags |= CSS_ONLINE;
4312                 css->cgroup->nr_css++;
4313                 rcu_assign_pointer(css->cgroup->subsys[ss->subsys_id], css);
4314         }
4315         return ret;
4316 }
4317 
4318 /* if the CSS is online, invoke ->css_offline() on it and mark it offline */
4319 static void offline_css(struct cgroup_subsys_state *css)
4320 {
4321         struct cgroup_subsys *ss = css->ss;
4322 
4323         lockdep_assert_held(&cgroup_mutex);
4324 
4325         if (!(css->flags & CSS_ONLINE))
4326                 return;
4327 
4328         if (ss->css_offline)
4329                 ss->css_offline(css);
4330 
4331         css->flags &= ~CSS_ONLINE;
4332         css->cgroup->nr_css--;
4333         RCU_INIT_POINTER(css->cgroup->subsys[ss->subsys_id], css);
4334 }
4335 
4336 /*
4337  * cgroup_create - create a cgroup
4338  * @parent: cgroup that will be parent of the new cgroup
4339  * @dentry: dentry of the new cgroup
4340  * @mode: mode to set on new inode
4341  *
4342  * Must be called with the mutex on the parent inode held
4343  */
4344 static long cgroup_create(struct cgroup *parent, struct dentry *dentry,
4345                              umode_t mode)
4346 {
4347         struct cgroup_subsys_state *css_ar[CGROUP_SUBSYS_COUNT] = { };
4348         struct cgroup *cgrp;
4349         struct cgroup_name *name;
4350         struct cgroupfs_root *root = parent->root;
4351         int err = 0;
4352         struct cgroup_subsys *ss;
4353         struct super_block *sb = root->sb;
4354 
4355         /* allocate the cgroup and its ID, 0 is reserved for the root */
4356         cgrp = kzalloc(sizeof(*cgrp), GFP_KERNEL);
4357         if (!cgrp)
4358                 return -ENOMEM;
4359 
4360         name = cgroup_alloc_name(dentry);
4361         if (!name)
4362                 goto err_free_cgrp;
4363         rcu_assign_pointer(cgrp->name, name);
4364 
4365         /*
4366          * Temporarily set the pointer to NULL, so idr_find() won't return
4367          * a half-baked cgroup.
4368          */
4369         cgrp->id = idr_alloc(&root->cgroup_idr, NULL, 1, 0, GFP_KERNEL);
4370         if (cgrp->id < 0)
4371                 goto err_free_name;
4372 
4373         /*
4374          * Only live parents can have children.  Note that the liveliness
4375          * check isn't strictly necessary because cgroup_mkdir() and
4376          * cgroup_rmdir() are fully synchronized by i_mutex; however, do it
4377          * anyway so that locking is contained inside cgroup proper and we
4378          * don't get nasty surprises if we ever grow another caller.
4379          */
4380         if (!cgroup_lock_live_group(parent)) {
4381                 err = -ENODEV;
4382                 goto err_free_id;
4383         }
4384 
4385         /* Grab a reference on the superblock so the hierarchy doesn't
4386          * get deleted on unmount if there are child cgroups.  This
4387          * can be done outside cgroup_mutex, since the sb can't
4388          * disappear while someone has an open control file on the
4389          * fs */
4390         atomic_inc(&sb->s_active);
4391 
4392         init_cgroup_housekeeping(cgrp);
4393 
4394         dentry->d_fsdata = cgrp;
4395         cgrp->dentry = dentry;
4396 
4397         cgrp->parent = parent;
4398         cgrp->dummy_css.parent = &parent->dummy_css;
4399         cgrp->root = parent->root;
4400 
4401         if (notify_on_release(parent))
4402                 set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
4403 
4404         if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &parent->flags))
4405                 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags);
4406 
4407         for_each_root_subsys(root, ss) {
4408                 struct cgroup_subsys_state *css;
4409 
4410                 css = ss->css_alloc(cgroup_css(parent, ss));
4411                 if (IS_ERR(css)) {
4412                         err = PTR_ERR(css);
4413                         goto err_free_all;
4414                 }
4415                 css_ar[ss->subsys_id] = css;
4416 
4417                 err = percpu_ref_init(&css->refcnt, css_release);
4418                 if (err)
4419                         goto err_free_all;
4420 
4421                 init_css(css, ss, cgrp);
4422         }
4423 
4424         /*
4425          * Create directory.  cgroup_create_file() returns with the new
4426          * directory locked on success so that it can be populated without
4427          * dropping cgroup_mutex.
4428          */
4429         err = cgroup_create_file(dentry, S_IFDIR | mode, sb);
4430         if (err < 0)
4431                 goto err_free_all;
4432         lockdep_assert_held(&dentry->d_inode->i_mutex);
4433 
4434         cgrp->serial_nr = cgroup_serial_nr_next++;
4435 
4436         /* allocation complete, commit to creation */
4437         list_add_tail_rcu(&cgrp->sibling, &cgrp->parent->children);
4438         root->number_of_cgroups++;
4439 
4440         /* hold a ref to the parent's dentry */
4441         dget(parent->dentry);
4442 
4443         /* creation succeeded, notify subsystems */
4444         for_each_root_subsys(root, ss) {
4445                 struct cgroup_subsys_state *css = css_ar[ss->subsys_id];
4446 
4447                 err = online_css(css);
4448                 if (err)
4449                         goto err_destroy;
4450 
4451                 /* each css holds a ref to the cgroup's dentry and parent css */
4452                 dget(dentry);
4453                 css_get(css->parent);
4454 
4455                 /* mark it consumed for error path */
4456                 css_ar[ss->subsys_id] = NULL;
4457 
4458                 if (ss->broken_hierarchy && !ss->warned_broken_hierarchy &&
4459                     parent->parent) {
4460                         pr_warning("cgroup: %s (%d) created nested cgroup for controller \"%s\" which has incomplete hierarchy support. Nested cgroups may change behavior in the future.\n",
4461                                    current->comm, current->pid, ss->name);
4462                         if (!strcmp(ss->name, "memory"))
4463                                 pr_warning("cgroup: \"memory\" requires setting use_hierarchy to 1 on the root.\n");
4464                         ss->warned_broken_hierarchy = true;
4465                 }
4466         }
4467 
4468         idr_replace(&root->cgroup_idr, cgrp, cgrp->id);
4469 
4470         err = cgroup_addrm_files(cgrp, cgroup_base_files, true);
4471         if (err)
4472                 goto err_destroy;
4473 
4474         err = cgroup_populate_dir(cgrp, root->subsys_mask);
4475         if (err)
4476                 goto err_destroy;
4477 
4478         mutex_unlock(&cgroup_mutex);
4479         mutex_unlock(&cgrp->dentry->d_inode->i_mutex);
4480 
4481         return 0;
4482 
4483 err_free_all:
4484         for_each_root_subsys(root, ss) {
4485                 struct cgroup_subsys_state *css = css_ar[ss->subsys_id];
4486 
4487                 if (css) {
4488                         percpu_ref_cancel_init(&css->refcnt);
4489                         ss->css_free(css);
4490                 }
4491         }
4492         mutex_unlock(&cgroup_mutex);
4493         /* Release the reference count that we took on the superblock */
4494         deactivate_super(sb);
4495 err_free_id:
4496         idr_remove(&root->cgroup_idr, cgrp->id);
4497 err_free_name:
4498         kfree(rcu_dereference_raw(cgrp->name));
4499 err_free_cgrp:
4500         kfree(cgrp);
4501         return err;
4502 
4503 err_destroy:
4504         for_each_root_subsys(root, ss) {
4505                 struct cgroup_subsys_state *css = css_ar[ss->subsys_id];
4506 
4507                 if (css) {
4508                         percpu_ref_cancel_init(&css->refcnt);
4509                         ss->css_free(css);
4510                 }
4511         }
4512         cgroup_destroy_locked(cgrp);
4513         mutex_unlock(&cgroup_mutex);
4514         mutex_unlock(&dentry->d_inode->i_mutex);
4515         return err;
4516 }
4517 
4518 static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
4519 {
4520         struct cgroup *c_parent = dentry->d_parent->d_fsdata;
4521 
4522         /* the vfs holds inode->i_mutex already */
4523         return cgroup_create(c_parent, dentry, mode | S_IFDIR);
4524 }
4525 
4526 /*
4527  * This is called when the refcnt of a css is confirmed to be killed.
