blob: cbbb46f783de0edcc5bc0d7b48c9274cfea20227 [file] [log] [blame]
Paul Menageddbcc7e2007-10-18 23:39:30 -07001/*
Paul Menageddbcc7e2007-10-18 23:39:30 -07002 * Generic process-grouping system.
3 *
4 * Based originally on the cpuset system, extracted by Paul Menage
5 * Copyright (C) 2006 Google, Inc
6 *
Kirill A. Shutemov0dea1162010-03-10 15:22:20 -08007 * Notifications support
8 * Copyright (C) 2009 Nokia Corporation
9 * Author: Kirill A. Shutemov
10 *
Paul Menageddbcc7e2007-10-18 23:39:30 -070011 * 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
Joe Perchesed3d2612014-04-25 18:28:03 -040029#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
30
Paul Menageddbcc7e2007-10-18 23:39:30 -070031#include <linux/cgroup.h>
eparis@redhat2ce97382011-06-02 21:20:51 +100032#include <linux/cred.h>
Paul Menagec6d57f32009-09-23 15:56:19 -070033#include <linux/ctype.h>
Paul Menageddbcc7e2007-10-18 23:39:30 -070034#include <linux/errno.h>
eparis@redhat2ce97382011-06-02 21:20:51 +100035#include <linux/init_task.h>
Paul Menageddbcc7e2007-10-18 23:39:30 -070036#include <linux/kernel.h>
37#include <linux/list.h>
Jianyu Zhanc9482a52014-04-26 15:40:28 +080038#include <linux/magic.h>
Paul Menageddbcc7e2007-10-18 23:39:30 -070039#include <linux/mm.h>
40#include <linux/mutex.h>
41#include <linux/mount.h>
42#include <linux/pagemap.h>
Paul Menagea4243162007-10-18 23:39:35 -070043#include <linux/proc_fs.h>
Paul Menageddbcc7e2007-10-18 23:39:30 -070044#include <linux/rcupdate.h>
45#include <linux/sched.h>
Paul Menageddbcc7e2007-10-18 23:39:30 -070046#include <linux/slab.h>
Paul Menageddbcc7e2007-10-18 23:39:30 -070047#include <linux/spinlock.h>
Tejun Heo96d365e2014-02-13 06:58:40 -050048#include <linux/rwsem.h>
Paul Menageddbcc7e2007-10-18 23:39:30 -070049#include <linux/string.h>
Paul Menagebbcb81d2007-10-18 23:39:32 -070050#include <linux/sort.h>
Paul Menage81a6a5c2007-10-18 23:39:38 -070051#include <linux/kmod.h>
Balbir Singh846c7bb2007-10-18 23:39:44 -070052#include <linux/delayacct.h>
53#include <linux/cgroupstats.h>
Li Zefan0ac801f2013-01-10 11:49:27 +080054#include <linux/hashtable.h>
Li Zefan096b7fe2009-07-29 15:04:04 -070055#include <linux/pid_namespace.h>
Paul Menage2c6ab6d2009-09-23 15:56:23 -070056#include <linux/idr.h>
Ben Blumd1d9fd32009-09-23 15:56:28 -070057#include <linux/vmalloc.h> /* TODO: replace with more sophisticated array */
Mike Galbraithc4c27fb2012-04-21 09:13:46 +020058#include <linux/kthread.h>
Tejun Heo776f02f2014-02-12 09:29:50 -050059#include <linux/delay.h>
Balbir Singh846c7bb2007-10-18 23:39:44 -070060
Arun Sharma600634972011-07-26 16:09:06 -070061#include <linux/atomic.h>
Paul Menageddbcc7e2007-10-18 23:39:30 -070062
Tejun Heoe25e2cb2011-12-12 18:12:21 -080063/*
Tejun Heob1a21362013-11-29 10:42:58 -050064 * pidlists linger the following amount before being destroyed. The goal
65 * is avoiding frequent destruction in the middle of consecutive read calls
66 * Expiring in the middle is a performance problem not a correctness one.
67 * 1 sec should be enough.
68 */
69#define CGROUP_PIDLIST_DESTROY_DELAY HZ
70
Tejun Heo8d7e6fb2014-02-11 11:52:48 -050071#define CGROUP_FILE_NAME_MAX (MAX_CGROUP_TYPE_NAMELEN + \
72 MAX_CFTYPE_NAME + 2)
73
Tejun Heob1a21362013-11-29 10:42:58 -050074/*
Tejun Heoe25e2cb2011-12-12 18:12:21 -080075 * cgroup_mutex is the master lock. Any modification to cgroup or its
76 * hierarchy must be performed while holding it.
77 *
Tejun Heo0e1d7682014-02-25 10:04:03 -050078 * css_set_rwsem protects task->cgroups pointer, the list of css_set
79 * objects, and the chain of tasks off each css_set.
Tejun Heoe25e2cb2011-12-12 18:12:21 -080080 *
Tejun Heo0e1d7682014-02-25 10:04:03 -050081 * These locks are exported if CONFIG_PROVE_RCU so that accessors in
82 * cgroup.h can use them for lockdep annotations.
Tejun Heoe25e2cb2011-12-12 18:12:21 -080083 */
Tejun Heo22194492013-04-07 09:29:51 -070084#ifdef CONFIG_PROVE_RCU
85DEFINE_MUTEX(cgroup_mutex);
Tejun Heo0e1d7682014-02-25 10:04:03 -050086DECLARE_RWSEM(css_set_rwsem);
87EXPORT_SYMBOL_GPL(cgroup_mutex);
88EXPORT_SYMBOL_GPL(css_set_rwsem);
Tejun Heo22194492013-04-07 09:29:51 -070089#else
Paul Menage81a6a5c2007-10-18 23:39:38 -070090static DEFINE_MUTEX(cgroup_mutex);
Tejun Heo0e1d7682014-02-25 10:04:03 -050091static DECLARE_RWSEM(css_set_rwsem);
Tejun Heo22194492013-04-07 09:29:51 -070092#endif
93
Tejun Heo69e943b2014-02-08 10:36:58 -050094/*
Tejun Heo15a4c832014-05-04 15:09:14 -040095 * Protects cgroup_idr and css_idr so that IDs can be released without
96 * grabbing cgroup_mutex.
Tejun Heo6fa49182014-05-04 15:09:13 -040097 */
98static DEFINE_SPINLOCK(cgroup_idr_lock);
99
100/*
Tejun Heo69e943b2014-02-08 10:36:58 -0500101 * Protects cgroup_subsys->release_agent_path. Modifying it also requires
102 * cgroup_mutex. Reading requires either cgroup_mutex or this spinlock.
103 */
104static DEFINE_SPINLOCK(release_agent_path_lock);
Paul Menage81a6a5c2007-10-18 23:39:38 -0700105
Tejun Heo8353da12014-05-13 12:19:23 -0400106#define cgroup_assert_mutex_or_rcu_locked() \
Tejun Heo87fb54f2013-12-06 15:11:55 -0500107 rcu_lockdep_assert(rcu_read_lock_held() || \
108 lockdep_is_held(&cgroup_mutex), \
Tejun Heo8353da12014-05-13 12:19:23 -0400109 "cgroup_mutex or RCU read lock required");
Tejun Heo780cd8b2013-12-06 15:11:56 -0500110
Ben Blumaae8aab2010-03-10 15:22:07 -0800111/*
Tejun Heoe5fca242013-11-22 17:14:39 -0500112 * 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 */
117static struct workqueue_struct *cgroup_destroy_wq;
118
119/*
Tejun Heob1a21362013-11-29 10:42:58 -0500120 * pidlist destructions need to be flushed on cgroup destruction. Use a
121 * separate workqueue as flush domain.
122 */
123static struct workqueue_struct *cgroup_pidlist_destroy_wq;
124
Tejun Heo3ed80a62014-02-08 10:36:58 -0500125/* generate an array of cgroup subsystem pointers */
Tejun Heo073219e2014-02-08 10:36:58 -0500126#define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys,
Tejun Heo3ed80a62014-02-08 10:36:58 -0500127static struct cgroup_subsys *cgroup_subsys[] = {
Paul Menageddbcc7e2007-10-18 23:39:30 -0700128#include <linux/cgroup_subsys.h>
129};
Tejun Heo073219e2014-02-08 10:36:58 -0500130#undef SUBSYS
131
132/* array of cgroup subsystem names */
133#define SUBSYS(_x) [_x ## _cgrp_id] = #_x,
134static const char *cgroup_subsys_name[] = {
135#include <linux/cgroup_subsys.h>
136};
137#undef SUBSYS
Paul Menageddbcc7e2007-10-18 23:39:30 -0700138
Paul Menageddbcc7e2007-10-18 23:39:30 -0700139/*
Tejun Heo3dd06ff2014-03-19 10:23:54 -0400140 * The default hierarchy, reserved for the subsystems that are otherwise
Tejun Heo9871bf92013-06-24 15:21:47 -0700141 * unattached - it never has more than a single cgroup, and all tasks are
142 * part of that cgroup.
Paul Menageddbcc7e2007-10-18 23:39:30 -0700143 */
Tejun Heoa2dd4242014-03-19 10:23:55 -0400144struct cgroup_root cgrp_dfl_root;
Tejun Heo9871bf92013-06-24 15:21:47 -0700145
Tejun Heoa2dd4242014-03-19 10:23:55 -0400146/*
147 * The default hierarchy always exists but is hidden until mounted for the
148 * first time. This is for backward compatibility.
149 */
150static bool cgrp_dfl_root_visible;
Paul Menageddbcc7e2007-10-18 23:39:30 -0700151
Tejun Heoa8ddc822014-07-15 11:05:10 -0400152/*
153 * Set by the boot param of the same name and makes subsystems with NULL
154 * ->dfl_files to use ->legacy_files on the default hierarchy.
155 */
156static bool cgroup_legacy_files_on_dfl;
157
Tejun Heo5533e012014-05-14 19:33:07 -0400158/* some controllers are not supported in the default hierarchy */
Tejun Heo5de4fa12014-07-15 11:05:10 -0400159static unsigned int cgrp_dfl_root_inhibit_ss_mask;
Tejun Heo5533e012014-05-14 19:33:07 -0400160
Paul Menageddbcc7e2007-10-18 23:39:30 -0700161/* The list of hierarchy roots */
162
Tejun Heo9871bf92013-06-24 15:21:47 -0700163static LIST_HEAD(cgroup_roots);
164static int cgroup_root_count;
Paul Menageddbcc7e2007-10-18 23:39:30 -0700165
Tejun Heo3417ae12014-02-08 10:37:01 -0500166/* hierarchy ID allocation and mapping, protected by cgroup_mutex */
Tejun Heo1a574232013-04-14 11:36:58 -0700167static DEFINE_IDR(cgroup_hierarchy_idr);
Paul Menage2c6ab6d2009-09-23 15:56:23 -0700168
Li Zefan794611a2013-06-18 18:53:53 +0800169/*
Tejun Heo0cb51d72014-05-16 13:22:49 -0400170 * Assign a monotonically increasing serial number to csses. It guarantees
171 * cgroups with bigger numbers are newer than those with smaller numbers.
172 * Also, as csses are always appended to the parent's ->children list, it
173 * guarantees that sibling csses are always sorted in the ascending serial
174 * number order on the list. Protected by cgroup_mutex.
Li Zefan794611a2013-06-18 18:53:53 +0800175 */
Tejun Heo0cb51d72014-05-16 13:22:49 -0400176static u64 css_serial_nr_next = 1;
Li Zefan794611a2013-06-18 18:53:53 +0800177
Paul Menageddbcc7e2007-10-18 23:39:30 -0700178/* This flag indicates whether tasks in the fork and exit paths should
Li Zefana043e3b2008-02-23 15:24:09 -0800179 * check for fork/exit handlers to call. This avoids us having to do
180 * extra work in the fork/exit path if none of the subsystems need to
181 * be called.
Paul Menageddbcc7e2007-10-18 23:39:30 -0700182 */
Li Zefan8947f9d2008-07-25 01:46:56 -0700183static int need_forkexit_callback __read_mostly;
Paul Menageddbcc7e2007-10-18 23:39:30 -0700184
Tejun Heoa14c6872014-07-15 11:05:09 -0400185static struct cftype cgroup_dfl_base_files[];
186static struct cftype cgroup_legacy_base_files[];
Tejun Heo628f7cd2013-06-28 16:24:11 -0700187
Tejun Heo3dd06ff2014-03-19 10:23:54 -0400188static int rebind_subsystems(struct cgroup_root *dst_root,
Tejun Heo69dfa002014-05-04 15:09:13 -0400189 unsigned int ss_mask);
Tejun Heo42809dd2012-11-19 08:13:37 -0800190static int cgroup_destroy_locked(struct cgroup *cgrp);
Tejun Heof63070d2014-07-08 18:02:57 -0400191static int create_css(struct cgroup *cgrp, struct cgroup_subsys *ss,
192 bool visible);
Tejun Heo9d755d32014-05-14 09:15:02 -0400193static void css_release(struct percpu_ref *ref);
Tejun Heof8f22e52014-04-23 11:13:16 -0400194static void kill_css(struct cgroup_subsys_state *css);
Tejun Heo2bb566c2013-08-08 20:11:23 -0400195static int cgroup_addrm_files(struct cgroup *cgrp, struct cftype cfts[],
196 bool is_add);
Tejun Heo42809dd2012-11-19 08:13:37 -0800197
Tejun Heo6fa49182014-05-04 15:09:13 -0400198/* IDR wrappers which synchronize using cgroup_idr_lock */
199static int cgroup_idr_alloc(struct idr *idr, void *ptr, int start, int end,
200 gfp_t gfp_mask)
201{
202 int ret;
203
204 idr_preload(gfp_mask);
Tejun Heo54504e92014-05-13 12:10:59 -0400205 spin_lock_bh(&cgroup_idr_lock);
Tejun Heo6fa49182014-05-04 15:09:13 -0400206 ret = idr_alloc(idr, ptr, start, end, gfp_mask);
Tejun Heo54504e92014-05-13 12:10:59 -0400207 spin_unlock_bh(&cgroup_idr_lock);
Tejun Heo6fa49182014-05-04 15:09:13 -0400208 idr_preload_end();
209 return ret;
210}
211
212static void *cgroup_idr_replace(struct idr *idr, void *ptr, int id)
213{
214 void *ret;
215
Tejun Heo54504e92014-05-13 12:10:59 -0400216 spin_lock_bh(&cgroup_idr_lock);
Tejun Heo6fa49182014-05-04 15:09:13 -0400217 ret = idr_replace(idr, ptr, id);
Tejun Heo54504e92014-05-13 12:10:59 -0400218 spin_unlock_bh(&cgroup_idr_lock);
Tejun Heo6fa49182014-05-04 15:09:13 -0400219 return ret;
220}
221
222static void cgroup_idr_remove(struct idr *idr, int id)
223{
Tejun Heo54504e92014-05-13 12:10:59 -0400224 spin_lock_bh(&cgroup_idr_lock);
Tejun Heo6fa49182014-05-04 15:09:13 -0400225 idr_remove(idr, id);
Tejun Heo54504e92014-05-13 12:10:59 -0400226 spin_unlock_bh(&cgroup_idr_lock);
Tejun Heo6fa49182014-05-04 15:09:13 -0400227}
228
Tejun Heod51f39b2014-05-16 13:22:48 -0400229static struct cgroup *cgroup_parent(struct cgroup *cgrp)
230{
231 struct cgroup_subsys_state *parent_css = cgrp->self.parent;
232
233 if (parent_css)
234 return container_of(parent_css, struct cgroup, self);
235 return NULL;
236}
237
Tejun Heo95109b62013-08-08 20:11:27 -0400238/**
239 * cgroup_css - obtain a cgroup's css for the specified subsystem
240 * @cgrp: the cgroup of interest
Tejun Heo9d800df2014-05-14 09:15:00 -0400241 * @ss: the subsystem of interest (%NULL returns @cgrp->self)
Tejun Heo95109b62013-08-08 20:11:27 -0400242 *
Tejun Heoca8bdca2013-08-26 18:40:56 -0400243 * Return @cgrp's css (cgroup_subsys_state) associated with @ss. This
244 * function must be called either under cgroup_mutex or rcu_read_lock() and
245 * the caller is responsible for pinning the returned css if it wants to
246 * keep accessing it outside the said locks. This function may return
247 * %NULL if @cgrp doesn't have @subsys_id enabled.
Tejun Heo95109b62013-08-08 20:11:27 -0400248 */
249static struct cgroup_subsys_state *cgroup_css(struct cgroup *cgrp,
Tejun Heoca8bdca2013-08-26 18:40:56 -0400250 struct cgroup_subsys *ss)
Tejun Heo95109b62013-08-08 20:11:27 -0400251{
Tejun Heoca8bdca2013-08-26 18:40:56 -0400252 if (ss)
Tejun Heoaec25022014-02-08 10:36:58 -0500253 return rcu_dereference_check(cgrp->subsys[ss->id],
Tejun Heoace2bee2014-02-11 11:52:47 -0500254 lockdep_is_held(&cgroup_mutex));
Tejun Heoca8bdca2013-08-26 18:40:56 -0400255 else
Tejun Heo9d800df2014-05-14 09:15:00 -0400256 return &cgrp->self;
Tejun Heo95109b62013-08-08 20:11:27 -0400257}
Paul Menageddbcc7e2007-10-18 23:39:30 -0700258
Tejun Heoaec3dfc2014-04-23 11:13:14 -0400259/**
260 * cgroup_e_css - obtain a cgroup's effective css for the specified subsystem
261 * @cgrp: the cgroup of interest
Tejun Heo9d800df2014-05-14 09:15:00 -0400262 * @ss: the subsystem of interest (%NULL returns @cgrp->self)
Tejun Heoaec3dfc2014-04-23 11:13:14 -0400263 *
264 * Similar to cgroup_css() but returns the effctive css, which is defined
265 * as the matching css of the nearest ancestor including self which has @ss
266 * enabled. If @ss is associated with the hierarchy @cgrp is on, this
267 * function is guaranteed to return non-NULL css.
268 */
269static struct cgroup_subsys_state *cgroup_e_css(struct cgroup *cgrp,
270 struct cgroup_subsys *ss)
271{
272 lockdep_assert_held(&cgroup_mutex);
273
274 if (!ss)
Tejun Heo9d800df2014-05-14 09:15:00 -0400275 return &cgrp->self;
Tejun Heoaec3dfc2014-04-23 11:13:14 -0400276
277 if (!(cgrp->root->subsys_mask & (1 << ss->id)))
278 return NULL;
279
Tejun Heod51f39b2014-05-16 13:22:48 -0400280 while (cgroup_parent(cgrp) &&
281 !(cgroup_parent(cgrp)->child_subsys_mask & (1 << ss->id)))
282 cgrp = cgroup_parent(cgrp);
Tejun Heoaec3dfc2014-04-23 11:13:14 -0400283
284 return cgroup_css(cgrp, ss);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700285}
286
Paul Menageddbcc7e2007-10-18 23:39:30 -0700287/* convenient tests for these bits */
Tejun Heo54766d42013-06-12 21:04:53 -0700288static inline bool cgroup_is_dead(const struct cgroup *cgrp)
Paul Menageddbcc7e2007-10-18 23:39:30 -0700289{
Tejun Heo184faf32014-05-16 13:22:51 -0400290 return !(cgrp->self.flags & CSS_ONLINE);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700291}
292
Tejun Heob4168642014-05-13 12:16:21 -0400293struct cgroup_subsys_state *of_css(struct kernfs_open_file *of)
Tejun Heo59f52962014-02-11 11:52:49 -0500294{
Tejun Heo2bd59d42014-02-11 11:52:49 -0500295 struct cgroup *cgrp = of->kn->parent->priv;
Tejun Heob4168642014-05-13 12:16:21 -0400296 struct cftype *cft = of_cft(of);
Tejun Heo2bd59d42014-02-11 11:52:49 -0500297
298 /*
299 * This is open and unprotected implementation of cgroup_css().
300 * seq_css() is only called from a kernfs file operation which has
301 * an active reference on the file. Because all the subsystem
302 * files are drained before a css is disassociated with a cgroup,
303 * the matching css from the cgroup's subsys table is guaranteed to
304 * be and stay valid until the enclosing operation is complete.
305 */
306 if (cft->ss)
307 return rcu_dereference_raw(cgrp->subsys[cft->ss->id]);
308 else
Tejun Heo9d800df2014-05-14 09:15:00 -0400309 return &cgrp->self;
Tejun Heo59f52962014-02-11 11:52:49 -0500310}
Tejun Heob4168642014-05-13 12:16:21 -0400311EXPORT_SYMBOL_GPL(of_css);
Tejun Heo59f52962014-02-11 11:52:49 -0500312
Li Zefan78574cf2013-04-08 19:00:38 -0700313/**
314 * cgroup_is_descendant - test ancestry
315 * @cgrp: the cgroup to be tested
316 * @ancestor: possible ancestor of @cgrp
317 *
318 * Test whether @cgrp is a descendant of @ancestor. It also returns %true
319 * if @cgrp == @ancestor. This function is safe to call as long as @cgrp
320 * and @ancestor are accessible.
321 */
322bool cgroup_is_descendant(struct cgroup *cgrp, struct cgroup *ancestor)
323{
324 while (cgrp) {
325 if (cgrp == ancestor)
326 return true;
Tejun Heod51f39b2014-05-16 13:22:48 -0400327 cgrp = cgroup_parent(cgrp);
Li Zefan78574cf2013-04-08 19:00:38 -0700328 }
329 return false;
330}
Paul Menageddbcc7e2007-10-18 23:39:30 -0700331
Adrian Bunke9685a02008-02-07 00:13:46 -0800332static int notify_on_release(const struct cgroup *cgrp)
Paul Menage81a6a5c2007-10-18 23:39:38 -0700333{
Paul Menagebd89aab2007-10-18 23:40:44 -0700334 return test_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
Paul Menage81a6a5c2007-10-18 23:39:38 -0700335}
336
Tejun Heo30159ec2013-06-25 11:53:37 -0700337/**
Tejun Heo1c6727a2013-12-06 15:11:56 -0500338 * for_each_css - iterate all css's of a cgroup
339 * @css: the iteration cursor
340 * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end
341 * @cgrp: the target cgroup to iterate css's of
Tejun Heo30159ec2013-06-25 11:53:37 -0700342 *
Tejun Heoaec3dfc2014-04-23 11:13:14 -0400343 * Should be called under cgroup_[tree_]mutex.
Tejun Heo30159ec2013-06-25 11:53:37 -0700344 */
Tejun Heo1c6727a2013-12-06 15:11:56 -0500345#define for_each_css(css, ssid, cgrp) \
346 for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \
347 if (!((css) = rcu_dereference_check( \
348 (cgrp)->subsys[(ssid)], \
349 lockdep_is_held(&cgroup_mutex)))) { } \
350 else
351
352/**
Tejun Heoaec3dfc2014-04-23 11:13:14 -0400353 * for_each_e_css - iterate all effective css's of a cgroup
354 * @css: the iteration cursor
355 * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end
356 * @cgrp: the target cgroup to iterate css's of
357 *
358 * Should be called under cgroup_[tree_]mutex.
359 */
360#define for_each_e_css(css, ssid, cgrp) \
361 for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \
362 if (!((css) = cgroup_e_css(cgrp, cgroup_subsys[(ssid)]))) \
363 ; \
364 else
365
366/**
Tejun Heo3ed80a62014-02-08 10:36:58 -0500367 * for_each_subsys - iterate all enabled cgroup subsystems
Tejun Heo30159ec2013-06-25 11:53:37 -0700368 * @ss: the iteration cursor
Tejun Heo780cd8b2013-12-06 15:11:56 -0500369 * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end
Tejun Heo30159ec2013-06-25 11:53:37 -0700370 */
Tejun Heo780cd8b2013-12-06 15:11:56 -0500371#define for_each_subsys(ss, ssid) \
Tejun Heo3ed80a62014-02-08 10:36:58 -0500372 for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT && \
373 (((ss) = cgroup_subsys[ssid]) || true); (ssid)++)
Tejun Heo30159ec2013-06-25 11:53:37 -0700374
Tejun Heo985ed672014-03-19 10:23:53 -0400375/* iterate across the hierarchies */
376#define for_each_root(root) \
Tejun Heo5549c492013-06-24 15:21:48 -0700377 list_for_each_entry((root), &cgroup_roots, root_list)
Paul Menageddbcc7e2007-10-18 23:39:30 -0700378
Tejun Heof8f22e52014-04-23 11:13:16 -0400379/* iterate over child cgrps, lock should be held throughout iteration */
380#define cgroup_for_each_live_child(child, cgrp) \
Tejun Heod5c419b2014-05-16 13:22:48 -0400381 list_for_each_entry((child), &(cgrp)->self.children, self.sibling) \
Tejun Heo8353da12014-05-13 12:19:23 -0400382 if (({ lockdep_assert_held(&cgroup_mutex); \
Tejun Heof8f22e52014-04-23 11:13:16 -0400383 cgroup_is_dead(child); })) \
384 ; \
385 else
Tejun Heo7ae1bad2013-04-07 09:29:51 -0700386
Paul Menage81a6a5c2007-10-18 23:39:38 -0700387static void cgroup_release_agent(struct work_struct *work);
Paul Menagebd89aab2007-10-18 23:40:44 -0700388static void check_for_release(struct cgroup *cgrp);
Paul Menage81a6a5c2007-10-18 23:39:38 -0700389
Tejun Heo69d02062013-06-12 21:04:50 -0700390/*
391 * A cgroup can be associated with multiple css_sets as different tasks may
392 * belong to different cgroups on different hierarchies. In the other
393 * direction, a css_set is naturally associated with multiple cgroups.
394 * This M:N relationship is represented by the following link structure
395 * which exists for each association and allows traversing the associations
396 * from both sides.
397 */
398struct cgrp_cset_link {
399 /* the cgroup and css_set this link associates */
400 struct cgroup *cgrp;
401 struct css_set *cset;
402
403 /* list of cgrp_cset_links anchored at cgrp->cset_links */
404 struct list_head cset_link;
405
406 /* list of cgrp_cset_links anchored at css_set->cgrp_links */
407 struct list_head cgrp_link;
Paul Menage817929e2007-10-18 23:39:36 -0700408};
409
Tejun Heo172a2c062014-03-19 10:23:53 -0400410/*
411 * The default css_set - used by init and its children prior to any
Paul Menage817929e2007-10-18 23:39:36 -0700412 * hierarchies being mounted. It contains a pointer to the root state
413 * for each subsystem. Also used to anchor the list of css_sets. Not
414 * reference-counted, to improve performance when child cgroups
415 * haven't been created.
416 */
Tejun Heo5024ae22014-05-07 21:31:17 -0400417struct css_set init_css_set = {
Tejun Heo172a2c062014-03-19 10:23:53 -0400418 .refcount = ATOMIC_INIT(1),
419 .cgrp_links = LIST_HEAD_INIT(init_css_set.cgrp_links),
420 .tasks = LIST_HEAD_INIT(init_css_set.tasks),
421 .mg_tasks = LIST_HEAD_INIT(init_css_set.mg_tasks),
422 .mg_preload_node = LIST_HEAD_INIT(init_css_set.mg_preload_node),
423 .mg_node = LIST_HEAD_INIT(init_css_set.mg_node),
424};
Paul Menage817929e2007-10-18 23:39:36 -0700425
Tejun Heo172a2c062014-03-19 10:23:53 -0400426static int css_set_count = 1; /* 1 for init_css_set */
Paul Menage817929e2007-10-18 23:39:36 -0700427
Tejun Heo842b5972014-04-25 18:28:02 -0400428/**
429 * cgroup_update_populated - updated populated count of a cgroup
430 * @cgrp: the target cgroup
431 * @populated: inc or dec populated count
432 *
433 * @cgrp is either getting the first task (css_set) or losing the last.
434 * Update @cgrp->populated_cnt accordingly. The count is propagated
435 * towards root so that a given cgroup's populated_cnt is zero iff the
436 * cgroup and all its descendants are empty.
437 *
438 * @cgrp's interface file "cgroup.populated" is zero if
439 * @cgrp->populated_cnt is zero and 1 otherwise. When @cgrp->populated_cnt
440 * changes from or to zero, userland is notified that the content of the
441 * interface file has changed. This can be used to detect when @cgrp and
442 * its descendants become populated or empty.
443 */
444static void cgroup_update_populated(struct cgroup *cgrp, bool populated)
445{
446 lockdep_assert_held(&css_set_rwsem);
447
448 do {
449 bool trigger;
450
451 if (populated)
452 trigger = !cgrp->populated_cnt++;
453 else
454 trigger = !--cgrp->populated_cnt;
455
456 if (!trigger)
457 break;
458
459 if (cgrp->populated_kn)
460 kernfs_notify(cgrp->populated_kn);
Tejun Heod51f39b2014-05-16 13:22:48 -0400461 cgrp = cgroup_parent(cgrp);
Tejun Heo842b5972014-04-25 18:28:02 -0400462 } while (cgrp);
463}
464
Paul Menage7717f7b2009-09-23 15:56:22 -0700465/*
466 * hash table for cgroup groups. This improves the performance to find
467 * an existing css_set. This hash doesn't (currently) take into
468 * account cgroups in empty hierarchies.
469 */
Li Zefan472b1052008-04-29 01:00:11 -0700470#define CSS_SET_HASH_BITS 7
Li Zefan0ac801f2013-01-10 11:49:27 +0800471static DEFINE_HASHTABLE(css_set_table, CSS_SET_HASH_BITS);
Li Zefan472b1052008-04-29 01:00:11 -0700472
Li Zefan0ac801f2013-01-10 11:49:27 +0800473static unsigned long css_set_hash(struct cgroup_subsys_state *css[])
Li Zefan472b1052008-04-29 01:00:11 -0700474{
Li Zefan0ac801f2013-01-10 11:49:27 +0800475 unsigned long key = 0UL;
Tejun Heo30159ec2013-06-25 11:53:37 -0700476 struct cgroup_subsys *ss;
477 int i;
Li Zefan472b1052008-04-29 01:00:11 -0700478
Tejun Heo30159ec2013-06-25 11:53:37 -0700479 for_each_subsys(ss, i)
Li Zefan0ac801f2013-01-10 11:49:27 +0800480 key += (unsigned long)css[i];
481 key = (key >> 16) ^ key;
Li Zefan472b1052008-04-29 01:00:11 -0700482
Li Zefan0ac801f2013-01-10 11:49:27 +0800483 return key;
Li Zefan472b1052008-04-29 01:00:11 -0700484}
485
Zefan Lia25eb522014-09-19 16:51:00 +0800486static void put_css_set_locked(struct css_set *cset)
Paul Menageb4f48b62007-10-18 23:39:33 -0700487{
Tejun Heo69d02062013-06-12 21:04:50 -0700488 struct cgrp_cset_link *link, *tmp_link;
Tejun Heo2d8f2432014-04-23 11:13:15 -0400489 struct cgroup_subsys *ss;
490 int ssid;
Tejun Heo5abb8852013-06-12 21:04:49 -0700491
Tejun Heo89c55092014-02-13 06:58:40 -0500492 lockdep_assert_held(&css_set_rwsem);
493
494 if (!atomic_dec_and_test(&cset->refcount))
Lai Jiangshan146aa1b2008-10-18 20:28:03 -0700495 return;
Paul Menage81a6a5c2007-10-18 23:39:38 -0700496
Paul Menage2c6ab6d2009-09-23 15:56:23 -0700497 /* This css_set is dead. unlink it and release cgroup refcounts */
Tejun Heo2d8f2432014-04-23 11:13:15 -0400498 for_each_subsys(ss, ssid)
499 list_del(&cset->e_cset_node[ssid]);
Tejun Heo5abb8852013-06-12 21:04:49 -0700500 hash_del(&cset->hlist);
Paul Menage2c6ab6d2009-09-23 15:56:23 -0700501 css_set_count--;
502
Tejun Heo69d02062013-06-12 21:04:50 -0700503 list_for_each_entry_safe(link, tmp_link, &cset->cgrp_links, cgrp_link) {
Paul Menage2c6ab6d2009-09-23 15:56:23 -0700504 struct cgroup *cgrp = link->cgrp;
Tejun Heo5abb8852013-06-12 21:04:49 -0700505
Tejun Heo69d02062013-06-12 21:04:50 -0700506 list_del(&link->cset_link);
507 list_del(&link->cgrp_link);
Li Zefan71b57072013-01-24 14:43:28 +0800508
Tejun Heo96d365e2014-02-13 06:58:40 -0500509 /* @cgrp can't go away while we're holding css_set_rwsem */
Tejun Heo842b5972014-04-25 18:28:02 -0400510 if (list_empty(&cgrp->cset_links)) {
511 cgroup_update_populated(cgrp, false);
Zefan Lia25eb522014-09-19 16:51:00 +0800512 check_for_release(cgrp);
Paul Menage81a6a5c2007-10-18 23:39:38 -0700513 }
Paul Menage2c6ab6d2009-09-23 15:56:23 -0700514
515 kfree(link);
Paul Menage81a6a5c2007-10-18 23:39:38 -0700516 }
Paul Menage2c6ab6d2009-09-23 15:56:23 -0700517
Tejun Heo5abb8852013-06-12 21:04:49 -0700518 kfree_rcu(cset, rcu_head);
Paul Menage817929e2007-10-18 23:39:36 -0700519}
520
Zefan Lia25eb522014-09-19 16:51:00 +0800521static void put_css_set(struct css_set *cset)
Tejun Heo89c55092014-02-13 06:58:40 -0500522{
523 /*
524 * Ensure that the refcount doesn't hit zero while any readers
525 * can see it. Similar to atomic_dec_and_lock(), but for an
526 * rwlock
527 */
528 if (atomic_add_unless(&cset->refcount, -1, 1))
529 return;
530
531 down_write(&css_set_rwsem);
Zefan Lia25eb522014-09-19 16:51:00 +0800532 put_css_set_locked(cset);
Tejun Heo89c55092014-02-13 06:58:40 -0500533 up_write(&css_set_rwsem);
534}
535
Paul Menage817929e2007-10-18 23:39:36 -0700536/*
537 * refcounted get/put for css_set objects
538 */
Tejun Heo5abb8852013-06-12 21:04:49 -0700539static inline void get_css_set(struct css_set *cset)
Paul Menage817929e2007-10-18 23:39:36 -0700540{
Tejun Heo5abb8852013-06-12 21:04:49 -0700541 atomic_inc(&cset->refcount);
Paul Menage817929e2007-10-18 23:39:36 -0700542}
543
Tejun Heob326f9d2013-06-24 15:21:48 -0700544/**
Paul Menage7717f7b2009-09-23 15:56:22 -0700545 * compare_css_sets - helper function for find_existing_css_set().
Tejun Heo5abb8852013-06-12 21:04:49 -0700546 * @cset: candidate css_set being tested
547 * @old_cset: existing css_set for a task
Paul Menage7717f7b2009-09-23 15:56:22 -0700548 * @new_cgrp: cgroup that's being entered by the task
549 * @template: desired set of css pointers in css_set (pre-calculated)
550 *
Li Zefan6f4b7e62013-07-31 16:18:36 +0800551 * Returns true if "cset" matches "old_cset" except for the hierarchy
Paul Menage7717f7b2009-09-23 15:56:22 -0700552 * which "new_cgrp" belongs to, for which it should match "new_cgrp".
553 */
Tejun Heo5abb8852013-06-12 21:04:49 -0700554static bool compare_css_sets(struct css_set *cset,
555 struct css_set *old_cset,
Paul Menage7717f7b2009-09-23 15:56:22 -0700556 struct cgroup *new_cgrp,
557 struct cgroup_subsys_state *template[])
558{
559 struct list_head *l1, *l2;
560
Tejun Heoaec3dfc2014-04-23 11:13:14 -0400561 /*
562 * On the default hierarchy, there can be csets which are
563 * associated with the same set of cgroups but different csses.
564 * Let's first ensure that csses match.
565 */
566 if (memcmp(template, cset->subsys, sizeof(cset->subsys)))
Paul Menage7717f7b2009-09-23 15:56:22 -0700567 return false;
Paul Menage7717f7b2009-09-23 15:56:22 -0700568
569 /*
570 * Compare cgroup pointers in order to distinguish between
Tejun Heoaec3dfc2014-04-23 11:13:14 -0400571 * different cgroups in hierarchies. As different cgroups may
572 * share the same effective css, this comparison is always
573 * necessary.
Paul Menage7717f7b2009-09-23 15:56:22 -0700574 */
Tejun Heo69d02062013-06-12 21:04:50 -0700575 l1 = &cset->cgrp_links;
576 l2 = &old_cset->cgrp_links;
Paul Menage7717f7b2009-09-23 15:56:22 -0700577 while (1) {
Tejun Heo69d02062013-06-12 21:04:50 -0700578 struct cgrp_cset_link *link1, *link2;
Tejun Heo5abb8852013-06-12 21:04:49 -0700579 struct cgroup *cgrp1, *cgrp2;
Paul Menage7717f7b2009-09-23 15:56:22 -0700580
581 l1 = l1->next;
582 l2 = l2->next;
583 /* See if we reached the end - both lists are equal length. */
Tejun Heo69d02062013-06-12 21:04:50 -0700584 if (l1 == &cset->cgrp_links) {
585 BUG_ON(l2 != &old_cset->cgrp_links);
Paul Menage7717f7b2009-09-23 15:56:22 -0700586 break;
587 } else {
Tejun Heo69d02062013-06-12 21:04:50 -0700588 BUG_ON(l2 == &old_cset->cgrp_links);
Paul Menage7717f7b2009-09-23 15:56:22 -0700589 }
590 /* Locate the cgroups associated with these links. */
Tejun Heo69d02062013-06-12 21:04:50 -0700591 link1 = list_entry(l1, struct cgrp_cset_link, cgrp_link);
592 link2 = list_entry(l2, struct cgrp_cset_link, cgrp_link);
593 cgrp1 = link1->cgrp;
594 cgrp2 = link2->cgrp;
Paul Menage7717f7b2009-09-23 15:56:22 -0700595 /* Hierarchies should be linked in the same order. */
Tejun Heo5abb8852013-06-12 21:04:49 -0700596 BUG_ON(cgrp1->root != cgrp2->root);
Paul Menage7717f7b2009-09-23 15:56:22 -0700597
598 /*
599 * If this hierarchy is the hierarchy of the cgroup
600 * that's changing, then we need to check that this
601 * css_set points to the new cgroup; if it's any other
602 * hierarchy, then this css_set should point to the
603 * same cgroup as the old css_set.
604 */
Tejun Heo5abb8852013-06-12 21:04:49 -0700605 if (cgrp1->root == new_cgrp->root) {
606 if (cgrp1 != new_cgrp)
Paul Menage7717f7b2009-09-23 15:56:22 -0700607 return false;
608 } else {
Tejun Heo5abb8852013-06-12 21:04:49 -0700609 if (cgrp1 != cgrp2)
Paul Menage7717f7b2009-09-23 15:56:22 -0700610 return false;
611 }
612 }
613 return true;
614}
615
Tejun Heob326f9d2013-06-24 15:21:48 -0700616/**
617 * find_existing_css_set - init css array and find the matching css_set
618 * @old_cset: the css_set that we're using before the cgroup transition
619 * @cgrp: the cgroup that we're moving into
620 * @template: out param for the new set of csses, should be clear on entry
Paul Menage817929e2007-10-18 23:39:36 -0700621 */
Tejun Heo5abb8852013-06-12 21:04:49 -0700622static struct css_set *find_existing_css_set(struct css_set *old_cset,
623 struct cgroup *cgrp,
624 struct cgroup_subsys_state *template[])
Paul Menage817929e2007-10-18 23:39:36 -0700625{
Tejun Heo3dd06ff2014-03-19 10:23:54 -0400626 struct cgroup_root *root = cgrp->root;
Tejun Heo30159ec2013-06-25 11:53:37 -0700627 struct cgroup_subsys *ss;
Tejun Heo5abb8852013-06-12 21:04:49 -0700628 struct css_set *cset;
Li Zefan0ac801f2013-01-10 11:49:27 +0800629 unsigned long key;
Tejun Heob326f9d2013-06-24 15:21:48 -0700630 int i;
Paul Menage817929e2007-10-18 23:39:36 -0700631
Ben Blumaae8aab2010-03-10 15:22:07 -0800632 /*
633 * Build the set of subsystem state objects that we want to see in the
634 * new css_set. while subsystems can change globally, the entries here
635 * won't change, so no need for locking.
636 */
Tejun Heo30159ec2013-06-25 11:53:37 -0700637 for_each_subsys(ss, i) {
Tejun Heof392e512014-04-23 11:13:14 -0400638 if (root->subsys_mask & (1UL << i)) {
Tejun Heoaec3dfc2014-04-23 11:13:14 -0400639 /*
640 * @ss is in this hierarchy, so we want the
641 * effective css from @cgrp.
642 */
643 template[i] = cgroup_e_css(cgrp, ss);
Paul Menage817929e2007-10-18 23:39:36 -0700644 } else {
Tejun Heoaec3dfc2014-04-23 11:13:14 -0400645 /*
646 * @ss is not in this hierarchy, so we don't want
647 * to change the css.
648 */
Tejun Heo5abb8852013-06-12 21:04:49 -0700649 template[i] = old_cset->subsys[i];
Paul Menage817929e2007-10-18 23:39:36 -0700650 }
651 }
652
Li Zefan0ac801f2013-01-10 11:49:27 +0800653 key = css_set_hash(template);
Tejun Heo5abb8852013-06-12 21:04:49 -0700654 hash_for_each_possible(css_set_table, cset, hlist, key) {
655 if (!compare_css_sets(cset, old_cset, cgrp, template))
Paul Menage7717f7b2009-09-23 15:56:22 -0700656 continue;
657
658 /* This css_set matches what we need */
Tejun Heo5abb8852013-06-12 21:04:49 -0700659 return cset;
Li Zefan472b1052008-04-29 01:00:11 -0700660 }
Paul Menage817929e2007-10-18 23:39:36 -0700661
662 /* No existing cgroup group matched */
663 return NULL;
664}
665
Tejun Heo69d02062013-06-12 21:04:50 -0700666static void free_cgrp_cset_links(struct list_head *links_to_free)
Paul Menage817929e2007-10-18 23:39:36 -0700667{
Tejun Heo69d02062013-06-12 21:04:50 -0700668 struct cgrp_cset_link *link, *tmp_link;
KOSAKI Motohiro71cbb942008-07-25 01:46:55 -0700669
Tejun Heo69d02062013-06-12 21:04:50 -0700670 list_for_each_entry_safe(link, tmp_link, links_to_free, cset_link) {
671 list_del(&link->cset_link);
Paul Menage817929e2007-10-18 23:39:36 -0700672 kfree(link);
673 }
674}
675
Tejun Heo69d02062013-06-12 21:04:50 -0700676/**
677 * allocate_cgrp_cset_links - allocate cgrp_cset_links
678 * @count: the number of links to allocate
679 * @tmp_links: list_head the allocated links are put on
680 *
681 * Allocate @count cgrp_cset_link structures and chain them on @tmp_links
682 * through ->cset_link. Returns 0 on success or -errno.
