Eric Lee / smarc-ti-linux-kernel

Download as
Browse Code ยป

Commit aec3dfcb2e43892180ee053e8c260dcdeccf4392

Authored by Tejun Heo
1 parent f392e51cd6
Exists in ti-lsk-linux-4.1.y and in 10 other branches dlt-processor-sdk-linux-03.00.00.04, processor-sdk-linux-02.00.01, smarc-ti-lsk-linux-4.1.y, smarct3x-processor-sdk-04.01.00.06, smarct3x-processor-sdk-linux-02.00.01, smarct3x-processor-sdk-linux-03.00.00.04, smarct4x-800-processor-sdk-linux-02.00.01, smarct4x-processor-sdk-04.01.00.06, smarct4x-processor-sdk-linux-02.00.01, smarct4x-processor-sdk-linux-03.00.00.04

cgroup: introduce effective cgroup_subsys_state 

In the planned default unified hierarchy, controllers may get
dynamically attached to and detached from a cgroup and a cgroup may
not have csses for all the controllers associated with the hierarchy.

When a cgroup doesn't have its own css for a given controller, the css
of the nearest ancestor with the controller enabled will be used,
which is called the effective css.  This patch introduces
cgroup_e_css() and for_each_e_css() to access the effective csses and
convert compare_css_sets(), find_existing_css_set() and
cgroup_migrate() to use the effective csses so that they can handle
cgroups with partial csses correctly.

This means that for two css_sets to be considered identical, they
should have both matching csses and cgroups.  compare_css_sets()
already compares both, not for correctness but for optimization.  As
this now becomes a matter of correctness, update the comments
accordingly.

For all !default hierarchies, cgroup_e_css() always equals
cgroup_css(), so this patch doesn't change behavior.

While at it, fix incorrect locking comment for for_each_css().

Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>

Showing 1 changed file with 64 additions and 19 deletions Inline Diff

1 /* 1 /*
2 * Generic process-grouping system. 2 * Generic process-grouping system.
3 * 3 *
4 * Based originally on the cpuset system, extracted by Paul Menage 4 * Based originally on the cpuset system, extracted by Paul Menage
5 * Copyright (C) 2006 Google, Inc 5 * Copyright (C) 2006 Google, Inc
6 * 6 *
7 * Notifications support 7 * Notifications support
8 * Copyright (C) 2009 Nokia Corporation 8 * Copyright (C) 2009 Nokia Corporation
9 * Author: Kirill A. Shutemov 9 * Author: Kirill A. Shutemov
10 * 10 *
11 * Copyright notices from the original cpuset code: 11 * Copyright notices from the original cpuset code:
12 * -------------------------------------------------- 12 * --------------------------------------------------
13 * Copyright (C) 2003 BULL SA. 13 * Copyright (C) 2003 BULL SA.
14 * Copyright (C) 2004-2006 Silicon Graphics, Inc. 14 * Copyright (C) 2004-2006 Silicon Graphics, Inc.
15 * 15 *
16 * Portions derived from Patrick Mochel's sysfs code. 16 * Portions derived from Patrick Mochel's sysfs code.
17 * sysfs is Copyright (c) 2001-3 Patrick Mochel 17 * sysfs is Copyright (c) 2001-3 Patrick Mochel
18 * 18 *
19 * 2003-10-10 Written by Simon Derr. 19 * 2003-10-10 Written by Simon Derr.
20 * 2003-10-22 Updates by Stephen Hemminger. 20 * 2003-10-22 Updates by Stephen Hemminger.
21 * 2004 May-July Rework by Paul Jackson. 21 * 2004 May-July Rework by Paul Jackson.
22 * --------------------------------------------------- 22 * ---------------------------------------------------
23 * 23 *
24 * This file is subject to the terms and conditions of the GNU General Public 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 25 * License. See the file COPYING in the main directory of the Linux
26 * distribution for more details. 26 * distribution for more details.
27 */ 27 */
28 28
29 #include <linux/cgroup.h> 29 #include <linux/cgroup.h>
30 #include <linux/cred.h> 30 #include <linux/cred.h>
31 #include <linux/ctype.h> 31 #include <linux/ctype.h>
32 #include <linux/errno.h> 32 #include <linux/errno.h>
33 #include <linux/init_task.h> 33 #include <linux/init_task.h>
34 #include <linux/kernel.h> 34 #include <linux/kernel.h>
35 #include <linux/list.h> 35 #include <linux/list.h>
36 #include <linux/mm.h> 36 #include <linux/mm.h>
37 #include <linux/mutex.h> 37 #include <linux/mutex.h>
38 #include <linux/mount.h> 38 #include <linux/mount.h>
39 #include <linux/pagemap.h> 39 #include <linux/pagemap.h>
40 #include <linux/proc_fs.h> 40 #include <linux/proc_fs.h>
41 #include <linux/rcupdate.h> 41 #include <linux/rcupdate.h>
42 #include <linux/sched.h> 42 #include <linux/sched.h>
43 #include <linux/slab.h> 43 #include <linux/slab.h>
44 #include <linux/spinlock.h> 44 #include <linux/spinlock.h>
45 #include <linux/rwsem.h> 45 #include <linux/rwsem.h>
46 #include <linux/string.h> 46 #include <linux/string.h>
47 #include <linux/sort.h> 47 #include <linux/sort.h>
48 #include <linux/kmod.h> 48 #include <linux/kmod.h>
49 #include <linux/delayacct.h> 49 #include <linux/delayacct.h>
50 #include <linux/cgroupstats.h> 50 #include <linux/cgroupstats.h>
51 #include <linux/hashtable.h> 51 #include <linux/hashtable.h>
52 #include <linux/pid_namespace.h> 52 #include <linux/pid_namespace.h>
53 #include <linux/idr.h> 53 #include <linux/idr.h>
54 #include <linux/vmalloc.h> /* TODO: replace with more sophisticated array */ 54 #include <linux/vmalloc.h> /* TODO: replace with more sophisticated array */
55 #include <linux/kthread.h> 55 #include <linux/kthread.h>
56 #include <linux/delay.h> 56 #include <linux/delay.h>
57 57
58 #include <linux/atomic.h> 58 #include <linux/atomic.h>
59 59
60 /* 60 /*
61 * pidlists linger the following amount before being destroyed. The goal 61 * pidlists linger the following amount before being destroyed. The goal
62 * is avoiding frequent destruction in the middle of consecutive read calls 62 * is avoiding frequent destruction in the middle of consecutive read calls
63 * Expiring in the middle is a performance problem not a correctness one. 63 * Expiring in the middle is a performance problem not a correctness one.
64 * 1 sec should be enough. 64 * 1 sec should be enough.
65 */ 65 */
66 #define CGROUP_PIDLIST_DESTROY_DELAY HZ 66 #define CGROUP_PIDLIST_DESTROY_DELAY HZ
67 67
68 #define CGROUP_FILE_NAME_MAX (MAX_CGROUP_TYPE_NAMELEN + \ 68 #define CGROUP_FILE_NAME_MAX (MAX_CGROUP_TYPE_NAMELEN + \
69 MAX_CFTYPE_NAME + 2) 69 MAX_CFTYPE_NAME + 2)
70 70
71 /* 71 /*
72 * cgroup_tree_mutex nests above cgroup_mutex and protects cftypes, file 72 * cgroup_tree_mutex nests above cgroup_mutex and protects cftypes, file
73 * creation/removal and hierarchy changing operations including cgroup 73 * creation/removal and hierarchy changing operations including cgroup
74 * creation, removal, css association and controller rebinding. This outer 74 * creation, removal, css association and controller rebinding. This outer
75 * lock is needed mainly to resolve the circular dependency between kernfs 75 * lock is needed mainly to resolve the circular dependency between kernfs
76 * active ref and cgroup_mutex. cgroup_tree_mutex nests above both. 76 * active ref and cgroup_mutex. cgroup_tree_mutex nests above both.
77 */ 77 */
78 static DEFINE_MUTEX(cgroup_tree_mutex); 78 static DEFINE_MUTEX(cgroup_tree_mutex);
79 79
80 /* 80 /*
81 * cgroup_mutex is the master lock. Any modification to cgroup or its 81 * cgroup_mutex is the master lock. Any modification to cgroup or its
82 * hierarchy must be performed while holding it. 82 * hierarchy must be performed while holding it.
83 * 83 *
84 * css_set_rwsem protects task->cgroups pointer, the list of css_set 84 * css_set_rwsem protects task->cgroups pointer, the list of css_set
85 * objects, and the chain of tasks off each css_set. 85 * objects, and the chain of tasks off each css_set.
86 * 86 *
87 * These locks are exported if CONFIG_PROVE_RCU so that accessors in 87 * These locks are exported if CONFIG_PROVE_RCU so that accessors in
88 * cgroup.h can use them for lockdep annotations. 88 * cgroup.h can use them for lockdep annotations.
89 */ 89 */
90 #ifdef CONFIG_PROVE_RCU 90 #ifdef CONFIG_PROVE_RCU
91 DEFINE_MUTEX(cgroup_mutex); 91 DEFINE_MUTEX(cgroup_mutex);
92 DECLARE_RWSEM(css_set_rwsem); 92 DECLARE_RWSEM(css_set_rwsem);
93 EXPORT_SYMBOL_GPL(cgroup_mutex); 93 EXPORT_SYMBOL_GPL(cgroup_mutex);
94 EXPORT_SYMBOL_GPL(css_set_rwsem); 94 EXPORT_SYMBOL_GPL(css_set_rwsem);
95 #else 95 #else
96 static DEFINE_MUTEX(cgroup_mutex); 96 static DEFINE_MUTEX(cgroup_mutex);
97 static DECLARE_RWSEM(css_set_rwsem); 97 static DECLARE_RWSEM(css_set_rwsem);
98 #endif 98 #endif
99 99
100 /* 100 /*
101 * Protects cgroup_subsys->release_agent_path. Modifying it also requires 101 * Protects cgroup_subsys->release_agent_path. Modifying it also requires
102 * cgroup_mutex. Reading requires either cgroup_mutex or this spinlock. 102 * cgroup_mutex. Reading requires either cgroup_mutex or this spinlock.
103 */ 103 */
104 static DEFINE_SPINLOCK(release_agent_path_lock); 104 static DEFINE_SPINLOCK(release_agent_path_lock);
105 105
106 #define cgroup_assert_mutexes_or_rcu_locked() \ 106 #define cgroup_assert_mutexes_or_rcu_locked() \
107 rcu_lockdep_assert(rcu_read_lock_held() || \ 107 rcu_lockdep_assert(rcu_read_lock_held() || \
108 lockdep_is_held(&cgroup_tree_mutex) || \ 108 lockdep_is_held(&cgroup_tree_mutex) || \
109 lockdep_is_held(&cgroup_mutex), \ 109 lockdep_is_held(&cgroup_mutex), \
110 "cgroup_[tree_]mutex or RCU read lock required"); 110 "cgroup_[tree_]mutex or RCU read lock required");
111 111
112 /* 112 /*
113 * cgroup destruction makes heavy use of work items and there can be a lot 113 * cgroup destruction makes heavy use of work items and there can be a lot
114 * of concurrent destructions. Use a separate workqueue so that cgroup 114 * of concurrent destructions. Use a separate workqueue so that cgroup
115 * destruction work items don't end up filling up max_active of system_wq 115 * destruction work items don't end up filling up max_active of system_wq
116 * which may lead to deadlock. 116 * which may lead to deadlock.