4528  * css_tryget() is now guaranteed to fail.
4529  */
4530 static void css_killed_work_fn(struct work_struct *work)
4531 {
4532         struct cgroup_subsys_state *css =
4533                 container_of(work, struct cgroup_subsys_state, destroy_work);
4534         struct cgroup *cgrp = css->cgroup;
4535 
4536         mutex_lock(&cgroup_mutex);
4537 
4538         /*
4539          * css_tryget() is guaranteed to fail now.  Tell subsystems to
4540          * initate destruction.
4541          */
4542         offline_css(css);
4543 
4544         /*
4545          * If @cgrp is marked dead, it's waiting for refs of all css's to
4546          * be disabled before proceeding to the second phase of cgroup
4547          * destruction.  If we are the last one, kick it off.
4548          */
4549         if (!cgrp->nr_css && cgroup_is_dead(cgrp))
4550                 cgroup_destroy_css_killed(cgrp);
4551 
4552         mutex_unlock(&cgroup_mutex);
4553 
4554         /*
4555          * Put the css refs from kill_css().  Each css holds an extra
4556          * reference to the cgroup's dentry and cgroup removal proceeds
4557          * regardless of css refs.  On the last put of each css, whenever
4558          * that may be, the extra dentry ref is put so that dentry
4559          * destruction happens only after all css's are released.
4560          */
4561         css_put(css);
4562 }
4563 
4564 /* css kill confirmation processing requires process context, bounce */
4565 static void css_killed_ref_fn(struct percpu_ref *ref)
4566 {
4567         struct cgroup_subsys_state *css =
4568                 container_of(ref, struct cgroup_subsys_state, refcnt);
4569 
4570         INIT_WORK(&css->destroy_work, css_killed_work_fn);
4571         queue_work(cgroup_destroy_wq, &css->destroy_work);
4572 }
4573 
4574 /**
4575  * kill_css - destroy a css
4576  * @css: css to destroy
4577  *
4578  * This function initiates destruction of @css by removing cgroup interface
4579  * files and putting its base reference.  ->css_offline() will be invoked
4580  * asynchronously once css_tryget() is guaranteed to fail and when the
4581  * reference count reaches zero, @css will be released.
4582  */
4583 static void kill_css(struct cgroup_subsys_state *css)
4584 {
4585         cgroup_clear_dir(css->cgroup, 1 << css->ss->subsys_id);
4586 
4587         /*
4588          * Killing would put the base ref, but we need to keep it alive
4589          * until after ->css_offline().
4590          */
4591         css_get(css);
4592 
4593         /*
4594          * cgroup core guarantees that, by the time ->css_offline() is
4595          * invoked, no new css reference will be given out via
4596          * css_tryget().  We can't simply call percpu_ref_kill() and
4597          * proceed to offlining css's because percpu_ref_kill() doesn't
4598          * guarantee that the ref is seen as killed on all CPUs on return.
4599          *
4600          * Use percpu_ref_kill_and_confirm() to get notifications as each
4601          * css is confirmed to be seen as killed on all CPUs.
4602          */
4603         percpu_ref_kill_and_confirm(&css->refcnt, css_killed_ref_fn);
4604 }
4605 
4606 /**
4607  * cgroup_destroy_locked - the first stage of cgroup destruction
4608  * @cgrp: cgroup to be destroyed
4609  *
4610  * css's make use of percpu refcnts whose killing latency shouldn't be
4611  * exposed to userland and are RCU protected.  Also, cgroup core needs to
4612  * guarantee that css_tryget() won't succeed by the time ->css_offline() is
4613  * invoked.  To satisfy all the requirements, destruction is implemented in
4614  * the following two steps.
4615  *
4616  * s1. Verify @cgrp can be destroyed and mark it dying.  Remove all
4617  *     userland visible parts and start killing the percpu refcnts of
4618  *     css's.  Set up so that the next stage will be kicked off once all
4619  *     the percpu refcnts are confirmed to be killed.
4620  *
4621  * s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the
4622  *     rest of destruction.  Once all cgroup references are gone, the
4623  *     cgroup is RCU-freed.
4624  *
4625  * This function implements s1.  After this step, @cgrp is gone as far as
4626  * the userland is concerned and a new cgroup with the same name may be
4627  * created.  As cgroup doesn't care about the names internally, this
4628  * doesn't cause any problem.
4629  */
4630 static int cgroup_destroy_locked(struct cgroup *cgrp)
4631         __releases(&cgroup_mutex) __acquires(&cgroup_mutex)
4632 {
4633         struct dentry *d = cgrp->dentry;
4634         struct cgroup_event *event, *tmp;
4635         struct cgroup_subsys *ss;
4636         struct cgroup *child;
4637         bool empty;
4638 
4639         lockdep_assert_held(&d->d_inode->i_mutex);
4640         lockdep_assert_held(&cgroup_mutex);
4641 
4642         /*
4643          * css_set_lock synchronizes access to ->cset_links and prevents
4644          * @cgrp from being removed while __put_css_set() is in progress.