Li Zefan36553432008-07-29 22:33:19 -0700683 */
Tejun Heo69d02062013-06-12 21:04:50 -0700684static int allocate_cgrp_cset_links(int count, struct list_head *tmp_links)
Li Zefan36553432008-07-29 22:33:19 -0700685{
Tejun Heo69d02062013-06-12 21:04:50 -0700686 struct cgrp_cset_link *link;
Li Zefan36553432008-07-29 22:33:19 -0700687 int i;
Tejun Heo69d02062013-06-12 21:04:50 -0700688
689 INIT_LIST_HEAD(tmp_links);
690
Li Zefan36553432008-07-29 22:33:19 -0700691 for (i = 0; i < count; i++) {
Tejun Heof4f4be22013-06-12 21:04:51 -0700692 link = kzalloc(sizeof(*link), GFP_KERNEL);
Li Zefan36553432008-07-29 22:33:19 -0700693 if (!link) {
Tejun Heo69d02062013-06-12 21:04:50 -0700694 free_cgrp_cset_links(tmp_links);
Li Zefan36553432008-07-29 22:33:19 -0700695 return -ENOMEM;
696 }
Tejun Heo69d02062013-06-12 21:04:50 -0700697 list_add(&link->cset_link, tmp_links);
Li Zefan36553432008-07-29 22:33:19 -0700698 }
699 return 0;
700}
701
Li Zefanc12f65d2009-01-07 18:07:42 -0800702/**
703 * link_css_set - a helper function to link a css_set to a cgroup
Tejun Heo69d02062013-06-12 21:04:50 -0700704 * @tmp_links: cgrp_cset_link objects allocated by allocate_cgrp_cset_links()
Tejun Heo5abb8852013-06-12 21:04:49 -0700705 * @cset: the css_set to be linked
Li Zefanc12f65d2009-01-07 18:07:42 -0800706 * @cgrp: the destination cgroup
707 */
Tejun Heo69d02062013-06-12 21:04:50 -0700708static void link_css_set(struct list_head *tmp_links, struct css_set *cset,
709 struct cgroup *cgrp)
Li Zefanc12f65d2009-01-07 18:07:42 -0800710{
Tejun Heo69d02062013-06-12 21:04:50 -0700711 struct cgrp_cset_link *link;
Li Zefanc12f65d2009-01-07 18:07:42 -0800712
Tejun Heo69d02062013-06-12 21:04:50 -0700713 BUG_ON(list_empty(tmp_links));
Tejun Heo6803c002014-04-23 11:13:16 -0400714
715 if (cgroup_on_dfl(cgrp))
716 cset->dfl_cgrp = cgrp;
717
Tejun Heo69d02062013-06-12 21:04:50 -0700718 link = list_first_entry(tmp_links, struct cgrp_cset_link, cset_link);
719 link->cset = cset;
Paul Menage7717f7b2009-09-23 15:56:22 -0700720 link->cgrp = cgrp;
Tejun Heo842b5972014-04-25 18:28:02 -0400721
722 if (list_empty(&cgrp->cset_links))
723 cgroup_update_populated(cgrp, true);
Tejun Heo69d02062013-06-12 21:04:50 -0700724 list_move(&link->cset_link, &cgrp->cset_links);
Tejun Heo842b5972014-04-25 18:28:02 -0400725
Paul Menage7717f7b2009-09-23 15:56:22 -0700726 /*
727 * Always add links to the tail of the list so that the list
728 * is sorted by order of hierarchy creation
729 */
Tejun Heo69d02062013-06-12 21:04:50 -0700730 list_add_tail(&link->cgrp_link, &cset->cgrp_links);
Li Zefanc12f65d2009-01-07 18:07:42 -0800731}
732
Tejun Heob326f9d2013-06-24 15:21:48 -0700733/**
734 * find_css_set - return a new css_set with one cgroup updated
735 * @old_cset: the baseline css_set
736 * @cgrp: the cgroup to be updated
737 *
738 * Return a new css_set that's equivalent to @old_cset, but with @cgrp
739 * substituted into the appropriate hierarchy.
Paul Menage817929e2007-10-18 23:39:36 -0700740 */
Tejun Heo5abb8852013-06-12 21:04:49 -0700741static struct css_set *find_css_set(struct css_set *old_cset,
742 struct cgroup *cgrp)
Paul Menage817929e2007-10-18 23:39:36 -0700743{
Tejun Heob326f9d2013-06-24 15:21:48 -0700744 struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT] = { };
Tejun Heo5abb8852013-06-12 21:04:49 -0700745 struct css_set *cset;
Tejun Heo69d02062013-06-12 21:04:50 -0700746 struct list_head tmp_links;
747 struct cgrp_cset_link *link;
Tejun Heo2d8f2432014-04-23 11:13:15 -0400748 struct cgroup_subsys *ss;
Li Zefan0ac801f2013-01-10 11:49:27 +0800749 unsigned long key;
Tejun Heo2d8f2432014-04-23 11:13:15 -0400750 int ssid;
Li Zefan472b1052008-04-29 01:00:11 -0700751
Tejun Heob326f9d2013-06-24 15:21:48 -0700752 lockdep_assert_held(&cgroup_mutex);
753
Paul Menage817929e2007-10-18 23:39:36 -0700754 /* First see if we already have a cgroup group that matches
755 * the desired set */
Tejun Heo96d365e2014-02-13 06:58:40 -0500756 down_read(&css_set_rwsem);
Tejun Heo5abb8852013-06-12 21:04:49 -0700757 cset = find_existing_css_set(old_cset, cgrp, template);
758 if (cset)
759 get_css_set(cset);
Tejun Heo96d365e2014-02-13 06:58:40 -0500760 up_read(&css_set_rwsem);
Paul Menage817929e2007-10-18 23:39:36 -0700761
Tejun Heo5abb8852013-06-12 21:04:49 -0700762 if (cset)
763 return cset;
Paul Menage817929e2007-10-18 23:39:36 -0700764
Tejun Heof4f4be22013-06-12 21:04:51 -0700765 cset = kzalloc(sizeof(*cset), GFP_KERNEL);
Tejun Heo5abb8852013-06-12 21:04:49 -0700766 if (!cset)
Paul Menage817929e2007-10-18 23:39:36 -0700767 return NULL;
768
Tejun Heo69d02062013-06-12 21:04:50 -0700769 /* Allocate all the cgrp_cset_link objects that we'll need */
Tejun Heo9871bf92013-06-24 15:21:47 -0700770 if (allocate_cgrp_cset_links(cgroup_root_count, &tmp_links) < 0) {
Tejun Heo5abb8852013-06-12 21:04:49 -0700771 kfree(cset);
Paul Menage817929e2007-10-18 23:39:36 -0700772 return NULL;
773 }
774
Tejun Heo5abb8852013-06-12 21:04:49 -0700775 atomic_set(&cset->refcount, 1);
Tejun Heo69d02062013-06-12 21:04:50 -0700776 INIT_LIST_HEAD(&cset->cgrp_links);
Tejun Heo5abb8852013-06-12 21:04:49 -0700777 INIT_LIST_HEAD(&cset->tasks);
Tejun Heoc7561122014-02-25 10:04:01 -0500778 INIT_LIST_HEAD(&cset->mg_tasks);
Tejun Heo1958d2d2014-02-25 10:04:03 -0500779 INIT_LIST_HEAD(&cset->mg_preload_node);
Tejun Heob3dc0942014-02-25 10:04:01 -0500780 INIT_LIST_HEAD(&cset->mg_node);
Tejun Heo5abb8852013-06-12 21:04:49 -0700781 INIT_HLIST_NODE(&cset->hlist);
Paul Menage817929e2007-10-18 23:39:36 -0700782
783 /* Copy the set of subsystem state objects generated in
784 * find_existing_css_set() */
Tejun Heo5abb8852013-06-12 21:04:49 -0700785 memcpy(cset->subsys, template, sizeof(cset->subsys));
Paul Menage817929e2007-10-18 23:39:36 -0700786
Tejun Heo96d365e2014-02-13 06:58:40 -0500787 down_write(&css_set_rwsem);
Paul Menage817929e2007-10-18 23:39:36 -0700788 /* Add reference counts and links from the new css_set. */
Tejun Heo69d02062013-06-12 21:04:50 -0700789 list_for_each_entry(link, &old_cset->cgrp_links, cgrp_link) {
Paul Menage7717f7b2009-09-23 15:56:22 -0700790 struct cgroup *c = link->cgrp;
Tejun Heo69d02062013-06-12 21:04:50 -0700791
Paul Menage7717f7b2009-09-23 15:56:22 -0700792 if (c->root == cgrp->root)
793 c = cgrp;
Tejun Heo69d02062013-06-12 21:04:50 -0700794 link_css_set(&tmp_links, cset, c);
Paul Menage7717f7b2009-09-23 15:56:22 -0700795 }
Paul Menage817929e2007-10-18 23:39:36 -0700796
Tejun Heo69d02062013-06-12 21:04:50 -0700797 BUG_ON(!list_empty(&tmp_links));
Paul Menage817929e2007-10-18 23:39:36 -0700798
Paul Menage817929e2007-10-18 23:39:36 -0700799 css_set_count++;
Li Zefan472b1052008-04-29 01:00:11 -0700800
Tejun Heo2d8f2432014-04-23 11:13:15 -0400801 /* Add @cset to the hash table */
Tejun Heo5abb8852013-06-12 21:04:49 -0700802 key = css_set_hash(cset->subsys);
803 hash_add(css_set_table, &cset->hlist, key);
Li Zefan472b1052008-04-29 01:00:11 -0700804
Tejun Heo2d8f2432014-04-23 11:13:15 -0400805 for_each_subsys(ss, ssid)
806 list_add_tail(&cset->e_cset_node[ssid],
807 &cset->subsys[ssid]->cgroup->e_csets[ssid]);
808
Tejun Heo96d365e2014-02-13 06:58:40 -0500809 up_write(&css_set_rwsem);
Paul Menage817929e2007-10-18 23:39:36 -0700810
Tejun Heo5abb8852013-06-12 21:04:49 -0700811 return cset;
Paul Menageb4f48b62007-10-18 23:39:33 -0700812}
813
Tejun Heo3dd06ff2014-03-19 10:23:54 -0400814static struct cgroup_root *cgroup_root_from_kf(struct kernfs_root *kf_root)
Paul Menage7717f7b2009-09-23 15:56:22 -0700815{
Tejun Heo3dd06ff2014-03-19 10:23:54 -0400816 struct cgroup *root_cgrp = kf_root->kn->priv;
Tejun Heo2bd59d42014-02-11 11:52:49 -0500817
Tejun Heo3dd06ff2014-03-19 10:23:54 -0400818 return root_cgrp->root;
Tejun Heo2bd59d42014-02-11 11:52:49 -0500819}
820
Tejun Heo3dd06ff2014-03-19 10:23:54 -0400821static int cgroup_init_root_id(struct cgroup_root *root)
Tejun Heof2e85d52014-02-11 11:52:49 -0500822{
823 int id;
824
825 lockdep_assert_held(&cgroup_mutex);
826
Tejun Heo985ed672014-03-19 10:23:53 -0400827 id = idr_alloc_cyclic(&cgroup_hierarchy_idr, root, 0, 0, GFP_KERNEL);
Tejun Heof2e85d52014-02-11 11:52:49 -0500828 if (id < 0)
829 return id;
830
831 root->hierarchy_id = id;
832 return 0;
833}
834
Tejun Heo3dd06ff2014-03-19 10:23:54 -0400835static void cgroup_exit_root_id(struct cgroup_root *root)
Tejun Heof2e85d52014-02-11 11:52:49 -0500836{
837 lockdep_assert_held(&cgroup_mutex);
838
839 if (root->hierarchy_id) {
840 idr_remove(&cgroup_hierarchy_idr, root->hierarchy_id);
841 root->hierarchy_id = 0;
842 }
843}
844
Tejun Heo3dd06ff2014-03-19 10:23:54 -0400845static void cgroup_free_root(struct cgroup_root *root)
Tejun Heof2e85d52014-02-11 11:52:49 -0500846{
847 if (root) {
848 /* hierarhcy ID shoulid already have been released */
849 WARN_ON_ONCE(root->hierarchy_id);
850
851 idr_destroy(&root->cgroup_idr);
852 kfree(root);
853 }
854}
855
Tejun Heo3dd06ff2014-03-19 10:23:54 -0400856static void cgroup_destroy_root(struct cgroup_root *root)
Tejun Heo59f52962014-02-11 11:52:49 -0500857{
Tejun Heo3dd06ff2014-03-19 10:23:54 -0400858 struct cgroup *cgrp = &root->cgrp;
Tejun Heof2e85d52014-02-11 11:52:49 -0500859 struct cgrp_cset_link *link, *tmp_link;
Tejun Heof2e85d52014-02-11 11:52:49 -0500860
Tejun Heo2bd59d42014-02-11 11:52:49 -0500861 mutex_lock(&cgroup_mutex);
Tejun Heof2e85d52014-02-11 11:52:49 -0500862
Tejun Heo776f02f2014-02-12 09:29:50 -0500863 BUG_ON(atomic_read(&root->nr_cgrps));
Tejun Heod5c419b2014-05-16 13:22:48 -0400864 BUG_ON(!list_empty(&cgrp->self.children));
Tejun Heof2e85d52014-02-11 11:52:49 -0500865
Tejun Heof2e85d52014-02-11 11:52:49 -0500866 /* Rebind all subsystems back to the default hierarchy */
Tejun Heof392e512014-04-23 11:13:14 -0400867 rebind_subsystems(&cgrp_dfl_root, root->subsys_mask);
Tejun Heof2e85d52014-02-11 11:52:49 -0500868
869 /*
870 * Release all the links from cset_links to this hierarchy's
871 * root cgroup
872 */
Tejun Heo96d365e2014-02-13 06:58:40 -0500873 down_write(&css_set_rwsem);
Tejun Heof2e85d52014-02-11 11:52:49 -0500874
875 list_for_each_entry_safe(link, tmp_link, &cgrp->cset_links, cset_link) {
876 list_del(&link->cset_link);
877 list_del(&link->cgrp_link);
878 kfree(link);
879 }
Tejun Heo96d365e2014-02-13 06:58:40 -0500880 up_write(&css_set_rwsem);
Tejun Heof2e85d52014-02-11 11:52:49 -0500881
882 if (!list_empty(&root->root_list)) {
883 list_del(&root->root_list);
884 cgroup_root_count--;
885 }
886
887 cgroup_exit_root_id(root);
888
889 mutex_unlock(&cgroup_mutex);
Tejun Heof2e85d52014-02-11 11:52:49 -0500890
Tejun Heo2bd59d42014-02-11 11:52:49 -0500891 kernfs_destroy_root(root->kf_root);
Tejun Heof2e85d52014-02-11 11:52:49 -0500892 cgroup_free_root(root);
893}
894
Tejun Heoceb6a082014-02-25 10:04:02 -0500895/* look up cgroup associated with given css_set on the specified hierarchy */
896static struct cgroup *cset_cgroup_from_root(struct css_set *cset,
Tejun Heo3dd06ff2014-03-19 10:23:54 -0400897 struct cgroup_root *root)
Paul Menage7717f7b2009-09-23 15:56:22 -0700898{
Paul Menage7717f7b2009-09-23 15:56:22 -0700899 struct cgroup *res = NULL;
900
Tejun Heo96d365e2014-02-13 06:58:40 -0500901 lockdep_assert_held(&cgroup_mutex);
902 lockdep_assert_held(&css_set_rwsem);
903
Tejun Heo5abb8852013-06-12 21:04:49 -0700904 if (cset == &init_css_set) {
Tejun Heo3dd06ff2014-03-19 10:23:54 -0400905 res = &root->cgrp;
Paul Menage7717f7b2009-09-23 15:56:22 -0700906 } else {
Tejun Heo69d02062013-06-12 21:04:50 -0700907 struct cgrp_cset_link *link;
908
909 list_for_each_entry(link, &cset->cgrp_links, cgrp_link) {
Paul Menage7717f7b2009-09-23 15:56:22 -0700910 struct cgroup *c = link->cgrp;
Tejun Heo69d02062013-06-12 21:04:50 -0700911
Paul Menage7717f7b2009-09-23 15:56:22 -0700912 if (c->root == root) {
913 res = c;
914 break;
915 }
916 }
917 }
Tejun Heo96d365e2014-02-13 06:58:40 -0500918
Paul Menage7717f7b2009-09-23 15:56:22 -0700919 BUG_ON(!res);
920 return res;
921}
922
923/*
Tejun Heoceb6a082014-02-25 10:04:02 -0500924 * Return the cgroup for "task" from the given hierarchy. Must be
925 * called with cgroup_mutex and css_set_rwsem held.
926 */
927static struct cgroup *task_cgroup_from_root(struct task_struct *task,
Tejun Heo3dd06ff2014-03-19 10:23:54 -0400928 struct cgroup_root *root)
Tejun Heoceb6a082014-02-25 10:04:02 -0500929{
930 /*
931 * No need to lock the task - since we hold cgroup_mutex the
932 * task can't change groups, so the only thing that can happen
933 * is that it exits and its css is set back to init_css_set.
934 */
935 return cset_cgroup_from_root(task_css_set(task), root);
936}
937
938/*
Paul Menageddbcc7e2007-10-18 23:39:30 -0700939 * A task must hold cgroup_mutex to modify cgroups.
940 *
941 * Any task can increment and decrement the count field without lock.
942 * So in general, code holding cgroup_mutex can't rely on the count
943 * field not changing. However, if the count goes to zero, then only
Cliff Wickman956db3c2008-02-07 00:14:43 -0800944 * cgroup_attach_task() can increment it again. Because a count of zero
Paul Menageddbcc7e2007-10-18 23:39:30 -0700945 * means that no tasks are currently attached, therefore there is no
946 * way a task attached to that cgroup can fork (the other way to
947 * increment the count). So code holding cgroup_mutex can safely
948 * assume that if the count is zero, it will stay zero. Similarly, if
949 * a task holds cgroup_mutex on a cgroup with zero count, it
950 * knows that the cgroup won't be removed, as cgroup_rmdir()
951 * needs that mutex.
952 *
Paul Menageddbcc7e2007-10-18 23:39:30 -0700953 * A cgroup can only be deleted if both its 'count' of using tasks
954 * is zero, and its list of 'children' cgroups is empty. Since all
955 * tasks in the system use _some_ cgroup, and since there is always at
Tejun Heo3dd06ff2014-03-19 10:23:54 -0400956 * least one task in the system (init, pid == 1), therefore, root cgroup
Paul Menageddbcc7e2007-10-18 23:39:30 -0700957 * always has either children cgroups and/or using tasks. So we don't
Tejun Heo3dd06ff2014-03-19 10:23:54 -0400958 * need a special hack to ensure that root cgroup cannot be deleted.
Paul Menageddbcc7e2007-10-18 23:39:30 -0700959 *
960 * P.S. One more locking exception. RCU is used to guard the
Cliff Wickman956db3c2008-02-07 00:14:43 -0800961 * update of a tasks cgroup pointer by cgroup_attach_task()
Paul Menageddbcc7e2007-10-18 23:39:30 -0700962 */
963
Tejun Heo69dfa002014-05-04 15:09:13 -0400964static int cgroup_populate_dir(struct cgroup *cgrp, unsigned int subsys_mask);
Tejun Heo2bd59d42014-02-11 11:52:49 -0500965static struct kernfs_syscall_ops cgroup_kf_syscall_ops;
Alexey Dobriyan828c0952009-10-01 15:43:56 -0700966static const struct file_operations proc_cgroupstats_operations;
Paul Menagea4243162007-10-18 23:39:35 -0700967
Tejun Heo8d7e6fb2014-02-11 11:52:48 -0500968static char *cgroup_file_name(struct cgroup *cgrp, const struct cftype *cft,
969 char *buf)
Paul Menageddbcc7e2007-10-18 23:39:30 -0700970{
Tejun Heo8d7e6fb2014-02-11 11:52:48 -0500971 if (cft->ss && !(cft->flags & CFTYPE_NO_PREFIX) &&
972 !(cgrp->root->flags & CGRP_ROOT_NOPREFIX))
973 snprintf(buf, CGROUP_FILE_NAME_MAX, "%s.%s",
974 cft->ss->name, cft->name);
975 else
976 strncpy(buf, cft->name, CGROUP_FILE_NAME_MAX);
977 return buf;
Paul Menageddbcc7e2007-10-18 23:39:30 -0700978}
979
Tejun Heof2e85d52014-02-11 11:52:49 -0500980/**
981 * cgroup_file_mode - deduce file mode of a control file
982 * @cft: the control file in question
983 *
984 * returns cft->mode if ->mode is not 0
985 * returns S_IRUGO|S_IWUSR if it has both a read and a write handler
986 * returns S_IRUGO if it has only a read handler
987 * returns S_IWUSR if it has only a write hander
988 */
989static umode_t cgroup_file_mode(const struct cftype *cft)
Li Zefan65dff752013-03-01 15:01:56 +0800990{
Tejun Heof2e85d52014-02-11 11:52:49 -0500991 umode_t mode = 0;
Li Zefan65dff752013-03-01 15:01:56 +0800992
Tejun Heof2e85d52014-02-11 11:52:49 -0500993 if (cft->mode)
994 return cft->mode;
995
996 if (cft->read_u64 || cft->read_s64 || cft->seq_show)
997 mode |= S_IRUGO;
998
Tejun Heo6770c642014-05-13 12:16:21 -0400999 if (cft->write_u64 || cft->write_s64 || cft->write)
Tejun Heof2e85d52014-02-11 11:52:49 -05001000 mode |= S_IWUSR;
1001
1002 return mode;
Li Zefan65dff752013-03-01 15:01:56 +08001003}
1004
Tejun Heo59f52962014-02-11 11:52:49 -05001005static void cgroup_get(struct cgroup *cgrp)
Paul Menageddbcc7e2007-10-18 23:39:30 -07001006{
Tejun Heo2bd59d42014-02-11 11:52:49 -05001007 WARN_ON_ONCE(cgroup_is_dead(cgrp));
Tejun Heo9d755d32014-05-14 09:15:02 -04001008 css_get(&cgrp->self);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001009}
1010
Li Zefanaa323622014-09-04 14:43:38 +08001011static bool cgroup_tryget(struct cgroup *cgrp)
1012{
1013 return css_tryget(&cgrp->self);
1014}
1015
Tejun Heo59f52962014-02-11 11:52:49 -05001016static void cgroup_put(struct cgroup *cgrp)
Paul Menageddbcc7e2007-10-18 23:39:30 -07001017{
Tejun Heo9d755d32014-05-14 09:15:02 -04001018 css_put(&cgrp->self);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001019}
1020
Tejun Heoa9746d82014-05-13 12:19:22 -04001021/**
Tejun Heo0f060de2014-11-18 02:49:50 -05001022 * cgroup_calc_child_subsys_mask - calculate child_subsys_mask
Tejun Heoaf0ba672014-07-08 18:02:57 -04001023 * @cgrp: the target cgroup
Tejun Heo0f060de2014-11-18 02:49:50 -05001024 * @subtree_control: the new subtree_control mask to consider
Tejun Heoaf0ba672014-07-08 18:02:57 -04001025 *
1026 * On the default hierarchy, a subsystem may request other subsystems to be
1027 * enabled together through its ->depends_on mask. In such cases, more
1028 * subsystems than specified in "cgroup.subtree_control" may be enabled.
1029 *
Tejun Heo0f060de2014-11-18 02:49:50 -05001030 * This function calculates which subsystems need to be enabled if
1031 * @subtree_control is to be applied to @cgrp. The returned mask is always
1032 * a superset of @subtree_control and follows the usual hierarchy rules.
Tejun Heoaf0ba672014-07-08 18:02:57 -04001033 */
Tejun Heo0f060de2014-11-18 02:49:50 -05001034static unsigned int cgroup_calc_child_subsys_mask(struct cgroup *cgrp,
1035 unsigned int subtree_control)
Tejun Heo667c2492014-07-08 18:02:56 -04001036{
Tejun Heoaf0ba672014-07-08 18:02:57 -04001037 struct cgroup *parent = cgroup_parent(cgrp);
Tejun Heo0f060de2014-11-18 02:49:50 -05001038 unsigned int cur_ss_mask = subtree_control;
Tejun Heoaf0ba672014-07-08 18:02:57 -04001039 struct cgroup_subsys *ss;
1040 int ssid;
1041
1042 lockdep_assert_held(&cgroup_mutex);
1043
Tejun Heo0f060de2014-11-18 02:49:50 -05001044 if (!cgroup_on_dfl(cgrp))
1045 return cur_ss_mask;
Tejun Heoaf0ba672014-07-08 18:02:57 -04001046
1047 while (true) {
1048 unsigned int new_ss_mask = cur_ss_mask;
1049
1050 for_each_subsys(ss, ssid)
1051 if (cur_ss_mask & (1 << ssid))
1052 new_ss_mask |= ss->depends_on;
1053
1054 /*
1055 * Mask out subsystems which aren't available. This can
1056 * happen only if some depended-upon subsystems were bound
1057 * to non-default hierarchies.
1058 */
1059 if (parent)
1060 new_ss_mask &= parent->child_subsys_mask;
1061 else
1062 new_ss_mask &= cgrp->root->subsys_mask;
1063
1064 if (new_ss_mask == cur_ss_mask)
1065 break;
1066 cur_ss_mask = new_ss_mask;
1067 }
1068
Tejun Heo0f060de2014-11-18 02:49:50 -05001069 return cur_ss_mask;
1070}
1071
1072/**
1073 * cgroup_refresh_child_subsys_mask - update child_subsys_mask
1074 * @cgrp: the target cgroup
1075 *
1076 * Update @cgrp->child_subsys_mask according to the current
1077 * @cgrp->subtree_control using cgroup_calc_child_subsys_mask().
1078 */
1079static void cgroup_refresh_child_subsys_mask(struct cgroup *cgrp)
1080{
1081 cgrp->child_subsys_mask =
1082 cgroup_calc_child_subsys_mask(cgrp, cgrp->subtree_control);
Tejun Heo667c2492014-07-08 18:02:56 -04001083}
1084
Tejun Heoa9746d82014-05-13 12:19:22 -04001085/**
1086 * cgroup_kn_unlock - unlocking helper for cgroup kernfs methods
1087 * @kn: the kernfs_node being serviced
1088 *
1089 * This helper undoes cgroup_kn_lock_live() and should be invoked before
1090 * the method finishes if locking succeeded. Note that once this function
1091 * returns the cgroup returned by cgroup_kn_lock_live() may become
1092 * inaccessible any time. If the caller intends to continue to access the
1093 * cgroup, it should pin it before invoking this function.
1094 */
1095static void cgroup_kn_unlock(struct kernfs_node *kn)
1096{
1097 struct cgroup *cgrp;
1098
1099 if (kernfs_type(kn) == KERNFS_DIR)
1100 cgrp = kn->priv;
1101 else
1102 cgrp = kn->parent->priv;
1103
1104 mutex_unlock(&cgroup_mutex);
Tejun Heoa9746d82014-05-13 12:19:22 -04001105
1106 kernfs_unbreak_active_protection(kn);
1107 cgroup_put(cgrp);
1108}
1109
1110/**
1111 * cgroup_kn_lock_live - locking helper for cgroup kernfs methods
1112 * @kn: the kernfs_node being serviced
1113 *
1114 * This helper is to be used by a cgroup kernfs method currently servicing
1115 * @kn. It breaks the active protection, performs cgroup locking and
1116 * verifies that the associated cgroup is alive. Returns the cgroup if
1117 * alive; otherwise, %NULL. A successful return should be undone by a
1118 * matching cgroup_kn_unlock() invocation.
1119 *
1120 * Any cgroup kernfs method implementation which requires locking the
1121 * associated cgroup should use this helper. It avoids nesting cgroup
1122 * locking under kernfs active protection and allows all kernfs operations
1123 * including self-removal.
1124 */
1125static struct cgroup *cgroup_kn_lock_live(struct kernfs_node *kn)
1126{
1127 struct cgroup *cgrp;
1128
1129 if (kernfs_type(kn) == KERNFS_DIR)
1130 cgrp = kn->priv;
1131 else
1132 cgrp = kn->parent->priv;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001133
1134 /*
Tejun Heo01f64742014-05-13 12:19:23 -04001135 * We're gonna grab cgroup_mutex which nests outside kernfs
Tejun Heoa9746d82014-05-13 12:19:22 -04001136 * active_ref. cgroup liveliness check alone provides enough
1137 * protection against removal. Ensure @cgrp stays accessible and
1138 * break the active_ref protection.
Paul Menageddbcc7e2007-10-18 23:39:30 -07001139 */
Li Zefanaa323622014-09-04 14:43:38 +08001140 if (!cgroup_tryget(cgrp))
1141 return NULL;
Tejun Heoa9746d82014-05-13 12:19:22 -04001142 kernfs_break_active_protection(kn);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001143
Tejun Heoa9746d82014-05-13 12:19:22 -04001144 mutex_lock(&cgroup_mutex);
1145
1146 if (!cgroup_is_dead(cgrp))
1147 return cgrp;
1148
1149 cgroup_kn_unlock(kn);
1150 return NULL;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001151}
1152
Li Zefan2739d3c2013-01-21 18:18:33 +08001153static void cgroup_rm_file(struct cgroup *cgrp, const struct cftype *cft)
Paul Menageddbcc7e2007-10-18 23:39:30 -07001154{
Tejun Heo2bd59d42014-02-11 11:52:49 -05001155 char name[CGROUP_FILE_NAME_MAX];
Paul Menageddbcc7e2007-10-18 23:39:30 -07001156
Tejun Heo01f64742014-05-13 12:19:23 -04001157 lockdep_assert_held(&cgroup_mutex);
Tejun Heo2bd59d42014-02-11 11:52:49 -05001158 kernfs_remove_by_name(cgrp->kn, cgroup_file_name(cgrp, cft, name));
Tejun Heo05ef1d72012-04-01 12:09:56 -07001159}
1160
Aristeu Rozanski13af07d2012-08-23 16:53:29 -04001161/**
Tejun Heo628f7cd2013-06-28 16:24:11 -07001162 * cgroup_clear_dir - remove subsys files in a cgroup directory
Tejun Heo8f891402013-06-28 16:24:10 -07001163 * @cgrp: target cgroup
Aristeu Rozanski13af07d2012-08-23 16:53:29 -04001164 * @subsys_mask: mask of the subsystem ids whose files should be removed
1165 */
Tejun Heo69dfa002014-05-04 15:09:13 -04001166static void cgroup_clear_dir(struct cgroup *cgrp, unsigned int subsys_mask)
Tejun Heo05ef1d72012-04-01 12:09:56 -07001167{
Aristeu Rozanski13af07d2012-08-23 16:53:29 -04001168 struct cgroup_subsys *ss;
Tejun Heob420ba72013-07-12 12:34:02 -07001169 int i;
Tejun Heo05ef1d72012-04-01 12:09:56 -07001170
Tejun Heob420ba72013-07-12 12:34:02 -07001171 for_each_subsys(ss, i) {
Tejun Heo0adb0702014-02-12 09:29:48 -05001172 struct cftype *cfts;
Tejun Heob420ba72013-07-12 12:34:02 -07001173
Tejun Heo69dfa002014-05-04 15:09:13 -04001174 if (!(subsys_mask & (1 << i)))
Aristeu Rozanski13af07d2012-08-23 16:53:29 -04001175 continue;
Tejun Heo0adb0702014-02-12 09:29:48 -05001176 list_for_each_entry(cfts, &ss->cfts, node)
1177 cgroup_addrm_files(cgrp, cfts, false);
Aristeu Rozanski13af07d2012-08-23 16:53:29 -04001178 }
Paul Menageddbcc7e2007-10-18 23:39:30 -07001179}
1180
Tejun Heo69dfa002014-05-04 15:09:13 -04001181static int rebind_subsystems(struct cgroup_root *dst_root, unsigned int ss_mask)
Paul Menageddbcc7e2007-10-18 23:39:30 -07001182{
Tejun Heo30159ec2013-06-25 11:53:37 -07001183 struct cgroup_subsys *ss;
Tejun Heo5533e012014-05-14 19:33:07 -04001184 unsigned int tmp_ss_mask;
Tejun Heo2d8f2432014-04-23 11:13:15 -04001185 int ssid, i, ret;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001186
Tejun Heoace2bee2014-02-11 11:52:47 -05001187 lockdep_assert_held(&cgroup_mutex);
Ben Blumaae8aab2010-03-10 15:22:07 -08001188
Tejun Heo5df36032014-03-19 10:23:54 -04001189 for_each_subsys(ss, ssid) {
1190 if (!(ss_mask & (1 << ssid)))
Paul Menageddbcc7e2007-10-18 23:39:30 -07001191 continue;
Tejun Heo30159ec2013-06-25 11:53:37 -07001192
Tejun Heo7fd8c562014-04-23 11:13:16 -04001193 /* if @ss has non-root csses attached to it, can't move */
1194 if (css_next_child(NULL, cgroup_css(&ss->root->cgrp, ss)))
Tejun Heo3ed80a62014-02-08 10:36:58 -05001195 return -EBUSY;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001196
Tejun Heo5df36032014-03-19 10:23:54 -04001197 /* can't move between two non-dummy roots either */
Tejun Heo7fd8c562014-04-23 11:13:16 -04001198 if (ss->root != &cgrp_dfl_root && dst_root != &cgrp_dfl_root)
Tejun Heo5df36032014-03-19 10:23:54 -04001199 return -EBUSY;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001200 }
1201
Tejun Heo5533e012014-05-14 19:33:07 -04001202 /* skip creating root files on dfl_root for inhibited subsystems */
1203 tmp_ss_mask = ss_mask;
1204 if (dst_root == &cgrp_dfl_root)
1205 tmp_ss_mask &= ~cgrp_dfl_root_inhibit_ss_mask;
1206
1207 ret = cgroup_populate_dir(&dst_root->cgrp, tmp_ss_mask);
Tejun Heoa2dd4242014-03-19 10:23:55 -04001208 if (ret) {
1209 if (dst_root != &cgrp_dfl_root)
Tejun Heo5df36032014-03-19 10:23:54 -04001210 return ret;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001211
Tejun Heoa2dd4242014-03-19 10:23:55 -04001212 /*
1213 * Rebinding back to the default root is not allowed to
1214 * fail. Using both default and non-default roots should
1215 * be rare. Moving subsystems back and forth even more so.
1216 * Just warn about it and continue.
1217 */
1218 if (cgrp_dfl_root_visible) {
Tejun Heo69dfa002014-05-04 15:09:13 -04001219 pr_warn("failed to create files (%d) while rebinding 0x%x to default root\n",
Jianyu Zhana2a1f9e2014-04-25 18:28:03 -04001220 ret, ss_mask);
Joe Perchesed3d2612014-04-25 18:28:03 -04001221 pr_warn("you may retry by moving them to a different hierarchy and unbinding\n");
Tejun Heoa2dd4242014-03-19 10:23:55 -04001222 }
Tejun Heo5df36032014-03-19 10:23:54 -04001223 }
Tejun Heo31261212013-06-28 17:07:30 -07001224
1225 /*
1226 * Nothing can fail from this point on. Remove files for the
1227 * removed subsystems and rebind each subsystem.
1228 */
Tejun Heo5df36032014-03-19 10:23:54 -04001229 for_each_subsys(ss, ssid)
Tejun Heoa2dd4242014-03-19 10:23:55 -04001230 if (ss_mask & (1 << ssid))
Tejun Heo3dd06ff2014-03-19 10:23:54 -04001231 cgroup_clear_dir(&ss->root->cgrp, 1 << ssid);
Tejun Heo31261212013-06-28 17:07:30 -07001232
Tejun Heo5df36032014-03-19 10:23:54 -04001233 for_each_subsys(ss, ssid) {
Tejun Heo3dd06ff2014-03-19 10:23:54 -04001234 struct cgroup_root *src_root;
Tejun Heo5df36032014-03-19 10:23:54 -04001235 struct cgroup_subsys_state *css;
Tejun Heo2d8f2432014-04-23 11:13:15 -04001236 struct css_set *cset;
Tejun Heo30159ec2013-06-25 11:53:37 -07001237
Tejun Heo5df36032014-03-19 10:23:54 -04001238 if (!(ss_mask & (1 << ssid)))
1239 continue;
Tejun Heoa8a648c2013-06-24 15:21:47 -07001240
Tejun Heo5df36032014-03-19 10:23:54 -04001241 src_root = ss->root;
Tejun Heo3dd06ff2014-03-19 10:23:54 -04001242 css = cgroup_css(&src_root->cgrp, ss);
Tejun Heo73e80ed2013-08-13 11:01:55 -04001243
Tejun Heo3dd06ff2014-03-19 10:23:54 -04001244 WARN_ON(!css || cgroup_css(&dst_root->cgrp, ss));
Tejun Heoa8a648c2013-06-24 15:21:47 -07001245
Tejun Heo3dd06ff2014-03-19 10:23:54 -04001246 RCU_INIT_POINTER(src_root->cgrp.subsys[ssid], NULL);
1247 rcu_assign_pointer(dst_root->cgrp.subsys[ssid], css);
Tejun Heo5df36032014-03-19 10:23:54 -04001248 ss->root = dst_root;
Tejun Heo3dd06ff2014-03-19 10:23:54 -04001249 css->cgroup = &dst_root->cgrp;
Tejun Heoa8a648c2013-06-24 15:21:47 -07001250
Tejun Heo2d8f2432014-04-23 11:13:15 -04001251 down_write(&css_set_rwsem);
1252 hash_for_each(css_set_table, i, cset, hlist)
1253 list_move_tail(&cset->e_cset_node[ss->id],
1254 &dst_root->cgrp.e_csets[ss->id]);
1255 up_write(&css_set_rwsem);
1256
Tejun Heof392e512014-04-23 11:13:14 -04001257 src_root->subsys_mask &= ~(1 << ssid);
Tejun Heo667c2492014-07-08 18:02:56 -04001258 src_root->cgrp.subtree_control &= ~(1 << ssid);
1259 cgroup_refresh_child_subsys_mask(&src_root->cgrp);
Tejun Heof392e512014-04-23 11:13:14 -04001260
Tejun Heobd53d612014-04-23 11:13:16 -04001261 /* default hierarchy doesn't enable controllers by default */
Tejun Heof392e512014-04-23 11:13:14 -04001262 dst_root->subsys_mask |= 1 << ssid;
Tejun Heo667c2492014-07-08 18:02:56 -04001263 if (dst_root != &cgrp_dfl_root) {
1264 dst_root->cgrp.subtree_control |= 1 << ssid;
1265 cgroup_refresh_child_subsys_mask(&dst_root->cgrp);
1266 }
Tejun Heo73e80ed2013-08-13 11:01:55 -04001267
Tejun Heo5df36032014-03-19 10:23:54 -04001268 if (ss->bind)
1269 ss->bind(css);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001270 }
Paul Menageddbcc7e2007-10-18 23:39:30 -07001271
Tejun Heoa2dd4242014-03-19 10:23:55 -04001272 kernfs_activate(dst_root->cgrp.kn);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001273 return 0;
1274}
1275
Tejun Heo2bd59d42014-02-11 11:52:49 -05001276static int cgroup_show_options(struct seq_file *seq,
1277 struct kernfs_root *kf_root)
Paul Menageddbcc7e2007-10-18 23:39:30 -07001278{
Tejun Heo3dd06ff2014-03-19 10:23:54 -04001279 struct cgroup_root *root = cgroup_root_from_kf(kf_root);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001280 struct cgroup_subsys *ss;
Tejun Heob85d2042013-12-06 15:11:57 -05001281 int ssid;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001282
Tejun Heob85d2042013-12-06 15:11:57 -05001283 for_each_subsys(ss, ssid)
Tejun Heof392e512014-04-23 11:13:14 -04001284 if (root->subsys_mask & (1 << ssid))
Tejun Heob85d2042013-12-06 15:11:57 -05001285 seq_printf(seq, ",%s", ss->name);
Tejun Heo93438622013-04-14 20:15:25 -07001286 if (root->flags & CGRP_ROOT_NOPREFIX)
Paul Menageddbcc7e2007-10-18 23:39:30 -07001287 seq_puts(seq, ",noprefix");
Tejun Heo93438622013-04-14 20:15:25 -07001288 if (root->flags & CGRP_ROOT_XATTR)
Aristeu Rozanski03b1cde2012-08-23 16:53:30 -04001289 seq_puts(seq, ",xattr");
Tejun Heo69e943b2014-02-08 10:36:58 -05001290
1291 spin_lock(&release_agent_path_lock);
Paul Menage81a6a5c2007-10-18 23:39:38 -07001292 if (strlen(root->release_agent_path))
1293 seq_printf(seq, ",release_agent=%s", root->release_agent_path);
Tejun Heo69e943b2014-02-08 10:36:58 -05001294 spin_unlock(&release_agent_path_lock);
1295
Tejun Heo3dd06ff2014-03-19 10:23:54 -04001296 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags))
Daniel Lezcano97978e62010-10-27 15:33:35 -07001297 seq_puts(seq, ",clone_children");
Paul Menagec6d57f32009-09-23 15:56:19 -07001298 if (strlen(root->name))
1299 seq_printf(seq, ",name=%s", root->name);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001300 return 0;
1301}
1302
1303struct cgroup_sb_opts {
Tejun Heo69dfa002014-05-04 15:09:13 -04001304 unsigned int subsys_mask;
1305 unsigned int flags;
Paul Menage81a6a5c2007-10-18 23:39:38 -07001306 char *release_agent;
Tejun Heo2260e7f2012-11-19 08:13:38 -08001307 bool cpuset_clone_children;
Paul Menagec6d57f32009-09-23 15:56:19 -07001308 char *name;
Paul Menage2c6ab6d2009-09-23 15:56:23 -07001309 /* User explicitly requested empty subsystem */
1310 bool none;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001311};
1312
Ben Blumcf5d5942010-03-10 15:22:09 -08001313static int parse_cgroupfs_options(char *data, struct cgroup_sb_opts *opts)
Paul Menageddbcc7e2007-10-18 23:39:30 -07001314{
Daniel Lezcano32a8cf232010-10-27 15:33:37 -07001315 char *token, *o = data;
1316 bool all_ss = false, one_ss = false;
Tejun Heo69dfa002014-05-04 15:09:13 -04001317 unsigned int mask = -1U;
Tejun Heo30159ec2013-06-25 11:53:37 -07001318 struct cgroup_subsys *ss;
Tejun Heo7b9a6ba2014-07-09 10:08:08 -04001319 int nr_opts = 0;
Tejun Heo30159ec2013-06-25 11:53:37 -07001320 int i;
Li Zefanf9ab5b52009-06-17 16:26:33 -07001321
1322#ifdef CONFIG_CPUSETS
Tejun Heo69dfa002014-05-04 15:09:13 -04001323 mask = ~(1U << cpuset_cgrp_id);
Li Zefanf9ab5b52009-06-17 16:26:33 -07001324#endif
Paul Menageddbcc7e2007-10-18 23:39:30 -07001325
Paul Menagec6d57f32009-09-23 15:56:19 -07001326 memset(opts, 0, sizeof(*opts));
Paul Menageddbcc7e2007-10-18 23:39:30 -07001327
1328 while ((token = strsep(&o, ",")) != NULL) {
Tejun Heo7b9a6ba2014-07-09 10:08:08 -04001329 nr_opts++;
1330
Paul Menageddbcc7e2007-10-18 23:39:30 -07001331 if (!*token)
1332 return -EINVAL;
Daniel Lezcano32a8cf232010-10-27 15:33:37 -07001333 if (!strcmp(token, "none")) {
Paul Menage2c6ab6d2009-09-23 15:56:23 -07001334 /* Explicitly have no subsystems */
1335 opts->none = true;
Daniel Lezcano32a8cf232010-10-27 15:33:37 -07001336 continue;
1337 }
1338 if (!strcmp(token, "all")) {
1339 /* Mutually exclusive option 'all' + subsystem name */
1340 if (one_ss)
1341 return -EINVAL;
1342 all_ss = true;
1343 continue;
1344 }
Tejun Heo873fe092013-04-14 20:15:26 -07001345 if (!strcmp(token, "__DEVEL__sane_behavior")) {
1346 opts->flags |= CGRP_ROOT_SANE_BEHAVIOR;
1347 continue;
1348 }
Daniel Lezcano32a8cf232010-10-27 15:33:37 -07001349 if (!strcmp(token, "noprefix")) {
Tejun Heo93438622013-04-14 20:15:25 -07001350 opts->flags |= CGRP_ROOT_NOPREFIX;
Daniel Lezcano32a8cf232010-10-27 15:33:37 -07001351 continue;
1352 }
1353 if (!strcmp(token, "clone_children")) {
Tejun Heo2260e7f2012-11-19 08:13:38 -08001354 opts->cpuset_clone_children = true;
Daniel Lezcano32a8cf232010-10-27 15:33:37 -07001355 continue;
1356 }
Aristeu Rozanski03b1cde2012-08-23 16:53:30 -04001357 if (!strcmp(token, "xattr")) {
Tejun Heo93438622013-04-14 20:15:25 -07001358 opts->flags |= CGRP_ROOT_XATTR;
Aristeu Rozanski03b1cde2012-08-23 16:53:30 -04001359 continue;
1360 }
Daniel Lezcano32a8cf232010-10-27 15:33:37 -07001361 if (!strncmp(token, "release_agent=", 14)) {
Paul Menage81a6a5c2007-10-18 23:39:38 -07001362 /* Specifying two release agents is forbidden */
1363 if (opts->release_agent)
1364 return -EINVAL;
Paul Menagec6d57f32009-09-23 15:56:19 -07001365 opts->release_agent =
Dan Carpentere400c282010-08-10 18:02:54 -07001366 kstrndup(token + 14, PATH_MAX - 1, GFP_KERNEL);
Paul Menage81a6a5c2007-10-18 23:39:38 -07001367 if (!opts->release_agent)
1368 return -ENOMEM;
Daniel Lezcano32a8cf232010-10-27 15:33:37 -07001369 continue;
1370 }
1371 if (!strncmp(token, "name=", 5)) {
Paul Menagec6d57f32009-09-23 15:56:19 -07001372 const char *name = token + 5;
1373 /* Can't specify an empty name */
1374 if (!strlen(name))
1375 return -EINVAL;
1376 /* Must match [\w.-]+ */
1377 for (i = 0; i < strlen(name); i++) {
1378 char c = name[i];
1379 if (isalnum(c))
1380 continue;
1381 if ((c == '.') || (c == '-') || (c == '_'))
1382 continue;
1383 return -EINVAL;
1384 }
1385 /* Specifying two names is forbidden */
1386 if (opts->name)
1387 return -EINVAL;
1388 opts->name = kstrndup(name,
Dan Carpentere400c282010-08-10 18:02:54 -07001389 MAX_CGROUP_ROOT_NAMELEN - 1,
Paul Menagec6d57f32009-09-23 15:56:19 -07001390 GFP_KERNEL);
1391 if (!opts->name)
1392 return -ENOMEM;
Daniel Lezcano32a8cf232010-10-27 15:33:37 -07001393
1394 continue;
1395 }
1396
Tejun Heo30159ec2013-06-25 11:53:37 -07001397 for_each_subsys(ss, i) {
Daniel Lezcano32a8cf232010-10-27 15:33:37 -07001398 if (strcmp(token, ss->name))
1399 continue;
1400 if (ss->disabled)
1401 continue;
1402
1403 /* Mutually exclusive option 'all' + subsystem name */
1404 if (all_ss)
1405 return -EINVAL;
Tejun Heo69dfa002014-05-04 15:09:13 -04001406 opts->subsys_mask |= (1 << i);
Daniel Lezcano32a8cf232010-10-27 15:33:37 -07001407 one_ss = true;
1408
1409 break;
1410 }
1411 if (i == CGROUP_SUBSYS_COUNT)
1412 return -ENOENT;
1413 }
1414
Tejun Heo873fe092013-04-14 20:15:26 -07001415 if (opts->flags & CGRP_ROOT_SANE_BEHAVIOR) {
Joe Perchesed3d2612014-04-25 18:28:03 -04001416 pr_warn("sane_behavior: this is still under development and its behaviors will change, proceed at your own risk\n");
Tejun Heo7b9a6ba2014-07-09 10:08:08 -04001417 if (nr_opts != 1) {
1418 pr_err("sane_behavior: no other mount options allowed\n");
Tejun Heo873fe092013-04-14 20:15:26 -07001419 return -EINVAL;
1420 }
Tejun Heo7b9a6ba2014-07-09 10:08:08 -04001421 return 0;
Tejun Heo873fe092013-04-14 20:15:26 -07001422 }
1423
Li Zefanf9ab5b52009-06-17 16:26:33 -07001424 /*
Tejun Heo7b9a6ba2014-07-09 10:08:08 -04001425 * If the 'all' option was specified select all the subsystems,
1426 * otherwise if 'none', 'name=' and a subsystem name options were
1427 * not specified, let's default to 'all'
1428 */
1429 if (all_ss || (!one_ss && !opts->none && !opts->name))
1430 for_each_subsys(ss, i)
1431 if (!ss->disabled)
1432 opts->subsys_mask |= (1 << i);
1433
1434 /*
1435 * We either have to specify by name or by subsystems. (So all
1436 * empty hierarchies must have a name).