117 */ 117 */
118 static struct workqueue_struct *cgroup_destroy_wq; 118 static struct workqueue_struct *cgroup_destroy_wq;
119 119
120 /* 120 /*
121 * pidlist destructions need to be flushed on cgroup destruction. Use a 121 * pidlist destructions need to be flushed on cgroup destruction. Use a
122 * separate workqueue as flush domain. 122 * separate workqueue as flush domain.
123 */ 123 */
124 static struct workqueue_struct *cgroup_pidlist_destroy_wq; 124 static struct workqueue_struct *cgroup_pidlist_destroy_wq;
125 125
126 /* generate an array of cgroup subsystem pointers */ 126 /* generate an array of cgroup subsystem pointers */
127 #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys, 127 #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys,
128 static struct cgroup_subsys *cgroup_subsys[] = { 128 static struct cgroup_subsys *cgroup_subsys[] = {
129 #include <linux/cgroup_subsys.h> 129 #include <linux/cgroup_subsys.h>
130 }; 130 };
131 #undef SUBSYS 131 #undef SUBSYS
132 132
133 /* array of cgroup subsystem names */ 133 /* array of cgroup subsystem names */
134 #define SUBSYS(_x) [_x ## _cgrp_id] = #_x, 134 #define SUBSYS(_x) [_x ## _cgrp_id] = #_x,
135 static const char *cgroup_subsys_name[] = { 135 static const char *cgroup_subsys_name[] = {
136 #include <linux/cgroup_subsys.h> 136 #include <linux/cgroup_subsys.h>
137 }; 137 };
138 #undef SUBSYS 138 #undef SUBSYS
139 139
140 /* 140 /*
141 * The default hierarchy, reserved for the subsystems that are otherwise 141 * The default hierarchy, reserved for the subsystems that are otherwise
142 * unattached - it never has more than a single cgroup, and all tasks are 142 * unattached - it never has more than a single cgroup, and all tasks are
143 * part of that cgroup. 143 * part of that cgroup.
144 */ 144 */
145 struct cgroup_root cgrp_dfl_root; 145 struct cgroup_root cgrp_dfl_root;
146 146
147 /* 147 /*
148 * The default hierarchy always exists but is hidden until mounted for the 148 * The default hierarchy always exists but is hidden until mounted for the
149 * first time. This is for backward compatibility. 149 * first time. This is for backward compatibility.
150 */ 150 */
151 static bool cgrp_dfl_root_visible; 151 static bool cgrp_dfl_root_visible;
152 152
153 /* The list of hierarchy roots */ 153 /* The list of hierarchy roots */
154 154
155 static LIST_HEAD(cgroup_roots); 155 static LIST_HEAD(cgroup_roots);
156 static int cgroup_root_count; 156 static int cgroup_root_count;
157 157
158 /* hierarchy ID allocation and mapping, protected by cgroup_mutex */ 158 /* hierarchy ID allocation and mapping, protected by cgroup_mutex */
159 static DEFINE_IDR(cgroup_hierarchy_idr); 159 static DEFINE_IDR(cgroup_hierarchy_idr);
160 160
161 /* 161 /*
162 * Assign a monotonically increasing serial number to cgroups. It 162 * Assign a monotonically increasing serial number to cgroups. It
163 * guarantees cgroups with bigger numbers are newer than those with smaller 163 * guarantees cgroups with bigger numbers are newer than those with smaller
164 * numbers. Also, as cgroups are always appended to the parent's 164 * numbers. Also, as cgroups are always appended to the parent's
165 * ->children list, it guarantees that sibling cgroups are always sorted in 165 * ->children list, it guarantees that sibling cgroups are always sorted in
166 * the ascending serial number order on the list. Protected by 166 * the ascending serial number order on the list. Protected by
167 * cgroup_mutex. 167 * cgroup_mutex.
168 */ 168 */
169 static u64 cgroup_serial_nr_next = 1; 169 static u64 cgroup_serial_nr_next = 1;
170 170
171 /* This flag indicates whether tasks in the fork and exit paths should 171 /* This flag indicates whether tasks in the fork and exit paths should
172 * check for fork/exit handlers to call. This avoids us having to do 172 * check for fork/exit handlers to call. This avoids us having to do
173 * extra work in the fork/exit path if none of the subsystems need to 173 * extra work in the fork/exit path if none of the subsystems need to
174 * be called. 174 * be called.
175 */ 175 */
176 static int need_forkexit_callback __read_mostly; 176 static int need_forkexit_callback __read_mostly;
177 177
178 static struct cftype cgroup_base_files[]; 178 static struct cftype cgroup_base_files[];
179 179
180 static void cgroup_put(struct cgroup *cgrp); 180 static void cgroup_put(struct cgroup *cgrp);
181 static int rebind_subsystems(struct cgroup_root *dst_root, 181 static int rebind_subsystems(struct cgroup_root *dst_root,
182 unsigned long ss_mask); 182 unsigned long ss_mask);
183 static void cgroup_destroy_css_killed(struct cgroup *cgrp); 183 static void cgroup_destroy_css_killed(struct cgroup *cgrp);
184 static int cgroup_destroy_locked(struct cgroup *cgrp); 184 static int cgroup_destroy_locked(struct cgroup *cgrp);
185 static int cgroup_addrm_files(struct cgroup *cgrp, struct cftype cfts[], 185 static int cgroup_addrm_files(struct cgroup *cgrp, struct cftype cfts[],
186 bool is_add); 186 bool is_add);
187 static void cgroup_pidlist_destroy_all(struct cgroup *cgrp); 187 static void cgroup_pidlist_destroy_all(struct cgroup *cgrp);
188 188
189 /** 189 /**
190 * cgroup_css - obtain a cgroup's css for the specified subsystem 190 * cgroup_css - obtain a cgroup's css for the specified subsystem
191 * @cgrp: the cgroup of interest 191 * @cgrp: the cgroup of interest
192 * @ss: the subsystem of interest (%NULL returns the dummy_css) 192 * @ss: the subsystem of interest (%NULL returns the dummy_css)
193 * 193 *
194 * Return @cgrp's css (cgroup_subsys_state) associated with @ss. This 194 * Return @cgrp's css (cgroup_subsys_state) associated with @ss. This
195 * function must be called either under cgroup_mutex or rcu_read_lock() and 195 * function must be called either under cgroup_mutex or rcu_read_lock() and
196 * the caller is responsible for pinning the returned css if it wants to 196 * the caller is responsible for pinning the returned css if it wants to
197 * keep accessing it outside the said locks. This function may return 197 * keep accessing it outside the said locks. This function may return
198 * %NULL if @cgrp doesn't have @subsys_id enabled. 198 * %NULL if @cgrp doesn't have @subsys_id enabled.
199 */ 199 */
200 static struct cgroup_subsys_state *cgroup_css(struct cgroup *cgrp, 200 static struct cgroup_subsys_state *cgroup_css(struct cgroup *cgrp,
201 struct cgroup_subsys *ss) 201 struct cgroup_subsys *ss)
202 { 202 {
203 if (ss) 203 if (ss)
204 return rcu_dereference_check(cgrp->subsys[ss->id], 204 return rcu_dereference_check(cgrp->subsys[ss->id],
205 lockdep_is_held(&cgroup_tree_mutex) || 205 lockdep_is_held(&cgroup_tree_mutex) ||
206 lockdep_is_held(&cgroup_mutex)); 206 lockdep_is_held(&cgroup_mutex));
207 else 207 else
208 return &cgrp->dummy_css; 208 return &cgrp->dummy_css;
209 } 209 }
210 210
211 /**
212 * cgroup_e_css - obtain a cgroup's effective css for the specified subsystem
213 * @cgrp: the cgroup of interest
214 * @ss: the subsystem of interest (%NULL returns the dummy_css)
215 *
216 * Similar to cgroup_css() but returns the effctive css, which is defined
217 * as the matching css of the nearest ancestor including self which has @ss
218 * enabled. If @ss is associated with the hierarchy @cgrp is on, this
219 * function is guaranteed to return non-NULL css.
220 */
221 static struct cgroup_subsys_state *cgroup_e_css(struct cgroup *cgrp,
222 struct cgroup_subsys *ss)
223 {
224 lockdep_assert_held(&cgroup_mutex);
225
226 if (!ss)
227 return &cgrp->dummy_css;
228
229 if (!(cgrp->root->subsys_mask & (1 << ss->id)))
230 return NULL;
231
232 while (cgrp->parent &&
233 !(cgrp->parent->child_subsys_mask & (1 << ss->id)))
234 cgrp = cgrp->parent;
235
236 return cgroup_css(cgrp, ss);
237 }
238
211 /* convenient tests for these bits */ 239 /* convenient tests for these bits */
212 static inline bool cgroup_is_dead(const struct cgroup *cgrp) 240 static inline bool cgroup_is_dead(const struct cgroup *cgrp)
213 { 241 {
214 return test_bit(CGRP_DEAD, &cgrp->flags); 242 return test_bit(CGRP_DEAD, &cgrp->flags);
215 } 243 }
216 244
217 struct cgroup_subsys_state *seq_css(struct seq_file *seq) 245 struct cgroup_subsys_state *seq_css(struct seq_file *seq)
218 { 246 {
219 struct kernfs_open_file *of = seq->private; 247 struct kernfs_open_file *of = seq->private;
220 struct cgroup *cgrp = of->kn->parent->priv; 248 struct cgroup *cgrp = of->kn->parent->priv;
221 struct cftype *cft = seq_cft(seq); 249 struct cftype *cft = seq_cft(seq);
222 250
223 /* 251 /*
224 * This is open and unprotected implementation of cgroup_css(). 252 * This is open and unprotected implementation of cgroup_css().
225 * seq_css() is only called from a kernfs file operation which has 253 * seq_css() is only called from a kernfs file operation which has
226 * an active reference on the file. Because all the subsystem 254 * an active reference on the file. Because all the subsystem
227 * files are drained before a css is disassociated with a cgroup, 255 * files are drained before a css is disassociated with a cgroup,
228 * the matching css from the cgroup's subsys table is guaranteed to 256 * the matching css from the cgroup's subsys table is guaranteed to
229 * be and stay valid until the enclosing operation is complete. 257 * be and stay valid until the enclosing operation is complete.