4645          */
4646         read_lock(&css_set_lock);
4647         empty = list_empty(&cgrp->cset_links);
4648         read_unlock(&css_set_lock);
4649         if (!empty)
4650                 return -EBUSY;
4651 
4652         /*
4653          * Make sure there's no live children.  We can't test ->children
4654          * emptiness as dead children linger on it while being destroyed;
4655          * otherwise, "rmdir parent/child parent" may fail with -EBUSY.
4656          */
4657         empty = true;
4658         rcu_read_lock();
4659         list_for_each_entry_rcu(child, &cgrp->children, sibling) {
4660                 empty = cgroup_is_dead(child);
4661                 if (!empty)
4662                         break;
4663         }
4664         rcu_read_unlock();
4665         if (!empty)
4666                 return -EBUSY;
4667 
4668         /*
4669          * Initiate massacre of all css's.  cgroup_destroy_css_killed()
4670          * will be invoked to perform the rest of destruction once the
4671          * percpu refs of all css's are confirmed to be killed.
4672          */
4673         for_each_root_subsys(cgrp->root, ss) {
4674                 struct cgroup_subsys_state *css = cgroup_css(cgrp, ss);
4675 
4676                 if (css)
4677                         kill_css(css);
4678         }
4679 
4680         /*
4681          * Mark @cgrp dead.  This prevents further task migration and child
4682          * creation by disabling cgroup_lock_live_group().  Note that
4683          * CGRP_DEAD assertion is depended upon by css_next_child() to
4684          * resume iteration after dropping RCU read lock.  See
4685          * css_next_child() for details.
4686          */
4687         set_bit(CGRP_DEAD, &cgrp->flags);
4688 
4689         /* CGRP_DEAD is set, remove from ->release_list for the last time */
4690         raw_spin_lock(&release_list_lock);
4691         if (!list_empty(&cgrp->release_list))
4692                 list_del_init(&cgrp->release_list);
4693         raw_spin_unlock(&release_list_lock);
4694 
4695         /*
4696          * If @cgrp has css's attached, the second stage of cgroup
4697          * destruction is kicked off from css_killed_work_fn() after the
4698          * refs of all attached css's are killed.  If @cgrp doesn't have
4699          * any css, we kick it off here.
4700          */
4701         if (!cgrp->nr_css)
4702                 cgroup_destroy_css_killed(cgrp);
4703 
4704         /*
4705          * Clear the base files and remove @cgrp directory.  The removal
4706          * puts the base ref but we aren't quite done with @cgrp yet, so
4707          * hold onto it.
4708          */
4709         cgroup_addrm_files(cgrp, cgroup_base_files, false);
4710         dget(d);
4711         cgroup_d_remove_dir(d);
4712 
4713         /*
4714          * Unregister events and notify userspace.
4715          * Notify userspace about cgroup removing only after rmdir of cgroup
4716          * directory to avoid race between userspace and kernelspace.
4717          */
4718         spin_lock(&cgrp->event_list_lock);
4719         list_for_each_entry_safe(event, tmp, &cgrp->event_list, list) {
4720                 list_del_init(&event->list);
4721                 schedule_work(&event->remove);
4722         }
4723         spin_unlock(&cgrp->event_list_lock);
4724 
4725         return 0;
4726 };
4727 
4728 /**
4729  * cgroup_destroy_css_killed - the second step of cgroup destruction
4730  * @work: cgroup->destroy_free_work
4731  *
4732  * This function is invoked from a work item for a cgroup which is being
4733  * destroyed after all css's are offlined and performs the rest of
4734  * destruction.  This is the second step of destruction described in the
4735  * comment above cgroup_destroy_locked().
4736  */
4737 static void cgroup_destroy_css_killed(struct cgroup *cgrp)
4738 {
4739         struct cgroup *parent = cgrp->parent;
4740         struct dentry *d = cgrp->dentry;
4741 
4742         lockdep_assert_held(&cgroup_mutex);
4743 
4744         /* delete this cgroup from parent->children */
4745         list_del_rcu(&cgrp->sibling);
4746 
4747         dput(d);
4748 
4749         set_bit(CGRP_RELEASABLE, &parent->flags);
4750         check_for_release(parent);
4751 }
4752 
4753 static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry)
4754 {
4755         int ret;
4756 
4757         mutex_lock(&cgroup_mutex);
4758         ret = cgroup_destroy_locked(dentry->d_fsdata);
4759         mutex_unlock(&cgroup_mutex);
4760 
4761         return ret;
4762 }
4763 
4764 static void __init_or_module cgroup_init_cftsets(struct cgroup_subsys *ss)
4765 {
4766         INIT_LIST_HEAD(&ss->cftsets);
4767 
4768         /*
4769          * base_cftset is embedded in subsys itself, no need to worry about
4770          * deregistration.
4771          */
4772         if (ss->base_cftypes) {
4773                 struct cftype *cft;
4774 
4775                 for (cft = ss->base_cftypes; cft->name[0] != '\0'; cft++)
4776                         cft->ss = ss;
4777 
4778                 ss->base_cftset.cfts = ss->base_cftypes;
4779                 list_add_tail(&ss->base_cftset.node, &ss->cftsets);
4780         }
4781 }
4782 
4783 static void __init cgroup_init_subsys(struct cgroup_subsys *ss)
4784 {
4785         struct cgroup_subsys_state *css;
4786 
4787         printk(KERN_INFO "Initializing cgroup subsys %s\n", ss->name);
4788 
4789         mutex_lock(&cgroup_mutex);
4790 
4791         /* init base cftset */
4792         cgroup_init_cftsets(ss);
4793 
4794         /* Create the top cgroup state for this subsystem */
4795         list_add(&ss->sibling, &cgroup_dummy_root.subsys_list);
4796         ss->root = &cgroup_dummy_root;
4797         css = ss->css_alloc(cgroup_css(cgroup_dummy_top, ss));
4798         /* We don't handle early failures gracefully */
4799         BUG_ON(IS_ERR(css));
4800         init_css(css, ss, cgroup_dummy_top);
4801 
4802         /* Update the init_css_set to contain a subsys
4803          * pointer to this state - since the subsystem is
4804          * newly registered, all tasks and hence the
4805          * init_css_set is in the subsystem's top cgroup. */
4806         init_css_set.subsys[ss->subsys_id] = css;
4807 
4808         need_forkexit_callback |= ss->fork || ss->exit;
4809 
4810         /* At system boot, before all subsystems have been
4811          * registered, no tasks have been forked, so we don't
4812          * need to invoke fork callbacks here. */
4813         BUG_ON(!list_empty(&init_task.tasks));
4814 
4815         BUG_ON(online_css(css));
4816 
4817         mutex_unlock(&cgroup_mutex);
4818 
4819         /* this function shouldn't be used with modular subsystems, since they
4820          * need to register a subsys_id, among other things */
4821         BUG_ON(ss->module);
4822 }
4823 
4824 /**
4825  * cgroup_load_subsys: load and register a modular subsystem at runtime
4826  * @ss: the subsystem to load
4827  *
4828  * This function should be called in a modular subsystem's initcall. If the
4829  * subsystem is built as a module, it will be assigned a new subsys_id and set
4830  * up for use. If the subsystem is built-in anyway, work is delegated to the
4831  * simpler cgroup_init_subsys.