1437 */
1438 if (!opts->subsys_mask && !opts->name)
1439 return -EINVAL;
1440
1441 /*
Li Zefanf9ab5b52009-06-17 16:26:33 -07001442 * Option noprefix was introduced just for backward compatibility
1443 * with the old cpuset, so we allow noprefix only if mounting just
1444 * the cpuset subsystem.
1445 */
Tejun Heo93438622013-04-14 20:15:25 -07001446 if ((opts->flags & CGRP_ROOT_NOPREFIX) && (opts->subsys_mask & mask))
Li Zefanf9ab5b52009-06-17 16:26:33 -07001447 return -EINVAL;
1448
Paul Menage2c6ab6d2009-09-23 15:56:23 -07001449 /* Can't specify "none" and some subsystems */
Aristeu Rozanskia1a71b42012-08-23 16:53:31 -04001450 if (opts->subsys_mask && opts->none)
Paul Menage2c6ab6d2009-09-23 15:56:23 -07001451 return -EINVAL;
1452
Paul Menageddbcc7e2007-10-18 23:39:30 -07001453 return 0;
1454}
1455
Tejun Heo2bd59d42014-02-11 11:52:49 -05001456static int cgroup_remount(struct kernfs_root *kf_root, int *flags, char *data)
Paul Menageddbcc7e2007-10-18 23:39:30 -07001457{
1458 int ret = 0;
Tejun Heo3dd06ff2014-03-19 10:23:54 -04001459 struct cgroup_root *root = cgroup_root_from_kf(kf_root);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001460 struct cgroup_sb_opts opts;
Tejun Heo69dfa002014-05-04 15:09:13 -04001461 unsigned int added_mask, removed_mask;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001462
Tejun Heoaa6ec292014-07-09 10:08:08 -04001463 if (root == &cgrp_dfl_root) {
1464 pr_err("remount is not allowed\n");
Tejun Heo873fe092013-04-14 20:15:26 -07001465 return -EINVAL;
1466 }
1467
Paul Menageddbcc7e2007-10-18 23:39:30 -07001468 mutex_lock(&cgroup_mutex);
1469
1470 /* See what subsystems are wanted */
1471 ret = parse_cgroupfs_options(data, &opts);
1472 if (ret)
1473 goto out_unlock;
1474
Tejun Heof392e512014-04-23 11:13:14 -04001475 if (opts.subsys_mask != root->subsys_mask || opts.release_agent)
Joe Perchesed3d2612014-04-25 18:28:03 -04001476 pr_warn("option changes via remount are deprecated (pid=%d comm=%s)\n",
Jianyu Zhana2a1f9e2014-04-25 18:28:03 -04001477 task_tgid_nr(current), current->comm);
Tejun Heo8b5a5a92012-04-01 12:09:54 -07001478
Tejun Heof392e512014-04-23 11:13:14 -04001479 added_mask = opts.subsys_mask & ~root->subsys_mask;
1480 removed_mask = root->subsys_mask & ~opts.subsys_mask;
Aristeu Rozanski13af07d2012-08-23 16:53:29 -04001481
Ben Blumcf5d5942010-03-10 15:22:09 -08001482 /* Don't allow flags or name to change at remount */
Tejun Heo7450e902014-07-09 10:08:07 -04001483 if ((opts.flags ^ root->flags) ||
Ben Blumcf5d5942010-03-10 15:22:09 -08001484 (opts.name && strcmp(opts.name, root->name))) {
Tejun Heo69dfa002014-05-04 15:09:13 -04001485 pr_err("option or name mismatch, new: 0x%x \"%s\", old: 0x%x \"%s\"\n",
Tejun Heo7450e902014-07-09 10:08:07 -04001486 opts.flags, opts.name ?: "", root->flags, root->name);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001487 ret = -EINVAL;
Paul Menagec6d57f32009-09-23 15:56:19 -07001488 goto out_unlock;
1489 }
1490
Tejun Heof172e672013-06-28 17:07:30 -07001491 /* remounting is not allowed for populated hierarchies */
Tejun Heod5c419b2014-05-16 13:22:48 -04001492 if (!list_empty(&root->cgrp.self.children)) {
Tejun Heof172e672013-06-28 17:07:30 -07001493 ret = -EBUSY;
Li Zefan0670e082009-04-02 16:57:30 -07001494 goto out_unlock;
Ben Blumcf5d5942010-03-10 15:22:09 -08001495 }
Paul Menageddbcc7e2007-10-18 23:39:30 -07001496
Tejun Heo5df36032014-03-19 10:23:54 -04001497 ret = rebind_subsystems(root, added_mask);
Tejun Heo31261212013-06-28 17:07:30 -07001498 if (ret)
Paul Menageddbcc7e2007-10-18 23:39:30 -07001499 goto out_unlock;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001500
Tejun Heo3dd06ff2014-03-19 10:23:54 -04001501 rebind_subsystems(&cgrp_dfl_root, removed_mask);
Tejun Heo5df36032014-03-19 10:23:54 -04001502
Tejun Heo69e943b2014-02-08 10:36:58 -05001503 if (opts.release_agent) {
1504 spin_lock(&release_agent_path_lock);
Paul Menage81a6a5c2007-10-18 23:39:38 -07001505 strcpy(root->release_agent_path, opts.release_agent);
Tejun Heo69e943b2014-02-08 10:36:58 -05001506 spin_unlock(&release_agent_path_lock);
1507 }
Paul Menageddbcc7e2007-10-18 23:39:30 -07001508 out_unlock:
Jesper Juhl66bdc9c2009-04-02 16:57:27 -07001509 kfree(opts.release_agent);
Paul Menagec6d57f32009-09-23 15:56:19 -07001510 kfree(opts.name);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001511 mutex_unlock(&cgroup_mutex);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001512 return ret;
1513}
1514
Tejun Heoafeb0f92014-02-13 06:58:39 -05001515/*
1516 * To reduce the fork() overhead for systems that are not actually using
1517 * their cgroups capability, we don't maintain the lists running through
1518 * each css_set to its tasks until we see the list actually used - in other
1519 * words after the first mount.
1520 */
1521static bool use_task_css_set_links __read_mostly;
1522
1523static void cgroup_enable_task_cg_lists(void)
1524{
1525 struct task_struct *p, *g;
1526
Tejun Heo96d365e2014-02-13 06:58:40 -05001527 down_write(&css_set_rwsem);
Tejun Heoafeb0f92014-02-13 06:58:39 -05001528
1529 if (use_task_css_set_links)
1530 goto out_unlock;
1531
1532 use_task_css_set_links = true;
1533
1534 /*
1535 * We need tasklist_lock because RCU is not safe against
1536 * while_each_thread(). Besides, a forking task that has passed
1537 * cgroup_post_fork() without seeing use_task_css_set_links = 1
1538 * is not guaranteed to have its child immediately visible in the
1539 * tasklist if we walk through it with RCU.
1540 */
1541 read_lock(&tasklist_lock);
1542 do_each_thread(g, p) {
Tejun Heoafeb0f92014-02-13 06:58:39 -05001543 WARN_ON_ONCE(!list_empty(&p->cg_list) ||
1544 task_css_set(p) != &init_css_set);
1545
1546 /*
1547 * We should check if the process is exiting, otherwise
1548 * it will race with cgroup_exit() in that the list
1549 * entry won't be deleted though the process has exited.
Tejun Heof153ad12014-02-25 09:56:49 -05001550 * Do it while holding siglock so that we don't end up
1551 * racing against cgroup_exit().
Tejun Heoafeb0f92014-02-13 06:58:39 -05001552 */
Tejun Heof153ad12014-02-25 09:56:49 -05001553 spin_lock_irq(&p->sighand->siglock);
Tejun Heoeaf797a2014-02-25 10:04:03 -05001554 if (!(p->flags & PF_EXITING)) {
1555 struct css_set *cset = task_css_set(p);
1556
1557 list_add(&p->cg_list, &cset->tasks);
1558 get_css_set(cset);
1559 }
Tejun Heof153ad12014-02-25 09:56:49 -05001560 spin_unlock_irq(&p->sighand->siglock);
Tejun Heoafeb0f92014-02-13 06:58:39 -05001561 } while_each_thread(g, p);
1562 read_unlock(&tasklist_lock);
1563out_unlock:
Tejun Heo96d365e2014-02-13 06:58:40 -05001564 up_write(&css_set_rwsem);
Tejun Heoafeb0f92014-02-13 06:58:39 -05001565}
Paul Menageddbcc7e2007-10-18 23:39:30 -07001566
Paul Menagecc31edc2008-10-18 20:28:04 -07001567static void init_cgroup_housekeeping(struct cgroup *cgrp)
1568{
Tejun Heo2d8f2432014-04-23 11:13:15 -04001569 struct cgroup_subsys *ss;
1570 int ssid;
1571
Tejun Heod5c419b2014-05-16 13:22:48 -04001572 INIT_LIST_HEAD(&cgrp->self.sibling);
1573 INIT_LIST_HEAD(&cgrp->self.children);
Tejun Heo69d02062013-06-12 21:04:50 -07001574 INIT_LIST_HEAD(&cgrp->cset_links);
Ben Blum72a8cb32009-09-23 15:56:27 -07001575 INIT_LIST_HEAD(&cgrp->pidlists);
1576 mutex_init(&cgrp->pidlist_mutex);
Tejun Heo9d800df2014-05-14 09:15:00 -04001577 cgrp->self.cgroup = cgrp;
Tejun Heo184faf32014-05-16 13:22:51 -04001578 cgrp->self.flags |= CSS_ONLINE;
Tejun Heo2d8f2432014-04-23 11:13:15 -04001579
1580 for_each_subsys(ss, ssid)
1581 INIT_LIST_HEAD(&cgrp->e_csets[ssid]);
Tejun Heof8f22e52014-04-23 11:13:16 -04001582
1583 init_waitqueue_head(&cgrp->offline_waitq);
Zefan Li971ff492014-09-18 16:06:19 +08001584 INIT_WORK(&cgrp->release_agent_work, cgroup_release_agent);
Paul Menagecc31edc2008-10-18 20:28:04 -07001585}
Paul Menagec6d57f32009-09-23 15:56:19 -07001586
Tejun Heo3dd06ff2014-03-19 10:23:54 -04001587static void init_cgroup_root(struct cgroup_root *root,
Tejun Heo172a2c062014-03-19 10:23:53 -04001588 struct cgroup_sb_opts *opts)
Paul Menageddbcc7e2007-10-18 23:39:30 -07001589{
Tejun Heo3dd06ff2014-03-19 10:23:54 -04001590 struct cgroup *cgrp = &root->cgrp;
Tejun Heob0ca5a82012-04-01 12:09:54 -07001591
Paul Menageddbcc7e2007-10-18 23:39:30 -07001592 INIT_LIST_HEAD(&root->root_list);
Tejun Heo3c9c8252014-02-12 09:29:50 -05001593 atomic_set(&root->nr_cgrps, 1);
Paul Menagebd89aab2007-10-18 23:40:44 -07001594 cgrp->root = root;
Paul Menagecc31edc2008-10-18 20:28:04 -07001595 init_cgroup_housekeeping(cgrp);
Li Zefan4e96ee8e2013-07-31 09:50:50 +08001596 idr_init(&root->cgroup_idr);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001597
Paul Menagec6d57f32009-09-23 15:56:19 -07001598 root->flags = opts->flags;
1599 if (opts->release_agent)
1600 strcpy(root->release_agent_path, opts->release_agent);
1601 if (opts->name)
1602 strcpy(root->name, opts->name);
Tejun Heo2260e7f2012-11-19 08:13:38 -08001603 if (opts->cpuset_clone_children)
Tejun Heo3dd06ff2014-03-19 10:23:54 -04001604 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags);
Paul Menagec6d57f32009-09-23 15:56:19 -07001605}
1606
Tejun Heo69dfa002014-05-04 15:09:13 -04001607static int cgroup_setup_root(struct cgroup_root *root, unsigned int ss_mask)
Paul Menage2c6ab6d2009-09-23 15:56:23 -07001608{
Tejun Heod427dfe2014-02-11 11:52:48 -05001609 LIST_HEAD(tmp_links);
Tejun Heo3dd06ff2014-03-19 10:23:54 -04001610 struct cgroup *root_cgrp = &root->cgrp;
Tejun Heoa14c6872014-07-15 11:05:09 -04001611 struct cftype *base_files;
Tejun Heod427dfe2014-02-11 11:52:48 -05001612 struct css_set *cset;
Tejun Heod427dfe2014-02-11 11:52:48 -05001613 int i, ret;
Paul Menage2c6ab6d2009-09-23 15:56:23 -07001614
Tejun Heod427dfe2014-02-11 11:52:48 -05001615 lockdep_assert_held(&cgroup_mutex);
Paul Menage2c6ab6d2009-09-23 15:56:23 -07001616
Tejun Heo6fa49182014-05-04 15:09:13 -04001617 ret = cgroup_idr_alloc(&root->cgroup_idr, root_cgrp, 1, 2, GFP_NOWAIT);
Tejun Heod427dfe2014-02-11 11:52:48 -05001618 if (ret < 0)
Tejun Heo2bd59d42014-02-11 11:52:49 -05001619 goto out;
Tejun Heod427dfe2014-02-11 11:52:48 -05001620 root_cgrp->id = ret;
Paul Menagec6d57f32009-09-23 15:56:19 -07001621
Tejun Heo2aad2a82014-09-24 13:31:50 -04001622 ret = percpu_ref_init(&root_cgrp->self.refcnt, css_release, 0,
1623 GFP_KERNEL);
Tejun Heo9d755d32014-05-14 09:15:02 -04001624 if (ret)
1625 goto out;
1626
Tejun Heod427dfe2014-02-11 11:52:48 -05001627 /*
Tejun Heo96d365e2014-02-13 06:58:40 -05001628 * We're accessing css_set_count without locking css_set_rwsem here,
Tejun Heod427dfe2014-02-11 11:52:48 -05001629 * but that's OK - it can only be increased by someone holding
1630 * cgroup_lock, and that's us. The worst that can happen is that we
1631 * have some link structures left over
1632 */
1633 ret = allocate_cgrp_cset_links(css_set_count, &tmp_links);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001634 if (ret)
Tejun Heo9d755d32014-05-14 09:15:02 -04001635 goto cancel_ref;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001636
Tejun Heo985ed672014-03-19 10:23:53 -04001637 ret = cgroup_init_root_id(root);
Tejun Heod427dfe2014-02-11 11:52:48 -05001638 if (ret)
Tejun Heo9d755d32014-05-14 09:15:02 -04001639 goto cancel_ref;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001640
Tejun Heo2bd59d42014-02-11 11:52:49 -05001641 root->kf_root = kernfs_create_root(&cgroup_kf_syscall_ops,
1642 KERNFS_ROOT_CREATE_DEACTIVATED,
1643 root_cgrp);
1644 if (IS_ERR(root->kf_root)) {
1645 ret = PTR_ERR(root->kf_root);
1646 goto exit_root_id;
1647 }
1648 root_cgrp->kn = root->kf_root->kn;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001649
Tejun Heoa14c6872014-07-15 11:05:09 -04001650 if (root == &cgrp_dfl_root)
1651 base_files = cgroup_dfl_base_files;
1652 else
1653 base_files = cgroup_legacy_base_files;
1654
1655 ret = cgroup_addrm_files(root_cgrp, base_files, true);
Tejun Heod427dfe2014-02-11 11:52:48 -05001656 if (ret)
Tejun Heo2bd59d42014-02-11 11:52:49 -05001657 goto destroy_root;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001658
Tejun Heo5df36032014-03-19 10:23:54 -04001659 ret = rebind_subsystems(root, ss_mask);
Tejun Heod427dfe2014-02-11 11:52:48 -05001660 if (ret)
Tejun Heo2bd59d42014-02-11 11:52:49 -05001661 goto destroy_root;
Al Viro0df6a632010-12-21 13:29:29 -05001662
Tejun Heod427dfe2014-02-11 11:52:48 -05001663 /*
1664 * There must be no failure case after here, since rebinding takes
1665 * care of subsystems' refcounts, which are explicitly dropped in
1666 * the failure exit path.
1667 */
1668 list_add(&root->root_list, &cgroup_roots);
1669 cgroup_root_count++;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001670
Tejun Heod427dfe2014-02-11 11:52:48 -05001671 /*
Tejun Heo3dd06ff2014-03-19 10:23:54 -04001672 * Link the root cgroup in this hierarchy into all the css_set
Tejun Heod427dfe2014-02-11 11:52:48 -05001673 * objects.
1674 */
Tejun Heo96d365e2014-02-13 06:58:40 -05001675 down_write(&css_set_rwsem);
Tejun Heod427dfe2014-02-11 11:52:48 -05001676 hash_for_each(css_set_table, i, cset, hlist)
1677 link_css_set(&tmp_links, cset, root_cgrp);
Tejun Heo96d365e2014-02-13 06:58:40 -05001678 up_write(&css_set_rwsem);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001679
Tejun Heod5c419b2014-05-16 13:22:48 -04001680 BUG_ON(!list_empty(&root_cgrp->self.children));
Tejun Heo3c9c8252014-02-12 09:29:50 -05001681 BUG_ON(atomic_read(&root->nr_cgrps) != 1);
Tejun Heod427dfe2014-02-11 11:52:48 -05001682
Tejun Heo2bd59d42014-02-11 11:52:49 -05001683 kernfs_activate(root_cgrp->kn);
Tejun Heod427dfe2014-02-11 11:52:48 -05001684 ret = 0;
Tejun Heo2bd59d42014-02-11 11:52:49 -05001685 goto out;
Tejun Heod427dfe2014-02-11 11:52:48 -05001686
Tejun Heo2bd59d42014-02-11 11:52:49 -05001687destroy_root:
1688 kernfs_destroy_root(root->kf_root);
1689 root->kf_root = NULL;
1690exit_root_id:
Tejun Heod427dfe2014-02-11 11:52:48 -05001691 cgroup_exit_root_id(root);
Tejun Heo9d755d32014-05-14 09:15:02 -04001692cancel_ref:
Tejun Heo9a1049d2014-06-28 08:10:14 -04001693 percpu_ref_exit(&root_cgrp->self.refcnt);
Tejun Heo2bd59d42014-02-11 11:52:49 -05001694out:
Tejun Heod427dfe2014-02-11 11:52:48 -05001695 free_cgrp_cset_links(&tmp_links);
1696 return ret;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001697}
1698
Al Virof7e83572010-07-26 13:23:11 +04001699static struct dentry *cgroup_mount(struct file_system_type *fs_type,
Paul Menageddbcc7e2007-10-18 23:39:30 -07001700 int flags, const char *unused_dev_name,
Al Virof7e83572010-07-26 13:23:11 +04001701 void *data)
Paul Menageddbcc7e2007-10-18 23:39:30 -07001702{
Li Zefan3a32bd72014-06-30 11:50:59 +08001703 struct super_block *pinned_sb = NULL;
Li Zefan970317a2014-06-30 11:49:58 +08001704 struct cgroup_subsys *ss;
Tejun Heo3dd06ff2014-03-19 10:23:54 -04001705 struct cgroup_root *root;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001706 struct cgroup_sb_opts opts;
Tejun Heo2bd59d42014-02-11 11:52:49 -05001707 struct dentry *dentry;
Tejun Heo8e30e2b2014-02-11 11:52:48 -05001708 int ret;
Li Zefan970317a2014-06-30 11:49:58 +08001709 int i;
Li Zefanc6b3d5b2014-04-04 17:14:41 +08001710 bool new_sb;
Paul Menagec6d57f32009-09-23 15:56:19 -07001711
1712 /*
Tejun Heo56fde9e2014-02-13 06:58:38 -05001713 * The first time anyone tries to mount a cgroup, enable the list
1714 * linking each css_set to its tasks and fix up all existing tasks.
Paul Menagec6d57f32009-09-23 15:56:19 -07001715 */
Tejun Heo56fde9e2014-02-13 06:58:38 -05001716 if (!use_task_css_set_links)
1717 cgroup_enable_task_cg_lists();
Li Zefane37a06f2014-04-17 13:53:08 +08001718
Tejun Heo8e30e2b2014-02-11 11:52:48 -05001719 mutex_lock(&cgroup_mutex);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001720
Paul Menageddbcc7e2007-10-18 23:39:30 -07001721 /* First find the desired set of subsystems */
Paul Menageddbcc7e2007-10-18 23:39:30 -07001722 ret = parse_cgroupfs_options(data, &opts);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001723 if (ret)
Tejun Heo8e30e2b2014-02-11 11:52:48 -05001724 goto out_unlock;
Tejun Heoa015edd2014-05-14 09:15:00 -04001725
Tejun Heo2bd59d42014-02-11 11:52:49 -05001726 /* look for a matching existing root */
Tejun Heo7b9a6ba2014-07-09 10:08:08 -04001727 if (opts.flags & CGRP_ROOT_SANE_BEHAVIOR) {
Tejun Heoa2dd4242014-03-19 10:23:55 -04001728 cgrp_dfl_root_visible = true;
1729 root = &cgrp_dfl_root;
1730 cgroup_get(&root->cgrp);
1731 ret = 0;
1732 goto out_unlock;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001733 }
1734
Li Zefan970317a2014-06-30 11:49:58 +08001735 /*
1736 * Destruction of cgroup root is asynchronous, so subsystems may
1737 * still be dying after the previous unmount. Let's drain the
1738 * dying subsystems. We just need to ensure that the ones
1739 * unmounted previously finish dying and don't care about new ones
1740 * starting. Testing ref liveliness is good enough.
1741 */
1742 for_each_subsys(ss, i) {
1743 if (!(opts.subsys_mask & (1 << i)) ||
1744 ss->root == &cgrp_dfl_root)
1745 continue;
1746
1747 if (!percpu_ref_tryget_live(&ss->root->cgrp.self.refcnt)) {
1748 mutex_unlock(&cgroup_mutex);
1749 msleep(10);
1750 ret = restart_syscall();
1751 goto out_free;
1752 }
1753 cgroup_put(&ss->root->cgrp);
1754 }
1755
Tejun Heo985ed672014-03-19 10:23:53 -04001756 for_each_root(root) {
Tejun Heo2bd59d42014-02-11 11:52:49 -05001757 bool name_match = false;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001758
Tejun Heo3dd06ff2014-03-19 10:23:54 -04001759 if (root == &cgrp_dfl_root)
Tejun Heo985ed672014-03-19 10:23:53 -04001760 continue;
Paul Menagec6d57f32009-09-23 15:56:19 -07001761
Paul Menage817929e2007-10-18 23:39:36 -07001762 /*
Tejun Heo2bd59d42014-02-11 11:52:49 -05001763 * If we asked for a name then it must match. Also, if
1764 * name matches but sybsys_mask doesn't, we should fail.
1765 * Remember whether name matched.
Paul Menage817929e2007-10-18 23:39:36 -07001766 */
Tejun Heo2bd59d42014-02-11 11:52:49 -05001767 if (opts.name) {
1768 if (strcmp(opts.name, root->name))
1769 continue;
1770 name_match = true;
1771 }
Tejun Heo31261212013-06-28 17:07:30 -07001772
1773 /*
Tejun Heo2bd59d42014-02-11 11:52:49 -05001774 * If we asked for subsystems (or explicitly for no
1775 * subsystems) then they must match.
Tejun Heo31261212013-06-28 17:07:30 -07001776 */
Tejun Heo2bd59d42014-02-11 11:52:49 -05001777 if ((opts.subsys_mask || opts.none) &&
Tejun Heof392e512014-04-23 11:13:14 -04001778 (opts.subsys_mask != root->subsys_mask)) {
Tejun Heo2bd59d42014-02-11 11:52:49 -05001779 if (!name_match)
1780 continue;
1781 ret = -EBUSY;
1782 goto out_unlock;
1783 }
Tejun Heo873fe092013-04-14 20:15:26 -07001784
Tejun Heo7b9a6ba2014-07-09 10:08:08 -04001785 if (root->flags ^ opts.flags)
1786 pr_warn("new mount options do not match the existing superblock, will be ignored\n");
Tejun Heo2bd59d42014-02-11 11:52:49 -05001787
Tejun Heo776f02f2014-02-12 09:29:50 -05001788 /*
Li Zefan3a32bd72014-06-30 11:50:59 +08001789 * We want to reuse @root whose lifetime is governed by its
1790 * ->cgrp. Let's check whether @root is alive and keep it
1791 * that way. As cgroup_kill_sb() can happen anytime, we
1792 * want to block it by pinning the sb so that @root doesn't
1793 * get killed before mount is complete.
1794 *
1795 * With the sb pinned, tryget_live can reliably indicate
1796 * whether @root can be reused. If it's being killed,
1797 * drain it. We can use wait_queue for the wait but this
1798 * path is super cold. Let's just sleep a bit and retry.
Tejun Heo776f02f2014-02-12 09:29:50 -05001799 */
Li Zefan3a32bd72014-06-30 11:50:59 +08001800 pinned_sb = kernfs_pin_sb(root->kf_root, NULL);
1801 if (IS_ERR(pinned_sb) ||
1802 !percpu_ref_tryget_live(&root->cgrp.self.refcnt)) {
Tejun Heo776f02f2014-02-12 09:29:50 -05001803 mutex_unlock(&cgroup_mutex);
Li Zefan3a32bd72014-06-30 11:50:59 +08001804 if (!IS_ERR_OR_NULL(pinned_sb))
1805 deactivate_super(pinned_sb);
Tejun Heo776f02f2014-02-12 09:29:50 -05001806 msleep(10);
Tejun Heoa015edd2014-05-14 09:15:00 -04001807 ret = restart_syscall();
1808 goto out_free;
Tejun Heo776f02f2014-02-12 09:29:50 -05001809 }
1810
1811 ret = 0;
Tejun Heo2bd59d42014-02-11 11:52:49 -05001812 goto out_unlock;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001813 }
1814
Tejun Heo172a2c062014-03-19 10:23:53 -04001815 /*
1816 * No such thing, create a new one. name= matching without subsys
1817 * specification is allowed for already existing hierarchies but we
1818 * can't create new one without subsys specification.
1819 */
1820 if (!opts.subsys_mask && !opts.none) {
1821 ret = -EINVAL;
Tejun Heo2bd59d42014-02-11 11:52:49 -05001822 goto out_unlock;
1823 }
Paul Menageddbcc7e2007-10-18 23:39:30 -07001824
Tejun Heo172a2c062014-03-19 10:23:53 -04001825 root = kzalloc(sizeof(*root), GFP_KERNEL);
1826 if (!root) {
1827 ret = -ENOMEM;
1828 goto out_unlock;
1829 }
1830
1831 init_cgroup_root(root, &opts);
1832
Tejun Heo35585572014-02-13 06:58:38 -05001833 ret = cgroup_setup_root(root, opts.subsys_mask);
Tejun Heo2bd59d42014-02-11 11:52:49 -05001834 if (ret)
1835 cgroup_free_root(root);
1836
Tejun Heo8e30e2b2014-02-11 11:52:48 -05001837out_unlock:
Tejun Heoe25e2cb2011-12-12 18:12:21 -08001838 mutex_unlock(&cgroup_mutex);
Tejun Heoa015edd2014-05-14 09:15:00 -04001839out_free:
Paul Menagec6d57f32009-09-23 15:56:19 -07001840 kfree(opts.release_agent);
1841 kfree(opts.name);
Tejun Heo8e30e2b2014-02-11 11:52:48 -05001842
Tejun Heo2bd59d42014-02-11 11:52:49 -05001843 if (ret)
Tejun Heo8e30e2b2014-02-11 11:52:48 -05001844 return ERR_PTR(ret);
Tejun Heo2bd59d42014-02-11 11:52:49 -05001845
Jianyu Zhanc9482a52014-04-26 15:40:28 +08001846 dentry = kernfs_mount(fs_type, flags, root->kf_root,
1847 CGROUP_SUPER_MAGIC, &new_sb);
Li Zefanc6b3d5b2014-04-04 17:14:41 +08001848 if (IS_ERR(dentry) || !new_sb)
Tejun Heo3dd06ff2014-03-19 10:23:54 -04001849 cgroup_put(&root->cgrp);
Li Zefan3a32bd72014-06-30 11:50:59 +08001850
1851 /*
1852 * If @pinned_sb, we're reusing an existing root and holding an
1853 * extra ref on its sb. Mount is complete. Put the extra ref.
1854 */
1855 if (pinned_sb) {
1856 WARN_ON(new_sb);
1857 deactivate_super(pinned_sb);
1858 }
1859
Tejun Heo2bd59d42014-02-11 11:52:49 -05001860 return dentry;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001861}
1862
SeongJae Parkdd4b0a42014-01-18 16:56:47 +09001863static void cgroup_kill_sb(struct super_block *sb)
1864{
Tejun Heo2bd59d42014-02-11 11:52:49 -05001865 struct kernfs_root *kf_root = kernfs_root_from_sb(sb);
Tejun Heo3dd06ff2014-03-19 10:23:54 -04001866 struct cgroup_root *root = cgroup_root_from_kf(kf_root);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001867
Tejun Heo9d755d32014-05-14 09:15:02 -04001868 /*
1869 * If @root doesn't have any mounts or children, start killing it.
1870 * This prevents new mounts by disabling percpu_ref_tryget_live().
1871 * cgroup_mount() may wait for @root's release.
Li Zefan1f779fb2014-06-04 16:48:15 +08001872 *
1873 * And don't kill the default root.
Tejun Heo9d755d32014-05-14 09:15:02 -04001874 */
Li Zefan1f779fb2014-06-04 16:48:15 +08001875 if (css_has_online_children(&root->cgrp.self) ||
1876 root == &cgrp_dfl_root)
Tejun Heo9d755d32014-05-14 09:15:02 -04001877 cgroup_put(&root->cgrp);
1878 else
1879 percpu_ref_kill(&root->cgrp.self.refcnt);
1880
Tejun Heo2bd59d42014-02-11 11:52:49 -05001881 kernfs_kill_sb(sb);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001882}
1883
1884static struct file_system_type cgroup_fs_type = {
1885 .name = "cgroup",
Al Virof7e83572010-07-26 13:23:11 +04001886 .mount = cgroup_mount,
Paul Menageddbcc7e2007-10-18 23:39:30 -07001887 .kill_sb = cgroup_kill_sb,
1888};
1889
Greg KH676db4a2010-08-05 13:53:35 -07001890static struct kobject *cgroup_kobj;
1891
Li Zefana043e3b2008-02-23 15:24:09 -08001892/**
Tejun Heo913ffdb2013-07-11 16:34:48 -07001893 * task_cgroup_path - cgroup path of a task in the first cgroup hierarchy
Tejun Heo857a2be2013-04-14 20:50:08 -07001894 * @task: target task
Tejun Heo857a2be2013-04-14 20:50:08 -07001895 * @buf: the buffer to write the path into
1896 * @buflen: the length of the buffer
1897 *
Tejun Heo913ffdb2013-07-11 16:34:48 -07001898 * Determine @task's cgroup on the first (the one with the lowest non-zero
1899 * hierarchy_id) cgroup hierarchy and copy its path into @buf. This
1900 * function grabs cgroup_mutex and shouldn't be used inside locks used by
1901 * cgroup controller callbacks.
1902 *
Tejun Heoe61734c2014-02-12 09:29:50 -05001903 * Return value is the same as kernfs_path().
Tejun Heo857a2be2013-04-14 20:50:08 -07001904 */
Tejun Heoe61734c2014-02-12 09:29:50 -05001905char *task_cgroup_path(struct task_struct *task, char *buf, size_t buflen)
Tejun Heo857a2be2013-04-14 20:50:08 -07001906{
Tejun Heo3dd06ff2014-03-19 10:23:54 -04001907 struct cgroup_root *root;
Tejun Heo913ffdb2013-07-11 16:34:48 -07001908 struct cgroup *cgrp;
Tejun Heoe61734c2014-02-12 09:29:50 -05001909 int hierarchy_id = 1;
1910 char *path = NULL;
Tejun Heo857a2be2013-04-14 20:50:08 -07001911
1912 mutex_lock(&cgroup_mutex);
Tejun Heo96d365e2014-02-13 06:58:40 -05001913 down_read(&css_set_rwsem);
Tejun Heo857a2be2013-04-14 20:50:08 -07001914
Tejun Heo913ffdb2013-07-11 16:34:48 -07001915 root = idr_get_next(&cgroup_hierarchy_idr, &hierarchy_id);
1916
Tejun Heo857a2be2013-04-14 20:50:08 -07001917 if (root) {
1918 cgrp = task_cgroup_from_root(task, root);
Tejun Heoe61734c2014-02-12 09:29:50 -05001919 path = cgroup_path(cgrp, buf, buflen);
Tejun Heo913ffdb2013-07-11 16:34:48 -07001920 } else {
1921 /* if no hierarchy exists, everyone is in "/" */
Tejun Heoe61734c2014-02-12 09:29:50 -05001922 if (strlcpy(buf, "/", buflen) < buflen)
1923 path = buf;
Tejun Heo857a2be2013-04-14 20:50:08 -07001924 }
1925
Tejun Heo96d365e2014-02-13 06:58:40 -05001926 up_read(&css_set_rwsem);
Tejun Heo857a2be2013-04-14 20:50:08 -07001927 mutex_unlock(&cgroup_mutex);
Tejun Heoe61734c2014-02-12 09:29:50 -05001928 return path;
Tejun Heo857a2be2013-04-14 20:50:08 -07001929}
Tejun Heo913ffdb2013-07-11 16:34:48 -07001930EXPORT_SYMBOL_GPL(task_cgroup_path);
Tejun Heo857a2be2013-04-14 20:50:08 -07001931
Tejun Heob3dc0942014-02-25 10:04:01 -05001932/* used to track tasks and other necessary states during migration */
Tejun Heo2f7ee562011-12-12 18:12:21 -08001933struct cgroup_taskset {
Tejun Heob3dc0942014-02-25 10:04:01 -05001934 /* the src and dst cset list running through cset->mg_node */
1935 struct list_head src_csets;
1936 struct list_head dst_csets;
1937
1938 /*
1939 * Fields for cgroup_taskset_*() iteration.
1940 *
1941 * Before migration is committed, the target migration tasks are on
1942 * ->mg_tasks of the csets on ->src_csets. After, on ->mg_tasks of
1943 * the csets on ->dst_csets. ->csets point to either ->src_csets
1944 * or ->dst_csets depending on whether migration is committed.
1945 *
1946 * ->cur_csets and ->cur_task point to the current task position
1947 * during iteration.
1948 */
1949 struct list_head *csets;
1950 struct css_set *cur_cset;
1951 struct task_struct *cur_task;
Tejun Heo2f7ee562011-12-12 18:12:21 -08001952};
1953
1954/**
1955 * cgroup_taskset_first - reset taskset and return the first task
1956 * @tset: taskset of interest
1957 *
1958 * @tset iteration is initialized and the first task is returned.
1959 */
1960struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset)
1961{
Tejun Heob3dc0942014-02-25 10:04:01 -05001962 tset->cur_cset = list_first_entry(tset->csets, struct css_set, mg_node);
1963 tset->cur_task = NULL;
1964
1965 return cgroup_taskset_next(tset);
Tejun Heo2f7ee562011-12-12 18:12:21 -08001966}
Tejun Heo2f7ee562011-12-12 18:12:21 -08001967
1968/**
1969 * cgroup_taskset_next - iterate to the next task in taskset
1970 * @tset: taskset of interest
1971 *
1972 * Return the next task in @tset. Iteration must have been initialized
1973 * with cgroup_taskset_first().
1974 */
1975struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset)
1976{
Tejun Heob3dc0942014-02-25 10:04:01 -05001977 struct css_set *cset = tset->cur_cset;
1978 struct task_struct *task = tset->cur_task;
Tejun Heo2f7ee562011-12-12 18:12:21 -08001979
Tejun Heob3dc0942014-02-25 10:04:01 -05001980 while (&cset->mg_node != tset->csets) {
1981 if (!task)
1982 task = list_first_entry(&cset->mg_tasks,
1983 struct task_struct, cg_list);
1984 else
1985 task = list_next_entry(task, cg_list);
Tejun Heo2f7ee562011-12-12 18:12:21 -08001986
Tejun Heob3dc0942014-02-25 10:04:01 -05001987 if (&task->cg_list != &cset->mg_tasks) {
1988 tset->cur_cset = cset;
1989 tset->cur_task = task;
1990 return task;
1991 }
1992
1993 cset = list_next_entry(cset, mg_node);
1994 task = NULL;
1995 }
1996
1997 return NULL;
Tejun Heo2f7ee562011-12-12 18:12:21 -08001998}
Tejun Heo2f7ee562011-12-12 18:12:21 -08001999
2000/**
Ben Blum74a11662011-05-26 16:25:20 -07002001 * cgroup_task_migrate - move a task from one cgroup to another.
Fabian Frederick60106942014-05-05 20:08:13 +02002002 * @old_cgrp: the cgroup @tsk is being migrated from
Tejun Heocb0f1fe2014-02-13 06:58:41 -05002003 * @tsk: the task being migrated
2004 * @new_cset: the new css_set @tsk is being attached to
Ben Blum74a11662011-05-26 16:25:20 -07002005 *
Tejun Heocb0f1fe2014-02-13 06:58:41 -05002006 * Must be called with cgroup_mutex, threadgroup and css_set_rwsem locked.
Ben Blum74a11662011-05-26 16:25:20 -07002007 */
Tejun Heo5abb8852013-06-12 21:04:49 -07002008static void cgroup_task_migrate(struct cgroup *old_cgrp,
2009 struct task_struct *tsk,
2010 struct css_set *new_cset)
Ben Blum74a11662011-05-26 16:25:20 -07002011{
Tejun Heo5abb8852013-06-12 21:04:49 -07002012 struct css_set *old_cset;
Ben Blum74a11662011-05-26 16:25:20 -07002013
Tejun Heocb0f1fe2014-02-13 06:58:41 -05002014 lockdep_assert_held(&cgroup_mutex);
2015 lockdep_assert_held(&css_set_rwsem);
2016
Ben Blum74a11662011-05-26 16:25:20 -07002017 /*
Mandeep Singh Baines026085e2011-12-21 20:18:35 -08002018 * We are synchronized through threadgroup_lock() against PF_EXITING
2019 * setting such that we can't race against cgroup_exit() changing the
2020 * css_set to init_css_set and dropping the old one.
Ben Blum74a11662011-05-26 16:25:20 -07002021 */
Frederic Weisbeckerc84cdf72011-12-21 20:03:18 +01002022 WARN_ON_ONCE(tsk->flags & PF_EXITING);
Tejun Heoa8ad8052013-06-21 15:52:04 -07002023 old_cset = task_css_set(tsk);
Ben Blum74a11662011-05-26 16:25:20 -07002024
Tejun Heob3dc0942014-02-25 10:04:01 -05002025 get_css_set(new_cset);
Tejun Heo5abb8852013-06-12 21:04:49 -07002026 rcu_assign_pointer(tsk->cgroups, new_cset);
Ben Blum74a11662011-05-26 16:25:20 -07002027
Tejun Heo1b9aba42014-03-19 17:43:21 -04002028 /*
2029 * Use move_tail so that cgroup_taskset_first() still returns the
2030 * leader after migration. This works because cgroup_migrate()
2031 * ensures that the dst_cset of the leader is the first on the
2032 * tset's dst_csets list.
2033 */
2034 list_move_tail(&tsk->cg_list, &new_cset->mg_tasks);
Ben Blum74a11662011-05-26 16:25:20 -07002035
2036 /*
Tejun Heo5abb8852013-06-12 21:04:49 -07002037 * We just gained a reference on old_cset by taking it from the
2038 * task. As trading it for new_cset is protected by cgroup_mutex,
2039 * we're safe to drop it here; it will be freed under RCU.
Ben Blum74a11662011-05-26 16:25:20 -07002040 */
Zefan Lia25eb522014-09-19 16:51:00 +08002041 put_css_set_locked(old_cset);
Ben Blum74a11662011-05-26 16:25:20 -07002042}
2043
Li Zefana043e3b2008-02-23 15:24:09 -08002044/**
Tejun Heo1958d2d2014-02-25 10:04:03 -05002045 * cgroup_migrate_finish - cleanup after attach
2046 * @preloaded_csets: list of preloaded css_sets
Ben Blum74a11662011-05-26 16:25:20 -07002047 *
Tejun Heo1958d2d2014-02-25 10:04:03 -05002048 * Undo cgroup_migrate_add_src() and cgroup_migrate_prepare_dst(). See
2049 * those functions for details.
Ben Blum74a11662011-05-26 16:25:20 -07002050 */
Tejun Heo1958d2d2014-02-25 10:04:03 -05002051static void cgroup_migrate_finish(struct list_head *preloaded_csets)
Ben Blum74a11662011-05-26 16:25:20 -07002052{
Tejun Heo1958d2d2014-02-25 10:04:03 -05002053 struct css_set *cset, *tmp_cset;
2054
2055 lockdep_assert_held(&cgroup_mutex);
2056
2057 down_write(&css_set_rwsem);
2058 list_for_each_entry_safe(cset, tmp_cset, preloaded_csets, mg_preload_node) {
2059 cset->mg_src_cgrp = NULL;
2060 cset->mg_dst_cset = NULL;
2061 list_del_init(&cset->mg_preload_node);
Zefan Lia25eb522014-09-19 16:51:00 +08002062 put_css_set_locked(cset);
Tejun Heo1958d2d2014-02-25 10:04:03 -05002063 }
2064 up_write(&css_set_rwsem);
2065}
2066
2067/**
2068 * cgroup_migrate_add_src - add a migration source css_set
2069 * @src_cset: the source css_set to add
2070 * @dst_cgrp: the destination cgroup
2071 * @preloaded_csets: list of preloaded css_sets
2072 *
2073 * Tasks belonging to @src_cset are about to be migrated to @dst_cgrp. Pin
2074 * @src_cset and add it to @preloaded_csets, which should later be cleaned
2075 * up by cgroup_migrate_finish().
