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

TOMOYO Linux Cross Reference
Linux/kernel/cgroup/cgroup.c

Version: ~ [ linux-5.16-rc3 ] ~ [ linux-5.15.5 ] ~ [ linux-5.14.21 ] ~ [ linux-5.13.19 ] ~ [ linux-5.12.19 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.82 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.162 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.218 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.256 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.291 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.293 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.18.140 ] ~ [ linux-3.16.85 ] ~ [ linux-3.14.79 ] ~ [ linux-3.12.74 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 /*
  2  *  Generic process-grouping system.
  3  *
  4  *  Based originally on the cpuset system, extracted by Paul Menage
  5  *  Copyright (C) 2006 Google, Inc
  6  *
  7  *  Notifications support
  8  *  Copyright (C) 2009 Nokia Corporation
  9  *  Author: Kirill A. Shutemov
 10  *
 11  *  Copyright notices from the original cpuset code:
 12  *  --------------------------------------------------
 13  *  Copyright (C) 2003 BULL SA.
 14  *  Copyright (C) 2004-2006 Silicon Graphics, Inc.
 15  *
 16  *  Portions derived from Patrick Mochel's sysfs code.
 17  *  sysfs is Copyright (c) 2001-3 Patrick Mochel
 18  *
 19  *  2003-10-10 Written by Simon Derr.
 20  *  2003-10-22 Updates by Stephen Hemminger.
 21  *  2004 May-July Rework by Paul Jackson.
 22  *  ---------------------------------------------------
 23  *
 24  *  This file is subject to the terms and conditions of the GNU General Public
 25  *  License.  See the file COPYING in the main directory of the Linux
 26  *  distribution for more details.
 27  */
 28 
 29 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
 30 
 31 #include "cgroup-internal.h"
 32 
 33 #include <linux/cred.h>
 34 #include <linux/errno.h>
 35 #include <linux/init_task.h>
 36 #include <linux/kernel.h>
 37 #include <linux/magic.h>
 38 #include <linux/mutex.h>
 39 #include <linux/mount.h>
 40 #include <linux/pagemap.h>
 41 #include <linux/proc_fs.h>
 42 #include <linux/rcupdate.h>
 43 #include <linux/sched.h>
 44 #include <linux/sched/task.h>
 45 #include <linux/slab.h>
 46 #include <linux/spinlock.h>
 47 #include <linux/percpu-rwsem.h>
 48 #include <linux/string.h>
 49 #include <linux/hashtable.h>
 50 #include <linux/idr.h>
 51 #include <linux/kthread.h>
 52 #include <linux/atomic.h>
 53 #include <linux/cpuset.h>
 54 #include <linux/proc_ns.h>
 55 #include <linux/nsproxy.h>
 56 #include <linux/file.h>
 57 #include <linux/fs_parser.h>
 58 #include <linux/sched/cputime.h>
 59 #include <linux/psi.h>
 60 #include <net/sock.h>
 61 
 62 #define CREATE_TRACE_POINTS
 63 #include <trace/events/cgroup.h>
 64 
 65 #define CGROUP_FILE_NAME_MAX            (MAX_CGROUP_TYPE_NAMELEN +      \
 66                                          MAX_CFTYPE_NAME + 2)
 67 /* let's not notify more than 100 times per second */
 68 #define CGROUP_FILE_NOTIFY_MIN_INTV     DIV_ROUND_UP(HZ, 100)
 69 
 70 /*
 71  * To avoid confusing the compiler (and generating warnings) with code
 72  * that attempts to access what would be a 0-element array (i.e. sized
 73  * to a potentially empty array when CGROUP_SUBSYS_COUNT == 0), this
 74  * constant expression can be added.
 75  */
 76 #define CGROUP_HAS_SUBSYS_CONFIG        (CGROUP_SUBSYS_COUNT > 0)
 77 
 78 /*
 79  * cgroup_mutex is the master lock.  Any modification to cgroup or its
 80  * hierarchy must be performed while holding it.
 81  *
 82  * css_set_lock protects task->cgroups pointer, the list of css_set
 83  * objects, and the chain of tasks off each css_set.
 84  *
 85  * These locks are exported if CONFIG_PROVE_RCU so that accessors in
 86  * cgroup.h can use them for lockdep annotations.
 87  */
 88 DEFINE_MUTEX(cgroup_mutex);
 89 DEFINE_SPINLOCK(css_set_lock);
 90 
 91 #ifdef CONFIG_PROVE_RCU
 92 EXPORT_SYMBOL_GPL(cgroup_mutex);
 93 EXPORT_SYMBOL_GPL(css_set_lock);
 94 #endif
 95 
 96 DEFINE_SPINLOCK(trace_cgroup_path_lock);
 97 char trace_cgroup_path[TRACE_CGROUP_PATH_LEN];
 98 bool cgroup_debug __read_mostly;
 99 
100 /*
101  * Protects cgroup_idr and css_idr so that IDs can be released without
102  * grabbing cgroup_mutex.
103  */
104 static DEFINE_SPINLOCK(cgroup_idr_lock);
105 
106 /*
107  * Protects cgroup_file->kn for !self csses.  It synchronizes notifications
108  * against file removal/re-creation across css hiding.
109  */
110 static DEFINE_SPINLOCK(cgroup_file_kn_lock);
111 
112 DEFINE_PERCPU_RWSEM(cgroup_threadgroup_rwsem);
113 
114 #define cgroup_assert_mutex_or_rcu_locked()                             \
115         RCU_LOCKDEP_WARN(!rcu_read_lock_held() &&                       \
116                            !lockdep_is_held(&cgroup_mutex),             \
117                            "cgroup_mutex or RCU read lock required");
118 
119 /*
120  * cgroup destruction makes heavy use of work items and there can be a lot
121  * of concurrent destructions.  Use a separate workqueue so that cgroup
122  * destruction work items don't end up filling up max_active of system_wq
123  * which may lead to deadlock.
124  */
125 static struct workqueue_struct *cgroup_destroy_wq;
126 
127 /* generate an array of cgroup subsystem pointers */
128 #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys,
129 struct cgroup_subsys *cgroup_subsys[] = {
130 #include <linux/cgroup_subsys.h>
131 };
132 #undef SUBSYS
133 
134 /* array of cgroup subsystem names */
135 #define SUBSYS(_x) [_x ## _cgrp_id] = #_x,
136 static const char *cgroup_subsys_name[] = {
137 #include <linux/cgroup_subsys.h>
138 };
139 #undef SUBSYS
140 
141 /* array of static_keys for cgroup_subsys_enabled() and cgroup_subsys_on_dfl() */
142 #define SUBSYS(_x)                                                              \
143         DEFINE_STATIC_KEY_TRUE(_x ## _cgrp_subsys_enabled_key);                 \
144         DEFINE_STATIC_KEY_TRUE(_x ## _cgrp_subsys_on_dfl_key);                  \
145         EXPORT_SYMBOL_GPL(_x ## _cgrp_subsys_enabled_key);                      \
146         EXPORT_SYMBOL_GPL(_x ## _cgrp_subsys_on_dfl_key);
147 #include <linux/cgroup_subsys.h>
148 #undef SUBSYS
149 
150 #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys_enabled_key,
151 static struct static_key_true *cgroup_subsys_enabled_key[] = {
152 #include <linux/cgroup_subsys.h>
153 };
154 #undef SUBSYS
155 
156 #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys_on_dfl_key,
157 static struct static_key_true *cgroup_subsys_on_dfl_key[] = {
158 #include <linux/cgroup_subsys.h>
159 };
160 #undef SUBSYS
161 
162 static DEFINE_PER_CPU(struct cgroup_rstat_cpu, cgrp_dfl_root_rstat_cpu);
163 
164 /* the default hierarchy */
165 struct cgroup_root cgrp_dfl_root = { .cgrp.rstat_cpu = &cgrp_dfl_root_rstat_cpu };
166 EXPORT_SYMBOL_GPL(cgrp_dfl_root);
167 
168 /*
169  * The default hierarchy always exists but is hidden until mounted for the
170  * first time.  This is for backward compatibility.
171  */
172 static bool cgrp_dfl_visible;
173 
174 /* some controllers are not supported in the default hierarchy */
175 static u16 cgrp_dfl_inhibit_ss_mask;
176 
177 /* some controllers are implicitly enabled on the default hierarchy */
178 static u16 cgrp_dfl_implicit_ss_mask;
179 
180 /* some controllers can be threaded on the default hierarchy */
181 static u16 cgrp_dfl_threaded_ss_mask;
182 
183 /* The list of hierarchy roots */
184 LIST_HEAD(cgroup_roots);
185 static int cgroup_root_count;
186 
187 /* hierarchy ID allocation and mapping, protected by cgroup_mutex */
188 static DEFINE_IDR(cgroup_hierarchy_idr);
189 
190 /*
191  * Assign a monotonically increasing serial number to csses.  It guarantees
192  * cgroups with bigger numbers are newer than those with smaller numbers.
193  * Also, as csses are always appended to the parent's ->children list, it
194  * guarantees that sibling csses are always sorted in the ascending serial
195  * number order on the list.  Protected by cgroup_mutex.
196  */
197 static u64 css_serial_nr_next = 1;
198 
199 /*
200  * These bitmasks identify subsystems with specific features to avoid
201  * having to do iterative checks repeatedly.
202  */
203 static u16 have_fork_callback __read_mostly;
204 static u16 have_exit_callback __read_mostly;
205 static u16 have_release_callback __read_mostly;
206 static u16 have_canfork_callback __read_mostly;
207 
208 /* cgroup namespace for init task */
209 struct cgroup_namespace init_cgroup_ns = {
210         .ns.count       = REFCOUNT_INIT(2),
211         .user_ns        = &init_user_ns,
212         .ns.ops         = &cgroupns_operations,
213         .ns.inum        = PROC_CGROUP_INIT_INO,
214         .root_cset      = &init_css_set,
215 };
216 
217 static struct file_system_type cgroup2_fs_type;
218 static struct cftype cgroup_base_files[];
219 
220 /* cgroup optional features */
221 enum cgroup_opt_features {
222 #ifdef CONFIG_PSI
223         OPT_FEATURE_PRESSURE,
224 #endif
225         OPT_FEATURE_COUNT
226 };
227 
228 static const char *cgroup_opt_feature_names[OPT_FEATURE_COUNT] = {
229 #ifdef CONFIG_PSI
230         "pressure",
231 #endif
232 };
233 
234 static u16 cgroup_feature_disable_mask __read_mostly;
235 
236 static int cgroup_apply_control(struct cgroup *cgrp);
237 static void cgroup_finalize_control(struct cgroup *cgrp, int ret);
238 static void css_task_iter_skip(struct css_task_iter *it,
239                                struct task_struct *task);
240 static int cgroup_destroy_locked(struct cgroup *cgrp);
241 static struct cgroup_subsys_state *css_create(struct cgroup *cgrp,
242                                               struct cgroup_subsys *ss);
243 static void css_release(struct percpu_ref *ref);
244 static void kill_css(struct cgroup_subsys_state *css);
245 static int cgroup_addrm_files(struct cgroup_subsys_state *css,
246                               struct cgroup *cgrp, struct cftype cfts[],
247                               bool is_add);
248 
249 /**
250  * cgroup_ssid_enabled - cgroup subsys enabled test by subsys ID
251  * @ssid: subsys ID of interest
252  *
253  * cgroup_subsys_enabled() can only be used with literal subsys names which
254  * is fine for individual subsystems but unsuitable for cgroup core.  This
255  * is slower static_key_enabled() based test indexed by @ssid.
256  */
257 bool cgroup_ssid_enabled(int ssid)
258 {
259         if (!CGROUP_HAS_SUBSYS_CONFIG)
260                 return false;
261 
262         return static_key_enabled(cgroup_subsys_enabled_key[ssid]);
263 }
264 
265 /**
266  * cgroup_on_dfl - test whether a cgroup is on the default hierarchy
267  * @cgrp: the cgroup of interest
268  *
269  * The default hierarchy is the v2 interface of cgroup and this function
270  * can be used to test whether a cgroup is on the default hierarchy for
271  * cases where a subsystem should behave differently depending on the
272  * interface version.
273  *
274  * List of changed behaviors:
275  *
276  * - Mount options "noprefix", "xattr", "clone_children", "release_agent"
277  *   and "name" are disallowed.
278  *
279  * - When mounting an existing superblock, mount options should match.
280  *
281  * - Remount is disallowed.
282  *
283  * - rename(2) is disallowed.
284  *
285  * - "tasks" is removed.  Everything should be at process granularity.  Use
286  *   "cgroup.procs" instead.
287  *
288  * - "cgroup.procs" is not sorted.  pids will be unique unless they got
289  *   recycled in-between reads.
290  *
291  * - "release_agent" and "notify_on_release" are removed.  Replacement
292  *   notification mechanism will be implemented.
293  *
294  * - "cgroup.clone_children" is removed.
295  *
296  * - "cgroup.subtree_populated" is available.  Its value is 0 if the cgroup
297  *   and its descendants contain no task; otherwise, 1.  The file also
298  *   generates kernfs notification which can be monitored through poll and
299  *   [di]notify when the value of the file changes.
300  *
301  * - cpuset: tasks will be kept in empty cpusets when hotplug happens and
302  *   take masks of ancestors with non-empty cpus/mems, instead of being
303  *   moved to an ancestor.
304  *
305  * - cpuset: a task can be moved into an empty cpuset, and again it takes
306  *   masks of ancestors.
307  *
308  * - blkcg: blk-throttle becomes properly hierarchical.
309  *
310  * - debug: disallowed on the default hierarchy.
311  */
312 bool cgroup_on_dfl(const struct cgroup *cgrp)
313 {
314         return cgrp->root == &cgrp_dfl_root;
315 }
316 
317 /* IDR wrappers which synchronize using cgroup_idr_lock */
318 static int cgroup_idr_alloc(struct idr *idr, void *ptr, int start, int end,
319                             gfp_t gfp_mask)
320 {
321         int ret;
322 
323         idr_preload(gfp_mask);
324         spin_lock_bh(&cgroup_idr_lock);
325         ret = idr_alloc(idr, ptr, start, end, gfp_mask & ~__GFP_DIRECT_RECLAIM);
326         spin_unlock_bh(&cgroup_idr_lock);
327         idr_preload_end();
328         return ret;
329 }
330 
331 static void *cgroup_idr_replace(struct idr *idr, void *ptr, int id)
332 {
333         void *ret;
334 
335         spin_lock_bh(&cgroup_idr_lock);
336         ret = idr_replace(idr, ptr, id);
337         spin_unlock_bh(&cgroup_idr_lock);
338         return ret;
339 }
340 
341 static void cgroup_idr_remove(struct idr *idr, int id)
342 {
343         spin_lock_bh(&cgroup_idr_lock);
344         idr_remove(idr, id);
345         spin_unlock_bh(&cgroup_idr_lock);
346 }
347 
348 static bool cgroup_has_tasks(struct cgroup *cgrp)
349 {
350         return cgrp->nr_populated_csets;
351 }
352 
353 bool cgroup_is_threaded(struct cgroup *cgrp)
354 {
355         return cgrp->dom_cgrp != cgrp;
356 }
357 
358 /* can @cgrp host both domain and threaded children? */
359 static bool cgroup_is_mixable(struct cgroup *cgrp)
360 {
361         /*
362          * Root isn't under domain level resource control exempting it from
363          * the no-internal-process constraint, so it can serve as a thread
364          * root and a parent of resource domains at the same time.
365          */
366         return !cgroup_parent(cgrp);
367 }
368 
369 /* can @cgrp become a thread root? Should always be true for a thread root */
370 static bool cgroup_can_be_thread_root(struct cgroup *cgrp)
371 {
372         /* mixables don't care */
373         if (cgroup_is_mixable(cgrp))
374                 return true;
375 
376         /* domain roots can't be nested under threaded */
377         if (cgroup_is_threaded(cgrp))
378                 return false;
379 
380         /* can only have either domain or threaded children */
381         if (cgrp->nr_populated_domain_children)
382                 return false;
383 
384         /* and no domain controllers can be enabled */
385         if (cgrp->subtree_control & ~cgrp_dfl_threaded_ss_mask)
386                 return false;
387 
388         return true;
389 }
390 
391 /* is @cgrp root of a threaded subtree? */
392 bool cgroup_is_thread_root(struct cgroup *cgrp)
393 {
394         /* thread root should be a domain */
395         if (cgroup_is_threaded(cgrp))
396                 return false;
397 
398         /* a domain w/ threaded children is a thread root */
399         if (cgrp->nr_threaded_children)
400                 return true;
401 
402         /*
403          * A domain which has tasks and explicit threaded controllers
404          * enabled is a thread root.
405          */
406         if (cgroup_has_tasks(cgrp) &&
407             (cgrp->subtree_control & cgrp_dfl_threaded_ss_mask))
408                 return true;
409 
410         return false;
411 }
412 
413 /* a domain which isn't connected to the root w/o brekage can't be used */
414 static bool cgroup_is_valid_domain(struct cgroup *cgrp)
415 {
416         /* the cgroup itself can be a thread root */
417         if (cgroup_is_threaded(cgrp))
418                 return false;
419 
420         /* but the ancestors can't be unless mixable */
421         while ((cgrp = cgroup_parent(cgrp))) {
422                 if (!cgroup_is_mixable(cgrp) && cgroup_is_thread_root(cgrp))
423                         return false;
424                 if (cgroup_is_threaded(cgrp))
425                         return false;
426         }
427 
428         return true;
429 }
430 
431 /* subsystems visibly enabled on a cgroup */
432 static u16 cgroup_control(struct cgroup *cgrp)
433 {
434         struct cgroup *parent = cgroup_parent(cgrp);
435         u16 root_ss_mask = cgrp->root->subsys_mask;
436 
437         if (parent) {
438                 u16 ss_mask = parent->subtree_control;
439 
440                 /* threaded cgroups can only have threaded controllers */
441                 if (cgroup_is_threaded(cgrp))
442                         ss_mask &= cgrp_dfl_threaded_ss_mask;
443                 return ss_mask;
444         }
445 
446         if (cgroup_on_dfl(cgrp))
447                 root_ss_mask &= ~(cgrp_dfl_inhibit_ss_mask |
448                                   cgrp_dfl_implicit_ss_mask);
449         return root_ss_mask;
450 }
451 
452 /* subsystems enabled on a cgroup */
453 static u16 cgroup_ss_mask(struct cgroup *cgrp)
454 {
455         struct cgroup *parent = cgroup_parent(cgrp);
456 
457         if (parent) {
458                 u16 ss_mask = parent->subtree_ss_mask;
459 
460                 /* threaded cgroups can only have threaded controllers */
461                 if (cgroup_is_threaded(cgrp))
462                         ss_mask &= cgrp_dfl_threaded_ss_mask;
463                 return ss_mask;
464         }
465 
466         return cgrp->root->subsys_mask;
467 }
468 
469 /**
470  * cgroup_css - obtain a cgroup's css for the specified subsystem
471  * @cgrp: the cgroup of interest
472  * @ss: the subsystem of interest (%NULL returns @cgrp->self)
473  *
474  * Return @cgrp's css (cgroup_subsys_state) associated with @ss.  This
475  * function must be called either under cgroup_mutex or rcu_read_lock() and
476  * the caller is responsible for pinning the returned css if it wants to
477  * keep accessing it outside the said locks.  This function may return
478  * %NULL if @cgrp doesn't have @subsys_id enabled.
479  */
480 static struct cgroup_subsys_state *cgroup_css(struct cgroup *cgrp,
481                                               struct cgroup_subsys *ss)
482 {
483         if (CGROUP_HAS_SUBSYS_CONFIG && ss)
484                 return rcu_dereference_check(cgrp->subsys[ss->id],
485                                         lockdep_is_held(&cgroup_mutex));
486         else
487                 return &cgrp->self;
488 }
489 
490 /**
491  * cgroup_tryget_css - try to get a cgroup's css for the specified subsystem
492  * @cgrp: the cgroup of interest
493  * @ss: the subsystem of interest
494  *
495  * Find and get @cgrp's css associated with @ss.  If the css doesn't exist
496  * or is offline, %NULL is returned.
497  */
498 static struct cgroup_subsys_state *cgroup_tryget_css(struct cgroup *cgrp,
499                                                      struct cgroup_subsys *ss)
500 {
501         struct cgroup_subsys_state *css;
502 
503         rcu_read_lock();
504         css = cgroup_css(cgrp, ss);
505         if (css && !css_tryget_online(css))
506                 css = NULL;
507         rcu_read_unlock();
508 
509         return css;
510 }
511 
512 /**
513  * cgroup_e_css_by_mask - obtain a cgroup's effective css for the specified ss
514  * @cgrp: the cgroup of interest
515  * @ss: the subsystem of interest (%NULL returns @cgrp->self)
516  *
517  * Similar to cgroup_css() but returns the effective css, which is defined
518  * as the matching css of the nearest ancestor including self which has @ss
519  * enabled.  If @ss is associated with the hierarchy @cgrp is on, this
520  * function is guaranteed to return non-NULL css.
521  */
522 static struct cgroup_subsys_state *cgroup_e_css_by_mask(struct cgroup *cgrp,
523                                                         struct cgroup_subsys *ss)
524 {
525         lockdep_assert_held(&cgroup_mutex);
526 
527         if (!ss)
528                 return &cgrp->self;
529 
530         /*
531          * This function is used while updating css associations and thus
532          * can't test the csses directly.  Test ss_mask.
533          */
534         while (!(cgroup_ss_mask(cgrp) & (1 << ss->id))) {
535                 cgrp = cgroup_parent(cgrp);
536                 if (!cgrp)
537                         return NULL;
538         }
539 
540         return cgroup_css(cgrp, ss);
541 }
542 
543 /**
544  * cgroup_e_css - obtain a cgroup's effective css for the specified subsystem
545  * @cgrp: the cgroup of interest
546  * @ss: the subsystem of interest
547  *
548  * Find and get the effective css of @cgrp for @ss.  The effective css is
549  * defined as the matching css of the nearest ancestor including self which
550  * has @ss enabled.  If @ss is not mounted on the hierarchy @cgrp is on,
551  * the root css is returned, so this function always returns a valid css.
552  *
553  * The returned css is not guaranteed to be online, and therefore it is the
554  * callers responsibility to try get a reference for it.
555  */
556 struct cgroup_subsys_state *cgroup_e_css(struct cgroup *cgrp,
557                                          struct cgroup_subsys *ss)
558 {
559         struct cgroup_subsys_state *css;
560 
561         if (!CGROUP_HAS_SUBSYS_CONFIG)
562                 return NULL;
563 
564         do {
565                 css = cgroup_css(cgrp, ss);
566 
567                 if (css)
568                         return css;
569                 cgrp = cgroup_parent(cgrp);
570         } while (cgrp);
571 
572         return init_css_set.subsys[ss->id];
573 }
574 
575 /**
576  * cgroup_get_e_css - get a cgroup's effective css for the specified subsystem
577  * @cgrp: the cgroup of interest
578  * @ss: the subsystem of interest
579  *
580  * Find and get the effective css of @cgrp for @ss.  The effective css is
581  * defined as the matching css of the nearest ancestor including self which
582  * has @ss enabled.  If @ss is not mounted on the hierarchy @cgrp is on,
583  * the root css is returned, so this function always returns a valid css.
584  * The returned css must be put using css_put().
585  */
586 struct cgroup_subsys_state *cgroup_get_e_css(struct cgroup *cgrp,
587                                              struct cgroup_subsys *ss)
588 {
589         struct cgroup_subsys_state *css;
590 
591         if (!CGROUP_HAS_SUBSYS_CONFIG)
592                 return NULL;
593 
594         rcu_read_lock();
595 
596         do {
597                 css = cgroup_css(cgrp, ss);
598 
599                 if (css && css_tryget_online(css))
600                         goto out_unlock;
601                 cgrp = cgroup_parent(cgrp);
602         } while (cgrp);
603 
604         css = init_css_set.subsys[ss->id];
605         css_get(css);
606 out_unlock:
607         rcu_read_unlock();
608         return css;
609 }
610 EXPORT_SYMBOL_GPL(cgroup_get_e_css);
611 
612 static void cgroup_get_live(struct cgroup *cgrp)
613 {
614         WARN_ON_ONCE(cgroup_is_dead(cgrp));
615         css_get(&cgrp->self);
616 }
617 
618 /**
619  * __cgroup_task_count - count the number of tasks in a cgroup. The caller
620  * is responsible for taking the css_set_lock.
621  * @cgrp: the cgroup in question
622  */
623 int __cgroup_task_count(const struct cgroup *cgrp)
624 {
625         int count = 0;
626         struct cgrp_cset_link *link;
627 
628         lockdep_assert_held(&css_set_lock);
629 
630         list_for_each_entry(link, &cgrp->cset_links, cset_link)
631                 count += link->cset->nr_tasks;
632 
633         return count;
634 }
635 
636 /**
637  * cgroup_task_count - count the number of tasks in a cgroup.
638  * @cgrp: the cgroup in question
639  */
640 int cgroup_task_count(const struct cgroup *cgrp)
641 {
642         int count;
643 
644         spin_lock_irq(&css_set_lock);
645         count = __cgroup_task_count(cgrp);
646         spin_unlock_irq(&css_set_lock);
647 
648         return count;
649 }
650 
651 struct cgroup_subsys_state *of_css(struct kernfs_open_file *of)
652 {
653         struct cgroup *cgrp = of->kn->parent->priv;
654         struct cftype *cft = of_cft(of);
655 
656         /*
657          * This is open and unprotected implementation of cgroup_css().
658          * seq_css() is only called from a kernfs file operation which has
659          * an active reference on the file.  Because all the subsystem
660          * files are drained before a css is disassociated with a cgroup,
661          * the matching css from the cgroup's subsys table is guaranteed to
662          * be and stay valid until the enclosing operation is complete.
663          */
664         if (CGROUP_HAS_SUBSYS_CONFIG && cft->ss)
665                 return rcu_dereference_raw(cgrp->subsys[cft->ss->id]);
666         else
667                 return &cgrp->self;
668 }
669 EXPORT_SYMBOL_GPL(of_css);
670 
671 /**
672  * for_each_css - iterate all css's of a cgroup
673  * @css: the iteration cursor
674  * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end
675  * @cgrp: the target cgroup to iterate css's of
676  *
677  * Should be called under cgroup_[tree_]mutex.
678  */
679 #define for_each_css(css, ssid, cgrp)                                   \
680         for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++)        \
681                 if (!((css) = rcu_dereference_check(                    \
682                                 (cgrp)->subsys[(ssid)],                 \
683                                 lockdep_is_held(&cgroup_mutex)))) { }   \
684                 else
685 
686 /**
687  * for_each_e_css - iterate all effective css's of a cgroup
688  * @css: the iteration cursor
689  * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end
690  * @cgrp: the target cgroup to iterate css's of
691  *
692  * Should be called under cgroup_[tree_]mutex.
693  */
694 #define for_each_e_css(css, ssid, cgrp)                                     \
695         for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++)            \
696                 if (!((css) = cgroup_e_css_by_mask(cgrp,                    \
697                                                    cgroup_subsys[(ssid)]))) \
698                         ;                                                   \
699                 else
700 
701 /**
702  * do_each_subsys_mask - filter for_each_subsys with a bitmask
703  * @ss: the iteration cursor
704  * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end
705  * @ss_mask: the bitmask
706  *
707  * The block will only run for cases where the ssid-th bit (1 << ssid) of
708  * @ss_mask is set.
709  */
710 #define do_each_subsys_mask(ss, ssid, ss_mask) do {                     \
711         unsigned long __ss_mask = (ss_mask);                            \
712         if (!CGROUP_HAS_SUBSYS_CONFIG) {                                \
713                 (ssid) = 0;                                             \
714                 break;                                                  \
715         }                                                               \
716         for_each_set_bit(ssid, &__ss_mask, CGROUP_SUBSYS_COUNT) {       \
717                 (ss) = cgroup_subsys[ssid];                             \
718                 {
719 
720 #define while_each_subsys_mask()                                        \
721                 }                                                       \
722         }                                                               \
723 } while (false)
724 
725 /* iterate over child cgrps, lock should be held throughout iteration */
726 #define cgroup_for_each_live_child(child, cgrp)                         \
727         list_for_each_entry((child), &(cgrp)->self.children, self.sibling) \
728                 if (({ lockdep_assert_held(&cgroup_mutex);              \
729                        cgroup_is_dead(child); }))                       \
730                         ;                                               \
731                 else
732 
733 /* walk live descendants in pre order */
734 #define cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp)          \
735         css_for_each_descendant_pre((d_css), cgroup_css((cgrp), NULL))  \
736                 if (({ lockdep_assert_held(&cgroup_mutex);              \
737                        (dsct) = (d_css)->cgroup;                        \
738                        cgroup_is_dead(dsct); }))                        \
739                         ;                                               \
740                 else
741 
742 /* walk live descendants in postorder */
743 #define cgroup_for_each_live_descendant_post(dsct, d_css, cgrp)         \
744         css_for_each_descendant_post((d_css), cgroup_css((cgrp), NULL)) \
745                 if (({ lockdep_assert_held(&cgroup_mutex);              \
746                        (dsct) = (d_css)->cgroup;                        \
747                        cgroup_is_dead(dsct); }))                        \
748                         ;                                               \
749                 else
750 
751 /*
752  * The default css_set - used by init and its children prior to any
753  * hierarchies being mounted. It contains a pointer to the root state
754  * for each subsystem. Also used to anchor the list of css_sets. Not
755  * reference-counted, to improve performance when child cgroups
756  * haven't been created.
757  */
758 struct css_set init_css_set = {
759         .refcount               = REFCOUNT_INIT(1),
760         .dom_cset               = &init_css_set,
761         .tasks                  = LIST_HEAD_INIT(init_css_set.tasks),
762         .mg_tasks               = LIST_HEAD_INIT(init_css_set.mg_tasks),
763         .dying_tasks            = LIST_HEAD_INIT(init_css_set.dying_tasks),
764         .task_iters             = LIST_HEAD_INIT(init_css_set.task_iters),
765         .threaded_csets         = LIST_HEAD_INIT(init_css_set.threaded_csets),
766         .cgrp_links             = LIST_HEAD_INIT(init_css_set.cgrp_links),
767         .mg_preload_node        = LIST_HEAD_INIT(init_css_set.mg_preload_node),
768         .mg_node                = LIST_HEAD_INIT(init_css_set.mg_node),
769 
770         /*
771          * The following field is re-initialized when this cset gets linked
772          * in cgroup_init().  However, let's initialize the field
773          * statically too so that the default cgroup can be accessed safely
774          * early during boot.
775          */
776         .dfl_cgrp               = &cgrp_dfl_root.cgrp,
777 };
778 
779 static int css_set_count        = 1;    /* 1 for init_css_set */
780 
781 static bool css_set_threaded(struct css_set *cset)
782 {
783         return cset->dom_cset != cset;
784 }
785 
786 /**
787  * css_set_populated - does a css_set contain any tasks?
788  * @cset: target css_set
789  *
790  * css_set_populated() should be the same as !!cset->nr_tasks at steady
791  * state. However, css_set_populated() can be called while a task is being
792  * added to or removed from the linked list before the nr_tasks is
793  * properly updated. Hence, we can't just look at ->nr_tasks here.
794  */
795 static bool css_set_populated(struct css_set *cset)
796 {
797         lockdep_assert_held(&css_set_lock);
798 
799         return !list_empty(&cset->tasks) || !list_empty(&cset->mg_tasks);
800 }
801 
802 /**
803  * cgroup_update_populated - update the populated count of a cgroup
804  * @cgrp: the target cgroup
805  * @populated: inc or dec populated count
806  *
807  * One of the css_sets associated with @cgrp is either getting its first
808  * task or losing the last.  Update @cgrp->nr_populated_* accordingly.  The
809  * count is propagated towards root so that a given cgroup's
810  * nr_populated_children is zero iff none of its descendants contain any
811  * tasks.
812  *
813  * @cgrp's interface file "cgroup.populated" is zero if both
814  * @cgrp->nr_populated_csets and @cgrp->nr_populated_children are zero and
815  * 1 otherwise.  When the sum changes from or to zero, userland is notified
816  * that the content of the interface file has changed.  This can be used to
817  * detect when @cgrp and its descendants become populated or empty.