4832  */
4833 int __init_or_module cgroup_load_subsys(struct cgroup_subsys *ss)
4834 {
4835         struct cgroup_subsys_state *css;
4836         int i, ret;
4837         struct hlist_node *tmp;
4838         struct css_set *cset;
4839         unsigned long key;
4840 
4841         /* check name and function validity */
4842         if (ss->name == NULL || strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN ||
4843             ss->css_alloc == NULL || ss->css_free == NULL)
4844                 return -EINVAL;
4845 
4846         /*
4847          * we don't support callbacks in modular subsystems. this check is
4848          * before the ss->module check for consistency; a subsystem that could
4849          * be a module should still have no callbacks even if the user isn't
4850          * compiling it as one.
4851          */
4852         if (ss->fork || ss->exit)
4853                 return -EINVAL;
4854 
4855         /*
4856          * an optionally modular subsystem is built-in: we want to do nothing,
4857          * since cgroup_init_subsys will have already taken care of it.
4858          */
4859         if (ss->module == NULL) {
4860                 /* a sanity check */
4861                 BUG_ON(cgroup_subsys[ss->subsys_id] != ss);
4862                 return 0;
4863         }
4864 
4865         /* init base cftset */
4866         cgroup_init_cftsets(ss);
4867 
4868         mutex_lock(&cgroup_mutex);
4869         cgroup_subsys[ss->subsys_id] = ss;
4870 
4871         /*
4872          * no ss->css_alloc seems to need anything important in the ss
4873          * struct, so this can happen first (i.e. before the dummy root
4874          * attachment).
4875          */
4876         css = ss->css_alloc(cgroup_css(cgroup_dummy_top, ss));
4877         if (IS_ERR(css)) {
4878                 /* failure case - need to deassign the cgroup_subsys[] slot. */
4879                 cgroup_subsys[ss->subsys_id] = NULL;
4880                 mutex_unlock(&cgroup_mutex);
4881                 return PTR_ERR(css);
4882         }
4883 
4884         list_add(&ss->sibling, &cgroup_dummy_root.subsys_list);
4885         ss->root = &cgroup_dummy_root;
4886 
4887         /* our new subsystem will be attached to the dummy hierarchy. */
4888         init_css(css, ss, cgroup_dummy_top);
4889 
4890         /*
4891          * Now we need to entangle the css into the existing css_sets. unlike
4892          * in cgroup_init_subsys, there are now multiple css_sets, so each one
4893          * will need a new pointer to it; done by iterating the css_set_table.
4894          * furthermore, modifying the existing css_sets will corrupt the hash
4895          * table state, so each changed css_set will need its hash recomputed.
4896          * this is all done under the css_set_lock.
4897          */
4898         write_lock(&css_set_lock);
4899         hash_for_each_safe(css_set_table, i, tmp, cset, hlist) {
4900                 /* skip entries that we already rehashed */
4901                 if (cset->subsys[ss->subsys_id])
4902                         continue;
4903                 /* remove existing entry */
4904                 hash_del(&cset->hlist);
4905                 /* set new value */
4906                 cset->subsys[ss->subsys_id] = css;
4907                 /* recompute hash and restore entry */
4908                 key = css_set_hash(cset->subsys);
4909                 hash_add(css_set_table, &cset->hlist, key);
4910         }
4911         write_unlock(&css_set_lock);
4912 
4913         ret = online_css(css);
4914         if (ret)
4915                 goto err_unload;
4916 
4917         /* success! */
4918         mutex_unlock(&cgroup_mutex);
4919         return 0;
4920 
4921 err_unload:
4922         mutex_unlock(&cgroup_mutex);
4923         /* @ss can't be mounted here as try_module_get() would fail */
4924         cgroup_unload_subsys(ss);
4925         return ret;
4926 }
4927 EXPORT_SYMBOL_GPL(cgroup_load_subsys);
4928 
4929 /**
4930  * cgroup_unload_subsys: unload a modular subsystem
4931  * @ss: the subsystem to unload
4932  *
4933  * This function should be called in a modular subsystem's exitcall. When this
4934  * function is invoked, the refcount on the subsystem's module will be 0, so
4935  * the subsystem will not be attached to any hierarchy.
4936  */
4937 void cgroup_unload_subsys(struct cgroup_subsys *ss)
4938 {
4939         struct cgrp_cset_link *link;
4940 
4941         BUG_ON(ss->module == NULL);
4942 
4943         /*
4944          * we shouldn't be called if the subsystem is in use, and the use of
4945          * try_module_get() in rebind_subsystems() should ensure that it
4946          * doesn't start being used while we're killing it off.
4947          */
4948         BUG_ON(ss->root != &cgroup_dummy_root);
4949 
4950         mutex_lock(&cgroup_mutex);
4951 
4952         offline_css(cgroup_css(cgroup_dummy_top, ss));
4953 
4954         /* deassign the subsys_id */
4955         cgroup_subsys[ss->subsys_id] = NULL;
4956 
4957         /* remove subsystem from the dummy root's list of subsystems */
4958         list_del_init(&ss->sibling);
4959 
4960         /*
4961          * disentangle the css from all css_sets attached to the dummy
4962          * top. as in loading, we need to pay our respects to the hashtable
4963          * gods.
4964          */
4965         write_lock(&css_set_lock);
4966         list_for_each_entry(link, &cgroup_dummy_top->cset_links, cset_link) {
4967                 struct css_set *cset = link->cset;
4968                 unsigned long key;
4969 
4970                 hash_del(&cset->hlist);
4971                 cset->subsys[ss->subsys_id] = NULL;
4972                 key = css_set_hash(cset->subsys);
4973                 hash_add(css_set_table, &cset->hlist, key);
4974         }
4975         write_unlock(&css_set_lock);
4976 
4977         /*
4978          * remove subsystem's css from the cgroup_dummy_top and free it -
4979          * need to free before marking as null because ss->css_free needs
4980          * the cgrp->subsys pointer to find their state.