230 */ 258 */
231 if (cft->ss) 259 if (cft->ss)
232 return rcu_dereference_raw(cgrp->subsys[cft->ss->id]); 260 return rcu_dereference_raw(cgrp->subsys[cft->ss->id]);
233 else 261 else
234 return &cgrp->dummy_css; 262 return &cgrp->dummy_css;
235 } 263 }
236 EXPORT_SYMBOL_GPL(seq_css); 264 EXPORT_SYMBOL_GPL(seq_css);
237 265
238 /** 266 /**
239 * cgroup_is_descendant - test ancestry 267 * cgroup_is_descendant - test ancestry
240 * @cgrp: the cgroup to be tested 268 * @cgrp: the cgroup to be tested
241 * @ancestor: possible ancestor of @cgrp 269 * @ancestor: possible ancestor of @cgrp
242 * 270 *
243 * Test whether @cgrp is a descendant of @ancestor. It also returns %true 271 * Test whether @cgrp is a descendant of @ancestor. It also returns %true
244 * if @cgrp == @ancestor. This function is safe to call as long as @cgrp 272 * if @cgrp == @ancestor. This function is safe to call as long as @cgrp
245 * and @ancestor are accessible. 273 * and @ancestor are accessible.
246 */ 274 */
247 bool cgroup_is_descendant(struct cgroup *cgrp, struct cgroup *ancestor) 275 bool cgroup_is_descendant(struct cgroup *cgrp, struct cgroup *ancestor)
248 { 276 {
249 while (cgrp) { 277 while (cgrp) {
250 if (cgrp == ancestor) 278 if (cgrp == ancestor)
251 return true; 279 return true;
252 cgrp = cgrp->parent; 280 cgrp = cgrp->parent;
253 } 281 }
254 return false; 282 return false;
255 } 283 }
256 284
257 static int cgroup_is_releasable(const struct cgroup *cgrp) 285 static int cgroup_is_releasable(const struct cgroup *cgrp)
258 { 286 {
259 const int bits = 287 const int bits =
260 (1 << CGRP_RELEASABLE) | 288 (1 << CGRP_RELEASABLE) |
261 (1 << CGRP_NOTIFY_ON_RELEASE); 289 (1 << CGRP_NOTIFY_ON_RELEASE);
262 return (cgrp->flags & bits) == bits; 290 return (cgrp->flags & bits) == bits;
263 } 291 }
264 292
265 static int notify_on_release(const struct cgroup *cgrp) 293 static int notify_on_release(const struct cgroup *cgrp)
266 { 294 {
267 return test_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); 295 return test_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
268 } 296 }
269 297
270 /** 298 /**
271 * for_each_css - iterate all css's of a cgroup 299 * for_each_css - iterate all css's of a cgroup
272 * @css: the iteration cursor 300 * @css: the iteration cursor
273 * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end 301 * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end
274 * @cgrp: the target cgroup to iterate css's of 302 * @cgrp: the target cgroup to iterate css's of
275 * 303 *
276 * Should be called under cgroup_mutex. 304 * Should be called under cgroup_[tree_]mutex.
277 */ 305 */
278 #define for_each_css(css, ssid, cgrp) \ 306 #define for_each_css(css, ssid, cgrp) \
279 for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \ 307 for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \
280 if (!((css) = rcu_dereference_check( \ 308 if (!((css) = rcu_dereference_check( \
281 (cgrp)->subsys[(ssid)], \ 309 (cgrp)->subsys[(ssid)], \
282 lockdep_is_held(&cgroup_tree_mutex) || \ 310 lockdep_is_held(&cgroup_tree_mutex) || \
283 lockdep_is_held(&cgroup_mutex)))) { } \ 311 lockdep_is_held(&cgroup_mutex)))) { } \
284 else 312 else
285 313
286 /** 314 /**
315 * for_each_e_css - iterate all effective css's of a cgroup
316 * @css: the iteration cursor
317 * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end
318 * @cgrp: the target cgroup to iterate css's of
319 *
320 * Should be called under cgroup_[tree_]mutex.
321 */
322 #define for_each_e_css(css, ssid, cgrp) \
323 for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \
324 if (!((css) = cgroup_e_css(cgrp, cgroup_subsys[(ssid)]))) \
325 ; \
326 else
327
328 /**
287 * for_each_subsys - iterate all enabled cgroup subsystems 329 * for_each_subsys - iterate all enabled cgroup subsystems
288 * @ss: the iteration cursor 330 * @ss: the iteration cursor
289 * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end 331 * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end
290 */ 332 */
291 #define for_each_subsys(ss, ssid) \ 333 #define for_each_subsys(ss, ssid) \
292 for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT && \ 334 for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT && \
293 (((ss) = cgroup_subsys[ssid]) || true); (ssid)++) 335 (((ss) = cgroup_subsys[ssid]) || true); (ssid)++)
294 336
295 /* iterate across the hierarchies */ 337 /* iterate across the hierarchies */
296 #define for_each_root(root) \ 338 #define for_each_root(root) \
297 list_for_each_entry((root), &cgroup_roots, root_list) 339 list_for_each_entry((root), &cgroup_roots, root_list)
298 340
299 /** 341 /**
300 * cgroup_lock_live_group - take cgroup_mutex and check that cgrp is alive. 342 * cgroup_lock_live_group - take cgroup_mutex and check that cgrp is alive.
301 * @cgrp: the cgroup to be checked for liveness 343 * @cgrp: the cgroup to be checked for liveness
302 * 344 *
303 * On success, returns true; the mutex should be later unlocked. On 345 * On success, returns true; the mutex should be later unlocked. On
304 * failure returns false with no lock held. 346 * failure returns false with no lock held.
305 */ 347 */
306 static bool cgroup_lock_live_group(struct cgroup *cgrp) 348 static bool cgroup_lock_live_group(struct cgroup *cgrp)
307 { 349 {
308 mutex_lock(&cgroup_mutex); 350 mutex_lock(&cgroup_mutex);
309 if (cgroup_is_dead(cgrp)) { 351 if (cgroup_is_dead(cgrp)) {
310 mutex_unlock(&cgroup_mutex); 352 mutex_unlock(&cgroup_mutex);
311 return false; 353 return false;
312 } 354 }
313 return true; 355 return true;
314 } 356 }
315 357
316 /* the list of cgroups eligible for automatic release. Protected by 358 /* the list of cgroups eligible for automatic release. Protected by
317 * release_list_lock */ 359 * release_list_lock */
318 static LIST_HEAD(release_list); 360 static LIST_HEAD(release_list);
319 static DEFINE_RAW_SPINLOCK(release_list_lock); 361 static DEFINE_RAW_SPINLOCK(release_list_lock);
320 static void cgroup_release_agent(struct work_struct *work); 362 static void cgroup_release_agent(struct work_struct *work);
321 static DECLARE_WORK(release_agent_work, cgroup_release_agent); 363 static DECLARE_WORK(release_agent_work, cgroup_release_agent);
322 static void check_for_release(struct cgroup *cgrp); 364 static void check_for_release(struct cgroup *cgrp);
323 365
324 /* 366 /*
325 * A cgroup can be associated with multiple css_sets as different tasks may 367 * A cgroup can be associated with multiple css_sets as different tasks may
326 * belong to different cgroups on different hierarchies. In the other 368 * belong to different cgroups on different hierarchies. In the other
327 * direction, a css_set is naturally associated with multiple cgroups. 369 * direction, a css_set is naturally associated with multiple cgroups.
328 * This M:N relationship is represented by the following link structure 370 * This M:N relationship is represented by the following link structure
329 * which exists for each association and allows traversing the associations 371 * which exists for each association and allows traversing the associations
330 * from both sides. 372 * from both sides.
331 */ 373 */
332 struct cgrp_cset_link { 374 struct cgrp_cset_link {
333 /* the cgroup and css_set this link associates */ 375 /* the cgroup and css_set this link associates */
334 struct cgroup *cgrp; 376 struct cgroup *cgrp;
335 struct css_set *cset; 377 struct css_set *cset;
336 378
337 /* list of cgrp_cset_links anchored at cgrp->cset_links */ 379 /* list of cgrp_cset_links anchored at cgrp->cset_links */
338 struct list_head cset_link; 380 struct list_head cset_link;
339 381
340 /* list of cgrp_cset_links anchored at css_set->cgrp_links */ 382 /* list of cgrp_cset_links anchored at css_set->cgrp_links */
341 struct list_head cgrp_link; 383 struct list_head cgrp_link;
342 }; 384 };
343 385
344 /* 386 /*
345 * The default css_set - used by init and its children prior to any 387 * The default css_set - used by init and its children prior to any
346 * hierarchies being mounted. It contains a pointer to the root state 388 * hierarchies being mounted. It contains a pointer to the root state
347 * for each subsystem. Also used to anchor the list of css_sets. Not 389 * for each subsystem. Also used to anchor the list of css_sets. Not
348 * reference-counted, to improve performance when child cgroups 390 * reference-counted, to improve performance when child cgroups
349 * haven't been created. 391 * haven't been created.
350 */ 392 */
351 static struct css_set init_css_set = { 393 static struct css_set init_css_set = {
352 .refcount = ATOMIC_INIT(1), 394 .refcount = ATOMIC_INIT(1),
353 .cgrp_links = LIST_HEAD_INIT(init_css_set.cgrp_links), 395 .cgrp_links = LIST_HEAD_INIT(init_css_set.cgrp_links),
354 .tasks = LIST_HEAD_INIT(init_css_set.tasks), 396 .tasks = LIST_HEAD_INIT(init_css_set.tasks),
355 .mg_tasks = LIST_HEAD_INIT(init_css_set.mg_tasks), 397 .mg_tasks = LIST_HEAD_INIT(init_css_set.mg_tasks),
356 .mg_preload_node = LIST_HEAD_INIT(init_css_set.mg_preload_node), 398 .mg_preload_node = LIST_HEAD_INIT(init_css_set.mg_preload_node),
357 .mg_node = LIST_HEAD_INIT(init_css_set.mg_node), 399 .mg_node = LIST_HEAD_INIT(init_css_set.mg_node),
358 }; 400 };
359 401
360 static int css_set_count = 1; /* 1 for init_css_set */ 402 static int css_set_count = 1; /* 1 for init_css_set */
361 403
362 /* 404 /*
363 * hash table for cgroup groups. This improves the performance to find 405 * hash table for cgroup groups. This improves the performance to find
364 * an existing css_set. This hash doesn't (currently) take into 406 * an existing css_set. This hash doesn't (currently) take into
365 * account cgroups in empty hierarchies. 407 * account cgroups in empty hierarchies.