818  */
819 static void cgroup_update_populated(struct cgroup *cgrp, bool populated)
820 {
821         struct cgroup *child = NULL;
822         int adj = populated ? 1 : -1;
823 
824         lockdep_assert_held(&css_set_lock);
825 
826         do {
827                 bool was_populated = cgroup_is_populated(cgrp);
828 
829                 if (!child) {
830                         cgrp->nr_populated_csets += adj;
831                 } else {
832                         if (cgroup_is_threaded(child))
833                                 cgrp->nr_populated_threaded_children += adj;
834                         else
835                                 cgrp->nr_populated_domain_children += adj;
836                 }
837 
838                 if (was_populated == cgroup_is_populated(cgrp))
839                         break;
840 
841                 cgroup1_check_for_release(cgrp);
842                 TRACE_CGROUP_PATH(notify_populated, cgrp,
843                                   cgroup_is_populated(cgrp));
844                 cgroup_file_notify(&cgrp->events_file);
845 
846                 child = cgrp;
847                 cgrp = cgroup_parent(cgrp);
848         } while (cgrp);
849 }
850 
851 /**
852  * css_set_update_populated - update populated state of a css_set
853  * @cset: target css_set
854  * @populated: whether @cset is populated or depopulated
855  *
856  * @cset is either getting the first task or losing the last.  Update the
857  * populated counters of all associated cgroups accordingly.
858  */
859 static void css_set_update_populated(struct css_set *cset, bool populated)
860 {
861         struct cgrp_cset_link *link;
862 
863         lockdep_assert_held(&css_set_lock);
864 
865         list_for_each_entry(link, &cset->cgrp_links, cgrp_link)
866                 cgroup_update_populated(link->cgrp, populated);
867 }
868 
869 /*
870  * @task is leaving, advance task iterators which are pointing to it so
871  * that they can resume at the next position.  Advancing an iterator might
872  * remove it from the list, use safe walk.  See css_task_iter_skip() for
873  * details.
874  */
875 static void css_set_skip_task_iters(struct css_set *cset,
876                                     struct task_struct *task)
877 {
878         struct css_task_iter *it, *pos;
879 
880         list_for_each_entry_safe(it, pos, &cset->task_iters, iters_node)
881                 css_task_iter_skip(it, task);
882 }
883 
884 /**
885  * css_set_move_task - move a task from one css_set to another
886  * @task: task being moved
887  * @from_cset: css_set @task currently belongs to (may be NULL)
888  * @to_cset: new css_set @task is being moved to (may be NULL)
889  * @use_mg_tasks: move to @to_cset->mg_tasks instead of ->tasks
890  *
891  * Move @task from @from_cset to @to_cset.  If @task didn't belong to any
892  * css_set, @from_cset can be NULL.  If @task is being disassociated
893  * instead of moved, @to_cset can be NULL.
894  *
895  * This function automatically handles populated counter updates and
896  * css_task_iter adjustments but the caller is responsible for managing
897  * @from_cset and @to_cset's reference counts.
898  */
899 static void css_set_move_task(struct task_struct *task,
900                               struct css_set *from_cset, struct css_set *to_cset,
901                               bool use_mg_tasks)
902 {
903         lockdep_assert_held(&css_set_lock);
904 
905         if (to_cset && !css_set_populated(to_cset))
906                 css_set_update_populated(to_cset, true);
907 
908         if (from_cset) {
909                 WARN_ON_ONCE(list_empty(&task->cg_list));
910 
911                 css_set_skip_task_iters(from_cset, task);
912                 list_del_init(&task->cg_list);
913                 if (!css_set_populated(from_cset))
914                         css_set_update_populated(from_cset, false);
915         } else {
916                 WARN_ON_ONCE(!list_empty(&task->cg_list));
917         }
918 
919         if (to_cset) {
920                 /*
921                  * We are synchronized through cgroup_threadgroup_rwsem
922                  * against PF_EXITING setting such that we can't race
923                  * against cgroup_exit()/cgroup_free() dropping the css_set.
924                  */
925                 WARN_ON_ONCE(task->flags & PF_EXITING);
926 
927                 cgroup_move_task(task, to_cset);
928                 list_add_tail(&task->cg_list, use_mg_tasks ? &to_cset->mg_tasks :
929                                                              &to_cset->tasks);
930         }
931 }
932 
933 /*
934  * hash table for cgroup groups. This improves the performance to find
935  * an existing css_set. This hash doesn't (currently) take into
936  * account cgroups in empty hierarchies.
937  */
938 #define CSS_SET_HASH_BITS       7
939 static DEFINE_HASHTABLE(css_set_table, CSS_SET_HASH_BITS);
940 
941 static unsigned long css_set_hash(struct cgroup_subsys_state *css[])
942 {
943         unsigned long key = 0UL;
944         struct cgroup_subsys *ss;
945         int i;
946 
947         for_each_subsys(ss, i)
948                 key += (unsigned long)css[i];
949         key = (key >> 16) ^ key;
950 
951         return key;
952 }
953 
954 void put_css_set_locked(struct css_set *cset)
955 {
956         struct cgrp_cset_link *link, *tmp_link;
957         struct cgroup_subsys *ss;
958         int ssid;
959 
960         lockdep_assert_held(&css_set_lock);
961 
962         if (!refcount_dec_and_test(&cset->refcount))
963                 return;
964 
965         WARN_ON_ONCE(!list_empty(&cset->threaded_csets));
966 
967         /* This css_set is dead. Unlink it and release cgroup and css refs */
968         for_each_subsys(ss, ssid) {
969                 list_del(&cset->e_cset_node[ssid]);
970                 css_put(cset->subsys[ssid]);
971         }
972         hash_del(&cset->hlist);
973         css_set_count--;
974 
975         list_for_each_entry_safe(link, tmp_link, &cset->cgrp_links, cgrp_link) {
976                 list_del(&link->cset_link);
977                 list_del(&link->cgrp_link);
978                 if (cgroup_parent(link->cgrp))
979                         cgroup_put(link->cgrp);
980                 kfree(link);
981         }
982 
983         if (css_set_threaded(cset)) {
984                 list_del(&cset->threaded_csets_node);
985                 put_css_set_locked(cset->dom_cset);
986         }
987 
988         kfree_rcu(cset, rcu_head);
989 }
990 
991 /**
992  * compare_css_sets - helper function for find_existing_css_set().
993  * @cset: candidate css_set being tested
994  * @old_cset: existing css_set for a task
995  * @new_cgrp: cgroup that's being entered by the task
996  * @template: desired set of css pointers in css_set (pre-calculated)
997  *
998  * Returns true if "cset" matches "old_cset" except for the hierarchy
999  * which "new_cgrp" belongs to, for which it should match "new_cgrp".
1000  */
1001 static bool compare_css_sets(struct css_set *cset,
1002                              struct css_set *old_cset,
1003                              struct cgroup *new_cgrp,
1004                              struct cgroup_subsys_state *template[])
1005 {
1006         struct cgroup *new_dfl_cgrp;
1007         struct list_head *l1, *l2;
1008 
1009         /*
1010          * On the default hierarchy, there can be csets which are
1011          * associated with the same set of cgroups but different csses.
1012          * Let's first ensure that csses match.
1013          */
1014         if (memcmp(template, cset->subsys, sizeof(cset->subsys)))
1015                 return false;
1016 
1017 
1018         /* @cset's domain should match the default cgroup's */
1019         if (cgroup_on_dfl(new_cgrp))
1020                 new_dfl_cgrp = new_cgrp;
1021         else
1022                 new_dfl_cgrp = old_cset->dfl_cgrp;
1023 
1024         if (new_dfl_cgrp->dom_cgrp != cset->dom_cset->dfl_cgrp)
1025                 return false;
1026 
1027         /*
1028          * Compare cgroup pointers in order to distinguish between
1029          * different cgroups in hierarchies.  As different cgroups may
1030          * share the same effective css, this comparison is always
1031          * necessary.
1032          */
1033         l1 = &cset->cgrp_links;
1034         l2 = &old_cset->cgrp_links;
1035         while (1) {
1036                 struct cgrp_cset_link *link1, *link2;
1037                 struct cgroup *cgrp1, *cgrp2;
1038 
1039                 l1 = l1->next;
1040                 l2 = l2->next;
1041                 /* See if we reached the end - both lists are equal length. */
1042                 if (l1 == &cset->cgrp_links) {
1043                         BUG_ON(l2 != &old_cset->cgrp_links);
1044                         break;
1045                 } else {
1046                         BUG_ON(l2 == &old_cset->cgrp_links);
1047                 }
1048                 /* Locate the cgroups associated with these links. */
1049                 link1 = list_entry(l1, struct cgrp_cset_link, cgrp_link);
1050                 link2 = list_entry(l2, struct cgrp_cset_link, cgrp_link);
1051                 cgrp1 = link1->cgrp;
1052                 cgrp2 = link2->cgrp;
1053                 /* Hierarchies should be linked in the same order. */
1054                 BUG_ON(cgrp1->root != cgrp2->root);
1055 
1056                 /*
1057                  * If this hierarchy is the hierarchy of the cgroup
1058                  * that's changing, then we need to check that this
1059                  * css_set points to the new cgroup; if it's any other
1060                  * hierarchy, then this css_set should point to the
1061                  * same cgroup as the old css_set.
1062                  */
1063                 if (cgrp1->root == new_cgrp->root) {
1064                         if (cgrp1 != new_cgrp)
1065                                 return false;
1066                 } else {
1067                         if (cgrp1 != cgrp2)
1068                                 return false;
1069                 }
1070         }
1071         return true;
1072 }
1073 
1074 /**
1075  * find_existing_css_set - init css array and find the matching css_set
1076  * @old_cset: the css_set that we're using before the cgroup transition
1077  * @cgrp: the cgroup that we're moving into
1078  * @template: out param for the new set of csses, should be clear on entry
1079  */
1080 static struct css_set *find_existing_css_set(struct css_set *old_cset,
1081                                         struct cgroup *cgrp,
1082                                         struct cgroup_subsys_state *template[])
1083 {
1084         struct cgroup_root *root = cgrp->root;
1085         struct cgroup_subsys *ss;
1086         struct css_set *cset;
1087         unsigned long key;
1088         int i;
1089 
1090         /*
1091          * Build the set of subsystem state objects that we want to see in the
1092          * new css_set. While subsystems can change globally, the entries here
1093          * won't change, so no need for locking.
1094          */
1095         for_each_subsys(ss, i) {
1096                 if (root->subsys_mask & (1UL << i)) {
1097                         /*
1098                          * @ss is in this hierarchy, so we want the
1099                          * effective css from @cgrp.
1100                          */
1101                         template[i] = cgroup_e_css_by_mask(cgrp, ss);
1102                 } else {
1103                         /*
1104                          * @ss is not in this hierarchy, so we don't want
1105                          * to change the css.
1106                          */
1107                         template[i] = old_cset->subsys[i];
1108                 }
1109         }
1110 
1111         key = css_set_hash(template);
1112         hash_for_each_possible(css_set_table, cset, hlist, key) {
1113                 if (!compare_css_sets(cset, old_cset, cgrp, template))
1114                         continue;
1115 
1116                 /* This css_set matches what we need */
1117                 return cset;
1118         }
1119 
1120         /* No existing cgroup group matched */
1121         return NULL;
1122 }
1123 
1124 static void free_cgrp_cset_links(struct list_head *links_to_free)
1125 {
1126         struct cgrp_cset_link *link, *tmp_link;
1127 
1128         list_for_each_entry_safe(link, tmp_link, links_to_free, cset_link) {
1129                 list_del(&link->cset_link);
1130                 kfree(link);
1131         }
1132 }
1133 
1134 /**
1135  * allocate_cgrp_cset_links - allocate cgrp_cset_links
1136  * @count: the number of links to allocate
1137  * @tmp_links: list_head the allocated links are put on
1138  *
1139  * Allocate @count cgrp_cset_link structures and chain them on @tmp_links
1140  * through ->cset_link.  Returns 0 on success or -errno.
1141  */
1142 static int allocate_cgrp_cset_links(int count, struct list_head *tmp_links)
1143 {
1144         struct cgrp_cset_link *link;
1145         int i;
1146 
1147         INIT_LIST_HEAD(tmp_links);
1148 
1149         for (i = 0; i < count; i++) {
1150                 link = kzalloc(sizeof(*link), GFP_KERNEL);
1151                 if (!link) {
1152                         free_cgrp_cset_links(tmp_links);
1153                         return -ENOMEM;
1154                 }
1155                 list_add(&link->cset_link, tmp_links);
1156         }
1157         return 0;
1158 }
1159 
1160 /**
1161  * link_css_set - a helper function to link a css_set to a cgroup
1162  * @tmp_links: cgrp_cset_link objects allocated by allocate_cgrp_cset_links()
1163  * @cset: the css_set to be linked
1164  * @cgrp: the destination cgroup
1165  */
1166 static void link_css_set(struct list_head *tmp_links, struct css_set *cset,
1167                          struct cgroup *cgrp)
1168 {
1169         struct cgrp_cset_link *link;
1170 
1171         BUG_ON(list_empty(tmp_links));
1172 
1173         if (cgroup_on_dfl(cgrp))
1174                 cset->dfl_cgrp = cgrp;
1175 
1176         link = list_first_entry(tmp_links, struct cgrp_cset_link, cset_link);
1177         link->cset = cset;
1178         link->cgrp = cgrp;
1179 
1180         /*
1181          * Always add links to the tail of the lists so that the lists are
1182          * in chronological order.
1183          */
1184         list_move_tail(&link->cset_link, &cgrp->cset_links);
1185         list_add_tail(&link->cgrp_link, &cset->cgrp_links);
1186 
1187         if (cgroup_parent(cgrp))
1188                 cgroup_get_live(cgrp);
1189 }
1190 
1191 /**
1192  * find_css_set - return a new css_set with one cgroup updated
1193  * @old_cset: the baseline css_set
1194  * @cgrp: the cgroup to be updated
1195  *
1196  * Return a new css_set that's equivalent to @old_cset, but with @cgrp
1197  * substituted into the appropriate hierarchy.
1198  */
1199 static struct css_set *find_css_set(struct css_set *old_cset,
1200                                     struct cgroup *cgrp)
1201 {
1202         struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT] = { };
1203         struct css_set *cset;
1204         struct list_head tmp_links;
1205         struct cgrp_cset_link *link;
1206         struct cgroup_subsys *ss;
1207         unsigned long key;
1208         int ssid;
1209 
1210         lockdep_assert_held(&cgroup_mutex);
1211 
1212         /* First see if we already have a cgroup group that matches
1213          * the desired set */
1214         spin_lock_irq(&css_set_lock);
1215         cset = find_existing_css_set(old_cset, cgrp, template);
1216         if (cset)
1217                 get_css_set(cset);
1218         spin_unlock_irq(&css_set_lock);
1219 
1220         if (cset)
1221                 return cset;
1222 
1223         cset = kzalloc(sizeof(*cset), GFP_KERNEL);
1224         if (!cset)
1225                 return NULL;
1226 
1227         /* Allocate all the cgrp_cset_link objects that we'll need */
1228         if (allocate_cgrp_cset_links(cgroup_root_count, &tmp_links) < 0) {
1229                 kfree(cset);
1230                 return NULL;
1231         }
1232 
1233         refcount_set(&cset->refcount, 1);
1234         cset->dom_cset = cset;
1235         INIT_LIST_HEAD(&cset->tasks);
1236         INIT_LIST_HEAD(&cset->mg_tasks);
1237         INIT_LIST_HEAD(&cset->dying_tasks);
1238         INIT_LIST_HEAD(&cset->task_iters);
1239         INIT_LIST_HEAD(&cset->threaded_csets);
1240         INIT_HLIST_NODE(&cset->hlist);
1241         INIT_LIST_HEAD(&cset->cgrp_links);
1242         INIT_LIST_HEAD(&cset->mg_preload_node);
1243         INIT_LIST_HEAD(&cset->mg_node);
1244 
1245         /* Copy the set of subsystem state objects generated in
1246          * find_existing_css_set() */
1247         memcpy(cset->subsys, template, sizeof(cset->subsys));
1248 
1249         spin_lock_irq(&css_set_lock);
1250         /* Add reference counts and links from the new css_set. */
1251         list_for_each_entry(link, &old_cset->cgrp_links, cgrp_link) {
1252                 struct cgroup *c = link->cgrp;
1253 
1254                 if (c->root == cgrp->root)
1255                         c = cgrp;
1256                 link_css_set(&tmp_links, cset, c);
1257         }
1258 
1259         BUG_ON(!list_empty(&tmp_links));
1260 
1261         css_set_count++;
1262 
1263         /* Add @cset to the hash table */
1264         key = css_set_hash(cset->subsys);
1265         hash_add(css_set_table, &cset->hlist, key);
1266 
1267         for_each_subsys(ss, ssid) {
1268                 struct cgroup_subsys_state *css = cset->subsys[ssid];
1269 
1270                 list_add_tail(&cset->e_cset_node[ssid],
1271                               &css->cgroup->e_csets[ssid]);
1272                 css_get(css);
1273         }
1274 
1275         spin_unlock_irq(&css_set_lock);
1276 
1277         /*
1278          * If @cset should be threaded, look up the matching dom_cset and
1279          * link them up.  We first fully initialize @cset then look for the
1280          * dom_cset.  It's simpler this way and safe as @cset is guaranteed
1281          * to stay empty until we return.
1282          */
1283         if (cgroup_is_threaded(cset->dfl_cgrp)) {
1284                 struct css_set *dcset;
1285 
1286                 dcset = find_css_set(cset, cset->dfl_cgrp->dom_cgrp);
1287                 if (!dcset) {
1288                         put_css_set(cset);
1289                         return NULL;
1290                 }
1291 
1292                 spin_lock_irq(&css_set_lock);
1293                 cset->dom_cset = dcset;
1294                 list_add_tail(&cset->threaded_csets_node,
1295                               &dcset->threaded_csets);
1296                 spin_unlock_irq(&css_set_lock);
1297         }
1298 
1299         return cset;
1300 }
1301 
1302 struct cgroup_root *cgroup_root_from_kf(struct kernfs_root *kf_root)
1303 {
1304         struct cgroup *root_cgrp = kf_root->kn->priv;
1305 
1306         return root_cgrp->root;
1307 }
1308 
1309 static int cgroup_init_root_id(struct cgroup_root *root)
1310 {
1311         int id;
1312 
1313         lockdep_assert_held(&cgroup_mutex);
1314 
1315         id = idr_alloc_cyclic(&cgroup_hierarchy_idr, root, 0, 0, GFP_KERNEL);
1316         if (id < 0)
1317                 return id;
1318 
1319         root->hierarchy_id = id;
1320         return 0;
1321 }
1322 
1323 static void cgroup_exit_root_id(struct cgroup_root *root)
1324 {
1325         lockdep_assert_held(&cgroup_mutex);
1326 
1327         idr_remove(&cgroup_hierarchy_idr, root->hierarchy_id);
1328 }
1329 
1330 void cgroup_free_root(struct cgroup_root *root)
1331 {
1332         kfree(root);
1333 }
1334 
1335 static void cgroup_destroy_root(struct cgroup_root *root)
1336 {
1337         struct cgroup *cgrp = &root->cgrp;
1338         struct cgrp_cset_link *link, *tmp_link;
1339 
1340         trace_cgroup_destroy_root(root);
1341 
1342         cgroup_lock_and_drain_offline(&cgrp_dfl_root.cgrp);
1343 
1344         BUG_ON(atomic_read(&root->nr_cgrps));
1345         BUG_ON(!list_empty(&cgrp->self.children));
1346 
1347         /* Rebind all subsystems back to the default hierarchy */
1348         WARN_ON(rebind_subsystems(&cgrp_dfl_root, root->subsys_mask));
1349 
1350         /*
1351          * Release all the links from cset_links to this hierarchy's
1352          * root cgroup
1353          */
1354         spin_lock_irq(&css_set_lock);
1355 
1356         list_for_each_entry_safe(link, tmp_link, &cgrp->cset_links, cset_link) {
1357                 list_del(&link->cset_link);
1358                 list_del(&link->cgrp_link);
1359                 kfree(link);
1360         }
1361 
1362         spin_unlock_irq(&css_set_lock);
1363 
1364         if (!list_empty(&root->root_list)) {
1365                 list_del(&root->root_list);
1366                 cgroup_root_count--;
1367         }
1368 
1369         cgroup_exit_root_id(root);
1370 
1371         mutex_unlock(&cgroup_mutex);
1372 
1373         cgroup_rstat_exit(cgrp);
1374         kernfs_destroy_root(root->kf_root);
1375         cgroup_free_root(root);
1376 }
1377 
1378 /*
1379  * look up cgroup associated with current task's cgroup namespace on the
1380  * specified hierarchy
1381  */
1382 static struct cgroup *
1383 current_cgns_cgroup_from_root(struct cgroup_root *root)
1384 {
1385         struct cgroup *res = NULL;
1386         struct css_set *cset;
1387 
1388         lockdep_assert_held(&css_set_lock);
1389 
1390         rcu_read_lock();
1391 
1392         cset = current->nsproxy->cgroup_ns->root_cset;
1393         if (cset == &init_css_set) {
1394                 res = &root->cgrp;
1395         } else if (root == &cgrp_dfl_root) {
1396                 res = cset->dfl_cgrp;
1397         } else {
1398                 struct cgrp_cset_link *link;
1399 
1400                 list_for_each_entry(link, &cset->cgrp_links, cgrp_link) {
1401                         struct cgroup *c = link->cgrp;
1402 
1403                         if (c->root == root) {
1404                                 res = c;
1405                                 break;
1406                         }
1407                 }
1408         }
1409         rcu_read_unlock();
1410 
1411         BUG_ON(!res);
1412         return res;
1413 }
1414 
1415 /* look up cgroup associated with given css_set on the specified hierarchy */
1416 static struct cgroup *cset_cgroup_from_root(struct css_set *cset,
1417                                             struct cgroup_root *root)
1418 {
1419         struct cgroup *res = NULL;
1420 
1421         lockdep_assert_held(&cgroup_mutex);
1422         lockdep_assert_held(&css_set_lock);
1423 
1424         if (cset == &init_css_set) {
1425                 res = &root->cgrp;
1426         } else if (root == &cgrp_dfl_root) {
1427                 res = cset->dfl_cgrp;
1428         } else {
1429                 struct cgrp_cset_link *link;
1430 
1431                 list_for_each_entry(link, &cset->cgrp_links, cgrp_link) {
1432                         struct cgroup *c = link->cgrp;
1433 
1434                         if (c->root == root) {
1435                                 res = c;
1436                                 break;
1437                         }
1438                 }
1439         }
1440 
1441         BUG_ON(!res);
1442         return res;
1443 }
1444 
1445 /*
1446  * Return the cgroup for "task" from the given hierarchy. Must be
1447  * called with cgroup_mutex and css_set_lock held.
1448  */
1449 struct cgroup *task_cgroup_from_root(struct task_struct *task,
1450                                      struct cgroup_root *root)
1451 {
1452         /*
1453          * No need to lock the task - since we hold css_set_lock the
1454          * task can't change groups.
1455          */
1456         return cset_cgroup_from_root(task_css_set(task), root);
1457 }
1458 
1459 /*
1460  * A task must hold cgroup_mutex to modify cgroups.
1461  *
1462  * Any task can increment and decrement the count field without lock.
1463  * So in general, code holding cgroup_mutex can't rely on the count
1464  * field not changing.  However, if the count goes to zero, then only
1465  * cgroup_attach_task() can increment it again.  Because a count of zero
1466  * means that no tasks are currently attached, therefore there is no
1467  * way a task attached to that cgroup can fork (the other way to
1468  * increment the count).  So code holding cgroup_mutex can safely
1469  * assume that if the count is zero, it will stay zero. Similarly, if
1470  * a task holds cgroup_mutex on a cgroup with zero count, it
1471  * knows that the cgroup won't be removed, as cgroup_rmdir()
1472  * needs that mutex.
1473  *
1474  * A cgroup can only be deleted if both its 'count' of using tasks
1475  * is zero, and its list of 'children' cgroups is empty.  Since all
1476  * tasks in the system use _some_ cgroup, and since there is always at
1477  * least one task in the system (init, pid == 1), therefore, root cgroup
1478  * always has either children cgroups and/or using tasks.  So we don't
1479  * need a special hack to ensure that root cgroup cannot be deleted.
1480  *
1481  * P.S.  One more locking exception.  RCU is used to guard the
1482  * update of a tasks cgroup pointer by cgroup_attach_task()
1483  */
1484 
1485 static struct kernfs_syscall_ops cgroup_kf_syscall_ops;
1486 
1487 static char *cgroup_file_name(struct cgroup *cgrp, const struct cftype *cft,
1488                               char *buf)
1489 {
1490         struct cgroup_subsys *ss = cft->ss;
1491 
1492         if (cft->ss && !(cft->flags & CFTYPE_NO_PREFIX) &&
1493             !(cgrp->root->flags & CGRP_ROOT_NOPREFIX)) {
1494                 const char *dbg = (cft->flags & CFTYPE_DEBUG) ? ".__DEBUG__." : "";
1495 
1496                 snprintf(buf, CGROUP_FILE_NAME_MAX, "%s%s.%s",
1497                          dbg, cgroup_on_dfl(cgrp) ? ss->name : ss->legacy_name,
1498                          cft->name);
1499         } else {
1500                 strscpy(buf, cft->name, CGROUP_FILE_NAME_MAX);
1501         }
1502         return buf;
1503 }
1504 
1505 /**
1506  * cgroup_file_mode - deduce file mode of a control file
1507  * @cft: the control file in question
1508  *
1509  * S_IRUGO for read, S_IWUSR for write.
1510  */
1511 static umode_t cgroup_file_mode(const struct cftype *cft)
1512 {
1513         umode_t mode = 0;
1514 
1515         if (cft->read_u64 || cft->read_s64 || cft->seq_show)
1516                 mode |= S_IRUGO;
1517 
1518         if (cft->write_u64 || cft->write_s64 || cft->write) {
1519                 if (cft->flags & CFTYPE_WORLD_WRITABLE)
1520                         mode |= S_IWUGO;
1521                 else
1522                         mode |= S_IWUSR;
1523         }
1524 
1525         return mode;
1526 }
1527 
1528 /**
1529  * cgroup_calc_subtree_ss_mask - calculate subtree_ss_mask
1530  * @subtree_control: the new subtree_control mask to consider
1531  * @this_ss_mask: available subsystems
1532  *
1533  * On the default hierarchy, a subsystem may request other subsystems to be
1534  * enabled together through its ->depends_on mask.  In such cases, more
1535  * subsystems than specified in "cgroup.subtree_control" may be enabled.
1536  *
1537  * This function calculates which subsystems need to be enabled if
1538  * @subtree_control is to be applied while restricted to @this_ss_mask.
1539  */
1540 static u16 cgroup_calc_subtree_ss_mask(u16 subtree_control, u16 this_ss_mask)
1541 {
1542         u16 cur_ss_mask = subtree_control;
1543         struct cgroup_subsys *ss;
1544         int ssid;
1545 
1546         lockdep_assert_held(&cgroup_mutex);
1547 
1548         cur_ss_mask |= cgrp_dfl_implicit_ss_mask;
1549 
1550         while (true) {
1551                 u16 new_ss_mask = cur_ss_mask;
1552 
1553                 do_each_subsys_mask(ss, ssid, cur_ss_mask) {
1554                         new_ss_mask |= ss->depends_on;
1555                 } while_each_subsys_mask();
1556 
1557                 /*
1558                  * Mask out subsystems which aren't available.  This can
1559                  * happen only if some depended-upon subsystems were bound
1560                  * to non-default hierarchies.
1561                  */
1562                 new_ss_mask &= this_ss_mask;
1563 
1564                 if (new_ss_mask == cur_ss_mask)
1565                         break;
1566                 cur_ss_mask = new_ss_mask;
1567         }
1568 
1569         return cur_ss_mask;
1570 }
1571 
1572 /**
1573  * cgroup_kn_unlock - unlocking helper for cgroup kernfs methods
1574  * @kn: the kernfs_node being serviced
1575  *
1576  * This helper undoes cgroup_kn_lock_live() and should be invoked before
1577  * the method finishes if locking succeeded.  Note that once this function
1578  * returns the cgroup returned by cgroup_kn_lock_live() may become
1579  * inaccessible any time.  If the caller intends to continue to access the
1580  * cgroup, it should pin it before invoking this function.
1581  */
1582 void cgroup_kn_unlock(struct kernfs_node *kn)
1583 {
1584         struct cgroup *cgrp;
1585 
1586         if (kernfs_type(kn) == KERNFS_DIR)
1587                 cgrp = kn->priv;
1588         else
1589                 cgrp = kn->parent->priv;
1590 
1591         mutex_unlock(&cgroup_mutex);
1592 
1593         kernfs_unbreak_active_protection(kn);
1594         cgroup_put(cgrp);
1595 }
1596 
1597 /**
1598  * cgroup_kn_lock_live - locking helper for cgroup kernfs methods
1599  * @kn: the kernfs_node being serviced
1600  * @drain_offline: perform offline draining on the cgroup
1601  *
1602  * This helper is to be used by a cgroup kernfs method currently servicing
1603  * @kn.  It breaks the active protection, performs cgroup locking and
1604  * verifies that the associated cgroup is alive.  Returns the cgroup if
1605  * alive; otherwise, %NULL.  A successful return should be undone by a
1606  * matching cgroup_kn_unlock() invocation.  If @drain_offline is %true, the
1607  * cgroup is drained of offlining csses before return.
1608  *
1609  * Any cgroup kernfs method implementation which requires locking the
1610  * associated cgroup should use this helper.  It avoids nesting cgroup
1611  * locking under kernfs active protection and allows all kernfs operations
1612  * including self-removal.
1613  */
1614 struct cgroup *cgroup_kn_lock_live(struct kernfs_node *kn, bool drain_offline)
1615 {
1616         struct cgroup *cgrp;
1617 
1618         if (kernfs_type(kn) == KERNFS_DIR)
1619                 cgrp = kn->priv;
1620         else
1621                 cgrp = kn->parent->priv;
1622 
1623         /*
1624          * We're gonna grab cgroup_mutex which nests outside kernfs
1625          * active_ref.  cgroup liveliness check alone provides enough
1626          * protection against removal.  Ensure @cgrp stays accessible and
1627          * break the active_ref protection.
1628          */
1629         if (!cgroup_tryget(cgrp))
1630                 return NULL;
1631         kernfs_break_active_protection(kn);
1632 
1633         if (drain_offline)
1634                 cgroup_lock_and_drain_offline(cgrp);
1635         else
1636                 mutex_lock(&cgroup_mutex);
1637 
1638         if (!cgroup_is_dead(cgrp))
1639                 return cgrp;
1640 
1641         cgroup_kn_unlock(kn);
1642         return NULL;
1643 }
1644 
1645 static void cgroup_rm_file(struct cgroup *cgrp, const struct cftype *cft)
1646 {
1647         char name[CGROUP_FILE_NAME_MAX];
1648 
1649         lockdep_assert_held(&cgroup_mutex);
1650 
1651         if (cft->file_offset) {
1652                 struct cgroup_subsys_state *css = cgroup_css(cgrp, cft->ss);
1653                 struct cgroup_file *cfile = (void *)css + cft->file_offset;
1654 
1655                 spin_lock_irq(&cgroup_file_kn_lock);
1656                 cfile->kn = NULL;
1657                 spin_unlock_irq(&cgroup_file_kn_lock);
1658 
1659                 del_timer_sync(&cfile->notify_timer);
1660         }
1661 
1662         kernfs_remove_by_name(cgrp->kn, cgroup_file_name(cgrp, cft, name));
1663 }
1664 
1665 /**
1666  * css_clear_dir - remove subsys files in a cgroup directory
1667  * @css: target css
1668  */
1669 static void css_clear_dir(struct cgroup_subsys_state *css)
1670 {
1671         struct cgroup *cgrp = css->cgroup;
1672         struct cftype *cfts;
1673 
1674         if (!(css->flags & CSS_VISIBLE))
1675                 return;
1676 
1677         css->flags &= ~CSS_VISIBLE;
1678 
1679         if (!css->ss) {
1680                 if (cgroup_on_dfl(cgrp))
1681                         cfts = cgroup_base_files;
1682                 else
1683                         cfts = cgroup1_base_files;
1684 
1685                 cgroup_addrm_files(css, cgrp, cfts, false);
1686         } else {
1687                 list_for_each_entry(cfts, &css->ss->cfts, node)
1688                         cgroup_addrm_files(css, cgrp, cfts, false);
1689         }
1690 }
1691 
1692 /**
1693  * css_populate_dir - create subsys files in a cgroup directory
1694  * @css: target css
1695  *
1696  * On failure, no file is added.