4981          */
4982         ss->css_free(cgroup_css(cgroup_dummy_top, ss));
4983         RCU_INIT_POINTER(cgroup_dummy_top->subsys[ss->subsys_id], NULL);
4984 
4985         mutex_unlock(&cgroup_mutex);
4986 }
4987 EXPORT_SYMBOL_GPL(cgroup_unload_subsys);
4988 
4989 /**
4990  * cgroup_init_early - cgroup initialization at system boot
4991  *
4992  * Initialize cgroups at system boot, and initialize any
4993  * subsystems that request early init.
4994  */
4995 int __init cgroup_init_early(void)
4996 {
4997         struct cgroup_subsys *ss;
4998         int i;
4999 
5000         atomic_set(&init_css_set.refcount, 1);
5001         INIT_LIST_HEAD(&init_css_set.cgrp_links);
5002         INIT_LIST_HEAD(&init_css_set.tasks);
5003         INIT_HLIST_NODE(&init_css_set.hlist);
5004         css_set_count = 1;
5005         init_cgroup_root(&cgroup_dummy_root);
5006         cgroup_root_count = 1;
5007         RCU_INIT_POINTER(init_task.cgroups, &init_css_set);
5008 
5009         init_cgrp_cset_link.cset = &init_css_set;
5010         init_cgrp_cset_link.cgrp = cgroup_dummy_top;
5011         list_add(&init_cgrp_cset_link.cset_link, &cgroup_dummy_top->cset_links);
5012         list_add(&init_cgrp_cset_link.cgrp_link, &init_css_set.cgrp_links);
5013 
5014         /* at bootup time, we don't worry about modular subsystems */
5015         for_each_builtin_subsys(ss, i) {
5016                 BUG_ON(!ss->name);
5017                 BUG_ON(strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN);
5018                 BUG_ON(!ss->css_alloc);
5019                 BUG_ON(!ss->css_free);
5020                 if (ss->subsys_id != i) {
5021                         printk(KERN_ERR "cgroup: Subsys %s id == %d\n",
5022                                ss->name, ss->subsys_id);
5023                         BUG();
5024                 }
5025 
5026                 if (ss->early_init)
5027                         cgroup_init_subsys(ss);
5028         }
5029         return 0;
5030 }
5031 
5032 /**
5033  * cgroup_init - cgroup initialization
5034  *
5035  * Register cgroup filesystem and /proc file, and initialize
5036  * any subsystems that didn't request early init.
5037  */
5038 int __init cgroup_init(void)
5039 {
5040         struct cgroup_subsys *ss;
5041         unsigned long key;
5042         int i, err;
5043 
5044         err = bdi_init(&cgroup_backing_dev_info);
5045         if (err)
5046                 return err;
5047 
5048         for_each_builtin_subsys(ss, i) {
5049                 if (!ss->early_init)
5050                         cgroup_init_subsys(ss);
5051         }
5052 
5053         /* allocate id for the dummy hierarchy */
5054         mutex_lock(&cgroup_mutex);
5055         mutex_lock(&cgroup_root_mutex);
5056 
5057         /* Add init_css_set to the hash table */
5058         key = css_set_hash(init_css_set.subsys);
5059         hash_add(css_set_table, &init_css_set.hlist, key);
5060 
5061         BUG_ON(cgroup_init_root_id(&cgroup_dummy_root, 0, 1));
5062 
5063         err = idr_alloc(&cgroup_dummy_root.cgroup_idr, cgroup_dummy_top,
5064                         0, 1, GFP_KERNEL);
5065         BUG_ON(err < 0);
5066 
5067         mutex_unlock(&cgroup_root_mutex);
5068         mutex_unlock(&cgroup_mutex);
5069 
5070         cgroup_kobj = kobject_create_and_add("cgroup", fs_kobj);
5071         if (!cgroup_kobj) {
5072                 err = -ENOMEM;
5073                 goto out;
5074         }
5075 
5076         err = register_filesystem(&cgroup_fs_type);
5077         if (err < 0) {
5078                 kobject_put(cgroup_kobj);
5079                 goto out;
5080         }
5081 
5082         proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations);
5083 
5084 out:
5085         if (err)
5086                 bdi_destroy(&cgroup_backing_dev_info);
5087 
5088         return err;
5089 }
5090 
5091 static int __init cgroup_wq_init(void)
5092 {
5093         /*
5094          * There isn't much point in executing destruction path in
5095          * parallel.  Good chunk is serialized with cgroup_mutex anyway.
5096          * Use 1 for @max_active.
5097          *
5098          * We would prefer to do this in cgroup_init() above, but that
5099          * is called before init_workqueues(): so leave this until after.
5100          */
5101         cgroup_destroy_wq = alloc_workqueue("cgroup_destroy", 0, 1);
5102         BUG_ON(!cgroup_destroy_wq);
5103         return 0;
5104 }
5105 core_initcall(cgroup_wq_init);
5106 
5107 /*
5108  * proc_cgroup_show()
5109  *  - Print task's cgroup paths into seq_file, one line for each hierarchy
5110  *  - Used for /proc/<pid>/cgroup.
5111  *  - No need to task_lock(tsk) on this tsk->cgroup reference, as it
5112  *    doesn't really matter if tsk->cgroup changes after we read it,
5113  *    and we take cgroup_mutex, keeping cgroup_attach_task() from changing it
5114  *    anyway.  No need to check that tsk->cgroup != NULL, thanks to
5115  *    the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks
5116  *    cgroup to top_cgroup.
5117  */
5118 
5119 /* TODO: Use a proper seq_file iterator */
5120 int proc_cgroup_show(struct seq_file *m, void *v)
5121 {
5122         struct pid *pid;
5123         struct task_struct *tsk;
5124         char *buf;
5125         int retval;
5126         struct cgroupfs_root *root;
5127 
5128         retval = -ENOMEM;
5129         buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
5130         if (!buf)
5131                 goto out;
5132 
5133         retval = -ESRCH;
5134         pid = m->private;
5135         tsk = get_pid_task(pid, PIDTYPE_PID);
5136         if (!tsk)
5137                 goto out_free;
5138 
5139         retval = 0;
5140 
5141         mutex_lock(&cgroup_mutex);
5142 
5143         for_each_active_root(root) {
5144                 struct cgroup_subsys *ss;
5145                 struct cgroup *cgrp;
5146                 int count = 0;
5147 
5148                 seq_printf(m, "%d:", root->hierarchy_id);
5149                 for_each_root_subsys(root, ss)
5150                         seq_printf(m, "%s%s", count++ ? "," : "", ss->name);
5151                 if (strlen(root->name))
5152                         seq_printf(m, "%sname=%s", count ? "," : "",
5153                                    root->name);
5154                 seq_putc(m, ':');
5155                 cgrp = task_cgroup_from_root(tsk, root);
5156                 retval = cgroup_path(cgrp, buf, PAGE_SIZE);
5157                 if (retval < 0)
5158                         goto out_unlock;
5159                 seq_puts(m, buf);
5160                 seq_putc(m, '\n');
5161         }
5162 
5163 out_unlock:
5164         mutex_unlock(&cgroup_mutex);
5165         put_task_struct(tsk);
5166 out_free:
5167         kfree(buf);
5168 out:
5169         return retval;
5170 }
5171 
5172 /* Display information about each subsystem and each hierarchy */
5173 static int proc_cgroupstats_show(struct seq_file *m, void *v)
5174 {
5175         struct cgroup_subsys *ss;
5176         int i;
5177 
5178         seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
5179         /*
5180          * ideally we don't want subsystems moving around while we do this.
