LCOV - code coverage report
Current view: top level - kernel/cgroup - cgroup.c (source / functions) Hit Total Coverage
Test: landlock.info Lines: 1339 2153 62.2 %
Date: 2021-04-22 12:43:58 Functions: 114 177 64.4 %

          Line data    Source code
       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             :  * cgroup_mutex is the master lock.  Any modification to cgroup or its
      72             :  * hierarchy must be performed while holding it.
      73             :  *
      74             :  * css_set_lock protects task->cgroups pointer, the list of css_set
      75             :  * objects, and the chain of tasks off each css_set.
      76             :  *
      77             :  * These locks are exported if CONFIG_PROVE_RCU so that accessors in
      78             :  * cgroup.h can use them for lockdep annotations.
      79             :  */
      80             : DEFINE_MUTEX(cgroup_mutex);
      81             : DEFINE_SPINLOCK(css_set_lock);
      82             : 
      83             : #ifdef CONFIG_PROVE_RCU
      84             : EXPORT_SYMBOL_GPL(cgroup_mutex);
      85             : EXPORT_SYMBOL_GPL(css_set_lock);
      86             : #endif
      87             : 
      88             : DEFINE_SPINLOCK(trace_cgroup_path_lock);
      89             : char trace_cgroup_path[TRACE_CGROUP_PATH_LEN];
      90             : bool cgroup_debug __read_mostly;
      91             : 
      92             : /*
      93             :  * Protects cgroup_idr and css_idr so that IDs can be released without
      94             :  * grabbing cgroup_mutex.
      95             :  */
      96             : static DEFINE_SPINLOCK(cgroup_idr_lock);
      97             : 
      98             : /*
      99             :  * Protects cgroup_file->kn for !self csses.  It synchronizes notifications
     100             :  * against file removal/re-creation across css hiding.
     101             :  */
     102             : static DEFINE_SPINLOCK(cgroup_file_kn_lock);
     103             : 
     104             : DEFINE_PERCPU_RWSEM(cgroup_threadgroup_rwsem);
     105             : 
     106             : #define cgroup_assert_mutex_or_rcu_locked()                             \
     107             :         RCU_LOCKDEP_WARN(!rcu_read_lock_held() &&                       \
     108             :                            !lockdep_is_held(&cgroup_mutex),         \
     109             :                            "cgroup_mutex or RCU read lock required");
     110             : 
     111             : /*
     112             :  * cgroup destruction makes heavy use of work items and there can be a lot
     113             :  * of concurrent destructions.  Use a separate workqueue so that cgroup
     114             :  * destruction work items don't end up filling up max_active of system_wq
     115             :  * which may lead to deadlock.
     116             :  */
     117             : static struct workqueue_struct *cgroup_destroy_wq;
     118             : 
     119             : /* generate an array of cgroup subsystem pointers */
     120             : #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys,
     121             : struct cgroup_subsys *cgroup_subsys[] = {
     122             : #include <linux/cgroup_subsys.h>
     123             : };
     124             : #undef SUBSYS
     125             : 
     126             : /* array of cgroup subsystem names */
     127             : #define SUBSYS(_x) [_x ## _cgrp_id] = #_x,
     128             : static const char *cgroup_subsys_name[] = {
     129             : #include <linux/cgroup_subsys.h>
     130             : };
     131             : #undef SUBSYS
     132             : 
     133             : /* array of static_keys for cgroup_subsys_enabled() and cgroup_subsys_on_dfl() */
     134             : #define SUBSYS(_x)                                                              \
     135             :         DEFINE_STATIC_KEY_TRUE(_x ## _cgrp_subsys_enabled_key);                 \
     136             :         DEFINE_STATIC_KEY_TRUE(_x ## _cgrp_subsys_on_dfl_key);                  \
     137             :         EXPORT_SYMBOL_GPL(_x ## _cgrp_subsys_enabled_key);                      \
     138             :         EXPORT_SYMBOL_GPL(_x ## _cgrp_subsys_on_dfl_key);
     139             : #include <linux/cgroup_subsys.h>
     140             : #undef SUBSYS
     141             : 
     142             : #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys_enabled_key,
     143             : static struct static_key_true *cgroup_subsys_enabled_key[] = {
     144             : #include <linux/cgroup_subsys.h>
     145             : };
     146             : #undef SUBSYS
     147             : 
     148             : #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys_on_dfl_key,
     149             : static struct static_key_true *cgroup_subsys_on_dfl_key[] = {
     150             : #include <linux/cgroup_subsys.h>
     151             : };
     152             : #undef SUBSYS
     153             : 
     154             : static DEFINE_PER_CPU(struct cgroup_rstat_cpu, cgrp_dfl_root_rstat_cpu);
     155             : 
     156             : /* the default hierarchy */
     157             : struct cgroup_root cgrp_dfl_root = { .cgrp.rstat_cpu = &cgrp_dfl_root_rstat_cpu };
     158             : EXPORT_SYMBOL_GPL(cgrp_dfl_root);
     159             : 
     160             : /*
     161             :  * The default hierarchy always exists but is hidden until mounted for the
     162             :  * first time.  This is for backward compatibility.
     163             :  */
     164             : static bool cgrp_dfl_visible;
     165             : 
     166             : /* some controllers are not supported in the default hierarchy */
     167             : static u16 cgrp_dfl_inhibit_ss_mask;
     168             : 
     169             : /* some controllers are implicitly enabled on the default hierarchy */
     170             : static u16 cgrp_dfl_implicit_ss_mask;
     171             : 
     172             : /* some controllers can be threaded on the default hierarchy */
     173             : static u16 cgrp_dfl_threaded_ss_mask;
     174             : 
     175             : /* The list of hierarchy roots */
     176             : LIST_HEAD(cgroup_roots);
     177             : static int cgroup_root_count;
     178             : 
     179             : /* hierarchy ID allocation and mapping, protected by cgroup_mutex */
     180             : static DEFINE_IDR(cgroup_hierarchy_idr);
     181             : 
     182             : /*
     183             :  * Assign a monotonically increasing serial number to csses.  It guarantees
     184             :  * cgroups with bigger numbers are newer than those with smaller numbers.
     185             :  * Also, as csses are always appended to the parent's ->children list, it
     186             :  * guarantees that sibling csses are always sorted in the ascending serial
     187             :  * number order on the list.  Protected by cgroup_mutex.
     188             :  */
     189             : static u64 css_serial_nr_next = 1;
     190             : 
     191             : /*
     192             :  * These bitmasks identify subsystems with specific features to avoid
     193             :  * having to do iterative checks repeatedly.
     194             :  */
     195             : static u16 have_fork_callback __read_mostly;
     196             : static u16 have_exit_callback __read_mostly;
     197             : static u16 have_release_callback __read_mostly;
     198             : static u16 have_canfork_callback __read_mostly;
     199             : 
     200             : /* cgroup namespace for init task */
     201             : struct cgroup_namespace init_cgroup_ns = {
     202             :         .ns.count       = REFCOUNT_INIT(2),
     203             :         .user_ns        = &init_user_ns,
     204             :         .ns.ops         = &cgroupns_operations,
     205             :         .ns.inum        = PROC_CGROUP_INIT_INO,
     206             :         .root_cset      = &init_css_set,
     207             : };
     208             : 
     209             : static struct file_system_type cgroup2_fs_type;
     210             : static struct cftype cgroup_base_files[];
     211             : 
     212             : static int cgroup_apply_control(struct cgroup *cgrp);
     213             : static void cgroup_finalize_control(struct cgroup *cgrp, int ret);
     214             : static void css_task_iter_skip(struct css_task_iter *it,
     215             :                                struct task_struct *task);
     216             : static int cgroup_destroy_locked(struct cgroup *cgrp);
     217             : static struct cgroup_subsys_state *css_create(struct cgroup *cgrp,
     218             :                                               struct cgroup_subsys *ss);
     219             : static void css_release(struct percpu_ref *ref);
     220             : static void kill_css(struct cgroup_subsys_state *css);
     221             : static int cgroup_addrm_files(struct cgroup_subsys_state *css,
     222             :                               struct cgroup *cgrp, struct cftype cfts[],
     223             :                               bool is_add);
     224             : 
     225             : /**
     226             :  * cgroup_ssid_enabled - cgroup subsys enabled test by subsys ID
     227             :  * @ssid: subsys ID of interest
     228             :  *
     229             :  * cgroup_subsys_enabled() can only be used with literal subsys names which
     230             :  * is fine for individual subsystems but unsuitable for cgroup core.  This
     231             :  * is slower static_key_enabled() based test indexed by @ssid.
     232             :  */
     233           0 : bool cgroup_ssid_enabled(int ssid)
     234             : {
     235           0 :         if (CGROUP_SUBSYS_COUNT == 0)
     236           0 :                 return false;
     237             : 
     238             :         return static_key_enabled(cgroup_subsys_enabled_key[ssid]);
     239             : }
     240             : 
     241             : /**
     242             :  * cgroup_on_dfl - test whether a cgroup is on the default hierarchy
     243             :  * @cgrp: the cgroup of interest
     244             :  *
     245             :  * The default hierarchy is the v2 interface of cgroup and this function
     246             :  * can be used to test whether a cgroup is on the default hierarchy for
     247             :  * cases where a subsystem should behave differently depending on the
     248             :  * interface version.
     249             :  *
     250             :  * List of changed behaviors:
     251             :  *
     252             :  * - Mount options "noprefix", "xattr", "clone_children", "release_agent"
     253             :  *   and "name" are disallowed.
     254             :  *
     255             :  * - When mounting an existing superblock, mount options should match.
     256             :  *
     257             :  * - Remount is disallowed.
     258             :  *
     259             :  * - rename(2) is disallowed.
     260             :  *
     261             :  * - "tasks" is removed.  Everything should be at process granularity.  Use
     262             :  *   "cgroup.procs" instead.
     263             :  *
     264             :  * - "cgroup.procs" is not sorted.  pids will be unique unless they got
     265             :  *   recycled in-between reads.
     266             :  *
     267             :  * - "release_agent" and "notify_on_release" are removed.  Replacement
     268             :  *   notification mechanism will be implemented.
     269             :  *
     270             :  * - "cgroup.clone_children" is removed.
     271             :  *
     272             :  * - "cgroup.subtree_populated" is available.  Its value is 0 if the cgroup
     273             :  *   and its descendants contain no task; otherwise, 1.  The file also
     274             :  *   generates kernfs notification which can be monitored through poll and
     275             :  *   [di]notify when the value of the file changes.
     276             :  *
     277             :  * - cpuset: tasks will be kept in empty cpusets when hotplug happens and
     278             :  *   take masks of ancestors with non-empty cpus/mems, instead of being
     279             :  *   moved to an ancestor.
     280             :  *
     281             :  * - cpuset: a task can be moved into an empty cpuset, and again it takes
     282             :  *   masks of ancestors.
     283             :  *
     284             :  * - blkcg: blk-throttle becomes properly hierarchical.
     285             :  *
     286             :  * - debug: disallowed on the default hierarchy.
     287             :  */
     288        2722 : bool cgroup_on_dfl(const struct cgroup *cgrp)
     289             : {
     290        2722 :         return cgrp->root == &cgrp_dfl_root;
     291             : }
     292             : 
     293             : /* IDR wrappers which synchronize using cgroup_idr_lock */
     294             : static int cgroup_idr_alloc(struct idr *idr, void *ptr, int start, int end,
     295             :                             gfp_t gfp_mask)
     296             : {
     297             :         int ret;
     298             : 
     299             :         idr_preload(gfp_mask);
     300             :         spin_lock_bh(&cgroup_idr_lock);
     301             :         ret = idr_alloc(idr, ptr, start, end, gfp_mask & ~__GFP_DIRECT_RECLAIM);
     302             :         spin_unlock_bh(&cgroup_idr_lock);
     303             :         idr_preload_end();
     304             :         return ret;
     305             : }
     306             : 
     307           0 : static void *cgroup_idr_replace(struct idr *idr, void *ptr, int id)
     308             : {
     309           0 :         void *ret;
     310             : 
     311           0 :         spin_lock_bh(&cgroup_idr_lock);
     312           0 :         ret = idr_replace(idr, ptr, id);
     313           0 :         spin_unlock_bh(&cgroup_idr_lock);
     314           0 :         return ret;
     315             : }
     316             : 
     317           0 : static void cgroup_idr_remove(struct idr *idr, int id)
     318             : {
     319           0 :         spin_lock_bh(&cgroup_idr_lock);
     320           0 :         idr_remove(idr, id);
     321           0 :         spin_unlock_bh(&cgroup_idr_lock);
     322           0 : }
     323             : 
     324         112 : static bool cgroup_has_tasks(struct cgroup *cgrp)
     325             : {
     326         112 :         return cgrp->nr_populated_csets;
     327             : }
     328             : 
     329        1133 : bool cgroup_is_threaded(struct cgroup *cgrp)
     330             : {
     331        1133 :         return cgrp->dom_cgrp != cgrp;
     332             : }
     333             : 
     334             : /* can @cgrp host both domain and threaded children? */
     335         409 : static bool cgroup_is_mixable(struct cgroup *cgrp)
     336             : {
     337             :         /*
     338             :          * Root isn't under domain level resource control exempting it from
     339             :          * the no-internal-process constraint, so it can serve as a thread
     340             :          * root and a parent of resource domains at the same time.
     341             :          */
     342        1128 :         return !cgroup_parent(cgrp);
     343             : }
     344             : 
     345             : /* can @cgrp become a thread root? Should always be true for a thread root */
     346          99 : static bool cgroup_can_be_thread_root(struct cgroup *cgrp)
     347             : {
     348             :         /* mixables don't care */
     349          99 :         if (cgroup_is_mixable(cgrp))
     350             :                 return true;
     351             : 
     352             :         /* domain roots can't be nested under threaded */
     353          99 :         if (cgroup_is_threaded(cgrp))
     354             :                 return false;
     355             : 
     356             :         /* can only have either domain or threaded children */
     357          99 :         if (cgrp->nr_populated_domain_children)
     358             :                 return false;
     359             : 
     360             :         /* and no domain controllers can be enabled */
     361          99 :         if (cgrp->subtree_control & ~cgrp_dfl_threaded_ss_mask)
     362           0 :                 return false;
     363             : 
     364             :         return true;
     365             : }
     366             : 
     367             : /* is @cgrp root of a threaded subtree? */
     368         112 : bool cgroup_is_thread_root(struct cgroup *cgrp)
     369             : {
     370             :         /* thread root should be a domain */
     371         112 :         if (cgroup_is_threaded(cgrp))
     372             :                 return false;
     373             : 
     374             :         /* a domain w/ threaded children is a thread root */
     375         112 :         if (cgrp->nr_threaded_children)
     376             :                 return true;
     377             : 
     378             :         /*
     379             :          * A domain which has tasks and explicit threaded controllers
     380             :          * enabled is a thread root.
     381             :          */
     382         112 :         if (cgroup_has_tasks(cgrp) &&
     383           2 :             (cgrp->subtree_control & cgrp_dfl_threaded_ss_mask))
     384           0 :                 return true;
     385             : 
     386             :         return false;
     387             : }
     388             : 
     389             : /* a domain which isn't connected to the root w/o brekage can't be used */
     390          99 : static bool cgroup_is_valid_domain(struct cgroup *cgrp)
     391             : {
     392             :         /* the cgroup itself can be a thread root */
     393          99 :         if (cgroup_is_threaded(cgrp))
     394             :                 return false;
     395             : 
     396             :         /* but the ancestors can't be unless mixable */
     397         310 :         while ((cgrp = cgroup_parent(cgrp))) {
     398         211 :                 if (!cgroup_is_mixable(cgrp) && cgroup_is_thread_root(cgrp))
     399             :                         return false;
     400         211 :                 if (cgroup_is_threaded(cgrp))
     401             :                         return false;
     402             :         }
     403             : 
     404             :         return true;
     405             : }
     406             : 
     407             : /* subsystems visibly enabled on a cgroup */
     408         147 : static u16 cgroup_control(struct cgroup *cgrp)
     409             : {
     410         147 :         struct cgroup *parent = cgroup_parent(cgrp);
     411         147 :         u16 root_ss_mask = cgrp->root->subsys_mask;
     412             : 
     413         147 :         if (parent) {
     414         147 :                 u16 ss_mask = parent->subtree_control;
     415             : 
     416             :                 /* threaded cgroups can only have threaded controllers */
     417         147 :                 if (cgroup_is_threaded(cgrp))
     418           0 :                         ss_mask &= cgrp_dfl_threaded_ss_mask;
     419         147 :                 return ss_mask;
     420             :         }
     421             : 
     422           0 :         if (cgroup_on_dfl(cgrp))
     423           0 :                 root_ss_mask &= ~(cgrp_dfl_inhibit_ss_mask |
     424             :                                   cgrp_dfl_implicit_ss_mask);
     425             :         return root_ss_mask;
     426             : }
     427             : 
     428             : /* subsystems enabled on a cgroup */
     429          98 : static u16 cgroup_ss_mask(struct cgroup *cgrp)
     430             : {
     431          98 :         struct cgroup *parent = cgroup_parent(cgrp);
     432             : 
     433          98 :         if (parent) {
     434          98 :                 u16 ss_mask = parent->subtree_ss_mask;
     435             : 
     436             :                 /* threaded cgroups can only have threaded controllers */
     437          98 :                 if (cgroup_is_threaded(cgrp))
     438           0 :                         ss_mask &= cgrp_dfl_threaded_ss_mask;
     439          98 :                 return ss_mask;
     440             :         }
     441             : 
     442           0 :         return cgrp->root->subsys_mask;
     443             : }
     444             : 
     445             : /**
     446             :  * cgroup_css - obtain a cgroup's css for the specified subsystem
     447             :  * @cgrp: the cgroup of interest
     448             :  * @ss: the subsystem of interest (%NULL returns @cgrp->self)
     449             :  *
     450             :  * Return @cgrp's css (cgroup_subsys_state) associated with @ss.  This
     451             :  * function must be called either under cgroup_mutex or rcu_read_lock() and
     452             :  * the caller is responsible for pinning the returned css if it wants to
     453             :  * keep accessing it outside the said locks.  This function may return
     454             :  * %NULL if @cgrp doesn't have @subsys_id enabled.
     455             :  */
     456         444 : static struct cgroup_subsys_state *cgroup_css(struct cgroup *cgrp,
     457             :                                               struct cgroup_subsys *ss)
     458             : {
     459          50 :         if (ss)
     460           0 :                 return rcu_dereference_check(cgrp->subsys[ss->id],
     461             :                                         lockdep_is_held(&cgroup_mutex));
     462             :         else
     463         149 :                 return &cgrp->self;
     464             : }
     465             : 
     466             : /**
     467             :  * cgroup_tryget_css - try to get a cgroup's css for the specified subsystem
     468             :  * @cgrp: the cgroup of interest
     469             :  * @ss: the subsystem of interest
     470             :  *
     471             :  * Find and get @cgrp's css assocaited with @ss.  If the css doesn't exist
     472             :  * or is offline, %NULL is returned.
     473             :  */
     474             : static struct cgroup_subsys_state *cgroup_tryget_css(struct cgroup *cgrp,
     475             :                                                      struct cgroup_subsys *ss)
     476             : {
     477             :         struct cgroup_subsys_state *css;
     478             : 
     479             :         rcu_read_lock();
     480             :         css = cgroup_css(cgrp, ss);
     481             :         if (css && !css_tryget_online(css))
     482             :                 css = NULL;
     483             :         rcu_read_unlock();
     484             : 
     485             :         return css;
     486             : }
     487             : 
     488             : /**
     489             :  * cgroup_e_css_by_mask - obtain a cgroup's effective css for the specified ss
     490             :  * @cgrp: the cgroup of interest
     491             :  * @ss: the subsystem of interest (%NULL returns @cgrp->self)
     492             :  *
     493             :  * Similar to cgroup_css() but returns the effective css, which is defined
     494             :  * as the matching css of the nearest ancestor including self which has @ss
     495             :  * enabled.  If @ss is associated with the hierarchy @cgrp is on, this
     496             :  * function is guaranteed to return non-NULL css.
     497             :  */
     498             : static struct cgroup_subsys_state *cgroup_e_css_by_mask(struct cgroup *cgrp,
     499             :                                                         struct cgroup_subsys *ss)
     500             : {
     501             :         lockdep_assert_held(&cgroup_mutex);
     502             : 
     503             :         if (!ss)
     504             :                 return &cgrp->self;
     505             : 
     506             :         /*
     507             :          * This function is used while updating css associations and thus
     508             :          * can't test the csses directly.  Test ss_mask.
     509             :          */
     510             :         while (!(cgroup_ss_mask(cgrp) & (1 << ss->id))) {
     511             :                 cgrp = cgroup_parent(cgrp);
     512             :                 if (!cgrp)
     513             :                         return NULL;
     514             :         }
     515             : 
     516             :         return cgroup_css(cgrp, ss);
     517             : }
     518             : 
     519             : /**
     520             :  * cgroup_e_css - obtain a cgroup's effective css for the specified subsystem
     521             :  * @cgrp: the cgroup of interest
     522             :  * @ss: the subsystem of interest
     523             :  *
     524             :  * Find and get the effective css of @cgrp for @ss.  The effective css is
     525             :  * defined as the matching css of the nearest ancestor including self which
     526             :  * has @ss enabled.  If @ss is not mounted on the hierarchy @cgrp is on,
     527             :  * the root css is returned, so this function always returns a valid css.
     528             :  *
     529             :  * The returned css is not guaranteed to be online, and therefore it is the
     530             :  * callers responsibility to try get a reference for it.
     531             :  */
     532           0 : struct cgroup_subsys_state *cgroup_e_css(struct cgroup *cgrp,
     533             :                                          struct cgroup_subsys *ss)
     534             : {
     535           0 :         struct cgroup_subsys_state *css;
     536             : 
     537           0 :         do {
     538           0 :                 css = cgroup_css(cgrp, ss);
     539             : 
     540           0 :                 if (css)
     541           0 :                         return css;
     542           0 :                 cgrp = cgroup_parent(cgrp);
     543           0 :         } while (cgrp);
     544             : 
     545           0 :         return init_css_set.subsys[ss->id];
     546             : }
     547             : 
     548             : /**
     549             :  * cgroup_get_e_css - get a cgroup's effective css for the specified subsystem
     550             :  * @cgrp: the cgroup of interest
     551             :  * @ss: the subsystem of interest
     552             :  *
     553             :  * Find and get the effective css of @cgrp for @ss.  The effective css is
     554             :  * defined as the matching css of the nearest ancestor including self which
     555             :  * has @ss enabled.  If @ss is not mounted on the hierarchy @cgrp is on,
     556             :  * the root css is returned, so this function always returns a valid css.
     557             :  * The returned css must be put using css_put().
     558             :  */
     559           0 : struct cgroup_subsys_state *cgroup_get_e_css(struct cgroup *cgrp,
     560             :                                              struct cgroup_subsys *ss)
     561             : {
     562           0 :         struct cgroup_subsys_state *css;
     563             : 
     564           0 :         rcu_read_lock();
     565             : 
     566           0 :         do {
     567           0 :                 css = cgroup_css(cgrp, ss);
     568             : 
     569           0 :                 if (css && css_tryget_online(css))
     570           0 :                         goto out_unlock;
     571           0 :                 cgrp = cgroup_parent(cgrp);
     572           0 :         } while (cgrp);
     573             : 
     574           0 :         css = init_css_set.subsys[ss->id];
     575           0 :         css_get(css);
     576           0 : out_unlock:
     577           0 :         rcu_read_unlock();
     578           0 :         return css;
     579             : }
     580             : 
     581         294 : static void cgroup_get_live(struct cgroup *cgrp)
     582             : {
     583         294 :         WARN_ON_ONCE(cgroup_is_dead(cgrp));
     584         294 :         css_get(&cgrp->self);
     585         294 : }
     586             : 
     587             : /**
     588             :  * __cgroup_task_count - count the number of tasks in a cgroup. The caller
     589             :  * is responsible for taking the css_set_lock.
     590             :  * @cgrp: the cgroup in question
     591             :  */
     592           0 : int __cgroup_task_count(const struct cgroup *cgrp)
     593             : {
     594           0 :         int count = 0;
     595           0 :         struct cgrp_cset_link *link;
     596             : 
     597           0 :         lockdep_assert_held(&css_set_lock);
     598             : 
     599           0 :         list_for_each_entry(link, &cgrp->cset_links, cset_link)
     600           0 :                 count += link->cset->nr_tasks;
     601             : 
     602           0 :         return count;
     603             : }
     604             : 
     605             : /**
     606             :  * cgroup_task_count - count the number of tasks in a cgroup.
     607             :  * @cgrp: the cgroup in question
     608             :  */
     609           0 : int cgroup_task_count(const struct cgroup *cgrp)
     610             : {
     611           0 :         int count;
     612             : 
     613           0 :         spin_lock_irq(&css_set_lock);
     614           0 :         count = __cgroup_task_count(cgrp);
     615           0 :         spin_unlock_irq(&css_set_lock);
     616             : 
     617           0 :         return count;
     618             : }
     619             : 
     620         235 : struct cgroup_subsys_state *of_css(struct kernfs_open_file *of)
     621             : {
     622         235 :         struct cgroup *cgrp = of->kn->parent->priv;
     623         235 :         struct cftype *cft = of_cft(of);
     624             : 
     625             :         /*
     626             :          * This is open and unprotected implementation of cgroup_css().
     627             :          * seq_css() is only called from a kernfs file operation which has
     628             :          * an active reference on the file.  Because all the subsystem
     629             :          * files are drained before a css is disassociated with a cgroup,
     630             :          * the matching css from the cgroup's subsys table is guaranteed to
     631             :          * be and stay valid until the enclosing operation is complete.
     632             :          */
     633             :         if (cft->ss)
     634           0 :                 return rcu_dereference_raw(cgrp->subsys[cft->ss->id]);
     635             :         else
     636         235 :                 return &cgrp->self;
     637             : }
     638             : EXPORT_SYMBOL_GPL(of_css);
     639             : 
     640             : /**
     641             :  * for_each_css - iterate all css's of a cgroup
     642             :  * @css: the iteration cursor
     643             :  * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end
     644             :  * @cgrp: the target cgroup to iterate css's of
     645             :  *
     646             :  * Should be called under cgroup_[tree_]mutex.
     647             :  */
     648             : #define for_each_css(css, ssid, cgrp)                                   \
     649             :         for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++)     \
     650             :                 if (!((css) = rcu_dereference_check(                    \
     651             :                                 (cgrp)->subsys[(ssid)],                      \
     652             :                                 lockdep_is_held(&cgroup_mutex)))) { }       \
     653             :                 else
     654             : 
     655             : /**
     656             :  * for_each_e_css - iterate all effective css's of a cgroup
     657             :  * @css: the iteration cursor
     658             :  * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end
     659             :  * @cgrp: the target cgroup to iterate css's of
     660             :  *
     661             :  * Should be called under cgroup_[tree_]mutex.
     662             :  */
     663             : #define for_each_e_css(css, ssid, cgrp)                                     \
     664             :         for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++)         \
     665             :                 if (!((css) = cgroup_e_css_by_mask(cgrp,                    \
     666             :                                                    cgroup_subsys[(ssid)]))) \
     667             :                         ;                                                   \
     668             :                 else
     669             : 
     670             : /**
     671             :  * do_each_subsys_mask - filter for_each_subsys with a bitmask
     672             :  * @ss: the iteration cursor
     673             :  * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end
     674             :  * @ss_mask: the bitmask
     675             :  *
     676             :  * The block will only run for cases where the ssid-th bit (1 << ssid) of
     677             :  * @ss_mask is set.
     678             :  */
     679             : #define do_each_subsys_mask(ss, ssid, ss_mask) do {                     \
     680             :         unsigned long __ss_mask = (ss_mask);                            \
     681             :         if (!CGROUP_SUBSYS_COUNT) { /* to avoid spurious gcc warning */ \
     682             :                 (ssid) = 0;                                             \
     683             :                 break;                                                  \
     684             :         }                                                               \
     685             :         for_each_set_bit(ssid, &__ss_mask, CGROUP_SUBSYS_COUNT) {   \
     686             :                 (ss) = cgroup_subsys[ssid];                             \
     687             :                 {
     688             : 
     689             : #define while_each_subsys_mask()                                        \
     690             :                 }                                                       \
     691             :         }                                                               \
     692             : } while (false)
     693             : 
     694             : /* iterate over child cgrps, lock should be held throughout iteration */
     695             : #define cgroup_for_each_live_child(child, cgrp)                         \
     696             :         list_for_each_entry((child), &(cgrp)->self.children, self.sibling) \
     697             :                 if (({ lockdep_assert_held(&cgroup_mutex);          \
     698             :                        cgroup_is_dead(child); }))                       \
     699             :                         ;                                               \
     700             :                 else
     701             : 
     702             : /* walk live descendants in pre order */
     703             : #define cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp)          \
     704             :         css_for_each_descendant_pre((d_css), cgroup_css((cgrp), NULL))  \
     705             :                 if (({ lockdep_assert_held(&cgroup_mutex);          \
     706             :                        (dsct) = (d_css)->cgroup;                     \
     707             :                        cgroup_is_dead(dsct); }))                        \
     708             :                         ;                                               \
     709             :                 else
     710             : 
     711             : /* walk live descendants in postorder */
     712             : #define cgroup_for_each_live_descendant_post(dsct, d_css, cgrp)         \
     713             :         css_for_each_descendant_post((d_css), cgroup_css((cgrp), NULL)) \
     714             :                 if (({ lockdep_assert_held(&cgroup_mutex);          \
     715             :                        (dsct) = (d_css)->cgroup;                     \
     716             :                        cgroup_is_dead(dsct); }))                        \
     717             :                         ;                                               \
     718             :                 else
     719             : 
     720             : /*
     721             :  * The default css_set - used by init and its children prior to any
     722             :  * hierarchies being mounted. It contains a pointer to the root state
     723             :  * for each subsystem. Also used to anchor the list of css_sets. Not
     724             :  * reference-counted, to improve performance when child cgroups
     725             :  * haven't been created.
     726             :  */
     727             : struct css_set init_css_set = {
     728             :         .refcount               = REFCOUNT_INIT(1),
     729             :         .dom_cset               = &init_css_set,
     730             :         .tasks                  = LIST_HEAD_INIT(init_css_set.tasks),
     731             :         .mg_tasks               = LIST_HEAD_INIT(init_css_set.mg_tasks),
     732             :         .dying_tasks            = LIST_HEAD_INIT(init_css_set.dying_tasks),
     733             :         .task_iters             = LIST_HEAD_INIT(init_css_set.task_iters),
     734             :         .threaded_csets         = LIST_HEAD_INIT(init_css_set.threaded_csets),
     735             :         .cgrp_links             = LIST_HEAD_INIT(init_css_set.cgrp_links),
     736             :         .mg_preload_node        = LIST_HEAD_INIT(init_css_set.mg_preload_node),
     737             :         .mg_node                = LIST_HEAD_INIT(init_css_set.mg_node),
     738             : 
     739             :         /*
     740             :          * The following field is re-initialized when this cset gets linked
     741             :          * in cgroup_init().  However, let's initialize the field
     742             :          * statically too so that the default cgroup can be accessed safely
     743             :          * early during boot.
     744             :          */
     745             :         .dfl_cgrp               = &cgrp_dfl_root.cgrp,
     746             : };
     747             : 
     748             : static int css_set_count        = 1;    /* 1 for init_css_set */
     749             : 
     750          81 : static bool css_set_threaded(struct css_set *cset)
     751             : {
     752          81 :         return cset->dom_cset != cset;
     753             : }
     754             : 
     755             : /**
     756             :  * css_set_populated - does a css_set contain any tasks?
     757             :  * @cset: target css_set
     758             :  *
     759             :  * css_set_populated() should be the same as !!cset->nr_tasks at steady
     760             :  * state. However, css_set_populated() can be called while a task is being
     761             :  * added to or removed from the linked list before the nr_tasks is
     762             :  * properly updated. Hence, we can't just look at ->nr_tasks here.
     763             :  */
     764        2456 : static bool css_set_populated(struct css_set *cset)
     765             : {
     766        7368 :         lockdep_assert_held(&css_set_lock);
     767             : 
     768        2456 :         return !list_empty(&cset->tasks) || !list_empty(&cset->mg_tasks);
     769             : }
     770             : 
     771             : /**
     772             :  * cgroup_update_populated - update the populated count of a cgroup
     773             :  * @cgrp: the target cgroup
     774             :  * @populated: inc or dec populated count
     775             :  *
     776             :  * One of the css_sets associated with @cgrp is either getting its first
     777             :  * task or losing the last.  Update @cgrp->nr_populated_* accordingly.  The
     778             :  * count is propagated towards root so that a given cgroup's
     779             :  * nr_populated_children is zero iff none of its descendants contain any
     780             :  * tasks.
     781             :  *
     782             :  * @cgrp's interface file "cgroup.populated" is zero if both
     783             :  * @cgrp->nr_populated_csets and @cgrp->nr_populated_children are zero and
     784             :  * 1 otherwise.  When the sum changes from or to zero, userland is notified
     785             :  * that the content of the interface file has changed.  This can be used to
     786             :  * detect when @cgrp and its descendants become populated or empty.
     787             :  */
     788         388 : static void cgroup_update_populated(struct cgroup *cgrp, bool populated)
     789             : {
     790         388 :         struct cgroup *child = NULL;
     791         388 :         int adj = populated ? 1 : -1;
     792             : 
     793        1164 :         lockdep_assert_held(&css_set_lock);
     794             : 
     795         576 :         do {
     796         576 :                 bool was_populated = cgroup_is_populated(cgrp);
     797             : 
     798         576 :                 if (!child) {
     799         388 :                         cgrp->nr_populated_csets += adj;
     800             :                 } else {
     801         188 :                         if (cgroup_is_threaded(child))
     802           0 :                                 cgrp->nr_populated_threaded_children += adj;
     803             :                         else
     804         188 :                                 cgrp->nr_populated_domain_children += adj;
     805             :                 }
     806             : 
     807         576 :                 if (was_populated == cgroup_is_populated(cgrp))
     808             :                         break;
     809             : 
     810         190 :                 cgroup1_check_for_release(cgrp);
     811         190 :                 TRACE_CGROUP_PATH(notify_populated, cgrp,
     812             :                                   cgroup_is_populated(cgrp));
     813         190 :                 cgroup_file_notify(&cgrp->events_file);
     814             : 
     815         190 :                 child = cgrp;
     816         578 :                 cgrp = cgroup_parent(cgrp);
     817             :         } while (cgrp);
     818         388 : }
     819             : 
     820             : /**
     821             :  * css_set_update_populated - update populated state of a css_set
     822             :  * @cset: target css_set
     823             :  * @populated: whether @cset is populated or depopulated
     824             :  *
     825             :  * @cset is either getting the first task or losing the last.  Update the
     826             :  * populated counters of all associated cgroups accordingly.
     827             :  */
     828         194 : static void css_set_update_populated(struct css_set *cset, bool populated)
     829             : {
     830         194 :         struct cgrp_cset_link *link;
     831             : 
     832         582 :         lockdep_assert_held(&css_set_lock);
     833             : 
     834         581 :         list_for_each_entry(link, &cset->cgrp_links, cgrp_link)
     835         387 :                 cgroup_update_populated(link->cgrp, populated);
     836         194 : }
     837             : 
     838             : /*
     839             :  * @task is leaving, advance task iterators which are pointing to it so
     840             :  * that they can resume at the next position.  Advancing an iterator might
     841             :  * remove it from the list, use safe walk.  See css_task_iter_skip() for
     842             :  * details.
     843             :  */
     844        2272 : static void css_set_skip_task_iters(struct css_set *cset,
     845             :                                     struct task_struct *task)
     846             : {
     847        2272 :         struct css_task_iter *it, *pos;
     848             : 
     849        2273 :         list_for_each_entry_safe(it, pos, &cset->task_iters, iters_node)
     850           1 :                 css_task_iter_skip(it, task);
     851        2272 : }
     852             : 
     853             : /**
     854             :  * css_set_move_task - move a task from one css_set to another
     855             :  * @task: task being moved
     856             :  * @from_cset: css_set @task currently belongs to (may be NULL)
     857             :  * @to_cset: new css_set @task is being moved to (may be NULL)
     858             :  * @use_mg_tasks: move to @to_cset->mg_tasks instead of ->tasks
     859             :  *
     860             :  * Move @task from @from_cset to @to_cset.  If @task didn't belong to any
     861             :  * css_set, @from_cset can be NULL.  If @task is being disassociated
     862             :  * instead of moved, @to_cset can be NULL.
     863             :  *
     864             :  * This function automatically handles populated counter updates and
     865             :  * css_task_iter adjustments but the caller is responsible for managing
     866             :  * @from_cset and @to_cset's reference counts.
     867             :  */
     868        2348 : static void css_set_move_task(struct task_struct *task,
     869             :                               struct css_set *from_cset, struct css_set *to_cset,
     870             :                               bool use_mg_tasks)
     871             : {
     872        7044 :         lockdep_assert_held(&css_set_lock);
     873             : 
     874        2348 :         if (to_cset && !css_set_populated(to_cset))
     875         106 :                 css_set_update_populated(to_cset, true);
     876             : 
     877        2348 :         if (from_cset) {
     878        1189 :                 WARN_ON_ONCE(list_empty(&task->cg_list));
     879             : 
     880        1189 :                 css_set_skip_task_iters(from_cset, task);
     881        1189 :                 list_del_init(&task->cg_list);
     882        1189 :                 if (!css_set_populated(from_cset))
     883          88 :                         css_set_update_populated(from_cset, false);
     884             :         } else {
     885        1159 :                 WARN_ON_ONCE(!list_empty(&task->cg_list));
     886             :         }
     887             : 
     888        2348 :         if (to_cset) {
     889             :                 /*
     890             :                  * We are synchronized through cgroup_threadgroup_rwsem
     891             :                  * against PF_EXITING setting such that we can't race
     892             :                  * against cgroup_exit()/cgroup_free() dropping the css_set.