366 */ 408 */
367 #define CSS_SET_HASH_BITS 7 409 #define CSS_SET_HASH_BITS 7
368 static DEFINE_HASHTABLE(css_set_table, CSS_SET_HASH_BITS); 410 static DEFINE_HASHTABLE(css_set_table, CSS_SET_HASH_BITS);
369 411
370 static unsigned long css_set_hash(struct cgroup_subsys_state *css[]) 412 static unsigned long css_set_hash(struct cgroup_subsys_state *css[])
371 { 413 {
372 unsigned long key = 0UL; 414 unsigned long key = 0UL;
373 struct cgroup_subsys *ss; 415 struct cgroup_subsys *ss;
374 int i; 416 int i;
375 417
376 for_each_subsys(ss, i) 418 for_each_subsys(ss, i)
377 key += (unsigned long)css[i]; 419 key += (unsigned long)css[i];
378 key = (key >> 16) ^ key; 420 key = (key >> 16) ^ key;
379 421
380 return key; 422 return key;
381 } 423 }
382 424
383 static void put_css_set_locked(struct css_set *cset, bool taskexit) 425 static void put_css_set_locked(struct css_set *cset, bool taskexit)
384 { 426 {
385 struct cgrp_cset_link *link, *tmp_link; 427 struct cgrp_cset_link *link, *tmp_link;
386 428
387 lockdep_assert_held(&css_set_rwsem); 429 lockdep_assert_held(&css_set_rwsem);
388 430
389 if (!atomic_dec_and_test(&cset->refcount)) 431 if (!atomic_dec_and_test(&cset->refcount))
390 return; 432 return;
391 433
392 /* This css_set is dead. unlink it and release cgroup refcounts */ 434 /* This css_set is dead. unlink it and release cgroup refcounts */
393 hash_del(&cset->hlist); 435 hash_del(&cset->hlist);
394 css_set_count--; 436 css_set_count--;
395 437
396 list_for_each_entry_safe(link, tmp_link, &cset->cgrp_links, cgrp_link) { 438 list_for_each_entry_safe(link, tmp_link, &cset->cgrp_links, cgrp_link) {
397 struct cgroup *cgrp = link->cgrp; 439 struct cgroup *cgrp = link->cgrp;
398 440
399 list_del(&link->cset_link); 441 list_del(&link->cset_link);
400 list_del(&link->cgrp_link); 442 list_del(&link->cgrp_link);
401 443
402 /* @cgrp can't go away while we're holding css_set_rwsem */ 444 /* @cgrp can't go away while we're holding css_set_rwsem */
403 if (list_empty(&cgrp->cset_links) && notify_on_release(cgrp)) { 445 if (list_empty(&cgrp->cset_links) && notify_on_release(cgrp)) {
404 if (taskexit) 446 if (taskexit)
405 set_bit(CGRP_RELEASABLE, &cgrp->flags); 447 set_bit(CGRP_RELEASABLE, &cgrp->flags);
406 check_for_release(cgrp); 448 check_for_release(cgrp);
407 } 449 }
408 450
409 kfree(link); 451 kfree(link);
410 } 452 }
411 453
412 kfree_rcu(cset, rcu_head); 454 kfree_rcu(cset, rcu_head);
413 } 455 }
414 456
415 static void put_css_set(struct css_set *cset, bool taskexit) 457 static void put_css_set(struct css_set *cset, bool taskexit)
416 { 458 {
417 /* 459 /*
418 * Ensure that the refcount doesn't hit zero while any readers 460 * Ensure that the refcount doesn't hit zero while any readers
419 * can see it. Similar to atomic_dec_and_lock(), but for an 461 * can see it. Similar to atomic_dec_and_lock(), but for an
420 * rwlock 462 * rwlock
421 */ 463 */
422 if (atomic_add_unless(&cset->refcount, -1, 1)) 464 if (atomic_add_unless(&cset->refcount, -1, 1))
423 return; 465 return;
424 466
425 down_write(&css_set_rwsem); 467 down_write(&css_set_rwsem);
426 put_css_set_locked(cset, taskexit); 468 put_css_set_locked(cset, taskexit);
427 up_write(&css_set_rwsem); 469 up_write(&css_set_rwsem);
428 } 470 }
429 471
430 /* 472 /*
431 * refcounted get/put for css_set objects 473 * refcounted get/put for css_set objects
432 */ 474 */
433 static inline void get_css_set(struct css_set *cset) 475 static inline void get_css_set(struct css_set *cset)
434 { 476 {
435 atomic_inc(&cset->refcount); 477 atomic_inc(&cset->refcount);
436 } 478 }
437 479
438 /** 480 /**
439 * compare_css_sets - helper function for find_existing_css_set(). 481 * compare_css_sets - helper function for find_existing_css_set().
440 * @cset: candidate css_set being tested 482 * @cset: candidate css_set being tested
441 * @old_cset: existing css_set for a task 483 * @old_cset: existing css_set for a task
442 * @new_cgrp: cgroup that's being entered by the task 484 * @new_cgrp: cgroup that's being entered by the task
443 * @template: desired set of css pointers in css_set (pre-calculated) 485 * @template: desired set of css pointers in css_set (pre-calculated)
444 * 486 *
445 * Returns true if "cset" matches "old_cset" except for the hierarchy 487 * Returns true if "cset" matches "old_cset" except for the hierarchy
446 * which "new_cgrp" belongs to, for which it should match "new_cgrp". 488 * which "new_cgrp" belongs to, for which it should match "new_cgrp".
447 */ 489 */
448 static bool compare_css_sets(struct css_set *cset, 490 static bool compare_css_sets(struct css_set *cset,
449 struct css_set *old_cset, 491 struct css_set *old_cset,
450 struct cgroup *new_cgrp, 492 struct cgroup *new_cgrp,
451 struct cgroup_subsys_state *template[]) 493 struct cgroup_subsys_state *template[])
452 { 494 {
453 struct list_head *l1, *l2; 495 struct list_head *l1, *l2;
454 496
455 if (memcmp(template, cset->subsys, sizeof(cset->subsys))) { 497 /*
456 /* Not all subsystems matched */ 498 * On the default hierarchy, there can be csets which are
499 * associated with the same set of cgroups but different csses.
500 * Let's first ensure that csses match.
501 */
502 if (memcmp(template, cset->subsys, sizeof(cset->subsys)))
457 return false; 503 return false;
458 }
459 504
460 /* 505 /*
461 * Compare cgroup pointers in order to distinguish between 506 * Compare cgroup pointers in order to distinguish between
462 * different cgroups in heirarchies with no subsystems. We 507 * different cgroups in hierarchies. As different cgroups may
463 * could get by with just this check alone (and skip the 508 * share the same effective css, this comparison is always
464 * memcmp above) but on most setups the memcmp check will 509 * necessary.
465 * avoid the need for this more expensive check on almost all
466 * candidates.
467 */ 510 */
468
469 l1 = &cset->cgrp_links; 511 l1 = &cset->cgrp_links;
470 l2 = &old_cset->cgrp_links; 512 l2 = &old_cset->cgrp_links;
471 while (1) { 513 while (1) {
472 struct cgrp_cset_link *link1, *link2; 514 struct cgrp_cset_link *link1, *link2;
473 struct cgroup *cgrp1, *cgrp2; 515 struct cgroup *cgrp1, *cgrp2;
474 516
475 l1 = l1->next; 517 l1 = l1->next;
476 l2 = l2->next; 518 l2 = l2->next;
477 /* See if we reached the end - both lists are equal length. */ 519 /* See if we reached the end - both lists are equal length. */
478 if (l1 == &cset->cgrp_links) { 520 if (l1 == &cset->cgrp_links) {
479 BUG_ON(l2 != &old_cset->cgrp_links); 521 BUG_ON(l2 != &old_cset->cgrp_links);
480 break; 522 break;
481 } else { 523 } else {
482 BUG_ON(l2 == &old_cset->cgrp_links); 524 BUG_ON(l2 == &old_cset->cgrp_links);
483 } 525 }
484 /* Locate the cgroups associated with these links. */ 526 /* Locate the cgroups associated with these links. */
485 link1 = list_entry(l1, struct cgrp_cset_link, cgrp_link); 527 link1 = list_entry(l1, struct cgrp_cset_link, cgrp_link);
486 link2 = list_entry(l2, struct cgrp_cset_link, cgrp_link); 528 link2 = list_entry(l2, struct cgrp_cset_link, cgrp_link);
487 cgrp1 = link1->cgrp; 529 cgrp1 = link1->cgrp;
488 cgrp2 = link2->cgrp; 530 cgrp2 = link2->cgrp;
489 /* Hierarchies should be linked in the same order. */ 531 /* Hierarchies should be linked in the same order. */
490 BUG_ON(cgrp1->root != cgrp2->root); 532 BUG_ON(cgrp1->root != cgrp2->root);
491 533
492 /* 534 /*
493 * If this hierarchy is the hierarchy of the cgroup 535 * If this hierarchy is the hierarchy of the cgroup
494 * that's changing, then we need to check that this 536 * that's changing, then we need to check that this
495 * css_set points to the new cgroup; if it's any other 537 * css_set points to the new cgroup; if it's any other
496 * hierarchy, then this css_set should point to the 538 * hierarchy, then this css_set should point to the
497 * same cgroup as the old css_set. 539 * same cgroup as the old css_set.
498 */ 540 */
499 if (cgrp1->root == new_cgrp->root) { 541 if (cgrp1->root == new_cgrp->root) {
500 if (cgrp1 != new_cgrp) 542 if (cgrp1 != new_cgrp)
501 return false; 543 return false;
502 } else { 544 } else {
503 if (cgrp1 != cgrp2) 545 if (cgrp1 != cgrp2)
504 return false; 546 return false;
505 } 547 }
506 } 548 }
507 return true; 549 return true;
508 } 550 }
509 551
510 /** 552 /**
511 * find_existing_css_set - init css array and find the matching css_set 553 * find_existing_css_set - init css array and find the matching css_set
512 * @old_cset: the css_set that we're using before the cgroup transition 554 * @old_cset: the css_set that we're using before the cgroup transition
513 * @cgrp: the cgroup that we're moving into 555 * @cgrp: the cgroup that we're moving into
514 * @template: out param for the new set of csses, should be clear on entry 556 * @template: out param for the new set of csses, should be clear on entry
515 */ 557 */
516 static struct css_set *find_existing_css_set(struct css_set *old_cset, 558 static struct css_set *find_existing_css_set(struct css_set *old_cset,
517 struct cgroup *cgrp, 559 struct cgroup *cgrp,
518 struct cgroup_subsys_state *template[]) 560 struct cgroup_subsys_state *template[])
519 { 561 {
520 struct cgroup_root *root = cgrp->root; 562 struct cgroup_root *root = cgrp->root;
521 struct cgroup_subsys *ss; 563 struct cgroup_subsys *ss;
522 struct css_set *cset; 564 struct css_set *cset;
523 unsigned long key; 565 unsigned long key;
524 int i; 566 int i;
525 567
526 /* 568 /*
527 * Build the set of subsystem state objects that we want to see in the 569 * Build the set of subsystem state objects that we want to see in the
528 * new css_set. while subsystems can change globally, the entries here 570 * new css_set. while subsystems can change globally, the entries here
529 * won't change, so no need for locking. 571 * won't change, so no need for locking.
530 */ 572 */
531 for_each_subsys(ss, i) { 573 for_each_subsys(ss, i) {
532 if (root->subsys_mask & (1UL << i)) { 574 if (root->subsys_mask & (1UL << i)) {
533 /* Subsystem is in this hierarchy. So we want 575 /*
534 * the subsystem state from the new 576 * @ss is in this hierarchy, so we want the
535 * cgroup */ 577 * effective css from @cgrp.