5181          * cgroup_mutex is also necessary to guarantee an atomic snapshot of
5182          * subsys/hierarchy state.
5183          */
5184         mutex_lock(&cgroup_mutex);
5185 
5186         for_each_subsys(ss, i)
5187                 seq_printf(m, "%s\t%d\t%d\t%d\n",
5188                            ss->name, ss->root->hierarchy_id,
5189                            ss->root->number_of_cgroups, !ss->disabled);
5190 
5191         mutex_unlock(&cgroup_mutex);
5192         return 0;
5193 }
5194 
5195 static int cgroupstats_open(struct inode *inode, struct file *file)
5196 {
5197         return single_open(file, proc_cgroupstats_show, NULL);
5198 }
5199 
5200 static const struct file_operations proc_cgroupstats_operations = {
5201         .open = cgroupstats_open,
5202         .read = seq_read,
5203         .llseek = seq_lseek,
5204         .release = single_release,
5205 };
5206 
5207 /**
5208  * cgroup_fork - attach newly forked task to its parents cgroup.
5209  * @child: pointer to task_struct of forking parent process.
5210  *
5211  * Description: A task inherits its parent's cgroup at fork().
5212  *
5213  * A pointer to the shared css_set was automatically copied in
5214  * fork.c by dup_task_struct().  However, we ignore that copy, since
5215  * it was not made under the protection of RCU or cgroup_mutex, so
5216  * might no longer be a valid cgroup pointer.  cgroup_attach_task() might
5217  * have already changed current->cgroups, allowing the previously
5218  * referenced cgroup group to be removed and freed.
5219  *
5220  * At the point that cgroup_fork() is called, 'current' is the parent
5221  * task, and the passed argument 'child' points to the child task.
5222  */
5223 void cgroup_fork(struct task_struct *child)
5224 {
5225         task_lock(current);
5226         get_css_set(task_css_set(current));
5227         child->cgroups = current->cgroups;
5228         task_unlock(current);
5229         INIT_LIST_HEAD(&child->cg_list);
5230 }
5231 
5232 /**
5233  * cgroup_post_fork - called on a new task after adding it to the task list
5234  * @child: the task in question
5235  *
5236  * Adds the task to the list running through its css_set if necessary and
5237  * call the subsystem fork() callbacks.  Has to be after the task is
5238  * visible on the task list in case we race with the first call to
5239  * cgroup_task_iter_start() - to guarantee that the new task ends up on its
5240  * list.
5241  */
5242 void cgroup_post_fork(struct task_struct *child)
5243 {
5244         struct cgroup_subsys *ss;
5245         int i;
5246 
5247         /*
5248          * use_task_css_set_links is set to 1 before we walk the tasklist
5249          * under the tasklist_lock and we read it here after we added the child
5250          * to the tasklist under the tasklist_lock as well. If the child wasn't
5251          * yet in the tasklist when we walked through it from
5252          * cgroup_enable_task_cg_lists(), then use_task_css_set_links value
5253          * should be visible now due to the paired locking and barriers implied
5254          * by LOCK/UNLOCK: it is written before the tasklist_lock unlock
5255          * in cgroup_enable_task_cg_lists() and read here after the tasklist_lock
5256          * lock on fork.
5257          */
5258         if (use_task_css_set_links) {
5259                 write_lock(&css_set_lock);
5260                 task_lock(child);
5261                 if (list_empty(&child->cg_list))
5262                         list_add(&child->cg_list, &task_css_set(child)->tasks);
5263                 task_unlock(child);
5264                 write_unlock(&css_set_lock);
5265         }
5266 
5267         /*
5268          * Call ss->fork().  This must happen after @child is linked on
5269          * css_set; otherwise, @child might change state between ->fork()
5270          * and addition to css_set.
5271          */
5272         if (need_forkexit_callback) {
5273                 /*
5274                  * fork/exit callbacks are supported only for builtin
5275                  * subsystems, and the builtin section of the subsys
5276                  * array is immutable, so we don't need to lock the
5277                  * subsys array here. On the other hand, modular section
5278                  * of the array can be freed at module unload, so we
5279                  * can't touch that.
5280                  */
5281                 for_each_builtin_subsys(ss, i)
5282                         if (ss->fork)
5283                                 ss->fork(child);
5284         }
5285 }
5286 
5287 /**
5288  * cgroup_exit - detach cgroup from exiting task
5289  * @tsk: pointer to task_struct of exiting process
5290  * @run_callback: run exit callbacks?
5291  *
5292  * Description: Detach cgroup from @tsk and release it.
5293  *
5294  * Note that cgroups marked notify_on_release force every task in
5295  * them to take the global cgroup_mutex mutex when exiting.
5296  * This could impact scaling on very large systems.  Be reluctant to
5297  * use notify_on_release cgroups where very high task exit scaling
5298  * is required on large systems.
5299  *
5300  * the_top_cgroup_hack:
5301  *
5302  *    Set the exiting tasks cgroup to the root cgroup (top_cgroup).
5303  *
5304  *    We call cgroup_exit() while the task is still competent to
5305  *    handle notify_on_release(), then leave the task attached to the
5306  *    root cgroup in each hierarchy for the remainder of its exit.
5307  *
5308  *    To do this properly, we would increment the reference count on
5309  *    top_cgroup, and near the very end of the kernel/exit.c do_exit()
5310  *    code we would add a second cgroup function call, to drop that
5311  *    reference.  This would just create an unnecessary hot spot on
5312  *    the top_cgroup reference count, to no avail.