1697  */
1698 static int css_populate_dir(struct cgroup_subsys_state *css)
1699 {
1700         struct cgroup *cgrp = css->cgroup;
1701         struct cftype *cfts, *failed_cfts;
1702         int ret;
1703 
1704         if ((css->flags & CSS_VISIBLE) || !cgrp->kn)
1705                 return 0;
1706 
1707         if (!css->ss) {
1708                 if (cgroup_on_dfl(cgrp))
1709                         cfts = cgroup_base_files;
1710                 else
1711                         cfts = cgroup1_base_files;
1712 
1713                 ret = cgroup_addrm_files(&cgrp->self, cgrp, cfts, true);
1714                 if (ret < 0)
1715                         return ret;
1716         } else {
1717                 list_for_each_entry(cfts, &css->ss->cfts, node) {
1718                         ret = cgroup_addrm_files(css, cgrp, cfts, true);
1719                         if (ret < 0) {
1720                                 failed_cfts = cfts;
1721                                 goto err;
1722                         }
1723                 }
1724         }
1725 
1726         css->flags |= CSS_VISIBLE;
1727 
1728         return 0;
1729 err:
1730         list_for_each_entry(cfts, &css->ss->cfts, node) {
1731                 if (cfts == failed_cfts)
1732                         break;
1733                 cgroup_addrm_files(css, cgrp, cfts, false);
1734         }
1735         return ret;
1736 }
1737 
1738 int rebind_subsystems(struct cgroup_root *dst_root, u16 ss_mask)
1739 {
1740         struct cgroup *dcgrp = &dst_root->cgrp;
1741         struct cgroup_subsys *ss;
1742         int ssid, i, ret;
1743         u16 dfl_disable_ss_mask = 0;
1744 
1745         lockdep_assert_held(&cgroup_mutex);
1746 
1747         do_each_subsys_mask(ss, ssid, ss_mask) {
1748                 /*
1749                  * If @ss has non-root csses attached to it, can't move.
1750                  * If @ss is an implicit controller, it is exempt from this
1751                  * rule and can be stolen.
1752                  */
1753                 if (css_next_child(NULL, cgroup_css(&ss->root->cgrp, ss)) &&
1754                     !ss->implicit_on_dfl)
1755                         return -EBUSY;
1756 
1757                 /* can't move between two non-dummy roots either */
1758                 if (ss->root != &cgrp_dfl_root && dst_root != &cgrp_dfl_root)
1759                         return -EBUSY;
1760 
1761                 /*
1762                  * Collect ssid's that need to be disabled from default
1763                  * hierarchy.
1764                  */
1765                 if (ss->root == &cgrp_dfl_root)
1766                         dfl_disable_ss_mask |= 1 << ssid;
1767 
1768         } while_each_subsys_mask();
1769 
1770         if (dfl_disable_ss_mask) {
1771                 struct cgroup *scgrp = &cgrp_dfl_root.cgrp;
1772 
1773                 /*
1774                  * Controllers from default hierarchy that need to be rebound
1775                  * are all disabled together in one go.
1776                  */
1777                 cgrp_dfl_root.subsys_mask &= ~dfl_disable_ss_mask;
1778                 WARN_ON(cgroup_apply_control(scgrp));
1779                 cgroup_finalize_control(scgrp, 0);
1780         }
1781 
1782         do_each_subsys_mask(ss, ssid, ss_mask) {
1783                 struct cgroup_root *src_root = ss->root;
1784                 struct cgroup *scgrp = &src_root->cgrp;
1785                 struct cgroup_subsys_state *css = cgroup_css(scgrp, ss);
1786                 struct css_set *cset;
1787 
1788                 WARN_ON(!css || cgroup_css(dcgrp, ss));
1789 
1790                 if (src_root != &cgrp_dfl_root) {
1791                         /* disable from the source */
1792                         src_root->subsys_mask &= ~(1 << ssid);
1793                         WARN_ON(cgroup_apply_control(scgrp));
1794                         cgroup_finalize_control(scgrp, 0);
1795                 }
1796 
1797                 /* rebind */
1798                 RCU_INIT_POINTER(scgrp->subsys[ssid], NULL);
1799                 rcu_assign_pointer(dcgrp->subsys[ssid], css);
1800                 ss->root = dst_root;
1801                 css->cgroup = dcgrp;
1802 
1803                 spin_lock_irq(&css_set_lock);
1804                 hash_for_each(css_set_table, i, cset, hlist)
1805                         list_move_tail(&cset->e_cset_node[ss->id],
1806                                        &dcgrp->e_csets[ss->id]);
1807                 spin_unlock_irq(&css_set_lock);
1808 
1809                 if (ss->css_rstat_flush) {
1810                         list_del_rcu(&css->rstat_css_node);
1811                         list_add_rcu(&css->rstat_css_node,
1812                                      &dcgrp->rstat_css_list);
1813                 }
1814 
1815                 /* default hierarchy doesn't enable controllers by default */
1816                 dst_root->subsys_mask |= 1 << ssid;
1817                 if (dst_root == &cgrp_dfl_root) {
1818                         static_branch_enable(cgroup_subsys_on_dfl_key[ssid]);
1819                 } else {
1820                         dcgrp->subtree_control |= 1 << ssid;
1821                         static_branch_disable(cgroup_subsys_on_dfl_key[ssid]);
1822                 }
1823 
1824                 ret = cgroup_apply_control(dcgrp);
1825                 if (ret)
1826                         pr_warn("partial failure to rebind %s controller (err=%d)\n",
1827                                 ss->name, ret);
1828 
1829                 if (ss->bind)
1830                         ss->bind(css);
1831         } while_each_subsys_mask();
1832 
1833         kernfs_activate(dcgrp->kn);
1834         return 0;
1835 }
1836 
1837 int cgroup_show_path(struct seq_file *sf, struct kernfs_node *kf_node,
1838                      struct kernfs_root *kf_root)
1839 {
1840         int len = 0;
1841         char *buf = NULL;
1842         struct cgroup_root *kf_cgroot = cgroup_root_from_kf(kf_root);
1843         struct cgroup *ns_cgroup;
1844 
1845         buf = kmalloc(PATH_MAX, GFP_KERNEL);
1846         if (!buf)
1847                 return -ENOMEM;
1848 
1849         spin_lock_irq(&css_set_lock);
1850         ns_cgroup = current_cgns_cgroup_from_root(kf_cgroot);
1851         len = kernfs_path_from_node(kf_node, ns_cgroup->kn, buf, PATH_MAX);
1852         spin_unlock_irq(&css_set_lock);
1853 
1854         if (len >= PATH_MAX)
1855                 len = -ERANGE;
1856         else if (len > 0) {
1857                 seq_escape(sf, buf, " \t\n\\");
1858                 len = 0;
1859         }
1860         kfree(buf);
1861         return len;
1862 }
1863 
1864 enum cgroup2_param {
1865         Opt_nsdelegate,
1866         Opt_memory_localevents,
1867         Opt_memory_recursiveprot,
1868         nr__cgroup2_params
1869 };
1870 
1871 static const struct fs_parameter_spec cgroup2_fs_parameters[] = {
1872         fsparam_flag("nsdelegate",              Opt_nsdelegate),
1873         fsparam_flag("memory_localevents",      Opt_memory_localevents),
1874         fsparam_flag("memory_recursiveprot",    Opt_memory_recursiveprot),
1875         {}
1876 };
1877 
1878 static int cgroup2_parse_param(struct fs_context *fc, struct fs_parameter *param)
1879 {
1880         struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
1881         struct fs_parse_result result;
1882         int opt;
1883 
1884         opt = fs_parse(fc, cgroup2_fs_parameters, param, &result);
1885         if (opt < 0)
1886                 return opt;
1887 
1888         switch (opt) {
1889         case Opt_nsdelegate:
1890                 ctx->flags |= CGRP_ROOT_NS_DELEGATE;
1891                 return 0;
1892         case Opt_memory_localevents:
1893                 ctx->flags |= CGRP_ROOT_MEMORY_LOCAL_EVENTS;
1894                 return 0;
1895         case Opt_memory_recursiveprot:
1896                 ctx->flags |= CGRP_ROOT_MEMORY_RECURSIVE_PROT;
1897                 return 0;
1898         }
1899         return -EINVAL;
1900 }
1901 
1902 static void apply_cgroup_root_flags(unsigned int root_flags)
1903 {
1904         if (current->nsproxy->cgroup_ns == &init_cgroup_ns) {
1905                 if (root_flags & CGRP_ROOT_NS_DELEGATE)
1906                         cgrp_dfl_root.flags |= CGRP_ROOT_NS_DELEGATE;
1907                 else
1908                         cgrp_dfl_root.flags &= ~CGRP_ROOT_NS_DELEGATE;
1909 
1910                 if (root_flags & CGRP_ROOT_MEMORY_LOCAL_EVENTS)
1911                         cgrp_dfl_root.flags |= CGRP_ROOT_MEMORY_LOCAL_EVENTS;
1912                 else
1913                         cgrp_dfl_root.flags &= ~CGRP_ROOT_MEMORY_LOCAL_EVENTS;
1914 
1915                 if (root_flags & CGRP_ROOT_MEMORY_RECURSIVE_PROT)
1916                         cgrp_dfl_root.flags |= CGRP_ROOT_MEMORY_RECURSIVE_PROT;
1917                 else
1918                         cgrp_dfl_root.flags &= ~CGRP_ROOT_MEMORY_RECURSIVE_PROT;
1919         }
1920 }
1921 
1922 static int cgroup_show_options(struct seq_file *seq, struct kernfs_root *kf_root)
1923 {
1924         if (cgrp_dfl_root.flags & CGRP_ROOT_NS_DELEGATE)
1925                 seq_puts(seq, ",nsdelegate");
1926         if (cgrp_dfl_root.flags & CGRP_ROOT_MEMORY_LOCAL_EVENTS)
1927                 seq_puts(seq, ",memory_localevents");
1928         if (cgrp_dfl_root.flags & CGRP_ROOT_MEMORY_RECURSIVE_PROT)
1929                 seq_puts(seq, ",memory_recursiveprot");
1930         return 0;
1931 }
1932 
1933 static int cgroup_reconfigure(struct fs_context *fc)
1934 {
1935         struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
1936 
1937         apply_cgroup_root_flags(ctx->flags);
1938         return 0;
1939 }
1940 
1941 static void init_cgroup_housekeeping(struct cgroup *cgrp)
1942 {
1943         struct cgroup_subsys *ss;
1944         int ssid;
1945 
1946         INIT_LIST_HEAD(&cgrp->self.sibling);
1947         INIT_LIST_HEAD(&cgrp->self.children);
1948         INIT_LIST_HEAD(&cgrp->cset_links);
1949         INIT_LIST_HEAD(&cgrp->pidlists);
1950         mutex_init(&cgrp->pidlist_mutex);
1951         cgrp->self.cgroup = cgrp;
1952         cgrp->self.flags |= CSS_ONLINE;
1953         cgrp->dom_cgrp = cgrp;
1954         cgrp->max_descendants = INT_MAX;
1955         cgrp->max_depth = INT_MAX;
1956         INIT_LIST_HEAD(&cgrp->rstat_css_list);
1957         prev_cputime_init(&cgrp->prev_cputime);
1958 
1959         for_each_subsys(ss, ssid)
1960                 INIT_LIST_HEAD(&cgrp->e_csets[ssid]);
1961 
1962         init_waitqueue_head(&cgrp->offline_waitq);
1963         INIT_WORK(&cgrp->release_agent_work, cgroup1_release_agent);
1964 }
1965 
1966 void init_cgroup_root(struct cgroup_fs_context *ctx)
1967 {
1968         struct cgroup_root *root = ctx->root;
1969         struct cgroup *cgrp = &root->cgrp;
1970 
1971         INIT_LIST_HEAD(&root->root_list);
1972         atomic_set(&root->nr_cgrps, 1);
1973         cgrp->root = root;
1974         init_cgroup_housekeeping(cgrp);
1975 
1976         root->flags = ctx->flags;
1977         if (ctx->release_agent)
1978                 strscpy(root->release_agent_path, ctx->release_agent, PATH_MAX);
1979         if (ctx->name)
1980                 strscpy(root->name, ctx->name, MAX_CGROUP_ROOT_NAMELEN);
1981         if (ctx->cpuset_clone_children)
1982                 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags);
1983 }
1984 
1985 int cgroup_setup_root(struct cgroup_root *root, u16 ss_mask)
1986 {
1987         LIST_HEAD(tmp_links);
1988         struct cgroup *root_cgrp = &root->cgrp;
1989         struct kernfs_syscall_ops *kf_sops;
1990         struct css_set *cset;
1991         int i, ret;
1992 
1993         lockdep_assert_held(&cgroup_mutex);
1994 
1995         ret = percpu_ref_init(&root_cgrp->self.refcnt, css_release,
1996                               0, GFP_KERNEL);
1997         if (ret)
1998                 goto out;
1999 
2000         /*
2001          * We're accessing css_set_count without locking css_set_lock here,
2002          * but that's OK - it can only be increased by someone holding
2003          * cgroup_lock, and that's us.  Later rebinding may disable
2004          * controllers on the default hierarchy and thus create new csets,
2005          * which can't be more than the existing ones.  Allocate 2x.
2006          */
2007         ret = allocate_cgrp_cset_links(2 * css_set_count, &tmp_links);
2008         if (ret)
2009                 goto cancel_ref;
2010 
2011         ret = cgroup_init_root_id(root);
2012         if (ret)
2013                 goto cancel_ref;
2014 
2015         kf_sops = root == &cgrp_dfl_root ?
2016                 &cgroup_kf_syscall_ops : &cgroup1_kf_syscall_ops;
2017 
2018         root->kf_root = kernfs_create_root(kf_sops,
2019                                            KERNFS_ROOT_CREATE_DEACTIVATED |
2020                                            KERNFS_ROOT_SUPPORT_EXPORTOP |
2021                                            KERNFS_ROOT_SUPPORT_USER_XATTR,
2022                                            root_cgrp);
2023         if (IS_ERR(root->kf_root)) {
2024                 ret = PTR_ERR(root->kf_root);
2025                 goto exit_root_id;
2026         }
2027         root_cgrp->kn = root->kf_root->kn;
2028         WARN_ON_ONCE(cgroup_ino(root_cgrp) != 1);
2029         root_cgrp->ancestor_ids[0] = cgroup_id(root_cgrp);
2030 
2031         ret = css_populate_dir(&root_cgrp->self);
2032         if (ret)
2033                 goto destroy_root;
2034 
2035         ret = cgroup_rstat_init(root_cgrp);
2036         if (ret)
2037                 goto destroy_root;
2038 
2039         ret = rebind_subsystems(root, ss_mask);
2040         if (ret)
2041                 goto exit_stats;
2042 
2043         ret = cgroup_bpf_inherit(root_cgrp);
2044         WARN_ON_ONCE(ret);
2045 
2046         trace_cgroup_setup_root(root);
2047 
2048         /*
2049          * There must be no failure case after here, since rebinding takes
2050          * care of subsystems' refcounts, which are explicitly dropped in
2051          * the failure exit path.
2052          */
2053         list_add(&root->root_list, &cgroup_roots);
2054         cgroup_root_count++;
2055 
2056         /*
2057          * Link the root cgroup in this hierarchy into all the css_set
2058          * objects.
2059          */
2060         spin_lock_irq(&css_set_lock);
2061         hash_for_each(css_set_table, i, cset, hlist) {
2062                 link_css_set(&tmp_links, cset, root_cgrp);
2063                 if (css_set_populated(cset))
2064                         cgroup_update_populated(root_cgrp, true);
2065         }
2066         spin_unlock_irq(&css_set_lock);
2067 
2068         BUG_ON(!list_empty(&root_cgrp->self.children));
2069         BUG_ON(atomic_read(&root->nr_cgrps) != 1);
2070 
2071         ret = 0;
2072         goto out;
2073 
2074 exit_stats:
2075         cgroup_rstat_exit(root_cgrp);
2076 destroy_root:
2077         kernfs_destroy_root(root->kf_root);
2078         root->kf_root = NULL;
2079 exit_root_id:
2080         cgroup_exit_root_id(root);
2081 cancel_ref:
2082         percpu_ref_exit(&root_cgrp->self.refcnt);
2083 out:
2084         free_cgrp_cset_links(&tmp_links);
2085         return ret;
2086 }
2087 
2088 int cgroup_do_get_tree(struct fs_context *fc)
2089 {
2090         struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
2091         int ret;
2092 
2093         ctx->kfc.root = ctx->root->kf_root;
2094         if (fc->fs_type == &cgroup2_fs_type)
2095                 ctx->kfc.magic = CGROUP2_SUPER_MAGIC;
2096         else
2097                 ctx->kfc.magic = CGROUP_SUPER_MAGIC;
2098         ret = kernfs_get_tree(fc);
2099 
2100         /*
2101          * In non-init cgroup namespace, instead of root cgroup's dentry,
2102          * we return the dentry corresponding to the cgroupns->root_cgrp.
2103          */
2104         if (!ret && ctx->ns != &init_cgroup_ns) {
2105                 struct dentry *nsdentry;
2106                 struct super_block *sb = fc->root->d_sb;
2107                 struct cgroup *cgrp;
2108 
2109                 mutex_lock(&cgroup_mutex);
2110                 spin_lock_irq(&css_set_lock);
2111 
2112                 cgrp = cset_cgroup_from_root(ctx->ns->root_cset, ctx->root);
2113 
2114                 spin_unlock_irq(&css_set_lock);
2115                 mutex_unlock(&cgroup_mutex);
2116 
2117                 nsdentry = kernfs_node_dentry(cgrp->kn, sb);
2118                 dput(fc->root);
2119                 if (IS_ERR(nsdentry)) {
2120                         deactivate_locked_super(sb);
2121                         ret = PTR_ERR(nsdentry);
2122                         nsdentry = NULL;
2123                 }
2124                 fc->root = nsdentry;
2125         }
2126 
2127         if (!ctx->kfc.new_sb_created)
2128                 cgroup_put(&ctx->root->cgrp);
2129 
2130         return ret;
2131 }
2132 
2133 /*
2134  * Destroy a cgroup filesystem context.
2135  */
2136 static void cgroup_fs_context_free(struct fs_context *fc)
2137 {
2138         struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
2139 
2140         kfree(ctx->name);
2141         kfree(ctx->release_agent);
2142         put_cgroup_ns(ctx->ns);
2143         kernfs_free_fs_context(fc);
2144         kfree(ctx);
2145 }
2146 
2147 static int cgroup_get_tree(struct fs_context *fc)
2148 {
2149         struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
2150         int ret;
2151 
2152         cgrp_dfl_visible = true;
2153         cgroup_get_live(&cgrp_dfl_root.cgrp);
2154         ctx->root = &cgrp_dfl_root;
2155 
2156         ret = cgroup_do_get_tree(fc);
2157         if (!ret)
2158                 apply_cgroup_root_flags(ctx->flags);
2159         return ret;
2160 }
2161 
2162 static const struct fs_context_operations cgroup_fs_context_ops = {
2163         .free           = cgroup_fs_context_free,
2164         .parse_param    = cgroup2_parse_param,
2165         .get_tree       = cgroup_get_tree,
2166         .reconfigure    = cgroup_reconfigure,
2167 };
2168 
2169 static const struct fs_context_operations cgroup1_fs_context_ops = {
2170         .free           = cgroup_fs_context_free,
2171         .parse_param    = cgroup1_parse_param,
2172         .get_tree       = cgroup1_get_tree,
2173         .reconfigure    = cgroup1_reconfigure,
2174 };
2175 
2176 /*
2177  * Initialise the cgroup filesystem creation/reconfiguration context.  Notably,
2178  * we select the namespace we're going to use.
2179  */
2180 static int cgroup_init_fs_context(struct fs_context *fc)
2181 {
2182         struct cgroup_fs_context *ctx;
2183 
2184         ctx = kzalloc(sizeof(struct cgroup_fs_context), GFP_KERNEL);
2185         if (!ctx)
2186                 return -ENOMEM;
2187 
2188         ctx->ns = current->nsproxy->cgroup_ns;
2189         get_cgroup_ns(ctx->ns);
2190         fc->fs_private = &ctx->kfc;
2191         if (fc->fs_type == &cgroup2_fs_type)
2192                 fc->ops = &cgroup_fs_context_ops;
2193         else
2194                 fc->ops = &cgroup1_fs_context_ops;
2195         put_user_ns(fc->user_ns);
2196         fc->user_ns = get_user_ns(ctx->ns->user_ns);
2197         fc->global = true;
2198         return 0;
2199 }
2200 
2201 static void cgroup_kill_sb(struct super_block *sb)
2202 {
2203         struct kernfs_root *kf_root = kernfs_root_from_sb(sb);
2204         struct cgroup_root *root = cgroup_root_from_kf(kf_root);
2205 
2206         /*
2207          * If @root doesn't have any children, start killing it.
2208          * This prevents new mounts by disabling percpu_ref_tryget_live().
2209          *
2210          * And don't kill the default root.
2211          */
2212         if (list_empty(&root->cgrp.self.children) && root != &cgrp_dfl_root &&
2213             !percpu_ref_is_dying(&root->cgrp.self.refcnt)) {
2214                 cgroup_bpf_offline(&root->cgrp);
2215                 percpu_ref_kill(&root->cgrp.self.refcnt);
2216         }
2217         cgroup_put(&root->cgrp);
2218         kernfs_kill_sb(sb);
2219 }
2220 
2221 struct file_system_type cgroup_fs_type = {
2222         .name                   = "cgroup",
2223         .init_fs_context        = cgroup_init_fs_context,
2224         .parameters             = cgroup1_fs_parameters,
2225         .kill_sb                = cgroup_kill_sb,
2226         .fs_flags               = FS_USERNS_MOUNT,
2227 };
2228 
2229 static struct file_system_type cgroup2_fs_type = {
2230         .name                   = "cgroup2",
2231         .init_fs_context        = cgroup_init_fs_context,
2232         .parameters             = cgroup2_fs_parameters,
2233         .kill_sb                = cgroup_kill_sb,
2234         .fs_flags               = FS_USERNS_MOUNT,
2235 };
2236 
2237 #ifdef CONFIG_CPUSETS
2238 static const struct fs_context_operations cpuset_fs_context_ops = {
2239         .get_tree       = cgroup1_get_tree,
2240         .free           = cgroup_fs_context_free,
2241 };
2242 
2243 /*
2244  * This is ugly, but preserves the userspace API for existing cpuset
2245  * users. If someone tries to mount the "cpuset" filesystem, we
2246  * silently switch it to mount "cgroup" instead
2247  */
2248 static int cpuset_init_fs_context(struct fs_context *fc)
2249 {
2250         char *agent = kstrdup("/sbin/cpuset_release_agent", GFP_USER);
2251         struct cgroup_fs_context *ctx;
2252         int err;
2253 
2254         err = cgroup_init_fs_context(fc);
2255         if (err) {
2256                 kfree(agent);
2257                 return err;
2258         }
2259 
2260         fc->ops = &cpuset_fs_context_ops;
2261 
2262         ctx = cgroup_fc2context(fc);
2263         ctx->subsys_mask = 1 << cpuset_cgrp_id;
2264         ctx->flags |= CGRP_ROOT_NOPREFIX;
2265         ctx->release_agent = agent;
2266 
2267         get_filesystem(&cgroup_fs_type);
2268         put_filesystem(fc->fs_type);
2269         fc->fs_type = &cgroup_fs_type;
2270 
2271         return 0;
2272 }
2273 
2274 static struct file_system_type cpuset_fs_type = {
2275         .name                   = "cpuset",
2276         .init_fs_context        = cpuset_init_fs_context,
2277         .fs_flags               = FS_USERNS_MOUNT,
2278 };
2279 #endif
2280 
2281 int cgroup_path_ns_locked(struct cgroup *cgrp, char *buf, size_t buflen,
2282                           struct cgroup_namespace *ns)
2283 {
2284         struct cgroup *root = cset_cgroup_from_root(ns->root_cset, cgrp->root);
2285 
2286         return kernfs_path_from_node(cgrp->kn, root->kn, buf, buflen);
2287 }
2288 
2289 int cgroup_path_ns(struct cgroup *cgrp, char *buf, size_t buflen,
2290                    struct cgroup_namespace *ns)
2291 {
2292         int ret;
2293 
2294         mutex_lock(&cgroup_mutex);
2295         spin_lock_irq(&css_set_lock);
2296 
2297         ret = cgroup_path_ns_locked(cgrp, buf, buflen, ns);
2298 
2299         spin_unlock_irq(&css_set_lock);
2300         mutex_unlock(&cgroup_mutex);
2301 
2302         return ret;
2303 }
2304 EXPORT_SYMBOL_GPL(cgroup_path_ns);
2305 
2306 /**
2307  * task_cgroup_path - cgroup path of a task in the first cgroup hierarchy
2308  * @task: target task
2309  * @buf: the buffer to write the path into
2310  * @buflen: the length of the buffer
2311  *
2312  * Determine @task's cgroup on the first (the one with the lowest non-zero
2313  * hierarchy_id) cgroup hierarchy and copy its path into @buf.  This
2314  * function grabs cgroup_mutex and shouldn't be used inside locks used by
2315  * cgroup controller callbacks.
2316  *
2317  * Return value is the same as kernfs_path().
2318  */
2319 int task_cgroup_path(struct task_struct *task, char *buf, size_t buflen)
2320 {
2321         struct cgroup_root *root;
2322         struct cgroup *cgrp;
2323         int hierarchy_id = 1;
2324         int ret;
2325 
2326         mutex_lock(&cgroup_mutex);
2327         spin_lock_irq(&css_set_lock);
2328 
2329         root = idr_get_next(&cgroup_hierarchy_idr, &hierarchy_id);
2330 
2331         if (root) {
2332                 cgrp = task_cgroup_from_root(task, root);
2333                 ret = cgroup_path_ns_locked(cgrp, buf, buflen, &init_cgroup_ns);
2334         } else {
2335                 /* if no hierarchy exists, everyone is in "/" */
2336                 ret = strlcpy(buf, "/", buflen);
2337         }
2338 
2339         spin_unlock_irq(&css_set_lock);
2340         mutex_unlock(&cgroup_mutex);
2341         return ret;
2342 }
2343 EXPORT_SYMBOL_GPL(task_cgroup_path);
2344 
2345 /**
2346  * cgroup_migrate_add_task - add a migration target task to a migration context
2347  * @task: target task
2348  * @mgctx: target migration context
2349  *
2350  * Add @task, which is a migration target, to @mgctx->tset.  This function
2351  * becomes noop if @task doesn't need to be migrated.  @task's css_set
2352  * should have been added as a migration source and @task->cg_list will be
2353  * moved from the css_set's tasks list to mg_tasks one.
2354  */
2355 static void cgroup_migrate_add_task(struct task_struct *task,
2356                                     struct cgroup_mgctx *mgctx)
2357 {
2358         struct css_set *cset;
2359 
2360         lockdep_assert_held(&css_set_lock);
2361 
2362         /* @task either already exited or can't exit until the end */
2363         if (task->flags & PF_EXITING)
2364                 return;
2365 
2366         /* cgroup_threadgroup_rwsem protects racing against forks */
2367         WARN_ON_ONCE(list_empty(&task->cg_list));
2368 
2369         cset = task_css_set(task);
2370         if (!cset->mg_src_cgrp)
2371                 return;
2372 
2373         mgctx->tset.nr_tasks++;
2374 
2375         list_move_tail(&task->cg_list, &cset->mg_tasks);
2376         if (list_empty(&cset->mg_node))
2377                 list_add_tail(&cset->mg_node,
2378                               &mgctx->tset.src_csets);
2379         if (list_empty(&cset->mg_dst_cset->mg_node))
2380                 list_add_tail(&cset->mg_dst_cset->mg_node,
2381                               &mgctx->tset.dst_csets);
2382 }
2383 
2384 /**
2385  * cgroup_taskset_first - reset taskset and return the first task
2386  * @tset: taskset of interest
2387  * @dst_cssp: output variable for the destination css
2388  *
2389  * @tset iteration is initialized and the first task is returned.
2390  */
2391 struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset,
2392                                          struct cgroup_subsys_state **dst_cssp)
2393 {
2394         tset->cur_cset = list_first_entry(tset->csets, struct css_set, mg_node);
2395         tset->cur_task = NULL;
2396 
2397         return cgroup_taskset_next(tset, dst_cssp);
2398 }
2399 
2400 /**
2401  * cgroup_taskset_next - iterate to the next task in taskset
2402  * @tset: taskset of interest
2403  * @dst_cssp: output variable for the destination css
2404  *
2405  * Return the next task in @tset.  Iteration must have been initialized
2406  * with cgroup_taskset_first().
2407  */
2408 struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset,
2409                                         struct cgroup_subsys_state **dst_cssp)
2410 {
2411         struct css_set *cset = tset->cur_cset;
2412         struct task_struct *task = tset->cur_task;
2413 
2414         while (CGROUP_HAS_SUBSYS_CONFIG && &cset->mg_node != tset->csets) {
2415                 if (!task)
2416                         task = list_first_entry(&cset->mg_tasks,
2417                                                 struct task_struct, cg_list);
2418                 else
2419                         task = list_next_entry(task, cg_list);
2420 
2421                 if (&task->cg_list != &cset->mg_tasks) {
2422                         tset->cur_cset = cset;
2423                         tset->cur_task = task;
2424 
2425                         /*
2426                          * This function may be called both before and
2427                          * after cgroup_taskset_migrate().  The two cases
2428                          * can be distinguished by looking at whether @cset
2429                          * has its ->mg_dst_cset set.
2430                          */
2431                         if (cset->mg_dst_cset)
2432                                 *dst_cssp = cset->mg_dst_cset->subsys[tset->ssid];
2433                         else
2434                                 *dst_cssp = cset->subsys[tset->ssid];
2435 
2436                         return task;
2437                 }
2438 
2439                 cset = list_next_entry(cset, mg_node);
2440                 task = NULL;
2441         }
2442 
2443         return NULL;
2444 }
2445 
2446 /**
2447  * cgroup_migrate_execute - migrate a taskset
2448  * @mgctx: migration context
2449  *
2450  * Migrate tasks in @mgctx as setup by migration preparation functions.
2451  * This function fails iff one of the ->can_attach callbacks fails and
2452  * guarantees that either all or none of the tasks in @mgctx are migrated.
2453  * @mgctx is consumed regardless of success.
2454  */
2455 static int cgroup_migrate_execute(struct cgroup_mgctx *mgctx)
2456 {
2457         struct cgroup_taskset *tset = &mgctx->tset;
2458         struct cgroup_subsys *ss;
2459         struct task_struct *task, *tmp_task;
2460         struct css_set *cset, *tmp_cset;
2461         int ssid, failed_ssid, ret;
2462 
2463         /* check that we can legitimately attach to the cgroup */
2464         if (tset->nr_tasks) {
2465                 do_each_subsys_mask(ss, ssid, mgctx->ss_mask) {
2466                         if (ss->can_attach) {
2467                                 tset->ssid = ssid;
2468                                 ret = ss->can_attach(tset);
2469                                 if (ret) {
2470                                         failed_ssid = ssid;
2471                                         goto out_cancel_attach;
2472                                 }
2473                         }
2474                 } while_each_subsys_mask();
2475         }
2476 
2477         /*
2478          * Now that we're guaranteed success, proceed to move all tasks to
2479          * the new cgroup.  There are no failure cases after here, so this
2480          * is the commit point.