     893             :                  */
     894        1265 :                 WARN_ON_ONCE(task->flags & PF_EXITING);
     895             : 
     896        1265 :                 cgroup_move_task(task, to_cset);
     897        1265 :                 list_add_tail(&task->cg_list, use_mg_tasks ? &to_cset->mg_tasks :
     898             :                                                              &to_cset->tasks);
     899             :         }
     900        2348 : }
     901             : 
     902             : /*
     903             :  * hash table for cgroup groups. This improves the performance to find
     904             :  * an existing css_set. This hash doesn't (currently) take into
     905             :  * account cgroups in empty hierarchies.
     906             :  */
     907             : #define CSS_SET_HASH_BITS       7
     908             : static DEFINE_HASHTABLE(css_set_table, CSS_SET_HASH_BITS);
     909             : 
     910         298 : static unsigned long css_set_hash(struct cgroup_subsys_state *css[])
     911             : {
     912         298 :         unsigned long key = 0UL;
     913         298 :         struct cgroup_subsys *ss;
     914         298 :         int i;
     915             : 
     916         298 :         for_each_subsys(ss, i)
     917             :                 key += (unsigned long)css[i];
     918         298 :         key = (key >> 16) ^ key;
     919             : 
     920         298 :         return key;
     921             : }
     922             : 
     923         371 : void put_css_set_locked(struct css_set *cset)
     924             : {
     925         371 :         struct cgrp_cset_link *link, *tmp_link;
     926         371 :         struct cgroup_subsys *ss;
     927         371 :         int ssid;
     928             : 
     929        1113 :         lockdep_assert_held(&css_set_lock);
     930             : 
     931         371 :         if (!refcount_dec_and_test(&cset->refcount))
     932             :                 return;
     933             : 
     934          81 :         WARN_ON_ONCE(!list_empty(&cset->threaded_csets));
     935             : 
     936             :         /* This css_set is dead. Unlink it and release cgroup and css refs */
     937          81 :         for_each_subsys(ss, ssid) {
     938             :                 list_del(&cset->e_cset_node[ssid]);
     939             :                 css_put(cset->subsys[ssid]);
     940             :         }
     941          81 :         hash_del(&cset->hlist);
     942          81 :         css_set_count--;
     943             : 
     944         243 :         list_for_each_entry_safe(link, tmp_link, &cset->cgrp_links, cgrp_link) {
     945         162 :                 list_del(&link->cset_link);
     946         162 :                 list_del(&link->cgrp_link);
     947         162 :                 if (cgroup_parent(link->cgrp))
     948         161 :                         cgroup_put(link->cgrp);
     949         162 :                 kfree(link);
     950             :         }
     951             : 
     952          81 :         if (css_set_threaded(cset)) {
     953           0 :                 list_del(&cset->threaded_csets_node);
     954           0 :                 put_css_set_locked(cset->dom_cset);
     955             :         }
     956             : 
     957          81 :         kfree_rcu(cset, rcu_head);
     958             : }
     959             : 
     960             : /**
     961             :  * compare_css_sets - helper function for find_existing_css_set().
     962             :  * @cset: candidate css_set being tested
     963             :  * @old_cset: existing css_set for a task
     964             :  * @new_cgrp: cgroup that's being entered by the task
     965             :  * @template: desired set of css pointers in css_set (pre-calculated)
     966             :  *
     967             :  * Returns true if "cset" matches "old_cset" except for the hierarchy
     968             :  * which "new_cgrp" belongs to, for which it should match "new_cgrp".
     969             :  */
     970        1291 : static bool compare_css_sets(struct css_set *cset,
     971             :                              struct css_set *old_cset,
     972             :                              struct cgroup *new_cgrp,
     973             :                              struct cgroup_subsys_state *template[])
     974             : {
     975        1291 :         struct cgroup *new_dfl_cgrp;
     976        1291 :         struct list_head *l1, *l2;
     977             : 
     978             :         /*
     979             :          * On the default hierarchy, there can be csets which are
     980             :          * associated with the same set of cgroups but different csses.
     981             :          * Let's first ensure that csses match.
     982             :          */
     983        1291 :         if (memcmp(template, cset->subsys, sizeof(cset->subsys)))
     984             :                 return false;
     985             : 
     986             : 
     987             :         /* @cset's domain should match the default cgroup's */
     988        1291 :         if (cgroup_on_dfl(new_cgrp))
     989             :                 new_dfl_cgrp = new_cgrp;
     990             :         else
     991         616 :                 new_dfl_cgrp = old_cset->dfl_cgrp;
     992             : 
     993        1291 :         if (new_dfl_cgrp->dom_cgrp != cset->dom_cset->dfl_cgrp)
     994             :                 return false;
     995             : 
     996             :         /*
     997             :          * Compare cgroup pointers in order to distinguish between
     998             :          * different cgroups in hierarchies.  As different cgroups may
     999             :          * share the same effective css, this comparison is always
    1000             :          * necessary.
    1001             :          */
    1002         161 :         l1 = &cset->cgrp_links;
    1003         161 :         l2 = &old_cset->cgrp_links;
    1004         422 :         while (1) {
    1005         422 :                 struct cgrp_cset_link *link1, *link2;
    1006         422 :                 struct cgroup *cgrp1, *cgrp2;
    1007             : 
    1008         422 :                 l1 = l1->next;
    1009         422 :                 l2 = l2->next;
    1010             :                 /* See if we reached the end - both lists are equal length. */
    1011         422 :                 if (l1 == &cset->cgrp_links) {
    1012         100 :                         BUG_ON(l2 != &old_cset->cgrp_links);
    1013             :                         break;
    1014             :                 } else {
    1015         322 :                         BUG_ON(l2 == &old_cset->cgrp_links);
    1016             :                 }
    1017             :                 /* Locate the cgroups associated with these links. */
    1018         322 :                 link1 = list_entry(l1, struct cgrp_cset_link, cgrp_link);
    1019         322 :                 link2 = list_entry(l2, struct cgrp_cset_link, cgrp_link);
    1020         322 :                 cgrp1 = link1->cgrp;
    1021         322 :                 cgrp2 = link2->cgrp;
    1022             :                 /* Hierarchies should be linked in the same order. */
    1023         322 :                 BUG_ON(cgrp1->root != cgrp2->root);
    1024             : 
    1025             :                 /*
    1026             :                  * If this hierarchy is the hierarchy of the cgroup
    1027             :                  * that's changing, then we need to check that this
    1028             :                  * css_set points to the new cgroup; if it's any other
    1029             :                  * hierarchy, then this css_set should point to the
    1030             :                  * same cgroup as the old css_set.
    1031             :                  */
    1032         322 :                 if (cgrp1->root == new_cgrp->root) {
    1033         161 :                         if (cgrp1 != new_cgrp)
    1034             :                                 return false;
    1035             :                 } else {
    1036         161 :                         if (cgrp1 != cgrp2)
    1037             :                                 return false;
    1038             :                 }
    1039             :         }
    1040             :         return true;
    1041             : }
    1042             : 
    1043             : /**
    1044             :  * find_existing_css_set - init css array and find the matching css_set
    1045             :  * @old_cset: the css_set that we're using before the cgroup transition
    1046             :  * @cgrp: the cgroup that we're moving into
    1047             :  * @template: out param for the new set of csses, should be clear on entry
    1048             :  */
    1049         198 : static struct css_set *find_existing_css_set(struct css_set *old_cset,
    1050             :                                         struct cgroup *cgrp,
    1051             :                                         struct cgroup_subsys_state *template[])
    1052             : {
    1053         198 :         struct cgroup_root *root = cgrp->root;
    1054         198 :         struct cgroup_subsys *ss;
    1055         198 :         struct css_set *cset;
    1056         198 :         unsigned long key;
    1057         198 :         int i;
    1058             : 
    1059             :         /*
    1060             :          * Build the set of subsystem state objects that we want to see in the
    1061             :          * new css_set. While subsystems can change globally, the entries here
    1062             :          * won't change, so no need for locking.
    1063             :          */
    1064         198 :         for_each_subsys(ss, i) {
    1065             :                 if (root->subsys_mask & (1UL << i)) {
    1066             :                         /*
    1067             :                          * @ss is in this hierarchy, so we want the
    1068             :                          * effective css from @cgrp.
    1069             :                          */
    1070             :                         template[i] = cgroup_e_css_by_mask(cgrp, ss);
    1071             :                 } else {
    1072             :                         /*
    1073             :                          * @ss is not in this hierarchy, so we don't want
    1074             :                          * to change the css.
    1075             :                          */
    1076             :                         template[i] = old_cset->subsys[i];
    1077             :                 }
    1078             :         }
    1079             : 
    1080         198 :         key = css_set_hash(template);
    1081        1489 :         hash_for_each_possible(css_set_table, cset, hlist, key) {
    1082        1291 :                 if (!compare_css_sets(cset, old_cset, cgrp, template))
    1083        1191 :                         continue;
    1084             : 
    1085             :                 /* This css_set matches what we need */
    1086             :                 return cset;
    1087             :         }
    1088             : 
    1089             :         /* No existing cgroup group matched */
    1090             :         return NULL;
    1091             : }
    1092             : 
    1093           2 : static void free_cgrp_cset_links(struct list_head *links_to_free)
    1094             : {
    1095           2 :         struct cgrp_cset_link *link, *tmp_link;
    1096             : 
    1097           4 :         list_for_each_entry_safe(link, tmp_link, links_to_free, cset_link) {
    1098           2 :                 list_del(&link->cset_link);
    1099           2 :                 kfree(link);
    1100             :         }
    1101           2 : }
    1102             : 
    1103             : /**
    1104             :  * allocate_cgrp_cset_links - allocate cgrp_cset_links
    1105             :  * @count: the number of links to allocate
    1106             :  * @tmp_links: list_head the allocated links are put on
    1107             :  *
    1108             :  * Allocate @count cgrp_cset_link structures and chain them on @tmp_links
    1109             :  * through ->cset_link.  Returns 0 on success or -errno.
    1110             :  */
    1111         100 : static int allocate_cgrp_cset_links(int count, struct list_head *tmp_links)
    1112             : {
    1113         100 :         struct cgrp_cset_link *link;
    1114         100 :         int i;
    1115             : 
    1116         100 :         INIT_LIST_HEAD(tmp_links);
    1117             : 
    1118         300 :         for (i = 0; i < count; i++) {
    1119         200 :                 link = kzalloc(sizeof(*link), GFP_KERNEL);
    1120         200 :                 if (!link) {
    1121           0 :                         free_cgrp_cset_links(tmp_links);
    1122           0 :                         return -ENOMEM;
    1123             :                 }
    1124         200 :                 list_add(&link->cset_link, tmp_links);
    1125             :         }
    1126             :         return 0;
    1127             : }
    1128             : 
    1129             : /**
    1130             :  * link_css_set - a helper function to link a css_set to a cgroup
    1131             :  * @tmp_links: cgrp_cset_link objects allocated by allocate_cgrp_cset_links()
    1132             :  * @cset: the css_set to be linked
    1133             :  * @cgrp: the destination cgroup
    1134             :  */
    1135         198 : static void link_css_set(struct list_head *tmp_links, struct css_set *cset,
    1136             :                          struct cgroup *cgrp)
    1137             : {
    1138         198 :         struct cgrp_cset_link *link;
    1139             : 
    1140         198 :         BUG_ON(list_empty(tmp_links));
    1141             : 
    1142         198 :         if (cgroup_on_dfl(cgrp))
    1143          99 :                 cset->dfl_cgrp = cgrp;
    1144             : 
    1145         198 :         link = list_first_entry(tmp_links, struct cgrp_cset_link, cset_link);
    1146         198 :         link->cset = cset;
    1147         198 :         link->cgrp = cgrp;
    1148             : 
    1149             :         /*
    1150             :          * Always add links to the tail of the lists so that the lists are
    1151             :          * in chronological order.
    1152             :          */
    1153         198 :         list_move_tail(&link->cset_link, &cgrp->cset_links);
    1154         198 :         list_add_tail(&link->cgrp_link, &cset->cgrp_links);
    1155             : 
    1156         198 :         if (cgroup_parent(cgrp))
    1157         195 :                 cgroup_get_live(cgrp);
    1158         198 : }
    1159             : 
    1160             : /**
    1161             :  * find_css_set - return a new css_set with one cgroup updated
    1162             :  * @old_cset: the baseline css_set
    1163             :  * @cgrp: the cgroup to be updated
    1164             :  *
    1165             :  * Return a new css_set that's equivalent to @old_cset, but with @cgrp
    1166             :  * substituted into the appropriate hierarchy.
    1167             :  */
    1168         198 : static struct css_set *find_css_set(struct css_set *old_cset,
    1169             :                                     struct cgroup *cgrp)
    1170             : {
    1171         198 :         struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT] = { };
    1172         198 :         struct css_set *cset;
    1173         198 :         struct list_head tmp_links;
    1174         198 :         struct cgrp_cset_link *link;
    1175         198 :         struct cgroup_subsys *ss;
    1176         198 :         unsigned long key;
    1177         198 :         int ssid;
    1178             : 
    1179         594 :         lockdep_assert_held(&cgroup_mutex);
    1180             : 
    1181             :         /* First see if we already have a cgroup group that matches
    1182             :          * the desired set */
    1183         198 :         spin_lock_irq(&css_set_lock);
    1184         198 :         cset = find_existing_css_set(old_cset, cgrp, template);
    1185         198 :         if (cset)
    1186         100 :                 get_css_set(cset);
    1187         198 :         spin_unlock_irq(&css_set_lock);
    1188             : 
    1189         198 :         if (cset)
    1190             :                 return cset;
    1191             : 
    1192          98 :         cset = kzalloc(sizeof(*cset), GFP_KERNEL);
    1193          98 :         if (!cset)
    1194             :                 return NULL;
    1195             : 
    1196             :         /* Allocate all the cgrp_cset_link objects that we'll need */
    1197          98 :         if (allocate_cgrp_cset_links(cgroup_root_count, &tmp_links) < 0) {
    1198           0 :                 kfree(cset);
    1199           0 :                 return NULL;
    1200             :         }
    1201             : 
    1202          98 :         refcount_set(&cset->refcount, 1);
    1203          98 :         cset->dom_cset = cset;
    1204          98 :         INIT_LIST_HEAD(&cset->tasks);
    1205          98 :         INIT_LIST_HEAD(&cset->mg_tasks);
    1206          98 :         INIT_LIST_HEAD(&cset->dying_tasks);
    1207          98 :         INIT_LIST_HEAD(&cset->task_iters);
    1208          98 :         INIT_LIST_HEAD(&cset->threaded_csets);
    1209          98 :         INIT_HLIST_NODE(&cset->hlist);
    1210          98 :         INIT_LIST_HEAD(&cset->cgrp_links);
    1211          98 :         INIT_LIST_HEAD(&cset->mg_preload_node);
    1212          98 :         INIT_LIST_HEAD(&cset->mg_node);
    1213             : 
    1214             :         /* Copy the set of subsystem state objects generated in
    1215             :          * find_existing_css_set() */
    1216          98 :         memcpy(cset->subsys, template, sizeof(cset->subsys));
    1217             : 
    1218          98 :         spin_lock_irq(&css_set_lock);
    1219             :         /* Add reference counts and links from the new css_set. */
    1220         294 :         list_for_each_entry(link, &old_cset->cgrp_links, cgrp_link) {
    1221         196 :                 struct cgroup *c = link->cgrp;
    1222             : 
    1223         196 :                 if (c->root == cgrp->root)
    1224          98 :                         c = cgrp;
    1225         196 :                 link_css_set(&tmp_links, cset, c);
    1226             :         }
    1227             : 
    1228          98 :         BUG_ON(!list_empty(&tmp_links));
    1229             : 
    1230          98 :         css_set_count++;
    1231             : 
    1232             :         /* Add @cset to the hash table */
    1233          98 :         key = css_set_hash(cset->subsys);
    1234          98 :         hash_add(css_set_table, &cset->hlist, key);
    1235             : 
    1236          98 :         for_each_subsys(ss, ssid) {
    1237             :                 struct cgroup_subsys_state *css = cset->subsys[ssid];
    1238             : 
    1239             :                 list_add_tail(&cset->e_cset_node[ssid],
    1240             :                               &css->cgroup->e_csets[ssid]);
    1241             :                 css_get(css);
    1242             :         }
    1243             : 
    1244          98 :         spin_unlock_irq(&css_set_lock);
    1245             : 
    1246             :         /*
    1247             :          * If @cset should be threaded, look up the matching dom_cset and
    1248             :          * link them up.  We first fully initialize @cset then look for the
    1249             :          * dom_cset.  It's simpler this way and safe as @cset is guaranteed
    1250             :          * to stay empty until we return.
    1251             :          */
    1252          98 :         if (cgroup_is_threaded(cset->dfl_cgrp)) {
    1253           0 :                 struct css_set *dcset;
    1254             : 
    1255           0 :                 dcset = find_css_set(cset, cset->dfl_cgrp->dom_cgrp);
    1256           0 :                 if (!dcset) {
    1257           0 :                         put_css_set(cset);
    1258           0 :                         return NULL;
    1259             :                 }
    1260             : 
    1261           0 :                 spin_lock_irq(&css_set_lock);
    1262           0 :                 cset->dom_cset = dcset;
    1263           0 :                 list_add_tail(&cset->threaded_csets_node,
    1264             :                               &dcset->threaded_csets);
    1265           0 :                 spin_unlock_irq(&css_set_lock);
    1266             :         }
    1267             : 
    1268             :         return cset;
    1269             : }
    1270             : 
    1271         862 : struct cgroup_root *cgroup_root_from_kf(struct kernfs_root *kf_root)
    1272             : {
    1273         862 :         struct cgroup *root_cgrp = kf_root->kn->priv;
    1274             : 
    1275         862 :         return root_cgrp->root;
    1276             : }
    1277             : 
    1278           2 : static int cgroup_init_root_id(struct cgroup_root *root)
    1279             : {
    1280           2 :         int id;
    1281             : 
    1282           6 :         lockdep_assert_held(&cgroup_mutex);
    1283             : 
    1284           2 :         id = idr_alloc_cyclic(&cgroup_hierarchy_idr, root, 0, 0, GFP_KERNEL);
    1285           2 :         if (id < 0)
    1286             :                 return id;
    1287             : 
    1288           2 :         root->hierarchy_id = id;
    1289           2 :         return 0;
    1290             : }
    1291             : 
    1292           0 : static void cgroup_exit_root_id(struct cgroup_root *root)
    1293             : {
    1294           0 :         lockdep_assert_held(&cgroup_mutex);
    1295             : 
    1296           0 :         idr_remove(&cgroup_hierarchy_idr, root->hierarchy_id);
    1297           0 : }
    1298             : 
    1299           0 : void cgroup_free_root(struct cgroup_root *root)
    1300             : {
    1301           0 :         kfree(root);
    1302           0 : }
    1303             : 
    1304           0 : static void cgroup_destroy_root(struct cgroup_root *root)
    1305             : {
    1306           0 :         struct cgroup *cgrp = &root->cgrp;
    1307           0 :         struct cgrp_cset_link *link, *tmp_link;
    1308             : 
    1309           0 :         trace_cgroup_destroy_root(root);
    1310             : 
    1311           0 :         cgroup_lock_and_drain_offline(&cgrp_dfl_root.cgrp);
    1312             : 
    1313           0 :         BUG_ON(atomic_read(&root->nr_cgrps));
    1314           0 :         BUG_ON(!list_empty(&cgrp->self.children));
    1315             : 
    1316             :         /* Rebind all subsystems back to the default hierarchy */
    1317           0 :         WARN_ON(rebind_subsystems(&cgrp_dfl_root, root->subsys_mask));
    1318             : 
    1319             :         /*
    1320             :          * Release all the links from cset_links to this hierarchy's
    1321             :          * root cgroup
    1322             :          */
    1323           0 :         spin_lock_irq(&css_set_lock);
    1324             : 
    1325           0 :         list_for_each_entry_safe(link, tmp_link, &cgrp->cset_links, cset_link) {
    1326           0 :                 list_del(&link->cset_link);
    1327           0 :                 list_del(&link->cgrp_link);
    1328           0 :                 kfree(link);
    1329             :         }
    1330             : 
    1331           0 :         spin_unlock_irq(&css_set_lock);
    1332             : 
    1333           0 :         if (!list_empty(&root->root_list)) {
    1334           0 :                 list_del(&root->root_list);
    1335           0 :                 cgroup_root_count--;
    1336             :         }
    1337             : 
    1338           0 :         cgroup_exit_root_id(root);
    1339             : 
    1340           0 :         mutex_unlock(&cgroup_mutex);
    1341             : 
    1342           0 :         kernfs_destroy_root(root->kf_root);
    1343           0 :         cgroup_free_root(root);
    1344           0 : }
    1345             : 
    1346             : /*
    1347             :  * look up cgroup associated with current task's cgroup namespace on the
    1348             :  * specified hierarchy
    1349             :  */
    1350             : static struct cgroup *
    1351         574 : current_cgns_cgroup_from_root(struct cgroup_root *root)
    1352             : {
    1353         574 :         struct cgroup *res = NULL;
    1354         574 :         struct css_set *cset;
    1355             : 
    1356        1722 :         lockdep_assert_held(&css_set_lock);
    1357             : 
    1358         574 :         rcu_read_lock();
    1359             : 
    1360         574 :         cset = current->nsproxy->cgroup_ns->root_cset;
    1361         574 :         if (cset == &init_css_set) {
    1362         574 :                 res = &root->cgrp;
    1363           0 :         } else if (root == &cgrp_dfl_root) {
    1364           0 :                 res = cset->dfl_cgrp;
    1365             :         } else {
    1366           0 :                 struct cgrp_cset_link *link;
    1367             : 
    1368           0 :                 list_for_each_entry(link, &cset->cgrp_links, cgrp_link) {
    1369           0 :                         struct cgroup *c = link->cgrp;
    1370             : 
    1371           0 :                         if (c->root == root) {
    1372             :                                 res = c;
    1373             :                                 break;
    1374             :                         }
    1375             :                 }
    1376             :         }
    1377         574 :         rcu_read_unlock();
    1378             : 
    1379         574 :         BUG_ON(!res);
    1380         574 :         return res;
    1381             : }
    1382             : 
    1383             : /* look up cgroup associated with given css_set on the specified hierarchy */
    1384         953 : static struct cgroup *cset_cgroup_from_root(struct css_set *cset,
    1385             :                                             struct cgroup_root *root)
    1386             : {
    1387         953 :         struct cgroup *res = NULL;
    1388             : 
    1389        2859 :         lockdep_assert_held(&cgroup_mutex);
    1390        2859 :         lockdep_assert_held(&css_set_lock);
    1391             : 
    1392         953 :         if (cset == &init_css_set) {
    1393         318 :                 res = &root->cgrp;
    1394         635 :         } else if (root == &cgrp_dfl_root) {
    1395         366 :                 res = cset->dfl_cgrp;
    1396             :         } else {
    1397         269 :                 struct cgrp_cset_link *link;
    1398             : 
    1399         538 :                 list_for_each_entry(link, &cset->cgrp_links, cgrp_link) {
    1400         538 :                         struct cgroup *c = link->cgrp;
    1401             : 
    1402         538 :                         if (c->root == root) {
    1403             :                                 res = c;
    1404             :                                 break;
    1405             :                         }
    1406             :                 }
    1407             :         }
    1408             : 
    1409         953 :         BUG_ON(!res);
    1410         953 :         return res;
    1411             : }
    1412             : 
    1413             : /*
    1414             :  * Return the cgroup for "task" from the given hierarchy. Must be
    1415             :  * called with cgroup_mutex and css_set_lock held.
    1416             :  */
    1417         443 : struct cgroup *task_cgroup_from_root(struct task_struct *task,
    1418             :                                      struct cgroup_root *root)
    1419             : {
    1420             :         /*
    1421             :          * No need to lock the task - since we hold css_set_lock the
    1422             :          * task can't change groups.
    1423             :          */
    1424         443 :         return cset_cgroup_from_root(task_css_set(task), root);
    1425             : }
    1426             : 
    1427             : /*
    1428             :  * A task must hold cgroup_mutex to modify cgroups.
    1429             :  *
    1430             :  * Any task can increment and decrement the count field without lock.
    1431             :  * So in general, code holding cgroup_mutex can't rely on the count
    1432             :  * field not changing.  However, if the count goes to zero, then only
    1433             :  * cgroup_attach_task() can increment it again.  Because a count of zero
    1434             :  * means that no tasks are currently attached, therefore there is no
    1435             :  * way a task attached to that cgroup can fork (the other way to
    1436             :  * increment the count).  So code holding cgroup_mutex can safely
    1437             :  * assume that if the count is zero, it will stay zero. Similarly, if
    1438             :  * a task holds cgroup_mutex on a cgroup with zero count, it
    1439             :  * knows that the cgroup won't be removed, as cgroup_rmdir()
    1440             :  * needs that mutex.
    1441             :  *
    1442             :  * A cgroup can only be deleted if both its 'count' of using tasks
    1443             :  * is zero, and its list of 'children' cgroups is empty.  Since all
    1444             :  * tasks in the system use _some_ cgroup, and since there is always at
    1445             :  * least one task in the system (init, pid == 1), therefore, root cgroup
    1446             :  * always has either children cgroups and/or using tasks.  So we don't
    1447             :  * need a special hack to ensure that root cgroup cannot be deleted.
    1448             :  *
    1449             :  * P.S.  One more locking exception.  RCU is used to guard the
    1450             :  * update of a tasks cgroup pointer by cgroup_attach_task()
    1451             :  */
    1452             : 
    1453             : static struct kernfs_syscall_ops cgroup_kf_syscall_ops;
    1454             : 
    1455        1124 : static char *cgroup_file_name(struct cgroup *cgrp, const struct cftype *cft,
    1456             :                               char *buf)
    1457             : {
    1458        1124 :         struct cgroup_subsys *ss = cft->ss;
    1459             : 
    1460        1124 :         if (cft->ss && !(cft->flags & CFTYPE_NO_PREFIX) &&
    1461           0 :             !(cgrp->root->flags & CGRP_ROOT_NOPREFIX)) {
    1462           0 :                 const char *dbg = (cft->flags & CFTYPE_DEBUG) ? ".__DEBUG__." : "";
    1463             : 
    1464           0 :                 snprintf(buf, CGROUP_FILE_NAME_MAX, "%s%s.%s",
    1465           0 :                          dbg, cgroup_on_dfl(cgrp) ? ss->name : ss->legacy_name,
    1466           0 :                          cft->name);
    1467             :         } else {
    1468        1124 :                 strscpy(buf, cft->name, CGROUP_FILE_NAME_MAX);
    1469             :         }
    1470        1124 :         return buf;
    1471             : }
    1472             : 
    1473             : /**
    1474             :  * cgroup_file_mode - deduce file mode of a control file
    1475             :  * @cft: the control file in question
    1476             :  *
    1477             :  * S_IRUGO for read, S_IWUSR for write.
    1478             :  */
    1479         749 : static umode_t cgroup_file_mode(const struct cftype *cft)
    1480             : {
    1481         749 :         umode_t mode = 0;
    1482             : 
    1483         749 :         if (cft->read_u64 || cft->read_s64 || cft->seq_show)
    1484         749 :                 mode |= S_IRUGO;
    1485             : 
    1486         749 :         if (cft->write_u64 || cft->write_s64 || cft->write) {
    1487         549 :                 if (cft->flags & CFTYPE_WORLD_WRITABLE)
    1488           0 :                         mode |= S_IWUGO;
    1489             :                 else
    1490         549 :                         mode |= S_IWUSR;
    1491             :         }
    1492             : 
    1493         749 :         return mode;
    1494             : }
    1495             : 
    1496             : /**
    1497             :  * cgroup_calc_subtree_ss_mask - calculate subtree_ss_mask
    1498             :  * @subtree_control: the new subtree_control mask to consider
    1499             :  * @this_ss_mask: available subsystems
    1500             :  *
    1501             :  * On the default hierarchy, a subsystem may request other subsystems to be
    1502             :  * enabled together through its ->depends_on mask.  In such cases, more
    1503             :  * subsystems than specified in "cgroup.subtree_control" may be enabled.
    1504             :  *
    1505             :  * This function calculates which subsystems need to be enabled if
    1506             :  * @subtree_control is to be applied while restricted to @this_ss_mask.
    1507             :  */
    1508          98 : static u16 cgroup_calc_subtree_ss_mask(u16 subtree_control, u16 this_ss_mask)
    1509             : {
    1510          98 :         u16 cur_ss_mask = subtree_control;
    1511          98 :         struct cgroup_subsys *ss;
    1512          98 :         int ssid;
    1513             : 
    1514         294 :         lockdep_assert_held(&cgroup_mutex);
    1515             : 
    1516          98 :         cur_ss_mask |= cgrp_dfl_implicit_ss_mask;
    1517             : 
    1518          98 :         while (true) {
    1519          98 :                 u16 new_ss_mask = cur_ss_mask;
    1520             : 
    1521          98 :                 do_each_subsys_mask(ss, ssid, cur_ss_mask) {
    1522             :                         new_ss_mask |= ss->depends_on;
    1523          98 :                 } while_each_subsys_mask();
    1524             : 
    1525             :                 /*
    1526             :                  * Mask out subsystems which aren't available.  This can
    1527             :                  * happen only if some depended-upon subsystems were bound
    1528             :                  * to non-default hierarchies.
    1529             :                  */
    1530          98 :                 new_ss_mask &= this_ss_mask;
    1531             : 
    1532          98 :                 if (new_ss_mask == cur_ss_mask)
    1533             :                         break;
    1534             :                 cur_ss_mask = new_ss_mask;
    1535             :         }
    1536             : 
    1537          98 :         return cur_ss_mask;
    1538             : }
    1539             : 
    1540             : /**
    1541             :  * cgroup_kn_unlock - unlocking helper for cgroup kernfs methods
    1542             :  * @kn: the kernfs_node being serviced
    1543             :  *
    1544             :  * This helper undoes cgroup_kn_lock_live() and should be invoked before
    1545             :  * the method finishes if locking succeeded.  Note that once this function
    1546             :  * returns the cgroup returned by cgroup_kn_lock_live() may become
    1547             :  * inaccessible any time.  If the caller intends to continue to access the
    1548             :  * cgroup, it should pin it before invoking this function.
    1549             :  */
    1550         347 : void cgroup_kn_unlock(struct kernfs_node *kn)
    1551             : {
    1552         347 :         struct cgroup *cgrp;
    1553             : 
    1554         347 :         if (kernfs_type(kn) == KERNFS_DIR)
    1555         149 :                 cgrp = kn->priv;
    1556             :         else
    1557         198 :                 cgrp = kn->parent->priv;
    1558             : 
    1559         347 :         mutex_unlock(&cgroup_mutex);
    1560             : 
    1561         347 :         kernfs_unbreak_active_protection(kn);
    1562         347 :         cgroup_put(cgrp);
    1563         347 : }
    1564             : 
    1565             : /**
    1566             :  * cgroup_kn_lock_live - locking helper for cgroup kernfs methods
    1567             :  * @kn: the kernfs_node being serviced
    1568             :  * @drain_offline: perform offline draining on the cgroup
    1569             :  *
    1570             :  * This helper is to be used by a cgroup kernfs method currently servicing
    1571             :  * @kn.  It breaks the active protection, performs cgroup locking and
    1572             :  * verifies that the associated cgroup is alive.  Returns the cgroup if
    1573             :  * alive; otherwise, %NULL.  A successful return should be undone by a
    1574             :  * matching cgroup_kn_unlock() invocation.  If @drain_offline is %true, the
    1575             :  * cgroup is drained of offlining csses before return.
    1576             :  *
    1577             :  * Any cgroup kernfs method implementation which requires locking the
    1578             :  * associated cgroup should use this helper.  It avoids nesting cgroup
    1579             :  * locking under kernfs active protection and allows all kernfs operations
    1580             :  * including self-removal.
    1581             :  */
    1582         347 : struct cgroup *cgroup_kn_lock_live(struct kernfs_node *kn, bool drain_offline)
    1583             : {
    1584         347 :         struct cgroup *cgrp;
    1585             : 
    1586         347 :         if (kernfs_type(kn) == KERNFS_DIR)
    1587         149 :                 cgrp = kn->priv;
    1588             :         else
    1589         198 :                 cgrp = kn->parent->priv;
    1590             : 
    1591             :         /*
    1592             :          * We're gonna grab cgroup_mutex which nests outside kernfs
    1593             :          * active_ref.  cgroup liveliness check alone provides enough
    1594             :          * protection against removal.  Ensure @cgrp stays accessible and
    1595             :          * break the active_ref protection.
    1596             :          */
    1597         347 :         if (!cgroup_tryget(cgrp))
    1598             :                 return NULL;
    1599         347 :         kernfs_break_active_protection(kn);
    1600             : 
    1601         347 :         if (drain_offline)
    1602           0 :                 cgroup_lock_and_drain_offline(cgrp);
    1603             :         else
    1604         347 :                 mutex_lock(&cgroup_mutex);
    1605             : 
    1606         347 :         if (!cgroup_is_dead(cgrp))
    1607             :                 return cgrp;
    1608             : 
    1609           0 :         cgroup_kn_unlock(kn);
    1610           0 :         return NULL;
    1611             : }
    1612             : 
    1613         375 : static void cgroup_rm_file(struct cgroup *cgrp, const struct cftype *cft)
    1614             : {
    1615         375 :         char name[CGROUP_FILE_NAME_MAX];
    1616             : 
    1617        1125 :         lockdep_assert_held(&cgroup_mutex);
    1618             : 
    1619         375 :         if (cft->file_offset) {
    1620          50 :                 struct cgroup_subsys_state *css = cgroup_css(cgrp, cft->ss);
    1621          50 :                 struct cgroup_file *cfile = (void *)css + cft->file_offset;
    1622             : 
    1623          50 :                 spin_lock_irq(&cgroup_file_kn_lock);
    1624          50 :                 cfile->kn = NULL;
    1625          50 :                 spin_unlock_irq(&cgroup_file_kn_lock);
    1626             : 
    1627          50 :                 del_timer_sync(&cfile->notify_timer);
    1628             :         }
    1629             : 
    1630         375 :         kernfs_remove_by_name(cgrp->kn, cgroup_file_name(cgrp, cft, name));
    1631         375 : }
    1632             : 
    1633             : /**
    1634             :  * css_clear_dir - remove subsys files in a cgroup directory
    1635             :  * @css: taget css
    1636             :  */
    1637          50 : static void css_clear_dir(struct cgroup_subsys_state *css)
    1638             : {
    1639          50 :         struct cgroup *cgrp = css->cgroup;
    1640          50 :         struct cftype *cfts;
    1641             : 
    1642          50 :         if (!(css->flags & CSS_VISIBLE))
    1643             :                 return;
    1644             : 
    1645          50 :         css->flags &= ~CSS_VISIBLE;
    1646             : 
    1647          50 :         if (!css->ss) {
    1648          50 :                 if (cgroup_on_dfl(cgrp))
    1649             :                         cfts = cgroup_base_files;
    1650             :                 else
    1651          25 :                         cfts = cgroup1_base_files;
    1652             : 
    1653          50 :                 cgroup_addrm_files(css, cgrp, cfts, false);
    1654             :         } else {
    1655           0 :                 list_for_each_entry(cfts, &css->ss->cfts, node)
    1656           0 :                         cgroup_addrm_files(css, cgrp, cfts, false);
    1657             :         }
    1658             : }
    1659             : 
    1660             : /**
    1661             :  * css_populate_dir - create subsys files in a cgroup directory
    1662             :  * @css: target css
    1663             :  *
    1664             :  * On failure, no file is added.
    1665             :  */
    1666         100 : static int css_populate_dir(struct cgroup_subsys_state *css)
    1667             : {
    1668         100 :         struct cgroup *cgrp = css->cgroup;
    1669         100 :         struct cftype *cfts, *failed_cfts;
    1670         100 :         int ret;
    1671             : 
    1672         100 :         if ((css->flags & CSS_VISIBLE) || !cgrp->kn)
    1673             :                 return 0;
    1674             : 
    1675         100 :         if (!css->ss) {
    1676         100 :                 if (cgroup_on_dfl(cgrp))
    1677             :                         cfts = cgroup_base_files;
    1678             :                 else
    1679          50 :                         cfts = cgroup1_base_files;
    1680             : 
    1681         100 :                 ret = cgroup_addrm_files(&cgrp->self, cgrp, cfts, true);
    1682         100 :                 if (ret < 0)
    1683             :                         return ret;
    1684             :         } else {
    1685           0 :                 list_for_each_entry(cfts, &css->ss->cfts, node) {
    1686           0 :                         ret = cgroup_addrm_files(css, cgrp, cfts, true);
    1687           0 :                         if (ret < 0) {
    1688           0 :                                 failed_cfts = cfts;
    1689           0 :                                 goto err;
    1690             :                         }
    1691             :                 }
    1692             :         }
    1693             : 
    1694         100 :         css->flags |= CSS_VISIBLE;
    1695             : 
    1696         100 :         return 0;
    1697           0 : err:
    1698           0 :         list_for_each_entry(cfts, &css->ss->cfts, node) {
    1699           0 :                 if (cfts == failed_cfts)
    1700             :                         break;
    1701           0 :                 cgroup_addrm_files(css, cgrp, cfts, false);
    1702             :         }
    1703             :         return ret;
    1704             : }
    1705             : 
    1706           2 : int rebind_subsystems(struct cgroup_root *dst_root, u16 ss_mask)
    1707             : {
    1708           2 :         struct cgroup *dcgrp = &dst_root->cgrp;
    1709           2 :         struct cgroup_subsys *ss;
    1710           2 :         int ssid, i, ret;
    1711             : 
    1712           6 :         lockdep_assert_held(&cgroup_mutex);
    1713             : 
    1714           2 :         do_each_subsys_mask(ss, ssid, ss_mask) {
    1715             :                 /*
    1716             :                  * If @ss has non-root csses attached to it, can't move.