5313  *
5314  *    Normally, holding a reference to a cgroup without bumping its
5315  *    count is unsafe.   The cgroup could go away, or someone could
5316  *    attach us to a different cgroup, decrementing the count on
5317  *    the first cgroup that we never incremented.  But in this case,
5318  *    top_cgroup isn't going away, and either task has PF_EXITING set,
5319  *    which wards off any cgroup_attach_task() attempts, or task is a failed
5320  *    fork, never visible to cgroup_attach_task.
5321  */
5322 void cgroup_exit(struct task_struct *tsk, int run_callbacks)
5323 {
5324         struct cgroup_subsys *ss;
5325         struct css_set *cset;
5326         int i;
5327 
5328         /*
5329          * Unlink from the css_set task list if necessary.
5330          * Optimistically check cg_list before taking
5331          * css_set_lock
5332          */
5333         if (!list_empty(&tsk->cg_list)) {
5334                 write_lock(&css_set_lock);
5335                 if (!list_empty(&tsk->cg_list))
5336                         list_del_init(&tsk->cg_list);
5337                 write_unlock(&css_set_lock);
5338         }
5339 
5340         /* Reassign the task to the init_css_set. */
5341         task_lock(tsk);
5342         cset = task_css_set(tsk);
5343         RCU_INIT_POINTER(tsk->cgroups, &init_css_set);
5344 
5345         if (run_callbacks && need_forkexit_callback) {
5346                 /*
5347                  * fork/exit callbacks are supported only for builtin
5348                  * subsystems, see cgroup_post_fork() for details.
5349                  */
5350                 for_each_builtin_subsys(ss, i) {
5351                         if (ss->exit) {
5352                                 struct cgroup_subsys_state *old_css = cset->subsys[i];
5353                                 struct cgroup_subsys_state *css = task_css(tsk, i);
5354 
5355                                 ss->exit(css, old_css, tsk);
5356                         }
5357                 }
5358         }
5359         task_unlock(tsk);
5360 
5361         put_css_set_taskexit(cset);
5362 }
5363 
5364 static void check_for_release(struct cgroup *cgrp)
5365 {
5366         if (cgroup_is_releasable(cgrp) &&
5367             list_empty(&cgrp->cset_links) && list_empty(&cgrp->children)) {
5368                 /*
5369                  * Control Group is currently removeable. If it's not
5370                  * already queued for a userspace notification, queue
5371                  * it now
5372                  */
5373                 int need_schedule_work = 0;
5374 
5375                 raw_spin_lock(&release_list_lock);
5376                 if (!cgroup_is_dead(cgrp) &&
5377                     list_empty(&cgrp->release_list)) {
5378                         list_add(&cgrp->release_list, &release_list);
5379                         need_schedule_work = 1;
5380                 }
5381                 raw_spin_unlock(&release_list_lock);
5382                 if (need_schedule_work)
5383                         schedule_work(&release_agent_work);
5384         }
5385 }
5386 
5387 /*
5388  * Notify userspace when a cgroup is released, by running the
5389  * configured release agent with the name of the cgroup (path
5390  * relative to the root of cgroup file system) as the argument.
5391  *
5392  * Most likely, this user command will try to rmdir this cgroup.
5393  *
5394  * This races with the possibility that some other task will be
5395  * attached to this cgroup before it is removed, or that some other
5396  * user task will 'mkdir' a child cgroup of this cgroup.  That's ok.
5397  * The presumed 'rmdir' will fail quietly if this cgroup is no longer
5398  * unused, and this cgroup will be reprieved from its death sentence,
5399  * to continue to serve a useful existence.  Next time it's released,
5400  * we will get notified again, if it still has 'notify_on_release' set.
5401  *
5402  * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
5403  * means only wait until the task is successfully execve()'d.  The
5404  * separate release agent task is forked by call_usermodehelper(),
5405  * then control in this thread returns here, without waiting for the
5406  * release agent task.  We don't bother to wait because the caller of
5407  * this routine has no use for the exit status of the release agent
5408  * task, so no sense holding our caller up for that.
5409  */
5410 static void cgroup_release_agent(struct work_struct *work)
5411 {
5412         BUG_ON(work != &release_agent_work);
5413         mutex_lock(&cgroup_mutex);
5414         raw_spin_lock(&release_list_lock);
5415         while (!list_empty(&release_list)) {
5416                 char *argv[3], *envp[3];
5417                 int i;
5418                 char *pathbuf = NULL, *agentbuf = NULL;
5419                 struct cgroup *cgrp = list_entry(release_list.next,
5420                                                     struct cgroup,
5421                                                     release_list);
5422                 list_del_init(&cgrp->release_list);
5423                 raw_spin_unlock(&release_list_lock);
5424                 pathbuf = kmalloc(PAGE_SIZE, GFP_KERNEL);
5425                 if (!pathbuf)
5426                         goto continue_free;
5427                 if (cgroup_path(cgrp, pathbuf, PAGE_SIZE) < 0)
5428                         goto continue_free;
5429                 agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL);
5430                 if (!agentbuf)
5431                         goto continue_free;
5432 
5433                 i = 0;
5434                 argv[i++] = agentbuf;
5435                 argv[i++] = pathbuf;
5436                 argv[i] = NULL;
5437 
5438                 i = 0;
5439                 /* minimal command environment */
5440                 envp[i++] = "HOME=/";
5441                 envp[i++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
5442                 envp[i] = NULL;
5443 
5444                 /* Drop the lock while we invoke the usermode helper,
5445                  * since the exec could involve hitting disk and hence
5446                  * be a slow process */
5447                 mutex_unlock(&cgroup_mutex);
5448                 call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC);
5449                 mutex_lock(&cgroup_mutex);
5450  continue_free:
5451                 kfree(pathbuf);
5452                 kfree(agentbuf);
5453                 raw_spin_lock(&release_list_lock);
5454         }
5455         raw_spin_unlock(&release_list_lock);
5456         mutex_unlock(&cgroup_mutex);
5457 }
5458 
5459 static int __init cgroup_disable(char *str)
5460 {
5461         struct cgroup_subsys *ss;
5462         char *token;
5463         int i;
5464 
5465         while ((token = strsep(&str, ",")) != NULL) {
5466                 if (!*token)
5467                         continue;
5468 
5469                 /*
5470                  * cgroup_disable, being at boot time, can't know about
5471                  * module subsystems, so we don't worry about them.
5472                  */
5473                 for_each_builtin_subsys(ss, i) {
5474                         if (!strcmp(token, ss->name)) {
5475                                 ss->disabled = 1;
5476                                 printk(KERN_INFO "Disabling %s control group"
5477                                         " subsystem\n", ss->name);
5478                                 break;
5479                         }
5480                 }
5481         }
5482         return 1;
5483 }
5484 __setup("cgroup_disable=", cgroup_disable);
5485 
5486 /**
5487  * css_from_dir - get corresponding css from the dentry of a cgroup dir
5488  * @dentry: directory dentry of interest
5489  * @ss: subsystem of interest
5490  *
5491  * Must be called under RCU read lock.  The caller is responsible for
5492  * pinning the returned css if it needs to be accessed outside the RCU
5493  * critical section.