2481          */
2482         spin_lock_irq(&css_set_lock);
2483         list_for_each_entry(cset, &tset->src_csets, mg_node) {
2484                 list_for_each_entry_safe(task, tmp_task, &cset->mg_tasks, cg_list) {
2485                         struct css_set *from_cset = task_css_set(task);
2486                         struct css_set *to_cset = cset->mg_dst_cset;
2487 
2488                         get_css_set(to_cset);
2489                         to_cset->nr_tasks++;
2490                         css_set_move_task(task, from_cset, to_cset, true);
2491                         from_cset->nr_tasks--;
2492                         /*
2493                          * If the source or destination cgroup is frozen,
2494                          * the task might require to change its state.
2495                          */
2496                         cgroup_freezer_migrate_task(task, from_cset->dfl_cgrp,
2497                                                     to_cset->dfl_cgrp);
2498                         put_css_set_locked(from_cset);
2499 
2500                 }
2501         }
2502         spin_unlock_irq(&css_set_lock);
2503 
2504         /*
2505          * Migration is committed, all target tasks are now on dst_csets.
2506          * Nothing is sensitive to fork() after this point.  Notify
2507          * controllers that migration is complete.
2508          */
2509         tset->csets = &tset->dst_csets;
2510 
2511         if (tset->nr_tasks) {
2512                 do_each_subsys_mask(ss, ssid, mgctx->ss_mask) {
2513                         if (ss->attach) {
2514                                 tset->ssid = ssid;
2515                                 ss->attach(tset);
2516                         }
2517                 } while_each_subsys_mask();
2518         }
2519 
2520         ret = 0;
2521         goto out_release_tset;
2522 
2523 out_cancel_attach:
2524         if (tset->nr_tasks) {
2525                 do_each_subsys_mask(ss, ssid, mgctx->ss_mask) {
2526                         if (ssid == failed_ssid)
2527                                 break;
2528                         if (ss->cancel_attach) {
2529                                 tset->ssid = ssid;
2530                                 ss->cancel_attach(tset);
2531                         }
2532                 } while_each_subsys_mask();
2533         }
2534 out_release_tset:
2535         spin_lock_irq(&css_set_lock);
2536         list_splice_init(&tset->dst_csets, &tset->src_csets);
2537         list_for_each_entry_safe(cset, tmp_cset, &tset->src_csets, mg_node) {
2538                 list_splice_tail_init(&cset->mg_tasks, &cset->tasks);
2539                 list_del_init(&cset->mg_node);
2540         }
2541         spin_unlock_irq(&css_set_lock);
2542 
2543         /*
2544          * Re-initialize the cgroup_taskset structure in case it is reused
2545          * again in another cgroup_migrate_add_task()/cgroup_migrate_execute()
2546          * iteration.
2547          */
2548         tset->nr_tasks = 0;
2549         tset->csets    = &tset->src_csets;
2550         return ret;
2551 }
2552 
2553 /**
2554  * cgroup_migrate_vet_dst - verify whether a cgroup can be migration destination
2555  * @dst_cgrp: destination cgroup to test
2556  *
2557  * On the default hierarchy, except for the mixable, (possible) thread root
2558  * and threaded cgroups, subtree_control must be zero for migration
2559  * destination cgroups with tasks so that child cgroups don't compete
2560  * against tasks.
2561  */
2562 int cgroup_migrate_vet_dst(struct cgroup *dst_cgrp)
2563 {
2564         /* v1 doesn't have any restriction */
2565         if (!cgroup_on_dfl(dst_cgrp))
2566                 return 0;
2567 
2568         /* verify @dst_cgrp can host resources */
2569         if (!cgroup_is_valid_domain(dst_cgrp->dom_cgrp))
2570                 return -EOPNOTSUPP;
2571 
2572         /* mixables don't care */
2573         if (cgroup_is_mixable(dst_cgrp))
2574                 return 0;
2575 
2576         /*
2577          * If @dst_cgrp is already or can become a thread root or is
2578          * threaded, it doesn't matter.
2579          */
2580         if (cgroup_can_be_thread_root(dst_cgrp) || cgroup_is_threaded(dst_cgrp))
2581                 return 0;
2582 
2583         /* apply no-internal-process constraint */
2584         if (dst_cgrp->subtree_control)
2585                 return -EBUSY;
2586 
2587         return 0;
2588 }
2589 
2590 /**
2591  * cgroup_migrate_finish - cleanup after attach
2592  * @mgctx: migration context
2593  *
2594  * Undo cgroup_migrate_add_src() and cgroup_migrate_prepare_dst().  See
2595  * those functions for details.
2596  */
2597 void cgroup_migrate_finish(struct cgroup_mgctx *mgctx)
2598 {
2599         LIST_HEAD(preloaded);
2600         struct css_set *cset, *tmp_cset;
2601 
2602         lockdep_assert_held(&cgroup_mutex);
2603 
2604         spin_lock_irq(&css_set_lock);
2605 
2606         list_splice_tail_init(&mgctx->preloaded_src_csets, &preloaded);
2607         list_splice_tail_init(&mgctx->preloaded_dst_csets, &preloaded);
2608 
2609         list_for_each_entry_safe(cset, tmp_cset, &preloaded, mg_preload_node) {
2610                 cset->mg_src_cgrp = NULL;
2611                 cset->mg_dst_cgrp = NULL;
2612                 cset->mg_dst_cset = NULL;
2613                 list_del_init(&cset->mg_preload_node);
2614                 put_css_set_locked(cset);
2615         }
2616 
2617         spin_unlock_irq(&css_set_lock);
2618 }
2619 
2620 /**
2621  * cgroup_migrate_add_src - add a migration source css_set
2622  * @src_cset: the source css_set to add
2623  * @dst_cgrp: the destination cgroup
2624  * @mgctx: migration context
2625  *
2626  * Tasks belonging to @src_cset are about to be migrated to @dst_cgrp.  Pin
2627  * @src_cset and add it to @mgctx->src_csets, which should later be cleaned
2628  * up by cgroup_migrate_finish().
2629  *
2630  * This function may be called without holding cgroup_threadgroup_rwsem
2631  * even if the target is a process.  Threads may be created and destroyed
2632  * but as long as cgroup_mutex is not dropped, no new css_set can be put
2633  * into play and the preloaded css_sets are guaranteed to cover all
2634  * migrations.
2635  */
2636 void cgroup_migrate_add_src(struct css_set *src_cset,
2637                             struct cgroup *dst_cgrp,
2638                             struct cgroup_mgctx *mgctx)
2639 {
2640         struct cgroup *src_cgrp;
2641 
2642         lockdep_assert_held(&cgroup_mutex);
2643         lockdep_assert_held(&css_set_lock);
2644 
2645         /*
2646          * If ->dead, @src_set is associated with one or more dead cgroups
2647          * and doesn't contain any migratable tasks.  Ignore it early so
2648          * that the rest of migration path doesn't get confused by it.
2649          */
2650         if (src_cset->dead)
2651                 return;
2652 
2653         src_cgrp = cset_cgroup_from_root(src_cset, dst_cgrp->root);
2654 
2655         if (!list_empty(&src_cset->mg_preload_node))
2656                 return;
2657 
2658         WARN_ON(src_cset->mg_src_cgrp);
2659         WARN_ON(src_cset->mg_dst_cgrp);
2660         WARN_ON(!list_empty(&src_cset->mg_tasks));
2661         WARN_ON(!list_empty(&src_cset->mg_node));
2662 
2663         src_cset->mg_src_cgrp = src_cgrp;
2664         src_cset->mg_dst_cgrp = dst_cgrp;
2665         get_css_set(src_cset);
2666         list_add_tail(&src_cset->mg_preload_node, &mgctx->preloaded_src_csets);
2667 }
2668 
2669 /**
2670  * cgroup_migrate_prepare_dst - prepare destination css_sets for migration
2671  * @mgctx: migration context
2672  *
2673  * Tasks are about to be moved and all the source css_sets have been
2674  * preloaded to @mgctx->preloaded_src_csets.  This function looks up and
2675  * pins all destination css_sets, links each to its source, and append them
2676  * to @mgctx->preloaded_dst_csets.
2677  *
2678  * This function must be called after cgroup_migrate_add_src() has been
2679  * called on each migration source css_set.  After migration is performed
2680  * using cgroup_migrate(), cgroup_migrate_finish() must be called on
2681  * @mgctx.
2682  */
2683 int cgroup_migrate_prepare_dst(struct cgroup_mgctx *mgctx)
2684 {
2685         struct css_set *src_cset, *tmp_cset;
2686 
2687         lockdep_assert_held(&cgroup_mutex);
2688 
2689         /* look up the dst cset for each src cset and link it to src */
2690         list_for_each_entry_safe(src_cset, tmp_cset, &mgctx->preloaded_src_csets,
2691                                  mg_preload_node) {
2692                 struct css_set *dst_cset;
2693                 struct cgroup_subsys *ss;
2694                 int ssid;
2695 
2696                 dst_cset = find_css_set(src_cset, src_cset->mg_dst_cgrp);
2697                 if (!dst_cset)
2698                         return -ENOMEM;
2699 
2700                 WARN_ON_ONCE(src_cset->mg_dst_cset || dst_cset->mg_dst_cset);
2701 
2702                 /*
2703                  * If src cset equals dst, it's noop.  Drop the src.
2704                  * cgroup_migrate() will skip the cset too.  Note that we
2705                  * can't handle src == dst as some nodes are used by both.
2706                  */
2707                 if (src_cset == dst_cset) {
2708                         src_cset->mg_src_cgrp = NULL;
2709                         src_cset->mg_dst_cgrp = NULL;
2710                         list_del_init(&src_cset->mg_preload_node);
2711                         put_css_set(src_cset);
2712                         put_css_set(dst_cset);
2713                         continue;
2714                 }
2715 
2716                 src_cset->mg_dst_cset = dst_cset;
2717 
2718                 if (list_empty(&dst_cset->mg_preload_node))
2719                         list_add_tail(&dst_cset->mg_preload_node,
2720                                       &mgctx->preloaded_dst_csets);
2721                 else
2722                         put_css_set(dst_cset);
2723 
2724                 for_each_subsys(ss, ssid)
2725                         if (src_cset->subsys[ssid] != dst_cset->subsys[ssid])
2726                                 mgctx->ss_mask |= 1 << ssid;
2727         }
2728 
2729         return 0;
2730 }
2731 
2732 /**
2733  * cgroup_migrate - migrate a process or task to a cgroup
2734  * @leader: the leader of the process or the task to migrate
2735  * @threadgroup: whether @leader points to the whole process or a single task
2736  * @mgctx: migration context
2737  *
2738  * Migrate a process or task denoted by @leader.  If migrating a process,
2739  * the caller must be holding cgroup_threadgroup_rwsem.  The caller is also
2740  * responsible for invoking cgroup_migrate_add_src() and
2741  * cgroup_migrate_prepare_dst() on the targets before invoking this
2742  * function and following up with cgroup_migrate_finish().
2743  *
2744  * As long as a controller's ->can_attach() doesn't fail, this function is
2745  * guaranteed to succeed.  This means that, excluding ->can_attach()
2746  * failure, when migrating multiple targets, the success or failure can be
2747  * decided for all targets by invoking group_migrate_prepare_dst() before
2748  * actually starting migrating.
2749  */
2750 int cgroup_migrate(struct task_struct *leader, bool threadgroup,
2751                    struct cgroup_mgctx *mgctx)
2752 {
2753         struct task_struct *task;
2754 
2755         /*
2756          * Prevent freeing of tasks while we take a snapshot. Tasks that are
2757          * already PF_EXITING could be freed from underneath us unless we
2758          * take an rcu_read_lock.
2759          */
2760         spin_lock_irq(&css_set_lock);
2761         rcu_read_lock();
2762         task = leader;
2763         do {
2764                 cgroup_migrate_add_task(task, mgctx);
2765                 if (!threadgroup)
2766                         break;
2767         } while_each_thread(leader, task);
2768         rcu_read_unlock();
2769         spin_unlock_irq(&css_set_lock);
2770 
2771         return cgroup_migrate_execute(mgctx);
2772 }
2773 
2774 /**
2775  * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup
2776  * @dst_cgrp: the cgroup to attach to
2777  * @leader: the task or the leader of the threadgroup to be attached
2778  * @threadgroup: attach the whole threadgroup?
2779  *
2780  * Call holding cgroup_mutex and cgroup_threadgroup_rwsem.
2781  */
2782 int cgroup_attach_task(struct cgroup *dst_cgrp, struct task_struct *leader,
2783                        bool threadgroup)
2784 {
2785         DEFINE_CGROUP_MGCTX(mgctx);
2786         struct task_struct *task;
2787         int ret = 0;
2788 
2789         /* look up all src csets */
2790         spin_lock_irq(&css_set_lock);
2791         rcu_read_lock();
2792         task = leader;
2793         do {
2794                 cgroup_migrate_add_src(task_css_set(task), dst_cgrp, &mgctx);
2795                 if (!threadgroup)
2796                         break;
2797         } while_each_thread(leader, task);
2798         rcu_read_unlock();
2799         spin_unlock_irq(&css_set_lock);
2800 
2801         /* prepare dst csets and commit */
2802         ret = cgroup_migrate_prepare_dst(&mgctx);
2803         if (!ret)
2804                 ret = cgroup_migrate(leader, threadgroup, &mgctx);
2805 
2806         cgroup_migrate_finish(&mgctx);
2807 
2808         if (!ret)
2809                 TRACE_CGROUP_PATH(attach_task, dst_cgrp, leader, threadgroup);
2810 
2811         return ret;
2812 }
2813 
2814 struct task_struct *cgroup_procs_write_start(char *buf, bool threadgroup,
2815                                              bool *locked)
2816         __acquires(&cgroup_threadgroup_rwsem)
2817 {
2818         struct task_struct *tsk;
2819         pid_t pid;
2820 
2821         if (kstrtoint(strstrip(buf), 0, &pid) || pid < 0)
2822                 return ERR_PTR(-EINVAL);
2823 
2824         /*
2825          * If we migrate a single thread, we don't care about threadgroup
2826          * stability. If the thread is `current`, it won't exit(2) under our
2827          * hands or change PID through exec(2). We exclude
2828          * cgroup_update_dfl_csses and other cgroup_{proc,thread}s_write
2829          * callers by cgroup_mutex.
2830          * Therefore, we can skip the global lock.
2831          */
2832         lockdep_assert_held(&cgroup_mutex);
2833         if (pid || threadgroup) {
2834                 percpu_down_write(&cgroup_threadgroup_rwsem);
2835                 *locked = true;
2836         } else {
2837                 *locked = false;
2838         }
2839 
2840         rcu_read_lock();
2841         if (pid) {
2842                 tsk = find_task_by_vpid(pid);
2843                 if (!tsk) {
2844                         tsk = ERR_PTR(-ESRCH);
2845                         goto out_unlock_threadgroup;
2846                 }
2847         } else {
2848                 tsk = current;
2849         }
2850 
2851         if (threadgroup)
2852                 tsk = tsk->group_leader;
2853 
2854         /*
2855          * kthreads may acquire PF_NO_SETAFFINITY during initialization.
2856          * If userland migrates such a kthread to a non-root cgroup, it can
2857          * become trapped in a cpuset, or RT kthread may be born in a
2858          * cgroup with no rt_runtime allocated.  Just say no.
2859          */
2860         if (tsk->no_cgroup_migration || (tsk->flags & PF_NO_SETAFFINITY)) {
2861                 tsk = ERR_PTR(-EINVAL);
2862                 goto out_unlock_threadgroup;
2863         }
2864 
2865         get_task_struct(tsk);
2866         goto out_unlock_rcu;
2867 
2868 out_unlock_threadgroup:
2869         if (*locked) {
2870                 percpu_up_write(&cgroup_threadgroup_rwsem);
2871                 *locked = false;
2872         }
2873 out_unlock_rcu:
2874         rcu_read_unlock();
2875         return tsk;
2876 }
2877 
2878 void cgroup_procs_write_finish(struct task_struct *task, bool locked)
2879         __releases(&cgroup_threadgroup_rwsem)
2880 {
2881         struct cgroup_subsys *ss;
2882         int ssid;
2883 
2884         /* release reference from cgroup_procs_write_start() */
2885         put_task_struct(task);
2886 
2887         if (locked)
2888                 percpu_up_write(&cgroup_threadgroup_rwsem);
2889         for_each_subsys(ss, ssid)
2890                 if (ss->post_attach)
2891                         ss->post_attach();
2892 }
2893 
2894 static void cgroup_print_ss_mask(struct seq_file *seq, u16 ss_mask)
2895 {
2896         struct cgroup_subsys *ss;
2897         bool printed = false;
2898         int ssid;
2899 
2900         do_each_subsys_mask(ss, ssid, ss_mask) {
2901                 if (printed)
2902                         seq_putc(seq, ' ');
2903                 seq_puts(seq, ss->name);
2904                 printed = true;
2905         } while_each_subsys_mask();
2906         if (printed)
2907                 seq_putc(seq, '\n');
2908 }
2909 
2910 /* show controllers which are enabled from the parent */
2911 static int cgroup_controllers_show(struct seq_file *seq, void *v)
2912 {
2913         struct cgroup *cgrp = seq_css(seq)->cgroup;
2914 
2915         cgroup_print_ss_mask(seq, cgroup_control(cgrp));
2916         return 0;
2917 }
2918 
2919 /* show controllers which are enabled for a given cgroup's children */
2920 static int cgroup_subtree_control_show(struct seq_file *seq, void *v)
2921 {
2922         struct cgroup *cgrp = seq_css(seq)->cgroup;
2923 
2924         cgroup_print_ss_mask(seq, cgrp->subtree_control);
2925         return 0;
2926 }
2927 
2928 /**
2929  * cgroup_update_dfl_csses - update css assoc of a subtree in default hierarchy
2930  * @cgrp: root of the subtree to update csses for
2931  *
2932  * @cgrp's control masks have changed and its subtree's css associations
2933  * need to be updated accordingly.  This function looks up all css_sets
2934  * which are attached to the subtree, creates the matching updated css_sets
2935  * and migrates the tasks to the new ones.
2936  */
2937 static int cgroup_update_dfl_csses(struct cgroup *cgrp)
2938 {
2939         DEFINE_CGROUP_MGCTX(mgctx);
2940         struct cgroup_subsys_state *d_css;
2941         struct cgroup *dsct;
2942         struct css_set *src_cset;
2943         int ret;
2944 
2945         lockdep_assert_held(&cgroup_mutex);
2946 
2947         percpu_down_write(&cgroup_threadgroup_rwsem);
2948 
2949         /* look up all csses currently attached to @cgrp's subtree */
2950         spin_lock_irq(&css_set_lock);
2951         cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) {
2952                 struct cgrp_cset_link *link;
2953 
2954                 list_for_each_entry(link, &dsct->cset_links, cset_link)
2955                         cgroup_migrate_add_src(link->cset, dsct, &mgctx);
2956         }
2957         spin_unlock_irq(&css_set_lock);
2958 
2959         /* NULL dst indicates self on default hierarchy */
2960         ret = cgroup_migrate_prepare_dst(&mgctx);
2961         if (ret)
2962                 goto out_finish;
2963 
2964         spin_lock_irq(&css_set_lock);
2965         list_for_each_entry(src_cset, &mgctx.preloaded_src_csets, mg_preload_node) {
2966                 struct task_struct *task, *ntask;
2967 
2968                 /* all tasks in src_csets need to be migrated */
2969                 list_for_each_entry_safe(task, ntask, &src_cset->tasks, cg_list)
2970                         cgroup_migrate_add_task(task, &mgctx);
2971         }
2972         spin_unlock_irq(&css_set_lock);
2973 
2974         ret = cgroup_migrate_execute(&mgctx);
2975 out_finish:
2976         cgroup_migrate_finish(&mgctx);
2977         percpu_up_write(&cgroup_threadgroup_rwsem);
2978         return ret;
2979 }
2980 
2981 /**
2982  * cgroup_lock_and_drain_offline - lock cgroup_mutex and drain offlined csses
2983  * @cgrp: root of the target subtree
2984  *
2985  * Because css offlining is asynchronous, userland may try to re-enable a
2986  * controller while the previous css is still around.  This function grabs
2987  * cgroup_mutex and drains the previous css instances of @cgrp's subtree.
2988  */
2989 void cgroup_lock_and_drain_offline(struct cgroup *cgrp)
2990         __acquires(&cgroup_mutex)
2991 {
2992         struct cgroup *dsct;
2993         struct cgroup_subsys_state *d_css;
2994         struct cgroup_subsys *ss;
2995         int ssid;
2996 
2997 restart:
2998         mutex_lock(&cgroup_mutex);
2999 
3000         cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) {
3001                 for_each_subsys(ss, ssid) {
3002                         struct cgroup_subsys_state *css = cgroup_css(dsct, ss);
3003                         DEFINE_WAIT(wait);
3004 
3005                         if (!css || !percpu_ref_is_dying(&css->refcnt))
3006                                 continue;
3007 
3008                         cgroup_get_live(dsct);
3009                         prepare_to_wait(&dsct->offline_waitq, &wait,
3010                                         TASK_UNINTERRUPTIBLE);
3011 
3012                         mutex_unlock(&cgroup_mutex);
3013                         schedule();
3014                         finish_wait(&dsct->offline_waitq, &wait);
3015 
3016                         cgroup_put(dsct);
3017                         goto restart;
3018                 }
3019         }
3020 }
3021 
3022 /**
3023  * cgroup_save_control - save control masks and dom_cgrp of a subtree
3024  * @cgrp: root of the target subtree
3025  *
3026  * Save ->subtree_control, ->subtree_ss_mask and ->dom_cgrp to the
3027  * respective old_ prefixed fields for @cgrp's subtree including @cgrp
3028  * itself.
3029  */
3030 static void cgroup_save_control(struct cgroup *cgrp)
3031 {
3032         struct cgroup *dsct;
3033         struct cgroup_subsys_state *d_css;
3034 
3035         cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) {
3036                 dsct->old_subtree_control = dsct->subtree_control;
3037                 dsct->old_subtree_ss_mask = dsct->subtree_ss_mask;
3038                 dsct->old_dom_cgrp = dsct->dom_cgrp;
3039         }
3040 }
3041 
3042 /**
3043  * cgroup_propagate_control - refresh control masks of a subtree
3044  * @cgrp: root of the target subtree
3045  *
3046  * For @cgrp and its subtree, ensure ->subtree_ss_mask matches
3047  * ->subtree_control and propagate controller availability through the
3048  * subtree so that descendants don't have unavailable controllers enabled.
3049  */
3050 static void cgroup_propagate_control(struct cgroup *cgrp)
3051 {
3052         struct cgroup *dsct;
3053         struct cgroup_subsys_state *d_css;
3054 
3055         cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) {
3056                 dsct->subtree_control &= cgroup_control(dsct);
3057                 dsct->subtree_ss_mask =
3058                         cgroup_calc_subtree_ss_mask(dsct->subtree_control,
3059                                                     cgroup_ss_mask(dsct));
3060         }
3061 }
3062 
3063 /**
3064  * cgroup_restore_control - restore control masks and dom_cgrp of a subtree
3065  * @cgrp: root of the target subtree
3066  *
3067  * Restore ->subtree_control, ->subtree_ss_mask and ->dom_cgrp from the
3068  * respective old_ prefixed fields for @cgrp's subtree including @cgrp
3069  * itself.
3070  */
3071 static void cgroup_restore_control(struct cgroup *cgrp)
3072 {
3073         struct cgroup *dsct;
3074         struct cgroup_subsys_state *d_css;
3075 
3076         cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) {
3077                 dsct->subtree_control = dsct->old_subtree_control;
3078                 dsct->subtree_ss_mask = dsct->old_subtree_ss_mask;
3079                 dsct->dom_cgrp = dsct->old_dom_cgrp;
3080         }
3081 }
3082 
3083 static bool css_visible(struct cgroup_subsys_state *css)
3084 {
3085         struct cgroup_subsys *ss = css->ss;
3086         struct cgroup *cgrp = css->cgroup;
3087 
3088         if (cgroup_control(cgrp) & (1 << ss->id))
3089                 return true;
3090         if (!(cgroup_ss_mask(cgrp) & (1 << ss->id)))
3091                 return false;
3092         return cgroup_on_dfl(cgrp) && ss->implicit_on_dfl;
3093 }
3094 
3095 /**
3096  * cgroup_apply_control_enable - enable or show csses according to control
3097  * @cgrp: root of the target subtree
3098  *
3099  * Walk @cgrp's subtree and create new csses or make the existing ones
3100  * visible.  A css is created invisible if it's being implicitly enabled
3101  * through dependency.  An invisible css is made visible when the userland
3102  * explicitly enables it.
3103  *
3104  * Returns 0 on success, -errno on failure.  On failure, csses which have
3105  * been processed already aren't cleaned up.  The caller is responsible for
3106  * cleaning up with cgroup_apply_control_disable().
3107  */
3108 static int cgroup_apply_control_enable(struct cgroup *cgrp)
3109 {
3110         struct cgroup *dsct;
3111         struct cgroup_subsys_state *d_css;
3112         struct cgroup_subsys *ss;
3113         int ssid, ret;
3114 
3115         cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) {
3116                 for_each_subsys(ss, ssid) {
3117                         struct cgroup_subsys_state *css = cgroup_css(dsct, ss);
3118 
3119                         if (!(cgroup_ss_mask(dsct) & (1 << ss->id)))
3120                                 continue;
3121 
3122                         if (!css) {
3123                                 css = css_create(dsct, ss);
3124                                 if (IS_ERR(css))
3125                                         return PTR_ERR(css);
3126                         }
3127 
3128                         WARN_ON_ONCE(percpu_ref_is_dying(&css->refcnt));
3129 
3130                         if (css_visible(css)) {
3131                                 ret = css_populate_dir(css);
3132                                 if (ret)
3133                                         return ret;
3134                         }
3135                 }
3136         }
3137 
3138         return 0;
3139 }
3140 
3141 /**
3142  * cgroup_apply_control_disable - kill or hide csses according to control
3143  * @cgrp: root of the target subtree
3144  *
3145  * Walk @cgrp's subtree and kill and hide csses so that they match
3146  * cgroup_ss_mask() and cgroup_visible_mask().
3147  *
3148  * A css is hidden when the userland requests it to be disabled while other
3149  * subsystems are still depending on it.  The css must not actively control
3150  * resources and be in the vanilla state if it's made visible again later.
3151  * Controllers which may be depended upon should provide ->css_reset() for
3152  * this purpose.
3153  */
3154 static void cgroup_apply_control_disable(struct cgroup *cgrp)
3155 {
3156         struct cgroup *dsct;
3157         struct cgroup_subsys_state *d_css;
3158         struct cgroup_subsys *ss;
3159         int ssid;
3160 
3161         cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) {
3162                 for_each_subsys(ss, ssid) {
3163                         struct cgroup_subsys_state *css = cgroup_css(dsct, ss);
3164 
3165                         if (!css)
3166                                 continue;
3167 
3168                         WARN_ON_ONCE(percpu_ref_is_dying(&css->refcnt));
3169 
3170                         if (css->parent &&
3171                             !(cgroup_ss_mask(dsct) & (1 << ss->id))) {
3172                                 kill_css(css);
3173                         } else if (!css_visible(css)) {
3174                                 css_clear_dir(css);
3175                                 if (ss->css_reset)
3176                                         ss->css_reset(css);
3177                         }
3178                 }
3179         }
3180 }
3181 
3182 /**
3183  * cgroup_apply_control - apply control mask updates to the subtree
3184  * @cgrp: root of the target subtree
3185  *
3186  * subsystems can be enabled and disabled in a subtree using the following
3187  * steps.
3188  *
3189  * 1. Call cgroup_save_control() to stash the current state.
3190  * 2. Update ->subtree_control masks in the subtree as desired.
3191  * 3. Call cgroup_apply_control() to apply the changes.
3192  * 4. Optionally perform other related operations.
3193  * 5. Call cgroup_finalize_control() to finish up.
3194  *
3195  * This function implements step 3 and propagates the mask changes
3196  * throughout @cgrp's subtree, updates csses accordingly and perform
3197  * process migrations.
3198  */
3199 static int cgroup_apply_control(struct cgroup *cgrp)
3200 {
3201         int ret;
3202 
3203         cgroup_propagate_control(cgrp);
3204 
3205         ret = cgroup_apply_control_enable(cgrp);
3206         if (ret)
3207                 return ret;
3208 
3209         /*
3210          * At this point, cgroup_e_css_by_mask() results reflect the new csses
3211          * making the following cgroup_update_dfl_csses() properly update
3212          * css associations of all tasks in the subtree.
3213          */
3214         ret = cgroup_update_dfl_csses(cgrp);
3215         if (ret)
3216                 return ret;
3217 
3218         return 0;
3219 }
3220 
3221 /**
3222  * cgroup_finalize_control - finalize control mask update
3223  * @cgrp: root of the target subtree
3224  * @ret: the result of the update
3225  *
3226  * Finalize control mask update.  See cgroup_apply_control() for more info.
3227  */
3228 static void cgroup_finalize_control(struct cgroup *cgrp, int ret)
3229 {
3230         if (ret) {
3231                 cgroup_restore_control(cgrp);
3232                 cgroup_propagate_control(cgrp);
3233         }
3234 
3235         cgroup_apply_control_disable(cgrp);
3236 }
3237 
3238 static int cgroup_vet_subtree_control_enable(struct cgroup *cgrp, u16 enable)
3239 {
3240         u16 domain_enable = enable & ~cgrp_dfl_threaded_ss_mask;
3241 
3242         /* if nothing is getting enabled, nothing to worry about */
3243         if (!enable)
3244                 return 0;
3245 
3246         /* can @cgrp host any resources? */
3247         if (!cgroup_is_valid_domain(cgrp->dom_cgrp))
3248                 return -EOPNOTSUPP;
3249 
3250         /* mixables don't care */
3251         if (cgroup_is_mixable(cgrp))
3252                 return 0;
3253 
3254         if (domain_enable) {
3255                 /* can't enable domain controllers inside a thread subtree */
3256                 if (cgroup_is_thread_root(cgrp) || cgroup_is_threaded(cgrp))
3257                         return -EOPNOTSUPP;
3258         } else {
3259                 /*
3260                  * Threaded controllers can handle internal competitions
3261                  * and are always allowed inside a (prospective) thread
3262                  * subtree.
3263                  */
3264                 if (cgroup_can_be_thread_root(cgrp) || cgroup_is_threaded(cgrp))
3265                         return 0;
3266         }
3267 
3268         /*
3269          * Controllers can't be enabled for a cgroup with tasks to avoid
3270          * child cgroups competing against tasks.
3271          */
3272         if (cgroup_has_tasks(cgrp))
3273                 return -EBUSY;
3274 
3275         return 0;
3276 }
3277 
3278 /* change the enabled child controllers for a cgroup in the default hierarchy */
3279 static ssize_t cgroup_subtree_control_write(struct kernfs_open_file *of,
3280                                             char *buf, size_t nbytes,
3281                                             loff_t off)
3282 {
3283         u16 enable = 0, disable = 0;
3284         struct cgroup *cgrp, *child;
3285         struct cgroup_subsys *ss;
3286         char *tok;
3287         int ssid, ret;
3288 
3289         /*
3290          * Parse input - space separated list of subsystem names prefixed
3291          * with either + or -.