    1717             :                  * If @ss is an implicit controller, it is exempt from this
    1718             :                  * rule and can be stolen.
    1719             :                  */
    1720             :                 if (css_next_child(NULL, cgroup_css(&ss->root->cgrp, ss)) &&
    1721             :                     !ss->implicit_on_dfl)
    1722             :                         return -EBUSY;
    1723             : 
    1724             :                 /* can't move between two non-dummy roots either */
    1725             :                 if (ss->root != &cgrp_dfl_root && dst_root != &cgrp_dfl_root)
    1726             :                         return -EBUSY;
    1727           2 :         } while_each_subsys_mask();
    1728             : 
    1729           2 :         do_each_subsys_mask(ss, ssid, ss_mask) {
    1730             :                 struct cgroup_root *src_root = ss->root;
    1731             :                 struct cgroup *scgrp = &src_root->cgrp;
    1732             :                 struct cgroup_subsys_state *css = cgroup_css(scgrp, ss);
    1733             :                 struct css_set *cset;
    1734             : 
    1735             :                 WARN_ON(!css || cgroup_css(dcgrp, ss));
    1736             : 
    1737             :                 /* disable from the source */
    1738             :                 src_root->subsys_mask &= ~(1 << ssid);
    1739             :                 WARN_ON(cgroup_apply_control(scgrp));
    1740             :                 cgroup_finalize_control(scgrp, 0);
    1741             : 
    1742             :                 /* rebind */
    1743             :                 RCU_INIT_POINTER(scgrp->subsys[ssid], NULL);
    1744             :                 rcu_assign_pointer(dcgrp->subsys[ssid], css);
    1745             :                 ss->root = dst_root;
    1746             :                 css->cgroup = dcgrp;
    1747             : 
    1748             :                 spin_lock_irq(&css_set_lock);
    1749             :                 hash_for_each(css_set_table, i, cset, hlist)
    1750             :                         list_move_tail(&cset->e_cset_node[ss->id],
    1751             :                                        &dcgrp->e_csets[ss->id]);
    1752             :                 spin_unlock_irq(&css_set_lock);
    1753             : 
    1754             :                 /* default hierarchy doesn't enable controllers by default */
    1755             :                 dst_root->subsys_mask |= 1 << ssid;
    1756             :                 if (dst_root == &cgrp_dfl_root) {
    1757             :                         static_branch_enable(cgroup_subsys_on_dfl_key[ssid]);
    1758             :                 } else {
    1759             :                         dcgrp->subtree_control |= 1 << ssid;
    1760             :                         static_branch_disable(cgroup_subsys_on_dfl_key[ssid]);
    1761             :                 }
    1762             : 
    1763             :                 ret = cgroup_apply_control(dcgrp);
    1764             :                 if (ret)
    1765             :                         pr_warn("partial failure to rebind %s controller (err=%d)\n",
    1766             :                                 ss->name, ret);
    1767             : 
    1768             :                 if (ss->bind)
    1769             :                         ss->bind(css);
    1770           2 :         } while_each_subsys_mask();
    1771             : 
    1772           2 :         kernfs_activate(dcgrp->kn);
    1773           2 :         return 0;
    1774             : }
    1775             : 
    1776         574 : int cgroup_show_path(struct seq_file *sf, struct kernfs_node *kf_node,
    1777             :                      struct kernfs_root *kf_root)
    1778             : {
    1779         574 :         int len = 0;
    1780         574 :         char *buf = NULL;
    1781         574 :         struct cgroup_root *kf_cgroot = cgroup_root_from_kf(kf_root);
    1782         574 :         struct cgroup *ns_cgroup;
    1783             : 
    1784         574 :         buf = kmalloc(PATH_MAX, GFP_KERNEL);
    1785         574 :         if (!buf)
    1786             :                 return -ENOMEM;
    1787             : 
    1788         574 :         spin_lock_irq(&css_set_lock);
    1789         574 :         ns_cgroup = current_cgns_cgroup_from_root(kf_cgroot);
    1790         574 :         len = kernfs_path_from_node(kf_node, ns_cgroup->kn, buf, PATH_MAX);
    1791         574 :         spin_unlock_irq(&css_set_lock);
    1792             : 
    1793         574 :         if (len >= PATH_MAX)
    1794             :                 len = -ERANGE;
    1795         574 :         else if (len > 0) {
    1796         574 :                 seq_escape(sf, buf, " \t\n\\");
    1797         574 :                 len = 0;
    1798             :         }
    1799         574 :         kfree(buf);
    1800         574 :         return len;
    1801             : }
    1802             : 
    1803             : enum cgroup2_param {
    1804             :         Opt_nsdelegate,
    1805             :         Opt_memory_localevents,
    1806             :         Opt_memory_recursiveprot,
    1807             :         nr__cgroup2_params
    1808             : };
    1809             : 
    1810             : static const struct fs_parameter_spec cgroup2_fs_parameters[] = {
    1811             :         fsparam_flag("nsdelegate",            Opt_nsdelegate),
    1812             :         fsparam_flag("memory_localevents",    Opt_memory_localevents),
    1813             :         fsparam_flag("memory_recursiveprot",  Opt_memory_recursiveprot),
    1814             :         {}
    1815             : };
    1816             : 
    1817           2 : static int cgroup2_parse_param(struct fs_context *fc, struct fs_parameter *param)
    1818             : {
    1819           2 :         struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
    1820           2 :         struct fs_parse_result result;
    1821           2 :         int opt;
    1822             : 
    1823           2 :         opt = fs_parse(fc, cgroup2_fs_parameters, param, &result);
    1824           2 :         if (opt < 0)
    1825             :                 return opt;
    1826             : 
    1827           1 :         switch (opt) {
    1828           1 :         case Opt_nsdelegate:
    1829           1 :                 ctx->flags |= CGRP_ROOT_NS_DELEGATE;
    1830           1 :                 return 0;
    1831           0 :         case Opt_memory_localevents:
    1832           0 :                 ctx->flags |= CGRP_ROOT_MEMORY_LOCAL_EVENTS;
    1833           0 :                 return 0;
    1834           0 :         case Opt_memory_recursiveprot:
    1835           0 :                 ctx->flags |= CGRP_ROOT_MEMORY_RECURSIVE_PROT;
    1836           0 :                 return 0;
    1837             :         }
    1838             :         return -EINVAL;
    1839             : }
    1840             : 
    1841           1 : static void apply_cgroup_root_flags(unsigned int root_flags)
    1842             : {
    1843           1 :         if (current->nsproxy->cgroup_ns == &init_cgroup_ns) {
    1844           1 :                 if (root_flags & CGRP_ROOT_NS_DELEGATE)
    1845           1 :                         cgrp_dfl_root.flags |= CGRP_ROOT_NS_DELEGATE;
    1846             :                 else
    1847           0 :                         cgrp_dfl_root.flags &= ~CGRP_ROOT_NS_DELEGATE;
    1848             : 
    1849           1 :                 if (root_flags & CGRP_ROOT_MEMORY_LOCAL_EVENTS)
    1850           0 :                         cgrp_dfl_root.flags |= CGRP_ROOT_MEMORY_LOCAL_EVENTS;
    1851             :                 else
    1852           1 :                         cgrp_dfl_root.flags &= ~CGRP_ROOT_MEMORY_LOCAL_EVENTS;
    1853             : 
    1854           1 :                 if (root_flags & CGRP_ROOT_MEMORY_RECURSIVE_PROT)
    1855           0 :                         cgrp_dfl_root.flags |= CGRP_ROOT_MEMORY_RECURSIVE_PROT;
    1856             :                 else
    1857           1 :                         cgrp_dfl_root.flags &= ~CGRP_ROOT_MEMORY_RECURSIVE_PROT;
    1858             :         }
    1859           1 : }
    1860             : 
    1861         288 : static int cgroup_show_options(struct seq_file *seq, struct kernfs_root *kf_root)
    1862             : {
    1863         288 :         if (cgrp_dfl_root.flags & CGRP_ROOT_NS_DELEGATE)
    1864         288 :                 seq_puts(seq, ",nsdelegate");
    1865         289 :         if (cgrp_dfl_root.flags & CGRP_ROOT_MEMORY_LOCAL_EVENTS)
    1866           0 :                 seq_puts(seq, ",memory_localevents");
    1867         289 :         if (cgrp_dfl_root.flags & CGRP_ROOT_MEMORY_RECURSIVE_PROT)
    1868           0 :                 seq_puts(seq, ",memory_recursiveprot");
    1869         289 :         return 0;
    1870             : }
    1871             : 
    1872           0 : static int cgroup_reconfigure(struct fs_context *fc)
    1873             : {
    1874           0 :         struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
    1875             : 
    1876           0 :         apply_cgroup_root_flags(ctx->flags);
    1877           0 :         return 0;
    1878             : }
    1879             : 
    1880         100 : static void init_cgroup_housekeeping(struct cgroup *cgrp)
    1881             : {
    1882         100 :         struct cgroup_subsys *ss;
    1883         100 :         int ssid;
    1884             : 
    1885         100 :         INIT_LIST_HEAD(&cgrp->self.sibling);
    1886         100 :         INIT_LIST_HEAD(&cgrp->self.children);
    1887         100 :         INIT_LIST_HEAD(&cgrp->cset_links);
    1888         100 :         INIT_LIST_HEAD(&cgrp->pidlists);
    1889         100 :         mutex_init(&cgrp->pidlist_mutex);
    1890         100 :         cgrp->self.cgroup = cgrp;
    1891         100 :         cgrp->self.flags |= CSS_ONLINE;
    1892         100 :         cgrp->dom_cgrp = cgrp;
    1893         100 :         cgrp->max_descendants = INT_MAX;
    1894         100 :         cgrp->max_depth = INT_MAX;
    1895         100 :         INIT_LIST_HEAD(&cgrp->rstat_css_list);
    1896         100 :         prev_cputime_init(&cgrp->prev_cputime);
    1897             : 
    1898         100 :         for_each_subsys(ss, ssid)
    1899             :                 INIT_LIST_HEAD(&cgrp->e_csets[ssid]);
    1900             : 
    1901         100 :         init_waitqueue_head(&cgrp->offline_waitq);
    1902         100 :         INIT_WORK(&cgrp->release_agent_work, cgroup1_release_agent);
    1903         100 : }
    1904             : 
    1905           2 : void init_cgroup_root(struct cgroup_fs_context *ctx)
    1906             : {
    1907           2 :         struct cgroup_root *root = ctx->root;
    1908           2 :         struct cgroup *cgrp = &root->cgrp;
    1909             : 
    1910           2 :         INIT_LIST_HEAD(&root->root_list);
    1911           2 :         atomic_set(&root->nr_cgrps, 1);
    1912           2 :         cgrp->root = root;
    1913           2 :         init_cgroup_housekeeping(cgrp);
    1914             : 
    1915           2 :         root->flags = ctx->flags;
    1916           2 :         if (ctx->release_agent)
    1917           0 :                 strscpy(root->release_agent_path, ctx->release_agent, PATH_MAX);
    1918           2 :         if (ctx->name)
    1919           1 :                 strscpy(root->name, ctx->name, MAX_CGROUP_ROOT_NAMELEN);
    1920           2 :         if (ctx->cpuset_clone_children)
    1921           0 :                 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags);
    1922           2 : }
    1923             : 
    1924           2 : int cgroup_setup_root(struct cgroup_root *root, u16 ss_mask)
    1925             : {
    1926           2 :         LIST_HEAD(tmp_links);
    1927           2 :         struct cgroup *root_cgrp = &root->cgrp;
    1928           2 :         struct kernfs_syscall_ops *kf_sops;
    1929           2 :         struct css_set *cset;
    1930           2 :         int i, ret;
    1931             : 
    1932           6 :         lockdep_assert_held(&cgroup_mutex);
    1933             : 
    1934           2 :         ret = percpu_ref_init(&root_cgrp->self.refcnt, css_release,
    1935             :                               0, GFP_KERNEL);
    1936           2 :         if (ret)
    1937           0 :                 goto out;
    1938             : 
    1939             :         /*
    1940             :          * We're accessing css_set_count without locking css_set_lock here,
    1941             :          * but that's OK - it can only be increased by someone holding
    1942             :          * cgroup_lock, and that's us.  Later rebinding may disable
    1943             :          * controllers on the default hierarchy and thus create new csets,
    1944             :          * which can't be more than the existing ones.  Allocate 2x.
    1945             :          */
    1946           2 :         ret = allocate_cgrp_cset_links(2 * css_set_count, &tmp_links);
    1947           2 :         if (ret)
    1948           0 :                 goto cancel_ref;
    1949             : 
    1950           2 :         ret = cgroup_init_root_id(root);
    1951           2 :         if (ret)
    1952           0 :                 goto cancel_ref;
    1953             : 
    1954           4 :         kf_sops = root == &cgrp_dfl_root ?
    1955           2 :                 &cgroup_kf_syscall_ops : &cgroup1_kf_syscall_ops;
    1956             : 
    1957           2 :         root->kf_root = kernfs_create_root(kf_sops,
    1958             :                                            KERNFS_ROOT_CREATE_DEACTIVATED |
    1959             :                                            KERNFS_ROOT_SUPPORT_EXPORTOP |
    1960             :                                            KERNFS_ROOT_SUPPORT_USER_XATTR,
    1961             :                                            root_cgrp);
    1962           2 :         if (IS_ERR(root->kf_root)) {
    1963           0 :                 ret = PTR_ERR(root->kf_root);
    1964           0 :                 goto exit_root_id;
    1965             :         }
    1966           2 :         root_cgrp->kn = root->kf_root->kn;
    1967           2 :         WARN_ON_ONCE(cgroup_ino(root_cgrp) != 1);
    1968           2 :         root_cgrp->ancestor_ids[0] = cgroup_id(root_cgrp);
    1969             : 
    1970           2 :         ret = css_populate_dir(&root_cgrp->self);
    1971           2 :         if (ret)
    1972           0 :                 goto destroy_root;
    1973             : 
    1974           2 :         ret = rebind_subsystems(root, ss_mask);
    1975           2 :         if (ret)
    1976           0 :                 goto destroy_root;
    1977             : 
    1978           2 :         ret = cgroup_bpf_inherit(root_cgrp);
    1979           2 :         WARN_ON_ONCE(ret);
    1980             : 
    1981           2 :         trace_cgroup_setup_root(root);
    1982             : 
    1983             :         /*
    1984             :          * There must be no failure case after here, since rebinding takes
    1985             :          * care of subsystems' refcounts, which are explicitly dropped in
    1986             :          * the failure exit path.
    1987             :          */
    1988           2 :         list_add(&root->root_list, &cgroup_roots);
    1989           2 :         cgroup_root_count++;
    1990             : 
    1991             :         /*
    1992             :          * Link the root cgroup in this hierarchy into all the css_set
    1993             :          * objects.
    1994             :          */
    1995           2 :         spin_lock_irq(&css_set_lock);
    1996         262 :         hash_for_each(css_set_table, i, cset, hlist) {
    1997           2 :                 link_css_set(&tmp_links, cset, root_cgrp);
    1998           2 :                 if (css_set_populated(cset))
    1999           1 :                         cgroup_update_populated(root_cgrp, true);
    2000             :         }
    2001           2 :         spin_unlock_irq(&css_set_lock);
    2002             : 
    2003           2 :         BUG_ON(!list_empty(&root_cgrp->self.children));
    2004           2 :         BUG_ON(atomic_read(&root->nr_cgrps) != 1);
    2005             : 
    2006           2 :         ret = 0;
    2007           2 :         goto out;
    2008             : 
    2009           0 : destroy_root:
    2010           0 :         kernfs_destroy_root(root->kf_root);
    2011           0 :         root->kf_root = NULL;
    2012           0 : exit_root_id:
    2013           0 :         cgroup_exit_root_id(root);
    2014           0 : cancel_ref:
    2015           0 :         percpu_ref_exit(&root_cgrp->self.refcnt);
    2016           2 : out:
    2017           2 :         free_cgrp_cset_links(&tmp_links);
    2018           2 :         return ret;
    2019             : }
    2020             : 
    2021           2 : int cgroup_do_get_tree(struct fs_context *fc)
    2022             : {
    2023           2 :         struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
    2024           2 :         int ret;
    2025             : 
    2026           2 :         ctx->kfc.root = ctx->root->kf_root;
    2027           2 :         if (fc->fs_type == &cgroup2_fs_type)
    2028           1 :                 ctx->kfc.magic = CGROUP2_SUPER_MAGIC;
    2029             :         else
    2030           1 :                 ctx->kfc.magic = CGROUP_SUPER_MAGIC;
    2031           2 :         ret = kernfs_get_tree(fc);
    2032             : 
    2033             :         /*
    2034             :          * In non-init cgroup namespace, instead of root cgroup's dentry,
    2035             :          * we return the dentry corresponding to the cgroupns->root_cgrp.
    2036             :          */
    2037           2 :         if (!ret && ctx->ns != &init_cgroup_ns) {
    2038           0 :                 struct dentry *nsdentry;
    2039           0 :                 struct super_block *sb = fc->root->d_sb;
    2040           0 :                 struct cgroup *cgrp;
    2041             : 
    2042           0 :                 mutex_lock(&cgroup_mutex);
    2043           0 :                 spin_lock_irq(&css_set_lock);
    2044             : 
    2045           0 :                 cgrp = cset_cgroup_from_root(ctx->ns->root_cset, ctx->root);
    2046             : 
    2047           0 :                 spin_unlock_irq(&css_set_lock);
    2048           0 :                 mutex_unlock(&cgroup_mutex);
    2049             : 
    2050           0 :                 nsdentry = kernfs_node_dentry(cgrp->kn, sb);
    2051           0 :                 dput(fc->root);
    2052           0 :                 if (IS_ERR(nsdentry)) {
    2053           0 :                         deactivate_locked_super(sb);
    2054           0 :                         ret = PTR_ERR(nsdentry);
    2055           0 :                         nsdentry = NULL;
    2056             :                 }
    2057           0 :                 fc->root = nsdentry;
    2058             :         }
    2059             : 
    2060           2 :         if (!ctx->kfc.new_sb_created)
    2061           0 :                 cgroup_put(&ctx->root->cgrp);
    2062             : 
    2063           2 :         return ret;
    2064             : }
    2065             : 
    2066             : /*
    2067             :  * Destroy a cgroup filesystem context.
    2068             :  */
    2069           2 : static void cgroup_fs_context_free(struct fs_context *fc)
    2070             : {
    2071           2 :         struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
    2072             : 
    2073           2 :         kfree(ctx->name);
    2074           2 :         kfree(ctx->release_agent);
    2075           2 :         put_cgroup_ns(ctx->ns);
    2076           2 :         kernfs_free_fs_context(fc);
    2077           2 :         kfree(ctx);
    2078           2 : }
    2079             : 
    2080           1 : static int cgroup_get_tree(struct fs_context *fc)
    2081             : {
    2082           1 :         struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
    2083           1 :         int ret;
    2084             : 
    2085           1 :         cgrp_dfl_visible = true;
    2086           1 :         cgroup_get_live(&cgrp_dfl_root.cgrp);
    2087           1 :         ctx->root = &cgrp_dfl_root;
    2088             : 
    2089           1 :         ret = cgroup_do_get_tree(fc);
    2090           1 :         if (!ret)
    2091           1 :                 apply_cgroup_root_flags(ctx->flags);
    2092           1 :         return ret;
    2093             : }
    2094             : 
    2095             : static const struct fs_context_operations cgroup_fs_context_ops = {
    2096             :         .free           = cgroup_fs_context_free,
    2097             :         .parse_param    = cgroup2_parse_param,
    2098             :         .get_tree       = cgroup_get_tree,
    2099             :         .reconfigure    = cgroup_reconfigure,
    2100             : };
    2101             : 
    2102             : static const struct fs_context_operations cgroup1_fs_context_ops = {
    2103             :         .free           = cgroup_fs_context_free,
    2104             :         .parse_param    = cgroup1_parse_param,
    2105             :         .get_tree       = cgroup1_get_tree,
    2106             :         .reconfigure    = cgroup1_reconfigure,
    2107             : };
    2108             : 
    2109             : /*
    2110             :  * Initialise the cgroup filesystem creation/reconfiguration context.  Notably,
    2111             :  * we select the namespace we're going to use.
    2112             :  */
    2113           2 : static int cgroup_init_fs_context(struct fs_context *fc)
    2114             : {
    2115           2 :         struct cgroup_fs_context *ctx;
    2116             : 
    2117           2 :         ctx = kzalloc(sizeof(struct cgroup_fs_context), GFP_KERNEL);
    2118           2 :         if (!ctx)
    2119             :                 return -ENOMEM;
    2120             : 
    2121           2 :         ctx->ns = current->nsproxy->cgroup_ns;
    2122           2 :         get_cgroup_ns(ctx->ns);
    2123           2 :         fc->fs_private = &ctx->kfc;
    2124           2 :         if (fc->fs_type == &cgroup2_fs_type)
    2125           1 :                 fc->ops = &cgroup_fs_context_ops;
    2126             :         else
    2127           1 :                 fc->ops = &cgroup1_fs_context_ops;
    2128           2 :         put_user_ns(fc->user_ns);
    2129           2 :         fc->user_ns = get_user_ns(ctx->ns->user_ns);
    2130           2 :         fc->global = true;
    2131           2 :         return 0;
    2132             : }
    2133             : 
    2134           0 : static void cgroup_kill_sb(struct super_block *sb)
    2135             : {
    2136           0 :         struct kernfs_root *kf_root = kernfs_root_from_sb(sb);
    2137           0 :         struct cgroup_root *root = cgroup_root_from_kf(kf_root);
    2138             : 
    2139             :         /*
    2140             :          * If @root doesn't have any children, start killing it.
    2141             :          * This prevents new mounts by disabling percpu_ref_tryget_live().
    2142             :          * cgroup_mount() may wait for @root's release.
    2143             :          *
    2144             :          * And don't kill the default root.
    2145             :          */
    2146           0 :         if (list_empty(&root->cgrp.self.children) && root != &cgrp_dfl_root &&
    2147           0 :             !percpu_ref_is_dying(&root->cgrp.self.refcnt))
    2148           0 :                 percpu_ref_kill(&root->cgrp.self.refcnt);
    2149           0 :         cgroup_put(&root->cgrp);
    2150           0 :         kernfs_kill_sb(sb);
    2151           0 : }
    2152             : 
    2153             : struct file_system_type cgroup_fs_type = {
    2154             :         .name                   = "cgroup",
    2155             :         .init_fs_context        = cgroup_init_fs_context,
    2156             :         .parameters             = cgroup1_fs_parameters,
    2157             :         .kill_sb                = cgroup_kill_sb,
    2158             :         .fs_flags               = FS_USERNS_MOUNT,
    2159             : };
    2160             : 
    2161             : static struct file_system_type cgroup2_fs_type = {
    2162             :         .name                   = "cgroup2",
    2163             :         .init_fs_context        = cgroup_init_fs_context,
    2164             :         .parameters             = cgroup2_fs_parameters,
    2165             :         .kill_sb                = cgroup_kill_sb,
    2166             :         .fs_flags               = FS_USERNS_MOUNT,
    2167             : };
    2168             : 
    2169             : #ifdef CONFIG_CPUSETS
    2170             : static const struct fs_context_operations cpuset_fs_context_ops = {
    2171             :         .get_tree       = cgroup1_get_tree,
    2172             :         .free           = cgroup_fs_context_free,
    2173             : };
    2174             : 
    2175             : /*
    2176             :  * This is ugly, but preserves the userspace API for existing cpuset
    2177             :  * users. If someone tries to mount the "cpuset" filesystem, we
    2178             :  * silently switch it to mount "cgroup" instead
    2179             :  */
    2180             : static int cpuset_init_fs_context(struct fs_context *fc)
    2181             : {
    2182             :         char *agent = kstrdup("/sbin/cpuset_release_agent", GFP_USER);
    2183             :         struct cgroup_fs_context *ctx;
    2184             :         int err;
    2185             : 
    2186             :         err = cgroup_init_fs_context(fc);
    2187             :         if (err) {
    2188             :                 kfree(agent);
    2189             :                 return err;
    2190             :         }
    2191             : 
    2192             :         fc->ops = &cpuset_fs_context_ops;
    2193             : 
    2194             :         ctx = cgroup_fc2context(fc);
    2195             :         ctx->subsys_mask = 1 << cpuset_cgrp_id;
    2196             :         ctx->flags |= CGRP_ROOT_NOPREFIX;
    2197             :         ctx->release_agent = agent;
    2198             : 
    2199             :         get_filesystem(&cgroup_fs_type);
    2200             :         put_filesystem(fc->fs_type);
    2201             :         fc->fs_type = &cgroup_fs_type;
    2202             : 
    2203             :         return 0;
    2204             : }
    2205             : 
    2206             : static struct file_system_type cpuset_fs_type = {
    2207             :         .name                   = "cpuset",
    2208             :         .init_fs_context        = cpuset_init_fs_context,
    2209             :         .fs_flags               = FS_USERNS_MOUNT,
    2210             : };
    2211             : #endif
    2212             : 
    2213         312 : int cgroup_path_ns_locked(struct cgroup *cgrp, char *buf, size_t buflen,
    2214             :                           struct cgroup_namespace *ns)
    2215             : {
    2216         312 :         struct cgroup *root = cset_cgroup_from_root(ns->root_cset, cgrp->root);
    2217             : 
    2218         312 :         return kernfs_path_from_node(cgrp->kn, root->kn, buf, buflen);
    2219             : }
    2220             : 
    2221           0 : int cgroup_path_ns(struct cgroup *cgrp, char *buf, size_t buflen,
    2222             :                    struct cgroup_namespace *ns)
    2223             : {
    2224           0 :         int ret;
    2225             : 
    2226           0 :         mutex_lock(&cgroup_mutex);
    2227           0 :         spin_lock_irq(&css_set_lock);
    2228             : 
    2229           0 :         ret = cgroup_path_ns_locked(cgrp, buf, buflen, ns);
    2230             : 
    2231           0 :         spin_unlock_irq(&css_set_lock);
    2232           0 :         mutex_unlock(&cgroup_mutex);
    2233             : 
    2234           0 :         return ret;
    2235             : }
    2236             : EXPORT_SYMBOL_GPL(cgroup_path_ns);
    2237             : 
    2238             : /**
    2239             :  * task_cgroup_path - cgroup path of a task in the first cgroup hierarchy
    2240             :  * @task: target task
    2241             :  * @buf: the buffer to write the path into
    2242             :  * @buflen: the length of the buffer
    2243             :  *
    2244             :  * Determine @task's cgroup on the first (the one with the lowest non-zero
    2245             :  * hierarchy_id) cgroup hierarchy and copy its path into @buf.  This
    2246             :  * function grabs cgroup_mutex and shouldn't be used inside locks used by
    2247             :  * cgroup controller callbacks.
    2248             :  *
    2249             :  * Return value is the same as kernfs_path().
    2250             :  */
    2251           0 : int task_cgroup_path(struct task_struct *task, char *buf, size_t buflen)
    2252             : {
    2253           0 :         struct cgroup_root *root;
    2254           0 :         struct cgroup *cgrp;
    2255           0 :         int hierarchy_id = 1;
    2256           0 :         int ret;
    2257             : 
    2258           0 :         mutex_lock(&cgroup_mutex);
    2259           0 :         spin_lock_irq(&css_set_lock);
    2260             : 
    2261           0 :         root = idr_get_next(&cgroup_hierarchy_idr, &hierarchy_id);
    2262             : 
    2263           0 :         if (root) {
    2264           0 :                 cgrp = task_cgroup_from_root(task, root);
    2265           0 :                 ret = cgroup_path_ns_locked(cgrp, buf, buflen, &init_cgroup_ns);
    2266             :         } else {
    2267             :                 /* if no hierarchy exists, everyone is in "/" */
    2268           0 :                 ret = strlcpy(buf, "/", buflen);
    2269             :         }
    2270             : 
    2271           0 :         spin_unlock_irq(&css_set_lock);
    2272           0 :         mutex_unlock(&cgroup_mutex);
    2273           0 :         return ret;
    2274             : }
    2275             : EXPORT_SYMBOL_GPL(task_cgroup_path);
    2276             : 
    2277             : /**
    2278             :  * cgroup_migrate_add_task - add a migration target task to a migration context
    2279             :  * @task: target task
    2280             :  * @mgctx: target migration context
    2281             :  *
    2282             :  * Add @task, which is a migration target, to @mgctx->tset.  This function
    2283             :  * becomes noop if @task doesn't need to be migrated.  @task's css_set
    2284             :  * should have been added as a migration source and @task->cg_list will be
    2285             :  * moved from the css_set's tasks list to mg_tasks one.
    2286             :  */
    2287         198 : static void cgroup_migrate_add_task(struct task_struct *task,
    2288             :                                     struct cgroup_mgctx *mgctx)
    2289             : {
    2290         198 :         struct css_set *cset;
    2291             : 
    2292         594 :         lockdep_assert_held(&css_set_lock);
    2293             : 
    2294             :         /* @task either already exited or can't exit until the end */
    2295         198 :         if (task->flags & PF_EXITING)
    2296             :                 return;
    2297             : 
    2298             :         /* cgroup_threadgroup_rwsem protects racing against forks */
    2299         198 :         WARN_ON_ONCE(list_empty(&task->cg_list));
    2300             : 
    2301         198 :         cset = task_css_set(task);
    2302         198 :         if (!cset->mg_src_cgrp)
    2303             :                 return;
    2304             : 
    2305         106 :         mgctx->tset.nr_tasks++;
    2306             : 
    2307         106 :         list_move_tail(&task->cg_list, &cset->mg_tasks);
    2308         106 :         if (list_empty(&cset->mg_node))
    2309         106 :                 list_add_tail(&cset->mg_node,
    2310             :                               &mgctx->tset.src_csets);
    2311         106 :         if (list_empty(&cset->mg_dst_cset->mg_node))
    2312         106 :                 list_add_tail(&cset->mg_dst_cset->mg_node,
    2313             :                               &mgctx->tset.dst_csets);
    2314             : }
    2315             : 
    2316             : /**
    2317             :  * cgroup_taskset_first - reset taskset and return the first task
    2318             :  * @tset: taskset of interest
    2319             :  * @dst_cssp: output variable for the destination css
    2320             :  *
    2321             :  * @tset iteration is initialized and the first task is returned.
    2322             :  */
    2323           0 : struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset,
    2324             :                                          struct cgroup_subsys_state **dst_cssp)
    2325             : {
    2326           0 :         tset->cur_cset = list_first_entry(tset->csets, struct css_set, mg_node);
    2327           0 :         tset->cur_task = NULL;
    2328             : 
    2329           0 :         return cgroup_taskset_next(tset, dst_cssp);
    2330             : }
    2331             : 
    2332             : /**
    2333             :  * cgroup_taskset_next - iterate to the next task in taskset
    2334             :  * @tset: taskset of interest
    2335             :  * @dst_cssp: output variable for the destination css
    2336             :  *
    2337             :  * Return the next task in @tset.  Iteration must have been initialized
    2338             :  * with cgroup_taskset_first().
    2339             :  */
    2340           0 : struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset,
    2341             :                                         struct cgroup_subsys_state **dst_cssp)
    2342             : {
    2343           0 :         struct css_set *cset = tset->cur_cset;
    2344           0 :         struct task_struct *task = tset->cur_task;
    2345             : 
    2346           0 :         while (&cset->mg_node != tset->csets) {
    2347           0 :                 if (!task)
    2348           0 :                         task = list_first_entry(&cset->mg_tasks,
    2349             :                                                 struct task_struct, cg_list);
    2350             :                 else
    2351           0 :                         task = list_next_entry(task, cg_list);
    2352             : 
    2353           0 :                 if (&task->cg_list != &cset->mg_tasks) {
    2354           0 :                         tset->cur_cset = cset;
    2355           0 :                         tset->cur_task = task;
    2356             : 
    2357             :                         /*
    2358             :                          * This function may be called both before and
    2359             :                          * after cgroup_taskset_migrate().  The two cases
    2360             :                          * can be distinguished by looking at whether @cset
    2361             :                          * has its ->mg_dst_cset set.
    2362             :                          */
    2363           0 :                         if (cset->mg_dst_cset)
    2364           0 :                                 *dst_cssp = cset->mg_dst_cset->subsys[tset->ssid];
    2365             :                         else
    2366           0 :                                 *dst_cssp = cset->subsys[tset->ssid];
    2367             : 
    2368           0 :                         return task;
    2369             :                 }
    2370             : 
    2371           0 :                 cset = list_next_entry(cset, mg_node);
    2372           0 :                 task = NULL;
    2373             :         }
    2374             : 
    2375             :         return NULL;
    2376             : }
    2377             : 
    2378             : /**
    2379             :  * cgroup_taskset_migrate - migrate a taskset
    2380             :  * @mgctx: migration context
    2381             :  *
    2382             :  * Migrate tasks in @mgctx as setup by migration preparation functions.
    2383             :  * This function fails iff one of the ->can_attach callbacks fails and
    2384             :  * guarantees that either all or none of the tasks in @mgctx are migrated.
    2385             :  * @mgctx is consumed regardless of success.
    2386             :  */
    2387         198 : static int cgroup_migrate_execute(struct cgroup_mgctx *mgctx)
    2388             : {
    2389         198 :         struct cgroup_taskset *tset = &mgctx->tset;
    2390         198 :         struct cgroup_subsys *ss;
    2391         198 :         struct task_struct *task, *tmp_task;
    2392         198 :         struct css_set *cset, *tmp_cset;
    2393         198 :         int ssid, failed_ssid, ret;
    2394             : 
    2395             :         /* check that we can legitimately attach to the cgroup */
    2396         198 :         if (tset->nr_tasks) {
    2397             :                 do_each_subsys_mask(ss, ssid, mgctx->ss_mask) {
    2398             :                         if (ss->can_attach) {
    2399             :                                 tset->ssid = ssid;
    2400             :                                 ret = ss->can_attach(tset);
    2401             :                                 if (ret) {
    2402             :                                         failed_ssid = ssid;
    2403             :                                         goto out_cancel_attach;
    2404             :                                 }
    2405             :                         }
    2406         198 :                 } while_each_subsys_mask();
    2407             :         }
    2408             : 
    2409             :         /*
    2410             :          * Now that we're guaranteed success, proceed to move all tasks to
    2411             :          * the new cgroup.  There are no failure cases after here, so this
    2412             :          * is the commit point.
    2413             :          */
    2414         198 :         spin_lock_irq(&css_set_lock);
    2415         304 :         list_for_each_entry(cset, &tset->src_csets, mg_node) {
    2416         212 :                 list_for_each_entry_safe(task, tmp_task, &cset->mg_tasks, cg_list) {
    2417         106 :                         struct css_set *from_cset = task_css_set(task);
    2418         106 :                         struct css_set *to_cset = cset->mg_dst_cset;
    2419             : 
    2420         106 :                         get_css_set(to_cset);
    2421         106 :                         to_cset->nr_tasks++;
    2422         106 :                         css_set_move_task(task, from_cset, to_cset, true);
    2423         106 :                         from_cset->nr_tasks--;
    2424             :                         /*
    2425             :                          * If the source or destination cgroup is frozen,
    2426             :                          * the task might require to change its state.
    2427             :                          */
    2428         106 :                         cgroup_freezer_migrate_task(task, from_cset->dfl_cgrp,
    2429             :                                                     to_cset->dfl_cgrp);
    2430         106 :                         put_css_set_locked(from_cset);
    2431             : 
    2432             :                 }
    2433             :         }
    2434         198 :         spin_unlock_irq(&css_set_lock);
    2435             : 
    2436             :         /*
    2437             :          * Migration is committed, all target tasks are now on dst_csets.
    2438             :          * Nothing is sensitive to fork() after this point.  Notify
    2439             :          * controllers that migration is complete.
    2440             :          */
    2441         198 :         tset->csets = &tset->dst_csets;
    2442             : 
    2443         198 :         if (tset->nr_tasks) {
    2444             :                 do_each_subsys_mask(ss, ssid, mgctx->ss_mask) {
    2445             :                         if (ss->attach) {
    2446             :                                 tset->ssid = ssid;
    2447             :                                 ss->attach(tset);
    2448             :                         }
    2449         198 :                 } while_each_subsys_mask();
    2450             :         }
    2451             : 
    2452         198 :         ret = 0;
    2453         198 :         goto out_release_tset;
    2454             : 
    2455             : out_cancel_attach:
    2456             :         if (tset->nr_tasks) {
    2457             :                 do_each_subsys_mask(ss, ssid, mgctx->ss_mask) {
    2458             :                         if (ssid == failed_ssid)
    2459             :                                 break;
    2460             :                         if (ss->cancel_attach) {
    2461             :                                 tset->ssid = ssid;
    2462             :                                 ss->cancel_attach(tset);
    2463             :                         }
    2464         198 :                 } while_each_subsys_mask();
    2465             :         }
    2466         198 : out_release_tset:
    2467         198 :         spin_lock_irq(&css_set_lock);
    2468         198 :         list_splice_init(&tset->dst_csets, &tset->src_csets);
    2469         410 :         list_for_each_entry_safe(cset, tmp_cset, &tset->src_csets, mg_node) {
    2470         212 :                 list_splice_tail_init(&cset->mg_tasks, &cset->tasks);
    2471         212 :                 list_del_init(&cset->mg_node);
    2472             :         }
    2473         198 :         spin_unlock_irq(&css_set_lock);
    2474             : 
    2475             :         /*
    2476             :          * Re-initialize the cgroup_taskset structure in case it is reused
    2477             :          * again in another cgroup_migrate_add_task()/cgroup_migrate_execute()
    2478             :          * iteration.