536 template[i] = cgroup_css(cgrp, ss); 578 */
579 template[i] = cgroup_e_css(cgrp, ss);
537 } else { 580 } else {
538 /* Subsystem is not in this hierarchy, so we 581 /*
539 * don't want to change the subsystem state */ 582 * @ss is not in this hierarchy, so we don't want
583 * to change the css.
584 */
540 template[i] = old_cset->subsys[i]; 585 template[i] = old_cset->subsys[i];
541 } 586 }
542 } 587 }
543 588
544 key = css_set_hash(template); 589 key = css_set_hash(template);
545 hash_for_each_possible(css_set_table, cset, hlist, key) { 590 hash_for_each_possible(css_set_table, cset, hlist, key) {
546 if (!compare_css_sets(cset, old_cset, cgrp, template)) 591 if (!compare_css_sets(cset, old_cset, cgrp, template))
547 continue; 592 continue;
548 593
549 /* This css_set matches what we need */ 594 /* This css_set matches what we need */
550 return cset; 595 return cset;
551 } 596 }
552 597
553 /* No existing cgroup group matched */ 598 /* No existing cgroup group matched */
554 return NULL; 599 return NULL;
555 } 600 }
556 601
557 static void free_cgrp_cset_links(struct list_head *links_to_free) 602 static void free_cgrp_cset_links(struct list_head *links_to_free)
558 { 603 {
559 struct cgrp_cset_link *link, *tmp_link; 604 struct cgrp_cset_link *link, *tmp_link;
560 605
561 list_for_each_entry_safe(link, tmp_link, links_to_free, cset_link) { 606 list_for_each_entry_safe(link, tmp_link, links_to_free, cset_link) {
562 list_del(&link->cset_link); 607 list_del(&link->cset_link);
563 kfree(link); 608 kfree(link);
564 } 609 }
565 } 610 }
566 611
567 /** 612 /**
568 * allocate_cgrp_cset_links - allocate cgrp_cset_links 613 * allocate_cgrp_cset_links - allocate cgrp_cset_links
569 * @count: the number of links to allocate 614 * @count: the number of links to allocate
570 * @tmp_links: list_head the allocated links are put on 615 * @tmp_links: list_head the allocated links are put on
571 * 616 *
572 * Allocate @count cgrp_cset_link structures and chain them on @tmp_links 617 * Allocate @count cgrp_cset_link structures and chain them on @tmp_links
573 * through ->cset_link. Returns 0 on success or -errno. 618 * through ->cset_link. Returns 0 on success or -errno.
574 */ 619 */
575 static int allocate_cgrp_cset_links(int count, struct list_head *tmp_links) 620 static int allocate_cgrp_cset_links(int count, struct list_head *tmp_links)
576 { 621 {
577 struct cgrp_cset_link *link; 622 struct cgrp_cset_link *link;
578 int i; 623 int i;
579 624
580 INIT_LIST_HEAD(tmp_links); 625 INIT_LIST_HEAD(tmp_links);
581 626
582 for (i = 0; i < count; i++) { 627 for (i = 0; i < count; i++) {
583 link = kzalloc(sizeof(*link), GFP_KERNEL); 628 link = kzalloc(sizeof(*link), GFP_KERNEL);
584 if (!link) { 629 if (!link) {
585 free_cgrp_cset_links(tmp_links); 630 free_cgrp_cset_links(tmp_links);
586 return -ENOMEM; 631 return -ENOMEM;
587 } 632 }
588 list_add(&link->cset_link, tmp_links); 633 list_add(&link->cset_link, tmp_links);
589 } 634 }
590 return 0; 635 return 0;
591 } 636 }
592 637
593 /** 638 /**
594 * link_css_set - a helper function to link a css_set to a cgroup 639 * link_css_set - a helper function to link a css_set to a cgroup
595 * @tmp_links: cgrp_cset_link objects allocated by allocate_cgrp_cset_links() 640 * @tmp_links: cgrp_cset_link objects allocated by allocate_cgrp_cset_links()
596 * @cset: the css_set to be linked 641 * @cset: the css_set to be linked
597 * @cgrp: the destination cgroup 642 * @cgrp: the destination cgroup
598 */ 643 */
599 static void link_css_set(struct list_head *tmp_links, struct css_set *cset, 644 static void link_css_set(struct list_head *tmp_links, struct css_set *cset,
600 struct cgroup *cgrp) 645 struct cgroup *cgrp)
601 { 646 {
602 struct cgrp_cset_link *link; 647 struct cgrp_cset_link *link;
603 648
604 BUG_ON(list_empty(tmp_links)); 649 BUG_ON(list_empty(tmp_links));
605 link = list_first_entry(tmp_links, struct cgrp_cset_link, cset_link); 650 link = list_first_entry(tmp_links, struct cgrp_cset_link, cset_link);
606 link->cset = cset; 651 link->cset = cset;
607 link->cgrp = cgrp; 652 link->cgrp = cgrp;
608 list_move(&link->cset_link, &cgrp->cset_links); 653 list_move(&link->cset_link, &cgrp->cset_links);
609 /* 654 /*
610 * Always add links to the tail of the list so that the list 655 * Always add links to the tail of the list so that the list
611 * is sorted by order of hierarchy creation 656 * is sorted by order of hierarchy creation
612 */ 657 */
613 list_add_tail(&link->cgrp_link, &cset->cgrp_links); 658 list_add_tail(&link->cgrp_link, &cset->cgrp_links);
614 } 659 }
615 660
616 /** 661 /**
617 * find_css_set - return a new css_set with one cgroup updated 662 * find_css_set - return a new css_set with one cgroup updated
618 * @old_cset: the baseline css_set 663 * @old_cset: the baseline css_set
619 * @cgrp: the cgroup to be updated 664 * @cgrp: the cgroup to be updated
620 * 665 *
621 * Return a new css_set that's equivalent to @old_cset, but with @cgrp 666 * Return a new css_set that's equivalent to @old_cset, but with @cgrp
622 * substituted into the appropriate hierarchy. 667 * substituted into the appropriate hierarchy.
623 */ 668 */
624 static struct css_set *find_css_set(struct css_set *old_cset, 669 static struct css_set *find_css_set(struct css_set *old_cset,
625 struct cgroup *cgrp) 670 struct cgroup *cgrp)
626 { 671 {
627 struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT] = { }; 672 struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT] = { };
628 struct css_set *cset; 673 struct css_set *cset;
629 struct list_head tmp_links; 674 struct list_head tmp_links;
630 struct cgrp_cset_link *link; 675 struct cgrp_cset_link *link;
631 unsigned long key; 676 unsigned long key;
632 677
633 lockdep_assert_held(&cgroup_mutex); 678 lockdep_assert_held(&cgroup_mutex);
634 679
635 /* First see if we already have a cgroup group that matches 680 /* First see if we already have a cgroup group that matches
636 * the desired set */ 681 * the desired set */
637 down_read(&css_set_rwsem); 682 down_read(&css_set_rwsem);
638 cset = find_existing_css_set(old_cset, cgrp, template); 683 cset = find_existing_css_set(old_cset, cgrp, template);
639 if (cset) 684 if (cset)
640 get_css_set(cset); 685 get_css_set(cset);
641 up_read(&css_set_rwsem); 686 up_read(&css_set_rwsem);
642 687
643 if (cset) 688 if (cset)
644 return cset; 689 return cset;
645 690
646 cset = kzalloc(sizeof(*cset), GFP_KERNEL); 691 cset = kzalloc(sizeof(*cset), GFP_KERNEL);
647 if (!cset) 692 if (!cset)
648 return NULL; 693 return NULL;
649 694
650 /* Allocate all the cgrp_cset_link objects that we'll need */ 695 /* Allocate all the cgrp_cset_link objects that we'll need */
651 if (allocate_cgrp_cset_links(cgroup_root_count, &tmp_links) < 0) { 696 if (allocate_cgrp_cset_links(cgroup_root_count, &tmp_links) < 0) {
652 kfree(cset); 697 kfree(cset);
653 return NULL; 698 return NULL;
654 } 699 }
655 700
656 atomic_set(&cset->refcount, 1); 701 atomic_set(&cset->refcount, 1);
657 INIT_LIST_HEAD(&cset->cgrp_links); 702 INIT_LIST_HEAD(&cset->cgrp_links);
658 INIT_LIST_HEAD(&cset->tasks); 703 INIT_LIST_HEAD(&cset->tasks);
659 INIT_LIST_HEAD(&cset->mg_tasks); 704 INIT_LIST_HEAD(&cset->mg_tasks);
660 INIT_LIST_HEAD(&cset->mg_preload_node); 705 INIT_LIST_HEAD(&cset->mg_preload_node);
661 INIT_LIST_HEAD(&cset->mg_node); 706 INIT_LIST_HEAD(&cset->mg_node);
662 INIT_HLIST_NODE(&cset->hlist); 707 INIT_HLIST_NODE(&cset->hlist);
663 708
664 /* Copy the set of subsystem state objects generated in 709 /* Copy the set of subsystem state objects generated in
665 * find_existing_css_set() */ 710 * find_existing_css_set() */
666 memcpy(cset->subsys, template, sizeof(cset->subsys)); 711 memcpy(cset->subsys, template, sizeof(cset->subsys));
667 712
668 down_write(&css_set_rwsem); 713 down_write(&css_set_rwsem);
669 /* Add reference counts and links from the new css_set. */ 714 /* Add reference counts and links from the new css_set. */
670 list_for_each_entry(link, &old_cset->cgrp_links, cgrp_link) { 715 list_for_each_entry(link, &old_cset->cgrp_links, cgrp_link) {
671 struct cgroup *c = link->cgrp; 716 struct cgroup *c = link->cgrp;
672 717
673 if (c->root == cgrp->root) 718 if (c->root == cgrp->root)
674 c = cgrp; 719 c = cgrp;
675 link_css_set(&tmp_links, cset, c); 720 link_css_set(&tmp_links, cset, c);
676 } 721 }
677 722
678 BUG_ON(!list_empty(&tmp_links)); 723 BUG_ON(!list_empty(&tmp_links));
679 724
680 css_set_count++; 725 css_set_count++;
681 726
682 /* Add this cgroup group to the hash table */ 727 /* Add this cgroup group to the hash table */
683 key = css_set_hash(cset->subsys); 728 key = css_set_hash(cset->subsys);
684 hash_add(css_set_table, &cset->hlist, key); 729 hash_add(css_set_table, &cset->hlist, key);
685 730
686 up_write(&css_set_rwsem); 731 up_write(&css_set_rwsem);
687 732
688 return cset; 733 return cset;
689 } 734 }
690 735
691 static struct cgroup_root *cgroup_root_from_kf(struct kernfs_root *kf_root) 736 static struct cgroup_root *cgroup_root_from_kf(struct kernfs_root *kf_root)
692 { 737 {
693 struct cgroup *root_cgrp = kf_root->kn->priv; 738 struct cgroup *root_cgrp = kf_root->kn->priv;
694 739
695 return root_cgrp->root; 740 return root_cgrp->root;
696 } 741 }
697 742
698 static int cgroup_init_root_id(struct cgroup_root *root) 743 static int cgroup_init_root_id(struct cgroup_root *root)