3292          */
3293         buf = strstrip(buf);
3294         while ((tok = strsep(&buf, " "))) {
3295                 if (tok[0] == '\0')
3296                         continue;
3297                 do_each_subsys_mask(ss, ssid, ~cgrp_dfl_inhibit_ss_mask) {
3298                         if (!cgroup_ssid_enabled(ssid) ||
3299                             strcmp(tok + 1, ss->name))
3300                                 continue;
3301 
3302                         if (*tok == '+') {
3303                                 enable |= 1 << ssid;
3304                                 disable &= ~(1 << ssid);
3305                         } else if (*tok == '-') {
3306                                 disable |= 1 << ssid;
3307                                 enable &= ~(1 << ssid);
3308                         } else {
3309                                 return -EINVAL;
3310                         }
3311                         break;
3312                 } while_each_subsys_mask();
3313                 if (ssid == CGROUP_SUBSYS_COUNT)
3314                         return -EINVAL;
3315         }
3316 
3317         cgrp = cgroup_kn_lock_live(of->kn, true);
3318         if (!cgrp)
3319                 return -ENODEV;
3320 
3321         for_each_subsys(ss, ssid) {
3322                 if (enable & (1 << ssid)) {
3323                         if (cgrp->subtree_control & (1 << ssid)) {
3324                                 enable &= ~(1 << ssid);
3325                                 continue;
3326                         }
3327 
3328                         if (!(cgroup_control(cgrp) & (1 << ssid))) {
3329                                 ret = -ENOENT;
3330                                 goto out_unlock;
3331                         }
3332                 } else if (disable & (1 << ssid)) {
3333                         if (!(cgrp->subtree_control & (1 << ssid))) {
3334                                 disable &= ~(1 << ssid);
3335                                 continue;
3336                         }
3337 
3338                         /* a child has it enabled? */
3339                         cgroup_for_each_live_child(child, cgrp) {
3340                                 if (child->subtree_control & (1 << ssid)) {
3341                                         ret = -EBUSY;
3342                                         goto out_unlock;
3343                                 }
3344                         }
3345                 }
3346         }
3347 
3348         if (!enable && !disable) {
3349                 ret = 0;
3350                 goto out_unlock;
3351         }
3352 
3353         ret = cgroup_vet_subtree_control_enable(cgrp, enable);
3354         if (ret)
3355                 goto out_unlock;
3356 
3357         /* save and update control masks and prepare csses */
3358         cgroup_save_control(cgrp);
3359 
3360         cgrp->subtree_control |= enable;
3361         cgrp->subtree_control &= ~disable;
3362 
3363         ret = cgroup_apply_control(cgrp);
3364         cgroup_finalize_control(cgrp, ret);
3365         if (ret)
3366                 goto out_unlock;
3367 
3368         kernfs_activate(cgrp->kn);
3369 out_unlock:
3370         cgroup_kn_unlock(of->kn);
3371         return ret ?: nbytes;
3372 }
3373 
3374 /**
3375  * cgroup_enable_threaded - make @cgrp threaded
3376  * @cgrp: the target cgroup
3377  *
3378  * Called when "threaded" is written to the cgroup.type interface file and
3379  * tries to make @cgrp threaded and join the parent's resource domain.
3380  * This function is never called on the root cgroup as cgroup.type doesn't
3381  * exist on it.
3382  */
3383 static int cgroup_enable_threaded(struct cgroup *cgrp)
3384 {
3385         struct cgroup *parent = cgroup_parent(cgrp);
3386         struct cgroup *dom_cgrp = parent->dom_cgrp;
3387         struct cgroup *dsct;
3388         struct cgroup_subsys_state *d_css;
3389         int ret;
3390 
3391         lockdep_assert_held(&cgroup_mutex);
3392 
3393         /* noop if already threaded */
3394         if (cgroup_is_threaded(cgrp))
3395                 return 0;
3396 
3397         /*
3398          * If @cgroup is populated or has domain controllers enabled, it
3399          * can't be switched.  While the below cgroup_can_be_thread_root()
3400          * test can catch the same conditions, that's only when @parent is
3401          * not mixable, so let's check it explicitly.
3402          */
3403         if (cgroup_is_populated(cgrp) ||
3404             cgrp->subtree_control & ~cgrp_dfl_threaded_ss_mask)
3405                 return -EOPNOTSUPP;
3406 
3407         /* we're joining the parent's domain, ensure its validity */
3408         if (!cgroup_is_valid_domain(dom_cgrp) ||
3409             !cgroup_can_be_thread_root(dom_cgrp))
3410                 return -EOPNOTSUPP;
3411 
3412         /*
3413          * The following shouldn't cause actual migrations and should
3414          * always succeed.
3415          */
3416         cgroup_save_control(cgrp);
3417 
3418         cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp)
3419                 if (dsct == cgrp || cgroup_is_threaded(dsct))
3420                         dsct->dom_cgrp = dom_cgrp;
3421 
3422         ret = cgroup_apply_control(cgrp);
3423         if (!ret)
3424                 parent->nr_threaded_children++;
3425 
3426         cgroup_finalize_control(cgrp, ret);
3427         return ret;
3428 }
3429 
3430 static int cgroup_type_show(struct seq_file *seq, void *v)
3431 {
3432         struct cgroup *cgrp = seq_css(seq)->cgroup;
3433 
3434         if (cgroup_is_threaded(cgrp))
3435                 seq_puts(seq, "threaded\n");
3436         else if (!cgroup_is_valid_domain(cgrp))
3437                 seq_puts(seq, "domain invalid\n");
3438         else if (cgroup_is_thread_root(cgrp))
3439                 seq_puts(seq, "domain threaded\n");
3440         else
3441                 seq_puts(seq, "domain\n");
3442 
3443         return 0;
3444 }
3445 
3446 static ssize_t cgroup_type_write(struct kernfs_open_file *of, char *buf,
3447                                  size_t nbytes, loff_t off)
3448 {
3449         struct cgroup *cgrp;
3450         int ret;
3451 
3452         /* only switching to threaded mode is supported */
3453         if (strcmp(strstrip(buf), "threaded"))
3454                 return -EINVAL;
3455 
3456         /* drain dying csses before we re-apply (threaded) subtree control */
3457         cgrp = cgroup_kn_lock_live(of->kn, true);
3458         if (!cgrp)
3459                 return -ENOENT;
3460 
3461         /* threaded can only be enabled */
3462         ret = cgroup_enable_threaded(cgrp);
3463 
3464         cgroup_kn_unlock(of->kn);
3465         return ret ?: nbytes;
3466 }
3467 
3468 static int cgroup_max_descendants_show(struct seq_file *seq, void *v)
3469 {
3470         struct cgroup *cgrp = seq_css(seq)->cgroup;
3471         int descendants = READ_ONCE(cgrp->max_descendants);
3472 
3473         if (descendants == INT_MAX)
3474                 seq_puts(seq, "max\n");
3475         else
3476                 seq_printf(seq, "%d\n", descendants);
3477 
3478         return 0;
3479 }
3480 
3481 static ssize_t cgroup_max_descendants_write(struct kernfs_open_file *of,
3482                                            char *buf, size_t nbytes, loff_t off)
3483 {
3484         struct cgroup *cgrp;
3485         int descendants;
3486         ssize_t ret;
3487 
3488         buf = strstrip(buf);
3489         if (!strcmp(buf, "max")) {
3490                 descendants = INT_MAX;
3491         } else {
3492                 ret = kstrtoint(buf, 0, &descendants);
3493                 if (ret)
3494                         return ret;
3495         }
3496 
3497         if (descendants < 0)
3498                 return -ERANGE;
3499 
3500         cgrp = cgroup_kn_lock_live(of->kn, false);
3501         if (!cgrp)
3502                 return -ENOENT;
3503 
3504         cgrp->max_descendants = descendants;
3505 
3506         cgroup_kn_unlock(of->kn);
3507 
3508         return nbytes;
3509 }
3510 
3511 static int cgroup_max_depth_show(struct seq_file *seq, void *v)
3512 {
3513         struct cgroup *cgrp = seq_css(seq)->cgroup;
3514         int depth = READ_ONCE(cgrp->max_depth);
3515 
3516         if (depth == INT_MAX)
3517                 seq_puts(seq, "max\n");
3518         else
3519                 seq_printf(seq, "%d\n", depth);
3520 
3521         return 0;
3522 }
3523 
3524 static ssize_t cgroup_max_depth_write(struct kernfs_open_file *of,
3525                                       char *buf, size_t nbytes, loff_t off)
3526 {
3527         struct cgroup *cgrp;
3528         ssize_t ret;
3529         int depth;
3530 
3531         buf = strstrip(buf);
3532         if (!strcmp(buf, "max")) {
3533                 depth = INT_MAX;
3534         } else {
3535                 ret = kstrtoint(buf, 0, &depth);
3536                 if (ret)
3537                         return ret;
3538         }
3539 
3540         if (depth < 0)
3541                 return -ERANGE;
3542 
3543         cgrp = cgroup_kn_lock_live(of->kn, false);
3544         if (!cgrp)
3545                 return -ENOENT;
3546 
3547         cgrp->max_depth = depth;
3548 
3549         cgroup_kn_unlock(of->kn);
3550 
3551         return nbytes;
3552 }
3553 
3554 static int cgroup_events_show(struct seq_file *seq, void *v)
3555 {
3556         struct cgroup *cgrp = seq_css(seq)->cgroup;
3557 
3558         seq_printf(seq, "populated %d\n", cgroup_is_populated(cgrp));
3559         seq_printf(seq, "frozen %d\n", test_bit(CGRP_FROZEN, &cgrp->flags));
3560 
3561         return 0;
3562 }
3563 
3564 static int cgroup_stat_show(struct seq_file *seq, void *v)
3565 {
3566         struct cgroup *cgroup = seq_css(seq)->cgroup;
3567 
3568         seq_printf(seq, "nr_descendants %d\n",
3569                    cgroup->nr_descendants);
3570         seq_printf(seq, "nr_dying_descendants %d\n",
3571                    cgroup->nr_dying_descendants);
3572 
3573         return 0;
3574 }
3575 
3576 static int __maybe_unused cgroup_extra_stat_show(struct seq_file *seq,
3577                                                  struct cgroup *cgrp, int ssid)
3578 {
3579         struct cgroup_subsys *ss = cgroup_subsys[ssid];
3580         struct cgroup_subsys_state *css;
3581         int ret;
3582 
3583         if (!ss->css_extra_stat_show)
3584                 return 0;
3585 
3586         css = cgroup_tryget_css(cgrp, ss);
3587         if (!css)
3588                 return 0;
3589 
3590         ret = ss->css_extra_stat_show(seq, css);
3591         css_put(css);
3592         return ret;
3593 }
3594 
3595 static int cpu_stat_show(struct seq_file *seq, void *v)
3596 {
3597         struct cgroup __maybe_unused *cgrp = seq_css(seq)->cgroup;
3598         int ret = 0;
3599 
3600         cgroup_base_stat_cputime_show(seq);
3601 #ifdef CONFIG_CGROUP_SCHED
3602         ret = cgroup_extra_stat_show(seq, cgrp, cpu_cgrp_id);
3603 #endif
3604         return ret;
3605 }
3606 
3607 #ifdef CONFIG_PSI
3608 static int cgroup_io_pressure_show(struct seq_file *seq, void *v)
3609 {
3610         struct cgroup *cgrp = seq_css(seq)->cgroup;
3611         struct psi_group *psi = cgroup_ino(cgrp) == 1 ? &psi_system : &cgrp->psi;
3612 
3613         return psi_show(seq, psi, PSI_IO);
3614 }
3615 static int cgroup_memory_pressure_show(struct seq_file *seq, void *v)
3616 {
3617         struct cgroup *cgrp = seq_css(seq)->cgroup;
3618         struct psi_group *psi = cgroup_ino(cgrp) == 1 ? &psi_system : &cgrp->psi;
3619 
3620         return psi_show(seq, psi, PSI_MEM);
3621 }
3622 static int cgroup_cpu_pressure_show(struct seq_file *seq, void *v)
3623 {
3624         struct cgroup *cgrp = seq_css(seq)->cgroup;
3625         struct psi_group *psi = cgroup_ino(cgrp) == 1 ? &psi_system : &cgrp->psi;
3626 
3627         return psi_show(seq, psi, PSI_CPU);
3628 }
3629 
3630 static ssize_t cgroup_pressure_write(struct kernfs_open_file *of, char *buf,
3631                                           size_t nbytes, enum psi_res res)
3632 {
3633         struct psi_trigger *new;
3634         struct cgroup *cgrp;
3635         struct psi_group *psi;
3636 
3637         cgrp = cgroup_kn_lock_live(of->kn, false);
3638         if (!cgrp)
3639                 return -ENODEV;
3640 
3641         cgroup_get(cgrp);
3642         cgroup_kn_unlock(of->kn);
3643 
3644         psi = cgroup_ino(cgrp) == 1 ? &psi_system : &cgrp->psi;
3645         new = psi_trigger_create(psi, buf, nbytes, res);
3646         if (IS_ERR(new)) {
3647                 cgroup_put(cgrp);
3648                 return PTR_ERR(new);
3649         }
3650 
3651         psi_trigger_replace(&of->priv, new);
3652 
3653         cgroup_put(cgrp);
3654 
3655         return nbytes;
3656 }
3657 
3658 static ssize_t cgroup_io_pressure_write(struct kernfs_open_file *of,
3659                                           char *buf, size_t nbytes,
3660                                           loff_t off)
3661 {
3662         return cgroup_pressure_write(of, buf, nbytes, PSI_IO);
3663 }
3664 
3665 static ssize_t cgroup_memory_pressure_write(struct kernfs_open_file *of,
3666                                           char *buf, size_t nbytes,
3667                                           loff_t off)
3668 {
3669         return cgroup_pressure_write(of, buf, nbytes, PSI_MEM);
3670 }
3671 
3672 static ssize_t cgroup_cpu_pressure_write(struct kernfs_open_file *of,
3673                                           char *buf, size_t nbytes,
3674                                           loff_t off)
3675 {
3676         return cgroup_pressure_write(of, buf, nbytes, PSI_CPU);
3677 }
3678 
3679 static __poll_t cgroup_pressure_poll(struct kernfs_open_file *of,
3680                                           poll_table *pt)
3681 {
3682         return psi_trigger_poll(&of->priv, of->file, pt);
3683 }
3684 
3685 static void cgroup_pressure_release(struct kernfs_open_file *of)
3686 {
3687         psi_trigger_replace(&of->priv, NULL);
3688 }
3689 
3690 bool cgroup_psi_enabled(void)
3691 {
3692         return (cgroup_feature_disable_mask & (1 << OPT_FEATURE_PRESSURE)) == 0;
3693 }
3694 
3695 #else /* CONFIG_PSI */
3696 bool cgroup_psi_enabled(void)
3697 {
3698         return false;
3699 }
3700 
3701 #endif /* CONFIG_PSI */
3702 
3703 static int cgroup_freeze_show(struct seq_file *seq, void *v)
3704 {
3705         struct cgroup *cgrp = seq_css(seq)->cgroup;
3706 
3707         seq_printf(seq, "%d\n", cgrp->freezer.freeze);
3708 
3709         return 0;
3710 }
3711 
3712 static ssize_t cgroup_freeze_write(struct kernfs_open_file *of,
3713                                    char *buf, size_t nbytes, loff_t off)
3714 {
3715         struct cgroup *cgrp;
3716         ssize_t ret;
3717         int freeze;
3718 
3719         ret = kstrtoint(strstrip(buf), 0, &freeze);
3720         if (ret)
3721                 return ret;
3722 
3723         if (freeze < 0 || freeze > 1)
3724                 return -ERANGE;
3725 
3726         cgrp = cgroup_kn_lock_live(of->kn, false);
3727         if (!cgrp)
3728                 return -ENOENT;
3729 
3730         cgroup_freeze(cgrp, freeze);
3731 
3732         cgroup_kn_unlock(of->kn);
3733 
3734         return nbytes;
3735 }
3736 
3737 static void __cgroup_kill(struct cgroup *cgrp)
3738 {
3739         struct css_task_iter it;
3740         struct task_struct *task;
3741 
3742         lockdep_assert_held(&cgroup_mutex);
3743 
3744         spin_lock_irq(&css_set_lock);
3745         set_bit(CGRP_KILL, &cgrp->flags);
3746         spin_unlock_irq(&css_set_lock);
3747 
3748         css_task_iter_start(&cgrp->self, CSS_TASK_ITER_PROCS | CSS_TASK_ITER_THREADED, &it);
3749         while ((task = css_task_iter_next(&it))) {
3750                 /* Ignore kernel threads here. */
3751                 if (task->flags & PF_KTHREAD)
3752                         continue;
3753 
3754                 /* Skip tasks that are already dying. */
3755                 if (__fatal_signal_pending(task))
3756                         continue;
3757 
3758                 send_sig(SIGKILL, task, 0);
3759         }
3760         css_task_iter_end(&it);
3761 
3762         spin_lock_irq(&css_set_lock);
3763         clear_bit(CGRP_KILL, &cgrp->flags);
3764         spin_unlock_irq(&css_set_lock);
3765 }
3766 
3767 static void cgroup_kill(struct cgroup *cgrp)
3768 {
3769         struct cgroup_subsys_state *css;
3770         struct cgroup *dsct;
3771 
3772         lockdep_assert_held(&cgroup_mutex);
3773 
3774         cgroup_for_each_live_descendant_pre(dsct, css, cgrp)
3775                 __cgroup_kill(dsct);
3776 }
3777 
3778 static ssize_t cgroup_kill_write(struct kernfs_open_file *of, char *buf,
3779                                  size_t nbytes, loff_t off)
3780 {
3781         ssize_t ret = 0;
3782         int kill;
3783         struct cgroup *cgrp;
3784 
3785         ret = kstrtoint(strstrip(buf), 0, &kill);
3786         if (ret)
3787                 return ret;
3788 
3789         if (kill != 1)
3790                 return -ERANGE;
3791 
3792         cgrp = cgroup_kn_lock_live(of->kn, false);
3793         if (!cgrp)
3794                 return -ENOENT;
3795 
3796         /*
3797          * Killing is a process directed operation, i.e. the whole thread-group
3798          * is taken down so act like we do for cgroup.procs and only make this
3799          * writable in non-threaded cgroups.
3800          */
3801         if (cgroup_is_threaded(cgrp))
3802                 ret = -EOPNOTSUPP;
3803         else
3804                 cgroup_kill(cgrp);
3805 
3806         cgroup_kn_unlock(of->kn);
3807 
3808         return ret ?: nbytes;
3809 }
3810 
3811 static int cgroup_file_open(struct kernfs_open_file *of)
3812 {
3813         struct cftype *cft = of_cft(of);
3814 
3815         if (cft->open)
3816                 return cft->open(of);
3817         return 0;
3818 }
3819 
3820 static void cgroup_file_release(struct kernfs_open_file *of)
3821 {
3822         struct cftype *cft = of_cft(of);
3823 
3824         if (cft->release)
3825                 cft->release(of);
3826 }
3827 
3828 static ssize_t cgroup_file_write(struct kernfs_open_file *of, char *buf,
3829                                  size_t nbytes, loff_t off)
3830 {
3831         struct cgroup_namespace *ns = current->nsproxy->cgroup_ns;
3832         struct cgroup *cgrp = of->kn->parent->priv;
3833         struct cftype *cft = of_cft(of);
3834         struct cgroup_subsys_state *css;
3835         int ret;
3836 
3837         if (!nbytes)
3838                 return 0;
3839 
3840         /*
3841          * If namespaces are delegation boundaries, disallow writes to
3842          * files in an non-init namespace root from inside the namespace
3843          * except for the files explicitly marked delegatable -
3844          * cgroup.procs and cgroup.subtree_control.
3845          */
3846         if ((cgrp->root->flags & CGRP_ROOT_NS_DELEGATE) &&
3847             !(cft->flags & CFTYPE_NS_DELEGATABLE) &&
3848             ns != &init_cgroup_ns && ns->root_cset->dfl_cgrp == cgrp)
3849                 return -EPERM;
3850 
3851         if (cft->write)
3852                 return cft->write(of, buf, nbytes, off);
3853 
3854         /*
3855          * kernfs guarantees that a file isn't deleted with operations in
3856          * flight, which means that the matching css is and stays alive and
3857          * doesn't need to be pinned.  The RCU locking is not necessary
3858          * either.  It's just for the convenience of using cgroup_css().
3859          */
3860         rcu_read_lock();
3861         css = cgroup_css(cgrp, cft->ss);
3862         rcu_read_unlock();
3863 
3864         if (cft->write_u64) {
3865                 unsigned long long v;
3866                 ret = kstrtoull(buf, 0, &v);
3867                 if (!ret)
3868                         ret = cft->write_u64(css, cft, v);
3869         } else if (cft->write_s64) {
3870                 long long v;
3871                 ret = kstrtoll(buf, 0, &v);
3872                 if (!ret)
3873                         ret = cft->write_s64(css, cft, v);
3874         } else {
3875                 ret = -EINVAL;
3876         }
3877 
3878         return ret ?: nbytes;
3879 }
3880 
3881 static __poll_t cgroup_file_poll(struct kernfs_open_file *of, poll_table *pt)
3882 {
3883         struct cftype *cft = of_cft(of);
3884 
3885         if (cft->poll)
3886                 return cft->poll(of, pt);
3887 
3888         return kernfs_generic_poll(of, pt);
3889 }
3890 
3891 static void *cgroup_seqfile_start(struct seq_file *seq, loff_t *ppos)
3892 {
3893         return seq_cft(seq)->seq_start(seq, ppos);
3894 }
3895 
3896 static void *cgroup_seqfile_next(struct seq_file *seq, void *v, loff_t *ppos)
3897 {
3898         return seq_cft(seq)->seq_next(seq, v, ppos);
3899 }
3900 
3901 static void cgroup_seqfile_stop(struct seq_file *seq, void *v)
3902 {
3903         if (seq_cft(seq)->seq_stop)
3904                 seq_cft(seq)->seq_stop(seq, v);
3905 }
3906 
3907 static int cgroup_seqfile_show(struct seq_file *m, void *arg)
3908 {
3909         struct cftype *cft = seq_cft(m);
3910         struct cgroup_subsys_state *css = seq_css(m);
3911 
3912         if (cft->seq_show)
3913                 return cft->seq_show(m, arg);
3914 
3915         if (cft->read_u64)
3916                 seq_printf(m, "%llu\n", cft->read_u64(css, cft));
3917         else if (cft->read_s64)
3918                 seq_printf(m, "%lld\n", cft->read_s64(css, cft));
3919         else
3920                 return -EINVAL;
3921         return 0;
3922 }
3923 
3924 static struct kernfs_ops cgroup_kf_single_ops = {
3925         .atomic_write_len       = PAGE_SIZE,
3926         .open                   = cgroup_file_open,
3927         .release                = cgroup_file_release,
3928         .write                  = cgroup_file_write,
3929         .poll                   = cgroup_file_poll,
3930         .seq_show               = cgroup_seqfile_show,
3931 };
3932 
3933 static struct kernfs_ops cgroup_kf_ops = {
3934         .atomic_write_len       = PAGE_SIZE,
3935         .open                   = cgroup_file_open,
3936         .release                = cgroup_file_release,
3937         .write                  = cgroup_file_write,
3938         .poll                   = cgroup_file_poll,
3939         .seq_start              = cgroup_seqfile_start,
3940         .seq_next               = cgroup_seqfile_next,
3941         .seq_stop               = cgroup_seqfile_stop,
3942         .seq_show               = cgroup_seqfile_show,
3943 };
3944 
3945 /* set uid and gid of cgroup dirs and files to that of the creator */
3946 static int cgroup_kn_set_ugid(struct kernfs_node *kn)
3947 {
3948         struct iattr iattr = { .ia_valid = ATTR_UID | ATTR_GID,
3949                                .ia_uid = current_fsuid(),
3950                                .ia_gid = current_fsgid(), };
3951 
3952         if (uid_eq(iattr.ia_uid, GLOBAL_ROOT_UID) &&
3953             gid_eq(iattr.ia_gid, GLOBAL_ROOT_GID))
3954                 return 0;
3955 
3956         return kernfs_setattr(kn, &iattr);
3957 }
3958 
3959 static void cgroup_file_notify_timer(struct timer_list *timer)
3960 {
3961         cgroup_file_notify(container_of(timer, struct cgroup_file,
3962                                         notify_timer));
3963 }
3964 
3965 static int cgroup_add_file(struct cgroup_subsys_state *css, struct cgroup *cgrp,
3966                            struct cftype *cft)
3967 {
3968         char name[CGROUP_FILE_NAME_MAX];
3969         struct kernfs_node *kn;
3970         struct lock_class_key *key = NULL;
3971         int ret;
3972 
3973 #ifdef CONFIG_DEBUG_LOCK_ALLOC
3974         key = &cft->lockdep_key;
3975 #endif
3976         kn = __kernfs_create_file(cgrp->kn, cgroup_file_name(cgrp, cft, name),
3977                                   cgroup_file_mode(cft),
3978                                   GLOBAL_ROOT_UID, GLOBAL_ROOT_GID,
3979                                   0, cft->kf_ops, cft,
3980                                   NULL, key);
3981         if (IS_ERR(kn))
3982                 return PTR_ERR(kn);
3983 
3984         ret = cgroup_kn_set_ugid(kn);
3985         if (ret) {
3986                 kernfs_remove(kn);
3987                 return ret;
3988         }
3989 
3990         if (cft->file_offset) {
3991                 struct cgroup_file *cfile = (void *)css + cft->file_offset;
3992 
3993                 timer_setup(&cfile->notify_timer, cgroup_file_notify_timer, 0);
3994 
3995                 spin_lock_irq(&cgroup_file_kn_lock);
3996                 cfile->kn = kn;
3997                 spin_unlock_irq(&cgroup_file_kn_lock);
3998         }
3999 
4000         return 0;
4001 }
4002 
4003 /**
4004  * cgroup_addrm_files - add or remove files to a cgroup directory
4005  * @css: the target css
4006  * @cgrp: the target cgroup (usually css->cgroup)
4007  * @cfts: array of cftypes to be added
4008  * @is_add: whether to add or remove
4009  *
4010  * Depending on @is_add, add or remove files defined by @cfts on @cgrp.
4011  * For removals, this function never fails.
4012  */
4013 static int cgroup_addrm_files(struct cgroup_subsys_state *css,
4014                               struct cgroup *cgrp, struct cftype cfts[],
4015                               bool is_add)
4016 {
4017         struct cftype *cft, *cft_end = NULL;
4018         int ret = 0;
4019 
4020         lockdep_assert_held(&cgroup_mutex);
4021 
4022 restart:
4023         for (cft = cfts; cft != cft_end && cft->name[0] != '\0'; cft++) {
4024                 /* does cft->flags tell us to skip this file on @cgrp? */
4025                 if ((cft->flags & CFTYPE_PRESSURE) && !cgroup_psi_enabled())
4026                         continue;
4027                 if ((cft->flags & __CFTYPE_ONLY_ON_DFL) && !cgroup_on_dfl(cgrp))
4028                         continue;
4029                 if ((cft->flags & __CFTYPE_NOT_ON_DFL) && cgroup_on_dfl(cgrp))
4030                         continue;
4031                 if ((cft->flags & CFTYPE_NOT_ON_ROOT) && !cgroup_parent(cgrp))
4032                         continue;
4033                 if ((cft->flags & CFTYPE_ONLY_ON_ROOT) && cgroup_parent(cgrp))
4034                         continue;
4035                 if ((cft->flags & CFTYPE_DEBUG) && !cgroup_debug)
4036                         continue;
4037                 if (is_add) {
4038                         ret = cgroup_add_file(css, cgrp, cft);
4039                         if (ret) {
4040                                 pr_warn("%s: failed to add %s, err=%d\n",
4041                                         __func__, cft->name, ret);
4042                                 cft_end = cft;
4043                                 is_add = false;
4044                                 goto restart;
4045                         }
4046                 } else {
4047                         cgroup_rm_file(cgrp, cft);
4048                 }
4049         }
4050         return ret;
4051 }
4052 
4053 static int cgroup_apply_cftypes(struct cftype *cfts, bool is_add)
4054 {
4055         struct cgroup_subsys *ss = cfts[0].ss;
4056         struct cgroup *root = &ss->root->cgrp;
4057         struct cgroup_subsys_state *css;
4058         int ret = 0;
4059 
4060         lockdep_assert_held(&cgroup_mutex);
4061 
4062         /* add/rm files for all cgroups created before */
4063         css_for_each_descendant_pre(css, cgroup_css(root, ss)) {
4064                 struct cgroup *cgrp = css->cgroup;
4065 
4066                 if (!(css->flags & CSS_VISIBLE))
4067                         continue;
4068 
4069                 ret = cgroup_addrm_files(css, cgrp, cfts, is_add);
4070                 if (ret)
4071                         break;
4072         }
4073 
4074         if (is_add && !ret)
4075                 kernfs_activate(root->kn);
4076         return ret;
4077 }
4078 
4079 static void cgroup_exit_cftypes(struct cftype *cfts)
4080 {
4081         struct cftype *cft;
4082 
4083         for (cft = cfts; cft->name[0] != '\0'; cft++) {
4084                 /* free copy for custom atomic_write_len, see init_cftypes() */
4085                 if (cft->max_write_len && cft->max_write_len != PAGE_SIZE)
4086                         kfree(cft->kf_ops);
4087                 cft->kf_ops = NULL;
4088                 cft->ss = NULL;
4089 
4090                 /* revert flags set by cgroup core while adding @cfts */
4091                 cft->flags &= ~(__CFTYPE_ONLY_ON_DFL | __CFTYPE_NOT_ON_DFL);
4092         }
4093 }
4094 
4095 static int cgroup_init_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
4096 {
4097         struct cftype *cft;
4098 
4099         for (cft = cfts; cft->name[0] != '\0'; cft++) {
4100                 struct kernfs_ops *kf_ops;
4101 
4102                 WARN_ON(cft->ss || cft->kf_ops);
4103 
4104                 if ((cft->flags & CFTYPE_PRESSURE) && !cgroup_psi_enabled())
4105                         continue;
4106 
4107                 if (cft->seq_start)
4108                         kf_ops = &cgroup_kf_ops;
4109                 else
4110                         kf_ops = &cgroup_kf_single_ops;
4111 
4112                 /*
4113                  * Ugh... if @cft wants a custom max_write_len, we need to
4114                  * make a copy of kf_ops to set its atomic_write_len.
4115                  */
4116                 if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) {
4117                         kf_ops = kmemdup(kf_ops, sizeof(*kf_ops), GFP_KERNEL);
4118                         if (!kf_ops) {
4119                                 cgroup_exit_cftypes(cfts);
4120                                 return -ENOMEM;
4121                         }
4122                         kf_ops->atomic_write_len = cft->max_write_len;
4123                 }
4124 
4125                 cft->kf_ops = kf_ops;
4126                 cft->ss = ss;
4127         }
4128 
4129         return 0;
4130 }
4131 
4132 static int cgroup_rm_cftypes_locked(struct cftype *cfts)
4133 {
4134         lockdep_assert_held(&cgroup_mutex);
4135 
4136         if (!cfts || !cfts[0].ss)
4137                 return -ENOENT;
4138 
4139         list_del(&cfts->node);
4140         cgroup_apply_cftypes(cfts, false);
4141         cgroup_exit_cftypes(cfts);
4142         return 0;
4143 }
4144 
4145 /**
4146  * cgroup_rm_cftypes - remove an array of cftypes from a subsystem
4147  * @cfts: zero-length name terminated array of cftypes
4148  *
4149  * Unregister @cfts.  Files described by @cfts are removed from all
4150  * existing cgroups and all future cgroups won't have them either.  This
4151  * function can be called anytime whether @cfts' subsys is attached or not.
4152  *
4153  * Returns 0 on successful unregistration, -ENOENT if @cfts is not
4154  * registered.
4155  */
4156 int cgroup_rm_cftypes(struct cftype *cfts)
4157 {
4158         int ret;
4159 
4160         mutex_lock(&cgroup_mutex);
4161         ret = cgroup_rm_cftypes_locked(cfts);
4162         mutex_unlock(&cgroup_mutex);
4163         return ret;
4164 }
4165 
4166 /**
4167  * cgroup_add_cftypes - add an array of cftypes to a subsystem
4168  * @ss: target cgroup subsystem
4169  * @cfts: zero-length name terminated array of cftypes
4170  *
4171  * Register @cfts to @ss.  Files described by @cfts are created for all
4172  * existing cgroups to which @ss is attached and all future cgroups will
4173  * have them too.  This function can be called anytime whether @ss is
4174  * attached or not.
4175  *
4176  * Returns 0 on successful registration, -errno on failure.  Note that this
4177  * function currently returns 0 as long as @cfts registration is successful
4178  * even if some file creation attempts on existing cgroups fail.