    2479             :          */
    2480         198 :         tset->nr_tasks = 0;
    2481         198 :         tset->csets    = &tset->src_csets;
    2482         198 :         return ret;
    2483             : }
    2484             : 
    2485             : /**
    2486             :  * cgroup_migrate_vet_dst - verify whether a cgroup can be migration destination
    2487             :  * @dst_cgrp: destination cgroup to test
    2488             :  *
    2489             :  * On the default hierarchy, except for the mixable, (possible) thread root
    2490             :  * and threaded cgroups, subtree_control must be zero for migration
    2491             :  * destination cgroups with tasks so that child cgroups don't compete
    2492             :  * against tasks.
    2493             :  */
    2494          99 : int cgroup_migrate_vet_dst(struct cgroup *dst_cgrp)
    2495             : {
    2496             :         /* v1 doesn't have any restriction */
    2497          99 :         if (!cgroup_on_dfl(dst_cgrp))
    2498             :                 return 0;
    2499             : 
    2500             :         /* verify @dst_cgrp can host resources */
    2501          99 :         if (!cgroup_is_valid_domain(dst_cgrp->dom_cgrp))
    2502             :                 return -EOPNOTSUPP;
    2503             : 
    2504             :         /* mixables don't care */
    2505          99 :         if (cgroup_is_mixable(dst_cgrp))
    2506             :                 return 0;
    2507             : 
    2508             :         /*
    2509             :          * If @dst_cgrp is already or can become a thread root or is
    2510             :          * threaded, it doesn't matter.
    2511             :          */
    2512          99 :         if (cgroup_can_be_thread_root(dst_cgrp) || cgroup_is_threaded(dst_cgrp))
    2513             :                 return 0;
    2514             : 
    2515             :         /* apply no-internal-process constraint */
    2516           0 :         if (dst_cgrp->subtree_control)
    2517           0 :                 return -EBUSY;
    2518             : 
    2519             :         return 0;
    2520             : }
    2521             : 
    2522             : /**
    2523             :  * cgroup_migrate_finish - cleanup after attach
    2524             :  * @mgctx: migration context
    2525             :  *
    2526             :  * Undo cgroup_migrate_add_src() and cgroup_migrate_prepare_dst().  See
    2527             :  * those functions for details.
    2528             :  */
    2529         198 : void cgroup_migrate_finish(struct cgroup_mgctx *mgctx)
    2530             : {
    2531         198 :         LIST_HEAD(preloaded);
    2532         198 :         struct css_set *cset, *tmp_cset;
    2533             : 
    2534         594 :         lockdep_assert_held(&cgroup_mutex);
    2535             : 
    2536         198 :         spin_lock_irq(&css_set_lock);
    2537             : 
    2538         198 :         list_splice_tail_init(&mgctx->preloaded_src_csets, &preloaded);
    2539         198 :         list_splice_tail_init(&mgctx->preloaded_dst_csets, &preloaded);
    2540             : 
    2541         410 :         list_for_each_entry_safe(cset, tmp_cset, &preloaded, mg_preload_node) {
    2542         212 :                 cset->mg_src_cgrp = NULL;
    2543         212 :                 cset->mg_dst_cgrp = NULL;
    2544         212 :                 cset->mg_dst_cset = NULL;
    2545         212 :                 list_del_init(&cset->mg_preload_node);
    2546         212 :                 put_css_set_locked(cset);
    2547             :         }
    2548             : 
    2549         198 :         spin_unlock_irq(&css_set_lock);
    2550         198 : }
    2551             : 
    2552             : /**
    2553             :  * cgroup_migrate_add_src - add a migration source css_set
    2554             :  * @src_cset: the source css_set to add
    2555             :  * @dst_cgrp: the destination cgroup
    2556             :  * @mgctx: migration context
    2557             :  *
    2558             :  * Tasks belonging to @src_cset are about to be migrated to @dst_cgrp.  Pin
    2559             :  * @src_cset and add it to @mgctx->src_csets, which should later be cleaned
    2560             :  * up by cgroup_migrate_finish().
    2561             :  *
    2562             :  * This function may be called without holding cgroup_threadgroup_rwsem
    2563             :  * even if the target is a process.  Threads may be created and destroyed
    2564             :  * but as long as cgroup_mutex is not dropped, no new css_set can be put
    2565             :  * into play and the preloaded css_sets are guaranteed to cover all
    2566             :  * migrations.
    2567             :  */
    2568         198 : void cgroup_migrate_add_src(struct css_set *src_cset,
    2569             :                             struct cgroup *dst_cgrp,
    2570             :                             struct cgroup_mgctx *mgctx)
    2571             : {
    2572         198 :         struct cgroup *src_cgrp;
    2573             : 
    2574         594 :         lockdep_assert_held(&cgroup_mutex);
    2575         594 :         lockdep_assert_held(&css_set_lock);
    2576             : 
    2577             :         /*
    2578             :          * If ->dead, @src_set is associated with one or more dead cgroups
    2579             :          * and doesn't contain any migratable tasks.  Ignore it early so
    2580             :          * that the rest of migration path doesn't get confused by it.
    2581             :          */
    2582         198 :         if (src_cset->dead)
    2583             :                 return;
    2584             : 
    2585         198 :         src_cgrp = cset_cgroup_from_root(src_cset, dst_cgrp->root);
    2586             : 
    2587         198 :         if (!list_empty(&src_cset->mg_preload_node))
    2588             :                 return;
    2589             : 
    2590         198 :         WARN_ON(src_cset->mg_src_cgrp);
    2591         198 :         WARN_ON(src_cset->mg_dst_cgrp);
    2592         198 :         WARN_ON(!list_empty(&src_cset->mg_tasks));
    2593         198 :         WARN_ON(!list_empty(&src_cset->mg_node));
    2594             : 
    2595         198 :         src_cset->mg_src_cgrp = src_cgrp;
    2596         198 :         src_cset->mg_dst_cgrp = dst_cgrp;
    2597         198 :         get_css_set(src_cset);
    2598         198 :         list_add_tail(&src_cset->mg_preload_node, &mgctx->preloaded_src_csets);
    2599             : }
    2600             : 
    2601             : /**
    2602             :  * cgroup_migrate_prepare_dst - prepare destination css_sets for migration
    2603             :  * @mgctx: migration context
    2604             :  *
    2605             :  * Tasks are about to be moved and all the source css_sets have been
    2606             :  * preloaded to @mgctx->preloaded_src_csets.  This function looks up and
    2607             :  * pins all destination css_sets, links each to its source, and append them
    2608             :  * to @mgctx->preloaded_dst_csets.
    2609             :  *
    2610             :  * This function must be called after cgroup_migrate_add_src() has been
    2611             :  * called on each migration source css_set.  After migration is performed
    2612             :  * using cgroup_migrate(), cgroup_migrate_finish() must be called on
    2613             :  * @mgctx.
    2614             :  */
    2615         198 : int cgroup_migrate_prepare_dst(struct cgroup_mgctx *mgctx)
    2616             : {
    2617         198 :         struct css_set *src_cset, *tmp_cset;
    2618             : 
    2619         594 :         lockdep_assert_held(&cgroup_mutex);
    2620             : 
    2621             :         /* look up the dst cset for each src cset and link it to src */
    2622         396 :         list_for_each_entry_safe(src_cset, tmp_cset, &mgctx->preloaded_src_csets,
    2623             :                                  mg_preload_node) {
    2624         198 :                 struct css_set *dst_cset;
    2625         198 :                 struct cgroup_subsys *ss;
    2626         198 :                 int ssid;
    2627             : 
    2628         198 :                 dst_cset = find_css_set(src_cset, src_cset->mg_dst_cgrp);
    2629         198 :                 if (!dst_cset)
    2630             :                         return -ENOMEM;
    2631             : 
    2632         396 :                 WARN_ON_ONCE(src_cset->mg_dst_cset || dst_cset->mg_dst_cset);
    2633             : 
    2634             :                 /*
    2635             :                  * If src cset equals dst, it's noop.  Drop the src.
    2636             :                  * cgroup_migrate() will skip the cset too.  Note that we
    2637             :                  * can't handle src == dst as some nodes are used by both.
    2638             :                  */
    2639         198 :                 if (src_cset == dst_cset) {
    2640          92 :                         src_cset->mg_src_cgrp = NULL;
    2641          92 :                         src_cset->mg_dst_cgrp = NULL;
    2642          92 :                         list_del_init(&src_cset->mg_preload_node);
    2643          92 :                         put_css_set(src_cset);
    2644          92 :                         put_css_set(dst_cset);
    2645          92 :                         continue;
    2646             :                 }
    2647             : 
    2648         106 :                 src_cset->mg_dst_cset = dst_cset;
    2649             : 
    2650         106 :                 if (list_empty(&dst_cset->mg_preload_node))
    2651         198 :                         list_add_tail(&dst_cset->mg_preload_node,
    2652             :                                       &mgctx->preloaded_dst_csets);
    2653             :                 else
    2654           0 :                         put_css_set(dst_cset);
    2655             : 
    2656             :                 for_each_subsys(ss, ssid)
    2657             :                         if (src_cset->subsys[ssid] != dst_cset->subsys[ssid])
    2658             :                                 mgctx->ss_mask |= 1 << ssid;
    2659             :         }
    2660             : 
    2661             :         return 0;
    2662             : }
    2663             : 
    2664             : /**
    2665             :  * cgroup_migrate - migrate a process or task to a cgroup
    2666             :  * @leader: the leader of the process or the task to migrate
    2667             :  * @threadgroup: whether @leader points to the whole process or a single task
    2668             :  * @mgctx: migration context
    2669             :  *
    2670             :  * Migrate a process or task denoted by @leader.  If migrating a process,
    2671             :  * the caller must be holding cgroup_threadgroup_rwsem.  The caller is also
    2672             :  * responsible for invoking cgroup_migrate_add_src() and
    2673             :  * cgroup_migrate_prepare_dst() on the targets before invoking this
    2674             :  * function and following up with cgroup_migrate_finish().
    2675             :  *
    2676             :  * As long as a controller's ->can_attach() doesn't fail, this function is
    2677             :  * guaranteed to succeed.  This means that, excluding ->can_attach()
    2678             :  * failure, when migrating multiple targets, the success or failure can be
    2679             :  * decided for all targets by invoking group_migrate_prepare_dst() before
    2680             :  * actually starting migrating.
    2681             :  */
    2682         198 : int cgroup_migrate(struct task_struct *leader, bool threadgroup,
    2683             :                    struct cgroup_mgctx *mgctx)
    2684             : {
    2685         198 :         struct task_struct *task;
    2686             : 
    2687             :         /*
    2688             :          * Prevent freeing of tasks while we take a snapshot. Tasks that are
    2689             :          * already PF_EXITING could be freed from underneath us unless we
    2690             :          * take an rcu_read_lock.
    2691             :          */
    2692         198 :         spin_lock_irq(&css_set_lock);
    2693         198 :         rcu_read_lock();
    2694             :         task = leader;
    2695         198 :         do {
    2696         198 :                 cgroup_migrate_add_task(task, mgctx);
    2697         198 :                 if (!threadgroup)
    2698             :                         break;
    2699         198 :         } while_each_thread(leader, task);
    2700         198 :         rcu_read_unlock();
    2701         198 :         spin_unlock_irq(&css_set_lock);
    2702             : 
    2703         198 :         return cgroup_migrate_execute(mgctx);
    2704             : }
    2705             : 
    2706             : /**
    2707             :  * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup
    2708             :  * @dst_cgrp: the cgroup to attach to
    2709             :  * @leader: the task or the leader of the threadgroup to be attached
    2710             :  * @threadgroup: attach the whole threadgroup?
    2711             :  *
    2712             :  * Call holding cgroup_mutex and cgroup_threadgroup_rwsem.
    2713             :  */
    2714         198 : int cgroup_attach_task(struct cgroup *dst_cgrp, struct task_struct *leader,
    2715             :                        bool threadgroup)
    2716             : {
    2717         198 :         DEFINE_CGROUP_MGCTX(mgctx);
    2718         198 :         struct task_struct *task;
    2719         198 :         int ret = 0;
    2720             : 
    2721             :         /* look up all src csets */
    2722         198 :         spin_lock_irq(&css_set_lock);
    2723         198 :         rcu_read_lock();
    2724             :         task = leader;
    2725         198 :         do {
    2726         198 :                 cgroup_migrate_add_src(task_css_set(task), dst_cgrp, &mgctx);
    2727         198 :                 if (!threadgroup)
    2728             :                         break;
    2729         198 :         } while_each_thread(leader, task);
    2730         198 :         rcu_read_unlock();
    2731         198 :         spin_unlock_irq(&css_set_lock);
    2732             : 
    2733             :         /* prepare dst csets and commit */
    2734         198 :         ret = cgroup_migrate_prepare_dst(&mgctx);
    2735         198 :         if (!ret)
    2736         198 :                 ret = cgroup_migrate(leader, threadgroup, &mgctx);
    2737             : 
    2738         198 :         cgroup_migrate_finish(&mgctx);
    2739             : 
    2740         198 :         if (!ret)
    2741         198 :                 TRACE_CGROUP_PATH(attach_task, dst_cgrp, leader, threadgroup);
    2742             : 
    2743         198 :         return ret;
    2744             : }
    2745             : 
    2746         198 : struct task_struct *cgroup_procs_write_start(char *buf, bool threadgroup,
    2747             :                                              bool *locked)
    2748             :         __acquires(&cgroup_threadgroup_rwsem)
    2749             : {
    2750         198 :         struct task_struct *tsk;
    2751         198 :         pid_t pid;
    2752             : 
    2753         198 :         if (kstrtoint(strstrip(buf), 0, &pid) || pid < 0)
    2754         198 :                 return ERR_PTR(-EINVAL);
    2755             : 
    2756             :         /*
    2757             :          * If we migrate a single thread, we don't care about threadgroup
    2758             :          * stability. If the thread is `current`, it won't exit(2) under our
    2759             :          * hands or change PID through exec(2). We exclude
    2760             :          * cgroup_update_dfl_csses and other cgroup_{proc,thread}s_write
    2761             :          * callers by cgroup_mutex.
    2762             :          * Therefore, we can skip the global lock.
    2763             :          */
    2764         594 :         lockdep_assert_held(&cgroup_mutex);
    2765         198 :         if (pid || threadgroup) {
    2766         198 :                 percpu_down_write(&cgroup_threadgroup_rwsem);
    2767         198 :                 *locked = true;
    2768             :         } else {
    2769           0 :                 *locked = false;
    2770             :         }
    2771             : 
    2772         198 :         rcu_read_lock();
    2773         198 :         if (pid) {
    2774         198 :                 tsk = find_task_by_vpid(pid);
    2775         198 :                 if (!tsk) {
    2776           0 :                         tsk = ERR_PTR(-ESRCH);
    2777           0 :                         goto out_unlock_threadgroup;
    2778             :                 }
    2779             :         } else {
    2780           0 :                 tsk = current;
    2781             :         }
    2782             : 
    2783         198 :         if (threadgroup)
    2784         198 :                 tsk = tsk->group_leader;
    2785             : 
    2786             :         /*
    2787             :          * kthreads may acquire PF_NO_SETAFFINITY during initialization.
    2788             :          * If userland migrates such a kthread to a non-root cgroup, it can
    2789             :          * become trapped in a cpuset, or RT kthread may be born in a
    2790             :          * cgroup with no rt_runtime allocated.  Just say no.
    2791             :          */
    2792         198 :         if (tsk->no_cgroup_migration || (tsk->flags & PF_NO_SETAFFINITY)) {
    2793           0 :                 tsk = ERR_PTR(-EINVAL);
    2794           0 :                 goto out_unlock_threadgroup;
    2795             :         }
    2796             : 
    2797         198 :         get_task_struct(tsk);
    2798         198 :         goto out_unlock_rcu;
    2799             : 
    2800           0 : out_unlock_threadgroup:
    2801           0 :         if (*locked) {
    2802           0 :                 percpu_up_write(&cgroup_threadgroup_rwsem);
    2803           0 :                 *locked = false;
    2804             :         }
    2805           0 : out_unlock_rcu:
    2806         198 :         rcu_read_unlock();
    2807         198 :         return tsk;
    2808             : }
    2809             : 
    2810         198 : void cgroup_procs_write_finish(struct task_struct *task, bool locked)
    2811             :         __releases(&cgroup_threadgroup_rwsem)
    2812             : {
    2813         198 :         struct cgroup_subsys *ss;
    2814         198 :         int ssid;
    2815             : 
    2816             :         /* release reference from cgroup_procs_write_start() */
    2817         198 :         put_task_struct(task);
    2818             : 
    2819         198 :         if (locked)
    2820         198 :                 percpu_up_write(&cgroup_threadgroup_rwsem);
    2821         198 :         for_each_subsys(ss, ssid)
    2822             :                 if (ss->post_attach)
    2823             :                         ss->post_attach();
    2824         198 : }
    2825             : 
    2826           0 : static void cgroup_print_ss_mask(struct seq_file *seq, u16 ss_mask)
    2827             : {
    2828           0 :         struct cgroup_subsys *ss;
    2829           0 :         bool printed = false;
    2830           0 :         int ssid;
    2831             : 
    2832           0 :         do_each_subsys_mask(ss, ssid, ss_mask) {
    2833             :                 if (printed)
    2834             :                         seq_putc(seq, ' ');
    2835             :                 seq_puts(seq, ss->name);
    2836             :                 printed = true;
    2837           0 :         } while_each_subsys_mask();
    2838           0 :         if (printed)
    2839             :                 seq_putc(seq, '\n');
    2840             : }
    2841             : 
    2842             : /* show controllers which are enabled from the parent */
    2843           0 : static int cgroup_controllers_show(struct seq_file *seq, void *v)
    2844             : {
    2845           0 :         struct cgroup *cgrp = seq_css(seq)->cgroup;
    2846             : 
    2847           0 :         cgroup_print_ss_mask(seq, cgroup_control(cgrp));
    2848           0 :         return 0;
    2849             : }
    2850             : 
    2851             : /* show controllers which are enabled for a given cgroup's children */
    2852           0 : static int cgroup_subtree_control_show(struct seq_file *seq, void *v)
    2853             : {
    2854           0 :         struct cgroup *cgrp = seq_css(seq)->cgroup;
    2855             : 
    2856           0 :         cgroup_print_ss_mask(seq, cgrp->subtree_control);
    2857           0 :         return 0;
    2858             : }
    2859             : 
    2860             : /**
    2861             :  * cgroup_update_dfl_csses - update css assoc of a subtree in default hierarchy
    2862             :  * @cgrp: root of the subtree to update csses for
    2863             :  *
    2864             :  * @cgrp's control masks have changed and its subtree's css associations
    2865             :  * need to be updated accordingly.  This function looks up all css_sets
    2866             :  * which are attached to the subtree, creates the matching updated css_sets
    2867             :  * and migrates the tasks to the new ones.
    2868             :  */
    2869           0 : static int cgroup_update_dfl_csses(struct cgroup *cgrp)
    2870             : {
    2871           0 :         DEFINE_CGROUP_MGCTX(mgctx);
    2872           0 :         struct cgroup_subsys_state *d_css;
    2873           0 :         struct cgroup *dsct;
    2874           0 :         struct css_set *src_cset;
    2875           0 :         int ret;
    2876             : 
    2877           0 :         lockdep_assert_held(&cgroup_mutex);
    2878             : 
    2879           0 :         percpu_down_write(&cgroup_threadgroup_rwsem);
    2880             : 
    2881             :         /* look up all csses currently attached to @cgrp's subtree */
    2882           0 :         spin_lock_irq(&css_set_lock);
    2883           0 :         cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) {
    2884           0 :                 struct cgrp_cset_link *link;
    2885             : 
    2886           0 :                 list_for_each_entry(link, &dsct->cset_links, cset_link)
    2887           0 :                         cgroup_migrate_add_src(link->cset, dsct, &mgctx);
    2888             :         }
    2889           0 :         spin_unlock_irq(&css_set_lock);
    2890             : 
    2891             :         /* NULL dst indicates self on default hierarchy */
    2892           0 :         ret = cgroup_migrate_prepare_dst(&mgctx);
    2893           0 :         if (ret)
    2894           0 :                 goto out_finish;
    2895             : 
    2896           0 :         spin_lock_irq(&css_set_lock);
    2897           0 :         list_for_each_entry(src_cset, &mgctx.preloaded_src_csets, mg_preload_node) {
    2898           0 :                 struct task_struct *task, *ntask;
    2899             : 
    2900             :                 /* all tasks in src_csets need to be migrated */
    2901           0 :                 list_for_each_entry_safe(task, ntask, &src_cset->tasks, cg_list)
    2902           0 :                         cgroup_migrate_add_task(task, &mgctx);
    2903             :         }
    2904           0 :         spin_unlock_irq(&css_set_lock);
    2905             : 
    2906           0 :         ret = cgroup_migrate_execute(&mgctx);
    2907           0 : out_finish:
    2908           0 :         cgroup_migrate_finish(&mgctx);
    2909           0 :         percpu_up_write(&cgroup_threadgroup_rwsem);
    2910           0 :         return ret;
    2911             : }
    2912             : 
    2913             : /**
    2914             :  * cgroup_lock_and_drain_offline - lock cgroup_mutex and drain offlined csses
    2915             :  * @cgrp: root of the target subtree
    2916             :  *
    2917             :  * Because css offlining is asynchronous, userland may try to re-enable a
    2918             :  * controller while the previous css is still around.  This function grabs
    2919             :  * cgroup_mutex and drains the previous css instances of @cgrp's subtree.
    2920             :  */
    2921           1 : void cgroup_lock_and_drain_offline(struct cgroup *cgrp)
    2922             :         __acquires(&cgroup_mutex)
    2923             : {
    2924           1 :         struct cgroup *dsct;
    2925           1 :         struct cgroup_subsys_state *d_css;
    2926           1 :         struct cgroup_subsys *ss;
    2927           1 :         int ssid;
    2928             : 
    2929           1 : restart:
    2930           1 :         mutex_lock(&cgroup_mutex);
    2931             : 
    2932           3 :         cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) {
    2933             :                 for_each_subsys(ss, ssid) {
    2934             :                         struct cgroup_subsys_state *css = cgroup_css(dsct, ss);
    2935             :                         DEFINE_WAIT(wait);
    2936             : 
    2937             :                         if (!css || !percpu_ref_is_dying(&css->refcnt))
    2938             :                                 continue;
    2939             : 
    2940             :                         cgroup_get_live(dsct);
    2941             :                         prepare_to_wait(&dsct->offline_waitq, &wait,
    2942             :                                         TASK_UNINTERRUPTIBLE);
    2943             : 
    2944             :                         mutex_unlock(&cgroup_mutex);
    2945             :                         schedule();
    2946             :                         finish_wait(&dsct->offline_waitq, &wait);
    2947             : 
    2948             :                         cgroup_put(dsct);
    2949             :                         goto restart;
    2950             :                 }
    2951             :         }
    2952           1 : }
    2953             : 
    2954             : /**
    2955             :  * cgroup_save_control - save control masks and dom_cgrp of a subtree
    2956             :  * @cgrp: root of the target subtree
    2957             :  *
    2958             :  * Save ->subtree_control, ->subtree_ss_mask and ->dom_cgrp to the
    2959             :  * respective old_ prefixed fields for @cgrp's subtree including @cgrp
    2960             :  * itself.
    2961             :  */
    2962           0 : static void cgroup_save_control(struct cgroup *cgrp)
    2963             : {
    2964           0 :         struct cgroup *dsct;
    2965           0 :         struct cgroup_subsys_state *d_css;
    2966             : 
    2967           0 :         cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) {
    2968           0 :                 dsct->old_subtree_control = dsct->subtree_control;
    2969           0 :                 dsct->old_subtree_ss_mask = dsct->subtree_ss_mask;
    2970           0 :                 dsct->old_dom_cgrp = dsct->dom_cgrp;
    2971             :         }
    2972           0 : }
    2973             : 
    2974             : /**
    2975             :  * cgroup_propagate_control - refresh control masks of a subtree
    2976             :  * @cgrp: root of the target subtree
    2977             :  *
    2978             :  * For @cgrp and its subtree, ensure ->subtree_ss_mask matches
    2979             :  * ->subtree_control and propagate controller availability through the
    2980             :  * subtree so that descendants don't have unavailable controllers enabled.
    2981             :  */
    2982          98 : static void cgroup_propagate_control(struct cgroup *cgrp)
    2983             : {
    2984          98 :         struct cgroup *dsct;
    2985          98 :         struct cgroup_subsys_state *d_css;
    2986             : 
    2987         392 :         cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) {
    2988          98 :                 dsct->subtree_control &= cgroup_control(dsct);
    2989          98 :                 dsct->subtree_ss_mask =
    2990          98 :                         cgroup_calc_subtree_ss_mask(dsct->subtree_control,
    2991          98 :                                                     cgroup_ss_mask(dsct));
    2992             :         }
    2993          98 : }
    2994             : 
    2995             : /**
    2996             :  * cgroup_restore_control - restore control masks and dom_cgrp of a subtree
    2997             :  * @cgrp: root of the target subtree
    2998             :  *
    2999             :  * Restore ->subtree_control, ->subtree_ss_mask and ->dom_cgrp from the
    3000             :  * respective old_ prefixed fields for @cgrp's subtree including @cgrp
    3001             :  * itself.
    3002             :  */
    3003           0 : static void cgroup_restore_control(struct cgroup *cgrp)
    3004             : {
    3005           0 :         struct cgroup *dsct;
    3006           0 :         struct cgroup_subsys_state *d_css;
    3007             : 
    3008           0 :         cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) {
    3009           0 :                 dsct->subtree_control = dsct->old_subtree_control;
    3010           0 :                 dsct->subtree_ss_mask = dsct->old_subtree_ss_mask;
    3011           0 :                 dsct->dom_cgrp = dsct->old_dom_cgrp;
    3012             :         }
    3013           0 : }
    3014             : 
    3015             : static bool css_visible(struct cgroup_subsys_state *css)
    3016             : {
    3017             :         struct cgroup_subsys *ss = css->ss;
    3018             :         struct cgroup *cgrp = css->cgroup;
    3019             : 
    3020             :         if (cgroup_control(cgrp) & (1 << ss->id))
    3021             :                 return true;
    3022             :         if (!(cgroup_ss_mask(cgrp) & (1 << ss->id)))
    3023             :                 return false;
    3024             :         return cgroup_on_dfl(cgrp) && ss->implicit_on_dfl;
    3025             : }
    3026             : 
    3027             : /**
    3028             :  * cgroup_apply_control_enable - enable or show csses according to control
    3029             :  * @cgrp: root of the target subtree
    3030             :  *
    3031             :  * Walk @cgrp's subtree and create new csses or make the existing ones
    3032             :  * visible.  A css is created invisible if it's being implicitly enabled
    3033             :  * through dependency.  An invisible css is made visible when the userland
    3034             :  * explicitly enables it.
    3035             :  *
    3036             :  * Returns 0 on success, -errno on failure.  On failure, csses which have
    3037             :  * been processed already aren't cleaned up.  The caller is responsible for
    3038             :  * cleaning up with cgroup_apply_control_disable().
    3039             :  */
    3040          98 : static int cgroup_apply_control_enable(struct cgroup *cgrp)
    3041             : {
    3042          98 :         struct cgroup *dsct;
    3043          98 :         struct cgroup_subsys_state *d_css;
    3044          98 :         struct cgroup_subsys *ss;
    3045          98 :         int ssid, ret;
    3046             : 
    3047         294 :         cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) {
    3048             :                 for_each_subsys(ss, ssid) {
    3049             :                         struct cgroup_subsys_state *css = cgroup_css(dsct, ss);
    3050             : 
    3051             :                         if (!(cgroup_ss_mask(dsct) & (1 << ss->id)))
    3052             :                                 continue;
    3053             : 
    3054             :                         if (!css) {
    3055             :                                 css = css_create(dsct, ss);
    3056             :                                 if (IS_ERR(css))
    3057             :                                         return PTR_ERR(css);
    3058             :                         }
    3059             : 
    3060             :                         WARN_ON_ONCE(percpu_ref_is_dying(&css->refcnt));
    3061             : 
    3062             :                         if (css_visible(css)) {
    3063             :                                 ret = css_populate_dir(css);
    3064             :                                 if (ret)
    3065             :                                         return ret;
    3066             :                         }
    3067             :                 }
    3068             :         }
    3069             : 
    3070          98 :         return 0;
    3071             : }
    3072             : 
    3073             : /**
    3074             :  * cgroup_apply_control_disable - kill or hide csses according to control
    3075             :  * @cgrp: root of the target subtree
    3076             :  *
    3077             :  * Walk @cgrp's subtree and kill and hide csses so that they match
    3078             :  * cgroup_ss_mask() and cgroup_visible_mask().
    3079             :  *
    3080             :  * A css is hidden when the userland requests it to be disabled while other
    3081             :  * subsystems are still depending on it.  The css must not actively control
    3082             :  * resources and be in the vanilla state if it's made visible again later.
    3083             :  * Controllers which may be depended upon should provide ->css_reset() for
    3084             :  * this purpose.
    3085             :  */
    3086           0 : static void cgroup_apply_control_disable(struct cgroup *cgrp)
    3087             : {
    3088           0 :         struct cgroup *dsct;
    3089           0 :         struct cgroup_subsys_state *d_css;
    3090           0 :         struct cgroup_subsys *ss;
    3091           0 :         int ssid;
    3092             : 
    3093           0 :         cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) {
    3094             :                 for_each_subsys(ss, ssid) {
    3095             :                         struct cgroup_subsys_state *css = cgroup_css(dsct, ss);
    3096             : 
    3097             :                         if (!css)
    3098             :                                 continue;
    3099             : 
    3100             :                         WARN_ON_ONCE(percpu_ref_is_dying(&css->refcnt));
    3101             : 
    3102             :                         if (css->parent &&
    3103             :                             !(cgroup_ss_mask(dsct) & (1 << ss->id))) {
    3104             :                                 kill_css(css);
    3105             :                         } else if (!css_visible(css)) {
    3106             :                                 css_clear_dir(css);
    3107             :                                 if (ss->css_reset)
    3108             :                                         ss->css_reset(css);
    3109             :                         }
    3110             :                 }
    3111             :         }
    3112           0 : }
    3113             : 
    3114             : /**
    3115             :  * cgroup_apply_control - apply control mask updates to the subtree
    3116             :  * @cgrp: root of the target subtree
    3117             :  *
    3118             :  * subsystems can be enabled and disabled in a subtree using the following
    3119             :  * steps.
    3120             :  *
    3121             :  * 1. Call cgroup_save_control() to stash the current state.
    3122             :  * 2. Update ->subtree_control masks in the subtree as desired.
    3123             :  * 3. Call cgroup_apply_control() to apply the changes.
    3124             :  * 4. Optionally perform other related operations.
    3125             :  * 5. Call cgroup_finalize_control() to finish up.
    3126             :  *
    3127             :  * This function implements step 3 and propagates the mask changes
    3128             :  * throughout @cgrp's subtree, updates csses accordingly and perform
    3129             :  * process migrations.
    3130             :  */
    3131           0 : static int cgroup_apply_control(struct cgroup *cgrp)
    3132             : {
    3133           0 :         int ret;
    3134             : 
    3135           0 :         cgroup_propagate_control(cgrp);
    3136             : 
    3137           0 :         ret = cgroup_apply_control_enable(cgrp);
    3138           0 :         if (ret)
    3139             :                 return ret;
    3140             : 
    3141             :         /*
    3142             :          * At this point, cgroup_e_css_by_mask() results reflect the new csses
    3143             :          * making the following cgroup_update_dfl_csses() properly update
    3144             :          * css associations of all tasks in the subtree.
    3145             :          */
    3146           0 :         ret = cgroup_update_dfl_csses(cgrp);
    3147           0 :         if (ret)
    3148           0 :                 return ret;
    3149             : 
    3150             :         return 0;
    3151             : }
    3152             : 
    3153             : /**
    3154             :  * cgroup_finalize_control - finalize control mask update
    3155             :  * @cgrp: root of the target subtree
    3156             :  * @ret: the result of the update
    3157             :  *
    3158             :  * Finalize control mask update.  See cgroup_apply_control() for more info.
    3159             :  */
    3160           0 : static void cgroup_finalize_control(struct cgroup *cgrp, int ret)
    3161             : {
    3162           0 :         if (ret) {
    3163           0 :                 cgroup_restore_control(cgrp);
    3164           0 :                 cgroup_propagate_control(cgrp);
    3165             :         }
    3166             : 
    3167           0 :         cgroup_apply_control_disable(cgrp);
    3168           0 : }
    3169             : 
    3170             : static int cgroup_vet_subtree_control_enable(struct cgroup *cgrp, u16 enable)
    3171             : {
    3172             :         u16 domain_enable = enable & ~cgrp_dfl_threaded_ss_mask;
    3173             : 
    3174             :         /* if nothing is getting enabled, nothing to worry about */
    3175             :         if (!enable)
    3176             :                 return 0;
    3177             : 
    3178             :         /* can @cgrp host any resources? */
    3179             :         if (!cgroup_is_valid_domain(cgrp->dom_cgrp))
    3180             :                 return -EOPNOTSUPP;
    3181             : 
    3182             :         /* mixables don't care */
    3183             :         if (cgroup_is_mixable(cgrp))
    3184             :                 return 0;
    3185             : 
    3186             :         if (domain_enable) {
    3187             :                 /* can't enable domain controllers inside a thread subtree */
    3188             :                 if (cgroup_is_thread_root(cgrp) || cgroup_is_threaded(cgrp))
    3189             :                         return -EOPNOTSUPP;
    3190             :         } else {
    3191             :                 /*
    3192             :                  * Threaded controllers can handle internal competitions
    3193             :                  * and are always allowed inside a (prospective) thread
    3194             :                  * subtree.
    3195             :                  */
    3196             :                 if (cgroup_can_be_thread_root(cgrp) || cgroup_is_threaded(cgrp))
    3197             :                         return 0;
    3198             :         }
    3199             : 
    3200             :         /*
    3201             :          * Controllers can't be enabled for a cgroup with tasks to avoid
    3202             :          * child cgroups competing against tasks.
    3203             :          */
    3204             :         if (cgroup_has_tasks(cgrp))
    3205             :                 return -EBUSY;
    3206             : 
    3207             :         return 0;
    3208             : }
    3209             : 
    3210             : /* change the enabled child controllers for a cgroup in the default hierarchy */
    3211           0 : static ssize_t cgroup_subtree_control_write(struct kernfs_open_file *of,
    3212             :                                             char *buf, size_t nbytes,
    3213             :                                             loff_t off)
    3214             : {
    3215           0 :         u16 enable = 0, disable = 0;
    3216           0 :         struct cgroup *cgrp, *child;
    3217           0 :         struct cgroup_subsys *ss;
    3218           0 :         char *tok;
    3219           0 :         int ssid, ret;
    3220             : 
    3221             :         /*
    3222             :          * Parse input - space separated list of subsystem names prefixed
    3223             :          * with either + or -.
    3224             :          */
    3225           0 :         buf = strstrip(buf);
    3226           0 :         while ((tok = strsep(&buf, " "))) {
    3227           0 :                 if (tok[0] == '\0')
    3228           0 :                         continue;
    3229           0 :                 do_each_subsys_mask(ss, ssid, ~cgrp_dfl_inhibit_ss_mask) {
    3230             :                         if (!cgroup_ssid_enabled(ssid) ||
    3231             :                             strcmp(tok + 1, ss->name))
    3232             :                                 continue;
    3233             : 
    3234             :                         if (*tok == '+') {
    3235             :                                 enable |= 1 << ssid;
    3236             :                                 disable &= ~(1 << ssid);
    3237             :                         } else if (*tok == '-') {
    3238             :                                 disable |= 1 << ssid;
    3239             :                                 enable &= ~(1 << ssid);
    3240             :                         } else {
    3241             :                                 return -EINVAL;
    3242             :                         }
    3243             :                         break;
    3244             :                 } while_each_subsys_mask();
    3245             :                 if (ssid == CGROUP_SUBSYS_COUNT)
    3246             :                         return -EINVAL;
    3247             :         }
    3248             : 
    3249           0 :         cgrp = cgroup_kn_lock_live(of->kn, true);
    3250           0 :         if (!cgrp)
    3251             :                 return -ENODEV;
    3252             : 
    3253           0 :         for_each_subsys(ss, ssid) {
    3254             :                 if (enable & (1 << ssid)) {
    3255             :                         if (cgrp->subtree_control & (1 << ssid)) {
    3256             :                                 enable &= ~(1 << ssid);
    3257             :                                 continue;
    3258             :                         }
    3259             : 
    3260             :                         if (!(cgroup_control(cgrp) & (1 << ssid))) {
    3261             :                                 ret = -ENOENT;
    3262             :                                 goto out_unlock;
    3263             :                         }
    3264             :                 } else if (disable & (1 << ssid)) {
    3265             :                         if (!(cgrp->subtree_control & (1 << ssid))) {
    3266             :                                 disable &= ~(1 << ssid);
    3267             :                                 continue;
    3268             :                         }
    3269             : 
    3270             :                         /* a child has it enabled? */
    3271             :                         cgroup_for_each_live_child(child, cgrp) {
    3272             :                                 if (child->subtree_control & (1 << ssid)) {
    3273             :                                         ret = -EBUSY;
    3274             :                                         goto out_unlock;
    3275             :                                 }
    3276             :                         }
    3277             :                 }
    3278             :         }
    3279             : 
    3280           0 :         if (!enable && !disable) {
    3281           0 :                 ret = 0;
    3282           0 :                 goto out_unlock;
    3283             :         }
    3284             : 
    3285             :         ret = cgroup_vet_subtree_control_enable(cgrp, enable);
    3286             :         if (ret)
    3287             :                 goto out_unlock;
    3288             : 
    3289             :         /* save and update control masks and prepare csses */
    3290             :         cgroup_save_control(cgrp);
    3291             : 
    3292             :         cgrp->subtree_control |= enable;
    3293             :         cgrp->subtree_control &= ~disable;
    3294             : 
    3295             :         ret = cgroup_apply_control(cgrp);
    3296             :         cgroup_finalize_control(cgrp, ret);
    3297             :         if (ret)
    3298             :                 goto out_unlock;
    3299             : 
    3300             :         kernfs_activate(cgrp->kn);
    3301           0 : out_unlock:
    3302           0 :         cgroup_kn_unlock(of->kn);
    3303           0 :         return ret ?: nbytes;
    3304             : }
    3305             : 
    3306             : /**
    3307             :  * cgroup_enable_threaded - make @cgrp threaded
    3308             :  * @cgrp: the target cgroup
    3309             :  *
    3310             :  * Called when "threaded" is written to the cgroup.type interface file and
    3311             :  * tries to make @cgrp threaded and join the parent's resource domain.