2076 *
2077 * This function may be called without holding threadgroup_lock even if the
2078 * target is a process. Threads may be created and destroyed but as long
2079 * as cgroup_mutex is not dropped, no new css_set can be put into play and
2080 * the preloaded css_sets are guaranteed to cover all migrations.
2081 */
2082static void cgroup_migrate_add_src(struct css_set *src_cset,
2083 struct cgroup *dst_cgrp,
2084 struct list_head *preloaded_csets)
2085{
2086 struct cgroup *src_cgrp;
2087
2088 lockdep_assert_held(&cgroup_mutex);
2089 lockdep_assert_held(&css_set_rwsem);
2090
2091 src_cgrp = cset_cgroup_from_root(src_cset, dst_cgrp->root);
2092
Tejun Heo1958d2d2014-02-25 10:04:03 -05002093 if (!list_empty(&src_cset->mg_preload_node))
2094 return;
2095
2096 WARN_ON(src_cset->mg_src_cgrp);
2097 WARN_ON(!list_empty(&src_cset->mg_tasks));
2098 WARN_ON(!list_empty(&src_cset->mg_node));
2099
2100 src_cset->mg_src_cgrp = src_cgrp;
2101 get_css_set(src_cset);
2102 list_add(&src_cset->mg_preload_node, preloaded_csets);
2103}
2104
2105/**
2106 * cgroup_migrate_prepare_dst - prepare destination css_sets for migration
Tejun Heof817de92014-04-23 11:13:16 -04002107 * @dst_cgrp: the destination cgroup (may be %NULL)
Tejun Heo1958d2d2014-02-25 10:04:03 -05002108 * @preloaded_csets: list of preloaded source css_sets
2109 *
2110 * Tasks are about to be moved to @dst_cgrp and all the source css_sets
2111 * have been preloaded to @preloaded_csets. This function looks up and
Tejun Heof817de92014-04-23 11:13:16 -04002112 * pins all destination css_sets, links each to its source, and append them
2113 * to @preloaded_csets. If @dst_cgrp is %NULL, the destination of each
2114 * source css_set is assumed to be its cgroup on the default hierarchy.
Tejun Heo1958d2d2014-02-25 10:04:03 -05002115 *
2116 * This function must be called after cgroup_migrate_add_src() has been
2117 * called on each migration source css_set. After migration is performed
2118 * using cgroup_migrate(), cgroup_migrate_finish() must be called on
2119 * @preloaded_csets.
2120 */
2121static int cgroup_migrate_prepare_dst(struct cgroup *dst_cgrp,
2122 struct list_head *preloaded_csets)
2123{
2124 LIST_HEAD(csets);
Tejun Heof817de92014-04-23 11:13:16 -04002125 struct css_set *src_cset, *tmp_cset;
Tejun Heo1958d2d2014-02-25 10:04:03 -05002126
2127 lockdep_assert_held(&cgroup_mutex);
2128
Tejun Heof8f22e52014-04-23 11:13:16 -04002129 /*
2130 * Except for the root, child_subsys_mask must be zero for a cgroup
2131 * with tasks so that child cgroups don't compete against tasks.
2132 */
Tejun Heod51f39b2014-05-16 13:22:48 -04002133 if (dst_cgrp && cgroup_on_dfl(dst_cgrp) && cgroup_parent(dst_cgrp) &&
Tejun Heof8f22e52014-04-23 11:13:16 -04002134 dst_cgrp->child_subsys_mask)
2135 return -EBUSY;
2136
Tejun Heo1958d2d2014-02-25 10:04:03 -05002137 /* look up the dst cset for each src cset and link it to src */
Tejun Heof817de92014-04-23 11:13:16 -04002138 list_for_each_entry_safe(src_cset, tmp_cset, preloaded_csets, mg_preload_node) {
Tejun Heo1958d2d2014-02-25 10:04:03 -05002139 struct css_set *dst_cset;
2140
Tejun Heof817de92014-04-23 11:13:16 -04002141 dst_cset = find_css_set(src_cset,
2142 dst_cgrp ?: src_cset->dfl_cgrp);
Tejun Heo1958d2d2014-02-25 10:04:03 -05002143 if (!dst_cset)
2144 goto err;
2145
2146 WARN_ON_ONCE(src_cset->mg_dst_cset || dst_cset->mg_dst_cset);
Tejun Heof817de92014-04-23 11:13:16 -04002147
2148 /*
2149 * If src cset equals dst, it's noop. Drop the src.
2150 * cgroup_migrate() will skip the cset too. Note that we
2151 * can't handle src == dst as some nodes are used by both.
2152 */
2153 if (src_cset == dst_cset) {
2154 src_cset->mg_src_cgrp = NULL;
2155 list_del_init(&src_cset->mg_preload_node);
Zefan Lia25eb522014-09-19 16:51:00 +08002156 put_css_set(src_cset);
2157 put_css_set(dst_cset);
Tejun Heof817de92014-04-23 11:13:16 -04002158 continue;
2159 }
2160
Tejun Heo1958d2d2014-02-25 10:04:03 -05002161 src_cset->mg_dst_cset = dst_cset;
2162
2163 if (list_empty(&dst_cset->mg_preload_node))
2164 list_add(&dst_cset->mg_preload_node, &csets);
2165 else
Zefan Lia25eb522014-09-19 16:51:00 +08002166 put_css_set(dst_cset);
Tejun Heo1958d2d2014-02-25 10:04:03 -05002167 }
2168
Tejun Heof817de92014-04-23 11:13:16 -04002169 list_splice_tail(&csets, preloaded_csets);
Tejun Heo1958d2d2014-02-25 10:04:03 -05002170 return 0;
2171err:
2172 cgroup_migrate_finish(&csets);
2173 return -ENOMEM;
2174}
2175
2176/**
2177 * cgroup_migrate - migrate a process or task to a cgroup
2178 * @cgrp: the destination cgroup
2179 * @leader: the leader of the process or the task to migrate
2180 * @threadgroup: whether @leader points to the whole process or a single task
2181 *
2182 * Migrate a process or task denoted by @leader to @cgrp. If migrating a
2183 * process, the caller must be holding threadgroup_lock of @leader. The
2184 * caller is also responsible for invoking cgroup_migrate_add_src() and
2185 * cgroup_migrate_prepare_dst() on the targets before invoking this
2186 * function and following up with cgroup_migrate_finish().
2187 *
2188 * As long as a controller's ->can_attach() doesn't fail, this function is
2189 * guaranteed to succeed. This means that, excluding ->can_attach()
2190 * failure, when migrating multiple targets, the success or failure can be
2191 * decided for all targets by invoking group_migrate_prepare_dst() before
2192 * actually starting migrating.
2193 */
2194static int cgroup_migrate(struct cgroup *cgrp, struct task_struct *leader,
2195 bool threadgroup)
Ben Blum74a11662011-05-26 16:25:20 -07002196{
Tejun Heob3dc0942014-02-25 10:04:01 -05002197 struct cgroup_taskset tset = {
2198 .src_csets = LIST_HEAD_INIT(tset.src_csets),
2199 .dst_csets = LIST_HEAD_INIT(tset.dst_csets),
2200 .csets = &tset.src_csets,
2201 };
Tejun Heo1c6727a2013-12-06 15:11:56 -05002202 struct cgroup_subsys_state *css, *failed_css = NULL;
Tejun Heob3dc0942014-02-25 10:04:01 -05002203 struct css_set *cset, *tmp_cset;
2204 struct task_struct *task, *tmp_task;
2205 int i, ret;
Ben Blum74a11662011-05-26 16:25:20 -07002206
2207 /*
Mandeep Singh Bainesfb5d2b42012-01-03 21:18:31 -08002208 * Prevent freeing of tasks while we take a snapshot. Tasks that are
2209 * already PF_EXITING could be freed from underneath us unless we
2210 * take an rcu_read_lock.
2211 */
Tejun Heob3dc0942014-02-25 10:04:01 -05002212 down_write(&css_set_rwsem);
Mandeep Singh Bainesfb5d2b42012-01-03 21:18:31 -08002213 rcu_read_lock();
Tejun Heo9db8de32014-02-13 06:58:43 -05002214 task = leader;
Ben Blum74a11662011-05-26 16:25:20 -07002215 do {
Tejun Heo9db8de32014-02-13 06:58:43 -05002216 /* @task either already exited or can't exit until the end */
2217 if (task->flags & PF_EXITING)
Anjana V Kumarea847532013-10-12 10:59:17 +08002218 goto next;
Tejun Heocd3d0952011-12-12 18:12:21 -08002219
Tejun Heoeaf797a2014-02-25 10:04:03 -05002220 /* leave @task alone if post_fork() hasn't linked it yet */
2221 if (list_empty(&task->cg_list))
Anjana V Kumarea847532013-10-12 10:59:17 +08002222 goto next;
Tejun Heoeaf797a2014-02-25 10:04:03 -05002223
Tejun Heob3dc0942014-02-25 10:04:01 -05002224 cset = task_css_set(task);
Tejun Heo1958d2d2014-02-25 10:04:03 -05002225 if (!cset->mg_src_cgrp)
Mandeep Singh Baines892a2b92011-12-21 20:18:37 -08002226 goto next;
Tejun Heob3dc0942014-02-25 10:04:01 -05002227
Mandeep Singh Baines61d1d212012-01-30 12:51:56 -08002228 /*
Tejun Heo1b9aba42014-03-19 17:43:21 -04002229 * cgroup_taskset_first() must always return the leader.
2230 * Take care to avoid disturbing the ordering.
Mandeep Singh Baines61d1d212012-01-30 12:51:56 -08002231 */
Tejun Heo1b9aba42014-03-19 17:43:21 -04002232 list_move_tail(&task->cg_list, &cset->mg_tasks);
2233 if (list_empty(&cset->mg_node))
2234 list_add_tail(&cset->mg_node, &tset.src_csets);
2235 if (list_empty(&cset->mg_dst_cset->mg_node))
2236 list_move_tail(&cset->mg_dst_cset->mg_node,
2237 &tset.dst_csets);
Anjana V Kumarea847532013-10-12 10:59:17 +08002238 next:
Li Zefan081aa452013-03-13 09:17:09 +08002239 if (!threadgroup)
2240 break;
Tejun Heo9db8de32014-02-13 06:58:43 -05002241 } while_each_thread(leader, task);
Mandeep Singh Bainesfb5d2b42012-01-03 21:18:31 -08002242 rcu_read_unlock();
Tejun Heob3dc0942014-02-25 10:04:01 -05002243 up_write(&css_set_rwsem);
Ben Blum74a11662011-05-26 16:25:20 -07002244
Tejun Heo134d3372011-12-12 18:12:21 -08002245 /* methods shouldn't be called if no task is actually migrating */
Tejun Heob3dc0942014-02-25 10:04:01 -05002246 if (list_empty(&tset.src_csets))
2247 return 0;
Tejun Heo134d3372011-12-12 18:12:21 -08002248
Tejun Heo1958d2d2014-02-25 10:04:03 -05002249 /* check that we can legitimately attach to the cgroup */
Tejun Heoaec3dfc2014-04-23 11:13:14 -04002250 for_each_e_css(css, i, cgrp) {
Tejun Heo1c6727a2013-12-06 15:11:56 -05002251 if (css->ss->can_attach) {
Tejun Heo9db8de32014-02-13 06:58:43 -05002252 ret = css->ss->can_attach(css, &tset);
2253 if (ret) {
Tejun Heo1c6727a2013-12-06 15:11:56 -05002254 failed_css = css;
Ben Blum74a11662011-05-26 16:25:20 -07002255 goto out_cancel_attach;
2256 }
2257 }
Ben Blum74a11662011-05-26 16:25:20 -07002258 }
2259
2260 /*
Tejun Heo1958d2d2014-02-25 10:04:03 -05002261 * Now that we're guaranteed success, proceed to move all tasks to
2262 * the new cgroup. There are no failure cases after here, so this
2263 * is the commit point.
Ben Blum74a11662011-05-26 16:25:20 -07002264 */
Tejun Heocb0f1fe2014-02-13 06:58:41 -05002265 down_write(&css_set_rwsem);
Tejun Heob3dc0942014-02-25 10:04:01 -05002266 list_for_each_entry(cset, &tset.src_csets, mg_node) {
2267 list_for_each_entry_safe(task, tmp_task, &cset->mg_tasks, cg_list)
2268 cgroup_task_migrate(cset->mg_src_cgrp, task,
2269 cset->mg_dst_cset);
Ben Blum74a11662011-05-26 16:25:20 -07002270 }
Tejun Heocb0f1fe2014-02-13 06:58:41 -05002271 up_write(&css_set_rwsem);
Ben Blum74a11662011-05-26 16:25:20 -07002272
2273 /*
Tejun Heo1958d2d2014-02-25 10:04:03 -05002274 * Migration is committed, all target tasks are now on dst_csets.
2275 * Nothing is sensitive to fork() after this point. Notify
2276 * controllers that migration is complete.
Ben Blum74a11662011-05-26 16:25:20 -07002277 */
Tejun Heob3dc0942014-02-25 10:04:01 -05002278 tset.csets = &tset.dst_csets;
Ben Blum74a11662011-05-26 16:25:20 -07002279
Tejun Heoaec3dfc2014-04-23 11:13:14 -04002280 for_each_e_css(css, i, cgrp)
Tejun Heo1c6727a2013-12-06 15:11:56 -05002281 if (css->ss->attach)
2282 css->ss->attach(css, &tset);
Ben Blum74a11662011-05-26 16:25:20 -07002283
Tejun Heo9db8de32014-02-13 06:58:43 -05002284 ret = 0;
Tejun Heob3dc0942014-02-25 10:04:01 -05002285 goto out_release_tset;
2286
Ben Blum74a11662011-05-26 16:25:20 -07002287out_cancel_attach:
Tejun Heoaec3dfc2014-04-23 11:13:14 -04002288 for_each_e_css(css, i, cgrp) {
Tejun Heob3dc0942014-02-25 10:04:01 -05002289 if (css == failed_css)
2290 break;
2291 if (css->ss->cancel_attach)
2292 css->ss->cancel_attach(css, &tset);
Ben Blum74a11662011-05-26 16:25:20 -07002293 }
Tejun Heob3dc0942014-02-25 10:04:01 -05002294out_release_tset:
2295 down_write(&css_set_rwsem);
2296 list_splice_init(&tset.dst_csets, &tset.src_csets);
2297 list_for_each_entry_safe(cset, tmp_cset, &tset.src_csets, mg_node) {
Tejun Heo1b9aba42014-03-19 17:43:21 -04002298 list_splice_tail_init(&cset->mg_tasks, &cset->tasks);
Tejun Heob3dc0942014-02-25 10:04:01 -05002299 list_del_init(&cset->mg_node);
Tejun Heob3dc0942014-02-25 10:04:01 -05002300 }
2301 up_write(&css_set_rwsem);
Tejun Heo9db8de32014-02-13 06:58:43 -05002302 return ret;
Ben Blum74a11662011-05-26 16:25:20 -07002303}
2304
Tejun Heo1958d2d2014-02-25 10:04:03 -05002305/**
2306 * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup
2307 * @dst_cgrp: the cgroup to attach to
2308 * @leader: the task or the leader of the threadgroup to be attached
2309 * @threadgroup: attach the whole threadgroup?
2310 *
Tejun Heo0e1d7682014-02-25 10:04:03 -05002311 * Call holding cgroup_mutex and threadgroup_lock of @leader.
Tejun Heo1958d2d2014-02-25 10:04:03 -05002312 */
2313static int cgroup_attach_task(struct cgroup *dst_cgrp,
2314 struct task_struct *leader, bool threadgroup)
2315{
2316 LIST_HEAD(preloaded_csets);
2317 struct task_struct *task;
2318 int ret;
2319
2320 /* look up all src csets */
2321 down_read(&css_set_rwsem);
2322 rcu_read_lock();
2323 task = leader;
2324 do {
2325 cgroup_migrate_add_src(task_css_set(task), dst_cgrp,
2326 &preloaded_csets);
2327 if (!threadgroup)
2328 break;
2329 } while_each_thread(leader, task);
2330 rcu_read_unlock();
2331 up_read(&css_set_rwsem);
2332
2333 /* prepare dst csets and commit */
2334 ret = cgroup_migrate_prepare_dst(dst_cgrp, &preloaded_csets);
2335 if (!ret)
2336 ret = cgroup_migrate(dst_cgrp, leader, threadgroup);
2337
2338 cgroup_migrate_finish(&preloaded_csets);
2339 return ret;
Ben Blum74a11662011-05-26 16:25:20 -07002340}
2341
2342/*
2343 * Find the task_struct of the task to attach by vpid and pass it along to the
Tejun Heocd3d0952011-12-12 18:12:21 -08002344 * function to attach either it or all tasks in its threadgroup. Will lock
Tejun Heo0e1d7682014-02-25 10:04:03 -05002345 * cgroup_mutex and threadgroup.
Ben Blum74a11662011-05-26 16:25:20 -07002346 */
Tejun Heoacbef752014-05-13 12:16:22 -04002347static ssize_t __cgroup_procs_write(struct kernfs_open_file *of, char *buf,
2348 size_t nbytes, loff_t off, bool threadgroup)
Paul Menagebbcb81d2007-10-18 23:39:32 -07002349{
Paul Menagebbcb81d2007-10-18 23:39:32 -07002350 struct task_struct *tsk;
David Howellsc69e8d92008-11-14 10:39:19 +11002351 const struct cred *cred = current_cred(), *tcred;
Tejun Heoe76ecae2014-05-13 12:19:23 -04002352 struct cgroup *cgrp;
Tejun Heoacbef752014-05-13 12:16:22 -04002353 pid_t pid;
Paul Menagebbcb81d2007-10-18 23:39:32 -07002354 int ret;
2355
Tejun Heoacbef752014-05-13 12:16:22 -04002356 if (kstrtoint(strstrip(buf), 0, &pid) || pid < 0)
2357 return -EINVAL;
2358
Tejun Heoe76ecae2014-05-13 12:19:23 -04002359 cgrp = cgroup_kn_lock_live(of->kn);
2360 if (!cgrp)
Ben Blum74a11662011-05-26 16:25:20 -07002361 return -ENODEV;
2362
Mandeep Singh Bainesb78949e2012-01-03 21:18:30 -08002363retry_find_task:
2364 rcu_read_lock();
Paul Menagebbcb81d2007-10-18 23:39:32 -07002365 if (pid) {
Pavel Emelyanov73507f32008-02-07 00:14:47 -08002366 tsk = find_task_by_vpid(pid);
Ben Blum74a11662011-05-26 16:25:20 -07002367 if (!tsk) {
Paul Menagebbcb81d2007-10-18 23:39:32 -07002368 rcu_read_unlock();
SeongJae Parkdd4b0a42014-01-18 16:56:47 +09002369 ret = -ESRCH;
Mandeep Singh Bainesb78949e2012-01-03 21:18:30 -08002370 goto out_unlock_cgroup;
Paul Menagebbcb81d2007-10-18 23:39:32 -07002371 }
Ben Blum74a11662011-05-26 16:25:20 -07002372 /*
2373 * even if we're attaching all tasks in the thread group, we
2374 * only need to check permissions on one of them.
2375 */
David Howellsc69e8d92008-11-14 10:39:19 +11002376 tcred = __task_cred(tsk);
Eric W. Biederman14a590c2012-03-12 15:44:39 -07002377 if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) &&
2378 !uid_eq(cred->euid, tcred->uid) &&
2379 !uid_eq(cred->euid, tcred->suid)) {
David Howellsc69e8d92008-11-14 10:39:19 +11002380 rcu_read_unlock();
Mandeep Singh Bainesb78949e2012-01-03 21:18:30 -08002381 ret = -EACCES;
2382 goto out_unlock_cgroup;
Paul Menagebbcb81d2007-10-18 23:39:32 -07002383 }
Mandeep Singh Bainesb78949e2012-01-03 21:18:30 -08002384 } else
2385 tsk = current;
Tejun Heocd3d0952011-12-12 18:12:21 -08002386
2387 if (threadgroup)
Mandeep Singh Bainesb78949e2012-01-03 21:18:30 -08002388 tsk = tsk->group_leader;
Mike Galbraithc4c27fb2012-04-21 09:13:46 +02002389
2390 /*
Tejun Heo14a40ff2013-03-19 13:45:20 -07002391 * Workqueue threads may acquire PF_NO_SETAFFINITY and become
Mike Galbraithc4c27fb2012-04-21 09:13:46 +02002392 * trapped in a cpuset, or RT worker may be born in a cgroup
2393 * with no rt_runtime allocated. Just say no.
2394 */
Tejun Heo14a40ff2013-03-19 13:45:20 -07002395 if (tsk == kthreadd_task || (tsk->flags & PF_NO_SETAFFINITY)) {
Mike Galbraithc4c27fb2012-04-21 09:13:46 +02002396 ret = -EINVAL;
2397 rcu_read_unlock();
2398 goto out_unlock_cgroup;
2399 }
2400
Mandeep Singh Bainesb78949e2012-01-03 21:18:30 -08002401 get_task_struct(tsk);
2402 rcu_read_unlock();
Tejun Heocd3d0952011-12-12 18:12:21 -08002403
Mandeep Singh Bainesb78949e2012-01-03 21:18:30 -08002404 threadgroup_lock(tsk);
2405 if (threadgroup) {
2406 if (!thread_group_leader(tsk)) {
2407 /*
2408 * a race with de_thread from another thread's exec()
2409 * may strip us of our leadership, if this happens,
2410 * there is no choice but to throw this task away and
2411 * try again; this is
2412 * "double-double-toil-and-trouble-check locking".
2413 */
2414 threadgroup_unlock(tsk);
2415 put_task_struct(tsk);
2416 goto retry_find_task;
2417 }
Li Zefan081aa452013-03-13 09:17:09 +08002418 }
2419
2420 ret = cgroup_attach_task(cgrp, tsk, threadgroup);
2421
Tejun Heocd3d0952011-12-12 18:12:21 -08002422 threadgroup_unlock(tsk);
2423
Paul Menagebbcb81d2007-10-18 23:39:32 -07002424 put_task_struct(tsk);
Mandeep Singh Bainesb78949e2012-01-03 21:18:30 -08002425out_unlock_cgroup:
Tejun Heoe76ecae2014-05-13 12:19:23 -04002426 cgroup_kn_unlock(of->kn);
Tejun Heoacbef752014-05-13 12:16:22 -04002427 return ret ?: nbytes;
Paul Menagebbcb81d2007-10-18 23:39:32 -07002428}
2429
Tejun Heo7ae1bad2013-04-07 09:29:51 -07002430/**
2431 * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from'
2432 * @from: attach to all cgroups of a given task
2433 * @tsk: the task to be attached
2434 */
2435int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk)
2436{
Tejun Heo3dd06ff2014-03-19 10:23:54 -04002437 struct cgroup_root *root;
Tejun Heo7ae1bad2013-04-07 09:29:51 -07002438 int retval = 0;
2439
Tejun Heo47cfcd02013-04-07 09:29:51 -07002440 mutex_lock(&cgroup_mutex);
Tejun Heo985ed672014-03-19 10:23:53 -04002441 for_each_root(root) {
Tejun Heo96d365e2014-02-13 06:58:40 -05002442 struct cgroup *from_cgrp;
2443
Tejun Heo3dd06ff2014-03-19 10:23:54 -04002444 if (root == &cgrp_dfl_root)
Tejun Heo985ed672014-03-19 10:23:53 -04002445 continue;
2446
Tejun Heo96d365e2014-02-13 06:58:40 -05002447 down_read(&css_set_rwsem);
2448 from_cgrp = task_cgroup_from_root(from, root);
2449 up_read(&css_set_rwsem);
Tejun Heo7ae1bad2013-04-07 09:29:51 -07002450
Li Zefan6f4b7e62013-07-31 16:18:36 +08002451 retval = cgroup_attach_task(from_cgrp, tsk, false);
Tejun Heo7ae1bad2013-04-07 09:29:51 -07002452 if (retval)
2453 break;
2454 }
Tejun Heo47cfcd02013-04-07 09:29:51 -07002455 mutex_unlock(&cgroup_mutex);
Tejun Heo7ae1bad2013-04-07 09:29:51 -07002456
2457 return retval;
2458}
2459EXPORT_SYMBOL_GPL(cgroup_attach_task_all);
2460
Tejun Heoacbef752014-05-13 12:16:22 -04002461static ssize_t cgroup_tasks_write(struct kernfs_open_file *of,
2462 char *buf, size_t nbytes, loff_t off)
Paul Menageaf351022008-07-25 01:47:01 -07002463{
Tejun Heoacbef752014-05-13 12:16:22 -04002464 return __cgroup_procs_write(of, buf, nbytes, off, false);
Ben Blum74a11662011-05-26 16:25:20 -07002465}
2466
Tejun Heoacbef752014-05-13 12:16:22 -04002467static ssize_t cgroup_procs_write(struct kernfs_open_file *of,
2468 char *buf, size_t nbytes, loff_t off)
Ben Blum74a11662011-05-26 16:25:20 -07002469{
Tejun Heoacbef752014-05-13 12:16:22 -04002470 return __cgroup_procs_write(of, buf, nbytes, off, true);
Paul Menageaf351022008-07-25 01:47:01 -07002471}
2472
Tejun Heo451af502014-05-13 12:16:21 -04002473static ssize_t cgroup_release_agent_write(struct kernfs_open_file *of,
2474 char *buf, size_t nbytes, loff_t off)
Paul Menagee788e0662008-07-25 01:46:59 -07002475{
Tejun Heoe76ecae2014-05-13 12:19:23 -04002476 struct cgroup *cgrp;
Tejun Heo5f469902014-02-11 11:52:48 -05002477
Tejun Heoe76ecae2014-05-13 12:19:23 -04002478 BUILD_BUG_ON(sizeof(cgrp->root->release_agent_path) < PATH_MAX);
2479
2480 cgrp = cgroup_kn_lock_live(of->kn);
2481 if (!cgrp)
Paul Menagee788e0662008-07-25 01:46:59 -07002482 return -ENODEV;
Tejun Heo69e943b2014-02-08 10:36:58 -05002483 spin_lock(&release_agent_path_lock);
Tejun Heoe76ecae2014-05-13 12:19:23 -04002484 strlcpy(cgrp->root->release_agent_path, strstrip(buf),
2485 sizeof(cgrp->root->release_agent_path));
Tejun Heo69e943b2014-02-08 10:36:58 -05002486 spin_unlock(&release_agent_path_lock);
Tejun Heoe76ecae2014-05-13 12:19:23 -04002487 cgroup_kn_unlock(of->kn);
Tejun Heo451af502014-05-13 12:16:21 -04002488 return nbytes;
Paul Menagee788e0662008-07-25 01:46:59 -07002489}
2490
Tejun Heo2da8ca82013-12-05 12:28:04 -05002491static int cgroup_release_agent_show(struct seq_file *seq, void *v)
Paul Menagee788e0662008-07-25 01:46:59 -07002492{
Tejun Heo2da8ca82013-12-05 12:28:04 -05002493 struct cgroup *cgrp = seq_css(seq)->cgroup;
Tejun Heo182446d2013-08-08 20:11:24 -04002494
Tejun Heo46cfeb02014-05-13 12:11:00 -04002495 spin_lock(&release_agent_path_lock);
Paul Menagee788e0662008-07-25 01:46:59 -07002496 seq_puts(seq, cgrp->root->release_agent_path);
Tejun Heo46cfeb02014-05-13 12:11:00 -04002497 spin_unlock(&release_agent_path_lock);
Paul Menagee788e0662008-07-25 01:46:59 -07002498 seq_putc(seq, '\n');
Paul Menagee788e0662008-07-25 01:46:59 -07002499 return 0;
2500}
2501
Tejun Heo2da8ca82013-12-05 12:28:04 -05002502static int cgroup_sane_behavior_show(struct seq_file *seq, void *v)
Tejun Heo873fe092013-04-14 20:15:26 -07002503{
Tejun Heoc1d5d422014-07-09 10:08:08 -04002504 seq_puts(seq, "0\n");
Paul Menage81a6a5c2007-10-18 23:39:38 -07002505 return 0;
Paul Menageddbcc7e2007-10-18 23:39:30 -07002506}
2507
Tejun Heof8f22e52014-04-23 11:13:16 -04002508static void cgroup_print_ss_mask(struct seq_file *seq, unsigned int ss_mask)
2509{
2510 struct cgroup_subsys *ss;
2511 bool printed = false;
2512 int ssid;
2513
2514 for_each_subsys(ss, ssid) {
2515 if (ss_mask & (1 << ssid)) {
2516 if (printed)
2517 seq_putc(seq, ' ');
2518 seq_printf(seq, "%s", ss->name);
2519 printed = true;
2520 }
2521 }
2522 if (printed)
2523 seq_putc(seq, '\n');
2524}
2525
2526/* show controllers which are currently attached to the default hierarchy */
2527static int cgroup_root_controllers_show(struct seq_file *seq, void *v)
2528{
2529 struct cgroup *cgrp = seq_css(seq)->cgroup;
2530
Tejun Heo5533e012014-05-14 19:33:07 -04002531 cgroup_print_ss_mask(seq, cgrp->root->subsys_mask &
2532 ~cgrp_dfl_root_inhibit_ss_mask);
Tejun Heof8f22e52014-04-23 11:13:16 -04002533 return 0;
2534}
2535
2536/* show controllers which are enabled from the parent */
2537static int cgroup_controllers_show(struct seq_file *seq, void *v)
2538{
2539 struct cgroup *cgrp = seq_css(seq)->cgroup;
2540
Tejun Heo667c2492014-07-08 18:02:56 -04002541 cgroup_print_ss_mask(seq, cgroup_parent(cgrp)->subtree_control);
Tejun Heof8f22e52014-04-23 11:13:16 -04002542 return 0;
2543}
2544
2545/* show controllers which are enabled for a given cgroup's children */
2546static int cgroup_subtree_control_show(struct seq_file *seq, void *v)
2547{
2548 struct cgroup *cgrp = seq_css(seq)->cgroup;
2549
Tejun Heo667c2492014-07-08 18:02:56 -04002550 cgroup_print_ss_mask(seq, cgrp->subtree_control);
Tejun Heof8f22e52014-04-23 11:13:16 -04002551 return 0;
2552}
2553
2554/**
2555 * cgroup_update_dfl_csses - update css assoc of a subtree in default hierarchy
2556 * @cgrp: root of the subtree to update csses for
2557 *
2558 * @cgrp's child_subsys_mask has changed and its subtree's (self excluded)
2559 * css associations need to be updated accordingly. This function looks up
2560 * all css_sets which are attached to the subtree, creates the matching
2561 * updated css_sets and migrates the tasks to the new ones.
2562 */
2563static int cgroup_update_dfl_csses(struct cgroup *cgrp)
2564{
2565 LIST_HEAD(preloaded_csets);
2566 struct cgroup_subsys_state *css;
2567 struct css_set *src_cset;
2568 int ret;
2569
Tejun Heof8f22e52014-04-23 11:13:16 -04002570 lockdep_assert_held(&cgroup_mutex);
2571
2572 /* look up all csses currently attached to @cgrp's subtree */
2573 down_read(&css_set_rwsem);
2574 css_for_each_descendant_pre(css, cgroup_css(cgrp, NULL)) {
2575 struct cgrp_cset_link *link;
2576
2577 /* self is not affected by child_subsys_mask change */
2578 if (css->cgroup == cgrp)
2579 continue;
2580
2581 list_for_each_entry(link, &css->cgroup->cset_links, cset_link)
2582 cgroup_migrate_add_src(link->cset, cgrp,
2583 &preloaded_csets);
2584 }
2585 up_read(&css_set_rwsem);
2586
2587 /* NULL dst indicates self on default hierarchy */
2588 ret = cgroup_migrate_prepare_dst(NULL, &preloaded_csets);
2589 if (ret)
2590 goto out_finish;
2591
2592 list_for_each_entry(src_cset, &preloaded_csets, mg_preload_node) {
2593 struct task_struct *last_task = NULL, *task;
2594
2595 /* src_csets precede dst_csets, break on the first dst_cset */
2596 if (!src_cset->mg_src_cgrp)
2597 break;
2598
2599 /*
2600 * All tasks in src_cset need to be migrated to the
2601 * matching dst_cset. Empty it process by process. We
2602 * walk tasks but migrate processes. The leader might even
2603 * belong to a different cset but such src_cset would also
2604 * be among the target src_csets because the default
2605 * hierarchy enforces per-process membership.
2606 */
2607 while (true) {
2608 down_read(&css_set_rwsem);
2609 task = list_first_entry_or_null(&src_cset->tasks,
2610 struct task_struct, cg_list);
2611 if (task) {
2612 task = task->group_leader;
2613 WARN_ON_ONCE(!task_css_set(task)->mg_src_cgrp);
2614 get_task_struct(task);
2615 }
2616 up_read(&css_set_rwsem);
2617
2618 if (!task)
2619 break;
2620
2621 /* guard against possible infinite loop */
2622 if (WARN(last_task == task,
2623 "cgroup: update_dfl_csses failed to make progress, aborting in inconsistent state\n"))
2624 goto out_finish;
2625 last_task = task;
2626
2627 threadgroup_lock(task);
2628 /* raced against de_thread() from another thread? */
2629 if (!thread_group_leader(task)) {
2630 threadgroup_unlock(task);
2631 put_task_struct(task);
2632 continue;
2633 }
2634
2635 ret = cgroup_migrate(src_cset->dfl_cgrp, task, true);
2636
2637 threadgroup_unlock(task);
2638 put_task_struct(task);
2639
2640 if (WARN(ret, "cgroup: failed to update controllers for the default hierarchy (%d), further operations may crash or hang\n", ret))
2641 goto out_finish;
2642 }
2643 }
2644
2645out_finish:
2646 cgroup_migrate_finish(&preloaded_csets);
2647 return ret;
2648}
2649
2650/* change the enabled child controllers for a cgroup in the default hierarchy */
Tejun Heo451af502014-05-13 12:16:21 -04002651static ssize_t cgroup_subtree_control_write(struct kernfs_open_file *of,
2652 char *buf, size_t nbytes,
2653 loff_t off)
Tejun Heof8f22e52014-04-23 11:13:16 -04002654{
Tejun Heo7d331fa2014-05-13 12:11:00 -04002655 unsigned int enable = 0, disable = 0;
Tejun Heo755bf5e2014-11-18 02:49:50 -05002656 unsigned int css_enable, css_disable, old_sc, new_sc, old_ss, new_ss;
Tejun Heoa9746d82014-05-13 12:19:22 -04002657 struct cgroup *cgrp, *child;
Tejun Heof8f22e52014-04-23 11:13:16 -04002658 struct cgroup_subsys *ss;
Tejun Heo451af502014-05-13 12:16:21 -04002659 char *tok;
Tejun Heof8f22e52014-04-23 11:13:16 -04002660 int ssid, ret;
2661
2662 /*
Tejun Heod37167a2014-05-13 12:10:59 -04002663 * Parse input - space separated list of subsystem names prefixed
2664 * with either + or -.
Tejun Heof8f22e52014-04-23 11:13:16 -04002665 */
Tejun Heo451af502014-05-13 12:16:21 -04002666 buf = strstrip(buf);
2667 while ((tok = strsep(&buf, " "))) {
Tejun Heod37167a2014-05-13 12:10:59 -04002668 if (tok[0] == '\0')
2669 continue;
Tejun Heof8f22e52014-04-23 11:13:16 -04002670 for_each_subsys(ss, ssid) {
Tejun Heo5533e012014-05-14 19:33:07 -04002671 if (ss->disabled || strcmp(tok + 1, ss->name) ||
2672 ((1 << ss->id) & cgrp_dfl_root_inhibit_ss_mask))
Tejun Heof8f22e52014-04-23 11:13:16 -04002673 continue;
2674
2675 if (*tok == '+') {
Tejun Heo7d331fa2014-05-13 12:11:00 -04002676 enable |= 1 << ssid;
2677 disable &= ~(1 << ssid);
Tejun Heof8f22e52014-04-23 11:13:16 -04002678 } else if (*tok == '-') {
Tejun Heo7d331fa2014-05-13 12:11:00 -04002679 disable |= 1 << ssid;
2680 enable &= ~(1 << ssid);
Tejun Heof8f22e52014-04-23 11:13:16 -04002681 } else {
2682 return -EINVAL;
2683 }
2684 break;
2685 }
2686 if (ssid == CGROUP_SUBSYS_COUNT)
2687 return -EINVAL;
2688 }
2689
Tejun Heoa9746d82014-05-13 12:19:22 -04002690 cgrp = cgroup_kn_lock_live(of->kn);
2691 if (!cgrp)
2692 return -ENODEV;
Tejun Heof8f22e52014-04-23 11:13:16 -04002693
2694 for_each_subsys(ss, ssid) {
2695 if (enable & (1 << ssid)) {
Tejun Heo667c2492014-07-08 18:02:56 -04002696 if (cgrp->subtree_control & (1 << ssid)) {
Tejun Heof8f22e52014-04-23 11:13:16 -04002697 enable &= ~(1 << ssid);
2698 continue;
2699 }
2700
Tejun Heoc29adf22014-07-08 18:02:56 -04002701 /* unavailable or not enabled on the parent? */
2702 if (!(cgrp_dfl_root.subsys_mask & (1 << ssid)) ||
2703 (cgroup_parent(cgrp) &&
Tejun Heo667c2492014-07-08 18:02:56 -04002704 !(cgroup_parent(cgrp)->subtree_control & (1 << ssid)))) {
Tejun Heoc29adf22014-07-08 18:02:56 -04002705 ret = -ENOENT;
2706 goto out_unlock;
2707 }
2708
Tejun Heof8f22e52014-04-23 11:13:16 -04002709 /*
Tejun Heof63070d2014-07-08 18:02:57 -04002710 * @ss is already enabled through dependency and
2711 * we'll just make it visible. Skip draining.
2712 */
2713 if (cgrp->child_subsys_mask & (1 << ssid))
2714 continue;
2715
Tejun Heof8f22e52014-04-23 11:13:16 -04002716 /*
2717 * Because css offlining is asynchronous, userland
2718 * might try to re-enable the same controller while
2719 * the previous instance is still around. In such
2720 * cases, wait till it's gone using offline_waitq.
2721 */
2722 cgroup_for_each_live_child(child, cgrp) {
Tejun Heo0cee8b72014-05-13 12:10:59 -04002723 DEFINE_WAIT(wait);
Tejun Heof8f22e52014-04-23 11:13:16 -04002724
2725 if (!cgroup_css(child, ss))
2726 continue;
2727
Tejun Heo0cee8b72014-05-13 12:10:59 -04002728 cgroup_get(child);
Tejun Heof8f22e52014-04-23 11:13:16 -04002729 prepare_to_wait(&child->offline_waitq, &wait,
2730 TASK_UNINTERRUPTIBLE);
Tejun Heoa9746d82014-05-13 12:19:22 -04002731 cgroup_kn_unlock(of->kn);
Tejun Heof8f22e52014-04-23 11:13:16 -04002732 schedule();
2733 finish_wait(&child->offline_waitq, &wait);
Tejun Heo0cee8b72014-05-13 12:10:59 -04002734 cgroup_put(child);
Tejun Heo7d331fa2014-05-13 12:11:00 -04002735
Tejun Heoa9746d82014-05-13 12:19:22 -04002736 return restart_syscall();
Tejun Heof8f22e52014-04-23 11:13:16 -04002737 }
Tejun Heof8f22e52014-04-23 11:13:16 -04002738 } else if (disable & (1 << ssid)) {
Tejun Heo667c2492014-07-08 18:02:56 -04002739 if (!(cgrp->subtree_control & (1 << ssid))) {
Tejun Heof8f22e52014-04-23 11:13:16 -04002740 disable &= ~(1 << ssid);
2741 continue;
2742 }
2743
2744 /* a child has it enabled? */
2745 cgroup_for_each_live_child(child, cgrp) {
Tejun Heo667c2492014-07-08 18:02:56 -04002746 if (child->subtree_control & (1 << ssid)) {
Tejun Heof8f22e52014-04-23 11:13:16 -04002747 ret = -EBUSY;
Tejun Heoddab2b62014-05-13 12:19:22 -04002748 goto out_unlock;
Tejun Heof8f22e52014-04-23 11:13:16 -04002749 }
2750 }
2751 }
2752 }
2753
2754 if (!enable && !disable) {
2755 ret = 0;
Tejun Heoddab2b62014-05-13 12:19:22 -04002756 goto out_unlock;
Tejun Heof8f22e52014-04-23 11:13:16 -04002757 }
2758
2759 /*
Tejun Heo667c2492014-07-08 18:02:56 -04002760 * Except for the root, subtree_control must be zero for a cgroup
Tejun Heof8f22e52014-04-23 11:13:16 -04002761 * with tasks so that child cgroups don't compete against tasks.
2762 */
Tejun Heod51f39b2014-05-16 13:22:48 -04002763 if (enable && cgroup_parent(cgrp) && !list_empty(&cgrp->cset_links)) {
Tejun Heof8f22e52014-04-23 11:13:16 -04002764 ret = -EBUSY;
2765 goto out_unlock;
2766 }
2767
2768 /*
Tejun Heof63070d2014-07-08 18:02:57 -04002769 * Update subsys masks and calculate what needs to be done. More
2770 * subsystems than specified may need to be enabled or disabled
2771 * depending on subsystem dependencies.
2772 */
Tejun Heo755bf5e2014-11-18 02:49:50 -05002773 old_sc = cgrp->subtree_control;
2774 old_ss = cgrp->child_subsys_mask;
2775 new_sc = (old_sc | enable) & ~disable;
2776 new_ss = cgroup_calc_child_subsys_mask(cgrp, new_sc);
Tejun Heoc29adf22014-07-08 18:02:56 -04002777
Tejun Heo755bf5e2014-11-18 02:49:50 -05002778 css_enable = ~old_ss & new_ss;
2779 css_disable = old_ss & ~new_ss;
Tejun Heof63070d2014-07-08 18:02:57 -04002780 enable |= css_enable;
2781 disable |= css_disable;
2782
Tejun Heo755bf5e2014-11-18 02:49:50 -05002783 cgrp->subtree_control = new_sc;
2784 cgrp->child_subsys_mask = new_ss;
2785
Tejun Heof63070d2014-07-08 18:02:57 -04002786 /*
2787 * Create new csses or make the existing ones visible. A css is
2788 * created invisible if it's being implicitly enabled through
2789 * dependency. An invisible css is made visible when the userland
2790 * explicitly enables it.
Tejun Heof8f22e52014-04-23 11:13:16 -04002791 */
2792 for_each_subsys(ss, ssid) {
2793 if (!(enable & (1 << ssid)))
2794 continue;
2795
2796 cgroup_for_each_live_child(child, cgrp) {
Tejun Heof63070d2014-07-08 18:02:57 -04002797 if (css_enable & (1 << ssid))
2798 ret = create_css(child, ss,
2799 cgrp->subtree_control & (1 << ssid));
2800 else
2801 ret = cgroup_populate_dir(child, 1 << ssid);
Tejun Heof8f22e52014-04-23 11:13:16 -04002802 if (ret)
2803 goto err_undo_css;
2804 }
2805 }
2806
Tejun Heoc29adf22014-07-08 18:02:56 -04002807 /*
2808 * At this point, cgroup_e_css() results reflect the new csses
2809 * making the following cgroup_update_dfl_csses() properly update
2810 * css associations of all tasks in the subtree.
2811 */
Tejun Heof8f22e52014-04-23 11:13:16 -04002812 ret = cgroup_update_dfl_csses(cgrp);
2813 if (ret)
2814 goto err_undo_css;
2815
Tejun Heof63070d2014-07-08 18:02:57 -04002816 /*
2817 * All tasks are migrated out of disabled csses. Kill or hide
2818 * them. A css is hidden when the userland requests it to be
Tejun Heob4536f0ca2014-07-08 18:02:57 -04002819 * disabled while other subsystems are still depending on it. The
2820 * css must not actively control resources and be in the vanilla
2821 * state if it's made visible again later. Controllers which may
2822 * be depended upon should provide ->css_reset() for this purpose.