4179  */
4180 static int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
4181 {
4182         int ret;
4183 
4184         if (!cgroup_ssid_enabled(ss->id))
4185                 return 0;
4186 
4187         if (!cfts || cfts[0].name[0] == '\0')
4188                 return 0;
4189 
4190         ret = cgroup_init_cftypes(ss, cfts);
4191         if (ret)
4192                 return ret;
4193 
4194         mutex_lock(&cgroup_mutex);
4195 
4196         list_add_tail(&cfts->node, &ss->cfts);
4197         ret = cgroup_apply_cftypes(cfts, true);
4198         if (ret)
4199                 cgroup_rm_cftypes_locked(cfts);
4200 
4201         mutex_unlock(&cgroup_mutex);
4202         return ret;
4203 }
4204 
4205 /**
4206  * cgroup_add_dfl_cftypes - add an array of cftypes for default hierarchy
4207  * @ss: target cgroup subsystem
4208  * @cfts: zero-length name terminated array of cftypes
4209  *
4210  * Similar to cgroup_add_cftypes() but the added files are only used for
4211  * the default hierarchy.
4212  */
4213 int cgroup_add_dfl_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
4214 {
4215         struct cftype *cft;
4216 
4217         for (cft = cfts; cft && cft->name[0] != '\0'; cft++)
4218                 cft->flags |= __CFTYPE_ONLY_ON_DFL;
4219         return cgroup_add_cftypes(ss, cfts);
4220 }
4221 
4222 /**
4223  * cgroup_add_legacy_cftypes - add an array of cftypes for legacy hierarchies
4224  * @ss: target cgroup subsystem
4225  * @cfts: zero-length name terminated array of cftypes
4226  *
4227  * Similar to cgroup_add_cftypes() but the added files are only used for
4228  * the legacy hierarchies.
4229  */
4230 int cgroup_add_legacy_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
4231 {
4232         struct cftype *cft;
4233 
4234         for (cft = cfts; cft && cft->name[0] != '\0'; cft++)
4235                 cft->flags |= __CFTYPE_NOT_ON_DFL;
4236         return cgroup_add_cftypes(ss, cfts);
4237 }
4238 
4239 /**
4240  * cgroup_file_notify - generate a file modified event for a cgroup_file
4241  * @cfile: target cgroup_file
4242  *
4243  * @cfile must have been obtained by setting cftype->file_offset.
4244  */
4245 void cgroup_file_notify(struct cgroup_file *cfile)
4246 {
4247         unsigned long flags;
4248 
4249         spin_lock_irqsave(&cgroup_file_kn_lock, flags);
4250         if (cfile->kn) {
4251                 unsigned long last = cfile->notified_at;
4252                 unsigned long next = last + CGROUP_FILE_NOTIFY_MIN_INTV;
4253 
4254                 if (time_in_range(jiffies, last, next)) {
4255                         timer_reduce(&cfile->notify_timer, next);
4256                 } else {
4257                         kernfs_notify(cfile->kn);
4258                         cfile->notified_at = jiffies;
4259                 }
4260         }
4261         spin_unlock_irqrestore(&cgroup_file_kn_lock, flags);
4262 }
4263 
4264 /**
4265  * css_next_child - find the next child of a given css
4266  * @pos: the current position (%NULL to initiate traversal)
4267  * @parent: css whose children to walk
4268  *
4269  * This function returns the next child of @parent and should be called
4270  * under either cgroup_mutex or RCU read lock.  The only requirement is
4271  * that @parent and @pos are accessible.  The next sibling is guaranteed to
4272  * be returned regardless of their states.
4273  *
4274  * If a subsystem synchronizes ->css_online() and the start of iteration, a
4275  * css which finished ->css_online() is guaranteed to be visible in the
4276  * future iterations and will stay visible until the last reference is put.
4277  * A css which hasn't finished ->css_online() or already finished
4278  * ->css_offline() may show up during traversal.  It's each subsystem's
4279  * responsibility to synchronize against on/offlining.
4280  */
4281 struct cgroup_subsys_state *css_next_child(struct cgroup_subsys_state *pos,
4282                                            struct cgroup_subsys_state *parent)
4283 {
4284         struct cgroup_subsys_state *next;
4285 
4286         cgroup_assert_mutex_or_rcu_locked();
4287 
4288         /*
4289          * @pos could already have been unlinked from the sibling list.
4290          * Once a cgroup is removed, its ->sibling.next is no longer
4291          * updated when its next sibling changes.  CSS_RELEASED is set when
4292          * @pos is taken off list, at which time its next pointer is valid,
4293          * and, as releases are serialized, the one pointed to by the next
4294          * pointer is guaranteed to not have started release yet.  This
4295          * implies that if we observe !CSS_RELEASED on @pos in this RCU
4296          * critical section, the one pointed to by its next pointer is
4297          * guaranteed to not have finished its RCU grace period even if we
4298          * have dropped rcu_read_lock() in-between iterations.
4299          *
4300          * If @pos has CSS_RELEASED set, its next pointer can't be
4301          * dereferenced; however, as each css is given a monotonically
4302          * increasing unique serial number and always appended to the
4303          * sibling list, the next one can be found by walking the parent's
4304          * children until the first css with higher serial number than
4305          * @pos's.  While this path can be slower, it happens iff iteration
4306          * races against release and the race window is very small.
4307          */
4308         if (!pos) {
4309                 next = list_entry_rcu(parent->children.next, struct cgroup_subsys_state, sibling);
4310         } else if (likely(!(pos->flags & CSS_RELEASED))) {
4311                 next = list_entry_rcu(pos->sibling.next, struct cgroup_subsys_state, sibling);
4312         } else {
4313                 list_for_each_entry_rcu(next, &parent->children, sibling,
4314                                         lockdep_is_held(&cgroup_mutex))
4315                         if (next->serial_nr > pos->serial_nr)
4316                                 break;
4317         }
4318 
4319         /*
4320          * @next, if not pointing to the head, can be dereferenced and is
4321          * the next sibling.
4322          */
4323         if (&next->sibling != &parent->children)
4324                 return next;
4325         return NULL;
4326 }
4327 
4328 /**
4329  * css_next_descendant_pre - find the next descendant for pre-order walk
4330  * @pos: the current position (%NULL to initiate traversal)
4331  * @root: css whose descendants to walk
4332  *
4333  * To be used by css_for_each_descendant_pre().  Find the next descendant
4334  * to visit for pre-order traversal of @root's descendants.  @root is
4335  * included in the iteration and the first node to be visited.
4336  *
4337  * While this function requires cgroup_mutex or RCU read locking, it
4338  * doesn't require the whole traversal to be contained in a single critical
4339  * section.  This function will return the correct next descendant as long
4340  * as both @pos and @root are accessible and @pos is a descendant of @root.
4341  *
4342  * If a subsystem synchronizes ->css_online() and the start of iteration, a
4343  * css which finished ->css_online() is guaranteed to be visible in the
4344  * future iterations and will stay visible until the last reference is put.
4345  * A css which hasn't finished ->css_online() or already finished
4346  * ->css_offline() may show up during traversal.  It's each subsystem's
4347  * responsibility to synchronize against on/offlining.
4348  */
4349 struct cgroup_subsys_state *
4350 css_next_descendant_pre(struct cgroup_subsys_state *pos,
4351                         struct cgroup_subsys_state *root)
4352 {
4353         struct cgroup_subsys_state *next;
4354 
4355         cgroup_assert_mutex_or_rcu_locked();
4356 
4357         /* if first iteration, visit @root */
4358         if (!pos)
4359                 return root;
4360 
4361         /* visit the first child if exists */
4362         next = css_next_child(NULL, pos);
4363         if (next)
4364                 return next;
4365 
4366         /* no child, visit my or the closest ancestor's next sibling */
4367         while (pos != root) {
4368                 next = css_next_child(pos, pos->parent);
4369                 if (next)
4370                         return next;
4371                 pos = pos->parent;
4372         }
4373 
4374         return NULL;
4375 }
4376 EXPORT_SYMBOL_GPL(css_next_descendant_pre);
4377 
4378 /**
4379  * css_rightmost_descendant - return the rightmost descendant of a css
4380  * @pos: css of interest
4381  *
4382  * Return the rightmost descendant of @pos.  If there's no descendant, @pos
4383  * is returned.  This can be used during pre-order traversal to skip
4384  * subtree of @pos.
4385  *
4386  * While this function requires cgroup_mutex or RCU read locking, it
4387  * doesn't require the whole traversal to be contained in a single critical
4388  * section.  This function will return the correct rightmost descendant as
4389  * long as @pos is accessible.
4390  */
4391 struct cgroup_subsys_state *
4392 css_rightmost_descendant(struct cgroup_subsys_state *pos)
4393 {
4394         struct cgroup_subsys_state *last, *tmp;
4395 
4396         cgroup_assert_mutex_or_rcu_locked();
4397 
4398         do {
4399                 last = pos;
4400                 /* ->prev isn't RCU safe, walk ->next till the end */
4401                 pos = NULL;
4402                 css_for_each_child(tmp, last)
4403                         pos = tmp;
4404         } while (pos);
4405 
4406         return last;
4407 }
4408 
4409 static struct cgroup_subsys_state *
4410 css_leftmost_descendant(struct cgroup_subsys_state *pos)
4411 {
4412         struct cgroup_subsys_state *last;
4413 
4414         do {
4415                 last = pos;
4416                 pos = css_next_child(NULL, pos);
4417         } while (pos);
4418 
4419         return last;
4420 }
4421 
4422 /**
4423  * css_next_descendant_post - find the next descendant for post-order walk
4424  * @pos: the current position (%NULL to initiate traversal)
4425  * @root: css whose descendants to walk
4426  *
4427  * To be used by css_for_each_descendant_post().  Find the next descendant
4428  * to visit for post-order traversal of @root's descendants.  @root is
4429  * included in the iteration and the last node to be visited.
4430  *
4431  * While this function requires cgroup_mutex or RCU read locking, it
4432  * doesn't require the whole traversal to be contained in a single critical
4433  * section.  This function will return the correct next descendant as long
4434  * as both @pos and @cgroup are accessible and @pos is a descendant of
4435  * @cgroup.
4436  *
4437  * If a subsystem synchronizes ->css_online() and the start of iteration, a
4438  * css which finished ->css_online() is guaranteed to be visible in the
4439  * future iterations and will stay visible until the last reference is put.
4440  * A css which hasn't finished ->css_online() or already finished
4441  * ->css_offline() may show up during traversal.  It's each subsystem's
4442  * responsibility to synchronize against on/offlining.
4443  */
4444 struct cgroup_subsys_state *
4445 css_next_descendant_post(struct cgroup_subsys_state *pos,
4446                          struct cgroup_subsys_state *root)
4447 {
4448         struct cgroup_subsys_state *next;
4449 
4450         cgroup_assert_mutex_or_rcu_locked();
4451 
4452         /* if first iteration, visit leftmost descendant which may be @root */
4453         if (!pos)
4454                 return css_leftmost_descendant(root);
4455 
4456         /* if we visited @root, we're done */
4457         if (pos == root)
4458                 return NULL;
4459 
4460         /* if there's an unvisited sibling, visit its leftmost descendant */
4461         next = css_next_child(pos, pos->parent);
4462         if (next)
4463                 return css_leftmost_descendant(next);
4464 
4465         /* no sibling left, visit parent */
4466         return pos->parent;
4467 }
4468 
4469 /**
4470  * css_has_online_children - does a css have online children
4471  * @css: the target css
4472  *
4473  * Returns %true if @css has any online children; otherwise, %false.  This
4474  * function can be called from any context but the caller is responsible
4475  * for synchronizing against on/offlining as necessary.
4476  */
4477 bool css_has_online_children(struct cgroup_subsys_state *css)
4478 {
4479         struct cgroup_subsys_state *child;
4480         bool ret = false;
4481 
4482         rcu_read_lock();
4483         css_for_each_child(child, css) {
4484                 if (child->flags & CSS_ONLINE) {
4485                         ret = true;
4486                         break;
4487                 }
4488         }
4489         rcu_read_unlock();
4490         return ret;
4491 }
4492 
4493 static struct css_set *css_task_iter_next_css_set(struct css_task_iter *it)
4494 {
4495         struct list_head *l;
4496         struct cgrp_cset_link *link;
4497         struct css_set *cset;
4498 
4499         lockdep_assert_held(&css_set_lock);
4500 
4501         /* find the next threaded cset */
4502         if (it->tcset_pos) {
4503                 l = it->tcset_pos->next;
4504 
4505                 if (l != it->tcset_head) {
4506                         it->tcset_pos = l;
4507                         return container_of(l, struct css_set,
4508                                             threaded_csets_node);
4509                 }
4510 
4511                 it->tcset_pos = NULL;
4512         }
4513 
4514         /* find the next cset */
4515         l = it->cset_pos;
4516         l = l->next;
4517         if (l == it->cset_head) {
4518                 it->cset_pos = NULL;
4519                 return NULL;
4520         }
4521 
4522         if (it->ss) {
4523                 cset = container_of(l, struct css_set, e_cset_node[it->ss->id]);
4524         } else {
4525                 link = list_entry(l, struct cgrp_cset_link, cset_link);
4526                 cset = link->cset;
4527         }
4528 
4529         it->cset_pos = l;
4530 
4531         /* initialize threaded css_set walking */
4532         if (it->flags & CSS_TASK_ITER_THREADED) {
4533                 if (it->cur_dcset)
4534                         put_css_set_locked(it->cur_dcset);
4535                 it->cur_dcset = cset;
4536                 get_css_set(cset);
4537 
4538                 it->tcset_head = &cset->threaded_csets;
4539                 it->tcset_pos = &cset->threaded_csets;
4540         }
4541 
4542         return cset;
4543 }
4544 
4545 /**
4546  * css_task_iter_advance_css_set - advance a task iterator to the next css_set
4547  * @it: the iterator to advance
4548  *
4549  * Advance @it to the next css_set to walk.
4550  */
4551 static void css_task_iter_advance_css_set(struct css_task_iter *it)
4552 {
4553         struct css_set *cset;
4554 
4555         lockdep_assert_held(&css_set_lock);
4556 
4557         /* Advance to the next non-empty css_set and find first non-empty tasks list*/
4558         while ((cset = css_task_iter_next_css_set(it))) {
4559                 if (!list_empty(&cset->tasks)) {
4560                         it->cur_tasks_head = &cset->tasks;
4561                         break;
4562                 } else if (!list_empty(&cset->mg_tasks)) {
4563                         it->cur_tasks_head = &cset->mg_tasks;
4564                         break;
4565                 } else if (!list_empty(&cset->dying_tasks)) {
4566                         it->cur_tasks_head = &cset->dying_tasks;
4567                         break;
4568                 }
4569         }
4570         if (!cset) {
4571                 it->task_pos = NULL;
4572                 return;
4573         }
4574         it->task_pos = it->cur_tasks_head->next;
4575 
4576         /*
4577          * We don't keep css_sets locked across iteration steps and thus
4578          * need to take steps to ensure that iteration can be resumed after
4579          * the lock is re-acquired.  Iteration is performed at two levels -
4580          * css_sets and tasks in them.
4581          *
4582          * Once created, a css_set never leaves its cgroup lists, so a
4583          * pinned css_set is guaranteed to stay put and we can resume
4584          * iteration afterwards.
4585          *
4586          * Tasks may leave @cset across iteration steps.  This is resolved
4587          * by registering each iterator with the css_set currently being
4588          * walked and making css_set_move_task() advance iterators whose
4589          * next task is leaving.
4590          */
4591         if (it->cur_cset) {
4592                 list_del(&it->iters_node);
4593                 put_css_set_locked(it->cur_cset);
4594         }
4595         get_css_set(cset);
4596         it->cur_cset = cset;
4597         list_add(&it->iters_node, &cset->task_iters);
4598 }
4599 
4600 static void css_task_iter_skip(struct css_task_iter *it,
4601                                struct task_struct *task)
4602 {
4603         lockdep_assert_held(&css_set_lock);
4604 
4605         if (it->task_pos == &task->cg_list) {
4606                 it->task_pos = it->task_pos->next;
4607                 it->flags |= CSS_TASK_ITER_SKIPPED;
4608         }
4609 }
4610 
4611 static void css_task_iter_advance(struct css_task_iter *it)
4612 {
4613         struct task_struct *task;
4614 
4615         lockdep_assert_held(&css_set_lock);
4616 repeat:
4617         if (it->task_pos) {
4618                 /*
4619                  * Advance iterator to find next entry. We go through cset
4620                  * tasks, mg_tasks and dying_tasks, when consumed we move onto
4621                  * the next cset.
4622                  */
4623                 if (it->flags & CSS_TASK_ITER_SKIPPED)
4624                         it->flags &= ~CSS_TASK_ITER_SKIPPED;
4625                 else
4626                         it->task_pos = it->task_pos->next;
4627 
4628                 if (it->task_pos == &it->cur_cset->tasks) {
4629                         it->cur_tasks_head = &it->cur_cset->mg_tasks;
4630                         it->task_pos = it->cur_tasks_head->next;
4631                 }
4632                 if (it->task_pos == &it->cur_cset->mg_tasks) {
4633                         it->cur_tasks_head = &it->cur_cset->dying_tasks;
4634                         it->task_pos = it->cur_tasks_head->next;
4635                 }
4636                 if (it->task_pos == &it->cur_cset->dying_tasks)
4637                         css_task_iter_advance_css_set(it);
4638         } else {
4639                 /* called from start, proceed to the first cset */
4640                 css_task_iter_advance_css_set(it);
4641         }
4642 
4643         if (!it->task_pos)
4644                 return;
4645 
4646         task = list_entry(it->task_pos, struct task_struct, cg_list);
4647 
4648         if (it->flags & CSS_TASK_ITER_PROCS) {
4649                 /* if PROCS, skip over tasks which aren't group leaders */
4650                 if (!thread_group_leader(task))
4651                         goto repeat;
4652 
4653                 /* and dying leaders w/o live member threads */
4654                 if (it->cur_tasks_head == &it->cur_cset->dying_tasks &&
4655                     !atomic_read(&task->signal->live))
4656                         goto repeat;
4657         } else {
4658                 /* skip all dying ones */
4659                 if (it->cur_tasks_head == &it->cur_cset->dying_tasks)
4660                         goto repeat;
4661         }
4662 }
4663 
4664 /**
4665  * css_task_iter_start - initiate task iteration
4666  * @css: the css to walk tasks of
4667  * @flags: CSS_TASK_ITER_* flags
4668  * @it: the task iterator to use
4669  *
4670  * Initiate iteration through the tasks of @css.  The caller can call
4671  * css_task_iter_next() to walk through the tasks until the function
4672  * returns NULL.  On completion of iteration, css_task_iter_end() must be
4673  * called.
4674  */
4675 void css_task_iter_start(struct cgroup_subsys_state *css, unsigned int flags,
4676                          struct css_task_iter *it)
4677 {
4678         memset(it, 0, sizeof(*it));
4679 
4680         spin_lock_irq(&css_set_lock);
4681 
4682         it->ss = css->ss;
4683         it->flags = flags;
4684 
4685         if (CGROUP_HAS_SUBSYS_CONFIG && it->ss)
4686                 it->cset_pos = &css->cgroup->e_csets[css->ss->id];
4687         else
4688                 it->cset_pos = &css->cgroup->cset_links;
4689 
4690         it->cset_head = it->cset_pos;
4691 
4692         css_task_iter_advance(it);
4693 
4694         spin_unlock_irq(&css_set_lock);
4695 }
4696 
4697 /**
4698  * css_task_iter_next - return the next task for the iterator
4699  * @it: the task iterator being iterated
4700  *
4701  * The "next" function for task iteration.  @it should have been
4702  * initialized via css_task_iter_start().  Returns NULL when the iteration
4703  * reaches the end.
4704  */
4705 struct task_struct *css_task_iter_next(struct css_task_iter *it)
4706 {
4707         if (it->cur_task) {
4708                 put_task_struct(it->cur_task);
4709                 it->cur_task = NULL;
4710         }
4711 
4712         spin_lock_irq(&css_set_lock);
4713 
4714         /* @it may be half-advanced by skips, finish advancing */
4715         if (it->flags & CSS_TASK_ITER_SKIPPED)
4716                 css_task_iter_advance(it);
4717 
4718         if (it->task_pos) {
4719                 it->cur_task = list_entry(it->task_pos, struct task_struct,
4720                                           cg_list);
4721                 get_task_struct(it->cur_task);
4722                 css_task_iter_advance(it);
4723         }
4724 
4725         spin_unlock_irq(&css_set_lock);
4726 
4727         return it->cur_task;
4728 }
4729 
4730 /**
4731  * css_task_iter_end - finish task iteration
4732  * @it: the task iterator to finish
4733  *
4734  * Finish task iteration started by css_task_iter_start().
4735  */
4736 void css_task_iter_end(struct css_task_iter *it)
4737 {
4738         if (it->cur_cset) {
4739                 spin_lock_irq(&css_set_lock);
4740                 list_del(&it->iters_node);
4741                 put_css_set_locked(it->cur_cset);
4742                 spin_unlock_irq(&css_set_lock);
4743         }
4744 
4745         if (it->cur_dcset)
4746                 put_css_set(it->cur_dcset);
4747 
4748         if (it->cur_task)
4749                 put_task_struct(it->cur_task);
4750 }
4751 
4752 static void cgroup_procs_release(struct kernfs_open_file *of)
4753 {
4754         if (of->priv) {
4755                 css_task_iter_end(of->priv);
4756                 kfree(of->priv);
4757         }
4758 }
4759 
4760 static void *cgroup_procs_next(struct seq_file *s, void *v, loff_t *pos)
4761 {
4762         struct kernfs_open_file *of = s->private;
4763         struct css_task_iter *it = of->priv;
4764 
4765         if (pos)
4766                 (*pos)++;
4767 
4768         return css_task_iter_next(it);
4769 }
4770 
4771 static void *__cgroup_procs_start(struct seq_file *s, loff_t *pos,
4772                                   unsigned int iter_flags)
4773 {
4774         struct kernfs_open_file *of = s->private;
4775         struct cgroup *cgrp = seq_css(s)->cgroup;
4776         struct css_task_iter *it = of->priv;
4777 
4778         /*
4779          * When a seq_file is seeked, it's always traversed sequentially
4780          * from position 0, so we can simply keep iterating on !0 *pos.
4781          */
4782         if (!it) {
4783                 if (WARN_ON_ONCE((*pos)))
4784                         return ERR_PTR(-EINVAL);
4785 
4786                 it = kzalloc(sizeof(*it), GFP_KERNEL);
4787                 if (!it)
4788                         return ERR_PTR(-ENOMEM);
4789                 of->priv = it;
4790                 css_task_iter_start(&cgrp->self, iter_flags, it);
4791         } else if (!(*pos)) {
4792                 css_task_iter_end(it);
4793                 css_task_iter_start(&cgrp->self, iter_flags, it);
4794         } else
4795                 return it->cur_task;
4796 
4797         return cgroup_procs_next(s, NULL, NULL);
4798 }
4799 
4800 static void *cgroup_procs_start(struct seq_file *s, loff_t *pos)
4801 {
4802         struct cgroup *cgrp = seq_css(s)->cgroup;
4803 
4804         /*
4805          * All processes of a threaded subtree belong to the domain cgroup
4806          * of the subtree.  Only threads can be distributed across the
4807          * subtree.  Reject reads on cgroup.procs in the subtree proper.
4808          * They're always empty anyway.
4809          */
4810         if (cgroup_is_threaded(cgrp))
4811                 return ERR_PTR(-EOPNOTSUPP);
4812 
4813         return __cgroup_procs_start(s, pos, CSS_TASK_ITER_PROCS |
4814                                             CSS_TASK_ITER_THREADED);
4815 }
4816 
4817 static int cgroup_procs_show(struct seq_file *s, void *v)
4818 {
4819         seq_printf(s, "%d\n", task_pid_vnr(v));
4820         return 0;
4821 }
4822 
4823 static int cgroup_may_write(const struct cgroup *cgrp, struct super_block *sb)
4824 {
4825         int ret;
4826         struct inode *inode;
4827 
4828         lockdep_assert_held(&cgroup_mutex);
4829 
4830         inode = kernfs_get_inode(sb, cgrp->procs_file.kn);
4831         if (!inode)
4832                 return -ENOMEM;
4833 
4834         ret = inode_permission(&init_user_ns, inode, MAY_WRITE);
4835         iput(inode);
4836         return ret;
4837 }
4838 
4839 static int cgroup_procs_write_permission(struct cgroup *src_cgrp,
4840                                          struct cgroup *dst_cgrp,
4841                                          struct super_block *sb)
4842 {
4843         struct cgroup_namespace *ns = current->nsproxy->cgroup_ns;
4844         struct cgroup *com_cgrp = src_cgrp;
4845         int ret;
4846 
4847         lockdep_assert_held(&cgroup_mutex);
4848 
4849         /* find the common ancestor */
4850         while (!cgroup_is_descendant(dst_cgrp, com_cgrp))
4851                 com_cgrp = cgroup_parent(com_cgrp);
4852 
4853         /* %current should be authorized to migrate to the common ancestor */
4854         ret = cgroup_may_write(com_cgrp, sb);
4855         if (ret)
4856                 return ret;
4857 
4858         /*
4859          * If namespaces are delegation boundaries, %current must be able
4860          * to see both source and destination cgroups from its namespace.
4861          */
4862         if ((cgrp_dfl_root.flags & CGRP_ROOT_NS_DELEGATE) &&
4863             (!cgroup_is_descendant(src_cgrp, ns->root_cset->dfl_cgrp) ||
4864              !cgroup_is_descendant(dst_cgrp, ns->root_cset->dfl_cgrp)))
4865                 return -ENOENT;
4866 
4867         return 0;
4868 }
4869 
4870 static int cgroup_attach_permissions(struct cgroup *src_cgrp,
4871                                      struct cgroup *dst_cgrp,
4872                                      struct super_block *sb, bool threadgroup)
4873 {
4874         int ret = 0;
4875 
4876         ret = cgroup_procs_write_permission(src_cgrp, dst_cgrp, sb);
4877         if (ret)
4878                 return ret;
4879 
4880         ret = cgroup_migrate_vet_dst(dst_cgrp);
4881         if (ret)
4882                 return ret;
4883 
4884         if (!threadgroup && (src_cgrp->dom_cgrp != dst_cgrp->dom_cgrp))
4885                 ret = -EOPNOTSUPP;
4886 
4887         return ret;
4888 }
4889 
4890 static ssize_t __cgroup_procs_write(struct kernfs_open_file *of, char *buf,
4891                                     bool threadgroup)
4892 {
4893         struct cgroup *src_cgrp, *dst_cgrp;
4894         struct task_struct *task;
4895         ssize_t ret;
4896         bool locked;
4897 
4898         dst_cgrp = cgroup_kn_lock_live(of->kn, false);
4899         if (!dst_cgrp)
4900                 return -ENODEV;
4901 
4902         task = cgroup_procs_write_start(buf, threadgroup, &locked);
4903         ret = PTR_ERR_OR_ZERO(task);
4904         if (ret)
4905                 goto out_unlock;
4906 
4907         /* find the source cgroup */
4908         spin_lock_irq(&css_set_lock);
4909         src_cgrp = task_cgroup_from_root(task, &cgrp_dfl_root);
4910         spin_unlock_irq(&css_set_lock);
4911 
4912         /* process and thread migrations follow same delegation rule */
4913         ret = cgroup_attach_permissions(src_cgrp, dst_cgrp,
4914                                         of->file->f_path.dentry->d_sb, threadgroup);
4915         if (ret)
4916                 goto out_finish;
4917 
4918         ret = cgroup_attach_task(dst_cgrp, task, threadgroup);
4919 
4920 out_finish:
4921         cgroup_procs_write_finish(task, locked);
4922 out_unlock:
4923         cgroup_kn_unlock(of->kn);
4924 
4925         return ret;
4926 }
4927 
4928 static ssize_t cgroup_procs_write(struct kernfs_open_file *of,
4929                                   char *buf, size_t nbytes, loff_t off)
4930 {
4931         return __cgroup_procs_write(of, buf, true) ?: nbytes;
4932 }
4933 
4934 static void *cgroup_threads_start(struct seq_file *s, loff_t *pos)
4935 {
4936         return __cgroup_procs_start(s, pos, 0);
4937 }
4938 
4939 static ssize_t cgroup_threads_write(struct kernfs_open_file *of,
4940                                     char *buf, size_t nbytes, loff_t off)
4941 {
4942         return __cgroup_procs_write(of, buf, false) ?: nbytes;
4943 }
4944 
4945 /* cgroup core interface files for the default hierarchy */
4946 static struct cftype cgroup_base_files[] = {
4947         {
4948                 .name = "cgroup.type",
4949                 .flags = CFTYPE_NOT_ON_ROOT,
4950                 .seq_show = cgroup_type_show,
4951                 .write = cgroup_type_write,
4952         },
4953         {
4954                 .name = "cgroup.procs",
4955                 .flags = CFTYPE_NS_DELEGATABLE,
4956                 .file_offset = offsetof(struct cgroup, procs_file),
4957                 .release = cgroup_procs_release,
4958                 .seq_start = cgroup_procs_start,
4959                 .seq_next = cgroup_procs_next,
4960                 .seq_show = cgroup_procs_show,
4961                 .write = cgroup_procs_write,
4962         },
4963         {
4964                 .name = "cgroup.threads",
4965                 .flags = CFTYPE_NS_DELEGATABLE,
4966                 .release = cgroup_procs_release,
4967                 .seq_start = cgroup_threads_start,
4968                 .seq_next = cgroup_procs_next,
4969                 .seq_show = cgroup_procs_show,
4970                 .write = cgroup_threads_write,
4971         },
4972         {
4973                 .name = "cgroup.controllers",
4974                 .seq_show = cgroup_controllers_show,
4975         },
4976         {
4977                 .name = "cgroup.subtree_control",
4978                 .flags = CFTYPE_NS_DELEGATABLE,
4979                 .seq_show = cgroup_subtree_control_show,
4980                 .write = cgroup_subtree_control_write,
4981         },
4982         {
4983                 .name = "cgroup.events",
4984                 .flags = CFTYPE_NOT_ON_ROOT,
4985                 .file_offset = offsetof(struct cgroup, events_file),
4986                 .seq_show = cgroup_events_show,
4987         },
4988         {
4989                 .name = "cgroup.max.descendants",
4990                 .seq_show = cgroup_max_descendants_show,
4991                 .write = cgroup_max_descendants_write,
4992         },
4993         {
4994                 .name = "cgroup.max.depth",
4995                 .seq_show = cgroup_max_depth_show,
4996                 .write = cgroup_max_depth_write,
4997         },
4998         {
4999                 .name = "cgroup.stat",
5000                 .seq_show = cgroup_stat_show,
5001         },
5002         {
5003                 .name = "cgroup.freeze",
5004                 .flags = CFTYPE_NOT_ON_ROOT,
5005                 .seq_show = cgroup_freeze_show,
5006                 .write = cgroup_freeze_write,
5007         },
5008         {
5009                 .name = "cgroup.kill",
5010                 .flags = CFTYPE_NOT_ON_ROOT,
5011                 .write = cgroup_kill_write,
5012         },
5013         {
5014                 .name = "cpu.stat",
5015                 .seq_show = cpu_stat_show,
5016         },
5017 #ifdef CONFIG_PSI
5018         {
5019                 .name = "io.pressure",
5020                 .flags = CFTYPE_PRESSURE,
5021                 .seq_show = cgroup_io_pressure_show,
5022                 .write = cgroup_io_pressure_write,
5023                 .poll = cgroup_pressure_poll,
5024                 .release = cgroup_pressure_release,
5025         },
5026         {
5027                 .name = "memory.pressure",
5028                 .flags = CFTYPE_PRESSURE,
5029                 .seq_show = cgroup_memory_pressure_show,
5030                 .write = cgroup_memory_pressure_write,
5031                 .poll = cgroup_pressure_poll,
5032                 .release = cgroup_pressure_release,
5033         },
5034         {
5035                 .name = "cpu.pressure",
5036                 .flags = CFTYPE_PRESSURE,
5037                 .seq_show = cgroup_cpu_pressure_show,
5038                 .write = cgroup_cpu_pressure_write,
5039                 .poll = cgroup_pressure_poll,
5040                 .release = cgroup_pressure_release,
5041         },
5042 #endif /* CONFIG_PSI */
5043         { }     /* terminate */
5044 };
5045 
5046 /*
5047  * css destruction is four-stage process.