    3312             :  * This function is never called on the root cgroup as cgroup.type doesn't
    3313             :  * exist on it.
    3314             :  */
    3315           0 : static int cgroup_enable_threaded(struct cgroup *cgrp)
    3316             : {
    3317           0 :         struct cgroup *parent = cgroup_parent(cgrp);
    3318           0 :         struct cgroup *dom_cgrp = parent->dom_cgrp;
    3319           0 :         struct cgroup *dsct;
    3320           0 :         struct cgroup_subsys_state *d_css;
    3321           0 :         int ret;
    3322             : 
    3323           0 :         lockdep_assert_held(&cgroup_mutex);
    3324             : 
    3325             :         /* noop if already threaded */
    3326           0 :         if (cgroup_is_threaded(cgrp))
    3327             :                 return 0;
    3328             : 
    3329             :         /*
    3330             :          * If @cgroup is populated or has domain controllers enabled, it
    3331             :          * can't be switched.  While the below cgroup_can_be_thread_root()
    3332             :          * test can catch the same conditions, that's only when @parent is
    3333             :          * not mixable, so let's check it explicitly.
    3334             :          */
    3335           0 :         if (cgroup_is_populated(cgrp) ||
    3336           0 :             cgrp->subtree_control & ~cgrp_dfl_threaded_ss_mask)
    3337             :                 return -EOPNOTSUPP;
    3338             : 
    3339             :         /* we're joining the parent's domain, ensure its validity */
    3340           0 :         if (!cgroup_is_valid_domain(dom_cgrp) ||
    3341           0 :             !cgroup_can_be_thread_root(dom_cgrp))
    3342             :                 return -EOPNOTSUPP;
    3343             : 
    3344             :         /*
    3345             :          * The following shouldn't cause actual migrations and should
    3346             :          * always succeed.
    3347             :          */
    3348           0 :         cgroup_save_control(cgrp);
    3349             : 
    3350           0 :         cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp)
    3351           0 :                 if (dsct == cgrp || cgroup_is_threaded(dsct))
    3352           0 :                         dsct->dom_cgrp = dom_cgrp;
    3353             : 
    3354           0 :         ret = cgroup_apply_control(cgrp);
    3355           0 :         if (!ret)
    3356           0 :                 parent->nr_threaded_children++;
    3357             : 
    3358           0 :         cgroup_finalize_control(cgrp, ret);
    3359           0 :         return ret;
    3360             : }
    3361             : 
    3362           0 : static int cgroup_type_show(struct seq_file *seq, void *v)
    3363             : {
    3364           0 :         struct cgroup *cgrp = seq_css(seq)->cgroup;
    3365             : 
    3366           0 :         if (cgroup_is_threaded(cgrp))
    3367           0 :                 seq_puts(seq, "threaded\n");
    3368           0 :         else if (!cgroup_is_valid_domain(cgrp))
    3369           0 :                 seq_puts(seq, "domain invalid\n");
    3370           0 :         else if (cgroup_is_thread_root(cgrp))
    3371           0 :                 seq_puts(seq, "domain threaded\n");
    3372             :         else
    3373           0 :                 seq_puts(seq, "domain\n");
    3374             : 
    3375           0 :         return 0;
    3376             : }
    3377             : 
    3378           0 : static ssize_t cgroup_type_write(struct kernfs_open_file *of, char *buf,
    3379             :                                  size_t nbytes, loff_t off)
    3380             : {
    3381           0 :         struct cgroup *cgrp;
    3382           0 :         int ret;
    3383             : 
    3384             :         /* only switching to threaded mode is supported */
    3385           0 :         if (strcmp(strstrip(buf), "threaded"))
    3386             :                 return -EINVAL;
    3387             : 
    3388             :         /* drain dying csses before we re-apply (threaded) subtree control */
    3389           0 :         cgrp = cgroup_kn_lock_live(of->kn, true);
    3390           0 :         if (!cgrp)
    3391             :                 return -ENOENT;
    3392             : 
    3393             :         /* threaded can only be enabled */
    3394           0 :         ret = cgroup_enable_threaded(cgrp);
    3395             : 
    3396           0 :         cgroup_kn_unlock(of->kn);
    3397           0 :         return ret ?: nbytes;
    3398             : }
    3399             : 
    3400           0 : static int cgroup_max_descendants_show(struct seq_file *seq, void *v)
    3401             : {
    3402           0 :         struct cgroup *cgrp = seq_css(seq)->cgroup;
    3403           0 :         int descendants = READ_ONCE(cgrp->max_descendants);
    3404             : 
    3405           0 :         if (descendants == INT_MAX)
    3406           0 :                 seq_puts(seq, "max\n");
    3407             :         else
    3408           0 :                 seq_printf(seq, "%d\n", descendants);
    3409             : 
    3410           0 :         return 0;
    3411             : }
    3412             : 
    3413           0 : static ssize_t cgroup_max_descendants_write(struct kernfs_open_file *of,
    3414             :                                            char *buf, size_t nbytes, loff_t off)
    3415             : {
    3416           0 :         struct cgroup *cgrp;
    3417           0 :         int descendants;
    3418           0 :         ssize_t ret;
    3419             : 
    3420           0 :         buf = strstrip(buf);
    3421           0 :         if (!strcmp(buf, "max")) {
    3422           0 :                 descendants = INT_MAX;
    3423             :         } else {
    3424           0 :                 ret = kstrtoint(buf, 0, &descendants);
    3425           0 :                 if (ret)
    3426             :                         return ret;
    3427             :         }
    3428             : 
    3429           0 :         if (descendants < 0)
    3430             :                 return -ERANGE;
    3431             : 
    3432           0 :         cgrp = cgroup_kn_lock_live(of->kn, false);
    3433           0 :         if (!cgrp)
    3434             :                 return -ENOENT;
    3435             : 
    3436           0 :         cgrp->max_descendants = descendants;
    3437             : 
    3438           0 :         cgroup_kn_unlock(of->kn);
    3439             : 
    3440           0 :         return nbytes;
    3441             : }
    3442             : 
    3443           0 : static int cgroup_max_depth_show(struct seq_file *seq, void *v)
    3444             : {
    3445           0 :         struct cgroup *cgrp = seq_css(seq)->cgroup;
    3446           0 :         int depth = READ_ONCE(cgrp->max_depth);
    3447             : 
    3448           0 :         if (depth == INT_MAX)
    3449           0 :                 seq_puts(seq, "max\n");
    3450             :         else
    3451           0 :                 seq_printf(seq, "%d\n", depth);
    3452             : 
    3453           0 :         return 0;
    3454             : }
    3455             : 
    3456           0 : static ssize_t cgroup_max_depth_write(struct kernfs_open_file *of,
    3457             :                                       char *buf, size_t nbytes, loff_t off)
    3458             : {
    3459           0 :         struct cgroup *cgrp;
    3460           0 :         ssize_t ret;
    3461           0 :         int depth;
    3462             : 
    3463           0 :         buf = strstrip(buf);
    3464           0 :         if (!strcmp(buf, "max")) {
    3465           0 :                 depth = INT_MAX;
    3466             :         } else {
    3467           0 :                 ret = kstrtoint(buf, 0, &depth);
    3468           0 :                 if (ret)
    3469             :                         return ret;
    3470             :         }
    3471             : 
    3472           0 :         if (depth < 0)
    3473             :                 return -ERANGE;
    3474             : 
    3475           0 :         cgrp = cgroup_kn_lock_live(of->kn, false);
    3476           0 :         if (!cgrp)
    3477             :                 return -ENOENT;
    3478             : 
    3479           0 :         cgrp->max_depth = depth;
    3480             : 
    3481           0 :         cgroup_kn_unlock(of->kn);
    3482             : 
    3483           0 :         return nbytes;
    3484             : }
    3485             : 
    3486          62 : static int cgroup_events_show(struct seq_file *seq, void *v)
    3487             : {
    3488          62 :         struct cgroup *cgrp = seq_css(seq)->cgroup;
    3489             : 
    3490          62 :         seq_printf(seq, "populated %d\n", cgroup_is_populated(cgrp));
    3491          62 :         seq_printf(seq, "frozen %d\n", test_bit(CGRP_FROZEN, &cgrp->flags));
    3492             : 
    3493          62 :         return 0;
    3494             : }
    3495             : 
    3496           0 : static int cgroup_stat_show(struct seq_file *seq, void *v)
    3497             : {
    3498           0 :         struct cgroup *cgroup = seq_css(seq)->cgroup;
    3499             : 
    3500           0 :         seq_printf(seq, "nr_descendants %d\n",
    3501             :                    cgroup->nr_descendants);
    3502           0 :         seq_printf(seq, "nr_dying_descendants %d\n",
    3503             :                    cgroup->nr_dying_descendants);
    3504             : 
    3505           0 :         return 0;
    3506             : }
    3507             : 
    3508             : static int __maybe_unused cgroup_extra_stat_show(struct seq_file *seq,
    3509             :                                                  struct cgroup *cgrp, int ssid)
    3510             : {
    3511             :         struct cgroup_subsys *ss = cgroup_subsys[ssid];
    3512             :         struct cgroup_subsys_state *css;
    3513             :         int ret;
    3514             : 
    3515             :         if (!ss->css_extra_stat_show)
    3516             :                 return 0;
    3517             : 
    3518             :         css = cgroup_tryget_css(cgrp, ss);
    3519             :         if (!css)
    3520             :                 return 0;
    3521             : 
    3522             :         ret = ss->css_extra_stat_show(seq, css);
    3523             :         css_put(css);
    3524             :         return ret;
    3525             : }
    3526             : 
    3527           0 : static int cpu_stat_show(struct seq_file *seq, void *v)
    3528             : {
    3529           0 :         struct cgroup __maybe_unused *cgrp = seq_css(seq)->cgroup;
    3530           0 :         int ret = 0;
    3531             : 
    3532           0 :         cgroup_base_stat_cputime_show(seq);
    3533             : #ifdef CONFIG_CGROUP_SCHED
    3534             :         ret = cgroup_extra_stat_show(seq, cgrp, cpu_cgrp_id);
    3535             : #endif
    3536           0 :         return ret;
    3537             : }
    3538             : 
    3539             : #ifdef CONFIG_PSI
    3540             : static int cgroup_io_pressure_show(struct seq_file *seq, void *v)
    3541             : {
    3542             :         struct cgroup *cgrp = seq_css(seq)->cgroup;
    3543             :         struct psi_group *psi = cgroup_ino(cgrp) == 1 ? &psi_system : &cgrp->psi;
    3544             : 
    3545             :         return psi_show(seq, psi, PSI_IO);
    3546             : }
    3547             : static int cgroup_memory_pressure_show(struct seq_file *seq, void *v)
    3548             : {
    3549             :         struct cgroup *cgrp = seq_css(seq)->cgroup;
    3550             :         struct psi_group *psi = cgroup_ino(cgrp) == 1 ? &psi_system : &cgrp->psi;
    3551             : 
    3552             :         return psi_show(seq, psi, PSI_MEM);
    3553             : }
    3554             : static int cgroup_cpu_pressure_show(struct seq_file *seq, void *v)
    3555             : {
    3556             :         struct cgroup *cgrp = seq_css(seq)->cgroup;
    3557             :         struct psi_group *psi = cgroup_ino(cgrp) == 1 ? &psi_system : &cgrp->psi;
    3558             : 
    3559             :         return psi_show(seq, psi, PSI_CPU);
    3560             : }
    3561             : 
    3562             : static ssize_t cgroup_pressure_write(struct kernfs_open_file *of, char *buf,
    3563             :                                           size_t nbytes, enum psi_res res)
    3564             : {
    3565             :         struct psi_trigger *new;
    3566             :         struct cgroup *cgrp;
    3567             :         struct psi_group *psi;
    3568             : 
    3569             :         cgrp = cgroup_kn_lock_live(of->kn, false);
    3570             :         if (!cgrp)
    3571             :                 return -ENODEV;
    3572             : 
    3573             :         cgroup_get(cgrp);
    3574             :         cgroup_kn_unlock(of->kn);
    3575             : 
    3576             :         psi = cgroup_ino(cgrp) == 1 ? &psi_system : &cgrp->psi;
    3577             :         new = psi_trigger_create(psi, buf, nbytes, res);
    3578             :         if (IS_ERR(new)) {
    3579             :                 cgroup_put(cgrp);
    3580             :                 return PTR_ERR(new);
    3581             :         }
    3582             : 
    3583             :         psi_trigger_replace(&of->priv, new);
    3584             : 
    3585             :         cgroup_put(cgrp);
    3586             : 
    3587             :         return nbytes;
    3588             : }
    3589             : 
    3590             : static ssize_t cgroup_io_pressure_write(struct kernfs_open_file *of,
    3591             :                                           char *buf, size_t nbytes,
    3592             :                                           loff_t off)
    3593             : {
    3594             :         return cgroup_pressure_write(of, buf, nbytes, PSI_IO);
    3595             : }
    3596             : 
    3597             : static ssize_t cgroup_memory_pressure_write(struct kernfs_open_file *of,
    3598             :                                           char *buf, size_t nbytes,
    3599             :                                           loff_t off)
    3600             : {
    3601             :         return cgroup_pressure_write(of, buf, nbytes, PSI_MEM);
    3602             : }
    3603             : 
    3604             : static ssize_t cgroup_cpu_pressure_write(struct kernfs_open_file *of,
    3605             :                                           char *buf, size_t nbytes,
    3606             :                                           loff_t off)
    3607             : {
    3608             :         return cgroup_pressure_write(of, buf, nbytes, PSI_CPU);
    3609             : }
    3610             : 
    3611             : static __poll_t cgroup_pressure_poll(struct kernfs_open_file *of,
    3612             :                                           poll_table *pt)
    3613             : {
    3614             :         return psi_trigger_poll(&of->priv, of->file, pt);
    3615             : }
    3616             : 
    3617             : static void cgroup_pressure_release(struct kernfs_open_file *of)
    3618             : {
    3619             :         psi_trigger_replace(&of->priv, NULL);
    3620             : }
    3621             : #endif /* CONFIG_PSI */
    3622             : 
    3623           0 : static int cgroup_freeze_show(struct seq_file *seq, void *v)
    3624             : {
    3625           0 :         struct cgroup *cgrp = seq_css(seq)->cgroup;
    3626             : 
    3627           0 :         seq_printf(seq, "%d\n", cgrp->freezer.freeze);
    3628             : 
    3629           0 :         return 0;
    3630             : }
    3631             : 
    3632           0 : static ssize_t cgroup_freeze_write(struct kernfs_open_file *of,
    3633             :                                    char *buf, size_t nbytes, loff_t off)
    3634             : {
    3635           0 :         struct cgroup *cgrp;
    3636           0 :         ssize_t ret;
    3637           0 :         int freeze;
    3638             : 
    3639           0 :         ret = kstrtoint(strstrip(buf), 0, &freeze);
    3640           0 :         if (ret)
    3641             :                 return ret;
    3642             : 
    3643           0 :         if (freeze < 0 || freeze > 1)
    3644             :                 return -ERANGE;
    3645             : 
    3646           0 :         cgrp = cgroup_kn_lock_live(of->kn, false);
    3647           0 :         if (!cgrp)
    3648             :                 return -ENOENT;
    3649             : 
    3650           0 :         cgroup_freeze(cgrp, freeze);
    3651             : 
    3652           0 :         cgroup_kn_unlock(of->kn);
    3653             : 
    3654           0 :         return nbytes;
    3655             : }
    3656             : 
    3657         286 : static int cgroup_file_open(struct kernfs_open_file *of)
    3658             : {
    3659         286 :         struct cftype *cft = of_cft(of);
    3660             : 
    3661         286 :         if (cft->open)
    3662           0 :                 return cft->open(of);
    3663             :         return 0;
    3664             : }
    3665             : 
    3666         286 : static void cgroup_file_release(struct kernfs_open_file *of)
    3667             : {
    3668         286 :         struct cftype *cft = of_cft(of);
    3669             : 
    3670         286 :         if (cft->release)
    3671         125 :                 cft->release(of);
    3672         286 : }
    3673             : 
    3674         198 : static ssize_t cgroup_file_write(struct kernfs_open_file *of, char *buf,
    3675             :                                  size_t nbytes, loff_t off)
    3676             : {
    3677         198 :         struct cgroup_namespace *ns = current->nsproxy->cgroup_ns;
    3678         198 :         struct cgroup *cgrp = of->kn->parent->priv;
    3679         198 :         struct cftype *cft = of_cft(of);
    3680         198 :         struct cgroup_subsys_state *css;
    3681         198 :         int ret;
    3682             : 
    3683         198 :         if (!nbytes)
    3684             :                 return 0;
    3685             : 
    3686             :         /*
    3687             :          * If namespaces are delegation boundaries, disallow writes to
    3688             :          * files in an non-init namespace root from inside the namespace
    3689             :          * except for the files explicitly marked delegatable -
    3690             :          * cgroup.procs and cgroup.subtree_control.
    3691             :          */
    3692         198 :         if ((cgrp->root->flags & CGRP_ROOT_NS_DELEGATE) &&
    3693          99 :             !(cft->flags & CFTYPE_NS_DELEGATABLE) &&
    3694           0 :             ns != &init_cgroup_ns && ns->root_cset->dfl_cgrp == cgrp)
    3695             :                 return -EPERM;
    3696             : 
    3697         198 :         if (cft->write)
    3698         198 :                 return cft->write(of, buf, nbytes, off);
    3699             : 
    3700             :         /*
    3701             :          * kernfs guarantees that a file isn't deleted with operations in
    3702             :          * flight, which means that the matching css is and stays alive and
    3703             :          * doesn't need to be pinned.  The RCU locking is not necessary
    3704             :          * either.  It's just for the convenience of using cgroup_css().
    3705             :          */
    3706           0 :         rcu_read_lock();
    3707           0 :         css = cgroup_css(cgrp, cft->ss);
    3708           0 :         rcu_read_unlock();
    3709             : 
    3710           0 :         if (cft->write_u64) {
    3711           0 :                 unsigned long long v;
    3712           0 :                 ret = kstrtoull(buf, 0, &v);
    3713           0 :                 if (!ret)
    3714           0 :                         ret = cft->write_u64(css, cft, v);
    3715           0 :         } else if (cft->write_s64) {
    3716           0 :                 long long v;
    3717           0 :                 ret = kstrtoll(buf, 0, &v);
    3718           0 :                 if (!ret)
    3719           0 :                         ret = cft->write_s64(css, cft, v);
    3720             :         } else {
    3721             :                 ret = -EINVAL;
    3722             :         }
    3723             : 
    3724           0 :         return ret ?: nbytes;
    3725             : }
    3726             : 
    3727           0 : static __poll_t cgroup_file_poll(struct kernfs_open_file *of, poll_table *pt)
    3728             : {
    3729           0 :         struct cftype *cft = of_cft(of);
    3730             : 
    3731           0 :         if (cft->poll)
    3732           0 :                 return cft->poll(of, pt);
    3733             : 
    3734           0 :         return kernfs_generic_poll(of, pt);
    3735             : }
    3736             : 
    3737          31 : static void *cgroup_seqfile_start(struct seq_file *seq, loff_t *ppos)
    3738             : {
    3739          31 :         return seq_cft(seq)->seq_start(seq, ppos);
    3740             : }
    3741             : 
    3742          49 : static void *cgroup_seqfile_next(struct seq_file *seq, void *v, loff_t *ppos)
    3743             : {
    3744          49 :         return seq_cft(seq)->seq_next(seq, v, ppos);
    3745             : }
    3746             : 
    3747          31 : static void cgroup_seqfile_stop(struct seq_file *seq, void *v)
    3748             : {
    3749          31 :         if (seq_cft(seq)->seq_stop)
    3750           0 :                 seq_cft(seq)->seq_stop(seq, v);
    3751          31 : }
    3752             : 
    3753         111 : static int cgroup_seqfile_show(struct seq_file *m, void *arg)
    3754             : {
    3755         111 :         struct cftype *cft = seq_cft(m);
    3756         111 :         struct cgroup_subsys_state *css = seq_css(m);
    3757             : 
    3758         111 :         if (cft->seq_show)
    3759         111 :                 return cft->seq_show(m, arg);
    3760             : 
    3761           0 :         if (cft->read_u64)
    3762           0 :                 seq_printf(m, "%llu\n", cft->read_u64(css, cft));
    3763           0 :         else if (cft->read_s64)
    3764           0 :                 seq_printf(m, "%lld\n", cft->read_s64(css, cft));
    3765             :         else
    3766             :                 return -EINVAL;
    3767             :         return 0;
    3768             : }
    3769             : 
    3770             : static struct kernfs_ops cgroup_kf_single_ops = {
    3771             :         .atomic_write_len       = PAGE_SIZE,
    3772             :         .open                   = cgroup_file_open,
    3773             :         .release                = cgroup_file_release,
    3774             :         .write                  = cgroup_file_write,
    3775             :         .poll                   = cgroup_file_poll,
    3776             :         .seq_show               = cgroup_seqfile_show,
    3777             : };
    3778             : 
    3779             : static struct kernfs_ops cgroup_kf_ops = {
    3780             :         .atomic_write_len       = PAGE_SIZE,
    3781             :         .open                   = cgroup_file_open,
    3782             :         .release                = cgroup_file_release,
    3783             :         .write                  = cgroup_file_write,
    3784             :         .poll                   = cgroup_file_poll,
    3785             :         .seq_start              = cgroup_seqfile_start,
    3786             :         .seq_next               = cgroup_seqfile_next,
    3787             :         .seq_stop               = cgroup_seqfile_stop,
    3788             :         .seq_show               = cgroup_seqfile_show,
    3789             : };
    3790             : 
    3791             : /* set uid and gid of cgroup dirs and files to that of the creator */
    3792         847 : static int cgroup_kn_set_ugid(struct kernfs_node *kn)
    3793             : {
    3794        2541 :         struct iattr iattr = { .ia_valid = ATTR_UID | ATTR_GID,
    3795         847 :                                .ia_uid = current_fsuid(),
    3796         847 :                                .ia_gid = current_fsgid(), };
    3797             : 
    3798         847 :         if (uid_eq(iattr.ia_uid, GLOBAL_ROOT_UID) &&
    3799         847 :             gid_eq(iattr.ia_gid, GLOBAL_ROOT_GID))
    3800             :                 return 0;
    3801             : 
    3802           0 :         return kernfs_setattr(kn, &iattr);
    3803             : }
    3804             : 
    3805           0 : static void cgroup_file_notify_timer(struct timer_list *timer)
    3806             : {
    3807           0 :         cgroup_file_notify(container_of(timer, struct cgroup_file,
    3808             :                                         notify_timer));
    3809           0 : }
    3810             : 
    3811         749 : static int cgroup_add_file(struct cgroup_subsys_state *css, struct cgroup *cgrp,
    3812             :                            struct cftype *cft)
    3813             : {
    3814         749 :         char name[CGROUP_FILE_NAME_MAX];
    3815         749 :         struct kernfs_node *kn;
    3816         749 :         struct lock_class_key *key = NULL;
    3817         749 :         int ret;
    3818             : 
    3819             : #ifdef CONFIG_DEBUG_LOCK_ALLOC
    3820         749 :         key = &cft->lockdep_key;
    3821             : #endif
    3822         749 :         kn = __kernfs_create_file(cgrp->kn, cgroup_file_name(cgrp, cft, name),
    3823         749 :                                   cgroup_file_mode(cft),
    3824         749 :                                   GLOBAL_ROOT_UID, GLOBAL_ROOT_GID,
    3825         749 :                                   0, cft->kf_ops, cft,
    3826             :                                   NULL, key);
    3827         749 :         if (IS_ERR(kn))
    3828           0 :                 return PTR_ERR(kn);
    3829             : 
    3830         749 :         ret = cgroup_kn_set_ugid(kn);
    3831         749 :         if (ret) {
    3832           0 :                 kernfs_remove(kn);
    3833           0 :                 return ret;
    3834             :         }
    3835             : 
    3836         749 :         if (cft->file_offset) {
    3837          99 :                 struct cgroup_file *cfile = (void *)css + cft->file_offset;
    3838             : 
    3839          99 :                 timer_setup(&cfile->notify_timer, cgroup_file_notify_timer, 0);
    3840             : 
    3841          99 :                 spin_lock_irq(&cgroup_file_kn_lock);
    3842          99 :                 cfile->kn = kn;
    3843          99 :                 spin_unlock_irq(&cgroup_file_kn_lock);
    3844             :         }
    3845             : 
    3846             :         return 0;
    3847             : }
    3848             : 
    3849             : /**
    3850             :  * cgroup_addrm_files - add or remove files to a cgroup directory
    3851             :  * @css: the target css
    3852             :  * @cgrp: the target cgroup (usually css->cgroup)
    3853             :  * @cfts: array of cftypes to be added
    3854             :  * @is_add: whether to add or remove
    3855             :  *
    3856             :  * Depending on @is_add, add or remove files defined by @cfts on @cgrp.
    3857             :  * For removals, this function never fails.
    3858             :  */
    3859         150 : static int cgroup_addrm_files(struct cgroup_subsys_state *css,
    3860             :                               struct cgroup *cgrp, struct cftype cfts[],
    3861             :                               bool is_add)
    3862             : {
    3863         150 :         struct cftype *cft, *cft_end = NULL;
    3864         150 :         int ret = 0;
    3865             : 
    3866         450 :         lockdep_assert_held(&cgroup_mutex);
    3867             : 
    3868             : restart:
    3869        1425 :         for (cft = cfts; cft != cft_end && cft->name[0] != '\0'; cft++) {
    3870             :                 /* does cft->flags tell us to skip this file on @cgrp? */
    3871        1275 :                 if ((cft->flags & __CFTYPE_ONLY_ON_DFL) && !cgroup_on_dfl(cgrp))
    3872           0 :                         continue;
    3873        1275 :                 if ((cft->flags & __CFTYPE_NOT_ON_DFL) && cgroup_on_dfl(cgrp))
    3874           0 :                         continue;
    3875        1275 :                 if ((cft->flags & CFTYPE_NOT_ON_ROOT) && !cgroup_parent(cgrp))
    3876           3 :                         continue;
    3877        1272 :                 if ((cft->flags & CFTYPE_ONLY_ON_ROOT) && cgroup_parent(cgrp))
    3878         148 :                         continue;
    3879        1124 :                 if ((cft->flags & CFTYPE_DEBUG) && !cgroup_debug)
    3880           0 :                         continue;
    3881        1124 :                 if (is_add) {
    3882         749 :                         ret = cgroup_add_file(css, cgrp, cft);
    3883         749 :                         if (ret) {
    3884           0 :                                 pr_warn("%s: failed to add %s, err=%d\n",
    3885             :                                         __func__, cft->name, ret);
    3886           0 :                                 cft_end = cft;
    3887           0 :                                 is_add = false;
    3888           0 :                                 goto restart;
    3889             :                         }
    3890             :                 } else {
    3891         375 :                         cgroup_rm_file(cgrp, cft);
    3892             :                 }
    3893             :         }
    3894         150 :         return ret;
    3895             : }
    3896             : 
    3897           0 : static int cgroup_apply_cftypes(struct cftype *cfts, bool is_add)
    3898             : {
    3899           0 :         struct cgroup_subsys *ss = cfts[0].ss;
    3900           0 :         struct cgroup *root = &ss->root->cgrp;
    3901           0 :         struct cgroup_subsys_state *css;
    3902           0 :         int ret = 0;
    3903             : 
    3904           0 :         lockdep_assert_held(&cgroup_mutex);
    3905             : 
    3906             :         /* add/rm files for all cgroups created before */
    3907           0 :         css_for_each_descendant_pre(css, cgroup_css(root, ss)) {
    3908           0 :                 struct cgroup *cgrp = css->cgroup;
    3909             : 
    3910           0 :                 if (!(css->flags & CSS_VISIBLE))
    3911           0 :                         continue;
    3912             : 
    3913           0 :                 ret = cgroup_addrm_files(css, cgrp, cfts, is_add);
    3914           0 :                 if (ret)
    3915             :                         break;
    3916             :         }
    3917             : 
    3918           0 :         if (is_add && !ret)
    3919           0 :                 kernfs_activate(root->kn);
    3920           0 :         return ret;
    3921             : }
    3922             : 
    3923           0 : static void cgroup_exit_cftypes(struct cftype *cfts)
    3924             : {
    3925           0 :         struct cftype *cft;
    3926             : 
    3927           0 :         for (cft = cfts; cft->name[0] != '\0'; cft++) {
    3928             :                 /* free copy for custom atomic_write_len, see init_cftypes() */
    3929           0 :                 if (cft->max_write_len && cft->max_write_len != PAGE_SIZE)
    3930           0 :                         kfree(cft->kf_ops);
    3931           0 :                 cft->kf_ops = NULL;
    3932           0 :                 cft->ss = NULL;
    3933             : 
    3934             :                 /* revert flags set by cgroup core while adding @cfts */
    3935           0 :                 cft->flags &= ~(__CFTYPE_ONLY_ON_DFL | __CFTYPE_NOT_ON_DFL);
    3936             :         }
    3937           0 : }
    3938             : 
    3939           2 : static int cgroup_init_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
    3940             : {
    3941           2 :         struct cftype *cft;
    3942             : 
    3943          19 :         for (cft = cfts; cft->name[0] != '\0'; cft++) {
    3944          17 :                 struct kernfs_ops *kf_ops;
    3945             : 
    3946          34 :                 WARN_ON(cft->ss || cft->kf_ops);
    3947             : 
    3948          17 :                 if (cft->seq_start)
    3949             :                         kf_ops = &cgroup_kf_ops;
    3950             :                 else
    3951          13 :                         kf_ops = &cgroup_kf_single_ops;
    3952             : 
    3953             :                 /*
    3954             :                  * Ugh... if @cft wants a custom max_write_len, we need to
    3955             :                  * make a copy of kf_ops to set its atomic_write_len.
    3956             :                  */
    3957          17 :                 if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) {
    3958           1 :                         kf_ops = kmemdup(kf_ops, sizeof(*kf_ops), GFP_KERNEL);
    3959           1 :                         if (!kf_ops) {
    3960           0 :                                 cgroup_exit_cftypes(cfts);
    3961           0 :                                 return -ENOMEM;
    3962             :                         }
    3963           1 :                         kf_ops->atomic_write_len = cft->max_write_len;
    3964             :                 }
    3965             : 
    3966          17 :                 cft->kf_ops = kf_ops;
    3967          17 :                 cft->ss = ss;
    3968             :         }
    3969             : 
    3970             :         return 0;
    3971             : }
    3972             : 
    3973           0 : static int cgroup_rm_cftypes_locked(struct cftype *cfts)
    3974             : {
    3975           0 :         lockdep_assert_held(&cgroup_mutex);
    3976             : 
    3977           0 :         if (!cfts || !cfts[0].ss)
    3978             :                 return -ENOENT;
    3979             : 
    3980           0 :         list_del(&cfts->node);
    3981           0 :         cgroup_apply_cftypes(cfts, false);
    3982           0 :         cgroup_exit_cftypes(cfts);
    3983           0 :         return 0;
    3984             : }
    3985             : 
    3986             : /**
    3987             :  * cgroup_rm_cftypes - remove an array of cftypes from a subsystem
    3988             :  * @cfts: zero-length name terminated array of cftypes
    3989             :  *
    3990             :  * Unregister @cfts.  Files described by @cfts are removed from all
    3991             :  * existing cgroups and all future cgroups won't have them either.  This
    3992             :  * function can be called anytime whether @cfts' subsys is attached or not.
    3993             :  *
    3994             :  * Returns 0 on successful unregistration, -ENOENT if @cfts is not
    3995             :  * registered.
    3996             :  */
    3997           0 : int cgroup_rm_cftypes(struct cftype *cfts)
    3998             : {
    3999           0 :         int ret;
    4000             : 
    4001           0 :         mutex_lock(&cgroup_mutex);
    4002           0 :         ret = cgroup_rm_cftypes_locked(cfts);
    4003           0 :         mutex_unlock(&cgroup_mutex);
    4004           0 :         return ret;
    4005             : }
    4006             : 
    4007             : /**
    4008             :  * cgroup_add_cftypes - add an array of cftypes to a subsystem
    4009             :  * @ss: target cgroup subsystem
    4010             :  * @cfts: zero-length name terminated array of cftypes
    4011             :  *
    4012             :  * Register @cfts to @ss.  Files described by @cfts are created for all
    4013             :  * existing cgroups to which @ss is attached and all future cgroups will
    4014             :  * have them too.  This function can be called anytime whether @ss is
    4015             :  * attached or not.
    4016             :  *
    4017             :  * Returns 0 on successful registration, -errno on failure.  Note that this
    4018             :  * function currently returns 0 as long as @cfts registration is successful
    4019             :  * even if some file creation attempts on existing cgroups fail.
    4020             :  */
    4021           0 : static int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
    4022             : {
    4023           0 :         int ret;
    4024             : 
    4025           0 :         if (!cgroup_ssid_enabled(ss->id))
    4026           0 :                 return 0;
    4027             : 
    4028             :         if (!cfts || cfts[0].name[0] == '\0')
    4029             :                 return 0;
    4030             : 
    4031             :         ret = cgroup_init_cftypes(ss, cfts);
    4032             :         if (ret)
    4033             :                 return ret;
    4034             : 
    4035             :         mutex_lock(&cgroup_mutex);
    4036             : 
    4037             :         list_add_tail(&cfts->node, &ss->cfts);
    4038             :         ret = cgroup_apply_cftypes(cfts, true);
    4039             :         if (ret)
    4040             :                 cgroup_rm_cftypes_locked(cfts);
    4041             : 
    4042             :         mutex_unlock(&cgroup_mutex);
    4043             :         return ret;
    4044             : }
    4045             : 
    4046             : /**
    4047             :  * cgroup_add_dfl_cftypes - add an array of cftypes for default hierarchy
    4048             :  * @ss: target cgroup subsystem
    4049             :  * @cfts: zero-length name terminated array of cftypes
    4050             :  *
    4051             :  * Similar to cgroup_add_cftypes() but the added files are only used for
    4052             :  * the default hierarchy.
    4053             :  */
    4054           0 : int cgroup_add_dfl_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
    4055             : {
    4056           0 :         struct cftype *cft;
    4057             : 
    4058           0 :         for (cft = cfts; cft && cft->name[0] != '\0'; cft++)
    4059           0 :                 cft->flags |= __CFTYPE_ONLY_ON_DFL;
    4060           0 :         return cgroup_add_cftypes(ss, cfts);
    4061             : }
    4062             : 
    4063             : /**
    4064             :  * cgroup_add_legacy_cftypes - add an array of cftypes for legacy hierarchies
    4065             :  * @ss: target cgroup subsystem
    4066             :  * @cfts: zero-length name terminated array of cftypes
    4067             :  *
    4068             :  * Similar to cgroup_add_cftypes() but the added files are only used for
    4069             :  * the legacy hierarchies.
    4070             :  */
    4071           0 : int cgroup_add_legacy_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
    4072             : {
    4073           0 :         struct cftype *cft;
    4074             : 
    4075           0 :         for (cft = cfts; cft && cft->name[0] != '\0'; cft++)
    4076           0 :                 cft->flags |= __CFTYPE_NOT_ON_DFL;
    4077           0 :         return cgroup_add_cftypes(ss, cfts);
    4078             : }
    4079             : 
    4080             : /**
    4081             :  * cgroup_file_notify - generate a file modified event for a cgroup_file
    4082             :  * @cfile: target cgroup_file
    4083             :  *
    4084             :  * @cfile must have been obtained by setting cftype->file_offset.
    4085             :  */
    4086         190 : void cgroup_file_notify(struct cgroup_file *cfile)
    4087             : {
    4088         190 :         unsigned long flags;
    4089             : 
    4090         190 :         spin_lock_irqsave(&cgroup_file_kn_lock, flags);
    4091         190 :         if (cfile->kn) {
    4092          94 :                 unsigned long last = cfile->notified_at;
    4093          94 :                 unsigned long next = last + CGROUP_FILE_NOTIFY_MIN_INTV;
    4094             : 
    4095          94 :                 if (time_in_range(jiffies, last, next)) {
    4096           0 :                         timer_reduce(&cfile->notify_timer, next);
    4097             :                 } else {
    4098          94 :                         kernfs_notify(cfile->kn);
    4099          94 :                         cfile->notified_at = jiffies;
    4100             :                 }
    4101             :         }
    4102         190 :         spin_unlock_irqrestore(&cgroup_file_kn_lock, flags);
    4103         190 : }
    4104             : 
    4105             : /**
    4106             :  * css_next_child - find the next child of a given css
    4107             :  * @pos: the current position (%NULL to initiate traversal)
    4108             :  * @parent: css whose children to walk
    4109             :  *
    4110             :  * This function returns the next child of @parent and should be called
    4111             :  * under either cgroup_mutex or RCU read lock.  The only requirement is
    4112             :  * that @parent and @pos are accessible.  The next sibling is guaranteed to
    4113             :  * be returned regardless of their states.
    4114             :  *
    4115             :  * If a subsystem synchronizes ->css_online() and the start of iteration, a
    4116             :  * css which finished ->css_online() is guaranteed to be visible in the
    4117             :  * future iterations and will stay visible until the last reference is put.