Tejun Heof63070d2014-07-08 18:02:57 -04002823 */
Tejun Heof8f22e52014-04-23 11:13:16 -04002824 for_each_subsys(ss, ssid) {
2825 if (!(disable & (1 << ssid)))
2826 continue;
2827
Tejun Heof63070d2014-07-08 18:02:57 -04002828 cgroup_for_each_live_child(child, cgrp) {
Tejun Heob4536f0ca2014-07-08 18:02:57 -04002829 struct cgroup_subsys_state *css = cgroup_css(child, ss);
2830
2831 if (css_disable & (1 << ssid)) {
2832 kill_css(css);
2833 } else {
Tejun Heof63070d2014-07-08 18:02:57 -04002834 cgroup_clear_dir(child, 1 << ssid);
Tejun Heob4536f0ca2014-07-08 18:02:57 -04002835 if (ss->css_reset)
2836 ss->css_reset(css);
2837 }
Tejun Heof63070d2014-07-08 18:02:57 -04002838 }
Tejun Heof8f22e52014-04-23 11:13:16 -04002839 }
2840
2841 kernfs_activate(cgrp->kn);
2842 ret = 0;
2843out_unlock:
Tejun Heoa9746d82014-05-13 12:19:22 -04002844 cgroup_kn_unlock(of->kn);
Tejun Heo451af502014-05-13 12:16:21 -04002845 return ret ?: nbytes;
Tejun Heof8f22e52014-04-23 11:13:16 -04002846
2847err_undo_css:
Tejun Heo755bf5e2014-11-18 02:49:50 -05002848 cgrp->subtree_control = old_sc;
2849 cgrp->child_subsys_mask = old_ss;
Tejun Heof8f22e52014-04-23 11:13:16 -04002850
2851 for_each_subsys(ss, ssid) {
2852 if (!(enable & (1 << ssid)))
2853 continue;
2854
2855 cgroup_for_each_live_child(child, cgrp) {
2856 struct cgroup_subsys_state *css = cgroup_css(child, ss);
Tejun Heof63070d2014-07-08 18:02:57 -04002857
2858 if (!css)
2859 continue;
2860
2861 if (css_enable & (1 << ssid))
Tejun Heof8f22e52014-04-23 11:13:16 -04002862 kill_css(css);
Tejun Heof63070d2014-07-08 18:02:57 -04002863 else
2864 cgroup_clear_dir(child, 1 << ssid);
Tejun Heof8f22e52014-04-23 11:13:16 -04002865 }
2866 }
2867 goto out_unlock;
2868}
2869
Tejun Heo842b5972014-04-25 18:28:02 -04002870static int cgroup_populated_show(struct seq_file *seq, void *v)
2871{
2872 seq_printf(seq, "%d\n", (bool)seq_css(seq)->cgroup->populated_cnt);
2873 return 0;
2874}
2875
Tejun Heo2bd59d42014-02-11 11:52:49 -05002876static ssize_t cgroup_file_write(struct kernfs_open_file *of, char *buf,
2877 size_t nbytes, loff_t off)
Paul Menageddbcc7e2007-10-18 23:39:30 -07002878{
Tejun Heo2bd59d42014-02-11 11:52:49 -05002879 struct cgroup *cgrp = of->kn->parent->priv;
2880 struct cftype *cft = of->kn->priv;
2881 struct cgroup_subsys_state *css;
Tejun Heoa742c592013-12-05 12:28:03 -05002882 int ret;
Paul Menageddbcc7e2007-10-18 23:39:30 -07002883
Tejun Heob4168642014-05-13 12:16:21 -04002884 if (cft->write)
2885 return cft->write(of, buf, nbytes, off);
2886
Tejun Heo2bd59d42014-02-11 11:52:49 -05002887 /*
2888 * kernfs guarantees that a file isn't deleted with operations in
2889 * flight, which means that the matching css is and stays alive and
2890 * doesn't need to be pinned. The RCU locking is not necessary
2891 * either. It's just for the convenience of using cgroup_css().
2892 */
2893 rcu_read_lock();
2894 css = cgroup_css(cgrp, cft->ss);
2895 rcu_read_unlock();
Paul Menageddbcc7e2007-10-18 23:39:30 -07002896
Tejun Heo451af502014-05-13 12:16:21 -04002897 if (cft->write_u64) {
Tejun Heoa742c592013-12-05 12:28:03 -05002898 unsigned long long v;
2899 ret = kstrtoull(buf, 0, &v);
2900 if (!ret)
2901 ret = cft->write_u64(css, cft, v);
2902 } else if (cft->write_s64) {
2903 long long v;
2904 ret = kstrtoll(buf, 0, &v);
2905 if (!ret)
2906 ret = cft->write_s64(css, cft, v);
Tejun Heoa742c592013-12-05 12:28:03 -05002907 } else {
2908 ret = -EINVAL;
2909 }
Tejun Heo2bd59d42014-02-11 11:52:49 -05002910
Tejun Heoa742c592013-12-05 12:28:03 -05002911 return ret ?: nbytes;
Paul Menageddbcc7e2007-10-18 23:39:30 -07002912}
2913
Tejun Heo6612f052013-12-05 12:28:04 -05002914static void *cgroup_seqfile_start(struct seq_file *seq, loff_t *ppos)
Paul Menage91796562008-04-29 01:00:01 -07002915{
Tejun Heo2bd59d42014-02-11 11:52:49 -05002916 return seq_cft(seq)->seq_start(seq, ppos);
Tejun Heo6612f052013-12-05 12:28:04 -05002917}
2918
2919static void *cgroup_seqfile_next(struct seq_file *seq, void *v, loff_t *ppos)
2920{
Tejun Heo2bd59d42014-02-11 11:52:49 -05002921 return seq_cft(seq)->seq_next(seq, v, ppos);
Tejun Heo6612f052013-12-05 12:28:04 -05002922}
2923
2924static void cgroup_seqfile_stop(struct seq_file *seq, void *v)
2925{
Tejun Heo2bd59d42014-02-11 11:52:49 -05002926 seq_cft(seq)->seq_stop(seq, v);
Paul Menage91796562008-04-29 01:00:01 -07002927}
2928
2929static int cgroup_seqfile_show(struct seq_file *m, void *arg)
2930{
Tejun Heo7da11272013-12-05 12:28:04 -05002931 struct cftype *cft = seq_cft(m);
2932 struct cgroup_subsys_state *css = seq_css(m);
Li Zefane0798ce2013-07-31 17:36:25 +08002933
Tejun Heo2da8ca82013-12-05 12:28:04 -05002934 if (cft->seq_show)
2935 return cft->seq_show(m, arg);
Paul Menage91796562008-04-29 01:00:01 -07002936
Tejun Heo896f5192013-12-05 12:28:04 -05002937 if (cft->read_u64)
2938 seq_printf(m, "%llu\n", cft->read_u64(css, cft));
2939 else if (cft->read_s64)
2940 seq_printf(m, "%lld\n", cft->read_s64(css, cft));
2941 else
2942 return -EINVAL;
2943 return 0;
Paul Menage91796562008-04-29 01:00:01 -07002944}
2945
Tejun Heo2bd59d42014-02-11 11:52:49 -05002946static struct kernfs_ops cgroup_kf_single_ops = {
2947 .atomic_write_len = PAGE_SIZE,
2948 .write = cgroup_file_write,
2949 .seq_show = cgroup_seqfile_show,
Paul Menage91796562008-04-29 01:00:01 -07002950};
2951
Tejun Heo2bd59d42014-02-11 11:52:49 -05002952static struct kernfs_ops cgroup_kf_ops = {
2953 .atomic_write_len = PAGE_SIZE,
2954 .write = cgroup_file_write,
2955 .seq_start = cgroup_seqfile_start,
2956 .seq_next = cgroup_seqfile_next,
2957 .seq_stop = cgroup_seqfile_stop,
2958 .seq_show = cgroup_seqfile_show,
2959};
Paul Menageddbcc7e2007-10-18 23:39:30 -07002960
2961/*
2962 * cgroup_rename - Only allow simple rename of directories in place.
2963 */
Tejun Heo2bd59d42014-02-11 11:52:49 -05002964static int cgroup_rename(struct kernfs_node *kn, struct kernfs_node *new_parent,
2965 const char *new_name_str)
Paul Menageddbcc7e2007-10-18 23:39:30 -07002966{
Tejun Heo2bd59d42014-02-11 11:52:49 -05002967 struct cgroup *cgrp = kn->priv;
Li Zefan65dff752013-03-01 15:01:56 +08002968 int ret;
Li Zefan65dff752013-03-01 15:01:56 +08002969
Tejun Heo2bd59d42014-02-11 11:52:49 -05002970 if (kernfs_type(kn) != KERNFS_DIR)
Paul Menageddbcc7e2007-10-18 23:39:30 -07002971 return -ENOTDIR;
Tejun Heo2bd59d42014-02-11 11:52:49 -05002972 if (kn->parent != new_parent)
Paul Menageddbcc7e2007-10-18 23:39:30 -07002973 return -EIO;
Li Zefan65dff752013-03-01 15:01:56 +08002974
Tejun Heo6db8e852013-06-14 11:18:22 -07002975 /*
2976 * This isn't a proper migration and its usefulness is very
Tejun Heoaa6ec292014-07-09 10:08:08 -04002977 * limited. Disallow on the default hierarchy.
Tejun Heo6db8e852013-06-14 11:18:22 -07002978 */
Tejun Heoaa6ec292014-07-09 10:08:08 -04002979 if (cgroup_on_dfl(cgrp))
Tejun Heo6db8e852013-06-14 11:18:22 -07002980 return -EPERM;
2981
Tejun Heoe1b2dc12014-03-20 11:10:15 -04002982 /*
Tejun Heo8353da12014-05-13 12:19:23 -04002983 * We're gonna grab cgroup_mutex which nests outside kernfs
Tejun Heoe1b2dc12014-03-20 11:10:15 -04002984 * active_ref. kernfs_rename() doesn't require active_ref
Tejun Heo8353da12014-05-13 12:19:23 -04002985 * protection. Break them before grabbing cgroup_mutex.
Tejun Heoe1b2dc12014-03-20 11:10:15 -04002986 */
2987 kernfs_break_active_protection(new_parent);
2988 kernfs_break_active_protection(kn);
Li Zefan65dff752013-03-01 15:01:56 +08002989
Tejun Heo2bd59d42014-02-11 11:52:49 -05002990 mutex_lock(&cgroup_mutex);
Li Zefan65dff752013-03-01 15:01:56 +08002991
Tejun Heo2bd59d42014-02-11 11:52:49 -05002992 ret = kernfs_rename(kn, new_parent, new_name_str);
Li Zefan65dff752013-03-01 15:01:56 +08002993
Tejun Heo2bd59d42014-02-11 11:52:49 -05002994 mutex_unlock(&cgroup_mutex);
Paul Menageddbcc7e2007-10-18 23:39:30 -07002995
Tejun Heoe1b2dc12014-03-20 11:10:15 -04002996 kernfs_unbreak_active_protection(kn);
2997 kernfs_unbreak_active_protection(new_parent);
Tejun Heo2bd59d42014-02-11 11:52:49 -05002998 return ret;
Li Zefan099fca32009-04-02 16:57:29 -07002999}
3000
Tejun Heo49957f82014-04-07 16:44:47 -04003001/* set uid and gid of cgroup dirs and files to that of the creator */
3002static int cgroup_kn_set_ugid(struct kernfs_node *kn)
3003{
3004 struct iattr iattr = { .ia_valid = ATTR_UID | ATTR_GID,
3005 .ia_uid = current_fsuid(),
3006 .ia_gid = current_fsgid(), };
3007
3008 if (uid_eq(iattr.ia_uid, GLOBAL_ROOT_UID) &&
3009 gid_eq(iattr.ia_gid, GLOBAL_ROOT_GID))
3010 return 0;
3011
3012 return kernfs_setattr(kn, &iattr);
3013}
3014
Tejun Heo2bb566c2013-08-08 20:11:23 -04003015static int cgroup_add_file(struct cgroup *cgrp, struct cftype *cft)
Paul Menageddbcc7e2007-10-18 23:39:30 -07003016{
Tejun Heo8d7e6fb2014-02-11 11:52:48 -05003017 char name[CGROUP_FILE_NAME_MAX];
Tejun Heo2bd59d42014-02-11 11:52:49 -05003018 struct kernfs_node *kn;
3019 struct lock_class_key *key = NULL;
Tejun Heo49957f82014-04-07 16:44:47 -04003020 int ret;
Tejun Heo8e3f6542012-04-01 12:09:55 -07003021
Tejun Heo2bd59d42014-02-11 11:52:49 -05003022#ifdef CONFIG_DEBUG_LOCK_ALLOC
3023 key = &cft->lockdep_key;
3024#endif
3025 kn = __kernfs_create_file(cgrp->kn, cgroup_file_name(cgrp, cft, name),
3026 cgroup_file_mode(cft), 0, cft->kf_ops, cft,
3027 NULL, false, key);
Tejun Heo49957f82014-04-07 16:44:47 -04003028 if (IS_ERR(kn))
3029 return PTR_ERR(kn);
3030
3031 ret = cgroup_kn_set_ugid(kn);
Tejun Heof8f22e52014-04-23 11:13:16 -04003032 if (ret) {
Tejun Heo49957f82014-04-07 16:44:47 -04003033 kernfs_remove(kn);
Tejun Heof8f22e52014-04-23 11:13:16 -04003034 return ret;
3035 }
3036
Tejun Heob7fc5ad2014-05-13 12:16:22 -04003037 if (cft->seq_show == cgroup_populated_show)
Tejun Heo842b5972014-04-25 18:28:02 -04003038 cgrp->populated_kn = kn;
Tejun Heof8f22e52014-04-23 11:13:16 -04003039 return 0;
Paul Menageddbcc7e2007-10-18 23:39:30 -07003040}
3041
Tejun Heob1f28d32013-06-28 16:24:10 -07003042/**
3043 * cgroup_addrm_files - add or remove files to a cgroup directory
3044 * @cgrp: the target cgroup
Tejun Heob1f28d32013-06-28 16:24:10 -07003045 * @cfts: array of cftypes to be added
3046 * @is_add: whether to add or remove
3047 *
3048 * Depending on @is_add, add or remove files defined by @cfts on @cgrp.
Tejun Heo2bb566c2013-08-08 20:11:23 -04003049 * For removals, this function never fails. If addition fails, this
3050 * function doesn't remove files already added. The caller is responsible
3051 * for cleaning up.
Tejun Heob1f28d32013-06-28 16:24:10 -07003052 */
Tejun Heo2bb566c2013-08-08 20:11:23 -04003053static int cgroup_addrm_files(struct cgroup *cgrp, struct cftype cfts[],
3054 bool is_add)
Paul Menageddbcc7e2007-10-18 23:39:30 -07003055{
Aristeu Rozanski03b1cde2012-08-23 16:53:30 -04003056 struct cftype *cft;
Tejun Heob1f28d32013-06-28 16:24:10 -07003057 int ret;
3058
Tejun Heo01f64742014-05-13 12:19:23 -04003059 lockdep_assert_held(&cgroup_mutex);
Tejun Heodb0416b2012-04-01 12:09:55 -07003060
3061 for (cft = cfts; cft->name[0] != '\0'; cft++) {
Gao fengf33fddc2012-12-06 14:38:57 +08003062 /* does cft->flags tell us to skip this file on @cgrp? */
Tejun Heo05ebb6e2014-07-15 11:05:10 -04003063 if ((cft->flags & __CFTYPE_ONLY_ON_DFL) && !cgroup_on_dfl(cgrp))
Tejun Heo8cbbf2c2014-03-19 10:23:55 -04003064 continue;
Tejun Heo05ebb6e2014-07-15 11:05:10 -04003065 if ((cft->flags & __CFTYPE_NOT_ON_DFL) && cgroup_on_dfl(cgrp))
Tejun Heo873fe092013-04-14 20:15:26 -07003066 continue;
Tejun Heod51f39b2014-05-16 13:22:48 -04003067 if ((cft->flags & CFTYPE_NOT_ON_ROOT) && !cgroup_parent(cgrp))
Gao fengf33fddc2012-12-06 14:38:57 +08003068 continue;
Tejun Heod51f39b2014-05-16 13:22:48 -04003069 if ((cft->flags & CFTYPE_ONLY_ON_ROOT) && cgroup_parent(cgrp))
Gao fengf33fddc2012-12-06 14:38:57 +08003070 continue;
3071
Li Zefan2739d3c2013-01-21 18:18:33 +08003072 if (is_add) {
Tejun Heo2bb566c2013-08-08 20:11:23 -04003073 ret = cgroup_add_file(cgrp, cft);
Tejun Heob1f28d32013-06-28 16:24:10 -07003074 if (ret) {
Joe Perchesed3d2612014-04-25 18:28:03 -04003075 pr_warn("%s: failed to add %s, err=%d\n",
3076 __func__, cft->name, ret);
Tejun Heob1f28d32013-06-28 16:24:10 -07003077 return ret;
3078 }
Li Zefan2739d3c2013-01-21 18:18:33 +08003079 } else {
3080 cgroup_rm_file(cgrp, cft);
Tejun Heodb0416b2012-04-01 12:09:55 -07003081 }
Paul Menageddbcc7e2007-10-18 23:39:30 -07003082 }
Tejun Heob1f28d32013-06-28 16:24:10 -07003083 return 0;
Paul Menageddbcc7e2007-10-18 23:39:30 -07003084}
3085
Tejun Heo21a2d342014-02-12 09:29:49 -05003086static int cgroup_apply_cftypes(struct cftype *cfts, bool is_add)
Tejun Heo8e3f6542012-04-01 12:09:55 -07003087{
3088 LIST_HEAD(pending);
Tejun Heo2bb566c2013-08-08 20:11:23 -04003089 struct cgroup_subsys *ss = cfts[0].ss;
Tejun Heo3dd06ff2014-03-19 10:23:54 -04003090 struct cgroup *root = &ss->root->cgrp;
Tejun Heo492eb212013-08-08 20:11:25 -04003091 struct cgroup_subsys_state *css;
Tejun Heo9ccece82013-06-28 16:24:11 -07003092 int ret = 0;
Tejun Heo8e3f6542012-04-01 12:09:55 -07003093
Tejun Heo01f64742014-05-13 12:19:23 -04003094 lockdep_assert_held(&cgroup_mutex);
Li Zefane8c82d22013-06-18 18:48:37 +08003095
Li Zefane8c82d22013-06-18 18:48:37 +08003096 /* add/rm files for all cgroups created before */
Tejun Heoca8bdca2013-08-26 18:40:56 -04003097 css_for_each_descendant_pre(css, cgroup_css(root, ss)) {
Tejun Heo492eb212013-08-08 20:11:25 -04003098 struct cgroup *cgrp = css->cgroup;
3099
Li Zefane8c82d22013-06-18 18:48:37 +08003100 if (cgroup_is_dead(cgrp))
3101 continue;
3102
Tejun Heo21a2d342014-02-12 09:29:49 -05003103 ret = cgroup_addrm_files(cgrp, cfts, is_add);
Tejun Heo9ccece82013-06-28 16:24:11 -07003104 if (ret)
3105 break;
Tejun Heo8e3f6542012-04-01 12:09:55 -07003106 }
Tejun Heo21a2d342014-02-12 09:29:49 -05003107
3108 if (is_add && !ret)
3109 kernfs_activate(root->kn);
Tejun Heo9ccece82013-06-28 16:24:11 -07003110 return ret;
Tejun Heo8e3f6542012-04-01 12:09:55 -07003111}
3112
Tejun Heo2da440a2014-02-11 11:52:48 -05003113static void cgroup_exit_cftypes(struct cftype *cfts)
3114{
3115 struct cftype *cft;
3116
Tejun Heo2bd59d42014-02-11 11:52:49 -05003117 for (cft = cfts; cft->name[0] != '\0'; cft++) {
3118 /* free copy for custom atomic_write_len, see init_cftypes() */
3119 if (cft->max_write_len && cft->max_write_len != PAGE_SIZE)
3120 kfree(cft->kf_ops);
3121 cft->kf_ops = NULL;
Tejun Heo2da440a2014-02-11 11:52:48 -05003122 cft->ss = NULL;
Tejun Heoa8ddc822014-07-15 11:05:10 -04003123
3124 /* revert flags set by cgroup core while adding @cfts */
Tejun Heo05ebb6e2014-07-15 11:05:10 -04003125 cft->flags &= ~(__CFTYPE_ONLY_ON_DFL | __CFTYPE_NOT_ON_DFL);
Tejun Heo2bd59d42014-02-11 11:52:49 -05003126 }
Tejun Heo2da440a2014-02-11 11:52:48 -05003127}
3128
Tejun Heo2bd59d42014-02-11 11:52:49 -05003129static int cgroup_init_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
Tejun Heo2da440a2014-02-11 11:52:48 -05003130{
3131 struct cftype *cft;
3132
Tejun Heo2bd59d42014-02-11 11:52:49 -05003133 for (cft = cfts; cft->name[0] != '\0'; cft++) {
3134 struct kernfs_ops *kf_ops;
3135
Tejun Heo0adb0702014-02-12 09:29:48 -05003136 WARN_ON(cft->ss || cft->kf_ops);
3137
Tejun Heo2bd59d42014-02-11 11:52:49 -05003138 if (cft->seq_start)
3139 kf_ops = &cgroup_kf_ops;
3140 else
3141 kf_ops = &cgroup_kf_single_ops;
3142
3143 /*
3144 * Ugh... if @cft wants a custom max_write_len, we need to
3145 * make a copy of kf_ops to set its atomic_write_len.
3146 */
3147 if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) {
3148 kf_ops = kmemdup(kf_ops, sizeof(*kf_ops), GFP_KERNEL);
3149 if (!kf_ops) {
3150 cgroup_exit_cftypes(cfts);
3151 return -ENOMEM;
3152 }
3153 kf_ops->atomic_write_len = cft->max_write_len;
3154 }
3155
3156 cft->kf_ops = kf_ops;
Tejun Heo2da440a2014-02-11 11:52:48 -05003157 cft->ss = ss;
Tejun Heo2bd59d42014-02-11 11:52:49 -05003158 }
3159
3160 return 0;
Tejun Heo2da440a2014-02-11 11:52:48 -05003161}
3162
Tejun Heo21a2d342014-02-12 09:29:49 -05003163static int cgroup_rm_cftypes_locked(struct cftype *cfts)
3164{
Tejun Heo01f64742014-05-13 12:19:23 -04003165 lockdep_assert_held(&cgroup_mutex);
Tejun Heo21a2d342014-02-12 09:29:49 -05003166
3167 if (!cfts || !cfts[0].ss)
3168 return -ENOENT;
3169
3170 list_del(&cfts->node);
3171 cgroup_apply_cftypes(cfts, false);
3172 cgroup_exit_cftypes(cfts);
3173 return 0;
3174}
3175
Tejun Heo8e3f6542012-04-01 12:09:55 -07003176/**
Tejun Heo80b13582014-02-12 09:29:48 -05003177 * cgroup_rm_cftypes - remove an array of cftypes from a subsystem
3178 * @cfts: zero-length name terminated array of cftypes
3179 *
3180 * Unregister @cfts. Files described by @cfts are removed from all
3181 * existing cgroups and all future cgroups won't have them either. This
3182 * function can be called anytime whether @cfts' subsys is attached or not.
3183 *
3184 * Returns 0 on successful unregistration, -ENOENT if @cfts is not
3185 * registered.
3186 */
3187int cgroup_rm_cftypes(struct cftype *cfts)
3188{
Tejun Heo21a2d342014-02-12 09:29:49 -05003189 int ret;
Tejun Heo80b13582014-02-12 09:29:48 -05003190
Tejun Heo01f64742014-05-13 12:19:23 -04003191 mutex_lock(&cgroup_mutex);
Tejun Heo21a2d342014-02-12 09:29:49 -05003192 ret = cgroup_rm_cftypes_locked(cfts);
Tejun Heo01f64742014-05-13 12:19:23 -04003193 mutex_unlock(&cgroup_mutex);
Tejun Heo8e3f6542012-04-01 12:09:55 -07003194 return ret;
3195}
3196
3197/**
3198 * cgroup_add_cftypes - add an array of cftypes to a subsystem
3199 * @ss: target cgroup subsystem
3200 * @cfts: zero-length name terminated array of cftypes
3201 *
3202 * Register @cfts to @ss. Files described by @cfts are created for all
3203 * existing cgroups to which @ss is attached and all future cgroups will
3204 * have them too. This function can be called anytime whether @ss is
3205 * attached or not.
3206 *
3207 * Returns 0 on successful registration, -errno on failure. Note that this
3208 * function currently returns 0 as long as @cfts registration is successful
3209 * even if some file creation attempts on existing cgroups fail.
3210 */
Tejun Heo2cf669a2014-07-15 11:05:09 -04003211static int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
Tejun Heo8e3f6542012-04-01 12:09:55 -07003212{
Tejun Heo9ccece82013-06-28 16:24:11 -07003213 int ret;
Tejun Heo8e3f6542012-04-01 12:09:55 -07003214
Li Zefanc731ae12014-06-05 17:16:30 +08003215 if (ss->disabled)
3216 return 0;
3217
Li Zefandc5736e2014-02-17 10:41:50 +08003218 if (!cfts || cfts[0].name[0] == '\0')
3219 return 0;
Tejun Heo8e3f6542012-04-01 12:09:55 -07003220
Tejun Heo2bd59d42014-02-11 11:52:49 -05003221 ret = cgroup_init_cftypes(ss, cfts);
Tejun Heo9ccece82013-06-28 16:24:11 -07003222 if (ret)
Tejun Heo2bd59d42014-02-11 11:52:49 -05003223 return ret;
Tejun Heo8e3f6542012-04-01 12:09:55 -07003224
Tejun Heo01f64742014-05-13 12:19:23 -04003225 mutex_lock(&cgroup_mutex);
Tejun Heo21a2d342014-02-12 09:29:49 -05003226
Tejun Heo0adb0702014-02-12 09:29:48 -05003227 list_add_tail(&cfts->node, &ss->cfts);
Tejun Heo21a2d342014-02-12 09:29:49 -05003228 ret = cgroup_apply_cftypes(cfts, true);
Tejun Heo9ccece82013-06-28 16:24:11 -07003229 if (ret)
Tejun Heo21a2d342014-02-12 09:29:49 -05003230 cgroup_rm_cftypes_locked(cfts);
3231
Tejun Heo01f64742014-05-13 12:19:23 -04003232 mutex_unlock(&cgroup_mutex);
Tejun Heo9ccece82013-06-28 16:24:11 -07003233 return ret;
Tejun Heo8e3f6542012-04-01 12:09:55 -07003234}
Tejun Heo79578622012-04-01 12:09:56 -07003235
3236/**
Tejun Heoa8ddc822014-07-15 11:05:10 -04003237 * cgroup_add_dfl_cftypes - add an array of cftypes for default hierarchy
3238 * @ss: target cgroup subsystem
3239 * @cfts: zero-length name terminated array of cftypes
3240 *
3241 * Similar to cgroup_add_cftypes() but the added files are only used for
3242 * the default hierarchy.
3243 */
3244int cgroup_add_dfl_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
3245{
3246 struct cftype *cft;
3247
3248 for (cft = cfts; cft && cft->name[0] != '\0'; cft++)
Tejun Heo05ebb6e2014-07-15 11:05:10 -04003249 cft->flags |= __CFTYPE_ONLY_ON_DFL;
Tejun Heoa8ddc822014-07-15 11:05:10 -04003250 return cgroup_add_cftypes(ss, cfts);
3251}
3252
3253/**
3254 * cgroup_add_legacy_cftypes - add an array of cftypes for legacy hierarchies
3255 * @ss: target cgroup subsystem
3256 * @cfts: zero-length name terminated array of cftypes
3257 *
3258 * Similar to cgroup_add_cftypes() but the added files are only used for
3259 * the legacy hierarchies.
3260 */
Tejun Heo2cf669a2014-07-15 11:05:09 -04003261int cgroup_add_legacy_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
3262{
Tejun Heoa8ddc822014-07-15 11:05:10 -04003263 struct cftype *cft;
3264
Vivek Goyalfa8137b2014-08-08 11:44:03 -04003265 /*
3266 * If legacy_flies_on_dfl, we want to show the legacy files on the
3267 * dfl hierarchy but iff the target subsystem hasn't been updated
3268 * for the dfl hierarchy yet.
3269 */
3270 if (!cgroup_legacy_files_on_dfl ||
3271 ss->dfl_cftypes != ss->legacy_cftypes) {
3272 for (cft = cfts; cft && cft->name[0] != '\0'; cft++)
3273 cft->flags |= __CFTYPE_NOT_ON_DFL;
3274 }
3275
Tejun Heo2cf669a2014-07-15 11:05:09 -04003276 return cgroup_add_cftypes(ss, cfts);
3277}
3278
Li Zefana043e3b2008-02-23 15:24:09 -08003279/**
3280 * cgroup_task_count - count the number of tasks in a cgroup.
3281 * @cgrp: the cgroup in question
3282 *
3283 * Return the number of tasks in the cgroup.
3284 */
Tejun Heo07bc3562014-02-13 06:58:39 -05003285static int cgroup_task_count(const struct cgroup *cgrp)
Paul Menagebbcb81d2007-10-18 23:39:32 -07003286{
3287 int count = 0;
Tejun Heo69d02062013-06-12 21:04:50 -07003288 struct cgrp_cset_link *link;
Paul Menagebbcb81d2007-10-18 23:39:32 -07003289
Tejun Heo96d365e2014-02-13 06:58:40 -05003290 down_read(&css_set_rwsem);
Tejun Heo69d02062013-06-12 21:04:50 -07003291 list_for_each_entry(link, &cgrp->cset_links, cset_link)
3292 count += atomic_read(&link->cset->refcount);
Tejun Heo96d365e2014-02-13 06:58:40 -05003293 up_read(&css_set_rwsem);
Paul Menagebbcb81d2007-10-18 23:39:32 -07003294 return count;
3295}
3296
Tejun Heo574bd9f2012-11-09 09:12:29 -08003297/**
Tejun Heo492eb212013-08-08 20:11:25 -04003298 * css_next_child - find the next child of a given css
Tejun Heoc2931b72014-05-16 13:22:51 -04003299 * @pos: the current position (%NULL to initiate traversal)
3300 * @parent: css whose children to walk
Tejun Heo53fa5262013-05-24 10:55:38 +09003301 *
Tejun Heoc2931b72014-05-16 13:22:51 -04003302 * This function returns the next child of @parent and should be called
Tejun Heo87fb54f2013-12-06 15:11:55 -05003303 * under either cgroup_mutex or RCU read lock. The only requirement is
Tejun Heoc2931b72014-05-16 13:22:51 -04003304 * that @parent and @pos are accessible. The next sibling is guaranteed to
3305 * be returned regardless of their states.
3306 *
3307 * If a subsystem synchronizes ->css_online() and the start of iteration, a
3308 * css which finished ->css_online() is guaranteed to be visible in the
3309 * future iterations and will stay visible until the last reference is put.
3310 * A css which hasn't finished ->css_online() or already finished
3311 * ->css_offline() may show up during traversal. It's each subsystem's
3312 * responsibility to synchronize against on/offlining.
Tejun Heo53fa5262013-05-24 10:55:38 +09003313 */
Tejun Heoc2931b72014-05-16 13:22:51 -04003314struct cgroup_subsys_state *css_next_child(struct cgroup_subsys_state *pos,
3315 struct cgroup_subsys_state *parent)
Tejun Heo53fa5262013-05-24 10:55:38 +09003316{
Tejun Heoc2931b72014-05-16 13:22:51 -04003317 struct cgroup_subsys_state *next;
Tejun Heo53fa5262013-05-24 10:55:38 +09003318
Tejun Heo8353da12014-05-13 12:19:23 -04003319 cgroup_assert_mutex_or_rcu_locked();
Tejun Heo53fa5262013-05-24 10:55:38 +09003320
3321 /*
Tejun Heode3f0342014-05-16 13:22:49 -04003322 * @pos could already have been unlinked from the sibling list.
3323 * Once a cgroup is removed, its ->sibling.next is no longer
3324 * updated when its next sibling changes. CSS_RELEASED is set when
3325 * @pos is taken off list, at which time its next pointer is valid,
3326 * and, as releases are serialized, the one pointed to by the next
3327 * pointer is guaranteed to not have started release yet. This
3328 * implies that if we observe !CSS_RELEASED on @pos in this RCU
3329 * critical section, the one pointed to by its next pointer is
3330 * guaranteed to not have finished its RCU grace period even if we
3331 * have dropped rcu_read_lock() inbetween iterations.
Tejun Heo3b287a52013-08-08 20:11:24 -04003332 *
Tejun Heode3f0342014-05-16 13:22:49 -04003333 * If @pos has CSS_RELEASED set, its next pointer can't be
3334 * dereferenced; however, as each css is given a monotonically
3335 * increasing unique serial number and always appended to the
3336 * sibling list, the next one can be found by walking the parent's
3337 * children until the first css with higher serial number than
3338 * @pos's. While this path can be slower, it happens iff iteration
3339 * races against release and the race window is very small.
Tejun Heo53fa5262013-05-24 10:55:38 +09003340 */
Tejun Heo3b287a52013-08-08 20:11:24 -04003341 if (!pos) {
Tejun Heoc2931b72014-05-16 13:22:51 -04003342 next = list_entry_rcu(parent->children.next, struct cgroup_subsys_state, sibling);
3343 } else if (likely(!(pos->flags & CSS_RELEASED))) {
3344 next = list_entry_rcu(pos->sibling.next, struct cgroup_subsys_state, sibling);
Tejun Heo3b287a52013-08-08 20:11:24 -04003345 } else {
Tejun Heoc2931b72014-05-16 13:22:51 -04003346 list_for_each_entry_rcu(next, &parent->children, sibling)
Tejun Heo3b287a52013-08-08 20:11:24 -04003347 if (next->serial_nr > pos->serial_nr)
3348 break;
Tejun Heo53fa5262013-05-24 10:55:38 +09003349 }
3350
Tejun Heo3b281af2014-04-23 11:13:15 -04003351 /*
3352 * @next, if not pointing to the head, can be dereferenced and is
Tejun Heoc2931b72014-05-16 13:22:51 -04003353 * the next sibling.
Tejun Heo3b281af2014-04-23 11:13:15 -04003354 */
Tejun Heoc2931b72014-05-16 13:22:51 -04003355 if (&next->sibling != &parent->children)
3356 return next;
Tejun Heo3b281af2014-04-23 11:13:15 -04003357 return NULL;
Tejun Heo53fa5262013-05-24 10:55:38 +09003358}
Tejun Heo53fa5262013-05-24 10:55:38 +09003359
3360/**
Tejun Heo492eb212013-08-08 20:11:25 -04003361 * css_next_descendant_pre - find the next descendant for pre-order walk
Tejun Heo574bd9f2012-11-09 09:12:29 -08003362 * @pos: the current position (%NULL to initiate traversal)
Tejun Heo492eb212013-08-08 20:11:25 -04003363 * @root: css whose descendants to walk
Tejun Heo574bd9f2012-11-09 09:12:29 -08003364 *
Tejun Heo492eb212013-08-08 20:11:25 -04003365 * To be used by css_for_each_descendant_pre(). Find the next descendant
Tejun Heobd8815a2013-08-08 20:11:27 -04003366 * to visit for pre-order traversal of @root's descendants. @root is
3367 * included in the iteration and the first node to be visited.
Tejun Heo75501a62013-05-24 10:55:38 +09003368 *
Tejun Heo87fb54f2013-12-06 15:11:55 -05003369 * While this function requires cgroup_mutex or RCU read locking, it
3370 * doesn't require the whole traversal to be contained in a single critical
3371 * section. This function will return the correct next descendant as long
3372 * as both @pos and @root are accessible and @pos is a descendant of @root.
Tejun Heoc2931b72014-05-16 13:22:51 -04003373 *
3374 * If a subsystem synchronizes ->css_online() and the start of iteration, a
3375 * css which finished ->css_online() is guaranteed to be visible in the
3376 * future iterations and will stay visible until the last reference is put.
3377 * A css which hasn't finished ->css_online() or already finished
3378 * ->css_offline() may show up during traversal. It's each subsystem's
3379 * responsibility to synchronize against on/offlining.
Tejun Heo574bd9f2012-11-09 09:12:29 -08003380 */
Tejun Heo492eb212013-08-08 20:11:25 -04003381struct cgroup_subsys_state *
3382css_next_descendant_pre(struct cgroup_subsys_state *pos,
3383 struct cgroup_subsys_state *root)
Tejun Heo574bd9f2012-11-09 09:12:29 -08003384{
Tejun Heo492eb212013-08-08 20:11:25 -04003385 struct cgroup_subsys_state *next;
Tejun Heo574bd9f2012-11-09 09:12:29 -08003386
Tejun Heo8353da12014-05-13 12:19:23 -04003387 cgroup_assert_mutex_or_rcu_locked();
Tejun Heo574bd9f2012-11-09 09:12:29 -08003388
Tejun Heobd8815a2013-08-08 20:11:27 -04003389 /* if first iteration, visit @root */
Tejun Heo7805d002013-05-24 10:50:24 +09003390 if (!pos)
Tejun Heobd8815a2013-08-08 20:11:27 -04003391 return root;
Tejun Heo574bd9f2012-11-09 09:12:29 -08003392
3393 /* visit the first child if exists */
Tejun Heo492eb212013-08-08 20:11:25 -04003394 next = css_next_child(NULL, pos);
Tejun Heo574bd9f2012-11-09 09:12:29 -08003395 if (next)
3396 return next;
3397
3398 /* no child, visit my or the closest ancestor's next sibling */
Tejun Heo492eb212013-08-08 20:11:25 -04003399 while (pos != root) {
Tejun Heo5c9d5352014-05-16 13:22:48 -04003400 next = css_next_child(pos, pos->parent);
Tejun Heo75501a62013-05-24 10:55:38 +09003401 if (next)
Tejun Heo574bd9f2012-11-09 09:12:29 -08003402 return next;
Tejun Heo5c9d5352014-05-16 13:22:48 -04003403 pos = pos->parent;
Tejun Heo7805d002013-05-24 10:50:24 +09003404 }
Tejun Heo574bd9f2012-11-09 09:12:29 -08003405
3406 return NULL;
3407}
Tejun Heo574bd9f2012-11-09 09:12:29 -08003408
Tejun Heo12a9d2f2013-01-07 08:49:33 -08003409/**
Tejun Heo492eb212013-08-08 20:11:25 -04003410 * css_rightmost_descendant - return the rightmost descendant of a css
3411 * @pos: css of interest
Tejun Heo12a9d2f2013-01-07 08:49:33 -08003412 *
Tejun Heo492eb212013-08-08 20:11:25 -04003413 * Return the rightmost descendant of @pos. If there's no descendant, @pos
3414 * is returned. This can be used during pre-order traversal to skip
Tejun Heo12a9d2f2013-01-07 08:49:33 -08003415 * subtree of @pos.
Tejun Heo75501a62013-05-24 10:55:38 +09003416 *
Tejun Heo87fb54f2013-12-06 15:11:55 -05003417 * While this function requires cgroup_mutex or RCU read locking, it
3418 * doesn't require the whole traversal to be contained in a single critical
3419 * section. This function will return the correct rightmost descendant as
3420 * long as @pos is accessible.
Tejun Heo12a9d2f2013-01-07 08:49:33 -08003421 */
Tejun Heo492eb212013-08-08 20:11:25 -04003422struct cgroup_subsys_state *
3423css_rightmost_descendant(struct cgroup_subsys_state *pos)
Tejun Heo12a9d2f2013-01-07 08:49:33 -08003424{
Tejun Heo492eb212013-08-08 20:11:25 -04003425 struct cgroup_subsys_state *last, *tmp;
Tejun Heo12a9d2f2013-01-07 08:49:33 -08003426
Tejun Heo8353da12014-05-13 12:19:23 -04003427 cgroup_assert_mutex_or_rcu_locked();
Tejun Heo12a9d2f2013-01-07 08:49:33 -08003428
3429 do {
3430 last = pos;
3431 /* ->prev isn't RCU safe, walk ->next till the end */
3432 pos = NULL;
Tejun Heo492eb212013-08-08 20:11:25 -04003433 css_for_each_child(tmp, last)
Tejun Heo12a9d2f2013-01-07 08:49:33 -08003434 pos = tmp;
3435 } while (pos);
3436
3437 return last;
3438}
Tejun Heo12a9d2f2013-01-07 08:49:33 -08003439
Tejun Heo492eb212013-08-08 20:11:25 -04003440static struct cgroup_subsys_state *
3441css_leftmost_descendant(struct cgroup_subsys_state *pos)
Tejun Heo574bd9f2012-11-09 09:12:29 -08003442{
Tejun Heo492eb212013-08-08 20:11:25 -04003443 struct cgroup_subsys_state *last;
Tejun Heo574bd9f2012-11-09 09:12:29 -08003444
3445 do {
3446 last = pos;
Tejun Heo492eb212013-08-08 20:11:25 -04003447 pos = css_next_child(NULL, pos);
Tejun Heo574bd9f2012-11-09 09:12:29 -08003448 } while (pos);
3449
3450 return last;
3451}
3452
3453/**
Tejun Heo492eb212013-08-08 20:11:25 -04003454 * css_next_descendant_post - find the next descendant for post-order walk
Tejun Heo574bd9f2012-11-09 09:12:29 -08003455 * @pos: the current position (%NULL to initiate traversal)
Tejun Heo492eb212013-08-08 20:11:25 -04003456 * @root: css whose descendants to walk
Tejun Heo574bd9f2012-11-09 09:12:29 -08003457 *
Tejun Heo492eb212013-08-08 20:11:25 -04003458 * To be used by css_for_each_descendant_post(). Find the next descendant
Tejun Heobd8815a2013-08-08 20:11:27 -04003459 * to visit for post-order traversal of @root's descendants. @root is
3460 * included in the iteration and the last node to be visited.
Tejun Heo75501a62013-05-24 10:55:38 +09003461 *
Tejun Heo87fb54f2013-12-06 15:11:55 -05003462 * While this function requires cgroup_mutex or RCU read locking, it
3463 * doesn't require the whole traversal to be contained in a single critical
3464 * section. This function will return the correct next descendant as long
3465 * as both @pos and @cgroup are accessible and @pos is a descendant of
3466 * @cgroup.
Tejun Heoc2931b72014-05-16 13:22:51 -04003467 *
3468 * If a subsystem synchronizes ->css_online() and the start of iteration, a
3469 * css which finished ->css_online() is guaranteed to be visible in the
3470 * future iterations and will stay visible until the last reference is put.
3471 * A css which hasn't finished ->css_online() or already finished
3472 * ->css_offline() may show up during traversal. It's each subsystem's
3473 * responsibility to synchronize against on/offlining.
Tejun Heo574bd9f2012-11-09 09:12:29 -08003474 */
Tejun Heo492eb212013-08-08 20:11:25 -04003475struct cgroup_subsys_state *
3476css_next_descendant_post(struct cgroup_subsys_state *pos,
3477 struct cgroup_subsys_state *root)
Tejun Heo574bd9f2012-11-09 09:12:29 -08003478{
Tejun Heo492eb212013-08-08 20:11:25 -04003479 struct cgroup_subsys_state *next;
Tejun Heo574bd9f2012-11-09 09:12:29 -08003480
Tejun Heo8353da12014-05-13 12:19:23 -04003481 cgroup_assert_mutex_or_rcu_locked();
Tejun Heo574bd9f2012-11-09 09:12:29 -08003482
Tejun Heo58b79a92013-09-06 15:31:08 -04003483 /* if first iteration, visit leftmost descendant which may be @root */
3484 if (!pos)
3485 return css_leftmost_descendant(root);
Tejun Heo574bd9f2012-11-09 09:12:29 -08003486
Tejun Heobd8815a2013-08-08 20:11:27 -04003487 /* if we visited @root, we're done */
3488 if (pos == root)
3489 return NULL;
3490
Tejun Heo574bd9f2012-11-09 09:12:29 -08003491 /* if there's an unvisited sibling, visit its leftmost descendant */
Tejun Heo5c9d5352014-05-16 13:22:48 -04003492 next = css_next_child(pos, pos->parent);
Tejun Heo75501a62013-05-24 10:55:38 +09003493 if (next)
Tejun Heo492eb212013-08-08 20:11:25 -04003494 return css_leftmost_descendant(next);
Tejun Heo574bd9f2012-11-09 09:12:29 -08003495
3496 /* no sibling left, visit parent */
Tejun Heo5c9d5352014-05-16 13:22:48 -04003497 return pos->parent;
Tejun Heo574bd9f2012-11-09 09:12:29 -08003498}
Tejun Heo574bd9f2012-11-09 09:12:29 -08003499
Tejun Heof3d46502014-05-16 13:22:52 -04003500/**
3501 * css_has_online_children - does a css have online children
3502 * @css: the target css
3503 *
3504 * Returns %true if @css has any online children; otherwise, %false. This
3505 * function can be called from any context but the caller is responsible
3506 * for synchronizing against on/offlining as necessary.