5048  *
5049  * 1. Destruction starts.  Killing of the percpu_ref is initiated.
5050  *    Implemented in kill_css().
5051  *
5052  * 2. When the percpu_ref is confirmed to be visible as killed on all CPUs
5053  *    and thus css_tryget_online() is guaranteed to fail, the css can be
5054  *    offlined by invoking offline_css().  After offlining, the base ref is
5055  *    put.  Implemented in css_killed_work_fn().
5056  *
5057  * 3. When the percpu_ref reaches zero, the only possible remaining
5058  *    accessors are inside RCU read sections.  css_release() schedules the
5059  *    RCU callback.
5060  *
5061  * 4. After the grace period, the css can be freed.  Implemented in
5062  *    css_free_work_fn().
5063  *
5064  * It is actually hairier because both step 2 and 4 require process context
5065  * and thus involve punting to css->destroy_work adding two additional
5066  * steps to the already complex sequence.
5067  */
5068 static void css_free_rwork_fn(struct work_struct *work)
5069 {
5070         struct cgroup_subsys_state *css = container_of(to_rcu_work(work),
5071                                 struct cgroup_subsys_state, destroy_rwork);
5072         struct cgroup_subsys *ss = css->ss;
5073         struct cgroup *cgrp = css->cgroup;
5074 
5075         percpu_ref_exit(&css->refcnt);
5076 
5077         if (ss) {
5078                 /* css free path */
5079                 struct cgroup_subsys_state *parent = css->parent;
5080                 int id = css->id;
5081 
5082                 ss->css_free(css);
5083                 cgroup_idr_remove(&ss->css_idr, id);
5084                 cgroup_put(cgrp);
5085 
5086                 if (parent)
5087                         css_put(parent);
5088         } else {
5089                 /* cgroup free path */
5090                 atomic_dec(&cgrp->root->nr_cgrps);
5091                 cgroup1_pidlist_destroy_all(cgrp);
5092                 cancel_work_sync(&cgrp->release_agent_work);
5093 
5094                 if (cgroup_parent(cgrp)) {
5095                         /*
5096                          * We get a ref to the parent, and put the ref when
5097                          * this cgroup is being freed, so it's guaranteed
5098                          * that the parent won't be destroyed before its
5099                          * children.
5100                          */
5101                         cgroup_put(cgroup_parent(cgrp));
5102                         kernfs_put(cgrp->kn);
5103                         psi_cgroup_free(cgrp);
5104                         cgroup_rstat_exit(cgrp);
5105                         kfree(cgrp);
5106                 } else {
5107                         /*
5108                          * This is root cgroup's refcnt reaching zero,
5109                          * which indicates that the root should be
5110                          * released.
5111                          */
5112                         cgroup_destroy_root(cgrp->root);
5113                 }
5114         }
5115 }
5116 
5117 static void css_release_work_fn(struct work_struct *work)
5118 {
5119         struct cgroup_subsys_state *css =
5120                 container_of(work, struct cgroup_subsys_state, destroy_work);
5121         struct cgroup_subsys *ss = css->ss;
5122         struct cgroup *cgrp = css->cgroup;
5123 
5124         mutex_lock(&cgroup_mutex);
5125 
5126         css->flags |= CSS_RELEASED;
5127         list_del_rcu(&css->sibling);
5128 
5129         if (ss) {
5130                 /* css release path */
5131                 if (!list_empty(&css->rstat_css_node)) {
5132                         cgroup_rstat_flush(cgrp);
5133                         list_del_rcu(&css->rstat_css_node);
5134                 }
5135 
5136                 cgroup_idr_replace(&ss->css_idr, NULL, css->id);
5137                 if (ss->css_released)
5138                         ss->css_released(css);
5139         } else {
5140                 struct cgroup *tcgrp;
5141 
5142                 /* cgroup release path */
5143                 TRACE_CGROUP_PATH(release, cgrp);
5144 
5145                 cgroup_rstat_flush(cgrp);
5146 
5147                 spin_lock_irq(&css_set_lock);
5148                 for (tcgrp = cgroup_parent(cgrp); tcgrp;
5149                      tcgrp = cgroup_parent(tcgrp))
5150                         tcgrp->nr_dying_descendants--;
5151                 spin_unlock_irq(&css_set_lock);
5152 
5153                 /*
5154                  * There are two control paths which try to determine
5155                  * cgroup from dentry without going through kernfs -
5156                  * cgroupstats_build() and css_tryget_online_from_dir().
5157                  * Those are supported by RCU protecting clearing of
5158                  * cgrp->kn->priv backpointer.
5159                  */
5160                 if (cgrp->kn)
5161                         RCU_INIT_POINTER(*(void __rcu __force **)&cgrp->kn->priv,
5162                                          NULL);
5163         }
5164 
5165         mutex_unlock(&cgroup_mutex);
5166 
5167         INIT_RCU_WORK(&css->destroy_rwork, css_free_rwork_fn);
5168         queue_rcu_work(cgroup_destroy_wq, &css->destroy_rwork);
5169 }
5170 
5171 static void css_release(struct percpu_ref *ref)
5172 {
5173         struct cgroup_subsys_state *css =
5174                 container_of(ref, struct cgroup_subsys_state, refcnt);
5175 
5176         INIT_WORK(&css->destroy_work, css_release_work_fn);
5177         queue_work(cgroup_destroy_wq, &css->destroy_work);
5178 }
5179 
5180 static void init_and_link_css(struct cgroup_subsys_state *css,
5181                               struct cgroup_subsys *ss, struct cgroup *cgrp)
5182 {
5183         lockdep_assert_held(&cgroup_mutex);
5184 
5185         cgroup_get_live(cgrp);
5186 
5187         memset(css, 0, sizeof(*css));
5188         css->cgroup = cgrp;
5189         css->ss = ss;
5190         css->id = -1;
5191         INIT_LIST_HEAD(&css->sibling);
5192         INIT_LIST_HEAD(&css->children);
5193         INIT_LIST_HEAD(&css->rstat_css_node);
5194         css->serial_nr = css_serial_nr_next++;
5195         atomic_set(&css->online_cnt, 0);
5196 
5197         if (cgroup_parent(cgrp)) {
5198                 css->parent = cgroup_css(cgroup_parent(cgrp), ss);
5199                 css_get(css->parent);
5200         }
5201 
5202         if (ss->css_rstat_flush)
5203                 list_add_rcu(&css->rstat_css_node, &cgrp->rstat_css_list);
5204 
5205         BUG_ON(cgroup_css(cgrp, ss));
5206 }
5207 
5208 /* invoke ->css_online() on a new CSS and mark it online if successful */
5209 static int online_css(struct cgroup_subsys_state *css)
5210 {
5211         struct cgroup_subsys *ss = css->ss;
5212         int ret = 0;
5213 
5214         lockdep_assert_held(&cgroup_mutex);
5215 
5216         if (ss->css_online)
5217                 ret = ss->css_online(css);
5218         if (!ret) {
5219                 css->flags |= CSS_ONLINE;
5220                 rcu_assign_pointer(css->cgroup->subsys[ss->id], css);
5221 
5222                 atomic_inc(&css->online_cnt);
5223                 if (css->parent)
5224                         atomic_inc(&css->parent->online_cnt);
5225         }
5226         return ret;
5227 }
5228 
5229 /* if the CSS is online, invoke ->css_offline() on it and mark it offline */
5230 static void offline_css(struct cgroup_subsys_state *css)
5231 {
5232         struct cgroup_subsys *ss = css->ss;
5233 
5234         lockdep_assert_held(&cgroup_mutex);
5235 
5236         if (!(css->flags & CSS_ONLINE))
5237                 return;
5238 
5239         if (ss->css_offline)
5240                 ss->css_offline(css);
5241 
5242         css->flags &= ~CSS_ONLINE;
5243         RCU_INIT_POINTER(css->cgroup->subsys[ss->id], NULL);
5244 
5245         wake_up_all(&css->cgroup->offline_waitq);
5246 }
5247 
5248 /**
5249  * css_create - create a cgroup_subsys_state
5250  * @cgrp: the cgroup new css will be associated with
5251  * @ss: the subsys of new css
5252  *
5253  * Create a new css associated with @cgrp - @ss pair.  On success, the new
5254  * css is online and installed in @cgrp.  This function doesn't create the
5255  * interface files.  Returns 0 on success, -errno on failure.
5256  */
5257 static struct cgroup_subsys_state *css_create(struct cgroup *cgrp,
5258                                               struct cgroup_subsys *ss)
5259 {
5260         struct cgroup *parent = cgroup_parent(cgrp);
5261         struct cgroup_subsys_state *parent_css = cgroup_css(parent, ss);
5262         struct cgroup_subsys_state *css;
5263         int err;
5264 
5265         lockdep_assert_held(&cgroup_mutex);
5266 
5267         css = ss->css_alloc(parent_css);
5268         if (!css)
5269                 css = ERR_PTR(-ENOMEM);
5270         if (IS_ERR(css))
5271                 return css;
5272 
5273         init_and_link_css(css, ss, cgrp);
5274 
5275         err = percpu_ref_init(&css->refcnt, css_release, 0, GFP_KERNEL);
5276         if (err)
5277                 goto err_free_css;
5278 
5279         err = cgroup_idr_alloc(&ss->css_idr, NULL, 2, 0, GFP_KERNEL);
5280         if (err < 0)
5281                 goto err_free_css;
5282         css->id = err;
5283 
5284         /* @css is ready to be brought online now, make it visible */
5285         list_add_tail_rcu(&css->sibling, &parent_css->children);
5286         cgroup_idr_replace(&ss->css_idr, css, css->id);
5287 
5288         err = online_css(css);
5289         if (err)
5290                 goto err_list_del;
5291 
5292         return css;
5293 
5294 err_list_del:
5295         list_del_rcu(&css->sibling);
5296 err_free_css:
5297         list_del_rcu(&css->rstat_css_node);
5298         INIT_RCU_WORK(&css->destroy_rwork, css_free_rwork_fn);
5299         queue_rcu_work(cgroup_destroy_wq, &css->destroy_rwork);
5300         return ERR_PTR(err);
5301 }
5302 
5303 /*
5304  * The returned cgroup is fully initialized including its control mask, but
5305  * it isn't associated with its kernfs_node and doesn't have the control
5306  * mask applied.
5307  */
5308 static struct cgroup *cgroup_create(struct cgroup *parent, const char *name,
5309                                     umode_t mode)
5310 {
5311         struct cgroup_root *root = parent->root;
5312         struct cgroup *cgrp, *tcgrp;
5313         struct kernfs_node *kn;
5314         int level = parent->level + 1;
5315         int ret;
5316 
5317         /* allocate the cgroup and its ID, 0 is reserved for the root */
5318         cgrp = kzalloc(struct_size(cgrp, ancestor_ids, (level + 1)),
5319                        GFP_KERNEL);
5320         if (!cgrp)
5321                 return ERR_PTR(-ENOMEM);
5322 
5323         ret = percpu_ref_init(&cgrp->self.refcnt, css_release, 0, GFP_KERNEL);
5324         if (ret)
5325                 goto out_free_cgrp;
5326 
5327         ret = cgroup_rstat_init(cgrp);
5328         if (ret)
5329                 goto out_cancel_ref;
5330 
5331         /* create the directory */
5332         kn = kernfs_create_dir(parent->kn, name, mode, cgrp);
5333         if (IS_ERR(kn)) {
5334                 ret = PTR_ERR(kn);
5335                 goto out_stat_exit;
5336         }
5337         cgrp->kn = kn;
5338 
5339         init_cgroup_housekeeping(cgrp);
5340 
5341         cgrp->self.parent = &parent->self;
5342         cgrp->root = root;
5343         cgrp->level = level;
5344 
5345         ret = psi_cgroup_alloc(cgrp);
5346         if (ret)
5347                 goto out_kernfs_remove;
5348 
5349         ret = cgroup_bpf_inherit(cgrp);
5350         if (ret)
5351                 goto out_psi_free;
5352 
5353         /*
5354          * New cgroup inherits effective freeze counter, and
5355          * if the parent has to be frozen, the child has too.
5356          */
5357         cgrp->freezer.e_freeze = parent->freezer.e_freeze;
5358         if (cgrp->freezer.e_freeze) {
5359                 /*
5360                  * Set the CGRP_FREEZE flag, so when a process will be
5361                  * attached to the child cgroup, it will become frozen.
5362                  * At this point the new cgroup is unpopulated, so we can
5363                  * consider it frozen immediately.
5364                  */
5365                 set_bit(CGRP_FREEZE, &cgrp->flags);
5366                 set_bit(CGRP_FROZEN, &cgrp->flags);
5367         }
5368 
5369         spin_lock_irq(&css_set_lock);
5370         for (tcgrp = cgrp; tcgrp; tcgrp = cgroup_parent(tcgrp)) {
5371                 cgrp->ancestor_ids[tcgrp->level] = cgroup_id(tcgrp);
5372 
5373                 if (tcgrp != cgrp) {
5374                         tcgrp->nr_descendants++;
5375 
5376                         /*
5377                          * If the new cgroup is frozen, all ancestor cgroups
5378                          * get a new frozen descendant, but their state can't
5379                          * change because of this.
5380                          */
5381                         if (cgrp->freezer.e_freeze)
5382                                 tcgrp->freezer.nr_frozen_descendants++;
5383                 }
5384         }
5385         spin_unlock_irq(&css_set_lock);
5386 
5387         if (notify_on_release(parent))
5388                 set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
5389 
5390         if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &parent->flags))
5391                 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags);
5392 
5393         cgrp->self.serial_nr = css_serial_nr_next++;
5394 
5395         /* allocation complete, commit to creation */
5396         list_add_tail_rcu(&cgrp->self.sibling, &cgroup_parent(cgrp)->self.children);
5397         atomic_inc(&root->nr_cgrps);
5398         cgroup_get_live(parent);
5399 
5400         /*
5401          * On the default hierarchy, a child doesn't automatically inherit
5402          * subtree_control from the parent.  Each is configured manually.
5403          */
5404         if (!cgroup_on_dfl(cgrp))
5405                 cgrp->subtree_control = cgroup_control(cgrp);
5406 
5407         cgroup_propagate_control(cgrp);
5408 
5409         return cgrp;
5410 
5411 out_psi_free:
5412         psi_cgroup_free(cgrp);
5413 out_kernfs_remove:
5414         kernfs_remove(cgrp->kn);
5415 out_stat_exit:
5416         cgroup_rstat_exit(cgrp);
5417 out_cancel_ref:
5418         percpu_ref_exit(&cgrp->self.refcnt);
5419 out_free_cgrp:
5420         kfree(cgrp);
5421         return ERR_PTR(ret);
5422 }
5423 
5424 static bool cgroup_check_hierarchy_limits(struct cgroup *parent)
5425 {
5426         struct cgroup *cgroup;
5427         int ret = false;
5428         int level = 1;
5429 
5430         lockdep_assert_held(&cgroup_mutex);
5431 
5432         for (cgroup = parent; cgroup; cgroup = cgroup_parent(cgroup)) {
5433                 if (cgroup->nr_descendants >= cgroup->max_descendants)
5434                         goto fail;
5435 
5436                 if (level > cgroup->max_depth)
5437                         goto fail;
5438 
5439                 level++;
5440         }
5441 
5442         ret = true;
5443 fail:
5444         return ret;
5445 }
5446 
5447 int cgroup_mkdir(struct kernfs_node *parent_kn, const char *name, umode_t mode)
5448 {
5449         struct cgroup *parent, *cgrp;
5450         int ret;
5451 
5452         /* do not accept '\n' to prevent making /proc/<pid>/cgroup unparsable */
5453         if (strchr(name, '\n'))
5454                 return -EINVAL;
5455 
5456         parent = cgroup_kn_lock_live(parent_kn, false);
5457         if (!parent)
5458                 return -ENODEV;
5459 
5460         if (!cgroup_check_hierarchy_limits(parent)) {
5461                 ret = -EAGAIN;
5462                 goto out_unlock;
5463         }
5464 
5465         cgrp = cgroup_create(parent, name, mode);
5466         if (IS_ERR(cgrp)) {
5467                 ret = PTR_ERR(cgrp);
5468                 goto out_unlock;
5469         }
5470 
5471         /*
5472          * This extra ref will be put in cgroup_free_fn() and guarantees
5473          * that @cgrp->kn is always accessible.
5474          */
5475         kernfs_get(cgrp->kn);
5476 
5477         ret = cgroup_kn_set_ugid(cgrp->kn);
5478         if (ret)
5479                 goto out_destroy;
5480 
5481         ret = css_populate_dir(&cgrp->self);
5482         if (ret)
5483                 goto out_destroy;
5484 
5485         ret = cgroup_apply_control_enable(cgrp);
5486         if (ret)
5487                 goto out_destroy;
5488 
5489         TRACE_CGROUP_PATH(mkdir, cgrp);
5490 
5491         /* let's create and online css's */
5492         kernfs_activate(cgrp->kn);
5493 
5494         ret = 0;
5495         goto out_unlock;
5496 
5497 out_destroy:
5498         cgroup_destroy_locked(cgrp);
5499 out_unlock:
5500         cgroup_kn_unlock(parent_kn);
5501         return ret;
5502 }
5503 
5504 /*
5505  * This is called when the refcnt of a css is confirmed to be killed.
5506  * css_tryget_online() is now guaranteed to fail.  Tell the subsystem to
5507  * initiate destruction and put the css ref from kill_css().
5508  */
5509 static void css_killed_work_fn(struct work_struct *work)
5510 {
5511         struct cgroup_subsys_state *css =
5512                 container_of(work, struct cgroup_subsys_state, destroy_work);
5513 
5514         mutex_lock(&cgroup_mutex);
5515 
5516         do {
5517                 offline_css(css);
5518                 css_put(css);
5519                 /* @css can't go away while we're holding cgroup_mutex */
5520                 css = css->parent;
5521         } while (css && atomic_dec_and_test(&css->online_cnt));
5522 
5523         mutex_unlock(&cgroup_mutex);
5524 }
5525 
5526 /* css kill confirmation processing requires process context, bounce */
5527 static void css_killed_ref_fn(struct percpu_ref *ref)
5528 {
5529         struct cgroup_subsys_state *css =
5530                 container_of(ref, struct cgroup_subsys_state, refcnt);
5531 
5532         if (atomic_dec_and_test(&css->online_cnt)) {
5533                 INIT_WORK(&css->destroy_work, css_killed_work_fn);
5534                 queue_work(cgroup_destroy_wq, &css->destroy_work);
5535         }
5536 }
5537 
5538 /**
5539  * kill_css - destroy a css
5540  * @css: css to destroy
5541  *
5542  * This function initiates destruction of @css by removing cgroup interface
5543  * files and putting its base reference.  ->css_offline() will be invoked
5544  * asynchronously once css_tryget_online() is guaranteed to fail and when
5545  * the reference count reaches zero, @css will be released.
5546  */
5547 static void kill_css(struct cgroup_subsys_state *css)
5548 {
5549         lockdep_assert_held(&cgroup_mutex);
5550 
5551         if (css->flags & CSS_DYING)
5552                 return;
5553 
5554         css->flags |= CSS_DYING;
5555 
5556         /*
5557          * This must happen before css is disassociated with its cgroup.
5558          * See seq_css() for details.
5559          */
5560         css_clear_dir(css);
5561 
5562         /*
5563          * Killing would put the base ref, but we need to keep it alive
5564          * until after ->css_offline().
5565          */
5566         css_get(css);
5567 
5568         /*
5569          * cgroup core guarantees that, by the time ->css_offline() is
5570          * invoked, no new css reference will be given out via
5571          * css_tryget_online().  We can't simply call percpu_ref_kill() and
5572          * proceed to offlining css's because percpu_ref_kill() doesn't
5573          * guarantee that the ref is seen as killed on all CPUs on return.
5574          *
5575          * Use percpu_ref_kill_and_confirm() to get notifications as each
5576          * css is confirmed to be seen as killed on all CPUs.
5577          */
5578         percpu_ref_kill_and_confirm(&css->refcnt, css_killed_ref_fn);
5579 }
5580 
5581 /**
5582  * cgroup_destroy_locked - the first stage of cgroup destruction
5583  * @cgrp: cgroup to be destroyed
5584  *
5585  * css's make use of percpu refcnts whose killing latency shouldn't be
5586  * exposed to userland and are RCU protected.  Also, cgroup core needs to
5587  * guarantee that css_tryget_online() won't succeed by the time
5588  * ->css_offline() is invoked.  To satisfy all the requirements,
5589  * destruction is implemented in the following two steps.
5590  *
5591  * s1. Verify @cgrp can be destroyed and mark it dying.  Remove all
5592  *     userland visible parts and start killing the percpu refcnts of
5593  *     css's.  Set up so that the next stage will be kicked off once all
5594  *     the percpu refcnts are confirmed to be killed.
5595  *
5596  * s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the
5597  *     rest of destruction.  Once all cgroup references are gone, the
5598  *     cgroup is RCU-freed.
5599  *
5600  * This function implements s1.  After this step, @cgrp is gone as far as
5601  * the userland is concerned and a new cgroup with the same name may be
5602  * created.  As cgroup doesn't care about the names internally, this
5603  * doesn't cause any problem.
5604  */
5605 static int cgroup_destroy_locked(struct cgroup *cgrp)
5606         __releases(&cgroup_mutex) __acquires(&cgroup_mutex)
5607 {
5608         struct cgroup *tcgrp, *parent = cgroup_parent(cgrp);
5609         struct cgroup_subsys_state *css;
5610         struct cgrp_cset_link *link;
5611         int ssid;
5612 
5613         lockdep_assert_held(&cgroup_mutex);
5614 
5615         /*
5616          * Only migration can raise populated from zero and we're already
5617          * holding cgroup_mutex.
5618          */
5619         if (cgroup_is_populated(cgrp))
5620                 return -EBUSY;
5621 
5622         /*
5623          * Make sure there's no live children.  We can't test emptiness of
5624          * ->self.children as dead children linger on it while being
5625          * drained; otherwise, "rmdir parent/child parent" may fail.
5626          */
5627         if (css_has_online_children(&cgrp->self))
5628                 return -EBUSY;
5629 
5630         /*
5631          * Mark @cgrp and the associated csets dead.  The former prevents
5632          * further task migration and child creation by disabling
5633          * cgroup_lock_live_group().  The latter makes the csets ignored by
5634          * the migration path.
5635          */
5636         cgrp->self.flags &= ~CSS_ONLINE;
5637 
5638         spin_lock_irq(&css_set_lock);
5639         list_for_each_entry(link, &cgrp->cset_links, cset_link)
5640                 link->cset->dead = true;
5641         spin_unlock_irq(&css_set_lock);
5642 
5643         /* initiate massacre of all css's */
5644         for_each_css(css, ssid, cgrp)
5645                 kill_css(css);
5646 
5647         /* clear and remove @cgrp dir, @cgrp has an extra ref on its kn */
5648         css_clear_dir(&cgrp->self);
5649         kernfs_remove(cgrp->kn);
5650 
5651         if (parent && cgroup_is_threaded(cgrp))
5652                 parent->nr_threaded_children--;
5653 
5654         spin_lock_irq(&css_set_lock);
5655         for (tcgrp = cgroup_parent(cgrp); tcgrp; tcgrp = cgroup_parent(tcgrp)) {
5656                 tcgrp->nr_descendants--;
5657                 tcgrp->nr_dying_descendants++;
5658                 /*
5659                  * If the dying cgroup is frozen, decrease frozen descendants
5660                  * counters of ancestor cgroups.
5661                  */
5662                 if (test_bit(CGRP_FROZEN, &cgrp->flags))
5663                         tcgrp->freezer.nr_frozen_descendants--;
5664         }
5665         spin_unlock_irq(&css_set_lock);
5666 
5667         cgroup1_check_for_release(parent);
5668 
5669         cgroup_bpf_offline(cgrp);
5670 
5671         /* put the base reference */
5672         percpu_ref_kill(&cgrp->self.refcnt);
5673 
5674         return 0;
5675 };
5676 
5677 int cgroup_rmdir(struct kernfs_node *kn)
5678 {
5679         struct cgroup *cgrp;
5680         int ret = 0;
5681 
5682         cgrp = cgroup_kn_lock_live(kn, false);
5683         if (!cgrp)
5684                 return 0;
5685 
5686         ret = cgroup_destroy_locked(cgrp);
5687         if (!ret)
5688                 TRACE_CGROUP_PATH(rmdir, cgrp);
5689 
5690         cgroup_kn_unlock(kn);
5691         return ret;
5692 }
5693 
5694 static struct kernfs_syscall_ops cgroup_kf_syscall_ops = {
5695         .show_options           = cgroup_show_options,
5696         .mkdir                  = cgroup_mkdir,
5697         .rmdir                  = cgroup_rmdir,
5698         .show_path              = cgroup_show_path,
5699 };
5700 
5701 static void __init cgroup_init_subsys(struct cgroup_subsys *ss, bool early)
5702 {
5703         struct cgroup_subsys_state *css;
5704 
5705         pr_debug("Initializing cgroup subsys %s\n", ss->name);
5706 
5707         mutex_lock(&cgroup_mutex);
5708 
5709         idr_init(&ss->css_idr);
5710         INIT_LIST_HEAD(&ss->cfts);
5711 
5712         /* Create the root cgroup state for this subsystem */
5713         ss->root = &cgrp_dfl_root;
5714         css = ss->css_alloc(cgroup_css(&cgrp_dfl_root.cgrp, ss));
5715         /* We don't handle early failures gracefully */
5716         BUG_ON(IS_ERR(css));
5717         init_and_link_css(css, ss, &cgrp_dfl_root.cgrp);
5718 
5719         /*
5720          * Root csses are never destroyed and we can't initialize
5721          * percpu_ref during early init.  Disable refcnting.
5722          */
5723         css->flags |= CSS_NO_REF;
5724 
5725         if (early) {
5726                 /* allocation can't be done safely during early init */
5727                 css->id = 1;
5728         } else {
5729                 css->id = cgroup_idr_alloc(&ss->css_idr, css, 1, 2, GFP_KERNEL);
5730                 BUG_ON(css->id < 0);
5731         }
5732 
5733         /* Update the init_css_set to contain a subsys
5734          * pointer to this state - since the subsystem is
5735          * newly registered, all tasks and hence the
5736          * init_css_set is in the subsystem's root cgroup. */
5737         init_css_set.subsys[ss->id] = css;
5738 
5739         have_fork_callback |= (bool)ss->fork << ss->id;
5740         have_exit_callback |= (bool)ss->exit << ss->id;
5741         have_release_callback |= (bool)ss->release << ss->id;
5742         have_canfork_callback |= (bool)ss->can_fork << ss->id;
5743 
5744         /* At system boot, before all subsystems have been
5745          * registered, no tasks have been forked, so we don't
5746          * need to invoke fork callbacks here. */
5747         BUG_ON(!list_empty(&init_task.tasks));
5748 
5749         BUG_ON(online_css(css));
5750 
5751         mutex_unlock(&cgroup_mutex);
5752 }
5753 
5754 /**
5755  * cgroup_init_early - cgroup initialization at system boot
5756  *
5757  * Initialize cgroups at system boot, and initialize any
5758  * subsystems that request early init.
5759  */
5760 int __init cgroup_init_early(void)
5761 {
5762         static struct cgroup_fs_context __initdata ctx;
5763         struct cgroup_subsys *ss;
5764         int i;
5765 
5766         ctx.root = &cgrp_dfl_root;
5767         init_cgroup_root(&ctx);
5768         cgrp_dfl_root.cgrp.self.flags |= CSS_NO_REF;
5769 
5770         RCU_INIT_POINTER(init_task.cgroups, &init_css_set);
5771 
5772         for_each_subsys(ss, i) {
5773                 WARN(!ss->css_alloc || !ss->css_free || ss->name || ss->id,
5774                      "invalid cgroup_subsys %d:%s css_alloc=%p css_free=%p id:name=%d:%s\n",
5775                      i, cgroup_subsys_name[i], ss->css_alloc, ss->css_free,
5776                      ss->id, ss->name);
5777                 WARN(strlen(cgroup_subsys_name[i]) > MAX_CGROUP_TYPE_NAMELEN,
5778                      "cgroup_subsys_name %s too long\n", cgroup_subsys_name[i]);
5779 
5780                 ss->id = i;
5781                 ss->name = cgroup_subsys_name[i];
5782                 if (!ss->legacy_name)
5783                         ss->legacy_name = cgroup_subsys_name[i];
5784 
5785                 if (ss->early_init)
5786                         cgroup_init_subsys(ss, true);
5787         }
5788         return 0;
5789 }
5790 
5791 /**
5792  * cgroup_init - cgroup initialization
5793  *
5794  * Register cgroup filesystem and /proc file, and initialize
5795  * any subsystems that didn't request early init.
5796  */
5797 int __init cgroup_init(void)
5798 {
5799         struct cgroup_subsys *ss;
5800         int ssid;
5801 
5802         BUILD_BUG_ON(CGROUP_SUBSYS_COUNT > 16);
5803         BUG_ON(cgroup_init_cftypes(NULL, cgroup_base_files));
5804         BUG_ON(cgroup_init_cftypes(NULL, cgroup1_base_files));
5805 
5806         cgroup_rstat_boot();
5807 
5808         /*
5809          * The latency of the synchronize_rcu() is too high for cgroups,
5810          * avoid it at the cost of forcing all readers into the slow path.
5811          */
5812         rcu_sync_enter_start(&cgroup_threadgroup_rwsem.rss);
5813 
5814         get_user_ns(init_cgroup_ns.user_ns);
5815 
5816         mutex_lock(&cgroup_mutex);
5817 
5818         /*
5819          * Add init_css_set to the hash table so that dfl_root can link to
5820          * it during init.
5821          */
5822         hash_add(css_set_table, &init_css_set.hlist,
5823                  css_set_hash(init_css_set.subsys));
5824 
5825         BUG_ON(cgroup_setup_root(&cgrp_dfl_root, 0));
5826 
5827         mutex_unlock(&cgroup_mutex);
5828 
5829         for_each_subsys(ss, ssid) {
5830                 if (ss->early_init) {
5831                         struct cgroup_subsys_state *css =
5832                                 init_css_set.subsys[ss->id];
5833 
5834                         css->id = cgroup_idr_alloc(&ss->css_idr, css, 1, 2,
5835                                                    GFP_KERNEL);
5836                         BUG_ON(css->id < 0);
5837                 } else {
5838                         cgroup_init_subsys(ss, false);
5839                 }
5840 
5841                 list_add_tail(&init_css_set.e_cset_node[ssid],
5842                               &cgrp_dfl_root.cgrp.e_csets[ssid]);
5843 
5844                 /*
5845                  * Setting dfl_root subsys_mask needs to consider the
5846                  * disabled flag and cftype registration needs kmalloc,
5847                  * both of which aren't available during early_init.