5494  */
5495 struct cgroup_subsys_state *css_from_dir(struct dentry *dentry,
5496                                          struct cgroup_subsys *ss)
5497 {
5498         struct cgroup *cgrp;
5499 
5500         WARN_ON_ONCE(!rcu_read_lock_held());
5501 
5502         /* is @dentry a cgroup dir? */
5503         if (!dentry->d_inode ||
5504             dentry->d_inode->i_op != &cgroup_dir_inode_operations)
5505                 return ERR_PTR(-EBADF);
5506 
5507         cgrp = __d_cgrp(dentry);
5508         return cgroup_css(cgrp, ss) ?: ERR_PTR(-ENOENT);
5509 }
5510 
5511 /**
5512  * css_from_id - lookup css by id
5513  * @id: the cgroup id
5514  * @ss: cgroup subsys to be looked into
5515  *
5516  * Returns the css if there's valid one with @id, otherwise returns NULL.
5517  * Should be called under rcu_read_lock().
5518  */
5519 struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss)
5520 {
5521         struct cgroup *cgrp;
5522 
5523         rcu_lockdep_assert(rcu_read_lock_held() ||
5524                            lockdep_is_held(&cgroup_mutex),
5525                            "css_from_id() needs proper protection");
5526 
5527         cgrp = idr_find(&ss->root->cgroup_idr, id);
5528         if (cgrp)
5529                 return cgroup_css(cgrp, ss);
5530         return NULL;
5531 }
5532 
5533 #ifdef CONFIG_CGROUP_DEBUG
5534 static struct cgroup_subsys_state *
5535 debug_css_alloc(struct cgroup_subsys_state *parent_css)
5536 {
5537         struct cgroup_subsys_state *css = kzalloc(sizeof(*css), GFP_KERNEL);
5538 
5539         if (!css)
5540                 return ERR_PTR(-ENOMEM);
5541 
5542         return css;
5543 }
5544 
5545 static void debug_css_free(struct cgroup_subsys_state *css)
5546 {
5547         kfree(css);
5548 }
5549 
5550 static u64 debug_taskcount_read(struct cgroup_subsys_state *css,
5551                                 struct cftype *cft)
5552 {
5553         return cgroup_task_count(css->cgroup);
5554 }
5555 
5556 static u64 current_css_set_read(struct cgroup_subsys_state *css,
5557                                 struct cftype *cft)
5558 {
5559         return (u64)(unsigned long)current->cgroups;
5560 }
5561 
5562 static u64 current_css_set_refcount_read(struct cgroup_subsys_state *css,
5563                                          struct cftype *cft)
5564 {
5565         u64 count;
5566 
5567         rcu_read_lock();
5568         count = atomic_read(&task_css_set(current)->refcount);
5569         rcu_read_unlock();
5570         return count;
5571 }
5572 
5573 static int current_css_set_cg_links_read(struct cgroup_subsys_state *css,
5574                                          struct cftype *cft,
5575                                          struct seq_file *seq)
5576 {
5577         struct cgrp_cset_link *link;
5578         struct css_set *cset;
5579 
5580         read_lock(&css_set_lock);
5581         rcu_read_lock();
5582         cset = rcu_dereference(current->cgroups);
5583         list_for_each_entry(link, &cset->cgrp_links, cgrp_link) {
5584                 struct cgroup *c = link->cgrp;
5585                 const char *name;
5586 
5587                 if (c->dentry)
5588                         name = c->dentry->d_name.name;
5589                 else
5590                         name = "?";
5591                 seq_printf(seq, "Root %d group %s\n",
5592                            c->root->hierarchy_id, name);
5593         }
5594         rcu_read_unlock();
5595         read_unlock(&css_set_lock);
5596         return 0;
5597 }
5598 
5599 #define MAX_TASKS_SHOWN_PER_CSS 25
5600 static int cgroup_css_links_read(struct cgroup_subsys_state *css,
5601                                  struct cftype *cft, struct seq_file *seq)
5602 {
5603         struct cgrp_cset_link *link;
5604 
5605         read_lock(&css_set_lock);
5606         list_for_each_entry(link, &css->cgroup->cset_links, cset_link) {
5607                 struct css_set *cset = link->cset;
5608                 struct task_struct *task;
5609                 int count = 0;
5610                 seq_printf(seq, "css_set %p\n", cset);
5611                 list_for_each_entry(task, &cset->tasks, cg_list) {
5612                         if (count++ > MAX_TASKS_SHOWN_PER_CSS) {
5613                                 seq_puts(seq, "  ...\n");
5614                                 break;
5615                         } else {
5616                                 seq_printf(seq, "  task %d\n",
5617                                            task_pid_vnr(task));
5618                         }
5619                 }
5620         }
5621         read_unlock(&css_set_lock);
5622         return 0;
5623 }
5624 
5625 static u64 releasable_read(struct cgroup_subsys_state *css, struct cftype *cft)
5626 {
5627         return test_bit(CGRP_RELEASABLE, &css->cgroup->flags);
5628 }
5629 
5630 static struct cftype debug_files[] =  {
5631         {
5632                 .name = "taskcount",
5633                 .read_u64 = debug_taskcount_read,
5634         },
5635 
5636         {
5637                 .name = "current_css_set",
5638                 .read_u64 = current_css_set_read,
5639         },
5640 
5641         {
5642                 .name = "current_css_set_refcount",
5643                 .read_u64 = current_css_set_refcount_read,
5644         },
5645 
5646         {
5647                 .name = "current_css_set_cg_links",
5648                 .read_seq_string = current_css_set_cg_links_read,
5649         },
5650 
5651         {
5652                 .name = "cgroup_css_links",
5653                 .read_seq_string = cgroup_css_links_read,
5654         },
5655 
5656         {
5657                 .name = "releasable",
5658                 .read_u64 = releasable_read,
5659         },
5660 
5661         { }     /* terminate */
5662 };
5663 
5664 struct cgroup_subsys debug_subsys = {
5665         .name = "debug",
5666         .css_alloc = debug_css_alloc,
5667         .css_free = debug_css_free,
5668         .subsys_id = debug_subsys_id,
5669         .base_cftypes = debug_files,
5670 };
5671 #endif /* CONFIG_CGROUP_DEBUG */
5672 

This page was automatically generated by LXR 0.3.1 (source).  •  Linux is a registered trademark of Linus Torvalds  •  Contact us