    4118             :  * A css which hasn't finished ->css_online() or already finished
    4119             :  * ->css_offline() may show up during traversal.  It's each subsystem's
    4120             :  * responsibility to synchronize against on/offlining.
    4121             :  */
    4122         247 : struct cgroup_subsys_state *css_next_child(struct cgroup_subsys_state *pos,
    4123             :                                            struct cgroup_subsys_state *parent)
    4124             : {
    4125         247 :         struct cgroup_subsys_state *next;
    4126             : 
    4127         444 :         cgroup_assert_mutex_or_rcu_locked();
    4128             : 
    4129             :         /*
    4130             :          * @pos could already have been unlinked from the sibling list.
    4131             :          * Once a cgroup is removed, its ->sibling.next is no longer
    4132             :          * updated when its next sibling changes.  CSS_RELEASED is set when
    4133             :          * @pos is taken off list, at which time its next pointer is valid,
    4134             :          * and, as releases are serialized, the one pointed to by the next
    4135             :          * pointer is guaranteed to not have started release yet.  This
    4136             :          * implies that if we observe !CSS_RELEASED on @pos in this RCU
    4137             :          * critical section, the one pointed to by its next pointer is
    4138             :          * guaranteed to not have finished its RCU grace period even if we
    4139             :          * have dropped rcu_read_lock() in-between iterations.
    4140             :          *
    4141             :          * If @pos has CSS_RELEASED set, its next pointer can't be
    4142             :          * dereferenced; however, as each css is given a monotonically
    4143             :          * increasing unique serial number and always appended to the
    4144             :          * sibling list, the next one can be found by walking the parent's
    4145             :          * children until the first css with higher serial number than
    4146             :          * @pos's.  While this path can be slower, it happens iff iteration
    4147             :          * races against release and the race window is very small.
    4148             :          */
    4149         247 :         if (!pos) {
    4150         247 :                 next = list_entry_rcu(parent->children.next, struct cgroup_subsys_state, sibling);
    4151           0 :         } else if (likely(!(pos->flags & CSS_RELEASED))) {
    4152           0 :                 next = list_entry_rcu(pos->sibling.next, struct cgroup_subsys_state, sibling);
    4153             :         } else {
    4154           0 :                 list_for_each_entry_rcu(next, &parent->children, sibling,
    4155             :                                         lockdep_is_held(&cgroup_mutex))
    4156           0 :                         if (next->serial_nr > pos->serial_nr)
    4157             :                                 break;
    4158             :         }
    4159             : 
    4160             :         /*
    4161             :          * @next, if not pointing to the head, can be dereferenced and is
    4162             :          * the next sibling.
    4163             :          */
    4164         247 :         if (&next->sibling != &parent->children)
    4165           0 :                 return next;
    4166             :         return NULL;
    4167             : }
    4168             : 
    4169             : /**
    4170             :  * css_next_descendant_pre - find the next descendant for pre-order walk
    4171             :  * @pos: the current position (%NULL to initiate traversal)
    4172             :  * @root: css whose descendants to walk
    4173             :  *
    4174             :  * To be used by css_for_each_descendant_pre().  Find the next descendant
    4175             :  * to visit for pre-order traversal of @root's descendants.  @root is
    4176             :  * included in the iteration and the first node to be visited.
    4177             :  *
    4178             :  * While this function requires cgroup_mutex or RCU read locking, it
    4179             :  * doesn't require the whole traversal to be contained in a single critical
    4180             :  * section.  This function will return the correct next descendant as long
    4181             :  * as both @pos and @root are accessible and @pos is a descendant of @root.
    4182             :  *
    4183             :  * If a subsystem synchronizes ->css_online() and the start of iteration, a
    4184             :  * css which finished ->css_online() is guaranteed to be visible in the
    4185             :  * future iterations and will stay visible until the last reference is put.
    4186             :  * A css which hasn't finished ->css_online() or already finished
    4187             :  * ->css_offline() may show up during traversal.  It's each subsystem's
    4188             :  * responsibility to synchronize against on/offlining.
    4189             :  */
    4190             : struct cgroup_subsys_state *
    4191         392 : css_next_descendant_pre(struct cgroup_subsys_state *pos,
    4192             :                         struct cgroup_subsys_state *root)
    4193             : {
    4194         392 :         struct cgroup_subsys_state *next;
    4195             : 
    4196         784 :         cgroup_assert_mutex_or_rcu_locked();
    4197             : 
    4198             :         /* if first iteration, visit @root */
    4199         392 :         if (!pos)
    4200             :                 return root;
    4201             : 
    4202             :         /* visit the first child if exists */
    4203         196 :         next = css_next_child(NULL, pos);
    4204         196 :         if (next)
    4205             :                 return next;
    4206             : 
    4207             :         /* no child, visit my or the closest ancestor's next sibling */
    4208         196 :         while (pos != root) {
    4209           0 :                 next = css_next_child(pos, pos->parent);
    4210           0 :                 if (next)
    4211           0 :                         return next;
    4212           0 :                 pos = pos->parent;
    4213             :         }
    4214             : 
    4215             :         return NULL;
    4216             : }
    4217             : EXPORT_SYMBOL_GPL(css_next_descendant_pre);
    4218             : 
    4219             : /**
    4220             :  * css_rightmost_descendant - return the rightmost descendant of a css
    4221             :  * @pos: css of interest
    4222             :  *
    4223             :  * Return the rightmost descendant of @pos.  If there's no descendant, @pos
    4224             :  * is returned.  This can be used during pre-order traversal to skip
    4225             :  * subtree of @pos.
    4226             :  *
    4227             :  * While this function requires cgroup_mutex or RCU read locking, it
    4228             :  * doesn't require the whole traversal to be contained in a single critical
    4229             :  * section.  This function will return the correct rightmost descendant as
    4230             :  * long as @pos is accessible.
    4231             :  */
    4232             : struct cgroup_subsys_state *
    4233           0 : css_rightmost_descendant(struct cgroup_subsys_state *pos)
    4234             : {
    4235           0 :         struct cgroup_subsys_state *last, *tmp;
    4236             : 
    4237           0 :         cgroup_assert_mutex_or_rcu_locked();
    4238             : 
    4239           0 :         do {
    4240           0 :                 last = pos;
    4241             :                 /* ->prev isn't RCU safe, walk ->next till the end */
    4242           0 :                 pos = NULL;
    4243           0 :                 css_for_each_child(tmp, last)
    4244           0 :                         pos = tmp;
    4245           0 :         } while (pos);
    4246             : 
    4247           0 :         return last;
    4248             : }
    4249             : 
    4250             : static struct cgroup_subsys_state *
    4251             : css_leftmost_descendant(struct cgroup_subsys_state *pos)
    4252             : {
    4253           1 :         struct cgroup_subsys_state *last;
    4254             : 
    4255           1 :         do {
    4256           1 :                 last = pos;
    4257           1 :                 pos = css_next_child(NULL, pos);
    4258           1 :         } while (pos);
    4259             : 
    4260           1 :         return last;
    4261             : }
    4262             : 
    4263             : /**
    4264             :  * css_next_descendant_post - find the next descendant for post-order walk
    4265             :  * @pos: the current position (%NULL to initiate traversal)
    4266             :  * @root: css whose descendants to walk
    4267             :  *
    4268             :  * To be used by css_for_each_descendant_post().  Find the next descendant
    4269             :  * to visit for post-order traversal of @root's descendants.  @root is
    4270             :  * included in the iteration and the last node to be visited.
    4271             :  *
    4272             :  * While this function requires cgroup_mutex or RCU read locking, it
    4273             :  * doesn't require the whole traversal to be contained in a single critical
    4274             :  * section.  This function will return the correct next descendant as long
    4275             :  * as both @pos and @cgroup are accessible and @pos is a descendant of
    4276             :  * @cgroup.
    4277             :  *
    4278             :  * If a subsystem synchronizes ->css_online() and the start of iteration, a
    4279             :  * css which finished ->css_online() is guaranteed to be visible in the
    4280             :  * future iterations and will stay visible until the last reference is put.
    4281             :  * A css which hasn't finished ->css_online() or already finished
    4282             :  * ->css_offline() may show up during traversal.  It's each subsystem's
    4283             :  * responsibility to synchronize against on/offlining.
    4284             :  */
    4285             : struct cgroup_subsys_state *
    4286           2 : css_next_descendant_post(struct cgroup_subsys_state *pos,
    4287             :                          struct cgroup_subsys_state *root)
    4288             : {
    4289           2 :         struct cgroup_subsys_state *next;
    4290             : 
    4291           4 :         cgroup_assert_mutex_or_rcu_locked();
    4292             : 
    4293             :         /* if first iteration, visit leftmost descendant which may be @root */
    4294           2 :         if (!pos)
    4295           1 :                 return css_leftmost_descendant(root);
    4296             : 
    4297             :         /* if we visited @root, we're done */
    4298           1 :         if (pos == root)
    4299             :                 return NULL;
    4300             : 
    4301             :         /* if there's an unvisited sibling, visit its leftmost descendant */
    4302           0 :         next = css_next_child(pos, pos->parent);
    4303           0 :         if (next)
    4304           0 :                 return css_leftmost_descendant(next);
    4305             : 
    4306             :         /* no sibling left, visit parent */
    4307           0 :         return pos->parent;
    4308             : }
    4309             : 
    4310             : /**
    4311             :  * css_has_online_children - does a css have online children
    4312             :  * @css: the target css
    4313             :  *
    4314             :  * Returns %true if @css has any online children; otherwise, %false.  This
    4315             :  * function can be called from any context but the caller is responsible
    4316             :  * for synchronizing against on/offlining as necessary.
    4317             :  */
    4318          50 : bool css_has_online_children(struct cgroup_subsys_state *css)
    4319             : {
    4320          50 :         struct cgroup_subsys_state *child;
    4321          50 :         bool ret = false;
    4322             : 
    4323          50 :         rcu_read_lock();
    4324          50 :         css_for_each_child(child, css) {
    4325           0 :                 if (child->flags & CSS_ONLINE) {
    4326             :                         ret = true;
    4327             :                         break;
    4328             :                 }
    4329             :         }
    4330          50 :         rcu_read_unlock();
    4331          50 :         return ret;
    4332             : }
    4333             : 
    4334          51 : static struct css_set *css_task_iter_next_css_set(struct css_task_iter *it)
    4335             : {
    4336          51 :         struct list_head *l;
    4337          51 :         struct cgrp_cset_link *link;
    4338          51 :         struct css_set *cset;
    4339             : 
    4340         153 :         lockdep_assert_held(&css_set_lock);
    4341             : 
    4342             :         /* find the next threaded cset */
    4343          51 :         if (it->tcset_pos) {
    4344          25 :                 l = it->tcset_pos->next;
    4345             : 
    4346          25 :                 if (l != it->tcset_head) {
    4347           0 :                         it->tcset_pos = l;
    4348           0 :                         return container_of(l, struct css_set,
    4349             :                                             threaded_csets_node);
    4350             :                 }
    4351             : 
    4352          25 :                 it->tcset_pos = NULL;
    4353             :         }
    4354             : 
    4355             :         /* find the next cset */
    4356          51 :         l = it->cset_pos;
    4357          51 :         l = l->next;
    4358          51 :         if (l == it->cset_head) {
    4359          26 :                 it->cset_pos = NULL;
    4360          26 :                 return NULL;
    4361             :         }
    4362             : 
    4363          25 :         if (it->ss) {
    4364           0 :                 cset = container_of(l, struct css_set, e_cset_node[it->ss->id]);
    4365             :         } else {
    4366          25 :                 link = list_entry(l, struct cgrp_cset_link, cset_link);
    4367          25 :                 cset = link->cset;
    4368             :         }
    4369             : 
    4370          25 :         it->cset_pos = l;
    4371             : 
    4372             :         /* initialize threaded css_set walking */
    4373          25 :         if (it->flags & CSS_TASK_ITER_THREADED) {
    4374          25 :                 if (it->cur_dcset)
    4375           0 :                         put_css_set_locked(it->cur_dcset);
    4376          25 :                 it->cur_dcset = cset;
    4377          25 :                 get_css_set(cset);
    4378             : 
    4379          25 :                 it->tcset_head = &cset->threaded_csets;
    4380          25 :                 it->tcset_pos = &cset->threaded_csets;
    4381             :         }
    4382             : 
    4383             :         return cset;
    4384             : }
    4385             : 
    4386             : /**
    4387             :  * css_task_iter_advance_css_set - advance a task iterator to the next css_set
    4388             :  * @it: the iterator to advance
    4389             :  *
    4390             :  * Advance @it to the next css_set to walk.
    4391             :  */
    4392          51 : static void css_task_iter_advance_css_set(struct css_task_iter *it)
    4393             : {
    4394          51 :         struct css_set *cset;
    4395             : 
    4396         153 :         lockdep_assert_held(&css_set_lock);
    4397             : 
    4398             :         /* Advance to the next non-empty css_set and find first non-empty tasks list*/
    4399          51 :         while ((cset = css_task_iter_next_css_set(it))) {
    4400          25 :                 if (!list_empty(&cset->tasks)) {
    4401           5 :                         it->cur_tasks_head = &cset->tasks;
    4402           5 :                         break;
    4403          20 :                 } else if (!list_empty(&cset->mg_tasks)) {
    4404           0 :                         it->cur_tasks_head = &cset->mg_tasks;
    4405           0 :                         break;
    4406          20 :                 } else if (!list_empty(&cset->dying_tasks)) {
    4407          20 :                         it->cur_tasks_head = &cset->dying_tasks;
    4408          20 :                         break;
    4409             :                 }
    4410             :         }
    4411          51 :         if (!cset) {
    4412          26 :                 it->task_pos = NULL;
    4413          26 :                 return;
    4414             :         }
    4415          25 :         it->task_pos = it->cur_tasks_head->next;
    4416             : 
    4417             :         /*
    4418             :          * We don't keep css_sets locked across iteration steps and thus
    4419             :          * need to take steps to ensure that iteration can be resumed after
    4420             :          * the lock is re-acquired.  Iteration is performed at two levels -
    4421             :          * css_sets and tasks in them.
    4422             :          *
    4423             :          * Once created, a css_set never leaves its cgroup lists, so a
    4424             :          * pinned css_set is guaranteed to stay put and we can resume
    4425             :          * iteration afterwards.
    4426             :          *
    4427             :          * Tasks may leave @cset across iteration steps.  This is resolved
    4428             :          * by registering each iterator with the css_set currently being
    4429             :          * walked and making css_set_move_task() advance iterators whose
    4430             :          * next task is leaving.
    4431             :          */
    4432          25 :         if (it->cur_cset) {
    4433           0 :                 list_del(&it->iters_node);
    4434           0 :                 put_css_set_locked(it->cur_cset);
    4435             :         }
    4436          25 :         get_css_set(cset);
    4437          25 :         it->cur_cset = cset;
    4438          25 :         list_add(&it->iters_node, &cset->task_iters);
    4439             : }
    4440             : 
    4441           1 : static void css_task_iter_skip(struct css_task_iter *it,
    4442             :                                struct task_struct *task)
    4443             : {
    4444           3 :         lockdep_assert_held(&css_set_lock);
    4445             : 
    4446           1 :         if (it->task_pos == &task->cg_list) {
    4447           0 :                 it->task_pos = it->task_pos->next;
    4448           0 :                 it->flags |= CSS_TASK_ITER_SKIPPED;
    4449             :         }
    4450           1 : }
    4451             : 
    4452          75 : static void css_task_iter_advance(struct css_task_iter *it)
    4453             : {
    4454          75 :         struct task_struct *task;
    4455             : 
    4456         225 :         lockdep_assert_held(&css_set_lock);
    4457             : repeat:
    4458         101 :         if (it->task_pos) {
    4459             :                 /*
    4460             :                  * Advance iterator to find next entry. We go through cset
    4461             :                  * tasks, mg_tasks and dying_tasks, when consumed we move onto
    4462             :                  * the next cset.
    4463             :                  */
    4464          75 :                 if (it->flags & CSS_TASK_ITER_SKIPPED)
    4465           0 :                         it->flags &= ~CSS_TASK_ITER_SKIPPED;
    4466             :                 else
    4467          75 :                         it->task_pos = it->task_pos->next;
    4468             : 
    4469          75 :                 if (it->task_pos == &it->cur_cset->tasks) {
    4470           5 :                         it->cur_tasks_head = &it->cur_cset->mg_tasks;
    4471           5 :                         it->task_pos = it->cur_tasks_head->next;
    4472             :                 }
    4473          75 :                 if (it->task_pos == &it->cur_cset->mg_tasks) {
    4474           5 :                         it->cur_tasks_head = &it->cur_cset->dying_tasks;
    4475           5 :                         it->task_pos = it->cur_tasks_head->next;
    4476             :                 }
    4477          75 :                 if (it->task_pos == &it->cur_cset->dying_tasks)
    4478          25 :                         css_task_iter_advance_css_set(it);
    4479             :         } else {
    4480             :                 /* called from start, proceed to the first cset */
    4481          26 :                 css_task_iter_advance_css_set(it);
    4482             :         }
    4483             : 
    4484         101 :         if (!it->task_pos)
    4485             :                 return;
    4486             : 
    4487          75 :         task = list_entry(it->task_pos, struct task_struct, cg_list);
    4488             : 
    4489          75 :         if (it->flags & CSS_TASK_ITER_PROCS) {
    4490             :                 /* if PROCS, skip over tasks which aren't group leaders */
    4491          75 :                 if (!thread_group_leader(task))
    4492           6 :                         goto repeat;
    4493             : 
    4494             :                 /* and dying leaders w/o live member threads */
    4495          69 :                 if (it->cur_tasks_head == &it->cur_cset->dying_tasks &&
    4496          20 :                     !atomic_read(&task->signal->live))
    4497          20 :                         goto repeat;
    4498             :         } else {
    4499             :                 /* skip all dying ones */
    4500           0 :                 if (it->cur_tasks_head == &it->cur_cset->dying_tasks)
    4501           0 :                         goto repeat;
    4502             :         }
    4503             : }
    4504             : 
    4505             : /**
    4506             :  * css_task_iter_start - initiate task iteration
    4507             :  * @css: the css to walk tasks of
    4508             :  * @flags: CSS_TASK_ITER_* flags
    4509             :  * @it: the task iterator to use
    4510             :  *
    4511             :  * Initiate iteration through the tasks of @css.  The caller can call
    4512             :  * css_task_iter_next() to walk through the tasks until the function
    4513             :  * returns NULL.  On completion of iteration, css_task_iter_end() must be
    4514             :  * called.
    4515             :  */
    4516          26 : void css_task_iter_start(struct cgroup_subsys_state *css, unsigned int flags,
    4517             :                          struct css_task_iter *it)
    4518             : {
    4519          26 :         memset(it, 0, sizeof(*it));
    4520             : 
    4521          26 :         spin_lock_irq(&css_set_lock);
    4522             : 
    4523          26 :         it->ss = css->ss;
    4524          26 :         it->flags = flags;
    4525             : 
    4526          26 :         if (it->ss)
    4527           0 :                 it->cset_pos = &css->cgroup->e_csets[css->ss->id];
    4528             :         else
    4529          26 :                 it->cset_pos = &css->cgroup->cset_links;
    4530             : 
    4531          26 :         it->cset_head = it->cset_pos;
    4532             : 
    4533          26 :         css_task_iter_advance(it);
    4534             : 
    4535          26 :         spin_unlock_irq(&css_set_lock);
    4536          26 : }
    4537             : 
    4538             : /**
    4539             :  * css_task_iter_next - return the next task for the iterator
    4540             :  * @it: the task iterator being iterated
    4541             :  *
    4542             :  * The "next" function for task iteration.  @it should have been
    4543             :  * initialized via css_task_iter_start().  Returns NULL when the iteration
    4544             :  * reaches the end.
    4545             :  */
    4546          75 : struct task_struct *css_task_iter_next(struct css_task_iter *it)
    4547             : {
    4548          75 :         if (it->cur_task) {
    4549          49 :                 put_task_struct(it->cur_task);
    4550          49 :                 it->cur_task = NULL;
    4551             :         }
    4552             : 
    4553          75 :         spin_lock_irq(&css_set_lock);
    4554             : 
    4555             :         /* @it may be half-advanced by skips, finish advancing */
    4556          75 :         if (it->flags & CSS_TASK_ITER_SKIPPED)
    4557           0 :                 css_task_iter_advance(it);
    4558             : 
    4559          75 :         if (it->task_pos) {
    4560          49 :                 it->cur_task = list_entry(it->task_pos, struct task_struct,
    4561             :                                           cg_list);
    4562          49 :                 get_task_struct(it->cur_task);
    4563          49 :                 css_task_iter_advance(it);
    4564             :         }
    4565             : 
    4566          75 :         spin_unlock_irq(&css_set_lock);
    4567             : 
    4568          75 :         return it->cur_task;
    4569             : }
    4570             : 
    4571             : /**
    4572             :  * css_task_iter_end - finish task iteration
    4573             :  * @it: the task iterator to finish
    4574             :  *
    4575             :  * Finish task iteration started by css_task_iter_start().
    4576             :  */
    4577          26 : void css_task_iter_end(struct css_task_iter *it)
    4578             : {
    4579          26 :         if (it->cur_cset) {
    4580          25 :                 spin_lock_irq(&css_set_lock);
    4581          25 :                 list_del(&it->iters_node);
    4582          25 :                 put_css_set_locked(it->cur_cset);
    4583          25 :                 spin_unlock_irq(&css_set_lock);
    4584             :         }
    4585             : 
    4586          26 :         if (it->cur_dcset)
    4587          25 :                 put_css_set(it->cur_dcset);
    4588             : 
    4589          26 :         if (it->cur_task)
    4590           0 :                 put_task_struct(it->cur_task);
    4591          26 : }
    4592             : 
    4593         125 : static void cgroup_procs_release(struct kernfs_open_file *of)
    4594             : {
    4595         125 :         if (of->priv) {
    4596          26 :                 css_task_iter_end(of->priv);
    4597          26 :                 kfree(of->priv);
    4598             :         }
    4599         125 : }
    4600             : 
    4601          75 : static void *cgroup_procs_next(struct seq_file *s, void *v, loff_t *pos)
    4602             : {
    4603          75 :         struct kernfs_open_file *of = s->private;
    4604          75 :         struct css_task_iter *it = of->priv;
    4605             : 
    4606          49 :         if (pos)
    4607          49 :                 (*pos)++;
    4608             : 
    4609          49 :         return css_task_iter_next(it);
    4610             : }
    4611             : 
    4612          31 : static void *__cgroup_procs_start(struct seq_file *s, loff_t *pos,
    4613             :                                   unsigned int iter_flags)
    4614             : {
    4615          31 :         struct kernfs_open_file *of = s->private;
    4616          31 :         struct cgroup *cgrp = seq_css(s)->cgroup;
    4617          31 :         struct css_task_iter *it = of->priv;
    4618             : 
    4619             :         /*
    4620             :          * When a seq_file is seeked, it's always traversed sequentially
    4621             :          * from position 0, so we can simply keep iterating on !0 *pos.
    4622             :          */
    4623          31 :         if (!it) {
    4624          26 :                 if (WARN_ON_ONCE((*pos)))
    4625          31 :                         return ERR_PTR(-EINVAL);
    4626             : 
    4627          26 :                 it = kzalloc(sizeof(*it), GFP_KERNEL);
    4628          26 :                 if (!it)
    4629          31 :                         return ERR_PTR(-ENOMEM);
    4630          26 :                 of->priv = it;
    4631          26 :                 css_task_iter_start(&cgrp->self, iter_flags, it);
    4632           5 :         } else if (!(*pos)) {
    4633           0 :                 css_task_iter_end(it);
    4634           0 :                 css_task_iter_start(&cgrp->self, iter_flags, it);
    4635             :         } else
    4636           5 :                 return it->cur_task;
    4637             : 
    4638          26 :         return cgroup_procs_next(s, NULL, NULL);
    4639             : }
    4640             : 
    4641          31 : static void *cgroup_procs_start(struct seq_file *s, loff_t *pos)
    4642             : {
    4643          31 :         struct cgroup *cgrp = seq_css(s)->cgroup;
    4644             : 
    4645             :         /*
    4646             :          * All processes of a threaded subtree belong to the domain cgroup
    4647             :          * of the subtree.  Only threads can be distributed across the
    4648             :          * subtree.  Reject reads on cgroup.procs in the subtree proper.
    4649             :          * They're always empty anyway.
    4650             :          */
    4651          31 :         if (cgroup_is_threaded(cgrp))
    4652          31 :                 return ERR_PTR(-EOPNOTSUPP);
    4653             : 
    4654          31 :         return __cgroup_procs_start(s, pos, CSS_TASK_ITER_PROCS |
    4655             :                                             CSS_TASK_ITER_THREADED);
    4656             : }
    4657             : 
    4658          49 : static int cgroup_procs_show(struct seq_file *s, void *v)
    4659             : {
    4660          49 :         seq_printf(s, "%d\n", task_pid_vnr(v));
    4661          49 :         return 0;
    4662             : }
    4663             : 
    4664          99 : static int cgroup_may_write(const struct cgroup *cgrp, struct super_block *sb)
    4665             : {
    4666          99 :         int ret;
    4667          99 :         struct inode *inode;
    4668             : 
    4669         297 :         lockdep_assert_held(&cgroup_mutex);
    4670             : 
    4671          99 :         inode = kernfs_get_inode(sb, cgrp->procs_file.kn);
    4672          99 :         if (!inode)
    4673             :                 return -ENOMEM;
    4674             : 
    4675          99 :         ret = inode_permission(&init_user_ns, inode, MAY_WRITE);
    4676          99 :         iput(inode);
    4677          99 :         return ret;
    4678             : }
    4679             : 
    4680          99 : static int cgroup_procs_write_permission(struct cgroup *src_cgrp,
    4681             :                                          struct cgroup *dst_cgrp,
    4682             :                                          struct super_block *sb)
    4683             : {
    4684          99 :         struct cgroup_namespace *ns = current->nsproxy->cgroup_ns;
    4685          99 :         struct cgroup *com_cgrp = src_cgrp;
    4686          99 :         int ret;
    4687             : 
    4688         297 :         lockdep_assert_held(&cgroup_mutex);
    4689             : 
    4690             :         /* find the common ancestor */
    4691         153 :         while (!cgroup_is_descendant(dst_cgrp, com_cgrp))
    4692         207 :                 com_cgrp = cgroup_parent(com_cgrp);
    4693             : 
    4694             :         /* %current should be authorized to migrate to the common ancestor */
    4695          99 :         ret = cgroup_may_write(com_cgrp, sb);
    4696          99 :         if (ret)
    4697             :                 return ret;
    4698             : 
    4699             :         /*
    4700             :          * If namespaces are delegation boundaries, %current must be able
    4701             :          * to see both source and destination cgroups from its namespace.
    4702             :          */
    4703          99 :         if ((cgrp_dfl_root.flags & CGRP_ROOT_NS_DELEGATE) &&
    4704         198 :             (!cgroup_is_descendant(src_cgrp, ns->root_cset->dfl_cgrp) ||
    4705         198 :              !cgroup_is_descendant(dst_cgrp, ns->root_cset->dfl_cgrp)))
    4706           0 :                 return -ENOENT;
    4707             : 
    4708             :         return 0;
    4709             : }
    4710             : 
    4711          99 : static int cgroup_attach_permissions(struct cgroup *src_cgrp,
    4712             :                                      struct cgroup *dst_cgrp,
    4713             :                                      struct super_block *sb, bool threadgroup)
    4714             : {
    4715          99 :         int ret = 0;
    4716             : 
    4717          99 :         ret = cgroup_procs_write_permission(src_cgrp, dst_cgrp, sb);
    4718          99 :         if (ret)
    4719             :                 return ret;
    4720             : 
    4721          99 :         ret = cgroup_migrate_vet_dst(dst_cgrp);
    4722          99 :         if (ret)
    4723             :                 return ret;
    4724             : 
    4725          99 :         if (!threadgroup && (src_cgrp->dom_cgrp != dst_cgrp->dom_cgrp))
    4726           0 :                 ret = -EOPNOTSUPP;
    4727             : 
    4728             :         return ret;
    4729             : }
    4730             : 
    4731          99 : static ssize_t __cgroup_procs_write(struct kernfs_open_file *of, char *buf,
    4732             :                                     bool threadgroup)
    4733             : {
    4734          99 :         struct cgroup *src_cgrp, *dst_cgrp;
    4735          99 :         struct task_struct *task;
    4736          99 :         ssize_t ret;
    4737          99 :         bool locked;
    4738             : 
    4739          99 :         dst_cgrp = cgroup_kn_lock_live(of->kn, false);
    4740          99 :         if (!dst_cgrp)
    4741             :                 return -ENODEV;
    4742             : 
    4743          99 :         task = cgroup_procs_write_start(buf, threadgroup, &locked);
    4744          99 :         ret = PTR_ERR_OR_ZERO(task);
    4745          99 :         if (ret)
    4746           0 :                 goto out_unlock;
    4747             : 
    4748             :         /* find the source cgroup */
    4749          99 :         spin_lock_irq(&css_set_lock);
    4750          99 :         src_cgrp = task_cgroup_from_root(task, &cgrp_dfl_root);
    4751          99 :         spin_unlock_irq(&css_set_lock);
    4752             : 
    4753             :         /* process and thread migrations follow same delegation rule */
    4754         198 :         ret = cgroup_attach_permissions(src_cgrp, dst_cgrp,
    4755          99 :                                         of->file->f_path.dentry->d_sb, threadgroup);
    4756          99 :         if (ret)
    4757           0 :                 goto out_finish;
    4758             : 
    4759          99 :         ret = cgroup_attach_task(dst_cgrp, task, threadgroup);
    4760             : 
    4761          99 : out_finish:
    4762          99 :         cgroup_procs_write_finish(task, locked);
    4763          99 : out_unlock:
    4764          99 :         cgroup_kn_unlock(of->kn);
    4765             : 
    4766          99 :         return ret;
    4767             : }
    4768             : 
    4769          99 : static ssize_t cgroup_procs_write(struct kernfs_open_file *of,
    4770             :                                   char *buf, size_t nbytes, loff_t off)
    4771             : {
    4772          99 :         return __cgroup_procs_write(of, buf, true) ?: nbytes;
    4773             : }
    4774             : 
    4775           0 : static void *cgroup_threads_start(struct seq_file *s, loff_t *pos)
    4776             : {
    4777           0 :         return __cgroup_procs_start(s, pos, 0);
    4778             : }
    4779             : 
    4780           0 : static ssize_t cgroup_threads_write(struct kernfs_open_file *of,
    4781             :                                     char *buf, size_t nbytes, loff_t off)
    4782             : {
    4783           0 :         return __cgroup_procs_write(of, buf, false) ?: nbytes;
    4784             : }
    4785             : 
    4786             : /* cgroup core interface files for the default hierarchy */
    4787             : static struct cftype cgroup_base_files[] = {
    4788             :         {
    4789             :                 .name = "cgroup.type",
    4790             :                 .flags = CFTYPE_NOT_ON_ROOT,
    4791             :                 .seq_show = cgroup_type_show,
    4792             :                 .write = cgroup_type_write,
    4793             :         },
    4794             :         {
    4795             :                 .name = "cgroup.procs",
    4796             :                 .flags = CFTYPE_NS_DELEGATABLE,
    4797             :                 .file_offset = offsetof(struct cgroup, procs_file),
    4798             :                 .release = cgroup_procs_release,
    4799             :                 .seq_start = cgroup_procs_start,
    4800             :                 .seq_next = cgroup_procs_next,
    4801             :                 .seq_show = cgroup_procs_show,
    4802             :                 .write = cgroup_procs_write,
    4803             :         },
    4804             :         {
    4805             :                 .name = "cgroup.threads",
    4806             :                 .flags = CFTYPE_NS_DELEGATABLE,
    4807             :                 .release = cgroup_procs_release,
    4808             :                 .seq_start = cgroup_threads_start,
    4809             :                 .seq_next = cgroup_procs_next,
    4810             :                 .seq_show = cgroup_procs_show,
    4811             :                 .write = cgroup_threads_write,
    4812             :         },
    4813             :         {
    4814             :                 .name = "cgroup.controllers",
    4815             :                 .seq_show = cgroup_controllers_show,
    4816             :         },
    4817             :         {
    4818             :                 .name = "cgroup.subtree_control",
    4819             :                 .flags = CFTYPE_NS_DELEGATABLE,
    4820             :                 .seq_show = cgroup_subtree_control_show,
    4821             :                 .write = cgroup_subtree_control_write,
    4822             :         },
    4823             :         {
    4824             :                 .name = "cgroup.events",
    4825             :                 .flags = CFTYPE_NOT_ON_ROOT,
    4826             :                 .file_offset = offsetof(struct cgroup, events_file),
    4827             :                 .seq_show = cgroup_events_show,
    4828             :         },
    4829             :         {
    4830             :                 .name = "cgroup.max.descendants",
    4831             :                 .seq_show = cgroup_max_descendants_show,
    4832             :                 .write = cgroup_max_descendants_write,
    4833             :         },
    4834             :         {
    4835             :                 .name = "cgroup.max.depth",
    4836             :                 .seq_show = cgroup_max_depth_show,
    4837             :                 .write = cgroup_max_depth_write,
    4838             :         },
    4839             :         {
    4840             :                 .name = "cgroup.stat",
    4841             :                 .seq_show = cgroup_stat_show,
    4842             :         },
    4843             :         {
    4844             :                 .name = "cgroup.freeze",
    4845             :                 .flags = CFTYPE_NOT_ON_ROOT,
    4846             :                 .seq_show = cgroup_freeze_show,
    4847             :                 .write = cgroup_freeze_write,
    4848             :         },
    4849             :         {
    4850             :                 .name = "cpu.stat",
    4851             :                 .seq_show = cpu_stat_show,
    4852             :         },
    4853             : #ifdef CONFIG_PSI
    4854             :         {
    4855             :                 .name = "io.pressure",
    4856             :                 .seq_show = cgroup_io_pressure_show,
    4857             :                 .write = cgroup_io_pressure_write,
    4858             :                 .poll = cgroup_pressure_poll,
    4859             :                 .release = cgroup_pressure_release,
    4860             :         },
    4861             :         {
    4862             :                 .name = "memory.pressure",
    4863             :                 .seq_show = cgroup_memory_pressure_show,
    4864             :                 .write = cgroup_memory_pressure_write,
    4865             :                 .poll = cgroup_pressure_poll,
    4866             :                 .release = cgroup_pressure_release,
    4867             :         },
    4868             :         {
    4869             :                 .name = "cpu.pressure",
    4870             :                 .seq_show = cgroup_cpu_pressure_show,
    4871             :                 .write = cgroup_cpu_pressure_write,
    4872             :                 .poll = cgroup_pressure_poll,
    4873             :                 .release = cgroup_pressure_release,
    4874             :         },
    4875             : #endif /* CONFIG_PSI */
    4876             :         { }     /* terminate */
    4877             : };
    4878             : 
    4879             : /*
    4880             :  * css destruction is four-stage process.
    4881             :  *
    4882             :  * 1. Destruction starts.  Killing of the percpu_ref is initiated.
    4883             :  *    Implemented in kill_css().
    4884             :  *
    4885             :  * 2. When the percpu_ref is confirmed to be visible as killed on all CPUs
    4886             :  *    and thus css_tryget_online() is guaranteed to fail, the css can be
    4887             :  *    offlined by invoking offline_css().  After offlining, the base ref is
    4888             :  *    put.  Implemented in css_killed_work_fn().
    4889             :  *
    4890             :  * 3. When the percpu_ref reaches zero, the only possible remaining
    4891             :  *    accessors are inside RCU read sections.  css_release() schedules the
    4892             :  *    RCU callback.
    4893             :  *
    4894             :  * 4. After the grace period, the css can be freed.  Implemented in
    4895             :  *    css_free_work_fn().
    4896             :  *
    4897             :  * It is actually hairier because both step 2 and 4 require process context
    4898             :  * and thus involve punting to css->destroy_work adding two additional
    4899             :  * steps to the already complex sequence.
    4900             :  */
    4901          50 : static void css_free_rwork_fn(struct work_struct *work)
    4902             : {
    4903          50 :         struct cgroup_subsys_state *css = container_of(to_rcu_work(work),
    4904             :                                 struct cgroup_subsys_state, destroy_rwork);
    4905          50 :         struct cgroup_subsys *ss = css->ss;
    4906          50 :         struct cgroup *cgrp = css->cgroup;
    4907             : 
    4908          50 :         percpu_ref_exit(&css->refcnt);
    4909             : 
    4910          50 :         if (ss) {
    4911             :                 /* css free path */
    4912           0 :                 struct cgroup_subsys_state *parent = css->parent;
    4913           0 :                 int id = css->id;
    4914             : 
    4915           0 :                 ss->css_free(css);
    4916           0 :                 cgroup_idr_remove(&ss->css_idr, id);
    4917           0 :                 cgroup_put(cgrp);
    4918             : 
    4919           0 :                 if (parent)
    4920           0 :                         css_put(parent);
    4921             :         } else {
    4922             :                 /* cgroup free path */
    4923          50 :                 atomic_dec(&cgrp->root->nr_cgrps);
    4924          50 :                 cgroup1_pidlist_destroy_all(cgrp);
    4925          50 :                 cancel_work_sync(&cgrp->release_agent_work);
    4926             : 
    4927          50 :                 if (cgroup_parent(cgrp)) {
    4928             :                         /*
    4929             :                          * We get a ref to the parent, and put the ref when
    4930             :                          * this cgroup is being freed, so it's guaranteed
    4931             :                          * that the parent won't be destroyed before its
    4932             :                          * children.