699 { 744 {
700 int id; 745 int id;
701 746
702 lockdep_assert_held(&cgroup_mutex); 747 lockdep_assert_held(&cgroup_mutex);
703 748
704 id = idr_alloc_cyclic(&cgroup_hierarchy_idr, root, 0, 0, GFP_KERNEL); 749 id = idr_alloc_cyclic(&cgroup_hierarchy_idr, root, 0, 0, GFP_KERNEL);
705 if (id < 0) 750 if (id < 0)
706 return id; 751 return id;
707 752
708 root->hierarchy_id = id; 753 root->hierarchy_id = id;
709 return 0; 754 return 0;
710 } 755 }
711 756
712 static void cgroup_exit_root_id(struct cgroup_root *root) 757 static void cgroup_exit_root_id(struct cgroup_root *root)
713 { 758 {
714 lockdep_assert_held(&cgroup_mutex); 759 lockdep_assert_held(&cgroup_mutex);
715 760
716 if (root->hierarchy_id) { 761 if (root->hierarchy_id) {
717 idr_remove(&cgroup_hierarchy_idr, root->hierarchy_id); 762 idr_remove(&cgroup_hierarchy_idr, root->hierarchy_id);
718 root->hierarchy_id = 0; 763 root->hierarchy_id = 0;
719 } 764 }
720 } 765 }
721 766
722 static void cgroup_free_root(struct cgroup_root *root) 767 static void cgroup_free_root(struct cgroup_root *root)
723 { 768 {
724 if (root) { 769 if (root) {
725 /* hierarhcy ID shoulid already have been released */ 770 /* hierarhcy ID shoulid already have been released */
726 WARN_ON_ONCE(root->hierarchy_id); 771 WARN_ON_ONCE(root->hierarchy_id);
727 772
728 idr_destroy(&root->cgroup_idr); 773 idr_destroy(&root->cgroup_idr);
729 kfree(root); 774 kfree(root);
730 } 775 }
731 } 776 }
732 777
733 static void cgroup_destroy_root(struct cgroup_root *root) 778 static void cgroup_destroy_root(struct cgroup_root *root)
734 { 779 {
735 struct cgroup *cgrp = &root->cgrp; 780 struct cgroup *cgrp = &root->cgrp;
736 struct cgrp_cset_link *link, *tmp_link; 781 struct cgrp_cset_link *link, *tmp_link;
737 782
738 mutex_lock(&cgroup_tree_mutex); 783 mutex_lock(&cgroup_tree_mutex);
739 mutex_lock(&cgroup_mutex); 784 mutex_lock(&cgroup_mutex);
740 785
741 BUG_ON(atomic_read(&root->nr_cgrps)); 786 BUG_ON(atomic_read(&root->nr_cgrps));
742 BUG_ON(!list_empty(&cgrp->children)); 787 BUG_ON(!list_empty(&cgrp->children));
743 788
744 /* Rebind all subsystems back to the default hierarchy */ 789 /* Rebind all subsystems back to the default hierarchy */
745 rebind_subsystems(&cgrp_dfl_root, root->subsys_mask); 790 rebind_subsystems(&cgrp_dfl_root, root->subsys_mask);
746 791
747 /* 792 /*
748 * Release all the links from cset_links to this hierarchy's 793 * Release all the links from cset_links to this hierarchy's
749 * root cgroup 794 * root cgroup
750 */ 795 */
751 down_write(&css_set_rwsem); 796 down_write(&css_set_rwsem);
752 797
753 list_for_each_entry_safe(link, tmp_link, &cgrp->cset_links, cset_link) { 798 list_for_each_entry_safe(link, tmp_link, &cgrp->cset_links, cset_link) {
754 list_del(&link->cset_link); 799 list_del(&link->cset_link);
755 list_del(&link->cgrp_link); 800 list_del(&link->cgrp_link);
756 kfree(link); 801 kfree(link);
757 } 802 }
758 up_write(&css_set_rwsem); 803 up_write(&css_set_rwsem);
759 804
760 if (!list_empty(&root->root_list)) { 805 if (!list_empty(&root->root_list)) {
761 list_del(&root->root_list); 806 list_del(&root->root_list);
762 cgroup_root_count--; 807 cgroup_root_count--;
763 } 808 }
764 809
765 cgroup_exit_root_id(root); 810 cgroup_exit_root_id(root);
766 811
767 mutex_unlock(&cgroup_mutex); 812 mutex_unlock(&cgroup_mutex);
768 mutex_unlock(&cgroup_tree_mutex); 813 mutex_unlock(&cgroup_tree_mutex);
769 814
770 kernfs_destroy_root(root->kf_root); 815 kernfs_destroy_root(root->kf_root);
771 cgroup_free_root(root); 816 cgroup_free_root(root);
772 } 817 }
773 818
774 /* look up cgroup associated with given css_set on the specified hierarchy */ 819 /* look up cgroup associated with given css_set on the specified hierarchy */
775 static struct cgroup *cset_cgroup_from_root(struct css_set *cset, 820 static struct cgroup *cset_cgroup_from_root(struct css_set *cset,
776 struct cgroup_root *root) 821 struct cgroup_root *root)
777 { 822 {
778 struct cgroup *res = NULL; 823 struct cgroup *res = NULL;
779 824
780 lockdep_assert_held(&cgroup_mutex); 825 lockdep_assert_held(&cgroup_mutex);
781 lockdep_assert_held(&css_set_rwsem); 826 lockdep_assert_held(&css_set_rwsem);
782 827
783 if (cset == &init_css_set) { 828 if (cset == &init_css_set) {
784 res = &root->cgrp; 829 res = &root->cgrp;
785 } else { 830 } else {
786 struct cgrp_cset_link *link; 831 struct cgrp_cset_link *link;
787 832
788 list_for_each_entry(link, &cset->cgrp_links, cgrp_link) { 833 list_for_each_entry(link, &cset->cgrp_links, cgrp_link) {
789 struct cgroup *c = link->cgrp; 834 struct cgroup *c = link->cgrp;
790 835
791 if (c->root == root) { 836 if (c->root == root) {
792 res = c; 837 res = c;
793 break; 838 break;
794 } 839 }
795 } 840 }
796 } 841 }
797 842
798 BUG_ON(!res); 843 BUG_ON(!res);
799 return res; 844 return res;
800 } 845 }
801 846
802 /* 847 /*
803 * Return the cgroup for "task" from the given hierarchy. Must be 848 * Return the cgroup for "task" from the given hierarchy. Must be
804 * called with cgroup_mutex and css_set_rwsem held. 849 * called with cgroup_mutex and css_set_rwsem held.
805 */ 850 */
806 static struct cgroup *task_cgroup_from_root(struct task_struct *task, 851 static struct cgroup *task_cgroup_from_root(struct task_struct *task,
807 struct cgroup_root *root) 852 struct cgroup_root *root)
808 { 853 {
809 /* 854 /*
810 * No need to lock the task - since we hold cgroup_mutex the 855 * No need to lock the task - since we hold cgroup_mutex the
811 * task can't change groups, so the only thing that can happen 856 * task can't change groups, so the only thing that can happen
812 * is that it exits and its css is set back to init_css_set. 857 * is that it exits and its css is set back to init_css_set.
813 */ 858 */
814 return cset_cgroup_from_root(task_css_set(task), root); 859 return cset_cgroup_from_root(task_css_set(task), root);
815 } 860 }
816 861
817 /* 862 /*
818 * A task must hold cgroup_mutex to modify cgroups. 863 * A task must hold cgroup_mutex to modify cgroups.
819 * 864 *
820 * Any task can increment and decrement the count field without lock. 865 * Any task can increment and decrement the count field without lock.
821 * So in general, code holding cgroup_mutex can't rely on the count 866 * So in general, code holding cgroup_mutex can't rely on the count
822 * field not changing. However, if the count goes to zero, then only 867 * field not changing. However, if the count goes to zero, then only
823 * cgroup_attach_task() can increment it again. Because a count of zero 868 * cgroup_attach_task() can increment it again. Because a count of zero
824 * means that no tasks are currently attached, therefore there is no 869 * means that no tasks are currently attached, therefore there is no
825 * way a task attached to that cgroup can fork (the other way to 870 * way a task attached to that cgroup can fork (the other way to
826 * increment the count). So code holding cgroup_mutex can safely 871 * increment the count). So code holding cgroup_mutex can safely
827 * assume that if the count is zero, it will stay zero. Similarly, if 872 * assume that if the count is zero, it will stay zero. Similarly, if
828 * a task holds cgroup_mutex on a cgroup with zero count, it 873 * a task holds cgroup_mutex on a cgroup with zero count, it
829 * knows that the cgroup won't be removed, as cgroup_rmdir() 874 * knows that the cgroup won't be removed, as cgroup_rmdir()
830 * needs that mutex. 875 * needs that mutex.
831 * 876 *
832 * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't 877 * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't
833 * (usually) take cgroup_mutex. These are the two most performance 878 * (usually) take cgroup_mutex. These are the two most performance
834 * critical pieces of code here. The exception occurs on cgroup_exit(), 879 * critical pieces of code here. The exception occurs on cgroup_exit(),
835 * when a task in a notify_on_release cgroup exits. Then cgroup_mutex 880 * when a task in a notify_on_release cgroup exits. Then cgroup_mutex
836 * is taken, and if the cgroup count is zero, a usermode call made 881 * is taken, and if the cgroup count is zero, a usermode call made
837 * to the release agent with the name of the cgroup (path relative to 882 * to the release agent with the name of the cgroup (path relative to
838 * the root of cgroup file system) as the argument. 883 * the root of cgroup file system) as the argument.
839 * 884 *
840 * A cgroup can only be deleted if both its 'count' of using tasks 885 * A cgroup can only be deleted if both its 'count' of using tasks
841 * is zero, and its list of 'children' cgroups is empty. Since all 886 * is zero, and its list of 'children' cgroups is empty. Since all
842 * tasks in the system use _some_ cgroup, and since there is always at 887 * tasks in the system use _some_ cgroup, and since there is always at
843 * least one task in the system (init, pid == 1), therefore, root cgroup 888 * least one task in the system (init, pid == 1), therefore, root cgroup
844 * always has either children cgroups and/or using tasks. So we don't 889 * always has either children cgroups and/or using tasks. So we don't
845 * need a special hack to ensure that root cgroup cannot be deleted. 890 * need a special hack to ensure that root cgroup cannot be deleted.