3507 */
3508bool css_has_online_children(struct cgroup_subsys_state *css)
Tejun Heocbc125e2014-05-14 09:15:01 -04003509{
Tejun Heof3d46502014-05-16 13:22:52 -04003510 struct cgroup_subsys_state *child;
3511 bool ret = false;
Tejun Heocbc125e2014-05-14 09:15:01 -04003512
3513 rcu_read_lock();
Tejun Heof3d46502014-05-16 13:22:52 -04003514 css_for_each_child(child, css) {
Li Zefan99bae5f2014-06-12 14:31:31 +08003515 if (child->flags & CSS_ONLINE) {
Tejun Heof3d46502014-05-16 13:22:52 -04003516 ret = true;
3517 break;
Tejun Heocbc125e2014-05-14 09:15:01 -04003518 }
3519 }
3520 rcu_read_unlock();
Tejun Heof3d46502014-05-16 13:22:52 -04003521 return ret;
Cliff Wickman31a7df02008-02-07 00:14:42 -08003522}
3523
Tejun Heo0942eee2013-08-08 20:11:26 -04003524/**
Tejun Heo72ec7022013-08-08 20:11:26 -04003525 * css_advance_task_iter - advance a task itererator to the next css_set
Tejun Heo0942eee2013-08-08 20:11:26 -04003526 * @it: the iterator to advance
3527 *
3528 * Advance @it to the next css_set to walk.
Tejun Heod5158762013-08-08 20:11:26 -04003529 */
Tejun Heo72ec7022013-08-08 20:11:26 -04003530static void css_advance_task_iter(struct css_task_iter *it)
Tejun Heod5158762013-08-08 20:11:26 -04003531{
Tejun Heo0f0a2b42014-04-23 11:13:15 -04003532 struct list_head *l = it->cset_pos;
Tejun Heod5158762013-08-08 20:11:26 -04003533 struct cgrp_cset_link *link;
3534 struct css_set *cset;
3535
3536 /* Advance to the next non-empty css_set */
3537 do {
3538 l = l->next;
Tejun Heo0f0a2b42014-04-23 11:13:15 -04003539 if (l == it->cset_head) {
3540 it->cset_pos = NULL;
Tejun Heod5158762013-08-08 20:11:26 -04003541 return;
3542 }
Tejun Heo3ebb2b62014-04-23 11:13:15 -04003543
3544 if (it->ss) {
3545 cset = container_of(l, struct css_set,
3546 e_cset_node[it->ss->id]);
3547 } else {
3548 link = list_entry(l, struct cgrp_cset_link, cset_link);
3549 cset = link->cset;
3550 }
Tejun Heoc7561122014-02-25 10:04:01 -05003551 } while (list_empty(&cset->tasks) && list_empty(&cset->mg_tasks));
3552
Tejun Heo0f0a2b42014-04-23 11:13:15 -04003553 it->cset_pos = l;
Tejun Heoc7561122014-02-25 10:04:01 -05003554
3555 if (!list_empty(&cset->tasks))
Tejun Heo0f0a2b42014-04-23 11:13:15 -04003556 it->task_pos = cset->tasks.next;
Tejun Heoc7561122014-02-25 10:04:01 -05003557 else
Tejun Heo0f0a2b42014-04-23 11:13:15 -04003558 it->task_pos = cset->mg_tasks.next;
3559
3560 it->tasks_head = &cset->tasks;
3561 it->mg_tasks_head = &cset->mg_tasks;
Tejun Heod5158762013-08-08 20:11:26 -04003562}
3563
Tejun Heo0942eee2013-08-08 20:11:26 -04003564/**
Tejun Heo72ec7022013-08-08 20:11:26 -04003565 * css_task_iter_start - initiate task iteration
3566 * @css: the css to walk tasks of
Tejun Heo0942eee2013-08-08 20:11:26 -04003567 * @it: the task iterator to use
3568 *
Tejun Heo72ec7022013-08-08 20:11:26 -04003569 * Initiate iteration through the tasks of @css. The caller can call
3570 * css_task_iter_next() to walk through the tasks until the function
3571 * returns NULL. On completion of iteration, css_task_iter_end() must be
3572 * called.
Tejun Heo0942eee2013-08-08 20:11:26 -04003573 *
3574 * Note that this function acquires a lock which is released when the
3575 * iteration finishes. The caller can't sleep while iteration is in
3576 * progress.
3577 */
Tejun Heo72ec7022013-08-08 20:11:26 -04003578void css_task_iter_start(struct cgroup_subsys_state *css,
3579 struct css_task_iter *it)
Tejun Heo96d365e2014-02-13 06:58:40 -05003580 __acquires(css_set_rwsem)
Paul Menage817929e2007-10-18 23:39:36 -07003581{
Tejun Heo56fde9e2014-02-13 06:58:38 -05003582 /* no one should try to iterate before mounting cgroups */
3583 WARN_ON_ONCE(!use_task_css_set_links);
Paul Menage817929e2007-10-18 23:39:36 -07003584
Tejun Heo96d365e2014-02-13 06:58:40 -05003585 down_read(&css_set_rwsem);
Tejun Heoc59cd3d2013-08-08 20:11:26 -04003586
Tejun Heo3ebb2b62014-04-23 11:13:15 -04003587 it->ss = css->ss;
3588
3589 if (it->ss)
3590 it->cset_pos = &css->cgroup->e_csets[css->ss->id];
3591 else
3592 it->cset_pos = &css->cgroup->cset_links;
3593
Tejun Heo0f0a2b42014-04-23 11:13:15 -04003594 it->cset_head = it->cset_pos;
Tejun Heoc59cd3d2013-08-08 20:11:26 -04003595
Tejun Heo72ec7022013-08-08 20:11:26 -04003596 css_advance_task_iter(it);
Paul Menagebd89aab2007-10-18 23:40:44 -07003597}
Paul Menage817929e2007-10-18 23:39:36 -07003598
Tejun Heo0942eee2013-08-08 20:11:26 -04003599/**
Tejun Heo72ec7022013-08-08 20:11:26 -04003600 * css_task_iter_next - return the next task for the iterator
Tejun Heo0942eee2013-08-08 20:11:26 -04003601 * @it: the task iterator being iterated
3602 *
3603 * The "next" function for task iteration. @it should have been
Tejun Heo72ec7022013-08-08 20:11:26 -04003604 * initialized via css_task_iter_start(). Returns NULL when the iteration
3605 * reaches the end.
Tejun Heo0942eee2013-08-08 20:11:26 -04003606 */
Tejun Heo72ec7022013-08-08 20:11:26 -04003607struct task_struct *css_task_iter_next(struct css_task_iter *it)
Paul Menage817929e2007-10-18 23:39:36 -07003608{
3609 struct task_struct *res;
Tejun Heo0f0a2b42014-04-23 11:13:15 -04003610 struct list_head *l = it->task_pos;
Paul Menage817929e2007-10-18 23:39:36 -07003611
3612 /* If the iterator cg is NULL, we have no tasks */
Tejun Heo0f0a2b42014-04-23 11:13:15 -04003613 if (!it->cset_pos)
Paul Menage817929e2007-10-18 23:39:36 -07003614 return NULL;
3615 res = list_entry(l, struct task_struct, cg_list);
Tejun Heoc7561122014-02-25 10:04:01 -05003616
3617 /*
3618 * Advance iterator to find next entry. cset->tasks is consumed
3619 * first and then ->mg_tasks. After ->mg_tasks, we move onto the
3620 * next cset.
3621 */
Paul Menage817929e2007-10-18 23:39:36 -07003622 l = l->next;
Tejun Heoc7561122014-02-25 10:04:01 -05003623
Tejun Heo0f0a2b42014-04-23 11:13:15 -04003624 if (l == it->tasks_head)
3625 l = it->mg_tasks_head->next;
Tejun Heoc7561122014-02-25 10:04:01 -05003626
Tejun Heo0f0a2b42014-04-23 11:13:15 -04003627 if (l == it->mg_tasks_head)
Tejun Heo72ec7022013-08-08 20:11:26 -04003628 css_advance_task_iter(it);
Tejun Heoc7561122014-02-25 10:04:01 -05003629 else
Tejun Heo0f0a2b42014-04-23 11:13:15 -04003630 it->task_pos = l;
Tejun Heoc7561122014-02-25 10:04:01 -05003631
Paul Menage817929e2007-10-18 23:39:36 -07003632 return res;
3633}
3634
Tejun Heo0942eee2013-08-08 20:11:26 -04003635/**
Tejun Heo72ec7022013-08-08 20:11:26 -04003636 * css_task_iter_end - finish task iteration
Tejun Heo0942eee2013-08-08 20:11:26 -04003637 * @it: the task iterator to finish
3638 *
Tejun Heo72ec7022013-08-08 20:11:26 -04003639 * Finish task iteration started by css_task_iter_start().
Tejun Heo0942eee2013-08-08 20:11:26 -04003640 */
Tejun Heo72ec7022013-08-08 20:11:26 -04003641void css_task_iter_end(struct css_task_iter *it)
Tejun Heo96d365e2014-02-13 06:58:40 -05003642 __releases(css_set_rwsem)
Paul Menage817929e2007-10-18 23:39:36 -07003643{
Tejun Heo96d365e2014-02-13 06:58:40 -05003644 up_read(&css_set_rwsem);
Tejun Heo8cc99342013-04-07 09:29:50 -07003645}
3646
3647/**
3648 * cgroup_trasnsfer_tasks - move tasks from one cgroup to another
3649 * @to: cgroup to which the tasks will be moved
3650 * @from: cgroup in which the tasks currently reside
Tejun Heoeaf797a2014-02-25 10:04:03 -05003651 *
3652 * Locking rules between cgroup_post_fork() and the migration path
3653 * guarantee that, if a task is forking while being migrated, the new child
3654 * is guaranteed to be either visible in the source cgroup after the
3655 * parent's migration is complete or put into the target cgroup. No task
3656 * can slip out of migration through forking.
Tejun Heo8cc99342013-04-07 09:29:50 -07003657 */
3658int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from)
3659{
Tejun Heo952aaa12014-02-25 10:04:03 -05003660 LIST_HEAD(preloaded_csets);
3661 struct cgrp_cset_link *link;
Tejun Heoe406d1c2014-02-13 06:58:39 -05003662 struct css_task_iter it;
3663 struct task_struct *task;
Tejun Heo952aaa12014-02-25 10:04:03 -05003664 int ret;
Tejun Heoe406d1c2014-02-13 06:58:39 -05003665
Tejun Heo952aaa12014-02-25 10:04:03 -05003666 mutex_lock(&cgroup_mutex);
3667
3668 /* all tasks in @from are being moved, all csets are source */
3669 down_read(&css_set_rwsem);
3670 list_for_each_entry(link, &from->cset_links, cset_link)
3671 cgroup_migrate_add_src(link->cset, to, &preloaded_csets);
3672 up_read(&css_set_rwsem);
3673
3674 ret = cgroup_migrate_prepare_dst(to, &preloaded_csets);
3675 if (ret)
3676 goto out_err;
3677
3678 /*
3679 * Migrate tasks one-by-one until @form is empty. This fails iff
3680 * ->can_attach() fails.
3681 */
Tejun Heoe406d1c2014-02-13 06:58:39 -05003682 do {
Tejun Heo9d800df2014-05-14 09:15:00 -04003683 css_task_iter_start(&from->self, &it);
Tejun Heoe406d1c2014-02-13 06:58:39 -05003684 task = css_task_iter_next(&it);
3685 if (task)
3686 get_task_struct(task);
3687 css_task_iter_end(&it);
3688
3689 if (task) {
Tejun Heo952aaa12014-02-25 10:04:03 -05003690 ret = cgroup_migrate(to, task, false);
Tejun Heoe406d1c2014-02-13 06:58:39 -05003691 put_task_struct(task);
3692 }
3693 } while (task && !ret);
Tejun Heo952aaa12014-02-25 10:04:03 -05003694out_err:
3695 cgroup_migrate_finish(&preloaded_csets);
3696 mutex_unlock(&cgroup_mutex);
Tejun Heoe406d1c2014-02-13 06:58:39 -05003697 return ret;
Tejun Heo8cc99342013-04-07 09:29:50 -07003698}
3699
Paul Menage817929e2007-10-18 23:39:36 -07003700/*
Ben Blum102a7752009-09-23 15:56:26 -07003701 * Stuff for reading the 'tasks'/'procs' files.
Paul Menagebbcb81d2007-10-18 23:39:32 -07003702 *
3703 * Reading this file can return large amounts of data if a cgroup has
3704 * *lots* of attached tasks. So it may need several calls to read(),
3705 * but we cannot guarantee that the information we produce is correct
3706 * unless we produce it entirely atomically.
3707 *
Paul Menagebbcb81d2007-10-18 23:39:32 -07003708 */
Paul Menagebbcb81d2007-10-18 23:39:32 -07003709
Li Zefan24528252012-01-20 11:58:43 +08003710/* which pidlist file are we talking about? */
3711enum cgroup_filetype {
3712 CGROUP_FILE_PROCS,
3713 CGROUP_FILE_TASKS,
3714};
3715
3716/*
3717 * A pidlist is a list of pids that virtually represents the contents of one
3718 * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists,
3719 * a pair (one each for procs, tasks) for each pid namespace that's relevant
3720 * to the cgroup.
3721 */
3722struct cgroup_pidlist {
3723 /*
3724 * used to find which pidlist is wanted. doesn't change as long as
3725 * this particular list stays in the list.
3726 */
3727 struct { enum cgroup_filetype type; struct pid_namespace *ns; } key;
3728 /* array of xids */
3729 pid_t *list;
3730 /* how many elements the above list has */
3731 int length;
Li Zefan24528252012-01-20 11:58:43 +08003732 /* each of these stored in a list by its cgroup */
3733 struct list_head links;
3734 /* pointer to the cgroup we belong to, for list removal purposes */
3735 struct cgroup *owner;
Tejun Heob1a21362013-11-29 10:42:58 -05003736 /* for delayed destruction */
3737 struct delayed_work destroy_dwork;
Li Zefan24528252012-01-20 11:58:43 +08003738};
3739
Paul Menagebbcb81d2007-10-18 23:39:32 -07003740/*
Ben Blumd1d9fd32009-09-23 15:56:28 -07003741 * The following two functions "fix" the issue where there are more pids
3742 * than kmalloc will give memory for; in such cases, we use vmalloc/vfree.
3743 * TODO: replace with a kernel-wide solution to this problem
3744 */
3745#define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2))
3746static void *pidlist_allocate(int count)
3747{
3748 if (PIDLIST_TOO_LARGE(count))
3749 return vmalloc(count * sizeof(pid_t));
3750 else
3751 return kmalloc(count * sizeof(pid_t), GFP_KERNEL);
3752}
Tejun Heob1a21362013-11-29 10:42:58 -05003753
Ben Blumd1d9fd32009-09-23 15:56:28 -07003754static void pidlist_free(void *p)
3755{
3756 if (is_vmalloc_addr(p))
3757 vfree(p);
3758 else
3759 kfree(p);
3760}
Ben Blumd1d9fd32009-09-23 15:56:28 -07003761
3762/*
Tejun Heob1a21362013-11-29 10:42:58 -05003763 * Used to destroy all pidlists lingering waiting for destroy timer. None
3764 * should be left afterwards.
3765 */
3766static void cgroup_pidlist_destroy_all(struct cgroup *cgrp)
3767{
3768 struct cgroup_pidlist *l, *tmp_l;
3769
3770 mutex_lock(&cgrp->pidlist_mutex);
3771 list_for_each_entry_safe(l, tmp_l, &cgrp->pidlists, links)
3772 mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, 0);
3773 mutex_unlock(&cgrp->pidlist_mutex);
3774
3775 flush_workqueue(cgroup_pidlist_destroy_wq);
3776 BUG_ON(!list_empty(&cgrp->pidlists));
3777}
3778
3779static void cgroup_pidlist_destroy_work_fn(struct work_struct *work)
3780{
3781 struct delayed_work *dwork = to_delayed_work(work);
3782 struct cgroup_pidlist *l = container_of(dwork, struct cgroup_pidlist,
3783 destroy_dwork);
3784 struct cgroup_pidlist *tofree = NULL;
3785
3786 mutex_lock(&l->owner->pidlist_mutex);
Tejun Heob1a21362013-11-29 10:42:58 -05003787
3788 /*
Tejun Heo04502362013-11-29 10:42:59 -05003789 * Destroy iff we didn't get queued again. The state won't change
3790 * as destroy_dwork can only be queued while locked.
Tejun Heob1a21362013-11-29 10:42:58 -05003791 */
Tejun Heo04502362013-11-29 10:42:59 -05003792 if (!delayed_work_pending(dwork)) {
Tejun Heob1a21362013-11-29 10:42:58 -05003793 list_del(&l->links);
3794 pidlist_free(l->list);
3795 put_pid_ns(l->key.ns);
3796 tofree = l;
3797 }
3798
Tejun Heob1a21362013-11-29 10:42:58 -05003799 mutex_unlock(&l->owner->pidlist_mutex);
3800 kfree(tofree);
3801}
3802
3803/*
Ben Blum102a7752009-09-23 15:56:26 -07003804 * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
Li Zefan6ee211a2013-03-12 15:36:00 -07003805 * Returns the number of unique elements.
Paul Menagebbcb81d2007-10-18 23:39:32 -07003806 */
Li Zefan6ee211a2013-03-12 15:36:00 -07003807static int pidlist_uniq(pid_t *list, int length)
Paul Menagebbcb81d2007-10-18 23:39:32 -07003808{
Ben Blum102a7752009-09-23 15:56:26 -07003809 int src, dest = 1;
Ben Blum102a7752009-09-23 15:56:26 -07003810
3811 /*
3812 * we presume the 0th element is unique, so i starts at 1. trivial
3813 * edge cases first; no work needs to be done for either
3814 */
3815 if (length == 0 || length == 1)
3816 return length;
3817 /* src and dest walk down the list; dest counts unique elements */
3818 for (src = 1; src < length; src++) {
3819 /* find next unique element */
3820 while (list[src] == list[src-1]) {
3821 src++;
3822 if (src == length)
3823 goto after;
3824 }
3825 /* dest always points to where the next unique element goes */
3826 list[dest] = list[src];
3827 dest++;
3828 }
3829after:
Ben Blum102a7752009-09-23 15:56:26 -07003830 return dest;
3831}
3832
Tejun Heoafb2bc12013-11-29 10:42:59 -05003833/*
3834 * The two pid files - task and cgroup.procs - guaranteed that the result
3835 * is sorted, which forced this whole pidlist fiasco. As pid order is
3836 * different per namespace, each namespace needs differently sorted list,
3837 * making it impossible to use, for example, single rbtree of member tasks
3838 * sorted by task pointer. As pidlists can be fairly large, allocating one
3839 * per open file is dangerous, so cgroup had to implement shared pool of
3840 * pidlists keyed by cgroup and namespace.
3841 *
3842 * All this extra complexity was caused by the original implementation
3843 * committing to an entirely unnecessary property. In the long term, we
Tejun Heoaa6ec292014-07-09 10:08:08 -04003844 * want to do away with it. Explicitly scramble sort order if on the
3845 * default hierarchy so that no such expectation exists in the new
3846 * interface.
Tejun Heoafb2bc12013-11-29 10:42:59 -05003847 *
3848 * Scrambling is done by swapping every two consecutive bits, which is
3849 * non-identity one-to-one mapping which disturbs sort order sufficiently.
3850 */
3851static pid_t pid_fry(pid_t pid)
3852{
3853 unsigned a = pid & 0x55555555;
3854 unsigned b = pid & 0xAAAAAAAA;
3855
3856 return (a << 1) | (b >> 1);
3857}
3858
3859static pid_t cgroup_pid_fry(struct cgroup *cgrp, pid_t pid)
3860{
Tejun Heoaa6ec292014-07-09 10:08:08 -04003861 if (cgroup_on_dfl(cgrp))
Tejun Heoafb2bc12013-11-29 10:42:59 -05003862 return pid_fry(pid);
3863 else
3864 return pid;
3865}
3866
Ben Blum102a7752009-09-23 15:56:26 -07003867static int cmppid(const void *a, const void *b)
3868{
3869 return *(pid_t *)a - *(pid_t *)b;
3870}
3871
Tejun Heoafb2bc12013-11-29 10:42:59 -05003872static int fried_cmppid(const void *a, const void *b)
3873{
3874 return pid_fry(*(pid_t *)a) - pid_fry(*(pid_t *)b);
3875}
3876
Ben Blum72a8cb32009-09-23 15:56:27 -07003877static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp,
3878 enum cgroup_filetype type)
3879{
3880 struct cgroup_pidlist *l;
3881 /* don't need task_nsproxy() if we're looking at ourself */
Eric W. Biederman17cf22c2010-03-02 14:51:53 -08003882 struct pid_namespace *ns = task_active_pid_ns(current);
Li Zefanb70cc5f2010-03-10 15:22:12 -08003883
Tejun Heoe6b81712013-11-29 10:42:59 -05003884 lockdep_assert_held(&cgrp->pidlist_mutex);
3885
3886 list_for_each_entry(l, &cgrp->pidlists, links)
3887 if (l->key.type == type && l->key.ns == ns)
Ben Blum72a8cb32009-09-23 15:56:27 -07003888 return l;
Tejun Heoe6b81712013-11-29 10:42:59 -05003889 return NULL;
3890}
3891
Ben Blum72a8cb32009-09-23 15:56:27 -07003892/*
3893 * find the appropriate pidlist for our purpose (given procs vs tasks)
3894 * returns with the lock on that pidlist already held, and takes care
3895 * of the use count, or returns NULL with no locks held if we're out of
3896 * memory.
3897 */
Tejun Heoe6b81712013-11-29 10:42:59 -05003898static struct cgroup_pidlist *cgroup_pidlist_find_create(struct cgroup *cgrp,
3899 enum cgroup_filetype type)
Ben Blum72a8cb32009-09-23 15:56:27 -07003900{
3901 struct cgroup_pidlist *l;
Ben Blum72a8cb32009-09-23 15:56:27 -07003902
Tejun Heoe6b81712013-11-29 10:42:59 -05003903 lockdep_assert_held(&cgrp->pidlist_mutex);
3904
3905 l = cgroup_pidlist_find(cgrp, type);
3906 if (l)
3907 return l;
3908
Ben Blum72a8cb32009-09-23 15:56:27 -07003909 /* entry not found; create a new one */
Tejun Heof4f4be22013-06-12 21:04:51 -07003910 l = kzalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL);
Tejun Heoe6b81712013-11-29 10:42:59 -05003911 if (!l)
Ben Blum72a8cb32009-09-23 15:56:27 -07003912 return l;
Tejun Heoe6b81712013-11-29 10:42:59 -05003913
Tejun Heob1a21362013-11-29 10:42:58 -05003914 INIT_DELAYED_WORK(&l->destroy_dwork, cgroup_pidlist_destroy_work_fn);
Ben Blum72a8cb32009-09-23 15:56:27 -07003915 l->key.type = type;
Tejun Heoe6b81712013-11-29 10:42:59 -05003916 /* don't need task_nsproxy() if we're looking at ourself */
3917 l->key.ns = get_pid_ns(task_active_pid_ns(current));
Ben Blum72a8cb32009-09-23 15:56:27 -07003918 l->owner = cgrp;
3919 list_add(&l->links, &cgrp->pidlists);
Ben Blum72a8cb32009-09-23 15:56:27 -07003920 return l;
3921}
3922
3923/*
Ben Blum102a7752009-09-23 15:56:26 -07003924 * Load a cgroup's pidarray with either procs' tgids or tasks' pids
3925 */
Ben Blum72a8cb32009-09-23 15:56:27 -07003926static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type,
3927 struct cgroup_pidlist **lp)
Ben Blum102a7752009-09-23 15:56:26 -07003928{
3929 pid_t *array;
3930 int length;
3931 int pid, n = 0; /* used for populating the array */
Tejun Heo72ec7022013-08-08 20:11:26 -04003932 struct css_task_iter it;
Paul Menage817929e2007-10-18 23:39:36 -07003933 struct task_struct *tsk;
Ben Blum102a7752009-09-23 15:56:26 -07003934 struct cgroup_pidlist *l;
3935
Tejun Heo4bac00d2013-11-29 10:42:59 -05003936 lockdep_assert_held(&cgrp->pidlist_mutex);
3937
Ben Blum102a7752009-09-23 15:56:26 -07003938 /*
3939 * If cgroup gets more users after we read count, we won't have
3940 * enough space - tough. This race is indistinguishable to the
3941 * caller from the case that the additional cgroup users didn't
3942 * show up until sometime later on.
3943 */
3944 length = cgroup_task_count(cgrp);
Ben Blumd1d9fd32009-09-23 15:56:28 -07003945 array = pidlist_allocate(length);
Ben Blum102a7752009-09-23 15:56:26 -07003946 if (!array)
3947 return -ENOMEM;
3948 /* now, populate the array */
Tejun Heo9d800df2014-05-14 09:15:00 -04003949 css_task_iter_start(&cgrp->self, &it);
Tejun Heo72ec7022013-08-08 20:11:26 -04003950 while ((tsk = css_task_iter_next(&it))) {
Ben Blum102a7752009-09-23 15:56:26 -07003951 if (unlikely(n == length))
Paul Menage817929e2007-10-18 23:39:36 -07003952 break;
Ben Blum102a7752009-09-23 15:56:26 -07003953 /* get tgid or pid for procs or tasks file respectively */
Ben Blum72a8cb32009-09-23 15:56:27 -07003954 if (type == CGROUP_FILE_PROCS)
3955 pid = task_tgid_vnr(tsk);
3956 else
3957 pid = task_pid_vnr(tsk);
Ben Blum102a7752009-09-23 15:56:26 -07003958 if (pid > 0) /* make sure to only use valid results */
3959 array[n++] = pid;
Paul Menage817929e2007-10-18 23:39:36 -07003960 }
Tejun Heo72ec7022013-08-08 20:11:26 -04003961 css_task_iter_end(&it);
Ben Blum102a7752009-09-23 15:56:26 -07003962 length = n;
3963 /* now sort & (if procs) strip out duplicates */
Tejun Heoaa6ec292014-07-09 10:08:08 -04003964 if (cgroup_on_dfl(cgrp))
Tejun Heoafb2bc12013-11-29 10:42:59 -05003965 sort(array, length, sizeof(pid_t), fried_cmppid, NULL);
3966 else
3967 sort(array, length, sizeof(pid_t), cmppid, NULL);
Ben Blum72a8cb32009-09-23 15:56:27 -07003968 if (type == CGROUP_FILE_PROCS)
Li Zefan6ee211a2013-03-12 15:36:00 -07003969 length = pidlist_uniq(array, length);
Tejun Heoe6b81712013-11-29 10:42:59 -05003970
Tejun Heoe6b81712013-11-29 10:42:59 -05003971 l = cgroup_pidlist_find_create(cgrp, type);
Ben Blum72a8cb32009-09-23 15:56:27 -07003972 if (!l) {
Ben Blumd1d9fd32009-09-23 15:56:28 -07003973 pidlist_free(array);
Ben Blum72a8cb32009-09-23 15:56:27 -07003974 return -ENOMEM;
Ben Blum102a7752009-09-23 15:56:26 -07003975 }
Tejun Heoe6b81712013-11-29 10:42:59 -05003976
3977 /* store array, freeing old if necessary */
Ben Blumd1d9fd32009-09-23 15:56:28 -07003978 pidlist_free(l->list);
Ben Blum102a7752009-09-23 15:56:26 -07003979 l->list = array;
3980 l->length = length;
Ben Blum72a8cb32009-09-23 15:56:27 -07003981 *lp = l;
Ben Blum102a7752009-09-23 15:56:26 -07003982 return 0;
Paul Menagebbcb81d2007-10-18 23:39:32 -07003983}
3984
Balbir Singh846c7bb2007-10-18 23:39:44 -07003985/**
Li Zefana043e3b2008-02-23 15:24:09 -08003986 * cgroupstats_build - build and fill cgroupstats
Balbir Singh846c7bb2007-10-18 23:39:44 -07003987 * @stats: cgroupstats to fill information into
3988 * @dentry: A dentry entry belonging to the cgroup for which stats have
3989 * been requested.
Li Zefana043e3b2008-02-23 15:24:09 -08003990 *
3991 * Build and fill cgroupstats so that taskstats can export it to user
3992 * space.
Balbir Singh846c7bb2007-10-18 23:39:44 -07003993 */
3994int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry)
3995{
Tejun Heo2bd59d42014-02-11 11:52:49 -05003996 struct kernfs_node *kn = kernfs_node_from_dentry(dentry);
Paul Menagebd89aab2007-10-18 23:40:44 -07003997 struct cgroup *cgrp;
Tejun Heo72ec7022013-08-08 20:11:26 -04003998 struct css_task_iter it;
Balbir Singh846c7bb2007-10-18 23:39:44 -07003999 struct task_struct *tsk;
Li Zefan33d283b2008-11-19 15:36:48 -08004000
Tejun Heo2bd59d42014-02-11 11:52:49 -05004001 /* it should be kernfs_node belonging to cgroupfs and is a directory */
4002 if (dentry->d_sb->s_type != &cgroup_fs_type || !kn ||
4003 kernfs_type(kn) != KERNFS_DIR)
4004 return -EINVAL;
Balbir Singh846c7bb2007-10-18 23:39:44 -07004005
Li Zefanbad34662014-02-14 16:54:28 +08004006 mutex_lock(&cgroup_mutex);
4007
Tejun Heo2bd59d42014-02-11 11:52:49 -05004008 /*
4009 * We aren't being called from kernfs and there's no guarantee on
Tejun Heoec903c02014-05-13 12:11:01 -04004010 * @kn->priv's validity. For this and css_tryget_online_from_dir(),
Tejun Heo2bd59d42014-02-11 11:52:49 -05004011 * @kn->priv is RCU safe. Let's do the RCU dancing.
4012 */
4013 rcu_read_lock();
4014 cgrp = rcu_dereference(kn->priv);
Li Zefanbad34662014-02-14 16:54:28 +08004015 if (!cgrp || cgroup_is_dead(cgrp)) {
Tejun Heo2bd59d42014-02-11 11:52:49 -05004016 rcu_read_unlock();
Li Zefanbad34662014-02-14 16:54:28 +08004017 mutex_unlock(&cgroup_mutex);
Tejun Heo2bd59d42014-02-11 11:52:49 -05004018 return -ENOENT;
4019 }
Li Zefanbad34662014-02-14 16:54:28 +08004020 rcu_read_unlock();
Balbir Singh846c7bb2007-10-18 23:39:44 -07004021
Tejun Heo9d800df2014-05-14 09:15:00 -04004022 css_task_iter_start(&cgrp->self, &it);
Tejun Heo72ec7022013-08-08 20:11:26 -04004023 while ((tsk = css_task_iter_next(&it))) {
Balbir Singh846c7bb2007-10-18 23:39:44 -07004024 switch (tsk->state) {
4025 case TASK_RUNNING:
4026 stats->nr_running++;
4027 break;
4028 case TASK_INTERRUPTIBLE:
4029 stats->nr_sleeping++;
4030 break;
4031 case TASK_UNINTERRUPTIBLE:
4032 stats->nr_uninterruptible++;
4033 break;
4034 case TASK_STOPPED:
4035 stats->nr_stopped++;
4036 break;
4037 default:
4038 if (delayacct_is_task_waiting_on_io(tsk))
4039 stats->nr_io_wait++;
4040 break;
4041 }
4042 }
Tejun Heo72ec7022013-08-08 20:11:26 -04004043 css_task_iter_end(&it);
Balbir Singh846c7bb2007-10-18 23:39:44 -07004044
Li Zefanbad34662014-02-14 16:54:28 +08004045 mutex_unlock(&cgroup_mutex);
Tejun Heo2bd59d42014-02-11 11:52:49 -05004046 return 0;
Balbir Singh846c7bb2007-10-18 23:39:44 -07004047}
4048
Paul Menage8f3ff202009-09-23 15:56:25 -07004049
Paul Menagecc31edc2008-10-18 20:28:04 -07004050/*
Ben Blum102a7752009-09-23 15:56:26 -07004051 * seq_file methods for the tasks/procs files. The seq_file position is the
Paul Menagecc31edc2008-10-18 20:28:04 -07004052 * next pid to display; the seq_file iterator is a pointer to the pid
Ben Blum102a7752009-09-23 15:56:26 -07004053 * in the cgroup->l->list array.
Paul Menagecc31edc2008-10-18 20:28:04 -07004054 */
4055
Ben Blum102a7752009-09-23 15:56:26 -07004056static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos)
Paul Menagecc31edc2008-10-18 20:28:04 -07004057{
4058 /*
4059 * Initially we receive a position value that corresponds to
4060 * one more than the last pid shown (or 0 on the first call or
4061 * after a seek to the start). Use a binary-search to find the
4062 * next pid to display, if any
4063 */
Tejun Heo2bd59d42014-02-11 11:52:49 -05004064 struct kernfs_open_file *of = s->private;
Tejun Heo7da11272013-12-05 12:28:04 -05004065 struct cgroup *cgrp = seq_css(s)->cgroup;
Tejun Heo4bac00d2013-11-29 10:42:59 -05004066 struct cgroup_pidlist *l;
Tejun Heo7da11272013-12-05 12:28:04 -05004067 enum cgroup_filetype type = seq_cft(s)->private;
Paul Menagecc31edc2008-10-18 20:28:04 -07004068 int index = 0, pid = *pos;
Tejun Heo4bac00d2013-11-29 10:42:59 -05004069 int *iter, ret;
Paul Menagecc31edc2008-10-18 20:28:04 -07004070
Tejun Heo4bac00d2013-11-29 10:42:59 -05004071 mutex_lock(&cgrp->pidlist_mutex);
4072
4073 /*
Tejun Heo5d224442013-12-05 12:28:04 -05004074 * !NULL @of->priv indicates that this isn't the first start()
Tejun Heo4bac00d2013-11-29 10:42:59 -05004075 * after open. If the matching pidlist is around, we can use that.
Tejun Heo5d224442013-12-05 12:28:04 -05004076 * Look for it. Note that @of->priv can't be used directly. It
Tejun Heo4bac00d2013-11-29 10:42:59 -05004077 * could already have been destroyed.
4078 */
Tejun Heo5d224442013-12-05 12:28:04 -05004079 if (of->priv)
4080 of->priv = cgroup_pidlist_find(cgrp, type);
Tejun Heo4bac00d2013-11-29 10:42:59 -05004081
4082 /*
4083 * Either this is the first start() after open or the matching
4084 * pidlist has been destroyed inbetween. Create a new one.
4085 */
Tejun Heo5d224442013-12-05 12:28:04 -05004086 if (!of->priv) {
4087 ret = pidlist_array_load(cgrp, type,
4088 (struct cgroup_pidlist **)&of->priv);
Tejun Heo4bac00d2013-11-29 10:42:59 -05004089 if (ret)
4090 return ERR_PTR(ret);
4091 }
Tejun Heo5d224442013-12-05 12:28:04 -05004092 l = of->priv;
Tejun Heo4bac00d2013-11-29 10:42:59 -05004093
Paul Menagecc31edc2008-10-18 20:28:04 -07004094 if (pid) {
Ben Blum102a7752009-09-23 15:56:26 -07004095 int end = l->length;
Stephen Rothwell20777762008-10-21 16:11:20 +11004096
Paul Menagecc31edc2008-10-18 20:28:04 -07004097 while (index < end) {
4098 int mid = (index + end) / 2;
Tejun Heoafb2bc12013-11-29 10:42:59 -05004099 if (cgroup_pid_fry(cgrp, l->list[mid]) == pid) {
Paul Menagecc31edc2008-10-18 20:28:04 -07004100 index = mid;
4101 break;
Tejun Heoafb2bc12013-11-29 10:42:59 -05004102 } else if (cgroup_pid_fry(cgrp, l->list[mid]) <= pid)
Paul Menagecc31edc2008-10-18 20:28:04 -07004103 index = mid + 1;
4104 else
4105 end = mid;
4106 }
4107 }
4108 /* If we're off the end of the array, we're done */
Ben Blum102a7752009-09-23 15:56:26 -07004109 if (index >= l->length)
Paul Menagecc31edc2008-10-18 20:28:04 -07004110 return NULL;
4111 /* Update the abstract position to be the actual pid that we found */
Ben Blum102a7752009-09-23 15:56:26 -07004112 iter = l->list + index;
Tejun Heoafb2bc12013-11-29 10:42:59 -05004113 *pos = cgroup_pid_fry(cgrp, *iter);
Paul Menagecc31edc2008-10-18 20:28:04 -07004114 return iter;
Paul Menagebbcb81d2007-10-18 23:39:32 -07004115}
4116
Ben Blum102a7752009-09-23 15:56:26 -07004117static void cgroup_pidlist_stop(struct seq_file *s, void *v)
Paul Menagecc31edc2008-10-18 20:28:04 -07004118{
Tejun Heo2bd59d42014-02-11 11:52:49 -05004119 struct kernfs_open_file *of = s->private;
Tejun Heo5d224442013-12-05 12:28:04 -05004120 struct cgroup_pidlist *l = of->priv;
Tejun Heo62236852013-11-29 10:42:58 -05004121
Tejun Heo5d224442013-12-05 12:28:04 -05004122 if (l)
4123 mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork,
Tejun Heo04502362013-11-29 10:42:59 -05004124 CGROUP_PIDLIST_DESTROY_DELAY);
Tejun Heo7da11272013-12-05 12:28:04 -05004125 mutex_unlock(&seq_css(s)->cgroup->pidlist_mutex);
Paul Menagecc31edc2008-10-18 20:28:04 -07004126}
4127
Ben Blum102a7752009-09-23 15:56:26 -07004128static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos)
Paul Menagecc31edc2008-10-18 20:28:04 -07004129{
Tejun Heo2bd59d42014-02-11 11:52:49 -05004130 struct kernfs_open_file *of = s->private;
Tejun Heo5d224442013-12-05 12:28:04 -05004131 struct cgroup_pidlist *l = of->priv;
Ben Blum102a7752009-09-23 15:56:26 -07004132 pid_t *p = v;
4133 pid_t *end = l->list + l->length;
Paul Menagecc31edc2008-10-18 20:28:04 -07004134 /*
4135 * Advance to the next pid in the array. If this goes off the
4136 * end, we're done
4137 */
4138 p++;
4139 if (p >= end) {
4140 return NULL;
4141 } else {
Tejun Heo7da11272013-12-05 12:28:04 -05004142 *pos = cgroup_pid_fry(seq_css(s)->cgroup, *p);
Paul Menagecc31edc2008-10-18 20:28:04 -07004143 return p;
4144 }
4145}
4146
Ben Blum102a7752009-09-23 15:56:26 -07004147static int cgroup_pidlist_show(struct seq_file *s, void *v)
Paul Menagecc31edc2008-10-18 20:28:04 -07004148{
4149 return seq_printf(s, "%d\n", *(int *)v);
4150}
4151
Tejun Heo182446d2013-08-08 20:11:24 -04004152static u64 cgroup_read_notify_on_release(struct cgroup_subsys_state *css,
4153 struct cftype *cft)
Paul Menage81a6a5c2007-10-18 23:39:38 -07004154{
Tejun Heo182446d2013-08-08 20:11:24 -04004155 return notify_on_release(css->cgroup);
Paul Menage81a6a5c2007-10-18 23:39:38 -07004156}
4157
Tejun Heo182446d2013-08-08 20:11:24 -04004158static int cgroup_write_notify_on_release(struct cgroup_subsys_state *css,
4159 struct cftype *cft, u64 val)
Paul Menage6379c102008-07-25 01:47:01 -07004160{
Paul Menage6379c102008-07-25 01:47:01 -07004161 if (val)
Tejun Heo182446d2013-08-08 20:11:24 -04004162 set_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags);
Paul Menage6379c102008-07-25 01:47:01 -07004163 else
Tejun Heo182446d2013-08-08 20:11:24 -04004164 clear_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags);
Paul Menage6379c102008-07-25 01:47:01 -07004165 return 0;
4166}
4167
Tejun Heo182446d2013-08-08 20:11:24 -04004168static u64 cgroup_clone_children_read(struct cgroup_subsys_state *css,
4169 struct cftype *cft)
Daniel Lezcano97978e62010-10-27 15:33:35 -07004170{
Tejun Heo182446d2013-08-08 20:11:24 -04004171 return test_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags);
Daniel Lezcano97978e62010-10-27 15:33:35 -07004172}
4173
Tejun Heo182446d2013-08-08 20:11:24 -04004174static int cgroup_clone_children_write(struct cgroup_subsys_state *css,
4175 struct cftype *cft, u64 val)
Daniel Lezcano97978e62010-10-27 15:33:35 -07004176{
4177 if (val)
Tejun Heo182446d2013-08-08 20:11:24 -04004178 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags);
Daniel Lezcano97978e62010-10-27 15:33:35 -07004179 else
Tejun Heo182446d2013-08-08 20:11:24 -04004180 clear_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags);
Daniel Lezcano97978e62010-10-27 15:33:35 -07004181 return 0;
4182}
4183
Tejun Heoa14c6872014-07-15 11:05:09 -04004184/* cgroup core interface files for the default hierarchy */
4185static struct cftype cgroup_dfl_base_files[] = {
4186 {
4187 .name = "cgroup.procs",
4188 .seq_start = cgroup_pidlist_start,
4189 .seq_next = cgroup_pidlist_next,
4190 .seq_stop = cgroup_pidlist_stop,
4191 .seq_show = cgroup_pidlist_show,
4192 .private = CGROUP_FILE_PROCS,
4193 .write = cgroup_procs_write,
4194 .mode = S_IRUGO | S_IWUSR,
4195 },
4196 {
4197 .name = "cgroup.controllers",
4198 .flags = CFTYPE_ONLY_ON_ROOT,
4199 .seq_show = cgroup_root_controllers_show,
4200 },
4201 {
4202 .name = "cgroup.controllers",
4203 .flags = CFTYPE_NOT_ON_ROOT,
4204 .seq_show = cgroup_controllers_show,
4205 },
4206 {
4207 .name = "cgroup.subtree_control",
4208 .seq_show = cgroup_subtree_control_show,
4209 .write = cgroup_subtree_control_write,
4210 },
4211 {
4212 .name = "cgroup.populated",
4213 .flags = CFTYPE_NOT_ON_ROOT,
4214 .seq_show = cgroup_populated_show,
4215 },
4216 { } /* terminate */
4217};
4218
4219/* cgroup core interface files for the legacy hierarchies */
4220static struct cftype cgroup_legacy_base_files[] = {
Paul Menage81a6a5c2007-10-18 23:39:38 -07004221 {
Tejun Heod5c56ce2013-06-03 19:14:34 -07004222 .name = "cgroup.procs",
Tejun Heo6612f052013-12-05 12:28:04 -05004223 .seq_start = cgroup_pidlist_start,
4224 .seq_next = cgroup_pidlist_next,
4225 .seq_stop = cgroup_pidlist_stop,
4226 .seq_show = cgroup_pidlist_show,
Tejun Heo5d224442013-12-05 12:28:04 -05004227 .private = CGROUP_FILE_PROCS,
Tejun Heoacbef752014-05-13 12:16:22 -04004228 .write = cgroup_procs_write,
Ben Blum74a11662011-05-26 16:25:20 -07004229 .mode = S_IRUGO | S_IWUSR,
Ben Blum102a7752009-09-23 15:56:26 -07004230 },
Paul Menage81a6a5c2007-10-18 23:39:38 -07004231 {
Daniel Lezcano97978e62010-10-27 15:33:35 -07004232 .name = "cgroup.clone_children",
4233 .read_u64 = cgroup_clone_children_read,
4234 .write_u64 = cgroup_clone_children_write,
4235 },
Tejun Heo6e6ff252012-04-01 12:09:55 -07004236 {
Tejun Heo873fe092013-04-14 20:15:26 -07004237 .name = "cgroup.sane_behavior",
4238 .flags = CFTYPE_ONLY_ON_ROOT,
Tejun Heo2da8ca82013-12-05 12:28:04 -05004239 .seq_show = cgroup_sane_behavior_show,
Tejun Heo873fe092013-04-14 20:15:26 -07004240 },
Tejun Heof8f22e52014-04-23 11:13:16 -04004241 {
Tejun Heod5c56ce2013-06-03 19:14:34 -07004242 .name = "tasks",
Tejun Heo6612f052013-12-05 12:28:04 -05004243 .seq_start = cgroup_pidlist_start,
4244 .seq_next = cgroup_pidlist_next,
4245 .seq_stop = cgroup_pidlist_stop,
4246 .seq_show = cgroup_pidlist_show,
Tejun Heo5d224442013-12-05 12:28:04 -05004247 .private = CGROUP_FILE_TASKS,
Tejun Heoacbef752014-05-13 12:16:22 -04004248 .write = cgroup_tasks_write,
Tejun Heod5c56ce2013-06-03 19:14:34 -07004249 .mode = S_IRUGO | S_IWUSR,
4250 },
4251 {
4252 .name = "notify_on_release",
Tejun Heod5c56ce2013-06-03 19:14:34 -07004253 .read_u64 = cgroup_read_notify_on_release,
4254 .write_u64 = cgroup_write_notify_on_release,
4255 },
Tejun Heo873fe092013-04-14 20:15:26 -07004256 {
Tejun Heo6e6ff252012-04-01 12:09:55 -07004257 .name = "release_agent",
Tejun Heoa14c6872014-07-15 11:05:09 -04004258 .flags = CFTYPE_ONLY_ON_ROOT,
Tejun Heo2da8ca82013-12-05 12:28:04 -05004259 .seq_show = cgroup_release_agent_show,
Tejun Heo451af502014-05-13 12:16:21 -04004260 .write = cgroup_release_agent_write,
Tejun Heo5f469902014-02-11 11:52:48 -05004261 .max_write_len = PATH_MAX - 1,
Tejun Heo6e6ff252012-04-01 12:09:55 -07004262 },
Tejun Heodb0416b2012-04-01 12:09:55 -07004263 { } /* terminate */
Paul Menagebbcb81d2007-10-18 23:39:32 -07004264};
4265
Aristeu Rozanski13af07d2012-08-23 16:53:29 -04004266/**
Tejun Heo628f7cd2013-06-28 16:24:11 -07004267 * cgroup_populate_dir - create subsys files in a cgroup directory
Aristeu Rozanski13af07d2012-08-23 16:53:29 -04004268 * @cgrp: target cgroup
Aristeu Rozanski13af07d2012-08-23 16:53:29 -04004269 * @subsys_mask: mask of the subsystem ids whose files should be added
Tejun Heobee55092013-06-28 16:24:11 -07004270 *
4271 * On failure, no file is added.