    4933             :                          */
    4934          50 :                         cgroup_put(cgroup_parent(cgrp));
    4935          50 :                         kernfs_put(cgrp->kn);
    4936          50 :                         psi_cgroup_free(cgrp);
    4937          50 :                         if (cgroup_on_dfl(cgrp))
    4938          25 :                                 cgroup_rstat_exit(cgrp);
    4939          50 :                         kfree(cgrp);
    4940             :                 } else {
    4941             :                         /*
    4942             :                          * This is root cgroup's refcnt reaching zero,
    4943             :                          * which indicates that the root should be
    4944             :                          * released.
    4945             :                          */
    4946           0 :                         cgroup_destroy_root(cgrp->root);
    4947             :                 }
    4948             :         }
    4949          50 : }
    4950             : 
    4951          50 : static void css_release_work_fn(struct work_struct *work)
    4952             : {
    4953          50 :         struct cgroup_subsys_state *css =
    4954          50 :                 container_of(work, struct cgroup_subsys_state, destroy_work);
    4955          50 :         struct cgroup_subsys *ss = css->ss;
    4956          50 :         struct cgroup *cgrp = css->cgroup;
    4957             : 
    4958          50 :         mutex_lock(&cgroup_mutex);
    4959             : 
    4960          50 :         css->flags |= CSS_RELEASED;
    4961          50 :         list_del_rcu(&css->sibling);
    4962             : 
    4963          50 :         if (ss) {
    4964             :                 /* css release path */
    4965           0 :                 if (!list_empty(&css->rstat_css_node)) {
    4966           0 :                         cgroup_rstat_flush(cgrp);
    4967           0 :                         list_del_rcu(&css->rstat_css_node);
    4968             :                 }
    4969             : 
    4970           0 :                 cgroup_idr_replace(&ss->css_idr, NULL, css->id);
    4971           0 :                 if (ss->css_released)
    4972           0 :                         ss->css_released(css);
    4973             :         } else {
    4974          50 :                 struct cgroup *tcgrp;
    4975             : 
    4976             :                 /* cgroup release path */
    4977          50 :                 TRACE_CGROUP_PATH(release, cgrp);
    4978             : 
    4979          50 :                 if (cgroup_on_dfl(cgrp))
    4980          25 :                         cgroup_rstat_flush(cgrp);
    4981             : 
    4982          50 :                 spin_lock_irq(&css_set_lock);
    4983         158 :                 for (tcgrp = cgroup_parent(cgrp); tcgrp;
    4984         108 :                      tcgrp = cgroup_parent(tcgrp))
    4985         108 :                         tcgrp->nr_dying_descendants--;
    4986          50 :                 spin_unlock_irq(&css_set_lock);
    4987             : 
    4988             :                 /*
    4989             :                  * There are two control paths which try to determine
    4990             :                  * cgroup from dentry without going through kernfs -
    4991             :                  * cgroupstats_build() and css_tryget_online_from_dir().
    4992             :                  * Those are supported by RCU protecting clearing of
    4993             :                  * cgrp->kn->priv backpointer.
    4994             :                  */
    4995          50 :                 if (cgrp->kn)
    4996          50 :                         RCU_INIT_POINTER(*(void __rcu __force **)&cgrp->kn->priv,
    4997             :                                          NULL);
    4998             :         }
    4999             : 
    5000          50 :         mutex_unlock(&cgroup_mutex);
    5001             : 
    5002          50 :         INIT_RCU_WORK(&css->destroy_rwork, css_free_rwork_fn);
    5003          50 :         queue_rcu_work(cgroup_destroy_wq, &css->destroy_rwork);
    5004          50 : }
    5005             : 
    5006          50 : static void css_release(struct percpu_ref *ref)
    5007             : {
    5008          50 :         struct cgroup_subsys_state *css =
    5009          50 :                 container_of(ref, struct cgroup_subsys_state, refcnt);
    5010             : 
    5011          50 :         INIT_WORK(&css->destroy_work, css_release_work_fn);
    5012          50 :         queue_work(cgroup_destroy_wq, &css->destroy_work);
    5013          50 : }
    5014             : 
    5015             : static void init_and_link_css(struct cgroup_subsys_state *css,
    5016             :                               struct cgroup_subsys *ss, struct cgroup *cgrp)
    5017             : {
    5018             :         lockdep_assert_held(&cgroup_mutex);
    5019             : 
    5020             :         cgroup_get_live(cgrp);
    5021             : 
    5022             :         memset(css, 0, sizeof(*css));
    5023             :         css->cgroup = cgrp;
    5024             :         css->ss = ss;
    5025             :         css->id = -1;
    5026             :         INIT_LIST_HEAD(&css->sibling);
    5027             :         INIT_LIST_HEAD(&css->children);
    5028             :         INIT_LIST_HEAD(&css->rstat_css_node);
    5029             :         css->serial_nr = css_serial_nr_next++;
    5030             :         atomic_set(&css->online_cnt, 0);
    5031             : 
    5032             :         if (cgroup_parent(cgrp)) {
    5033             :                 css->parent = cgroup_css(cgroup_parent(cgrp), ss);
    5034             :                 css_get(css->parent);
    5035             :         }
    5036             : 
    5037             :         if (cgroup_on_dfl(cgrp) && ss->css_rstat_flush)
    5038             :                 list_add_rcu(&css->rstat_css_node, &cgrp->rstat_css_list);
    5039             : 
    5040             :         BUG_ON(cgroup_css(cgrp, ss));
    5041             : }
    5042             : 
    5043             : /* invoke ->css_online() on a new CSS and mark it online if successful */
    5044             : static int online_css(struct cgroup_subsys_state *css)
    5045             : {
    5046             :         struct cgroup_subsys *ss = css->ss;
    5047             :         int ret = 0;
    5048             : 
    5049             :         lockdep_assert_held(&cgroup_mutex);
    5050             : 
    5051             :         if (ss->css_online)
    5052             :                 ret = ss->css_online(css);
    5053             :         if (!ret) {
    5054             :                 css->flags |= CSS_ONLINE;
    5055             :                 rcu_assign_pointer(css->cgroup->subsys[ss->id], css);
    5056             : 
    5057             :                 atomic_inc(&css->online_cnt);
    5058             :                 if (css->parent)
    5059             :                         atomic_inc(&css->parent->online_cnt);
    5060             :         }
    5061             :         return ret;
    5062             : }
    5063             : 
    5064             : /* if the CSS is online, invoke ->css_offline() on it and mark it offline */
    5065             : static void offline_css(struct cgroup_subsys_state *css)
    5066             : {
    5067             :         struct cgroup_subsys *ss = css->ss;
    5068             : 
    5069             :         lockdep_assert_held(&cgroup_mutex);
    5070             : 
    5071             :         if (!(css->flags & CSS_ONLINE))
    5072             :                 return;
    5073             : 
    5074             :         if (ss->css_offline)
    5075             :                 ss->css_offline(css);
    5076             : 
    5077             :         css->flags &= ~CSS_ONLINE;
    5078             :         RCU_INIT_POINTER(css->cgroup->subsys[ss->id], NULL);
    5079             : 
    5080             :         wake_up_all(&css->cgroup->offline_waitq);
    5081             : }
    5082             : 
    5083             : /**
    5084             :  * css_create - create a cgroup_subsys_state
    5085             :  * @cgrp: the cgroup new css will be associated with
    5086             :  * @ss: the subsys of new css
    5087             :  *
    5088             :  * Create a new css associated with @cgrp - @ss pair.  On success, the new
    5089             :  * css is online and installed in @cgrp.  This function doesn't create the
    5090             :  * interface files.  Returns 0 on success, -errno on failure.
    5091             :  */
    5092             : static struct cgroup_subsys_state *css_create(struct cgroup *cgrp,
    5093             :                                               struct cgroup_subsys *ss)
    5094             : {
    5095             :         struct cgroup *parent = cgroup_parent(cgrp);
    5096             :         struct cgroup_subsys_state *parent_css = cgroup_css(parent, ss);
    5097             :         struct cgroup_subsys_state *css;
    5098             :         int err;
    5099             : 
    5100             :         lockdep_assert_held(&cgroup_mutex);
    5101             : 
    5102             :         css = ss->css_alloc(parent_css);
    5103             :         if (!css)
    5104             :                 css = ERR_PTR(-ENOMEM);
    5105             :         if (IS_ERR(css))
    5106             :                 return css;
    5107             : 
    5108             :         init_and_link_css(css, ss, cgrp);
    5109             : 
    5110             :         err = percpu_ref_init(&css->refcnt, css_release, 0, GFP_KERNEL);
    5111             :         if (err)
    5112             :                 goto err_free_css;
    5113             : 
    5114             :         err = cgroup_idr_alloc(&ss->css_idr, NULL, 2, 0, GFP_KERNEL);
    5115             :         if (err < 0)
    5116             :                 goto err_free_css;
    5117             :         css->id = err;
    5118             : 
    5119             :         /* @css is ready to be brought online now, make it visible */
    5120             :         list_add_tail_rcu(&css->sibling, &parent_css->children);
    5121             :         cgroup_idr_replace(&ss->css_idr, css, css->id);
    5122             : 
    5123             :         err = online_css(css);
    5124             :         if (err)
    5125             :                 goto err_list_del;
    5126             : 
    5127             :         return css;
    5128             : 
    5129             : err_list_del:
    5130             :         list_del_rcu(&css->sibling);
    5131             : err_free_css:
    5132             :         list_del_rcu(&css->rstat_css_node);
    5133             :         INIT_RCU_WORK(&css->destroy_rwork, css_free_rwork_fn);
    5134             :         queue_rcu_work(cgroup_destroy_wq, &css->destroy_rwork);
    5135             :         return ERR_PTR(err);
    5136             : }
    5137             : 
    5138             : /*
    5139             :  * The returned cgroup is fully initialized including its control mask, but
    5140             :  * it isn't associated with its kernfs_node and doesn't have the control
    5141             :  * mask applied.
    5142             :  */
    5143          98 : static struct cgroup *cgroup_create(struct cgroup *parent, const char *name,
    5144             :                                     umode_t mode)
    5145             : {
    5146          98 :         struct cgroup_root *root = parent->root;
    5147          98 :         struct cgroup *cgrp, *tcgrp;
    5148          98 :         struct kernfs_node *kn;
    5149          98 :         int level = parent->level + 1;
    5150          98 :         int ret;
    5151             : 
    5152             :         /* allocate the cgroup and its ID, 0 is reserved for the root */
    5153         196 :         cgrp = kzalloc(struct_size(cgrp, ancestor_ids, (level + 1)),
    5154             :                        GFP_KERNEL);
    5155          98 :         if (!cgrp)
    5156          98 :                 return ERR_PTR(-ENOMEM);
    5157             : 
    5158          98 :         ret = percpu_ref_init(&cgrp->self.refcnt, css_release, 0, GFP_KERNEL);
    5159          98 :         if (ret)
    5160           0 :                 goto out_free_cgrp;
    5161             : 
    5162          98 :         if (cgroup_on_dfl(parent)) {
    5163          49 :                 ret = cgroup_rstat_init(cgrp);
    5164          49 :                 if (ret)
    5165           0 :                         goto out_cancel_ref;
    5166             :         }
    5167             : 
    5168             :         /* create the directory */
    5169          98 :         kn = kernfs_create_dir(parent->kn, name, mode, cgrp);
    5170          98 :         if (IS_ERR(kn)) {
    5171           0 :                 ret = PTR_ERR(kn);
    5172           0 :                 goto out_stat_exit;
    5173             :         }
    5174          98 :         cgrp->kn = kn;
    5175             : 
    5176          98 :         init_cgroup_housekeeping(cgrp);
    5177             : 
    5178          98 :         cgrp->self.parent = &parent->self;
    5179          98 :         cgrp->root = root;
    5180          98 :         cgrp->level = level;
    5181             : 
    5182          98 :         ret = psi_cgroup_alloc(cgrp);
    5183          98 :         if (ret)
    5184             :                 goto out_kernfs_remove;
    5185             : 
    5186          98 :         ret = cgroup_bpf_inherit(cgrp);
    5187          98 :         if (ret)
    5188             :                 goto out_psi_free;
    5189             : 
    5190             :         /*
    5191             :          * New cgroup inherits effective freeze counter, and
    5192             :          * if the parent has to be frozen, the child has too.
    5193             :          */
    5194          98 :         cgrp->freezer.e_freeze = parent->freezer.e_freeze;
    5195          98 :         if (cgrp->freezer.e_freeze) {
    5196             :                 /*
    5197             :                  * Set the CGRP_FREEZE flag, so when a process will be
    5198             :                  * attached to the child cgroup, it will become frozen.
    5199             :                  * At this point the new cgroup is unpopulated, so we can
    5200             :                  * consider it frozen immediately.
    5201             :                  */
    5202           0 :                 set_bit(CGRP_FREEZE, &cgrp->flags);
    5203           0 :                 set_bit(CGRP_FROZEN, &cgrp->flags);
    5204             :         }
    5205             : 
    5206          98 :         spin_lock_irq(&css_set_lock);
    5207         502 :         for (tcgrp = cgrp; tcgrp; tcgrp = cgroup_parent(tcgrp)) {
    5208         306 :                 cgrp->ancestor_ids[tcgrp->level] = cgroup_id(tcgrp);
    5209             : 
    5210         306 :                 if (tcgrp != cgrp) {
    5211         208 :                         tcgrp->nr_descendants++;
    5212             : 
    5213             :                         /*
    5214             :                          * If the new cgroup is frozen, all ancestor cgroups
    5215             :                          * get a new frozen descendant, but their state can't
    5216             :                          * change because of this.
    5217             :                          */
    5218         208 :                         if (cgrp->freezer.e_freeze)
    5219           0 :                                 tcgrp->freezer.nr_frozen_descendants++;
    5220             :                 }
    5221             :         }
    5222          98 :         spin_unlock_irq(&css_set_lock);
    5223             : 
    5224          98 :         if (notify_on_release(parent))
    5225           0 :                 set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
    5226             : 
    5227          98 :         if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &parent->flags))
    5228           0 :                 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags);
    5229             : 
    5230          98 :         cgrp->self.serial_nr = css_serial_nr_next++;
    5231             : 
    5232             :         /* allocation complete, commit to creation */
    5233         196 :         list_add_tail_rcu(&cgrp->self.sibling, &cgroup_parent(cgrp)->self.children);
    5234          98 :         atomic_inc(&root->nr_cgrps);
    5235          98 :         cgroup_get_live(parent);
    5236             : 
    5237             :         /*
    5238             :          * On the default hierarchy, a child doesn't automatically inherit
    5239             :          * subtree_control from the parent.  Each is configured manually.
    5240             :          */
    5241          98 :         if (!cgroup_on_dfl(cgrp))
    5242          49 :                 cgrp->subtree_control = cgroup_control(cgrp);
    5243             : 
    5244          98 :         cgroup_propagate_control(cgrp);
    5245             : 
    5246          98 :         return cgrp;
    5247             : 
    5248             : out_psi_free:
    5249             :         psi_cgroup_free(cgrp);
    5250             : out_kernfs_remove:
    5251             :         kernfs_remove(cgrp->kn);
    5252           0 : out_stat_exit:
    5253           0 :         if (cgroup_on_dfl(parent))
    5254           0 :                 cgroup_rstat_exit(cgrp);
    5255           0 : out_cancel_ref:
    5256           0 :         percpu_ref_exit(&cgrp->self.refcnt);
    5257           0 : out_free_cgrp:
    5258           0 :         kfree(cgrp);
    5259           0 :         return ERR_PTR(ret);
    5260             : }
    5261             : 
    5262          98 : static bool cgroup_check_hierarchy_limits(struct cgroup *parent)
    5263             : {
    5264          98 :         struct cgroup *cgroup;
    5265          98 :         int ret = false;
    5266          98 :         int level = 1;
    5267             : 
    5268         294 :         lockdep_assert_held(&cgroup_mutex);
    5269             : 
    5270         306 :         for (cgroup = parent; cgroup; cgroup = cgroup_parent(cgroup)) {
    5271         208 :                 if (cgroup->nr_descendants >= cgroup->max_descendants)
    5272           0 :                         goto fail;
    5273             : 
    5274         208 :                 if (level > cgroup->max_depth)
    5275           0 :                         goto fail;
    5276             : 
    5277         208 :                 level++;
    5278             :         }
    5279             : 
    5280             :         ret = true;
    5281          98 : fail:
    5282          98 :         return ret;
    5283             : }
    5284             : 
    5285          98 : int cgroup_mkdir(struct kernfs_node *parent_kn, const char *name, umode_t mode)
    5286             : {
    5287          98 :         struct cgroup *parent, *cgrp;
    5288          98 :         int ret;
    5289             : 
    5290             :         /* do not accept '\n' to prevent making /proc/<pid>/cgroup unparsable */
    5291          98 :         if (strchr(name, '\n'))
    5292             :                 return -EINVAL;
    5293             : 
    5294          98 :         parent = cgroup_kn_lock_live(parent_kn, false);
    5295          98 :         if (!parent)
    5296             :                 return -ENODEV;
    5297             : 
    5298          98 :         if (!cgroup_check_hierarchy_limits(parent)) {
    5299           0 :                 ret = -EAGAIN;
    5300           0 :                 goto out_unlock;
    5301             :         }
    5302             : 
    5303          98 :         cgrp = cgroup_create(parent, name, mode);
    5304          98 :         if (IS_ERR(cgrp)) {
    5305           0 :                 ret = PTR_ERR(cgrp);
    5306           0 :                 goto out_unlock;
    5307             :         }
    5308             : 
    5309             :         /*
    5310             :          * This extra ref will be put in cgroup_free_fn() and guarantees
    5311             :          * that @cgrp->kn is always accessible.
    5312             :          */
    5313          98 :         kernfs_get(cgrp->kn);
    5314             : 
    5315          98 :         ret = cgroup_kn_set_ugid(cgrp->kn);
    5316          98 :         if (ret)
    5317           0 :                 goto out_destroy;
    5318             : 
    5319          98 :         ret = css_populate_dir(&cgrp->self);
    5320          98 :         if (ret)
    5321           0 :                 goto out_destroy;
    5322             : 
    5323          98 :         ret = cgroup_apply_control_enable(cgrp);
    5324          98 :         if (ret)
    5325           0 :                 goto out_destroy;
    5326             : 
    5327          98 :         TRACE_CGROUP_PATH(mkdir, cgrp);
    5328             : 
    5329             :         /* let's create and online css's */
    5330          98 :         kernfs_activate(cgrp->kn);
    5331             : 
    5332          98 :         ret = 0;
    5333          98 :         goto out_unlock;
    5334             : 
    5335           0 : out_destroy:
    5336           0 :         cgroup_destroy_locked(cgrp);
    5337          98 : out_unlock:
    5338          98 :         cgroup_kn_unlock(parent_kn);
    5339          98 :         return ret;
    5340             : }
    5341             : 
    5342             : /*
    5343             :  * This is called when the refcnt of a css is confirmed to be killed.
    5344             :  * css_tryget_online() is now guaranteed to fail.  Tell the subsystem to
    5345             :  * initate destruction and put the css ref from kill_css().
    5346             :  */
    5347             : static void css_killed_work_fn(struct work_struct *work)
    5348             : {
    5349             :         struct cgroup_subsys_state *css =
    5350             :                 container_of(work, struct cgroup_subsys_state, destroy_work);
    5351             : 
    5352             :         mutex_lock(&cgroup_mutex);
    5353             : 
    5354             :         do {
    5355             :                 offline_css(css);
    5356             :                 css_put(css);
    5357             :                 /* @css can't go away while we're holding cgroup_mutex */
    5358             :                 css = css->parent;
    5359             :         } while (css && atomic_dec_and_test(&css->online_cnt));
    5360             : 
    5361             :         mutex_unlock(&cgroup_mutex);
    5362             : }
    5363             : 
    5364             : /* css kill confirmation processing requires process context, bounce */
    5365             : static void css_killed_ref_fn(struct percpu_ref *ref)
    5366             : {
    5367             :         struct cgroup_subsys_state *css =
    5368             :                 container_of(ref, struct cgroup_subsys_state, refcnt);
    5369             : 
    5370             :         if (atomic_dec_and_test(&css->online_cnt)) {
    5371             :                 INIT_WORK(&css->destroy_work, css_killed_work_fn);
    5372             :                 queue_work(cgroup_destroy_wq, &css->destroy_work);
    5373             :         }
    5374             : }
    5375             : 
    5376             : /**
    5377             :  * kill_css - destroy a css
    5378             :  * @css: css to destroy
    5379             :  *
    5380             :  * This function initiates destruction of @css by removing cgroup interface
    5381             :  * files and putting its base reference.  ->css_offline() will be invoked
    5382             :  * asynchronously once css_tryget_online() is guaranteed to fail and when
    5383             :  * the reference count reaches zero, @css will be released.
    5384             :  */
    5385             : static void kill_css(struct cgroup_subsys_state *css)
    5386             : {
    5387             :         lockdep_assert_held(&cgroup_mutex);
    5388             : 
    5389             :         if (css->flags & CSS_DYING)
    5390             :                 return;
    5391             : 
    5392             :         css->flags |= CSS_DYING;
    5393             : 
    5394             :         /*
    5395             :          * This must happen before css is disassociated with its cgroup.
    5396             :          * See seq_css() for details.
    5397             :          */
    5398             :         css_clear_dir(css);
    5399             : 
    5400             :         /*
    5401             :          * Killing would put the base ref, but we need to keep it alive
    5402             :          * until after ->css_offline().
    5403             :          */
    5404             :         css_get(css);
    5405             : 
    5406             :         /*
    5407             :          * cgroup core guarantees that, by the time ->css_offline() is
    5408             :          * invoked, no new css reference will be given out via
    5409             :          * css_tryget_online().  We can't simply call percpu_ref_kill() and
    5410             :          * proceed to offlining css's because percpu_ref_kill() doesn't
    5411             :          * guarantee that the ref is seen as killed on all CPUs on return.
    5412             :          *
    5413             :          * Use percpu_ref_kill_and_confirm() to get notifications as each
    5414             :          * css is confirmed to be seen as killed on all CPUs.
    5415             :          */
    5416             :         percpu_ref_kill_and_confirm(&css->refcnt, css_killed_ref_fn);
    5417             : }
    5418             : 
    5419             : /**
    5420             :  * cgroup_destroy_locked - the first stage of cgroup destruction
    5421             :  * @cgrp: cgroup to be destroyed
    5422             :  *
    5423             :  * css's make use of percpu refcnts whose killing latency shouldn't be
    5424             :  * exposed to userland and are RCU protected.  Also, cgroup core needs to
    5425             :  * guarantee that css_tryget_online() won't succeed by the time
    5426             :  * ->css_offline() is invoked.  To satisfy all the requirements,
    5427             :  * destruction is implemented in the following two steps.
    5428             :  *
    5429             :  * s1. Verify @cgrp can be destroyed and mark it dying.  Remove all
    5430             :  *     userland visible parts and start killing the percpu refcnts of
    5431             :  *     css's.  Set up so that the next stage will be kicked off once all
    5432             :  *     the percpu refcnts are confirmed to be killed.
    5433             :  *
    5434             :  * s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the
    5435             :  *     rest of destruction.  Once all cgroup references are gone, the
    5436             :  *     cgroup is RCU-freed.
    5437             :  *
    5438             :  * This function implements s1.  After this step, @cgrp is gone as far as
    5439             :  * the userland is concerned and a new cgroup with the same name may be
    5440             :  * created.  As cgroup doesn't care about the names internally, this
    5441             :  * doesn't cause any problem.
    5442             :  */
    5443          51 : static int cgroup_destroy_locked(struct cgroup *cgrp)
    5444             :         __releases(&cgroup_mutex) __acquires(&cgroup_mutex)
    5445             : {
    5446          51 :         struct cgroup *tcgrp, *parent = cgroup_parent(cgrp);
    5447          51 :         struct cgroup_subsys_state *css;
    5448          51 :         struct cgrp_cset_link *link;
    5449          51 :         int ssid;
    5450             : 
    5451         153 :         lockdep_assert_held(&cgroup_mutex);
    5452             : 
    5453             :         /*
    5454             :          * Only migration can raise populated from zero and we're already
    5455             :          * holding cgroup_mutex.
    5456             :          */
    5457          51 :         if (cgroup_is_populated(cgrp))
    5458             :                 return -EBUSY;
    5459             : 
    5460             :         /*
    5461             :          * Make sure there's no live children.  We can't test emptiness of
    5462             :          * ->self.children as dead children linger on it while being
    5463             :          * drained; otherwise, "rmdir parent/child parent" may fail.
    5464             :          */
    5465          50 :         if (css_has_online_children(&cgrp->self))
    5466             :                 return -EBUSY;
    5467             : 
    5468             :         /*
    5469             :          * Mark @cgrp and the associated csets dead.  The former prevents
    5470             :          * further task migration and child creation by disabling
    5471             :          * cgroup_lock_live_group().  The latter makes the csets ignored by
    5472             :          * the migration path.
    5473             :          */
    5474          50 :         cgrp->self.flags &= ~CSS_ONLINE;
    5475             : 
    5476          50 :         spin_lock_irq(&css_set_lock);
    5477         100 :         list_for_each_entry(link, &cgrp->cset_links, cset_link)
    5478          50 :                 link->cset->dead = true;
    5479          50 :         spin_unlock_irq(&css_set_lock);
    5480             : 
    5481             :         /* initiate massacre of all css's */
    5482          50 :         for_each_css(css, ssid, cgrp)
    5483             :                 kill_css(css);
    5484             : 
    5485             :         /* clear and remove @cgrp dir, @cgrp has an extra ref on its kn */
    5486          50 :         css_clear_dir(&cgrp->self);
    5487          50 :         kernfs_remove(cgrp->kn);
    5488             : 
    5489          50 :         if (parent && cgroup_is_threaded(cgrp))
    5490           0 :                 parent->nr_threaded_children--;
    5491             : 
    5492          50 :         spin_lock_irq(&css_set_lock);
    5493         208 :         for (tcgrp = cgroup_parent(cgrp); tcgrp; tcgrp = cgroup_parent(tcgrp)) {
    5494         108 :                 tcgrp->nr_descendants--;
    5495         108 :                 tcgrp->nr_dying_descendants++;
    5496             :                 /*
    5497             :                  * If the dying cgroup is frozen, decrease frozen descendants
    5498             :                  * counters of ancestor cgroups.
    5499             :                  */
    5500         108 :                 if (test_bit(CGRP_FROZEN, &cgrp->flags))
    5501           0 :                         tcgrp->freezer.nr_frozen_descendants--;
    5502             :         }
    5503          50 :         spin_unlock_irq(&css_set_lock);
    5504             : 
    5505          50 :         cgroup1_check_for_release(parent);
    5506             : 
    5507          50 :         cgroup_bpf_offline(cgrp);
    5508             : 
    5509             :         /* put the base reference */
    5510          50 :         percpu_ref_kill(&cgrp->self.refcnt);
    5511             : 
    5512          50 :         return 0;
    5513             : };
    5514             : 
    5515          51 : int cgroup_rmdir(struct kernfs_node *kn)
    5516             : {
    5517          51 :         struct cgroup *cgrp;
    5518          51 :         int ret = 0;
    5519             : 
    5520          51 :         cgrp = cgroup_kn_lock_live(kn, false);
    5521          51 :         if (!cgrp)
    5522             :                 return 0;
    5523             : 
    5524          51 :         ret = cgroup_destroy_locked(cgrp);
    5525          51 :         if (!ret)
    5526          50 :                 TRACE_CGROUP_PATH(rmdir, cgrp);
    5527             : 
    5528          51 :         cgroup_kn_unlock(kn);
    5529          51 :         return ret;
    5530             : }
    5531             : 
    5532             : static struct kernfs_syscall_ops cgroup_kf_syscall_ops = {
    5533             :         .show_options           = cgroup_show_options,
    5534             :         .mkdir                  = cgroup_mkdir,
    5535             :         .rmdir                  = cgroup_rmdir,
    5536             :         .show_path              = cgroup_show_path,
    5537             : };
    5538             : 
    5539             : static void __init cgroup_init_subsys(struct cgroup_subsys *ss, bool early)
    5540             : {
    5541             :         struct cgroup_subsys_state *css;
    5542             : 
    5543             :         pr_debug("Initializing cgroup subsys %s\n", ss->name);
    5544             : 
    5545             :         mutex_lock(&cgroup_mutex);
    5546             : 
    5547             :         idr_init(&ss->css_idr);
    5548             :         INIT_LIST_HEAD(&ss->cfts);
    5549             : 
    5550             :         /* Create the root cgroup state for this subsystem */
    5551             :         ss->root = &cgrp_dfl_root;
    5552             :         css = ss->css_alloc(cgroup_css(&cgrp_dfl_root.cgrp, ss));
    5553             :         /* We don't handle early failures gracefully */
    5554             :         BUG_ON(IS_ERR(css));
    5555             :         init_and_link_css(css, ss, &cgrp_dfl_root.cgrp);
    5556             : 
    5557             :         /*
    5558             :          * Root csses are never destroyed and we can't initialize
    5559             :          * percpu_ref during early init.  Disable refcnting.
    5560             :          */
    5561             :         css->flags |= CSS_NO_REF;
    5562             : 
    5563             :         if (early) {
    5564             :                 /* allocation can't be done safely during early init */
    5565             :                 css->id = 1;
    5566             :         } else {
    5567             :                 css->id = cgroup_idr_alloc(&ss->css_idr, css, 1, 2, GFP_KERNEL);
    5568             :                 BUG_ON(css->id < 0);
    5569             :         }
    5570             : 
    5571             :         /* Update the init_css_set to contain a subsys
    5572             :          * pointer to this state - since the subsystem is
    5573             :          * newly registered, all tasks and hence the
    5574             :          * init_css_set is in the subsystem's root cgroup. */
    5575             :         init_css_set.subsys[ss->id] = css;
    5576             : 
    5577             :         have_fork_callback |= (bool)ss->fork << ss->id;
    5578             :         have_exit_callback |= (bool)ss->exit << ss->id;
    5579             :         have_release_callback |= (bool)ss->release << ss->id;
    5580             :         have_canfork_callback |= (bool)ss->can_fork << ss->id;
    5581             : 
    5582             :         /* At system boot, before all subsystems have been
    5583             :          * registered, no tasks have been forked, so we don't
    5584             :          * need to invoke fork callbacks here. */
    5585             :         BUG_ON(!list_empty(&init_task.tasks));
    5586             : 
    5587             :         BUG_ON(online_css(css));
    5588             : 
    5589             :         mutex_unlock(&cgroup_mutex);
    5590             : }
    5591             : 
    5592             : /**
    5593             :  * cgroup_init_early - cgroup initialization at system boot
    5594             :  *
    5595             :  * Initialize cgroups at system boot, and initialize any
    5596             :  * subsystems that request early init.
    5597             :  */
    5598           1 : int __init cgroup_init_early(void)
    5599             : {
    5600           1 :         static struct cgroup_fs_context __initdata ctx;
    5601           1 :         struct cgroup_subsys *ss;
    5602           1 :         int i;
    5603             : 
    5604           1 :         ctx.root = &cgrp_dfl_root;
    5605           1 :         init_cgroup_root(&ctx);
    5606           1 :         cgrp_dfl_root.cgrp.self.flags |= CSS_NO_REF;
    5607             : 
    5608           1 :         RCU_INIT_POINTER(init_task.cgroups, &init_css_set);
    5609             : 
    5610           1 :         for_each_subsys(ss, i) {
    5611             :                 WARN(!ss->css_alloc || !ss->css_free || ss->name || ss->id,
    5612             :                      "invalid cgroup_subsys %d:%s css_alloc=%p css_free=%p id:name=%d:%s\n",
    5613             :                      i, cgroup_subsys_name[i], ss->css_alloc, ss->css_free,
    5614             :                      ss->id, ss->name);
    5615             :                 WARN(strlen(cgroup_subsys_name[i]) > MAX_CGROUP_TYPE_NAMELEN,
    5616             :                      "cgroup_subsys_name %s too long\n", cgroup_subsys_name[i]);
    5617             : 
    5618             :                 ss->id = i;
    5619             :                 ss->name = cgroup_subsys_name[i];
    5620             :                 if (!ss->legacy_name)
    5621             :                         ss->legacy_name = cgroup_subsys_name[i];
    5622             : 
    5623             :                 if (ss->early_init)
    5624             :                         cgroup_init_subsys(ss, true);
    5625             :         }
    5626           1 :         return 0;
    5627             : }
    5628             : 
    5629             : static u16 cgroup_disable_mask __initdata;
    5630             : 
    5631             : /**
    5632             :  * cgroup_init - cgroup initialization
    5633             :  *
    5634             :  * Register cgroup filesystem and /proc file, and initialize
    5635             :  * any subsystems that didn't request early init.
    5636             :  */
    5637           1 : int __init cgroup_init(void)
    5638             : {
    5639           1 :         struct cgroup_subsys *ss;
    5640           1 :         int ssid;
    5641             : 
    5642           1 :         BUILD_BUG_ON(CGROUP_SUBSYS_COUNT > 16);
    5643           1 :         BUG_ON(cgroup_init_cftypes(NULL, cgroup_base_files));
    5644           1 :         BUG_ON(cgroup_init_cftypes(NULL, cgroup1_base_files));
    5645             : 
    5646           1 :         cgroup_rstat_boot();
    5647             : 
    5648             :         /*
    5649             :          * The latency of the synchronize_rcu() is too high for cgroups,
    5650             :          * avoid it at the cost of forcing all readers into the slow path.
    5651             :          */
    5652           1 :         rcu_sync_enter_start(&cgroup_threadgroup_rwsem.rss);
    5653             : 
    5654           1 :         get_user_ns(init_cgroup_ns.user_ns);
    5655             : 
    5656           1 :         mutex_lock(&cgroup_mutex);
    5657             : 
    5658             :         /*
    5659             :          * Add init_css_set to the hash table so that dfl_root can link to
    5660             :          * it during init.
    5661             :          */
    5662           1 :         hash_add(css_set_table, &init_css_set.hlist,
    5663             :                  css_set_hash(init_css_set.subsys));
    5664             : 
    5665           1 :         BUG_ON(cgroup_setup_root(&cgrp_dfl_root, 0));
    5666             : 
    5667           1 :         mutex_unlock(&cgroup_mutex);
    5668             : 
    5669           1 :         for_each_subsys(ss, ssid) {
    5670             :                 if (ss->early_init) {
    5671             :                         struct cgroup_subsys_state *css =
    5672             :                                 init_css_set.subsys[ss->id];
    5673             : 
    5674             :                         css->id = cgroup_idr_alloc(&ss->css_idr, css, 1, 2,
    5675             :                                                    GFP_KERNEL);
    5676             :                         BUG_ON(css->id < 0);
    5677             :                 } else {
    5678             :                         cgroup_init_subsys(ss, false);
    5679             :                 }
    5680             : 
    5681             :                 list_add_tail(&init_css_set.e_cset_node[ssid],
    5682             :                               &cgrp_dfl_root.cgrp.e_csets[ssid]);
    5683             : 
    5684             :                 /*
    5685             :                  * Setting dfl_root subsys_mask needs to consider the
    5686             :                  * disabled flag and cftype registration needs kmalloc,
    5687             :                  * both of which aren't available during early_init.
    5688             :                  */
    5689             :                 if (cgroup_disable_mask & (1 << ssid)) {
    5690             :                         static_branch_disable(cgroup_subsys_enabled_key[ssid]);
    5691             :                         printk(KERN_INFO "Disabling %s control group subsystem\n",
    5692             :                                ss->name);
    5693             :                         continue;
    5694             :                 }
    5695             : 
    5696             :                 if (cgroup1_ssid_disabled(ssid))
    5697             :                         printk(KERN_INFO "Disabling %s control group subsystem in v1 mounts\n",
    5698             :                                ss->name);
    5699             : 
    5700             :                 cgrp_dfl_root.subsys_mask |= 1 << ss->id;
    5701             : 
    5702             :                 /* implicit controllers must be threaded too */
    5703             :                 WARN_ON(ss->implicit_on_dfl && !ss->threaded);
    5704             : 
    5705             :                 if (ss->implicit_on_dfl)
    5706             :                         cgrp_dfl_implicit_ss_mask |= 1 << ss->id;
    5707             :                 else if (!ss->dfl_cftypes)
    5708             :                         cgrp_dfl_inhibit_ss_mask |= 1 << ss->id;
    5709             : 
    5710             :                 if (ss->threaded)
    5711             :                         cgrp_dfl_threaded_ss_mask |= 1 << ss->id;
    5712             : 
    5713             :                 if (ss->dfl_cftypes == ss->legacy_cftypes) {
    5714             :                         WARN_ON(cgroup_add_cftypes(ss, ss->dfl_cftypes));
    5715             :                 } else {
    5716             :                         WARN_ON(cgroup_add_dfl_cftypes(ss, ss->dfl_cftypes));
    5717             :                         WARN_ON(cgroup_add_legacy_cftypes(ss, ss->legacy_cftypes));
    5718             :                 }
    5719             : 
    5720             :                 if (ss->bind)
    5721             :                         ss->bind(init_css_set.subsys[ssid]);
    5722             : 
    5723             :                 mutex_lock(&cgroup_mutex);
    5724             :                 css_populate_dir(init_css_set.subsys[ssid]);
    5725             :                 mutex_unlock(&cgroup_mutex);
    5726             :         }
    5727             : 
    5728             :         /* init_css_set.subsys[] has been updated, re-hash */
    5729           1 :         hash_del(&init_css_set.hlist);
    5730           1 :         hash_add(css_set_table, &init_css_set.hlist,
    5731             :                  css_set_hash(init_css_set.subsys));
    5732             : 
    5733           1 :         WARN_ON(sysfs_create_mount_point(fs_kobj, "cgroup"));
    5734           1 :         WARN_ON(register_filesystem(&cgroup_fs_type));
    5735           1 :         WARN_ON(register_filesystem(&cgroup2_fs_type));
    5736           1 :         WARN_ON(!proc_create_single("cgroups", 0, NULL, proc_cgroupstats_show));
    5737             : #ifdef CONFIG_CPUSETS
    5738             :         WARN_ON(register_filesystem(&cpuset_fs_type));
    5739             : #endif
    5740             : 
    5741           1 :         return 0;
    5742             : }
    5743             : 
    5744           1 : static int __init cgroup_wq_init(void)
    5745             : {
    5746             :         /*
    5747             :          * There isn't much point in executing destruction path in
    5748             :          * parallel.  Good chunk is serialized with cgroup_mutex anyway.