846 * 891 *
847 * P.S. One more locking exception. RCU is used to guard the 892 * P.S. One more locking exception. RCU is used to guard the
848 * update of a tasks cgroup pointer by cgroup_attach_task() 893 * update of a tasks cgroup pointer by cgroup_attach_task()
849 */ 894 */
850 895
851 static int cgroup_populate_dir(struct cgroup *cgrp, unsigned long subsys_mask); 896 static int cgroup_populate_dir(struct cgroup *cgrp, unsigned long subsys_mask);
852 static struct kernfs_syscall_ops cgroup_kf_syscall_ops; 897 static struct kernfs_syscall_ops cgroup_kf_syscall_ops;
853 static const struct file_operations proc_cgroupstats_operations; 898 static const struct file_operations proc_cgroupstats_operations;
854 899
855 static char *cgroup_file_name(struct cgroup *cgrp, const struct cftype *cft, 900 static char *cgroup_file_name(struct cgroup *cgrp, const struct cftype *cft,
856 char *buf) 901 char *buf)
857 { 902 {
858 if (cft->ss && !(cft->flags & CFTYPE_NO_PREFIX) && 903 if (cft->ss && !(cft->flags & CFTYPE_NO_PREFIX) &&
859 !(cgrp->root->flags & CGRP_ROOT_NOPREFIX)) 904 !(cgrp->root->flags & CGRP_ROOT_NOPREFIX))
860 snprintf(buf, CGROUP_FILE_NAME_MAX, "%s.%s", 905 snprintf(buf, CGROUP_FILE_NAME_MAX, "%s.%s",
861 cft->ss->name, cft->name); 906 cft->ss->name, cft->name);
862 else 907 else
863 strncpy(buf, cft->name, CGROUP_FILE_NAME_MAX); 908 strncpy(buf, cft->name, CGROUP_FILE_NAME_MAX);
864 return buf; 909 return buf;
865 } 910 }
866 911
867 /** 912 /**
868 * cgroup_file_mode - deduce file mode of a control file 913 * cgroup_file_mode - deduce file mode of a control file
869 * @cft: the control file in question 914 * @cft: the control file in question
870 * 915 *
871 * returns cft->mode if ->mode is not 0 916 * returns cft->mode if ->mode is not 0
872 * returns S_IRUGO|S_IWUSR if it has both a read and a write handler 917 * returns S_IRUGO|S_IWUSR if it has both a read and a write handler
873 * returns S_IRUGO if it has only a read handler 918 * returns S_IRUGO if it has only a read handler
874 * returns S_IWUSR if it has only a write hander 919 * returns S_IWUSR if it has only a write hander
875 */ 920 */
876 static umode_t cgroup_file_mode(const struct cftype *cft) 921 static umode_t cgroup_file_mode(const struct cftype *cft)
877 { 922 {
878 umode_t mode = 0; 923 umode_t mode = 0;
879 924
880 if (cft->mode) 925 if (cft->mode)
881 return cft->mode; 926 return cft->mode;
882 927
883 if (cft->read_u64 || cft->read_s64 || cft->seq_show) 928 if (cft->read_u64 || cft->read_s64 || cft->seq_show)
884 mode |= S_IRUGO; 929 mode |= S_IRUGO;
885 930
886 if (cft->write_u64 || cft->write_s64 || cft->write_string || 931 if (cft->write_u64 || cft->write_s64 || cft->write_string ||
887 cft->trigger) 932 cft->trigger)
888 mode |= S_IWUSR; 933 mode |= S_IWUSR;
889 934
890 return mode; 935 return mode;
891 } 936 }
892 937
893 static void cgroup_free_fn(struct work_struct *work) 938 static void cgroup_free_fn(struct work_struct *work)
894 { 939 {
895 struct cgroup *cgrp = container_of(work, struct cgroup, destroy_work); 940 struct cgroup *cgrp = container_of(work, struct cgroup, destroy_work);
896 941
897 atomic_dec(&cgrp->root->nr_cgrps); 942 atomic_dec(&cgrp->root->nr_cgrps);
898 cgroup_pidlist_destroy_all(cgrp); 943 cgroup_pidlist_destroy_all(cgrp);
899 944
900 if (cgrp->parent) { 945 if (cgrp->parent) {
901 /* 946 /*
902 * We get a ref to the parent, and put the ref when this 947 * We get a ref to the parent, and put the ref when this
903 * cgroup is being freed, so it's guaranteed that the 948 * cgroup is being freed, so it's guaranteed that the
904 * parent won't be destroyed before its children. 949 * parent won't be destroyed before its children.
905 */ 950 */
906 cgroup_put(cgrp->parent); 951 cgroup_put(cgrp->parent);
907 kernfs_put(cgrp->kn); 952 kernfs_put(cgrp->kn);
908 kfree(cgrp); 953 kfree(cgrp);
909 } else { 954 } else {
910 /* 955 /*
911 * This is root cgroup's refcnt reaching zero, which 956 * This is root cgroup's refcnt reaching zero, which
912 * indicates that the root should be released. 957 * indicates that the root should be released.
913 */ 958 */
914 cgroup_destroy_root(cgrp->root); 959 cgroup_destroy_root(cgrp->root);
915 } 960 }
916 } 961 }
917 962
918 static void cgroup_free_rcu(struct rcu_head *head) 963 static void cgroup_free_rcu(struct rcu_head *head)
919 { 964 {
920 struct cgroup *cgrp = container_of(head, struct cgroup, rcu_head); 965 struct cgroup *cgrp = container_of(head, struct cgroup, rcu_head);
921 966
922 INIT_WORK(&cgrp->destroy_work, cgroup_free_fn); 967 INIT_WORK(&cgrp->destroy_work, cgroup_free_fn);
923 queue_work(cgroup_destroy_wq, &cgrp->destroy_work); 968 queue_work(cgroup_destroy_wq, &cgrp->destroy_work);
924 } 969 }
925 970
926 static void cgroup_get(struct cgroup *cgrp) 971 static void cgroup_get(struct cgroup *cgrp)
927 { 972 {
928 WARN_ON_ONCE(cgroup_is_dead(cgrp)); 973 WARN_ON_ONCE(cgroup_is_dead(cgrp));
929 WARN_ON_ONCE(atomic_read(&cgrp->refcnt) <= 0); 974 WARN_ON_ONCE(atomic_read(&cgrp->refcnt) <= 0);
930 atomic_inc(&cgrp->refcnt); 975 atomic_inc(&cgrp->refcnt);
931 } 976 }
932 977
933 static void cgroup_put(struct cgroup *cgrp) 978 static void cgroup_put(struct cgroup *cgrp)
934 { 979 {
935 if (!atomic_dec_and_test(&cgrp->refcnt)) 980 if (!atomic_dec_and_test(&cgrp->refcnt))
936 return; 981 return;
937 if (WARN_ON_ONCE(cgrp->parent && !cgroup_is_dead(cgrp))) 982 if (WARN_ON_ONCE(cgrp->parent && !cgroup_is_dead(cgrp)))
938 return; 983 return;
939 984
940 /* 985 /*
941 * XXX: cgrp->id is only used to look up css's. As cgroup and 986 * XXX: cgrp->id is only used to look up css's. As cgroup and
942 * css's lifetimes will be decoupled, it should be made 987 * css's lifetimes will be decoupled, it should be made
943 * per-subsystem and moved to css->id so that lookups are 988 * per-subsystem and moved to css->id so that lookups are
944 * successful until the target css is released. 989 * successful until the target css is released.
945 */ 990 */
946 mutex_lock(&cgroup_mutex); 991 mutex_lock(&cgroup_mutex);
947 idr_remove(&cgrp->root->cgroup_idr, cgrp->id); 992 idr_remove(&cgrp->root->cgroup_idr, cgrp->id);
948 mutex_unlock(&cgroup_mutex); 993 mutex_unlock(&cgroup_mutex);
949 cgrp->id = -1; 994 cgrp->id = -1;
950 995
951 call_rcu(&cgrp->rcu_head, cgroup_free_rcu); 996 call_rcu(&cgrp->rcu_head, cgroup_free_rcu);
952 } 997 }
953 998
954 static void cgroup_rm_file(struct cgroup *cgrp, const struct cftype *cft) 999 static void cgroup_rm_file(struct cgroup *cgrp, const struct cftype *cft)
955 { 1000 {
956 char name[CGROUP_FILE_NAME_MAX]; 1001 char name[CGROUP_FILE_NAME_MAX];
957 1002
958 lockdep_assert_held(&cgroup_tree_mutex); 1003 lockdep_assert_held(&cgroup_tree_mutex);
959 kernfs_remove_by_name(cgrp->kn, cgroup_file_name(cgrp, cft, name)); 1004 kernfs_remove_by_name(cgrp->kn, cgroup_file_name(cgrp, cft, name));
960 } 1005 }
961 1006
962 /** 1007 /**
963 * cgroup_clear_dir - remove subsys files in a cgroup directory 1008 * cgroup_clear_dir - remove subsys files in a cgroup directory
964 * @cgrp: target cgroup 1009 * @cgrp: target cgroup
965 * @subsys_mask: mask of the subsystem ids whose files should be removed 1010 * @subsys_mask: mask of the subsystem ids whose files should be removed
966 */ 1011 */
967 static void cgroup_clear_dir(struct cgroup *cgrp, unsigned long subsys_mask) 1012 static void cgroup_clear_dir(struct cgroup *cgrp, unsigned long subsys_mask)
968 { 1013 {
969 struct cgroup_subsys *ss; 1014 struct cgroup_subsys *ss;
970 int i; 1015 int i;
971 1016
972 for_each_subsys(ss, i) { 1017 for_each_subsys(ss, i) {
973 struct cftype *cfts; 1018 struct cftype *cfts;
974 1019
975 if (!test_bit(i, &subsys_mask)) 1020 if (!test_bit(i, &subsys_mask))
976 continue; 1021 continue;
977 list_for_each_entry(cfts, &ss->cfts, node) 1022 list_for_each_entry(cfts, &ss->cfts, node)
978 cgroup_addrm_files(cgrp, cfts, false); 1023 cgroup_addrm_files(cgrp, cfts, false);
979 } 1024 }
980 } 1025 }
981 1026
982 static int rebind_subsystems(struct cgroup_root *dst_root, 1027 static int rebind_subsystems(struct cgroup_root *dst_root,
983 unsigned long ss_mask) 1028 unsigned long ss_mask)
984 { 1029 {
985 struct cgroup_subsys *ss; 1030 struct cgroup_subsys *ss;
986 int ssid, ret; 1031 int ssid, ret;
987 1032
988 lockdep_assert_held(&cgroup_tree_mutex); 1033 lockdep_assert_held(&cgroup_tree_mutex);
989 lockdep_assert_held(&cgroup_mutex); 1034 lockdep_assert_held(&cgroup_mutex);
990 1035
991 for_each_subsys(ss, ssid) { 1036 for_each_subsys(ss, ssid) {
992 if (!(ss_mask & (1 << ssid))) 1037 if (!(ss_mask & (1 << ssid)))
993 continue; 1038 continue;
994 1039
995 /* if @ss is on the dummy_root, we can always move it */ 1040 /* if @ss is on the dummy_root, we can always move it */
996 if (ss->root == &cgrp_dfl_root) 1041 if (ss->root == &cgrp_dfl_root)
997 continue; 1042 continue;
998 1043
999 /* if @ss has non-root cgroups attached to it, can't move */ 1044 /* if @ss has non-root cgroups attached to it, can't move */
1000 if (!list_empty(&ss->root->cgrp.children)) 1045 if (!list_empty(&ss->root->cgrp.children))
1001 return -EBUSY; 1046 return -EBUSY;
1002 1047
1003 /* can't move between two non-dummy roots either */ 1048 /* can't move between two non-dummy roots either */
1004 if (dst_root != &cgrp_dfl_root) 1049 if (dst_root != &cgrp_dfl_root)
1005 return -EBUSY; 1050 return -EBUSY;
1006 } 1051 }
1007 1052
1008 ret = cgroup_populate_dir(&dst_root->cgrp, ss_mask); 1053 ret = cgroup_populate_dir(&dst_root->cgrp, ss_mask);
1009 if (ret) { 1054 if (ret) {
1010 if (dst_root != &cgrp_dfl_root) 1055 if (dst_root != &cgrp_dfl_root)
1011 return ret; 1056 return ret;
1012 1057
1013 /* 1058 /*
1014 * Rebinding back to the default root is not allowed to 1059 * Rebinding back to the default root is not allowed to
1015 * fail. Using both default and non-default roots should 1060 * fail. Using both default and non-default roots should
1016 * be rare. Moving subsystems back and forth even more so. 1061 * be rare. Moving subsystems back and forth even more so.