Aristeu Rozanski13af07d2012-08-23 16:53:29 -04004272 */
Tejun Heo69dfa002014-05-04 15:09:13 -04004273static int cgroup_populate_dir(struct cgroup *cgrp, unsigned int subsys_mask)
Paul Menageddbcc7e2007-10-18 23:39:30 -07004274{
Paul Menageddbcc7e2007-10-18 23:39:30 -07004275 struct cgroup_subsys *ss;
Tejun Heob420ba72013-07-12 12:34:02 -07004276 int i, ret = 0;
Paul Menagebbcb81d2007-10-18 23:39:32 -07004277
Tejun Heo8e3f6542012-04-01 12:09:55 -07004278 /* process cftsets of each subsystem */
Tejun Heob420ba72013-07-12 12:34:02 -07004279 for_each_subsys(ss, i) {
Tejun Heo0adb0702014-02-12 09:29:48 -05004280 struct cftype *cfts;
Tejun Heob420ba72013-07-12 12:34:02 -07004281
Tejun Heo69dfa002014-05-04 15:09:13 -04004282 if (!(subsys_mask & (1 << i)))
Aristeu Rozanski13af07d2012-08-23 16:53:29 -04004283 continue;
Tejun Heo8e3f6542012-04-01 12:09:55 -07004284
Tejun Heo0adb0702014-02-12 09:29:48 -05004285 list_for_each_entry(cfts, &ss->cfts, node) {
4286 ret = cgroup_addrm_files(cgrp, cfts, true);
Tejun Heobee55092013-06-28 16:24:11 -07004287 if (ret < 0)
4288 goto err;
4289 }
Paul Menageddbcc7e2007-10-18 23:39:30 -07004290 }
Paul Menageddbcc7e2007-10-18 23:39:30 -07004291 return 0;
Tejun Heobee55092013-06-28 16:24:11 -07004292err:
4293 cgroup_clear_dir(cgrp, subsys_mask);
4294 return ret;
Paul Menageddbcc7e2007-10-18 23:39:30 -07004295}
4296
Tejun Heo0c21ead2013-08-13 20:22:51 -04004297/*
4298 * css destruction is four-stage process.
4299 *
4300 * 1. Destruction starts. Killing of the percpu_ref is initiated.
4301 * Implemented in kill_css().
4302 *
4303 * 2. When the percpu_ref is confirmed to be visible as killed on all CPUs
Tejun Heoec903c02014-05-13 12:11:01 -04004304 * and thus css_tryget_online() is guaranteed to fail, the css can be
4305 * offlined by invoking offline_css(). After offlining, the base ref is
4306 * put. Implemented in css_killed_work_fn().
Tejun Heo0c21ead2013-08-13 20:22:51 -04004307 *
4308 * 3. When the percpu_ref reaches zero, the only possible remaining
4309 * accessors are inside RCU read sections. css_release() schedules the
4310 * RCU callback.
4311 *
4312 * 4. After the grace period, the css can be freed. Implemented in
4313 * css_free_work_fn().
4314 *
4315 * It is actually hairier because both step 2 and 4 require process context
4316 * and thus involve punting to css->destroy_work adding two additional
4317 * steps to the already complex sequence.
4318 */
Tejun Heo35ef10d2013-08-13 11:01:54 -04004319static void css_free_work_fn(struct work_struct *work)
Tejun Heo48ddbe12012-04-01 12:09:56 -07004320{
4321 struct cgroup_subsys_state *css =
Tejun Heo35ef10d2013-08-13 11:01:54 -04004322 container_of(work, struct cgroup_subsys_state, destroy_work);
Tejun Heo0c21ead2013-08-13 20:22:51 -04004323 struct cgroup *cgrp = css->cgroup;
Tejun Heo48ddbe12012-04-01 12:09:56 -07004324
Tejun Heo9a1049d2014-06-28 08:10:14 -04004325 percpu_ref_exit(&css->refcnt);
4326
Tejun Heo9d755d32014-05-14 09:15:02 -04004327 if (css->ss) {
4328 /* css free path */
4329 if (css->parent)
4330 css_put(css->parent);
Tejun Heo0ae78e02013-08-13 11:01:54 -04004331
Tejun Heo9d755d32014-05-14 09:15:02 -04004332 css->ss->css_free(css);
4333 cgroup_put(cgrp);
4334 } else {
4335 /* cgroup free path */
4336 atomic_dec(&cgrp->root->nr_cgrps);
4337 cgroup_pidlist_destroy_all(cgrp);
Zefan Li971ff492014-09-18 16:06:19 +08004338 cancel_work_sync(&cgrp->release_agent_work);
Tejun Heo9d755d32014-05-14 09:15:02 -04004339
Tejun Heod51f39b2014-05-16 13:22:48 -04004340 if (cgroup_parent(cgrp)) {
Tejun Heo9d755d32014-05-14 09:15:02 -04004341 /*
4342 * We get a ref to the parent, and put the ref when
4343 * this cgroup is being freed, so it's guaranteed
4344 * that the parent won't be destroyed before its
4345 * children.
4346 */
Tejun Heod51f39b2014-05-16 13:22:48 -04004347 cgroup_put(cgroup_parent(cgrp));
Tejun Heo9d755d32014-05-14 09:15:02 -04004348 kernfs_put(cgrp->kn);
4349 kfree(cgrp);
4350 } else {
4351 /*
4352 * This is root cgroup's refcnt reaching zero,
4353 * which indicates that the root should be
4354 * released.
4355 */
4356 cgroup_destroy_root(cgrp->root);
4357 }
4358 }
Tejun Heo0c21ead2013-08-13 20:22:51 -04004359}
4360
4361static void css_free_rcu_fn(struct rcu_head *rcu_head)
4362{
4363 struct cgroup_subsys_state *css =
4364 container_of(rcu_head, struct cgroup_subsys_state, rcu_head);
4365
Tejun Heo0c21ead2013-08-13 20:22:51 -04004366 INIT_WORK(&css->destroy_work, css_free_work_fn);
Tejun Heoe5fca242013-11-22 17:14:39 -05004367 queue_work(cgroup_destroy_wq, &css->destroy_work);
Tejun Heo48ddbe12012-04-01 12:09:56 -07004368}
4369
Tejun Heo25e15d82014-05-14 09:15:02 -04004370static void css_release_work_fn(struct work_struct *work)
Tejun Heod3daf282013-06-13 19:39:16 -07004371{
4372 struct cgroup_subsys_state *css =
Tejun Heo25e15d82014-05-14 09:15:02 -04004373 container_of(work, struct cgroup_subsys_state, destroy_work);
Tejun Heo15a4c832014-05-04 15:09:14 -04004374 struct cgroup_subsys *ss = css->ss;
Tejun Heo9d755d32014-05-14 09:15:02 -04004375 struct cgroup *cgrp = css->cgroup;
Tejun Heod3daf282013-06-13 19:39:16 -07004376
Tejun Heo1fed1b22014-05-16 13:22:49 -04004377 mutex_lock(&cgroup_mutex);
4378
Tejun Heode3f0342014-05-16 13:22:49 -04004379 css->flags |= CSS_RELEASED;
Tejun Heo1fed1b22014-05-16 13:22:49 -04004380 list_del_rcu(&css->sibling);
4381
Tejun Heo9d755d32014-05-14 09:15:02 -04004382 if (ss) {
4383 /* css release path */
4384 cgroup_idr_remove(&ss->css_idr, css->id);
4385 } else {
4386 /* cgroup release path */
Tejun Heo9d755d32014-05-14 09:15:02 -04004387 cgroup_idr_remove(&cgrp->root->cgroup_idr, cgrp->id);
4388 cgrp->id = -1;
Li Zefana4189482014-09-04 14:43:07 +08004389
4390 /*
4391 * There are two control paths which try to determine
4392 * cgroup from dentry without going through kernfs -
4393 * cgroupstats_build() and css_tryget_online_from_dir().
4394 * Those are supported by RCU protecting clearing of
4395 * cgrp->kn->priv backpointer.
4396 */
4397 RCU_INIT_POINTER(*(void __rcu __force **)&cgrp->kn->priv, NULL);
Tejun Heo9d755d32014-05-14 09:15:02 -04004398 }
Tejun Heo15a4c832014-05-04 15:09:14 -04004399
Tejun Heo1fed1b22014-05-16 13:22:49 -04004400 mutex_unlock(&cgroup_mutex);
4401
Tejun Heo0c21ead2013-08-13 20:22:51 -04004402 call_rcu(&css->rcu_head, css_free_rcu_fn);
Tejun Heod3daf282013-06-13 19:39:16 -07004403}
4404
Paul Menageddbcc7e2007-10-18 23:39:30 -07004405static void css_release(struct percpu_ref *ref)
4406{
4407 struct cgroup_subsys_state *css =
4408 container_of(ref, struct cgroup_subsys_state, refcnt);
4409
Tejun Heo25e15d82014-05-14 09:15:02 -04004410 INIT_WORK(&css->destroy_work, css_release_work_fn);
4411 queue_work(cgroup_destroy_wq, &css->destroy_work);
Paul Menageddbcc7e2007-10-18 23:39:30 -07004412}
4413
Tejun Heoddfcada2014-05-04 15:09:14 -04004414static void init_and_link_css(struct cgroup_subsys_state *css,
4415 struct cgroup_subsys *ss, struct cgroup *cgrp)
Paul Menageddbcc7e2007-10-18 23:39:30 -07004416{
Tejun Heo0cb51d72014-05-16 13:22:49 -04004417 lockdep_assert_held(&cgroup_mutex);
4418
Tejun Heoddfcada2014-05-04 15:09:14 -04004419 cgroup_get(cgrp);
4420
Tejun Heod5c419b2014-05-16 13:22:48 -04004421 memset(css, 0, sizeof(*css));
Paul Menagebd89aab2007-10-18 23:40:44 -07004422 css->cgroup = cgrp;
Tejun Heo72c97e52013-08-08 20:11:22 -04004423 css->ss = ss;
Tejun Heod5c419b2014-05-16 13:22:48 -04004424 INIT_LIST_HEAD(&css->sibling);
4425 INIT_LIST_HEAD(&css->children);
Tejun Heo0cb51d72014-05-16 13:22:49 -04004426 css->serial_nr = css_serial_nr_next++;
Tejun Heo48ddbe12012-04-01 12:09:56 -07004427
Tejun Heod51f39b2014-05-16 13:22:48 -04004428 if (cgroup_parent(cgrp)) {
4429 css->parent = cgroup_css(cgroup_parent(cgrp), ss);
Tejun Heoddfcada2014-05-04 15:09:14 -04004430 css_get(css->parent);
Tejun Heoddfcada2014-05-04 15:09:14 -04004431 }
Tejun Heo0ae78e02013-08-13 11:01:54 -04004432
Tejun Heoca8bdca2013-08-26 18:40:56 -04004433 BUG_ON(cgroup_css(cgrp, ss));
Paul Menageddbcc7e2007-10-18 23:39:30 -07004434}
4435
Li Zefan2a4ac632013-07-31 16:16:40 +08004436/* invoke ->css_online() on a new CSS and mark it online if successful */
Tejun Heo623f9262013-08-13 11:01:55 -04004437static int online_css(struct cgroup_subsys_state *css)
Tejun Heoa31f2d32012-11-19 08:13:37 -08004438{
Tejun Heo623f9262013-08-13 11:01:55 -04004439 struct cgroup_subsys *ss = css->ss;
Tejun Heob1929db2012-11-19 08:13:38 -08004440 int ret = 0;
4441
Tejun Heoa31f2d32012-11-19 08:13:37 -08004442 lockdep_assert_held(&cgroup_mutex);
4443
Tejun Heo92fb9742012-11-19 08:13:38 -08004444 if (ss->css_online)
Tejun Heoeb954192013-08-08 20:11:23 -04004445 ret = ss->css_online(css);
Tejun Heoae7f1642013-08-13 20:22:50 -04004446 if (!ret) {
Tejun Heoeb954192013-08-08 20:11:23 -04004447 css->flags |= CSS_ONLINE;
Tejun Heoaec25022014-02-08 10:36:58 -05004448 rcu_assign_pointer(css->cgroup->subsys[ss->id], css);
Tejun Heoae7f1642013-08-13 20:22:50 -04004449 }
Tejun Heob1929db2012-11-19 08:13:38 -08004450 return ret;
Tejun Heoa31f2d32012-11-19 08:13:37 -08004451}
4452
Li Zefan2a4ac632013-07-31 16:16:40 +08004453/* if the CSS is online, invoke ->css_offline() on it and mark it offline */
Tejun Heo623f9262013-08-13 11:01:55 -04004454static void offline_css(struct cgroup_subsys_state *css)
Tejun Heoa31f2d32012-11-19 08:13:37 -08004455{
Tejun Heo623f9262013-08-13 11:01:55 -04004456 struct cgroup_subsys *ss = css->ss;
Tejun Heoa31f2d32012-11-19 08:13:37 -08004457
4458 lockdep_assert_held(&cgroup_mutex);
4459
4460 if (!(css->flags & CSS_ONLINE))
4461 return;
4462
Li Zefand7eeac12013-03-12 15:35:59 -07004463 if (ss->css_offline)
Tejun Heoeb954192013-08-08 20:11:23 -04004464 ss->css_offline(css);
Tejun Heoa31f2d32012-11-19 08:13:37 -08004465
Tejun Heoeb954192013-08-08 20:11:23 -04004466 css->flags &= ~CSS_ONLINE;
Tejun Heoe3297802014-04-23 11:13:15 -04004467 RCU_INIT_POINTER(css->cgroup->subsys[ss->id], NULL);
Tejun Heof8f22e52014-04-23 11:13:16 -04004468
4469 wake_up_all(&css->cgroup->offline_waitq);
Tejun Heoa31f2d32012-11-19 08:13:37 -08004470}
4471
Tejun Heoc81c925a2013-12-06 15:11:56 -05004472/**
4473 * create_css - create a cgroup_subsys_state
4474 * @cgrp: the cgroup new css will be associated with
4475 * @ss: the subsys of new css
Tejun Heof63070d2014-07-08 18:02:57 -04004476 * @visible: whether to create control knobs for the new css or not
Tejun Heoc81c925a2013-12-06 15:11:56 -05004477 *
4478 * Create a new css associated with @cgrp - @ss pair. On success, the new
Tejun Heof63070d2014-07-08 18:02:57 -04004479 * css is online and installed in @cgrp with all interface files created if
4480 * @visible. Returns 0 on success, -errno on failure.
Tejun Heoc81c925a2013-12-06 15:11:56 -05004481 */
Tejun Heof63070d2014-07-08 18:02:57 -04004482static int create_css(struct cgroup *cgrp, struct cgroup_subsys *ss,
4483 bool visible)
Tejun Heoc81c925a2013-12-06 15:11:56 -05004484{
Tejun Heod51f39b2014-05-16 13:22:48 -04004485 struct cgroup *parent = cgroup_parent(cgrp);
Tejun Heo1fed1b22014-05-16 13:22:49 -04004486 struct cgroup_subsys_state *parent_css = cgroup_css(parent, ss);
Tejun Heoc81c925a2013-12-06 15:11:56 -05004487 struct cgroup_subsys_state *css;
4488 int err;
4489
Tejun Heoc81c925a2013-12-06 15:11:56 -05004490 lockdep_assert_held(&cgroup_mutex);
4491
Tejun Heo1fed1b22014-05-16 13:22:49 -04004492 css = ss->css_alloc(parent_css);
Tejun Heoc81c925a2013-12-06 15:11:56 -05004493 if (IS_ERR(css))
4494 return PTR_ERR(css);
4495
Tejun Heoddfcada2014-05-04 15:09:14 -04004496 init_and_link_css(css, ss, cgrp);
Tejun Heoa2bed822014-05-04 15:09:14 -04004497
Tejun Heo2aad2a82014-09-24 13:31:50 -04004498 err = percpu_ref_init(&css->refcnt, css_release, 0, GFP_KERNEL);
Tejun Heoc81c925a2013-12-06 15:11:56 -05004499 if (err)
Li Zefan3eb59ec2014-03-18 17:02:36 +08004500 goto err_free_css;
Tejun Heoc81c925a2013-12-06 15:11:56 -05004501
Tejun Heo15a4c832014-05-04 15:09:14 -04004502 err = cgroup_idr_alloc(&ss->css_idr, NULL, 2, 0, GFP_NOWAIT);
4503 if (err < 0)
4504 goto err_free_percpu_ref;
4505 css->id = err;
Tejun Heoc81c925a2013-12-06 15:11:56 -05004506
Tejun Heof63070d2014-07-08 18:02:57 -04004507 if (visible) {
4508 err = cgroup_populate_dir(cgrp, 1 << ss->id);
4509 if (err)
4510 goto err_free_id;
4511 }
Tejun Heo15a4c832014-05-04 15:09:14 -04004512
4513 /* @css is ready to be brought online now, make it visible */
Tejun Heo1fed1b22014-05-16 13:22:49 -04004514 list_add_tail_rcu(&css->sibling, &parent_css->children);
Tejun Heo15a4c832014-05-04 15:09:14 -04004515 cgroup_idr_replace(&ss->css_idr, css, css->id);
Tejun Heoc81c925a2013-12-06 15:11:56 -05004516
4517 err = online_css(css);
4518 if (err)
Tejun Heo1fed1b22014-05-16 13:22:49 -04004519 goto err_list_del;
Tejun Heo94419622014-03-19 10:23:54 -04004520
Tejun Heoc81c925a2013-12-06 15:11:56 -05004521 if (ss->broken_hierarchy && !ss->warned_broken_hierarchy &&
Tejun Heod51f39b2014-05-16 13:22:48 -04004522 cgroup_parent(parent)) {
Joe Perchesed3d2612014-04-25 18:28:03 -04004523 pr_warn("%s (%d) created nested cgroup for controller \"%s\" which has incomplete hierarchy support. Nested cgroups may change behavior in the future.\n",
Jianyu Zhana2a1f9e2014-04-25 18:28:03 -04004524 current->comm, current->pid, ss->name);
Tejun Heoc81c925a2013-12-06 15:11:56 -05004525 if (!strcmp(ss->name, "memory"))
Joe Perchesed3d2612014-04-25 18:28:03 -04004526 pr_warn("\"memory\" requires setting use_hierarchy to 1 on the root\n");
Tejun Heoc81c925a2013-12-06 15:11:56 -05004527 ss->warned_broken_hierarchy = true;
4528 }
4529
4530 return 0;
4531
Tejun Heo1fed1b22014-05-16 13:22:49 -04004532err_list_del:
4533 list_del_rcu(&css->sibling);
Linus Torvalds32d01dc2014-04-03 13:05:42 -07004534 cgroup_clear_dir(css->cgroup, 1 << css->ss->id);
Tejun Heo15a4c832014-05-04 15:09:14 -04004535err_free_id:
4536 cgroup_idr_remove(&ss->css_idr, css->id);
Li Zefan3eb59ec2014-03-18 17:02:36 +08004537err_free_percpu_ref:
Tejun Heo9a1049d2014-06-28 08:10:14 -04004538 percpu_ref_exit(&css->refcnt);
Li Zefan3eb59ec2014-03-18 17:02:36 +08004539err_free_css:
Tejun Heoa2bed822014-05-04 15:09:14 -04004540 call_rcu(&css->rcu_head, css_free_rcu_fn);
Tejun Heoc81c925a2013-12-06 15:11:56 -05004541 return err;
4542}
4543
Tejun Heob3bfd982014-05-13 12:19:22 -04004544static int cgroup_mkdir(struct kernfs_node *parent_kn, const char *name,
4545 umode_t mode)
Paul Menageddbcc7e2007-10-18 23:39:30 -07004546{
Tejun Heoa9746d82014-05-13 12:19:22 -04004547 struct cgroup *parent, *cgrp;
4548 struct cgroup_root *root;
Paul Menageddbcc7e2007-10-18 23:39:30 -07004549 struct cgroup_subsys *ss;
Tejun Heo2bd59d42014-02-11 11:52:49 -05004550 struct kernfs_node *kn;
Tejun Heoa14c6872014-07-15 11:05:09 -04004551 struct cftype *base_files;
Tejun Heob3bfd982014-05-13 12:19:22 -04004552 int ssid, ret;
Paul Menageddbcc7e2007-10-18 23:39:30 -07004553
Alban Crequy71b1fb52014-08-18 12:20:20 +01004554 /* Do not accept '\n' to prevent making /proc/<pid>/cgroup unparsable.
4555 */
4556 if (strchr(name, '\n'))
4557 return -EINVAL;
4558
Tejun Heoa9746d82014-05-13 12:19:22 -04004559 parent = cgroup_kn_lock_live(parent_kn);
4560 if (!parent)
4561 return -ENODEV;
4562 root = parent->root;
Paul Menageddbcc7e2007-10-18 23:39:30 -07004563
Tejun Heo0a950f62012-11-19 09:02:12 -08004564 /* allocate the cgroup and its ID, 0 is reserved for the root */
Paul Menagebd89aab2007-10-18 23:40:44 -07004565 cgrp = kzalloc(sizeof(*cgrp), GFP_KERNEL);
Tejun Heoba0f4d72014-05-13 12:19:22 -04004566 if (!cgrp) {
4567 ret = -ENOMEM;
4568 goto out_unlock;
Li Zefan0ab02ca2014-02-11 16:05:46 +08004569 }
4570
Tejun Heo2aad2a82014-09-24 13:31:50 -04004571 ret = percpu_ref_init(&cgrp->self.refcnt, css_release, 0, GFP_KERNEL);
Tejun Heo9d755d32014-05-14 09:15:02 -04004572 if (ret)
4573 goto out_free_cgrp;
4574
Li Zefan0ab02ca2014-02-11 16:05:46 +08004575 /*
4576 * Temporarily set the pointer to NULL, so idr_find() won't return
4577 * a half-baked cgroup.
4578 */
Tejun Heo6fa49182014-05-04 15:09:13 -04004579 cgrp->id = cgroup_idr_alloc(&root->cgroup_idr, NULL, 2, 0, GFP_NOWAIT);
Li Zefan0ab02ca2014-02-11 16:05:46 +08004580 if (cgrp->id < 0) {
Tejun Heoba0f4d72014-05-13 12:19:22 -04004581 ret = -ENOMEM;
Tejun Heo9d755d32014-05-14 09:15:02 -04004582 goto out_cancel_ref;
Tejun Heo976c06b2012-11-05 09:16:59 -08004583 }
4584
Paul Menagecc31edc2008-10-18 20:28:04 -07004585 init_cgroup_housekeeping(cgrp);
Paul Menageddbcc7e2007-10-18 23:39:30 -07004586
Tejun Heo9d800df2014-05-14 09:15:00 -04004587 cgrp->self.parent = &parent->self;
Tejun Heoba0f4d72014-05-13 12:19:22 -04004588 cgrp->root = root;
Paul Menageddbcc7e2007-10-18 23:39:30 -07004589
Li Zefanb6abdb02008-03-04 14:28:19 -08004590 if (notify_on_release(parent))
4591 set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
4592
Tejun Heo2260e7f2012-11-19 08:13:38 -08004593 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &parent->flags))
4594 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags);
Daniel Lezcano97978e62010-10-27 15:33:35 -07004595
Tejun Heo2bd59d42014-02-11 11:52:49 -05004596 /* create the directory */
Tejun Heoe61734c2014-02-12 09:29:50 -05004597 kn = kernfs_create_dir(parent->kn, name, mode, cgrp);
Tejun Heo2bd59d42014-02-11 11:52:49 -05004598 if (IS_ERR(kn)) {
Tejun Heoba0f4d72014-05-13 12:19:22 -04004599 ret = PTR_ERR(kn);
4600 goto out_free_id;
Tejun Heo2bd59d42014-02-11 11:52:49 -05004601 }
4602 cgrp->kn = kn;
Paul Menageddbcc7e2007-10-18 23:39:30 -07004603
Tejun Heo6f305582014-02-12 09:29:50 -05004604 /*
4605 * This extra ref will be put in cgroup_free_fn() and guarantees
4606 * that @cgrp->kn is always accessible.
4607 */
4608 kernfs_get(kn);
Paul Menageddbcc7e2007-10-18 23:39:30 -07004609
Tejun Heo0cb51d72014-05-16 13:22:49 -04004610 cgrp->self.serial_nr = css_serial_nr_next++;
Tejun Heo53fa5262013-05-24 10:55:38 +09004611
Tejun Heo4e139af2012-11-19 08:13:36 -08004612 /* allocation complete, commit to creation */
Tejun Heod5c419b2014-05-16 13:22:48 -04004613 list_add_tail_rcu(&cgrp->self.sibling, &cgroup_parent(cgrp)->self.children);
Tejun Heo3c9c8252014-02-12 09:29:50 -05004614 atomic_inc(&root->nr_cgrps);
Tejun Heo59f52962014-02-11 11:52:49 -05004615 cgroup_get(parent);
Li Zefan415cf072013-04-08 14:35:02 +08004616
Tejun Heo0d802552013-12-06 15:11:56 -05004617 /*
4618 * @cgrp is now fully operational. If something fails after this
4619 * point, it'll be released via the normal destruction path.
4620 */
Tejun Heo6fa49182014-05-04 15:09:13 -04004621 cgroup_idr_replace(&root->cgroup_idr, cgrp, cgrp->id);
Li Zefan4e96ee8e2013-07-31 09:50:50 +08004622
Tejun Heoba0f4d72014-05-13 12:19:22 -04004623 ret = cgroup_kn_set_ugid(kn);
4624 if (ret)
4625 goto out_destroy;
Tejun Heo49957f82014-04-07 16:44:47 -04004626
Tejun Heoa14c6872014-07-15 11:05:09 -04004627 if (cgroup_on_dfl(cgrp))
4628 base_files = cgroup_dfl_base_files;
4629 else
4630 base_files = cgroup_legacy_base_files;
4631
4632 ret = cgroup_addrm_files(cgrp, base_files, true);
Tejun Heoba0f4d72014-05-13 12:19:22 -04004633 if (ret)
4634 goto out_destroy;
Tejun Heo628f7cd2013-06-28 16:24:11 -07004635
Tejun Heo9d403e92013-12-06 15:11:56 -05004636 /* let's create and online css's */
Tejun Heob85d2042013-12-06 15:11:57 -05004637 for_each_subsys(ss, ssid) {
Tejun Heof392e512014-04-23 11:13:14 -04004638 if (parent->child_subsys_mask & (1 << ssid)) {
Tejun Heof63070d2014-07-08 18:02:57 -04004639 ret = create_css(cgrp, ss,
4640 parent->subtree_control & (1 << ssid));
Tejun Heoba0f4d72014-05-13 12:19:22 -04004641 if (ret)
4642 goto out_destroy;
Tejun Heob85d2042013-12-06 15:11:57 -05004643 }
Tejun Heoa8638032012-11-09 09:12:29 -08004644 }
Paul Menageddbcc7e2007-10-18 23:39:30 -07004645
Tejun Heobd53d612014-04-23 11:13:16 -04004646 /*
4647 * On the default hierarchy, a child doesn't automatically inherit
Tejun Heo667c2492014-07-08 18:02:56 -04004648 * subtree_control from the parent. Each is configured manually.
Tejun Heobd53d612014-04-23 11:13:16 -04004649 */
Tejun Heo667c2492014-07-08 18:02:56 -04004650 if (!cgroup_on_dfl(cgrp)) {
4651 cgrp->subtree_control = parent->subtree_control;
4652 cgroup_refresh_child_subsys_mask(cgrp);
4653 }
Tejun Heof392e512014-04-23 11:13:14 -04004654
Tejun Heo2bd59d42014-02-11 11:52:49 -05004655 kernfs_activate(kn);
4656
Tejun Heoba0f4d72014-05-13 12:19:22 -04004657 ret = 0;
4658 goto out_unlock;
Paul Menageddbcc7e2007-10-18 23:39:30 -07004659
Tejun Heoba0f4d72014-05-13 12:19:22 -04004660out_free_id:
Tejun Heo6fa49182014-05-04 15:09:13 -04004661 cgroup_idr_remove(&root->cgroup_idr, cgrp->id);
Tejun Heo9d755d32014-05-14 09:15:02 -04004662out_cancel_ref:
Tejun Heo9a1049d2014-06-28 08:10:14 -04004663 percpu_ref_exit(&cgrp->self.refcnt);
Tejun Heoba0f4d72014-05-13 12:19:22 -04004664out_free_cgrp:
Paul Menagebd89aab2007-10-18 23:40:44 -07004665 kfree(cgrp);
Tejun Heoba0f4d72014-05-13 12:19:22 -04004666out_unlock:
Tejun Heoa9746d82014-05-13 12:19:22 -04004667 cgroup_kn_unlock(parent_kn);
Tejun Heoe1b2dc12014-03-20 11:10:15 -04004668 return ret;
Tejun Heoba0f4d72014-05-13 12:19:22 -04004669
4670out_destroy:
4671 cgroup_destroy_locked(cgrp);
4672 goto out_unlock;
Paul Menageddbcc7e2007-10-18 23:39:30 -07004673}
4674
Tejun Heo223dbc32013-08-13 20:22:50 -04004675/*
4676 * This is called when the refcnt of a css is confirmed to be killed.
Tejun Heo249f3462014-05-14 09:15:01 -04004677 * css_tryget_online() is now guaranteed to fail. Tell the subsystem to
4678 * initate destruction and put the css ref from kill_css().
Tejun Heo223dbc32013-08-13 20:22:50 -04004679 */
4680static void css_killed_work_fn(struct work_struct *work)
Tejun Heod3daf282013-06-13 19:39:16 -07004681{
Tejun Heo223dbc32013-08-13 20:22:50 -04004682 struct cgroup_subsys_state *css =
4683 container_of(work, struct cgroup_subsys_state, destroy_work);
Tejun Heod3daf282013-06-13 19:39:16 -07004684
Tejun Heof20104d2013-08-13 20:22:50 -04004685 mutex_lock(&cgroup_mutex);
Tejun Heo09a503ea2013-08-13 20:22:50 -04004686 offline_css(css);
Tejun Heof20104d2013-08-13 20:22:50 -04004687 mutex_unlock(&cgroup_mutex);
Tejun Heo09a503ea2013-08-13 20:22:50 -04004688
Tejun Heo09a503ea2013-08-13 20:22:50 -04004689 css_put(css);
Tejun Heod3daf282013-06-13 19:39:16 -07004690}
4691
Tejun Heo223dbc32013-08-13 20:22:50 -04004692/* css kill confirmation processing requires process context, bounce */
4693static void css_killed_ref_fn(struct percpu_ref *ref)
Tejun Heod3daf282013-06-13 19:39:16 -07004694{
4695 struct cgroup_subsys_state *css =
4696 container_of(ref, struct cgroup_subsys_state, refcnt);
4697
Tejun Heo223dbc32013-08-13 20:22:50 -04004698 INIT_WORK(&css->destroy_work, css_killed_work_fn);
Tejun Heoe5fca242013-11-22 17:14:39 -05004699 queue_work(cgroup_destroy_wq, &css->destroy_work);
Tejun Heod3daf282013-06-13 19:39:16 -07004700}
4701
Tejun Heof392e512014-04-23 11:13:14 -04004702/**
4703 * kill_css - destroy a css
4704 * @css: css to destroy
4705 *
4706 * This function initiates destruction of @css by removing cgroup interface
4707 * files and putting its base reference. ->css_offline() will be invoked
Tejun Heoec903c02014-05-13 12:11:01 -04004708 * asynchronously once css_tryget_online() is guaranteed to fail and when
4709 * the reference count reaches zero, @css will be released.
Tejun Heof392e512014-04-23 11:13:14 -04004710 */
4711static void kill_css(struct cgroup_subsys_state *css)
Tejun Heoedae0c32013-08-13 20:22:51 -04004712{
Tejun Heo01f64742014-05-13 12:19:23 -04004713 lockdep_assert_held(&cgroup_mutex);
Tejun Heo94419622014-03-19 10:23:54 -04004714
Tejun Heo2bd59d42014-02-11 11:52:49 -05004715 /*
4716 * This must happen before css is disassociated with its cgroup.
4717 * See seq_css() for details.
4718 */
Tejun Heoaec25022014-02-08 10:36:58 -05004719 cgroup_clear_dir(css->cgroup, 1 << css->ss->id);
Tejun Heo3c14f8b2013-08-13 20:22:51 -04004720
Tejun Heoedae0c32013-08-13 20:22:51 -04004721 /*
4722 * Killing would put the base ref, but we need to keep it alive
4723 * until after ->css_offline().
4724 */
4725 css_get(css);
4726
4727 /*
4728 * cgroup core guarantees that, by the time ->css_offline() is
4729 * invoked, no new css reference will be given out via
Tejun Heoec903c02014-05-13 12:11:01 -04004730 * css_tryget_online(). We can't simply call percpu_ref_kill() and
Tejun Heoedae0c32013-08-13 20:22:51 -04004731 * proceed to offlining css's because percpu_ref_kill() doesn't
4732 * guarantee that the ref is seen as killed on all CPUs on return.
4733 *
4734 * Use percpu_ref_kill_and_confirm() to get notifications as each
4735 * css is confirmed to be seen as killed on all CPUs.
4736 */
4737 percpu_ref_kill_and_confirm(&css->refcnt, css_killed_ref_fn);
Tejun Heod3daf282013-06-13 19:39:16 -07004738}
4739
4740/**
4741 * cgroup_destroy_locked - the first stage of cgroup destruction
4742 * @cgrp: cgroup to be destroyed
4743 *
4744 * css's make use of percpu refcnts whose killing latency shouldn't be
4745 * exposed to userland and are RCU protected. Also, cgroup core needs to
Tejun Heoec903c02014-05-13 12:11:01 -04004746 * guarantee that css_tryget_online() won't succeed by the time
4747 * ->css_offline() is invoked. To satisfy all the requirements,
4748 * destruction is implemented in the following two steps.
Tejun Heod3daf282013-06-13 19:39:16 -07004749 *
4750 * s1. Verify @cgrp can be destroyed and mark it dying. Remove all
4751 * userland visible parts and start killing the percpu refcnts of
4752 * css's. Set up so that the next stage will be kicked off once all
4753 * the percpu refcnts are confirmed to be killed.
4754 *
4755 * s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the
4756 * rest of destruction. Once all cgroup references are gone, the
4757 * cgroup is RCU-freed.
4758 *
4759 * This function implements s1. After this step, @cgrp is gone as far as
4760 * the userland is concerned and a new cgroup with the same name may be
4761 * created. As cgroup doesn't care about the names internally, this
4762 * doesn't cause any problem.
4763 */
Tejun Heo42809dd2012-11-19 08:13:37 -08004764static int cgroup_destroy_locked(struct cgroup *cgrp)
4765 __releases(&cgroup_mutex) __acquires(&cgroup_mutex)
Paul Menageddbcc7e2007-10-18 23:39:30 -07004766{
Tejun Heo2bd59d42014-02-11 11:52:49 -05004767 struct cgroup_subsys_state *css;
Tejun Heoddd69142013-06-12 21:04:54 -07004768 bool empty;
Tejun Heo1c6727a2013-12-06 15:11:56 -05004769 int ssid;
Paul Menageddbcc7e2007-10-18 23:39:30 -07004770
Tejun Heo42809dd2012-11-19 08:13:37 -08004771 lockdep_assert_held(&cgroup_mutex);
4772
Tejun Heoddd69142013-06-12 21:04:54 -07004773 /*
Tejun Heo96d365e2014-02-13 06:58:40 -05004774 * css_set_rwsem synchronizes access to ->cset_links and prevents
Tejun Heo89c55092014-02-13 06:58:40 -05004775 * @cgrp from being removed while put_css_set() is in progress.
Tejun Heoddd69142013-06-12 21:04:54 -07004776 */
Tejun Heo96d365e2014-02-13 06:58:40 -05004777 down_read(&css_set_rwsem);
Hugh Dickinsbb78a922013-08-28 16:31:23 -07004778 empty = list_empty(&cgrp->cset_links);
Tejun Heo96d365e2014-02-13 06:58:40 -05004779 up_read(&css_set_rwsem);
Tejun Heoddd69142013-06-12 21:04:54 -07004780 if (!empty)
Paul Menageddbcc7e2007-10-18 23:39:30 -07004781 return -EBUSY;
Tejun Heoed9577932012-11-05 09:16:58 -08004782
Tejun Heo1a90dd52012-11-05 09:16:59 -08004783 /*
Tejun Heod5c419b2014-05-16 13:22:48 -04004784 * Make sure there's no live children. We can't test emptiness of
4785 * ->self.children as dead children linger on it while being
4786 * drained; otherwise, "rmdir parent/child parent" may fail.
Hugh Dickinsbb78a922013-08-28 16:31:23 -07004787 */
Tejun Heof3d46502014-05-16 13:22:52 -04004788 if (css_has_online_children(&cgrp->self))
Hugh Dickinsbb78a922013-08-28 16:31:23 -07004789 return -EBUSY;
4790
4791 /*
Tejun Heo455050d2013-06-13 19:27:41 -07004792 * Mark @cgrp dead. This prevents further task migration and child
Tejun Heode3f0342014-05-16 13:22:49 -04004793 * creation by disabling cgroup_lock_live_group().
Tejun Heo455050d2013-06-13 19:27:41 -07004794 */
Tejun Heo184faf32014-05-16 13:22:51 -04004795 cgrp->self.flags &= ~CSS_ONLINE;
Tejun Heo1a90dd52012-11-05 09:16:59 -08004796
Tejun Heo249f3462014-05-14 09:15:01 -04004797 /* initiate massacre of all css's */
Tejun Heo1a90dd52012-11-05 09:16:59 -08004798 for_each_css(css, ssid, cgrp)
Tejun Heo455050d2013-06-13 19:27:41 -07004799 kill_css(css);
4800
Tejun Heo455050d2013-06-13 19:27:41 -07004801 /*
Tejun Heo01f64742014-05-13 12:19:23 -04004802 * Remove @cgrp directory along with the base files. @cgrp has an
4803 * extra ref on its kn.
Tejun Heo455050d2013-06-13 19:27:41 -07004804 */
Tejun Heo01f64742014-05-13 12:19:23 -04004805 kernfs_remove(cgrp->kn);
Tejun Heof20104d2013-08-13 20:22:50 -04004806
Tejun Heod51f39b2014-05-16 13:22:48 -04004807 check_for_release(cgroup_parent(cgrp));
Tejun Heo2bd59d42014-02-11 11:52:49 -05004808
Tejun Heo249f3462014-05-14 09:15:01 -04004809 /* put the base reference */
Tejun Heo9d755d32014-05-14 09:15:02 -04004810 percpu_ref_kill(&cgrp->self.refcnt);
Tejun Heo455050d2013-06-13 19:27:41 -07004811
Tejun Heoea15f8c2013-06-13 19:27:42 -07004812 return 0;
4813};
4814
Tejun Heo2bd59d42014-02-11 11:52:49 -05004815static int cgroup_rmdir(struct kernfs_node *kn)
Tejun Heo42809dd2012-11-19 08:13:37 -08004816{
Tejun Heoa9746d82014-05-13 12:19:22 -04004817 struct cgroup *cgrp;
Tejun Heo2bd59d42014-02-11 11:52:49 -05004818 int ret = 0;
Tejun Heo42809dd2012-11-19 08:13:37 -08004819
Tejun Heoa9746d82014-05-13 12:19:22 -04004820 cgrp = cgroup_kn_lock_live(kn);
4821 if (!cgrp)
4822 return 0;
Tejun Heo42809dd2012-11-19 08:13:37 -08004823
Tejun Heoa9746d82014-05-13 12:19:22 -04004824 ret = cgroup_destroy_locked(cgrp);
Tejun Heo42809dd2012-11-19 08:13:37 -08004825
Tejun Heoa9746d82014-05-13 12:19:22 -04004826 cgroup_kn_unlock(kn);
Tejun Heo42809dd2012-11-19 08:13:37 -08004827 return ret;
4828}
4829
Tejun Heo2bd59d42014-02-11 11:52:49 -05004830static struct kernfs_syscall_ops cgroup_kf_syscall_ops = {
4831 .remount_fs = cgroup_remount,
4832 .show_options = cgroup_show_options,
4833 .mkdir = cgroup_mkdir,
4834 .rmdir = cgroup_rmdir,
4835 .rename = cgroup_rename,
4836};
Tejun Heo8e3f6542012-04-01 12:09:55 -07004837
Tejun Heo15a4c832014-05-04 15:09:14 -04004838static void __init cgroup_init_subsys(struct cgroup_subsys *ss, bool early)
Paul Menageddbcc7e2007-10-18 23:39:30 -07004839{
Paul Menageddbcc7e2007-10-18 23:39:30 -07004840 struct cgroup_subsys_state *css;
Diego Callejacfe36bd2007-11-14 16:58:54 -08004841
4842 printk(KERN_INFO "Initializing cgroup subsys %s\n", ss->name);
Paul Menageddbcc7e2007-10-18 23:39:30 -07004843
Tejun Heo648bb562012-11-19 08:13:36 -08004844 mutex_lock(&cgroup_mutex);
4845
Tejun Heo15a4c832014-05-04 15:09:14 -04004846 idr_init(&ss->css_idr);
Tejun Heo0adb0702014-02-12 09:29:48 -05004847 INIT_LIST_HEAD(&ss->cfts);
Tejun Heo8e3f6542012-04-01 12:09:55 -07004848
Tejun Heo3dd06ff2014-03-19 10:23:54 -04004849 /* Create the root cgroup state for this subsystem */
4850 ss->root = &cgrp_dfl_root;
4851 css = ss->css_alloc(cgroup_css(&cgrp_dfl_root.cgrp, ss));
Paul Menageddbcc7e2007-10-18 23:39:30 -07004852 /* We don't handle early failures gracefully */
4853 BUG_ON(IS_ERR(css));
Tejun Heoddfcada2014-05-04 15:09:14 -04004854 init_and_link_css(css, ss, &cgrp_dfl_root.cgrp);
Tejun Heo3b514d22014-05-16 13:22:47 -04004855
4856 /*
4857 * Root csses are never destroyed and we can't initialize
4858 * percpu_ref during early init. Disable refcnting.