    5749             :          * Use 1 for @max_active.
    5750             :          *
    5751             :          * We would prefer to do this in cgroup_init() above, but that
    5752             :          * is called before init_workqueues(): so leave this until after.
    5753             :          */
    5754           1 :         cgroup_destroy_wq = alloc_workqueue("cgroup_destroy", 0, 1);
    5755           1 :         BUG_ON(!cgroup_destroy_wq);
    5756           1 :         return 0;
    5757             : }
    5758             : core_initcall(cgroup_wq_init);
    5759             : 
    5760           0 : void cgroup_path_from_kernfs_id(u64 id, char *buf, size_t buflen)
    5761             : {
    5762           0 :         struct kernfs_node *kn;
    5763             : 
    5764           0 :         kn = kernfs_find_and_get_node_by_id(cgrp_dfl_root.kf_root, id);
    5765           0 :         if (!kn)
    5766             :                 return;
    5767           0 :         kernfs_path(kn, buf, buflen);
    5768           0 :         kernfs_put(kn);
    5769             : }
    5770             : 
    5771             : /*
    5772             :  * proc_cgroup_show()
    5773             :  *  - Print task's cgroup paths into seq_file, one line for each hierarchy
    5774             :  *  - Used for /proc/<pid>/cgroup.
    5775             :  */
    5776         172 : int proc_cgroup_show(struct seq_file *m, struct pid_namespace *ns,
    5777             :                      struct pid *pid, struct task_struct *tsk)
    5778             : {
    5779         172 :         char *buf;
    5780         172 :         int retval;
    5781         172 :         struct cgroup_root *root;
    5782             : 
    5783         172 :         retval = -ENOMEM;
    5784         172 :         buf = kmalloc(PATH_MAX, GFP_KERNEL);
    5785         172 :         if (!buf)
    5786           0 :                 goto out;
    5787             : 
    5788         172 :         mutex_lock(&cgroup_mutex);
    5789         172 :         spin_lock_irq(&css_set_lock);
    5790             : 
    5791         516 :         for_each_root(root) {
    5792         344 :                 struct cgroup_subsys *ss;
    5793         344 :                 struct cgroup *cgrp;
    5794         344 :                 int ssid, count = 0;
    5795             : 
    5796         344 :                 if (root == &cgrp_dfl_root && !cgrp_dfl_visible)
    5797           0 :                         continue;
    5798             : 
    5799         344 :                 seq_printf(m, "%d:", root->hierarchy_id);
    5800         344 :                 if (root != &cgrp_dfl_root)
    5801             :                         for_each_subsys(ss, ssid)
    5802             :                                 if (root->subsys_mask & (1 << ssid))
    5803             :                                         seq_printf(m, "%s%s", count++ ? "," : "",
    5804             :                                                    ss->legacy_name);
    5805         344 :                 if (strlen(root->name))
    5806         172 :                         seq_printf(m, "%sname=%s", count ? "," : "",
    5807         172 :                                    root->name);
    5808         344 :                 seq_putc(m, ':');
    5809             : 
    5810         344 :                 cgrp = task_cgroup_from_root(tsk, root);
    5811             : 
    5812             :                 /*
    5813             :                  * On traditional hierarchies, all zombie tasks show up as
    5814             :                  * belonging to the root cgroup.  On the default hierarchy,
    5815             :                  * while a zombie doesn't show up in "cgroup.procs" and
    5816             :                  * thus can't be migrated, its /proc/PID/cgroup keeps
    5817             :                  * reporting the cgroup it belonged to before exiting.  If
    5818             :                  * the cgroup is removed before the zombie is reaped,
    5819             :                  * " (deleted)" is appended to the cgroup path.
    5820             :                  */
    5821         344 :                 if (cgroup_on_dfl(cgrp) || !(tsk->flags & PF_EXITING)) {
    5822         312 :                         retval = cgroup_path_ns_locked(cgrp, buf, PATH_MAX,
    5823         312 :                                                 current->nsproxy->cgroup_ns);
    5824         312 :                         if (retval >= PATH_MAX)
    5825             :                                 retval = -ENAMETOOLONG;
    5826         312 :                         if (retval < 0)
    5827           0 :                                 goto out_unlock;
    5828             : 
    5829         312 :                         seq_puts(m, buf);
    5830             :                 } else {
    5831          32 :                         seq_puts(m, "/");
    5832             :                 }
    5833             : 
    5834         344 :                 if (cgroup_on_dfl(cgrp) && cgroup_is_dead(cgrp))
    5835           0 :                         seq_puts(m, " (deleted)\n");
    5836             :                 else
    5837         344 :                         seq_putc(m, '\n');
    5838             :         }
    5839             : 
    5840             :         retval = 0;
    5841         172 : out_unlock:
    5842         172 :         spin_unlock_irq(&css_set_lock);
    5843         172 :         mutex_unlock(&cgroup_mutex);
    5844         172 :         kfree(buf);
    5845         172 : out:
    5846         172 :         return retval;
    5847             : }
    5848             : 
    5849             : /**
    5850             :  * cgroup_fork - initialize cgroup related fields during copy_process()
    5851             :  * @child: pointer to task_struct of forking parent process.
    5852             :  *
    5853             :  * A task is associated with the init_css_set until cgroup_post_fork()
    5854             :  * attaches it to the target css_set.
    5855             :  */
    5856        1162 : void cgroup_fork(struct task_struct *child)
    5857             : {
    5858        1162 :         RCU_INIT_POINTER(child->cgroups, &init_css_set);
    5859        1162 :         INIT_LIST_HEAD(&child->cg_list);
    5860        1162 : }
    5861             : 
    5862           0 : static struct cgroup *cgroup_get_from_file(struct file *f)
    5863             : {
    5864           0 :         struct cgroup_subsys_state *css;
    5865           0 :         struct cgroup *cgrp;
    5866             : 
    5867           0 :         css = css_tryget_online_from_dir(f->f_path.dentry, NULL);
    5868           0 :         if (IS_ERR(css))
    5869           0 :                 return ERR_CAST(css);
    5870             : 
    5871           0 :         cgrp = css->cgroup;
    5872           0 :         if (!cgroup_on_dfl(cgrp)) {
    5873           0 :                 cgroup_put(cgrp);
    5874           0 :                 return ERR_PTR(-EBADF);
    5875             :         }
    5876             : 
    5877             :         return cgrp;
    5878             : }
    5879             : 
    5880             : /**
    5881             :  * cgroup_css_set_fork - find or create a css_set for a child process
    5882             :  * @kargs: the arguments passed to create the child process
    5883             :  *
    5884             :  * This functions finds or creates a new css_set which the child
    5885             :  * process will be attached to in cgroup_post_fork(). By default,
    5886             :  * the child process will be given the same css_set as its parent.
    5887             :  *
    5888             :  * If CLONE_INTO_CGROUP is specified this function will try to find an
    5889             :  * existing css_set which includes the requested cgroup and if not create
    5890             :  * a new css_set that the child will be attached to later. If this function
    5891             :  * succeeds it will hold cgroup_threadgroup_rwsem on return. If
    5892             :  * CLONE_INTO_CGROUP is requested this function will grab cgroup mutex
    5893             :  * before grabbing cgroup_threadgroup_rwsem and will hold a reference
    5894             :  * to the target cgroup.
    5895             :  */
    5896        1162 : static int cgroup_css_set_fork(struct kernel_clone_args *kargs)
    5897             :         __acquires(&cgroup_mutex) __acquires(&cgroup_threadgroup_rwsem)
    5898             : {
    5899        1162 :         int ret;
    5900        1162 :         struct cgroup *dst_cgrp = NULL;
    5901        1162 :         struct css_set *cset;
    5902        1162 :         struct super_block *sb;
    5903        1162 :         struct file *f;
    5904             : 
    5905        1162 :         if (kargs->flags & CLONE_INTO_CGROUP)
    5906           0 :                 mutex_lock(&cgroup_mutex);
    5907             : 
    5908        1162 :         cgroup_threadgroup_change_begin(current);
    5909             : 
    5910        1162 :         spin_lock_irq(&css_set_lock);
    5911        1162 :         cset = task_css_set(current);
    5912        1162 :         get_css_set(cset);
    5913        1162 :         spin_unlock_irq(&css_set_lock);
    5914             : 
    5915        1162 :         if (!(kargs->flags & CLONE_INTO_CGROUP)) {
    5916        1162 :                 kargs->cset = cset;
    5917        1162 :                 return 0;
    5918             :         }
    5919             : 
    5920           0 :         f = fget_raw(kargs->cgroup);
    5921           0 :         if (!f) {
    5922           0 :                 ret = -EBADF;
    5923           0 :                 goto err;
    5924             :         }
    5925           0 :         sb = f->f_path.dentry->d_sb;
    5926             : 
    5927           0 :         dst_cgrp = cgroup_get_from_file(f);
    5928           0 :         if (IS_ERR(dst_cgrp)) {
    5929           0 :                 ret = PTR_ERR(dst_cgrp);
    5930           0 :                 dst_cgrp = NULL;
    5931           0 :                 goto err;
    5932             :         }
    5933             : 
    5934           0 :         if (cgroup_is_dead(dst_cgrp)) {
    5935           0 :                 ret = -ENODEV;
    5936           0 :                 goto err;
    5937             :         }
    5938             : 
    5939             :         /*
    5940             :          * Verify that we the target cgroup is writable for us. This is
    5941             :          * usually done by the vfs layer but since we're not going through
    5942             :          * the vfs layer here we need to do it "manually".
    5943             :          */
    5944           0 :         ret = cgroup_may_write(dst_cgrp, sb);
    5945           0 :         if (ret)
    5946           0 :                 goto err;
    5947             : 
    5948           0 :         ret = cgroup_attach_permissions(cset->dfl_cgrp, dst_cgrp, sb,
    5949           0 :                                         !(kargs->flags & CLONE_THREAD));
    5950           0 :         if (ret)
    5951           0 :                 goto err;
    5952             : 
    5953           0 :         kargs->cset = find_css_set(cset, dst_cgrp);
    5954           0 :         if (!kargs->cset) {
    5955           0 :                 ret = -ENOMEM;
    5956           0 :                 goto err;
    5957             :         }
    5958             : 
    5959           0 :         put_css_set(cset);
    5960           0 :         fput(f);
    5961           0 :         kargs->cgrp = dst_cgrp;
    5962           0 :         return ret;
    5963             : 
    5964           0 : err:
    5965           0 :         cgroup_threadgroup_change_end(current);
    5966           0 :         mutex_unlock(&cgroup_mutex);
    5967           0 :         if (f)
    5968           0 :                 fput(f);
    5969           0 :         if (dst_cgrp)
    5970           0 :                 cgroup_put(dst_cgrp);
    5971           0 :         put_css_set(cset);
    5972           0 :         if (kargs->cset)
    5973           0 :                 put_css_set(kargs->cset);
    5974             :         return ret;
    5975             : }
    5976             : 
    5977             : /**
    5978             :  * cgroup_css_set_put_fork - drop references we took during fork
    5979             :  * @kargs: the arguments passed to create the child process
    5980             :  *
    5981             :  * Drop references to the prepared css_set and target cgroup if
    5982             :  * CLONE_INTO_CGROUP was requested.
    5983             :  */
    5984        1162 : static void cgroup_css_set_put_fork(struct kernel_clone_args *kargs)
    5985             :         __releases(&cgroup_threadgroup_rwsem) __releases(&cgroup_mutex)
    5986             : {
    5987        1162 :         cgroup_threadgroup_change_end(current);
    5988             : 
    5989        1162 :         if (kargs->flags & CLONE_INTO_CGROUP) {
    5990           0 :                 struct cgroup *cgrp = kargs->cgrp;
    5991           0 :                 struct css_set *cset = kargs->cset;
    5992             : 
    5993           0 :                 mutex_unlock(&cgroup_mutex);
    5994             : 
    5995           0 :                 if (cset) {
    5996           0 :                         put_css_set(cset);
    5997           0 :                         kargs->cset = NULL;
    5998             :                 }
    5999             : 
    6000           0 :                 if (cgrp) {
    6001           0 :                         cgroup_put(cgrp);
    6002           0 :                         kargs->cgrp = NULL;
    6003             :                 }
    6004             :         }
    6005        1162 : }
    6006             : 
    6007             : /**
    6008             :  * cgroup_can_fork - called on a new task before the process is exposed
    6009             :  * @child: the child process
    6010             :  *
    6011             :  * This prepares a new css_set for the child process which the child will
    6012             :  * be attached to in cgroup_post_fork().
    6013             :  * This calls the subsystem can_fork() callbacks. If the cgroup_can_fork()
    6014             :  * callback returns an error, the fork aborts with that error code. This
    6015             :  * allows for a cgroup subsystem to conditionally allow or deny new forks.
    6016             :  */
    6017        1162 : int cgroup_can_fork(struct task_struct *child, struct kernel_clone_args *kargs)
    6018             : {
    6019        1162 :         struct cgroup_subsys *ss;
    6020        1162 :         int i, j, ret;
    6021             : 
    6022        1162 :         ret = cgroup_css_set_fork(kargs);
    6023        1162 :         if (ret)
    6024           0 :                 return ret;
    6025             : 
    6026        1162 :         do_each_subsys_mask(ss, i, have_canfork_callback) {
    6027             :                 ret = ss->can_fork(child, kargs->cset);
    6028             :                 if (ret)
    6029             :                         goto out_revert;
    6030             :         } while_each_subsys_mask();
    6031             : 
    6032             :         return 0;
    6033             : 
    6034             : out_revert:
    6035             :         for_each_subsys(ss, j) {
    6036             :                 if (j >= i)
    6037             :                         break;
    6038             :                 if (ss->cancel_fork)
    6039             :                         ss->cancel_fork(child, kargs->cset);
    6040             :         }
    6041             : 
    6042             :         cgroup_css_set_put_fork(kargs);
    6043             : 
    6044             :         return ret;
    6045             : }
    6046             : 
    6047             : /**
    6048             :  * cgroup_cancel_fork - called if a fork failed after cgroup_can_fork()
    6049             :  * @child: the child process
    6050             :  * @kargs: the arguments passed to create the child process
    6051             :  *
    6052             :  * This calls the cancel_fork() callbacks if a fork failed *after*
    6053             :  * cgroup_can_fork() succeded and cleans up references we took to
    6054             :  * prepare a new css_set for the child process in cgroup_can_fork().
    6055             :  */
    6056           0 : void cgroup_cancel_fork(struct task_struct *child,
    6057             :                         struct kernel_clone_args *kargs)
    6058             : {
    6059           0 :         struct cgroup_subsys *ss;
    6060           0 :         int i;
    6061             : 
    6062           0 :         for_each_subsys(ss, i)
    6063             :                 if (ss->cancel_fork)
    6064             :                         ss->cancel_fork(child, kargs->cset);
    6065             : 
    6066           0 :         cgroup_css_set_put_fork(kargs);
    6067           0 : }
    6068             : 
    6069             : /**
    6070             :  * cgroup_post_fork - finalize cgroup setup for the child process
    6071             :  * @child: the child process
    6072             :  *
    6073             :  * Attach the child process to its css_set calling the subsystem fork()
    6074             :  * callbacks.
    6075             :  */
    6076        1162 : void cgroup_post_fork(struct task_struct *child,
    6077             :                       struct kernel_clone_args *kargs)
    6078             :         __releases(&cgroup_threadgroup_rwsem) __releases(&cgroup_mutex)
    6079             : {
    6080        1162 :         struct cgroup_subsys *ss;
    6081        1162 :         struct css_set *cset;
    6082        1162 :         int i;
    6083             : 
    6084        1162 :         cset = kargs->cset;
    6085        1162 :         kargs->cset = NULL;
    6086             : 
    6087        1162 :         spin_lock_irq(&css_set_lock);
    6088             : 
    6089             :         /* init tasks are special, only link regular threads */
    6090        1162 :         if (likely(child->pid)) {
    6091        1159 :                 WARN_ON_ONCE(!list_empty(&child->cg_list));
    6092        1159 :                 cset->nr_tasks++;
    6093        1159 :                 css_set_move_task(child, NULL, cset, false);
    6094             :         } else {
    6095           3 :                 put_css_set(cset);
    6096           3 :                 cset = NULL;
    6097             :         }
    6098             : 
    6099             :         /*
    6100             :          * If the cgroup has to be frozen, the new task has too.  Let's set
    6101             :          * the JOBCTL_TRAP_FREEZE jobctl bit to get the task into the
    6102             :          * frozen state.
    6103             :          */
    6104        1162 :         if (unlikely(cgroup_task_freeze(child))) {
    6105           0 :                 spin_lock(&child->sighand->siglock);
    6106           0 :                 WARN_ON_ONCE(child->frozen);
    6107           0 :                 child->jobctl |= JOBCTL_TRAP_FREEZE;
    6108           0 :                 spin_unlock(&child->sighand->siglock);
    6109             : 
    6110             :                 /*
    6111             :                  * Calling cgroup_update_frozen() isn't required here,
    6112             :                  * because it will be called anyway a bit later from
    6113             :                  * do_freezer_trap(). So we avoid cgroup's transient switch
    6114             :                  * from the frozen state and back.
    6115             :                  */
    6116             :         }
    6117             : 
    6118        1162 :         spin_unlock_irq(&css_set_lock);
    6119             : 
    6120             :         /*
    6121             :          * Call ss->fork().  This must happen after @child is linked on
    6122             :          * css_set; otherwise, @child might change state between ->fork()
    6123             :          * and addition to css_set.
    6124             :          */
    6125        1162 :         do_each_subsys_mask(ss, i, have_fork_callback) {
    6126             :                 ss->fork(child);
    6127        1162 :         } while_each_subsys_mask();
    6128             : 
    6129             :         /* Make the new cset the root_cset of the new cgroup namespace. */
    6130        1162 :         if (kargs->flags & CLONE_NEWCGROUP) {
    6131           0 :                 struct css_set *rcset = child->nsproxy->cgroup_ns->root_cset;
    6132             : 
    6133           0 :                 get_css_set(cset);
    6134           0 :                 child->nsproxy->cgroup_ns->root_cset = cset;
    6135           0 :                 put_css_set(rcset);
    6136             :         }
    6137             : 
    6138        1162 :         cgroup_css_set_put_fork(kargs);
    6139        1162 : }
    6140             : 
    6141             : /**
    6142             :  * cgroup_exit - detach cgroup from exiting task
    6143             :  * @tsk: pointer to task_struct of exiting process
    6144             :  *
    6145             :  * Description: Detach cgroup from @tsk.
    6146             :  *
    6147             :  */
    6148        1083 : void cgroup_exit(struct task_struct *tsk)
    6149             : {
    6150        1083 :         struct cgroup_subsys *ss;
    6151        1083 :         struct css_set *cset;
    6152        1083 :         int i;
    6153             : 
    6154        1083 :         spin_lock_irq(&css_set_lock);
    6155             : 
    6156        1083 :         WARN_ON_ONCE(list_empty(&tsk->cg_list));
    6157        1083 :         cset = task_css_set(tsk);
    6158        1083 :         css_set_move_task(tsk, cset, NULL, false);
    6159        1083 :         list_add_tail(&tsk->cg_list, &cset->dying_tasks);
    6160        1083 :         cset->nr_tasks--;
    6161             : 
    6162        1083 :         WARN_ON_ONCE(cgroup_task_frozen(tsk));
    6163        1083 :         if (unlikely(cgroup_task_freeze(tsk)))
    6164           0 :                 cgroup_update_frozen(task_dfl_cgroup(tsk));
    6165             : 
    6166        1083 :         spin_unlock_irq(&css_set_lock);
    6167             : 
    6168             :         /* see cgroup_post_fork() for details */
    6169        1083 :         do_each_subsys_mask(ss, i, have_exit_callback) {
    6170             :                 ss->exit(tsk);
    6171        1083 :         } while_each_subsys_mask();
    6172        1083 : }
    6173             : 
    6174        1083 : void cgroup_release(struct task_struct *task)
    6175             : {
    6176        1083 :         struct cgroup_subsys *ss;
    6177        1083 :         int ssid;
    6178             : 
    6179        1083 :         do_each_subsys_mask(ss, ssid, have_release_callback) {
    6180             :                 ss->release(task);
    6181        1083 :         } while_each_subsys_mask();
    6182             : 
    6183        1083 :         spin_lock_irq(&css_set_lock);
    6184        1083 :         css_set_skip_task_iters(task_css_set(task), task);
    6185        1083 :         list_del_init(&task->cg_list);
    6186        1083 :         spin_unlock_irq(&css_set_lock);
    6187        1083 : }
    6188             : 
    6189        1083 : void cgroup_free(struct task_struct *task)
    6190             : {
    6191        1083 :         struct css_set *cset = task_css_set(task);
    6192        1083 :         put_css_set(cset);
    6193        1083 : }
    6194             : 
    6195           0 : static int __init cgroup_disable(char *str)
    6196             : {
    6197           0 :         struct cgroup_subsys *ss;
    6198           0 :         char *token;
    6199           0 :         int i;
    6200             : 
    6201           0 :         while ((token = strsep(&str, ",")) != NULL) {
    6202             :                 if (!*token)
    6203             :                         continue;
    6204             : 
    6205             :                 for_each_subsys(ss, i) {
    6206             :                         if (strcmp(token, ss->name) &&
    6207             :                             strcmp(token, ss->legacy_name))
    6208             :                                 continue;
    6209             :                         cgroup_disable_mask |= 1 << i;
    6210             :                 }
    6211             :         }
    6212           0 :         return 1;
    6213             : }
    6214             : __setup("cgroup_disable=", cgroup_disable);
    6215             : 
    6216           0 : void __init __weak enable_debug_cgroup(void) { }
    6217             : 
    6218           0 : static int __init enable_cgroup_debug(char *str)
    6219             : {
    6220           0 :         cgroup_debug = true;
    6221           0 :         enable_debug_cgroup();
    6222           0 :         return 1;
    6223             : }
    6224             : __setup("cgroup_debug", enable_cgroup_debug);
    6225             : 
    6226             : /**
    6227             :  * css_tryget_online_from_dir - get corresponding css from a cgroup dentry
    6228             :  * @dentry: directory dentry of interest
    6229             :  * @ss: subsystem of interest
    6230             :  *
    6231             :  * If @dentry is a directory for a cgroup which has @ss enabled on it, try
    6232             :  * to get the corresponding css and return it.  If such css doesn't exist
    6233             :  * or can't be pinned, an ERR_PTR value is returned.
    6234             :  */
    6235           0 : struct cgroup_subsys_state *css_tryget_online_from_dir(struct dentry *dentry,
    6236             :                                                        struct cgroup_subsys *ss)
    6237             : {
    6238           0 :         struct kernfs_node *kn = kernfs_node_from_dentry(dentry);
    6239           0 :         struct file_system_type *s_type = dentry->d_sb->s_type;
    6240           0 :         struct cgroup_subsys_state *css = NULL;
    6241           0 :         struct cgroup *cgrp;
    6242             : 
    6243             :         /* is @dentry a cgroup dir? */
    6244           0 :         if ((s_type != &cgroup_fs_type && s_type != &cgroup2_fs_type) ||
    6245           0 :             !kn || kernfs_type(kn) != KERNFS_DIR)
    6246           0 :                 return ERR_PTR(-EBADF);
    6247             : 
    6248           0 :         rcu_read_lock();
    6249             : 
    6250             :         /*
    6251             :          * This path doesn't originate from kernfs and @kn could already
    6252             :          * have been or be removed at any point.  @kn->priv is RCU
    6253             :          * protected for this access.  See css_release_work_fn() for details.
    6254             :          */
    6255           0 :         cgrp = rcu_dereference(*(void __rcu __force **)&kn->priv);
    6256           0 :         if (cgrp)
    6257           0 :                 css = cgroup_css(cgrp, ss);
    6258             : 
    6259           0 :         if (!css || !css_tryget_online(css))
    6260           0 :                 css = ERR_PTR(-ENOENT);
    6261             : 
    6262           0 :         rcu_read_unlock();
    6263           0 :         return css;
    6264             : }
    6265             : 
    6266             : /**
    6267             :  * css_from_id - lookup css by id
    6268             :  * @id: the cgroup id
    6269             :  * @ss: cgroup subsys to be looked into
    6270             :  *
    6271             :  * Returns the css if there's valid one with @id, otherwise returns NULL.
    6272             :  * Should be called under rcu_read_lock().
    6273             :  */
    6274           0 : struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss)
    6275             : {
    6276           0 :         WARN_ON_ONCE(!rcu_read_lock_held());
    6277           0 :         return idr_find(&ss->css_idr, id);
    6278             : }
    6279             : 
    6280             : /**
    6281             :  * cgroup_get_from_path - lookup and get a cgroup from its default hierarchy path
    6282             :  * @path: path on the default hierarchy
    6283             :  *
    6284             :  * Find the cgroup at @path on the default hierarchy, increment its
    6285             :  * reference count and return it.  Returns pointer to the found cgroup on
    6286             :  * success, ERR_PTR(-ENOENT) if @path doesn't exist and ERR_PTR(-ENOTDIR)
    6287             :  * if @path points to a non-directory.
    6288             :  */
    6289           0 : struct cgroup *cgroup_get_from_path(const char *path)
    6290             : {
    6291           0 :         struct kernfs_node *kn;
    6292           0 :         struct cgroup *cgrp;
    6293             : 
    6294           0 :         mutex_lock(&cgroup_mutex);
    6295             : 
    6296           0 :         kn = kernfs_walk_and_get(cgrp_dfl_root.cgrp.kn, path);
    6297           0 :         if (kn) {
    6298           0 :                 if (kernfs_type(kn) == KERNFS_DIR) {
    6299           0 :                         cgrp = kn->priv;
    6300           0 :                         cgroup_get_live(cgrp);
    6301             :                 } else {
    6302           0 :                         cgrp = ERR_PTR(-ENOTDIR);
    6303             :                 }
    6304           0 :                 kernfs_put(kn);
    6305             :         } else {
    6306           0 :                 cgrp = ERR_PTR(-ENOENT);
    6307             :         }
    6308             : 
    6309           0 :         mutex_unlock(&cgroup_mutex);
    6310           0 :         return cgrp;
    6311             : }
    6312             : EXPORT_SYMBOL_GPL(cgroup_get_from_path);
    6313             : 
    6314             : /**
    6315             :  * cgroup_get_from_fd - get a cgroup pointer from a fd
    6316             :  * @fd: fd obtained by open(cgroup2_dir)
    6317             :  *
    6318             :  * Find the cgroup from a fd which should be obtained
    6319             :  * by opening a cgroup directory.  Returns a pointer to the
    6320             :  * cgroup on success. ERR_PTR is returned if the cgroup
    6321             :  * cannot be found.
    6322             :  */
    6323           0 : struct cgroup *cgroup_get_from_fd(int fd)
    6324             : {
    6325           0 :         struct cgroup *cgrp;
    6326           0 :         struct file *f;
    6327             : 
    6328           0 :         f = fget_raw(fd);
    6329           0 :         if (!f)
    6330           0 :                 return ERR_PTR(-EBADF);
    6331             : 
    6332           0 :         cgrp = cgroup_get_from_file(f);
    6333           0 :         fput(f);
    6334           0 :         return cgrp;
    6335             : }
    6336             : EXPORT_SYMBOL_GPL(cgroup_get_from_fd);
    6337             : 
    6338           0 : static u64 power_of_ten(int power)
    6339             : {
    6340           0 :         u64 v = 1;
    6341           0 :         while (power--)
    6342           0 :                 v *= 10;
    6343           0 :         return v;
    6344             : }
    6345             : 
    6346             : /**
    6347             :  * cgroup_parse_float - parse a floating number
    6348             :  * @input: input string
    6349             :  * @dec_shift: number of decimal digits to shift
    6350             :  * @v: output
    6351             :  *
    6352             :  * Parse a decimal floating point number in @input and store the result in
    6353             :  * @v with decimal point right shifted @dec_shift times.  For example, if
    6354             :  * @input is "12.3456" and @dec_shift is 3, *@v will be set to 12345.
    6355             :  * Returns 0 on success, -errno otherwise.
    6356             :  *
    6357             :  * There's nothing cgroup specific about this function except that it's
    6358             :  * currently the only user.
    6359             :  */
    6360           0 : int cgroup_parse_float(const char *input, unsigned dec_shift, s64 *v)
    6361             : {
    6362           0 :         s64 whole, frac = 0;
    6363           0 :         int fstart = 0, fend = 0, flen;
    6364             : 
    6365           0 :         if (!sscanf(input, "%lld.%n%lld%n", &whole, &fstart, &frac, &fend))
    6366             :                 return -EINVAL;
    6367           0 :         if (frac < 0)
    6368             :                 return -EINVAL;
    6369             : 
    6370           0 :         flen = fend > fstart ? fend - fstart : 0;
    6371           0 :         if (flen < dec_shift)
    6372           0 :                 frac *= power_of_ten(dec_shift - flen);
    6373             :         else
    6374           0 :                 frac = DIV_ROUND_CLOSEST_ULL(frac, power_of_ten(flen - dec_shift));
    6375             : 
    6376           0 :         *v = whole * power_of_ten(dec_shift) + frac;
    6377           0 :         return 0;
    6378             : }
    6379             : 
    6380             : /*
    6381             :  * sock->sk_cgrp_data handling.  For more info, see sock_cgroup_data
    6382             :  * definition in cgroup-defs.h.
    6383             :  */
    6384             : #ifdef CONFIG_SOCK_CGROUP_DATA
    6385             : 
    6386             : #if defined(CONFIG_CGROUP_NET_PRIO) || defined(CONFIG_CGROUP_NET_CLASSID)
    6387             : 
    6388             : DEFINE_SPINLOCK(cgroup_sk_update_lock);
    6389             : static bool cgroup_sk_alloc_disabled __read_mostly;
    6390             : 
    6391             : void cgroup_sk_alloc_disable(void)
    6392             : {
    6393             :         if (cgroup_sk_alloc_disabled)
    6394             :                 return;
    6395             :         pr_info("cgroup: disabling cgroup2 socket matching due to net_prio or net_cls activation\n");
    6396             :         cgroup_sk_alloc_disabled = true;
    6397             : }
    6398             : 
    6399             : #else
    6400             : 
    6401             : #define cgroup_sk_alloc_disabled        false
    6402             : 
    6403             : #endif
    6404             : 
    6405             : void cgroup_sk_alloc(struct sock_cgroup_data *skcd)
    6406             : {
    6407             :         if (cgroup_sk_alloc_disabled) {
    6408             :                 skcd->no_refcnt = 1;
    6409             :                 return;
    6410             :         }
    6411             : 
    6412             :         /* Don't associate the sock with unrelated interrupted task's cgroup. */
    6413             :         if (in_interrupt())
    6414             :                 return;
    6415             : 
    6416             :         rcu_read_lock();
    6417             : 
    6418             :         while (true) {
    6419             :                 struct css_set *cset;
    6420             : 
    6421             :                 cset = task_css_set(current);
    6422             :                 if (likely(cgroup_tryget(cset->dfl_cgrp))) {
    6423             :                         skcd->val = (unsigned long)cset->dfl_cgrp;
    6424             :                         cgroup_bpf_get(cset->dfl_cgrp);
    6425             :                         break;
    6426             :                 }
    6427             :                 cpu_relax();
    6428             :         }
    6429             : 
    6430             :         rcu_read_unlock();
    6431             : }
    6432             : 
    6433             : void cgroup_sk_clone(struct sock_cgroup_data *skcd)
    6434             : {
    6435             :         if (skcd->val) {
    6436             :                 if (skcd->no_refcnt)
    6437             :                         return;
    6438             :                 /*
    6439             :                  * We might be cloning a socket which is left in an empty
    6440             :                  * cgroup and the cgroup might have already been rmdir'd.
    6441             :                  * Don't use cgroup_get_live().
    6442             :                  */
    6443             :                 cgroup_get(sock_cgroup_ptr(skcd));
    6444             :                 cgroup_bpf_get(sock_cgroup_ptr(skcd));
    6445             :         }
    6446             : }
    6447             : 
    6448             : void cgroup_sk_free(struct sock_cgroup_data *skcd)
    6449             : {
    6450             :         struct cgroup *cgrp = sock_cgroup_ptr(skcd);
    6451             : 
    6452             :         if (skcd->no_refcnt)
    6453             :                 return;
    6454             :         cgroup_bpf_put(cgrp);
    6455             :         cgroup_put(cgrp);
    6456             : }
    6457             : 
    6458             : #endif  /* CONFIG_SOCK_CGROUP_DATA */
    6459             : 
    6460             : #ifdef CONFIG_CGROUP_BPF
    6461             : int cgroup_bpf_attach(struct cgroup *cgrp,
    6462             :                       struct bpf_prog *prog, struct bpf_prog *replace_prog,
    6463             :                       struct bpf_cgroup_link *link,
    6464             :                       enum bpf_attach_type type,
    6465             :                       u32 flags)
    6466             : {
    6467             :         int ret;
    6468             : 
    6469             :         mutex_lock(&cgroup_mutex);
    6470             :         ret = __cgroup_bpf_attach(cgrp, prog, replace_prog, link, type, flags);
    6471             :         mutex_unlock(&cgroup_mutex);
    6472             :         return ret;
    6473             : }
    6474             : 
    6475             : int cgroup_bpf_detach(struct cgroup *cgrp, struct bpf_prog *prog,
    6476             :                       enum bpf_attach_type type)
    6477             : {
    6478             :         int ret;
    6479             : 
    6480             :         mutex_lock(&cgroup_mutex);
    6481             :         ret = __cgroup_bpf_detach(cgrp, prog, NULL, type);
    6482             :         mutex_unlock(&cgroup_mutex);
    6483             :         return ret;
    6484             : }
    6485             : 
    6486             : int cgroup_bpf_query(struct cgroup *cgrp, const union bpf_attr *attr,
    6487             :                      union bpf_attr __user *uattr)
    6488             : {
    6489             :         int ret;
    6490             : 
    6491             :         mutex_lock(&cgroup_mutex);
    6492             :         ret = __cgroup_bpf_query(cgrp, attr, uattr);
    6493             :         mutex_unlock(&cgroup_mutex);
    6494             :         return ret;
    6495             : }
    6496             : #endif /* CONFIG_CGROUP_BPF */
    6497             : 
    6498             : #ifdef CONFIG_SYSFS
    6499           0 : static ssize_t show_delegatable_files(struct cftype *files, char *buf,
    6500             :                                       ssize_t size, const char *prefix)
    6501             : {
    6502           0 :         struct cftype *cft;
    6503           0 :         ssize_t ret = 0;
    6504             : 
    6505           0 :         for (cft = files; cft && cft->name[0] != '\0'; cft++) {
    6506           0 :                 if (!(cft->flags & CFTYPE_NS_DELEGATABLE))
    6507           0 :                         continue;
    6508             : 
    6509           0 :                 if (prefix)
    6510           0 :                         ret += snprintf(buf + ret, size - ret, "%s.", prefix);
    6511             : 
    6512           0 :                 ret += snprintf(buf + ret, size - ret, "%s\n", cft->name);
    6513             : 
    6514           0 :                 if (WARN_ON(ret >= size))
    6515             :                         break;
    6516             :         }
    6517             : 
    6518           0 :         return ret;
    6519             : }
    6520             : 
    6521           0 : static ssize_t delegate_show(struct kobject *kobj, struct kobj_attribute *attr,
    6522             :                               char *buf)
    6523             : {
    6524           0 :         struct cgroup_subsys *ss;
    6525           0 :         int ssid;
    6526           0 :         ssize_t ret = 0;
    6527             : 
    6528           0 :         ret = show_delegatable_files(cgroup_base_files, buf, PAGE_SIZE - ret,
    6529             :                                      NULL);
    6530             : 
    6531           0 :         for_each_subsys(ss, ssid)
    6532             :                 ret += show_delegatable_files(ss->dfl_cftypes, buf + ret,
    6533             :                                               PAGE_SIZE - ret,
    6534             :                                               cgroup_subsys_name[ssid]);
    6535             : 
    6536           0 :         return ret;
    6537             : }
    6538             : static struct kobj_attribute cgroup_delegate_attr = __ATTR_RO(delegate);
    6539             : 
    6540           0 : static ssize_t features_show(struct kobject *kobj, struct kobj_attribute *attr,
    6541             :                              char *buf)
    6542             : {
    6543           0 :         return snprintf(buf, PAGE_SIZE,
    6544             :                         "nsdelegate\n"
    6545             :                         "memory_localevents\n"
    6546             :                         "memory_recursiveprot\n");
    6547             : }
    6548             : static struct kobj_attribute cgroup_features_attr = __ATTR_RO(features);
    6549             : 
    6550             : static struct attribute *cgroup_sysfs_attrs[] = {
    6551             :         &cgroup_delegate_attr.attr,
    6552             :         &cgroup_features_attr.attr,
    6553             :         NULL,
    6554             : };
    6555             : 
    6556             : static const struct attribute_group cgroup_sysfs_attr_group = {
    6557             :         .attrs = cgroup_sysfs_attrs,
    6558             :         .name = "cgroup",
    6559             : };
    6560             : 
    6561           1 : static int __init cgroup_sysfs_init(void)
    6562             : {
    6563           1 :         return sysfs_create_group(kernel_kobj, &cgroup_sysfs_attr_group);
    6564             : }
    6565             : subsys_initcall(cgroup_sysfs_init);
    6566             : 
    6567             : #endif /* CONFIG_SYSFS */

Generated by: LCOV version 1.14