5848                  */
5849                 if (!cgroup_ssid_enabled(ssid))
5850                         continue;
5851 
5852                 if (cgroup1_ssid_disabled(ssid))
5853                         printk(KERN_INFO "Disabling %s control group subsystem in v1 mounts\n",
5854                                ss->name);
5855 
5856                 cgrp_dfl_root.subsys_mask |= 1 << ss->id;
5857 
5858                 /* implicit controllers must be threaded too */
5859                 WARN_ON(ss->implicit_on_dfl && !ss->threaded);
5860 
5861                 if (ss->implicit_on_dfl)
5862                         cgrp_dfl_implicit_ss_mask |= 1 << ss->id;
5863                 else if (!ss->dfl_cftypes)
5864                         cgrp_dfl_inhibit_ss_mask |= 1 << ss->id;
5865 
5866                 if (ss->threaded)
5867                         cgrp_dfl_threaded_ss_mask |= 1 << ss->id;
5868 
5869                 if (ss->dfl_cftypes == ss->legacy_cftypes) {
5870                         WARN_ON(cgroup_add_cftypes(ss, ss->dfl_cftypes));
5871                 } else {
5872                         WARN_ON(cgroup_add_dfl_cftypes(ss, ss->dfl_cftypes));
5873                         WARN_ON(cgroup_add_legacy_cftypes(ss, ss->legacy_cftypes));
5874                 }
5875 
5876                 if (ss->bind)
5877                         ss->bind(init_css_set.subsys[ssid]);
5878 
5879                 mutex_lock(&cgroup_mutex);
5880                 css_populate_dir(init_css_set.subsys[ssid]);
5881                 mutex_unlock(&cgroup_mutex);
5882         }
5883 
5884         /* init_css_set.subsys[] has been updated, re-hash */
5885         hash_del(&init_css_set.hlist);
5886         hash_add(css_set_table, &init_css_set.hlist,
5887                  css_set_hash(init_css_set.subsys));
5888 
5889         WARN_ON(sysfs_create_mount_point(fs_kobj, "cgroup"));
5890         WARN_ON(register_filesystem(&cgroup_fs_type));
5891         WARN_ON(register_filesystem(&cgroup2_fs_type));
5892         WARN_ON(!proc_create_single("cgroups", 0, NULL, proc_cgroupstats_show));
5893 #ifdef CONFIG_CPUSETS
5894         WARN_ON(register_filesystem(&cpuset_fs_type));
5895 #endif
5896 
5897         return 0;
5898 }
5899 
5900 static int __init cgroup_wq_init(void)
5901 {
5902         /*
5903          * There isn't much point in executing destruction path in
5904          * parallel.  Good chunk is serialized with cgroup_mutex anyway.
5905          * Use 1 for @max_active.
5906          *
5907          * We would prefer to do this in cgroup_init() above, but that
5908          * is called before init_workqueues(): so leave this until after.
5909          */
5910         cgroup_destroy_wq = alloc_workqueue("cgroup_destroy", 0, 1);
5911         BUG_ON(!cgroup_destroy_wq);
5912         return 0;
5913 }
5914 core_initcall(cgroup_wq_init);
5915 
5916 void cgroup_path_from_kernfs_id(u64 id, char *buf, size_t buflen)
5917 {
5918         struct kernfs_node *kn;
5919 
5920         kn = kernfs_find_and_get_node_by_id(cgrp_dfl_root.kf_root, id);
5921         if (!kn)
5922                 return;
5923         kernfs_path(kn, buf, buflen);
5924         kernfs_put(kn);
5925 }
5926 
5927 /*
5928  * cgroup_get_from_id : get the cgroup associated with cgroup id
5929  * @id: cgroup id
5930  * On success return the cgrp, on failure return NULL
5931  */
5932 struct cgroup *cgroup_get_from_id(u64 id)
5933 {
5934         struct kernfs_node *kn;
5935         struct cgroup *cgrp = NULL;
5936 
5937         kn = kernfs_find_and_get_node_by_id(cgrp_dfl_root.kf_root, id);
5938         if (!kn)
5939                 goto out;
5940 
5941         rcu_read_lock();
5942 
5943         cgrp = rcu_dereference(*(void __rcu __force **)&kn->priv);
5944         if (cgrp && !cgroup_tryget(cgrp))
5945                 cgrp = NULL;
5946 
5947         rcu_read_unlock();
5948 
5949         kernfs_put(kn);
5950 out:
5951         return cgrp;
5952 }
5953 EXPORT_SYMBOL_GPL(cgroup_get_from_id);
5954 
5955 /*
5956  * proc_cgroup_show()
5957  *  - Print task's cgroup paths into seq_file, one line for each hierarchy
5958  *  - Used for /proc/<pid>/cgroup.
5959  */
5960 int proc_cgroup_show(struct seq_file *m, struct pid_namespace *ns,
5961                      struct pid *pid, struct task_struct *tsk)
5962 {
5963         char *buf;
5964         int retval;
5965         struct cgroup_root *root;
5966 
5967         retval = -ENOMEM;
5968         buf = kmalloc(PATH_MAX, GFP_KERNEL);
5969         if (!buf)
5970                 goto out;
5971 
5972         mutex_lock(&cgroup_mutex);
5973         spin_lock_irq(&css_set_lock);
5974 
5975         for_each_root(root) {
5976                 struct cgroup_subsys *ss;
5977                 struct cgroup *cgrp;
5978                 int ssid, count = 0;
5979 
5980                 if (root == &cgrp_dfl_root && !cgrp_dfl_visible)
5981                         continue;
5982 
5983                 seq_printf(m, "%d:", root->hierarchy_id);
5984                 if (root != &cgrp_dfl_root)
5985                         for_each_subsys(ss, ssid)
5986                                 if (root->subsys_mask & (1 << ssid))
5987                                         seq_printf(m, "%s%s", count++ ? "," : "",
5988                                                    ss->legacy_name);
5989                 if (strlen(root->name))
5990                         seq_printf(m, "%sname=%s", count ? "," : "",
5991                                    root->name);
5992                 seq_putc(m, ':');
5993 
5994                 cgrp = task_cgroup_from_root(tsk, root);
5995 
5996                 /*
5997                  * On traditional hierarchies, all zombie tasks show up as
5998                  * belonging to the root cgroup.  On the default hierarchy,
5999                  * while a zombie doesn't show up in "cgroup.procs" and
6000                  * thus can't be migrated, its /proc/PID/cgroup keeps
6001                  * reporting the cgroup it belonged to before exiting.  If
6002                  * the cgroup is removed before the zombie is reaped,
6003                  * " (deleted)" is appended to the cgroup path.
6004                  */
6005                 if (cgroup_on_dfl(cgrp) || !(tsk->flags & PF_EXITING)) {
6006                         retval = cgroup_path_ns_locked(cgrp, buf, PATH_MAX,
6007                                                 current->nsproxy->cgroup_ns);
6008                         if (retval >= PATH_MAX)
6009                                 retval = -ENAMETOOLONG;
6010                         if (retval < 0)
6011                                 goto out_unlock;
6012 
6013                         seq_puts(m, buf);
6014                 } else {
6015                         seq_puts(m, "/");
6016                 }
6017 
6018                 if (cgroup_on_dfl(cgrp) && cgroup_is_dead(cgrp))
6019                         seq_puts(m, " (deleted)\n");
6020                 else
6021                         seq_putc(m, '\n');
6022         }
6023 
6024         retval = 0;
6025 out_unlock:
6026         spin_unlock_irq(&css_set_lock);
6027         mutex_unlock(&cgroup_mutex);
6028         kfree(buf);
6029 out:
6030         return retval;
6031 }
6032 
6033 /**
6034  * cgroup_fork - initialize cgroup related fields during copy_process()
6035  * @child: pointer to task_struct of forking parent process.
6036  *
6037  * A task is associated with the init_css_set until cgroup_post_fork()
6038  * attaches it to the target css_set.
6039  */
6040 void cgroup_fork(struct task_struct *child)
6041 {
6042         RCU_INIT_POINTER(child->cgroups, &init_css_set);
6043         INIT_LIST_HEAD(&child->cg_list);
6044 }
6045 
6046 static struct cgroup *cgroup_get_from_file(struct file *f)
6047 {
6048         struct cgroup_subsys_state *css;
6049         struct cgroup *cgrp;
6050 
6051         css = css_tryget_online_from_dir(f->f_path.dentry, NULL);
6052         if (IS_ERR(css))
6053                 return ERR_CAST(css);
6054 
6055         cgrp = css->cgroup;
6056         if (!cgroup_on_dfl(cgrp)) {
6057                 cgroup_put(cgrp);
6058                 return ERR_PTR(-EBADF);
6059         }
6060 
6061         return cgrp;
6062 }
6063 
6064 /**
6065  * cgroup_css_set_fork - find or create a css_set for a child process
6066  * @kargs: the arguments passed to create the child process
6067  *
6068  * This functions finds or creates a new css_set which the child
6069  * process will be attached to in cgroup_post_fork(). By default,
6070  * the child process will be given the same css_set as its parent.
6071  *
6072  * If CLONE_INTO_CGROUP is specified this function will try to find an
6073  * existing css_set which includes the requested cgroup and if not create
6074  * a new css_set that the child will be attached to later. If this function
6075  * succeeds it will hold cgroup_threadgroup_rwsem on return. If
6076  * CLONE_INTO_CGROUP is requested this function will grab cgroup mutex
6077  * before grabbing cgroup_threadgroup_rwsem and will hold a reference
6078  * to the target cgroup.
6079  */
6080 static int cgroup_css_set_fork(struct kernel_clone_args *kargs)
6081         __acquires(&cgroup_mutex) __acquires(&cgroup_threadgroup_rwsem)
6082 {
6083         int ret;
6084         struct cgroup *dst_cgrp = NULL;
6085         struct css_set *cset;
6086         struct super_block *sb;
6087         struct file *f;
6088 
6089         if (kargs->flags & CLONE_INTO_CGROUP)
6090                 mutex_lock(&cgroup_mutex);
6091 
6092         cgroup_threadgroup_change_begin(current);
6093 
6094         spin_lock_irq(&css_set_lock);
6095         cset = task_css_set(current);
6096         get_css_set(cset);
6097         spin_unlock_irq(&css_set_lock);
6098 
6099         if (!(kargs->flags & CLONE_INTO_CGROUP)) {
6100                 kargs->cset = cset;
6101                 return 0;
6102         }
6103 
6104         f = fget_raw(kargs->cgroup);
6105         if (!f) {
6106                 ret = -EBADF;
6107                 goto err;
6108         }
6109         sb = f->f_path.dentry->d_sb;
6110 
6111         dst_cgrp = cgroup_get_from_file(f);
6112         if (IS_ERR(dst_cgrp)) {
6113                 ret = PTR_ERR(dst_cgrp);
6114                 dst_cgrp = NULL;
6115                 goto err;
6116         }
6117 
6118         if (cgroup_is_dead(dst_cgrp)) {
6119                 ret = -ENODEV;
6120                 goto err;
6121         }
6122 
6123         /*
6124          * Verify that we the target cgroup is writable for us. This is
6125          * usually done by the vfs layer but since we're not going through
6126          * the vfs layer here we need to do it "manually".
6127          */
6128         ret = cgroup_may_write(dst_cgrp, sb);
6129         if (ret)
6130                 goto err;
6131 
6132         ret = cgroup_attach_permissions(cset->dfl_cgrp, dst_cgrp, sb,
6133                                         !(kargs->flags & CLONE_THREAD));
6134         if (ret)
6135                 goto err;
6136 
6137         kargs->cset = find_css_set(cset, dst_cgrp);
6138         if (!kargs->cset) {
6139                 ret = -ENOMEM;
6140                 goto err;
6141         }
6142 
6143         put_css_set(cset);
6144         fput(f);
6145         kargs->cgrp = dst_cgrp;
6146         return ret;
6147 
6148 err:
6149         cgroup_threadgroup_change_end(current);
6150         mutex_unlock(&cgroup_mutex);
6151         if (f)
6152                 fput(f);
6153         if (dst_cgrp)
6154                 cgroup_put(dst_cgrp);
6155         put_css_set(cset);
6156         if (kargs->cset)
6157                 put_css_set(kargs->cset);
6158         return ret;
6159 }
6160 
6161 /**
6162  * cgroup_css_set_put_fork - drop references we took during fork
6163  * @kargs: the arguments passed to create the child process
6164  *
6165  * Drop references to the prepared css_set and target cgroup if
6166  * CLONE_INTO_CGROUP was requested.
6167  */
6168 static void cgroup_css_set_put_fork(struct kernel_clone_args *kargs)
6169         __releases(&cgroup_threadgroup_rwsem) __releases(&cgroup_mutex)
6170 {
6171         cgroup_threadgroup_change_end(current);
6172 
6173         if (kargs->flags & CLONE_INTO_CGROUP) {
6174                 struct cgroup *cgrp = kargs->cgrp;
6175                 struct css_set *cset = kargs->cset;
6176 
6177                 mutex_unlock(&cgroup_mutex);
6178 
6179                 if (cset) {
6180                         put_css_set(cset);
6181                         kargs->cset = NULL;
6182                 }
6183 
6184                 if (cgrp) {
6185                         cgroup_put(cgrp);
6186                         kargs->cgrp = NULL;
6187                 }
6188         }
6189 }
6190 
6191 /**
6192  * cgroup_can_fork - called on a new task before the process is exposed
6193  * @child: the child process
6194  *
6195  * This prepares a new css_set for the child process which the child will
6196  * be attached to in cgroup_post_fork().
6197  * This calls the subsystem can_fork() callbacks. If the cgroup_can_fork()
6198  * callback returns an error, the fork aborts with that error code. This
6199  * allows for a cgroup subsystem to conditionally allow or deny new forks.
6200  */
6201 int cgroup_can_fork(struct task_struct *child, struct kernel_clone_args *kargs)
6202 {
6203         struct cgroup_subsys *ss;
6204         int i, j, ret;
6205 
6206         ret = cgroup_css_set_fork(kargs);
6207         if (ret)
6208                 return ret;
6209 
6210         do_each_subsys_mask(ss, i, have_canfork_callback) {
6211                 ret = ss->can_fork(child, kargs->cset);
6212                 if (ret)
6213                         goto out_revert;
6214         } while_each_subsys_mask();
6215 
6216         return 0;
6217 
6218 out_revert:
6219         for_each_subsys(ss, j) {
6220                 if (j >= i)
6221                         break;
6222                 if (ss->cancel_fork)
6223                         ss->cancel_fork(child, kargs->cset);
6224         }
6225 
6226         cgroup_css_set_put_fork(kargs);
6227 
6228         return ret;
6229 }
6230 
6231 /**
6232  * cgroup_cancel_fork - called if a fork failed after cgroup_can_fork()
6233  * @child: the child process
6234  * @kargs: the arguments passed to create the child process
6235  *
6236  * This calls the cancel_fork() callbacks if a fork failed *after*
6237  * cgroup_can_fork() succeeded and cleans up references we took to
6238  * prepare a new css_set for the child process in cgroup_can_fork().
6239  */
6240 void cgroup_cancel_fork(struct task_struct *child,
6241                         struct kernel_clone_args *kargs)
6242 {
6243         struct cgroup_subsys *ss;
6244         int i;
6245 
6246         for_each_subsys(ss, i)
6247                 if (ss->cancel_fork)
6248                         ss->cancel_fork(child, kargs->cset);
6249 
6250         cgroup_css_set_put_fork(kargs);
6251 }
6252 
6253 /**
6254  * cgroup_post_fork - finalize cgroup setup for the child process
6255  * @child: the child process
6256  *
6257  * Attach the child process to its css_set calling the subsystem fork()
6258  * callbacks.
6259  */
6260 void cgroup_post_fork(struct task_struct *child,
6261                       struct kernel_clone_args *kargs)
6262         __releases(&cgroup_threadgroup_rwsem) __releases(&cgroup_mutex)
6263 {
6264         unsigned long cgrp_flags = 0;
6265         bool kill = false;
6266         struct cgroup_subsys *ss;
6267         struct css_set *cset;
6268         int i;
6269 
6270         cset = kargs->cset;
6271         kargs->cset = NULL;
6272 
6273         spin_lock_irq(&css_set_lock);
6274 
6275         /* init tasks are special, only link regular threads */
6276         if (likely(child->pid)) {
6277                 if (kargs->cgrp)
6278                         cgrp_flags = kargs->cgrp->flags;
6279                 else
6280                         cgrp_flags = cset->dfl_cgrp->flags;
6281 
6282                 WARN_ON_ONCE(!list_empty(&child->cg_list));
6283                 cset->nr_tasks++;
6284                 css_set_move_task(child, NULL, cset, false);
6285         } else {
6286                 put_css_set(cset);
6287                 cset = NULL;
6288         }
6289 
6290         if (!(child->flags & PF_KTHREAD)) {
6291                 if (unlikely(test_bit(CGRP_FREEZE, &cgrp_flags))) {
6292                         /*
6293                          * If the cgroup has to be frozen, the new task has
6294                          * too. Let's set the JOBCTL_TRAP_FREEZE jobctl bit to
6295                          * get the task into the frozen state.
6296                          */
6297                         spin_lock(&child->sighand->siglock);
6298                         WARN_ON_ONCE(child->frozen);
6299                         child->jobctl |= JOBCTL_TRAP_FREEZE;
6300                         spin_unlock(&child->sighand->siglock);
6301 
6302                         /*
6303                          * Calling cgroup_update_frozen() isn't required here,
6304                          * because it will be called anyway a bit later from
6305                          * do_freezer_trap(). So we avoid cgroup's transient
6306                          * switch from the frozen state and back.
6307                          */
6308                 }
6309 
6310                 /*
6311                  * If the cgroup is to be killed notice it now and take the
6312                  * child down right after we finished preparing it for
6313                  * userspace.
6314                  */
6315                 kill = test_bit(CGRP_KILL, &cgrp_flags);
6316         }
6317 
6318         spin_unlock_irq(&css_set_lock);
6319 
6320         /*
6321          * Call ss->fork().  This must happen after @child is linked on
6322          * css_set; otherwise, @child might change state between ->fork()
6323          * and addition to css_set.
6324          */
6325         do_each_subsys_mask(ss, i, have_fork_callback) {
6326                 ss->fork(child);
6327         } while_each_subsys_mask();
6328 
6329         /* Make the new cset the root_cset of the new cgroup namespace. */
6330         if (kargs->flags & CLONE_NEWCGROUP) {
6331                 struct css_set *rcset = child->nsproxy->cgroup_ns->root_cset;
6332 
6333                 get_css_set(cset);
6334                 child->nsproxy->cgroup_ns->root_cset = cset;
6335                 put_css_set(rcset);
6336         }
6337 
6338         /* Cgroup has to be killed so take down child immediately. */
6339         if (unlikely(kill))
6340                 do_send_sig_info(SIGKILL, SEND_SIG_NOINFO, child, PIDTYPE_TGID);
6341 
6342         cgroup_css_set_put_fork(kargs);
6343 }
6344 
6345 /**
6346  * cgroup_exit - detach cgroup from exiting task
6347  * @tsk: pointer to task_struct of exiting process
6348  *
6349  * Description: Detach cgroup from @tsk.
6350  *
6351  */
6352 void cgroup_exit(struct task_struct *tsk)
6353 {
6354         struct cgroup_subsys *ss;
6355         struct css_set *cset;
6356         int i;
6357 
6358         spin_lock_irq(&css_set_lock);
6359 
6360         WARN_ON_ONCE(list_empty(&tsk->cg_list));
6361         cset = task_css_set(tsk);
6362         css_set_move_task(tsk, cset, NULL, false);
6363         list_add_tail(&tsk->cg_list, &cset->dying_tasks);
6364         cset->nr_tasks--;
6365 
6366         WARN_ON_ONCE(cgroup_task_frozen(tsk));
6367         if (unlikely(!(tsk->flags & PF_KTHREAD) &&
6368                      test_bit(CGRP_FREEZE, &task_dfl_cgroup(tsk)->flags)))
6369                 cgroup_update_frozen(task_dfl_cgroup(tsk));
6370 
6371         spin_unlock_irq(&css_set_lock);
6372 
6373         /* see cgroup_post_fork() for details */
6374         do_each_subsys_mask(ss, i, have_exit_callback) {
6375                 ss->exit(tsk);
6376         } while_each_subsys_mask();
6377 }
6378 
6379 void cgroup_release(struct task_struct *task)
6380 {
6381         struct cgroup_subsys *ss;
6382         int ssid;
6383 
6384         do_each_subsys_mask(ss, ssid, have_release_callback) {
6385                 ss->release(task);
6386         } while_each_subsys_mask();
6387 
6388         spin_lock_irq(&css_set_lock);
6389         css_set_skip_task_iters(task_css_set(task), task);
6390         list_del_init(&task->cg_list);
6391         spin_unlock_irq(&css_set_lock);
6392 }
6393 
6394 void cgroup_free(struct task_struct *task)
6395 {
6396         struct css_set *cset = task_css_set(task);
6397         put_css_set(cset);
6398 }
6399 
6400 static int __init cgroup_disable(char *str)
6401 {
6402         struct cgroup_subsys *ss;
6403         char *token;
6404         int i;
6405 
6406         while ((token = strsep(&str, ",")) != NULL) {
6407                 if (!*token)
6408                         continue;
6409 
6410                 for_each_subsys(ss, i) {
6411                         if (strcmp(token, ss->name) &&
6412                             strcmp(token, ss->legacy_name))
6413                                 continue;
6414 
6415                         static_branch_disable(cgroup_subsys_enabled_key[i]);
6416                         pr_info("Disabling %s control group subsystem\n",
6417                                 ss->name);
6418                 }
6419 
6420                 for (i = 0; i < OPT_FEATURE_COUNT; i++) {
6421                         if (strcmp(token, cgroup_opt_feature_names[i]))
6422                                 continue;
6423                         cgroup_feature_disable_mask |= 1 << i;
6424                         pr_info("Disabling %s control group feature\n",
6425                                 cgroup_opt_feature_names[i]);
6426                         break;
6427                 }
6428         }
6429         return 1;
6430 }
6431 __setup("cgroup_disable=", cgroup_disable);
6432 
6433 void __init __weak enable_debug_cgroup(void) { }
6434 
6435 static int __init enable_cgroup_debug(char *str)
6436 {
6437         cgroup_debug = true;
6438         enable_debug_cgroup();
6439         return 1;
6440 }
6441 __setup("cgroup_debug", enable_cgroup_debug);
6442 
6443 /**
6444  * css_tryget_online_from_dir - get corresponding css from a cgroup dentry
6445  * @dentry: directory dentry of interest
6446  * @ss: subsystem of interest
6447  *
6448  * If @dentry is a directory for a cgroup which has @ss enabled on it, try
6449  * to get the corresponding css and return it.  If such css doesn't exist
6450  * or can't be pinned, an ERR_PTR value is returned.
6451  */
6452 struct cgroup_subsys_state *css_tryget_online_from_dir(struct dentry *dentry,
6453                                                        struct cgroup_subsys *ss)
6454 {
6455         struct kernfs_node *kn = kernfs_node_from_dentry(dentry);
6456         struct file_system_type *s_type = dentry->d_sb->s_type;
6457         struct cgroup_subsys_state *css = NULL;
6458         struct cgroup *cgrp;
6459 
6460         /* is @dentry a cgroup dir? */
6461         if ((s_type != &cgroup_fs_type && s_type != &cgroup2_fs_type) ||
6462             !kn || kernfs_type(kn) != KERNFS_DIR)
6463                 return ERR_PTR(-EBADF);
6464 
6465         rcu_read_lock();
6466 
6467         /*
6468          * This path doesn't originate from kernfs and @kn could already
6469          * have been or be removed at any point.  @kn->priv is RCU
6470          * protected for this access.  See css_release_work_fn() for details.
6471          */
6472         cgrp = rcu_dereference(*(void __rcu __force **)&kn->priv);
6473         if (cgrp)
6474                 css = cgroup_css(cgrp, ss);
6475 
6476         if (!css || !css_tryget_online(css))
6477                 css = ERR_PTR(-ENOENT);
6478 
6479         rcu_read_unlock();
6480         return css;
6481 }
6482 
6483 /**
6484  * css_from_id - lookup css by id
6485  * @id: the cgroup id
6486  * @ss: cgroup subsys to be looked into
6487  *
6488  * Returns the css if there's valid one with @id, otherwise returns NULL.
6489  * Should be called under rcu_read_lock().
6490  */
6491 struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss)
6492 {
6493         WARN_ON_ONCE(!rcu_read_lock_held());
6494         return idr_find(&ss->css_idr, id);
6495 }
6496 
6497 /**
6498  * cgroup_get_from_path - lookup and get a cgroup from its default hierarchy path
6499  * @path: path on the default hierarchy
6500  *
6501  * Find the cgroup at @path on the default hierarchy, increment its
6502  * reference count and return it.  Returns pointer to the found cgroup on
6503  * success, ERR_PTR(-ENOENT) if @path doesn't exist or if the cgroup has already
6504  * been released and ERR_PTR(-ENOTDIR) if @path points to a non-directory.
6505  */
6506 struct cgroup *cgroup_get_from_path(const char *path)
6507 {
6508         struct kernfs_node *kn;
6509         struct cgroup *cgrp = ERR_PTR(-ENOENT);
6510 
6511         kn = kernfs_walk_and_get(cgrp_dfl_root.cgrp.kn, path);
6512         if (!kn)
6513                 goto out;
6514 
6515         if (kernfs_type(kn) != KERNFS_DIR) {
6516                 cgrp = ERR_PTR(-ENOTDIR);
6517                 goto out_kernfs;
6518         }
6519 
6520         rcu_read_lock();
6521 
6522         cgrp = rcu_dereference(*(void __rcu __force **)&kn->priv);
6523         if (!cgrp || !cgroup_tryget(cgrp))
6524                 cgrp = ERR_PTR(-ENOENT);
6525 
6526         rcu_read_unlock();
6527 
6528 out_kernfs:
6529         kernfs_put(kn);
6530 out:
6531         return cgrp;
6532 }
6533 EXPORT_SYMBOL_GPL(cgroup_get_from_path);
6534 
6535 /**
6536  * cgroup_get_from_fd - get a cgroup pointer from a fd
6537  * @fd: fd obtained by open(cgroup2_dir)
6538  *
6539  * Find the cgroup from a fd which should be obtained
6540  * by opening a cgroup directory.  Returns a pointer to the
6541  * cgroup on success. ERR_PTR is returned if the cgroup
6542  * cannot be found.
6543  */
6544 struct cgroup *cgroup_get_from_fd(int fd)
6545 {
6546         struct cgroup *cgrp;
6547         struct file *f;
6548 
6549         f = fget_raw(fd);
6550         if (!f)
6551                 return ERR_PTR(-EBADF);
6552 
6553         cgrp = cgroup_get_from_file(f);
6554         fput(f);
6555         return cgrp;
6556 }
6557 EXPORT_SYMBOL_GPL(cgroup_get_from_fd);
6558 
6559 static u64 power_of_ten(int power)
6560 {
6561         u64 v = 1;
6562         while (power--)
6563                 v *= 10;
6564         return v;
6565 }
6566 
6567 /**
6568  * cgroup_parse_float - parse a floating number
6569  * @input: input string
6570  * @dec_shift: number of decimal digits to shift
6571  * @v: output
6572  *
6573  * Parse a decimal floating point number in @input and store the result in
6574  * @v with decimal point right shifted @dec_shift times.  For example, if
6575  * @input is "12.3456" and @dec_shift is 3, *@v will be set to 12345.
6576  * Returns 0 on success, -errno otherwise.
6577  *
6578  * There's nothing cgroup specific about this function except that it's
6579  * currently the only user.
6580  */
6581 int cgroup_parse_float(const char *input, unsigned dec_shift, s64 *v)
6582 {
6583         s64 whole, frac = 0;
6584         int fstart = 0, fend = 0, flen;
6585 
6586         if (!sscanf(input, "%lld.%n%lld%n", &whole, &fstart, &frac, &fend))
6587                 return -EINVAL;
6588         if (frac < 0)
6589                 return -EINVAL;
6590 
6591         flen = fend > fstart ? fend - fstart : 0;
6592         if (flen < dec_shift)
6593                 frac *= power_of_ten(dec_shift - flen);
6594         else
6595                 frac = DIV_ROUND_CLOSEST_ULL(frac, power_of_ten(flen - dec_shift));
6596 
6597         *v = whole * power_of_ten(dec_shift) + frac;
6598         return 0;
6599 }
6600 
6601 /*
6602  * sock->sk_cgrp_data handling.  For more info, see sock_cgroup_data
6603  * definition in cgroup-defs.h.
6604  */
6605 #ifdef CONFIG_SOCK_CGROUP_DATA
6606 
6607 void cgroup_sk_alloc(struct sock_cgroup_data *skcd)
6608 {
6609         struct cgroup *cgroup;
6610 
6611         rcu_read_lock();
6612         /* Don't associate the sock with unrelated interrupted task's cgroup. */
6613         if (in_interrupt()) {
6614                 cgroup = &cgrp_dfl_root.cgrp;
6615                 cgroup_get(cgroup);
6616                 goto out;
6617         }
6618 
6619         while (true) {
6620                 struct css_set *cset;
6621 
6622                 cset = task_css_set(current);
6623                 if (likely(cgroup_tryget(cset->dfl_cgrp))) {
6624                         cgroup = cset->dfl_cgrp;
6625                         break;
6626                 }
6627                 cpu_relax();
6628         }
6629 out:
6630         skcd->cgroup = cgroup;
6631         cgroup_bpf_get(cgroup);
6632         rcu_read_unlock();
6633 }
6634 
6635 void cgroup_sk_clone(struct sock_cgroup_data *skcd)
6636 {
6637         struct cgroup *cgrp = sock_cgroup_ptr(skcd);
6638 
6639         /*
6640          * We might be cloning a socket which is left in an empty
6641          * cgroup and the cgroup might have already been rmdir'd.
6642          * Don't use cgroup_get_live().
6643          */
6644         cgroup_get(cgrp);
6645         cgroup_bpf_get(cgrp);
6646 }
6647 
6648 void cgroup_sk_free(struct sock_cgroup_data *skcd)
6649 {
6650         struct cgroup *cgrp = sock_cgroup_ptr(skcd);
6651 
6652         cgroup_bpf_put(cgrp);
6653         cgroup_put(cgrp);
6654 }
6655 
6656 #endif  /* CONFIG_SOCK_CGROUP_DATA */
6657 
6658 #ifdef CONFIG_SYSFS
6659 static ssize_t show_delegatable_files(struct cftype *files, char *buf,
6660                                       ssize_t size, const char *prefix)
6661 {
6662         struct cftype *cft;
6663         ssize_t ret = 0;
6664 
6665         for (cft = files; cft && cft->name[0] != '\0'; cft++) {
6666                 if (!(cft->flags & CFTYPE_NS_DELEGATABLE))
6667                         continue;
6668 
6669                 if ((cft->flags & CFTYPE_PRESSURE) && !cgroup_psi_enabled())
6670                         continue;
6671 
6672                 if (prefix)
6673                         ret += snprintf(buf + ret, size - ret, "%s.", prefix);
6674 
6675                 ret += snprintf(buf + ret, size - ret, "%s\n", cft->name);
6676 
6677                 if (WARN_ON(ret >= size))
6678                         break;
6679         }
6680 
6681         return ret;
6682 }
6683 
6684 static ssize_t delegate_show(struct kobject *kobj, struct kobj_attribute *attr,
6685                               char *buf)
6686 {
6687         struct cgroup_subsys *ss;
6688         int ssid;
6689         ssize_t ret = 0;
6690 
6691         ret = show_delegatable_files(cgroup_base_files, buf, PAGE_SIZE - ret,
6692                                      NULL);
6693 
6694         for_each_subsys(ss, ssid)
6695                 ret += show_delegatable_files(ss->dfl_cftypes, buf + ret,
6696                                               PAGE_SIZE - ret,
6697                                               cgroup_subsys_name[ssid]);
6698 
6699         return ret;
6700 }
6701 static struct kobj_attribute cgroup_delegate_attr = __ATTR_RO(delegate);
6702 
6703 static ssize_t features_show(struct kobject *kobj, struct kobj_attribute *attr,
6704                              char *buf)
6705 {
6706         return snprintf(buf, PAGE_SIZE,
6707                         "nsdelegate\n"
6708                         "memory_localevents\n"
6709                         "memory_recursiveprot\n");
6710 }
6711 static struct kobj_attribute cgroup_features_attr = __ATTR_RO(features);
6712 
6713 static struct attribute *cgroup_sysfs_attrs[] = {
6714         &cgroup_delegate_attr.attr,
6715         &cgroup_features_attr.attr,
6716         NULL,
6717 };
6718 
6719 static const struct attribute_group cgroup_sysfs_attr_group = {
6720         .attrs = cgroup_sysfs_attrs,
6721         .name = "cgroup",
6722 };
6723 
6724 static int __init cgroup_sysfs_init(void)
6725 {
6726         return sysfs_create_group(kernel_kobj, &cgroup_sysfs_attr_group);
6727 }
6728 subsys_initcall(cgroup_sysfs_init);
6729 
6730 #endif /* CONFIG_SYSFS */
6731 

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

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

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

osdn.jp