4859 */
4860 css->flags |= CSS_NO_REF;
4861
Tejun Heo15a4c832014-05-04 15:09:14 -04004862 if (early) {
Tejun Heo9395a452014-05-14 09:15:02 -04004863 /* allocation can't be done safely during early init */
Tejun Heo15a4c832014-05-04 15:09:14 -04004864 css->id = 1;
4865 } else {
4866 css->id = cgroup_idr_alloc(&ss->css_idr, css, 1, 2, GFP_KERNEL);
4867 BUG_ON(css->id < 0);
4868 }
Paul Menageddbcc7e2007-10-18 23:39:30 -07004869
Li Zefane8d55fd2008-04-29 01:00:13 -07004870 /* Update the init_css_set to contain a subsys
Paul Menage817929e2007-10-18 23:39:36 -07004871 * pointer to this state - since the subsystem is
Li Zefane8d55fd2008-04-29 01:00:13 -07004872 * newly registered, all tasks and hence the
Tejun Heo3dd06ff2014-03-19 10:23:54 -04004873 * init_css_set is in the subsystem's root cgroup. */
Tejun Heoaec25022014-02-08 10:36:58 -05004874 init_css_set.subsys[ss->id] = css;
Paul Menageddbcc7e2007-10-18 23:39:30 -07004875
4876 need_forkexit_callback |= ss->fork || ss->exit;
4877
Li Zefane8d55fd2008-04-29 01:00:13 -07004878 /* At system boot, before all subsystems have been
4879 * registered, no tasks have been forked, so we don't
4880 * need to invoke fork callbacks here. */
4881 BUG_ON(!list_empty(&init_task.tasks));
4882
Tejun Heoae7f1642013-08-13 20:22:50 -04004883 BUG_ON(online_css(css));
Tejun Heoa8638032012-11-09 09:12:29 -08004884
Tejun Heo648bb562012-11-19 08:13:36 -08004885 mutex_unlock(&cgroup_mutex);
Paul Menageddbcc7e2007-10-18 23:39:30 -07004886}
4887
4888/**
Li Zefana043e3b2008-02-23 15:24:09 -08004889 * cgroup_init_early - cgroup initialization at system boot
4890 *
4891 * Initialize cgroups at system boot, and initialize any
4892 * subsystems that request early init.
Paul Menageddbcc7e2007-10-18 23:39:30 -07004893 */
4894int __init cgroup_init_early(void)
4895{
Tejun Heo7b9a6ba2014-07-09 10:08:08 -04004896 static struct cgroup_sb_opts __initdata opts;
Tejun Heo30159ec2013-06-25 11:53:37 -07004897 struct cgroup_subsys *ss;
Paul Menageddbcc7e2007-10-18 23:39:30 -07004898 int i;
Tejun Heo30159ec2013-06-25 11:53:37 -07004899
Tejun Heo3dd06ff2014-03-19 10:23:54 -04004900 init_cgroup_root(&cgrp_dfl_root, &opts);
Tejun Heo3b514d22014-05-16 13:22:47 -04004901 cgrp_dfl_root.cgrp.self.flags |= CSS_NO_REF;
4902
Tejun Heoa4ea1cc2013-06-21 15:52:33 -07004903 RCU_INIT_POINTER(init_task.cgroups, &init_css_set);
Paul Menage817929e2007-10-18 23:39:36 -07004904
Tejun Heo3ed80a62014-02-08 10:36:58 -05004905 for_each_subsys(ss, i) {
Tejun Heoaec25022014-02-08 10:36:58 -05004906 WARN(!ss->css_alloc || !ss->css_free || ss->name || ss->id,
Tejun Heo073219e2014-02-08 10:36:58 -05004907 "invalid cgroup_subsys %d:%s css_alloc=%p css_free=%p name:id=%d:%s\n",
4908 i, cgroup_subsys_name[i], ss->css_alloc, ss->css_free,
Tejun Heoaec25022014-02-08 10:36:58 -05004909 ss->id, ss->name);
Tejun Heo073219e2014-02-08 10:36:58 -05004910 WARN(strlen(cgroup_subsys_name[i]) > MAX_CGROUP_TYPE_NAMELEN,
4911 "cgroup_subsys_name %s too long\n", cgroup_subsys_name[i]);
Paul Menageddbcc7e2007-10-18 23:39:30 -07004912
Tejun Heoaec25022014-02-08 10:36:58 -05004913 ss->id = i;
Tejun Heo073219e2014-02-08 10:36:58 -05004914 ss->name = cgroup_subsys_name[i];
Paul Menageddbcc7e2007-10-18 23:39:30 -07004915
4916 if (ss->early_init)
Tejun Heo15a4c832014-05-04 15:09:14 -04004917 cgroup_init_subsys(ss, true);
Paul Menageddbcc7e2007-10-18 23:39:30 -07004918 }
4919 return 0;
4920}
4921
4922/**
Li Zefana043e3b2008-02-23 15:24:09 -08004923 * cgroup_init - cgroup initialization
4924 *
4925 * Register cgroup filesystem and /proc file, and initialize
4926 * any subsystems that didn't request early init.
Paul Menageddbcc7e2007-10-18 23:39:30 -07004927 */
4928int __init cgroup_init(void)
4929{
Tejun Heo30159ec2013-06-25 11:53:37 -07004930 struct cgroup_subsys *ss;
Li Zefan0ac801f2013-01-10 11:49:27 +08004931 unsigned long key;
Tejun Heo172a2c062014-03-19 10:23:53 -04004932 int ssid, err;
Paul Menagea4243162007-10-18 23:39:35 -07004933
Tejun Heoa14c6872014-07-15 11:05:09 -04004934 BUG_ON(cgroup_init_cftypes(NULL, cgroup_dfl_base_files));
4935 BUG_ON(cgroup_init_cftypes(NULL, cgroup_legacy_base_files));
Paul Menageddbcc7e2007-10-18 23:39:30 -07004936
Tejun Heo54e7b4e2013-04-14 11:36:57 -07004937 mutex_lock(&cgroup_mutex);
Tejun Heo54e7b4e2013-04-14 11:36:57 -07004938
Tejun Heo82fe9b02013-06-25 11:53:37 -07004939 /* Add init_css_set to the hash table */
4940 key = css_set_hash(init_css_set.subsys);
4941 hash_add(css_set_table, &init_css_set.hlist, key);
4942
Tejun Heo3dd06ff2014-03-19 10:23:54 -04004943 BUG_ON(cgroup_setup_root(&cgrp_dfl_root, 0));
Greg KH676db4a2010-08-05 13:53:35 -07004944
Tejun Heo54e7b4e2013-04-14 11:36:57 -07004945 mutex_unlock(&cgroup_mutex);
4946
Tejun Heo172a2c062014-03-19 10:23:53 -04004947 for_each_subsys(ss, ssid) {
Tejun Heo15a4c832014-05-04 15:09:14 -04004948 if (ss->early_init) {
4949 struct cgroup_subsys_state *css =
4950 init_css_set.subsys[ss->id];
4951
4952 css->id = cgroup_idr_alloc(&ss->css_idr, css, 1, 2,
4953 GFP_KERNEL);
4954 BUG_ON(css->id < 0);
4955 } else {
4956 cgroup_init_subsys(ss, false);
4957 }
Tejun Heo172a2c062014-03-19 10:23:53 -04004958
Tejun Heo2d8f2432014-04-23 11:13:15 -04004959 list_add_tail(&init_css_set.e_cset_node[ssid],
4960 &cgrp_dfl_root.cgrp.e_csets[ssid]);
Tejun Heo172a2c062014-03-19 10:23:53 -04004961
4962 /*
Li Zefanc731ae12014-06-05 17:16:30 +08004963 * Setting dfl_root subsys_mask needs to consider the
4964 * disabled flag and cftype registration needs kmalloc,
4965 * both of which aren't available during early_init.
Tejun Heo172a2c062014-03-19 10:23:53 -04004966 */
Tejun Heoa8ddc822014-07-15 11:05:10 -04004967 if (ss->disabled)
4968 continue;
4969
4970 cgrp_dfl_root.subsys_mask |= 1 << ss->id;
4971
4972 if (cgroup_legacy_files_on_dfl && !ss->dfl_cftypes)
4973 ss->dfl_cftypes = ss->legacy_cftypes;
4974
Tejun Heo5de4fa12014-07-15 11:05:10 -04004975 if (!ss->dfl_cftypes)
4976 cgrp_dfl_root_inhibit_ss_mask |= 1 << ss->id;
4977
Tejun Heoa8ddc822014-07-15 11:05:10 -04004978 if (ss->dfl_cftypes == ss->legacy_cftypes) {
4979 WARN_ON(cgroup_add_cftypes(ss, ss->dfl_cftypes));
4980 } else {
4981 WARN_ON(cgroup_add_dfl_cftypes(ss, ss->dfl_cftypes));
4982 WARN_ON(cgroup_add_legacy_cftypes(ss, ss->legacy_cftypes));
Li Zefanc731ae12014-06-05 17:16:30 +08004983 }
Tejun Heo172a2c062014-03-19 10:23:53 -04004984 }
Greg KH676db4a2010-08-05 13:53:35 -07004985
Paul Menageddbcc7e2007-10-18 23:39:30 -07004986 cgroup_kobj = kobject_create_and_add("cgroup", fs_kobj);
Tejun Heo2bd59d42014-02-11 11:52:49 -05004987 if (!cgroup_kobj)
4988 return -ENOMEM;
Paul Menagea4243162007-10-18 23:39:35 -07004989
4990 err = register_filesystem(&cgroup_fs_type);
Paul Menageddbcc7e2007-10-18 23:39:30 -07004991 if (err < 0) {
4992 kobject_put(cgroup_kobj);
Tejun Heo2bd59d42014-02-11 11:52:49 -05004993 return err;
Paul Menagea4243162007-10-18 23:39:35 -07004994 }
4995
4996 proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations);
Tejun Heo2bd59d42014-02-11 11:52:49 -05004997 return 0;
Paul Menageddbcc7e2007-10-18 23:39:30 -07004998}
Paul Menageb4f48b62007-10-18 23:39:33 -07004999
Tejun Heoe5fca242013-11-22 17:14:39 -05005000static int __init cgroup_wq_init(void)
5001{
5002 /*
5003 * There isn't much point in executing destruction path in
5004 * parallel. Good chunk is serialized with cgroup_mutex anyway.
Tejun Heo1a115332014-02-12 19:06:19 -05005005 * Use 1 for @max_active.
Tejun Heoe5fca242013-11-22 17:14:39 -05005006 *
5007 * We would prefer to do this in cgroup_init() above, but that
5008 * is called before init_workqueues(): so leave this until after.
5009 */
Tejun Heo1a115332014-02-12 19:06:19 -05005010 cgroup_destroy_wq = alloc_workqueue("cgroup_destroy", 0, 1);
Tejun Heoe5fca242013-11-22 17:14:39 -05005011 BUG_ON(!cgroup_destroy_wq);
Tejun Heob1a21362013-11-29 10:42:58 -05005012
5013 /*
5014 * Used to destroy pidlists and separate to serve as flush domain.
5015 * Cap @max_active to 1 too.
5016 */
5017 cgroup_pidlist_destroy_wq = alloc_workqueue("cgroup_pidlist_destroy",
5018 0, 1);
5019 BUG_ON(!cgroup_pidlist_destroy_wq);
5020
Tejun Heoe5fca242013-11-22 17:14:39 -05005021 return 0;
5022}
5023core_initcall(cgroup_wq_init);
5024
Paul Menagea4243162007-10-18 23:39:35 -07005025/*
5026 * proc_cgroup_show()
5027 * - Print task's cgroup paths into seq_file, one line for each hierarchy
5028 * - Used for /proc/<pid>/cgroup.
Paul Menagea4243162007-10-18 23:39:35 -07005029 */
Zefan Li006f4ac2014-09-18 16:03:15 +08005030int proc_cgroup_show(struct seq_file *m, struct pid_namespace *ns,
5031 struct pid *pid, struct task_struct *tsk)
Paul Menagea4243162007-10-18 23:39:35 -07005032{
Tejun Heoe61734c2014-02-12 09:29:50 -05005033 char *buf, *path;
Paul Menagea4243162007-10-18 23:39:35 -07005034 int retval;
Tejun Heo3dd06ff2014-03-19 10:23:54 -04005035 struct cgroup_root *root;
Paul Menagea4243162007-10-18 23:39:35 -07005036
5037 retval = -ENOMEM;
Tejun Heoe61734c2014-02-12 09:29:50 -05005038 buf = kmalloc(PATH_MAX, GFP_KERNEL);
Paul Menagea4243162007-10-18 23:39:35 -07005039 if (!buf)
5040 goto out;
5041
Paul Menagea4243162007-10-18 23:39:35 -07005042 mutex_lock(&cgroup_mutex);
Tejun Heo96d365e2014-02-13 06:58:40 -05005043 down_read(&css_set_rwsem);
Paul Menagea4243162007-10-18 23:39:35 -07005044
Tejun Heo985ed672014-03-19 10:23:53 -04005045 for_each_root(root) {
Paul Menagea4243162007-10-18 23:39:35 -07005046 struct cgroup_subsys *ss;
Paul Menagebd89aab2007-10-18 23:40:44 -07005047 struct cgroup *cgrp;
Tejun Heob85d2042013-12-06 15:11:57 -05005048 int ssid, count = 0;
Paul Menagea4243162007-10-18 23:39:35 -07005049
Tejun Heoa2dd4242014-03-19 10:23:55 -04005050 if (root == &cgrp_dfl_root && !cgrp_dfl_root_visible)
Tejun Heo985ed672014-03-19 10:23:53 -04005051 continue;
5052
Paul Menage2c6ab6d2009-09-23 15:56:23 -07005053 seq_printf(m, "%d:", root->hierarchy_id);
Tejun Heob85d2042013-12-06 15:11:57 -05005054 for_each_subsys(ss, ssid)
Tejun Heof392e512014-04-23 11:13:14 -04005055 if (root->subsys_mask & (1 << ssid))
Tejun Heob85d2042013-12-06 15:11:57 -05005056 seq_printf(m, "%s%s", count++ ? "," : "", ss->name);
Paul Menagec6d57f32009-09-23 15:56:19 -07005057 if (strlen(root->name))
5058 seq_printf(m, "%sname=%s", count ? "," : "",
5059 root->name);
Paul Menagea4243162007-10-18 23:39:35 -07005060 seq_putc(m, ':');
Paul Menage7717f7b2009-09-23 15:56:22 -07005061 cgrp = task_cgroup_from_root(tsk, root);
Tejun Heoe61734c2014-02-12 09:29:50 -05005062 path = cgroup_path(cgrp, buf, PATH_MAX);
5063 if (!path) {
5064 retval = -ENAMETOOLONG;
Paul Menagea4243162007-10-18 23:39:35 -07005065 goto out_unlock;
Tejun Heoe61734c2014-02-12 09:29:50 -05005066 }
5067 seq_puts(m, path);
Paul Menagea4243162007-10-18 23:39:35 -07005068 seq_putc(m, '\n');
5069 }
5070
Zefan Li006f4ac2014-09-18 16:03:15 +08005071 retval = 0;
Paul Menagea4243162007-10-18 23:39:35 -07005072out_unlock:
Tejun Heo96d365e2014-02-13 06:58:40 -05005073 up_read(&css_set_rwsem);
Paul Menagea4243162007-10-18 23:39:35 -07005074 mutex_unlock(&cgroup_mutex);
Paul Menagea4243162007-10-18 23:39:35 -07005075 kfree(buf);
5076out:
5077 return retval;
5078}
5079
Paul Menagea4243162007-10-18 23:39:35 -07005080/* Display information about each subsystem and each hierarchy */
5081static int proc_cgroupstats_show(struct seq_file *m, void *v)
5082{
Tejun Heo30159ec2013-06-25 11:53:37 -07005083 struct cgroup_subsys *ss;
Paul Menagea4243162007-10-18 23:39:35 -07005084 int i;
Paul Menagea4243162007-10-18 23:39:35 -07005085
Paul Menage8bab8dd2008-04-04 14:29:57 -07005086 seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
Ben Blumaae8aab2010-03-10 15:22:07 -08005087 /*
5088 * ideally we don't want subsystems moving around while we do this.
5089 * cgroup_mutex is also necessary to guarantee an atomic snapshot of
5090 * subsys/hierarchy state.
5091 */
Paul Menagea4243162007-10-18 23:39:35 -07005092 mutex_lock(&cgroup_mutex);
Tejun Heo30159ec2013-06-25 11:53:37 -07005093
5094 for_each_subsys(ss, i)
Paul Menage2c6ab6d2009-09-23 15:56:23 -07005095 seq_printf(m, "%s\t%d\t%d\t%d\n",
5096 ss->name, ss->root->hierarchy_id,
Tejun Heo3c9c8252014-02-12 09:29:50 -05005097 atomic_read(&ss->root->nr_cgrps), !ss->disabled);
Tejun Heo30159ec2013-06-25 11:53:37 -07005098
Paul Menagea4243162007-10-18 23:39:35 -07005099 mutex_unlock(&cgroup_mutex);
5100 return 0;
5101}
5102
5103static int cgroupstats_open(struct inode *inode, struct file *file)
5104{
Al Viro9dce07f2008-03-29 03:07:28 +00005105 return single_open(file, proc_cgroupstats_show, NULL);
Paul Menagea4243162007-10-18 23:39:35 -07005106}
5107
Alexey Dobriyan828c0952009-10-01 15:43:56 -07005108static const struct file_operations proc_cgroupstats_operations = {
Paul Menagea4243162007-10-18 23:39:35 -07005109 .open = cgroupstats_open,
5110 .read = seq_read,
5111 .llseek = seq_lseek,
5112 .release = single_release,
5113};
5114
Paul Menageb4f48b62007-10-18 23:39:33 -07005115/**
Tejun Heoeaf797a2014-02-25 10:04:03 -05005116 * cgroup_fork - initialize cgroup related fields during copy_process()
Li Zefana043e3b2008-02-23 15:24:09 -08005117 * @child: pointer to task_struct of forking parent process.
Paul Menageb4f48b62007-10-18 23:39:33 -07005118 *
Tejun Heoeaf797a2014-02-25 10:04:03 -05005119 * A task is associated with the init_css_set until cgroup_post_fork()
5120 * attaches it to the parent's css_set. Empty cg_list indicates that
5121 * @child isn't holding reference to its css_set.
Paul Menageb4f48b62007-10-18 23:39:33 -07005122 */
5123void cgroup_fork(struct task_struct *child)
5124{
Tejun Heoeaf797a2014-02-25 10:04:03 -05005125 RCU_INIT_POINTER(child->cgroups, &init_css_set);
Paul Menage817929e2007-10-18 23:39:36 -07005126 INIT_LIST_HEAD(&child->cg_list);
Paul Menageb4f48b62007-10-18 23:39:33 -07005127}
5128
5129/**
Li Zefana043e3b2008-02-23 15:24:09 -08005130 * cgroup_post_fork - called on a new task after adding it to the task list
5131 * @child: the task in question
5132 *
Tejun Heo5edee612012-10-16 15:03:14 -07005133 * Adds the task to the list running through its css_set if necessary and
5134 * call the subsystem fork() callbacks. Has to be after the task is
5135 * visible on the task list in case we race with the first call to
Tejun Heo0942eee2013-08-08 20:11:26 -04005136 * cgroup_task_iter_start() - to guarantee that the new task ends up on its
Tejun Heo5edee612012-10-16 15:03:14 -07005137 * list.
Li Zefana043e3b2008-02-23 15:24:09 -08005138 */
Paul Menage817929e2007-10-18 23:39:36 -07005139void cgroup_post_fork(struct task_struct *child)
5140{
Tejun Heo30159ec2013-06-25 11:53:37 -07005141 struct cgroup_subsys *ss;
Tejun Heo5edee612012-10-16 15:03:14 -07005142 int i;
5143
Frederic Weisbecker3ce32302012-02-08 03:37:27 +01005144 /*
Dongsheng Yang251f8c02014-08-25 19:27:52 +08005145 * This may race against cgroup_enable_task_cg_lists(). As that
Tejun Heoeaf797a2014-02-25 10:04:03 -05005146 * function sets use_task_css_set_links before grabbing
5147 * tasklist_lock and we just went through tasklist_lock to add
5148 * @child, it's guaranteed that either we see the set
5149 * use_task_css_set_links or cgroup_enable_task_cg_lists() sees
5150 * @child during its iteration.
5151 *
5152 * If we won the race, @child is associated with %current's
5153 * css_set. Grabbing css_set_rwsem guarantees both that the
5154 * association is stable, and, on completion of the parent's
5155 * migration, @child is visible in the source of migration or
5156 * already in the destination cgroup. This guarantee is necessary
5157 * when implementing operations which need to migrate all tasks of
5158 * a cgroup to another.
5159 *
Dongsheng Yang251f8c02014-08-25 19:27:52 +08005160 * Note that if we lose to cgroup_enable_task_cg_lists(), @child
Tejun Heoeaf797a2014-02-25 10:04:03 -05005161 * will remain in init_css_set. This is safe because all tasks are
5162 * in the init_css_set before cg_links is enabled and there's no
5163 * operation which transfers all tasks out of init_css_set.
Frederic Weisbecker3ce32302012-02-08 03:37:27 +01005164 */
Paul Menage817929e2007-10-18 23:39:36 -07005165 if (use_task_css_set_links) {
Tejun Heoeaf797a2014-02-25 10:04:03 -05005166 struct css_set *cset;
5167
Tejun Heo96d365e2014-02-13 06:58:40 -05005168 down_write(&css_set_rwsem);
Tejun Heo0e1d7682014-02-25 10:04:03 -05005169 cset = task_css_set(current);
Tejun Heoeaf797a2014-02-25 10:04:03 -05005170 if (list_empty(&child->cg_list)) {
5171 rcu_assign_pointer(child->cgroups, cset);
5172 list_add(&child->cg_list, &cset->tasks);
5173 get_css_set(cset);
5174 }
Tejun Heo96d365e2014-02-13 06:58:40 -05005175 up_write(&css_set_rwsem);
Paul Menage817929e2007-10-18 23:39:36 -07005176 }
Tejun Heo5edee612012-10-16 15:03:14 -07005177
5178 /*
5179 * Call ss->fork(). This must happen after @child is linked on
5180 * css_set; otherwise, @child might change state between ->fork()
5181 * and addition to css_set.
5182 */
5183 if (need_forkexit_callback) {
Tejun Heo3ed80a62014-02-08 10:36:58 -05005184 for_each_subsys(ss, i)
Tejun Heo5edee612012-10-16 15:03:14 -07005185 if (ss->fork)
5186 ss->fork(child);
Tejun Heo5edee612012-10-16 15:03:14 -07005187 }
Paul Menage817929e2007-10-18 23:39:36 -07005188}
Tejun Heo5edee612012-10-16 15:03:14 -07005189
Paul Menage817929e2007-10-18 23:39:36 -07005190/**
Paul Menageb4f48b62007-10-18 23:39:33 -07005191 * cgroup_exit - detach cgroup from exiting task
5192 * @tsk: pointer to task_struct of exiting process
5193 *
5194 * Description: Detach cgroup from @tsk and release it.
5195 *
5196 * Note that cgroups marked notify_on_release force every task in
5197 * them to take the global cgroup_mutex mutex when exiting.
5198 * This could impact scaling on very large systems. Be reluctant to
5199 * use notify_on_release cgroups where very high task exit scaling
5200 * is required on large systems.
5201 *
Tejun Heo0e1d7682014-02-25 10:04:03 -05005202 * We set the exiting tasks cgroup to the root cgroup (top_cgroup). We
5203 * call cgroup_exit() while the task is still competent to handle
5204 * notify_on_release(), then leave the task attached to the root cgroup in
5205 * each hierarchy for the remainder of its exit. No need to bother with
5206 * init_css_set refcnting. init_css_set never goes away and we can't race
Li Zefane8604cb2014-03-28 15:18:27 +08005207 * with migration path - PF_EXITING is visible to migration path.
Paul Menageb4f48b62007-10-18 23:39:33 -07005208 */
Li Zefan1ec41832014-03-28 15:22:19 +08005209void cgroup_exit(struct task_struct *tsk)
Paul Menageb4f48b62007-10-18 23:39:33 -07005210{
Tejun Heo30159ec2013-06-25 11:53:37 -07005211 struct cgroup_subsys *ss;
Tejun Heo5abb8852013-06-12 21:04:49 -07005212 struct css_set *cset;
Tejun Heoeaf797a2014-02-25 10:04:03 -05005213 bool put_cset = false;
Peter Zijlstrad41d5a02011-02-07 17:02:20 +01005214 int i;
Paul Menage817929e2007-10-18 23:39:36 -07005215
5216 /*
Tejun Heo0e1d7682014-02-25 10:04:03 -05005217 * Unlink from @tsk from its css_set. As migration path can't race
5218 * with us, we can check cg_list without grabbing css_set_rwsem.
Paul Menage817929e2007-10-18 23:39:36 -07005219 */
5220 if (!list_empty(&tsk->cg_list)) {
Tejun Heo96d365e2014-02-13 06:58:40 -05005221 down_write(&css_set_rwsem);
Tejun Heo0e1d7682014-02-25 10:04:03 -05005222 list_del_init(&tsk->cg_list);
Tejun Heo96d365e2014-02-13 06:58:40 -05005223 up_write(&css_set_rwsem);
Tejun Heo0e1d7682014-02-25 10:04:03 -05005224 put_cset = true;
Paul Menage817929e2007-10-18 23:39:36 -07005225 }
5226
Paul Menageb4f48b62007-10-18 23:39:33 -07005227 /* Reassign the task to the init_css_set. */
Tejun Heoa8ad8052013-06-21 15:52:04 -07005228 cset = task_css_set(tsk);
5229 RCU_INIT_POINTER(tsk->cgroups, &init_css_set);
Peter Zijlstrad41d5a02011-02-07 17:02:20 +01005230
Li Zefan1ec41832014-03-28 15:22:19 +08005231 if (need_forkexit_callback) {
Tejun Heo3ed80a62014-02-08 10:36:58 -05005232 /* see cgroup_post_fork() for details */
5233 for_each_subsys(ss, i) {
Peter Zijlstrad41d5a02011-02-07 17:02:20 +01005234 if (ss->exit) {
Tejun Heoeb954192013-08-08 20:11:23 -04005235 struct cgroup_subsys_state *old_css = cset->subsys[i];
5236 struct cgroup_subsys_state *css = task_css(tsk, i);
Tejun Heo30159ec2013-06-25 11:53:37 -07005237
Tejun Heoeb954192013-08-08 20:11:23 -04005238 ss->exit(css, old_css, tsk);
Peter Zijlstrad41d5a02011-02-07 17:02:20 +01005239 }
5240 }
5241 }
Peter Zijlstrad41d5a02011-02-07 17:02:20 +01005242
Tejun Heoeaf797a2014-02-25 10:04:03 -05005243 if (put_cset)
Zefan Lia25eb522014-09-19 16:51:00 +08005244 put_css_set(cset);
Paul Menageb4f48b62007-10-18 23:39:33 -07005245}
Paul Menage697f4162007-10-18 23:39:34 -07005246
Paul Menagebd89aab2007-10-18 23:40:44 -07005247static void check_for_release(struct cgroup *cgrp)
Paul Menage81a6a5c2007-10-18 23:39:38 -07005248{
Zefan Lia25eb522014-09-19 16:51:00 +08005249 if (notify_on_release(cgrp) && !cgroup_has_tasks(cgrp) &&
Zefan Li971ff492014-09-18 16:06:19 +08005250 !css_has_online_children(&cgrp->self) && !cgroup_is_dead(cgrp))
5251 schedule_work(&cgrp->release_agent_work);
Paul Menage81a6a5c2007-10-18 23:39:38 -07005252}
5253
Paul Menage81a6a5c2007-10-18 23:39:38 -07005254/*
5255 * Notify userspace when a cgroup is released, by running the
5256 * configured release agent with the name of the cgroup (path
5257 * relative to the root of cgroup file system) as the argument.
5258 *
5259 * Most likely, this user command will try to rmdir this cgroup.
5260 *
5261 * This races with the possibility that some other task will be
5262 * attached to this cgroup before it is removed, or that some other
5263 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
5264 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
5265 * unused, and this cgroup will be reprieved from its death sentence,
5266 * to continue to serve a useful existence. Next time it's released,
5267 * we will get notified again, if it still has 'notify_on_release' set.
5268 *
5269 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
5270 * means only wait until the task is successfully execve()'d. The
5271 * separate release agent task is forked by call_usermodehelper(),
5272 * then control in this thread returns here, without waiting for the
5273 * release agent task. We don't bother to wait because the caller of
5274 * this routine has no use for the exit status of the release agent
5275 * task, so no sense holding our caller up for that.
Paul Menage81a6a5c2007-10-18 23:39:38 -07005276 */
Paul Menage81a6a5c2007-10-18 23:39:38 -07005277static void cgroup_release_agent(struct work_struct *work)
5278{
Zefan Li971ff492014-09-18 16:06:19 +08005279 struct cgroup *cgrp =
5280 container_of(work, struct cgroup, release_agent_work);
5281 char *pathbuf = NULL, *agentbuf = NULL, *path;
5282 char *argv[3], *envp[3];
5283
Paul Menage81a6a5c2007-10-18 23:39:38 -07005284 mutex_lock(&cgroup_mutex);
Paul Menage81a6a5c2007-10-18 23:39:38 -07005285
Zefan Li971ff492014-09-18 16:06:19 +08005286 pathbuf = kmalloc(PATH_MAX, GFP_KERNEL);
5287 agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL);
5288 if (!pathbuf || !agentbuf)
5289 goto out;
Paul Menage81a6a5c2007-10-18 23:39:38 -07005290
Zefan Li971ff492014-09-18 16:06:19 +08005291 path = cgroup_path(cgrp, pathbuf, PATH_MAX);
5292 if (!path)
5293 goto out;
Paul Menage81a6a5c2007-10-18 23:39:38 -07005294
Zefan Li971ff492014-09-18 16:06:19 +08005295 argv[0] = agentbuf;
5296 argv[1] = path;
5297 argv[2] = NULL;
5298
5299 /* minimal command environment */
5300 envp[0] = "HOME=/";
5301 envp[1] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
5302 envp[2] = NULL;
5303
Paul Menage81a6a5c2007-10-18 23:39:38 -07005304 mutex_unlock(&cgroup_mutex);
Zefan Li971ff492014-09-18 16:06:19 +08005305 call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC);
Zefan Li3e2cd912014-09-20 14:35:43 +08005306 goto out_free;
Zefan Li971ff492014-09-18 16:06:19 +08005307out:
Zefan Li3e2cd912014-09-20 14:35:43 +08005308 mutex_unlock(&cgroup_mutex);
5309out_free:
Zefan Li971ff492014-09-18 16:06:19 +08005310 kfree(agentbuf);
5311 kfree(pathbuf);
Paul Menage81a6a5c2007-10-18 23:39:38 -07005312}
Paul Menage8bab8dd2008-04-04 14:29:57 -07005313
5314static int __init cgroup_disable(char *str)
5315{
Tejun Heo30159ec2013-06-25 11:53:37 -07005316 struct cgroup_subsys *ss;
Paul Menage8bab8dd2008-04-04 14:29:57 -07005317 char *token;
Tejun Heo30159ec2013-06-25 11:53:37 -07005318 int i;
Paul Menage8bab8dd2008-04-04 14:29:57 -07005319
5320 while ((token = strsep(&str, ",")) != NULL) {
5321 if (!*token)
5322 continue;
Paul Menage8bab8dd2008-04-04 14:29:57 -07005323
Tejun Heo3ed80a62014-02-08 10:36:58 -05005324 for_each_subsys(ss, i) {
Paul Menage8bab8dd2008-04-04 14:29:57 -07005325 if (!strcmp(token, ss->name)) {
5326 ss->disabled = 1;
5327 printk(KERN_INFO "Disabling %s control group"
5328 " subsystem\n", ss->name);
5329 break;
5330 }
5331 }
5332 }
5333 return 1;
5334}
5335__setup("cgroup_disable=", cgroup_disable);
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -07005336
Tejun Heoa8ddc822014-07-15 11:05:10 -04005337static int __init cgroup_set_legacy_files_on_dfl(char *str)
5338{
5339 printk("cgroup: using legacy files on the default hierarchy\n");
5340 cgroup_legacy_files_on_dfl = true;
5341 return 0;
5342}
5343__setup("cgroup__DEVEL__legacy_files_on_dfl", cgroup_set_legacy_files_on_dfl);
5344
Tejun Heob77d7b62013-08-13 11:01:54 -04005345/**
Tejun Heoec903c02014-05-13 12:11:01 -04005346 * css_tryget_online_from_dir - get corresponding css from a cgroup dentry
Tejun Heo35cf0832013-08-26 18:40:56 -04005347 * @dentry: directory dentry of interest
5348 * @ss: subsystem of interest
Tejun Heob77d7b62013-08-13 11:01:54 -04005349 *
Tejun Heo5a17f542014-02-11 11:52:47 -05005350 * If @dentry is a directory for a cgroup which has @ss enabled on it, try
5351 * to get the corresponding css and return it. If such css doesn't exist
5352 * or can't be pinned, an ERR_PTR value is returned.
Stephane Eraniane5d13672011-02-14 11:20:01 +02005353 */
Tejun Heoec903c02014-05-13 12:11:01 -04005354struct cgroup_subsys_state *css_tryget_online_from_dir(struct dentry *dentry,
5355 struct cgroup_subsys *ss)
Stephane Eraniane5d13672011-02-14 11:20:01 +02005356{
Tejun Heo2bd59d42014-02-11 11:52:49 -05005357 struct kernfs_node *kn = kernfs_node_from_dentry(dentry);
5358 struct cgroup_subsys_state *css = NULL;
Stephane Eraniane5d13672011-02-14 11:20:01 +02005359 struct cgroup *cgrp;
Stephane Eraniane5d13672011-02-14 11:20:01 +02005360
Tejun Heo35cf0832013-08-26 18:40:56 -04005361 /* is @dentry a cgroup dir? */
Tejun Heo2bd59d42014-02-11 11:52:49 -05005362 if (dentry->d_sb->s_type != &cgroup_fs_type || !kn ||
5363 kernfs_type(kn) != KERNFS_DIR)
Stephane Eraniane5d13672011-02-14 11:20:01 +02005364 return ERR_PTR(-EBADF);
5365
Tejun Heo5a17f542014-02-11 11:52:47 -05005366 rcu_read_lock();
5367
Tejun Heo2bd59d42014-02-11 11:52:49 -05005368 /*
5369 * This path doesn't originate from kernfs and @kn could already
5370 * have been or be removed at any point. @kn->priv is RCU
Li Zefana4189482014-09-04 14:43:07 +08005371 * protected for this access. See css_release_work_fn() for details.
Tejun Heo2bd59d42014-02-11 11:52:49 -05005372 */
5373 cgrp = rcu_dereference(kn->priv);
5374 if (cgrp)
5375 css = cgroup_css(cgrp, ss);
Tejun Heo5a17f542014-02-11 11:52:47 -05005376
Tejun Heoec903c02014-05-13 12:11:01 -04005377 if (!css || !css_tryget_online(css))
Tejun Heo5a17f542014-02-11 11:52:47 -05005378 css = ERR_PTR(-ENOENT);
5379
5380 rcu_read_unlock();
5381 return css;
Stephane Eraniane5d13672011-02-14 11:20:01 +02005382}
Stephane Eraniane5d13672011-02-14 11:20:01 +02005383
Li Zefan1cb650b2013-08-19 10:05:24 +08005384/**
5385 * css_from_id - lookup css by id
5386 * @id: the cgroup id
5387 * @ss: cgroup subsys to be looked into
5388 *
5389 * Returns the css if there's valid one with @id, otherwise returns NULL.
5390 * Should be called under rcu_read_lock().
5391 */
5392struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss)
5393{
Tejun Heo6fa49182014-05-04 15:09:13 -04005394 WARN_ON_ONCE(!rcu_read_lock_held());
Tejun Heo15a4c832014-05-04 15:09:14 -04005395 return idr_find(&ss->css_idr, id);
Stephane Eraniane5d13672011-02-14 11:20:01 +02005396}
5397
Paul Menagefe693432009-09-23 15:56:20 -07005398#ifdef CONFIG_CGROUP_DEBUG
Tejun Heoeb954192013-08-08 20:11:23 -04005399static struct cgroup_subsys_state *
5400debug_css_alloc(struct cgroup_subsys_state *parent_css)
Paul Menagefe693432009-09-23 15:56:20 -07005401{
5402 struct cgroup_subsys_state *css = kzalloc(sizeof(*css), GFP_KERNEL);
5403
5404 if (!css)
5405 return ERR_PTR(-ENOMEM);
5406
5407 return css;
5408}
5409
Tejun Heoeb954192013-08-08 20:11:23 -04005410static void debug_css_free(struct cgroup_subsys_state *css)
Paul Menagefe693432009-09-23 15:56:20 -07005411{
Tejun Heoeb954192013-08-08 20:11:23 -04005412 kfree(css);
Paul Menagefe693432009-09-23 15:56:20 -07005413}
5414
Tejun Heo182446d2013-08-08 20:11:24 -04005415static u64 debug_taskcount_read(struct cgroup_subsys_state *css,
5416 struct cftype *cft)
Paul Menagefe693432009-09-23 15:56:20 -07005417{
Tejun Heo182446d2013-08-08 20:11:24 -04005418 return cgroup_task_count(css->cgroup);
Paul Menagefe693432009-09-23 15:56:20 -07005419}
5420
Tejun Heo182446d2013-08-08 20:11:24 -04005421static u64 current_css_set_read(struct cgroup_subsys_state *css,
5422 struct cftype *cft)
Paul Menagefe693432009-09-23 15:56:20 -07005423{
5424 return (u64)(unsigned long)current->cgroups;
5425}
5426
Tejun Heo182446d2013-08-08 20:11:24 -04005427static u64 current_css_set_refcount_read(struct cgroup_subsys_state *css,
Li Zefan03c78cb2013-06-14 11:17:19 +08005428 struct cftype *cft)
Paul Menagefe693432009-09-23 15:56:20 -07005429{
5430 u64 count;
5431
5432 rcu_read_lock();
Tejun Heoa8ad8052013-06-21 15:52:04 -07005433 count = atomic_read(&task_css_set(current)->refcount);
Paul Menagefe693432009-09-23 15:56:20 -07005434 rcu_read_unlock();
5435 return count;
5436}
5437
Tejun Heo2da8ca82013-12-05 12:28:04 -05005438static int current_css_set_cg_links_read(struct seq_file *seq, void *v)
Paul Menage7717f7b2009-09-23 15:56:22 -07005439{
Tejun Heo69d02062013-06-12 21:04:50 -07005440 struct cgrp_cset_link *link;
Tejun Heo5abb8852013-06-12 21:04:49 -07005441 struct css_set *cset;
Tejun Heoe61734c2014-02-12 09:29:50 -05005442 char *name_buf;
Paul Menage7717f7b2009-09-23 15:56:22 -07005443
Tejun Heoe61734c2014-02-12 09:29:50 -05005444 name_buf = kmalloc(NAME_MAX + 1, GFP_KERNEL);
5445 if (!name_buf)
5446 return -ENOMEM;
Paul Menage7717f7b2009-09-23 15:56:22 -07005447
Tejun Heo96d365e2014-02-13 06:58:40 -05005448 down_read(&css_set_rwsem);
Paul Menage7717f7b2009-09-23 15:56:22 -07005449 rcu_read_lock();
Tejun Heo5abb8852013-06-12 21:04:49 -07005450 cset = rcu_dereference(current->cgroups);
Tejun Heo69d02062013-06-12 21:04:50 -07005451 list_for_each_entry(link, &cset->cgrp_links, cgrp_link) {
Paul Menage7717f7b2009-09-23 15:56:22 -07005452 struct cgroup *c = link->cgrp;
Paul Menage7717f7b2009-09-23 15:56:22 -07005453
Tejun Heoa2dd4242014-03-19 10:23:55 -04005454 cgroup_name(c, name_buf, NAME_MAX + 1);
Paul Menage2c6ab6d2009-09-23 15:56:23 -07005455 seq_printf(seq, "Root %d group %s\n",
Tejun Heoa2dd4242014-03-19 10:23:55 -04005456 c->root->hierarchy_id, name_buf);
Paul Menage7717f7b2009-09-23 15:56:22 -07005457 }
5458 rcu_read_unlock();
Tejun Heo96d365e2014-02-13 06:58:40 -05005459 up_read(&css_set_rwsem);
Tejun Heoe61734c2014-02-12 09:29:50 -05005460 kfree(name_buf);
Paul Menage7717f7b2009-09-23 15:56:22 -07005461 return 0;
5462}
5463
5464#define MAX_TASKS_SHOWN_PER_CSS 25
Tejun Heo2da8ca82013-12-05 12:28:04 -05005465static int cgroup_css_links_read(struct seq_file *seq, void *v)
Paul Menage7717f7b2009-09-23 15:56:22 -07005466{
Tejun Heo2da8ca82013-12-05 12:28:04 -05005467 struct cgroup_subsys_state *css = seq_css(seq);
Tejun Heo69d02062013-06-12 21:04:50 -07005468 struct cgrp_cset_link *link;
Paul Menage7717f7b2009-09-23 15:56:22 -07005469
Tejun Heo96d365e2014-02-13 06:58:40 -05005470 down_read(&css_set_rwsem);
Tejun Heo182446d2013-08-08 20:11:24 -04005471 list_for_each_entry(link, &css->cgroup->cset_links, cset_link) {
Tejun Heo69d02062013-06-12 21:04:50 -07005472 struct css_set *cset = link->cset;
Paul Menage7717f7b2009-09-23 15:56:22 -07005473 struct task_struct *task;
5474 int count = 0;
Tejun Heoc7561122014-02-25 10:04:01 -05005475
Tejun Heo5abb8852013-06-12 21:04:49 -07005476 seq_printf(seq, "css_set %p\n", cset);
Tejun Heoc7561122014-02-25 10:04:01 -05005477
Tejun Heo5abb8852013-06-12 21:04:49 -07005478 list_for_each_entry(task, &cset->tasks, cg_list) {
Tejun Heoc7561122014-02-25 10:04:01 -05005479 if (count++ > MAX_TASKS_SHOWN_PER_CSS)
5480 goto overflow;
5481 seq_printf(seq, " task %d\n", task_pid_vnr(task));
Paul Menage7717f7b2009-09-23 15:56:22 -07005482 }
Tejun Heoc7561122014-02-25 10:04:01 -05005483
5484 list_for_each_entry(task, &cset->mg_tasks, cg_list) {
5485 if (count++ > MAX_TASKS_SHOWN_PER_CSS)
5486 goto overflow;
5487 seq_printf(seq, " task %d\n", task_pid_vnr(task));
5488 }
5489 continue;
5490 overflow:
5491 seq_puts(seq, " ...\n");
Paul Menage7717f7b2009-09-23 15:56:22 -07005492 }
Tejun Heo96d365e2014-02-13 06:58:40 -05005493 up_read(&css_set_rwsem);
Paul Menage7717f7b2009-09-23 15:56:22 -07005494 return 0;
5495}
5496
Tejun Heo182446d2013-08-08 20:11:24 -04005497static u64 releasable_read(struct cgroup_subsys_state *css, struct cftype *cft)
Paul Menagefe693432009-09-23 15:56:20 -07005498{
Zefan Lia25eb522014-09-19 16:51:00 +08005499 return (!cgroup_has_tasks(css->cgroup) &&
5500 !css_has_online_children(&css->cgroup->self));
Paul Menagefe693432009-09-23 15:56:20 -07005501}
5502
5503static struct cftype debug_files[] = {
5504 {
Paul Menagefe693432009-09-23 15:56:20 -07005505 .name = "taskcount",
5506 .read_u64 = debug_taskcount_read,
5507 },
5508
5509 {
5510 .name = "current_css_set",
5511 .read_u64 = current_css_set_read,
5512 },
5513
5514 {
5515 .name = "current_css_set_refcount",
5516 .read_u64 = current_css_set_refcount_read,
5517 },
5518
5519 {
Paul Menage7717f7b2009-09-23 15:56:22 -07005520 .name = "current_css_set_cg_links",
Tejun Heo2da8ca82013-12-05 12:28:04 -05005521 .seq_show = current_css_set_cg_links_read,
Paul Menage7717f7b2009-09-23 15:56:22 -07005522 },
5523
5524 {
5525 .name = "cgroup_css_links",
Tejun Heo2da8ca82013-12-05 12:28:04 -05005526 .seq_show = cgroup_css_links_read,
Paul Menage7717f7b2009-09-23 15:56:22 -07005527 },
5528
5529 {
Paul Menagefe693432009-09-23 15:56:20 -07005530 .name = "releasable",
5531 .read_u64 = releasable_read,
5532 },
Paul Menagefe693432009-09-23 15:56:20 -07005533
Tejun Heo4baf6e32012-04-01 12:09:55 -07005534 { } /* terminate */
5535};
Paul Menagefe693432009-09-23 15:56:20 -07005536
Tejun Heo073219e2014-02-08 10:36:58 -05005537struct cgroup_subsys debug_cgrp_subsys = {
Tejun Heo92fb9742012-11-19 08:13:38 -08005538 .css_alloc = debug_css_alloc,
5539 .css_free = debug_css_free,
Tejun Heo55779642014-07-15 11:05:09 -04005540 .legacy_cftypes = debug_files,
Paul Menagefe693432009-09-23 15:56:20 -07005541};
5542#endif /* CONFIG_CGROUP_DEBUG */