1017 * Just warn about it and continue. 1062 * Just warn about it and continue.
1018 */ 1063 */
1019 if (cgrp_dfl_root_visible) { 1064 if (cgrp_dfl_root_visible) {
1020 pr_warning("cgroup: failed to create files (%d) while rebinding 0x%lx to default root\n", 1065 pr_warning("cgroup: failed to create files (%d) while rebinding 0x%lx to default root\n",
1021 ret, ss_mask); 1066 ret, ss_mask);
1022 pr_warning("cgroup: you may retry by moving them to a different hierarchy and unbinding\n"); 1067 pr_warning("cgroup: you may retry by moving them to a different hierarchy and unbinding\n");
1023 } 1068 }
1024 } 1069 }
1025 1070
1026 /* 1071 /*
1027 * Nothing can fail from this point on. Remove files for the 1072 * Nothing can fail from this point on. Remove files for the
1028 * removed subsystems and rebind each subsystem. 1073 * removed subsystems and rebind each subsystem.
1029 */ 1074 */
1030 mutex_unlock(&cgroup_mutex); 1075 mutex_unlock(&cgroup_mutex);
1031 for_each_subsys(ss, ssid) 1076 for_each_subsys(ss, ssid)
1032 if (ss_mask & (1 << ssid)) 1077 if (ss_mask & (1 << ssid))
1033 cgroup_clear_dir(&ss->root->cgrp, 1 << ssid); 1078 cgroup_clear_dir(&ss->root->cgrp, 1 << ssid);
1034 mutex_lock(&cgroup_mutex); 1079 mutex_lock(&cgroup_mutex);
1035 1080
1036 for_each_subsys(ss, ssid) { 1081 for_each_subsys(ss, ssid) {
1037 struct cgroup_root *src_root; 1082 struct cgroup_root *src_root;
1038 struct cgroup_subsys_state *css; 1083 struct cgroup_subsys_state *css;
1039 1084
1040 if (!(ss_mask & (1 << ssid))) 1085 if (!(ss_mask & (1 << ssid)))
1041 continue; 1086 continue;
1042 1087
1043 src_root = ss->root; 1088 src_root = ss->root;
1044 css = cgroup_css(&src_root->cgrp, ss); 1089 css = cgroup_css(&src_root->cgrp, ss);
1045 1090
1046 WARN_ON(!css || cgroup_css(&dst_root->cgrp, ss)); 1091 WARN_ON(!css || cgroup_css(&dst_root->cgrp, ss));
1047 1092
1048 RCU_INIT_POINTER(src_root->cgrp.subsys[ssid], NULL); 1093 RCU_INIT_POINTER(src_root->cgrp.subsys[ssid], NULL);
1049 rcu_assign_pointer(dst_root->cgrp.subsys[ssid], css); 1094 rcu_assign_pointer(dst_root->cgrp.subsys[ssid], css);
1050 ss->root = dst_root; 1095 ss->root = dst_root;
1051 css->cgroup = &dst_root->cgrp; 1096 css->cgroup = &dst_root->cgrp;
1052 1097
1053 src_root->subsys_mask &= ~(1 << ssid); 1098 src_root->subsys_mask &= ~(1 << ssid);
1054 src_root->cgrp.child_subsys_mask &= ~(1 << ssid); 1099 src_root->cgrp.child_subsys_mask &= ~(1 << ssid);
1055 1100
1056 dst_root->subsys_mask |= 1 << ssid; 1101 dst_root->subsys_mask |= 1 << ssid;
1057 dst_root->cgrp.child_subsys_mask |= 1 << ssid; 1102 dst_root->cgrp.child_subsys_mask |= 1 << ssid;
1058 1103
1059 if (ss->bind) 1104 if (ss->bind)
1060 ss->bind(css); 1105 ss->bind(css);
1061 } 1106 }
1062 1107
1063 kernfs_activate(dst_root->cgrp.kn); 1108 kernfs_activate(dst_root->cgrp.kn);
1064 return 0; 1109 return 0;
1065 } 1110 }
1066 1111
1067 static int cgroup_show_options(struct seq_file *seq, 1112 static int cgroup_show_options(struct seq_file *seq,
1068 struct kernfs_root *kf_root) 1113 struct kernfs_root *kf_root)
1069 { 1114 {
1070 struct cgroup_root *root = cgroup_root_from_kf(kf_root); 1115 struct cgroup_root *root = cgroup_root_from_kf(kf_root);
1071 struct cgroup_subsys *ss; 1116 struct cgroup_subsys *ss;
1072 int ssid; 1117 int ssid;
1073 1118
1074 for_each_subsys(ss, ssid) 1119 for_each_subsys(ss, ssid)
1075 if (root->subsys_mask & (1 << ssid)) 1120 if (root->subsys_mask & (1 << ssid))
1076 seq_printf(seq, ",%s", ss->name); 1121 seq_printf(seq, ",%s", ss->name);
1077 if (root->flags & CGRP_ROOT_SANE_BEHAVIOR) 1122 if (root->flags & CGRP_ROOT_SANE_BEHAVIOR)
1078 seq_puts(seq, ",sane_behavior"); 1123 seq_puts(seq, ",sane_behavior");
1079 if (root->flags & CGRP_ROOT_NOPREFIX) 1124 if (root->flags & CGRP_ROOT_NOPREFIX)
1080 seq_puts(seq, ",noprefix"); 1125 seq_puts(seq, ",noprefix");
1081 if (root->flags & CGRP_ROOT_XATTR) 1126 if (root->flags & CGRP_ROOT_XATTR)
1082 seq_puts(seq, ",xattr"); 1127 seq_puts(seq, ",xattr");
1083 1128
1084 spin_lock(&release_agent_path_lock); 1129 spin_lock(&release_agent_path_lock);
1085 if (strlen(root->release_agent_path)) 1130 if (strlen(root->release_agent_path))
1086 seq_printf(seq, ",release_agent=%s", root->release_agent_path); 1131 seq_printf(seq, ",release_agent=%s", root->release_agent_path);
1087 spin_unlock(&release_agent_path_lock); 1132 spin_unlock(&release_agent_path_lock);
1088 1133
1089 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags)) 1134 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags))
1090 seq_puts(seq, ",clone_children"); 1135 seq_puts(seq, ",clone_children");
1091 if (strlen(root->name)) 1136 if (strlen(root->name))
1092 seq_printf(seq, ",name=%s", root->name); 1137 seq_printf(seq, ",name=%s", root->name);
1093 return 0; 1138 return 0;
1094 } 1139 }
1095 1140
1096 struct cgroup_sb_opts { 1141 struct cgroup_sb_opts {
1097 unsigned long subsys_mask; 1142 unsigned long subsys_mask;
1098 unsigned long flags; 1143 unsigned long flags;
1099 char *release_agent; 1144 char *release_agent;
1100 bool cpuset_clone_children; 1145 bool cpuset_clone_children;
1101 char *name; 1146 char *name;
1102 /* User explicitly requested empty subsystem */ 1147 /* User explicitly requested empty subsystem */
1103 bool none; 1148 bool none;
1104 }; 1149 };
1105 1150
1106 /* 1151 /*
1107 * Convert a hierarchy specifier into a bitmask of subsystems and 1152 * Convert a hierarchy specifier into a bitmask of subsystems and
1108 * flags. Call with cgroup_mutex held to protect the cgroup_subsys[] 1153 * flags. Call with cgroup_mutex held to protect the cgroup_subsys[]
1109 * array. This function takes refcounts on subsystems to be used, unless it 1154 * array. This function takes refcounts on subsystems to be used, unless it
1110 * returns error, in which case no refcounts are taken. 1155 * returns error, in which case no refcounts are taken.
1111 */ 1156 */
1112 static int parse_cgroupfs_options(char *data, struct cgroup_sb_opts *opts) 1157 static int parse_cgroupfs_options(char *data, struct cgroup_sb_opts *opts)
1113 { 1158 {
1114 char *token, *o = data; 1159 char *token, *o = data;
1115 bool all_ss = false, one_ss = false; 1160 bool all_ss = false, one_ss = false;
1116 unsigned long mask = (unsigned long)-1; 1161 unsigned long mask = (unsigned long)-1;
1117 struct cgroup_subsys *ss; 1162 struct cgroup_subsys *ss;
1118 int i; 1163 int i;
1119 1164
1120 BUG_ON(!mutex_is_locked(&cgroup_mutex)); 1165 BUG_ON(!mutex_is_locked(&cgroup_mutex));
1121 1166
1122 #ifdef CONFIG_CPUSETS 1167 #ifdef CONFIG_CPUSETS
1123 mask = ~(1UL << cpuset_cgrp_id); 1168 mask = ~(1UL << cpuset_cgrp_id);
1124 #endif 1169 #endif
1125 1170
1126 memset(opts, 0, sizeof(*opts)); 1171 memset(opts, 0, sizeof(*opts));
1127 1172
1128 while ((token = strsep(&o, ",")) != NULL) { 1173 while ((token = strsep(&o, ",")) != NULL) {
1129 if (!*token) 1174 if (!*token)
1130 return -EINVAL; 1175 return -EINVAL;
1131 if (!strcmp(token, "none")) { 1176 if (!strcmp(token, "none")) {
1132 /* Explicitly have no subsystems */ 1177 /* Explicitly have no subsystems */
1133 opts->none = true; 1178 opts->none = true;
1134 continue; 1179 continue;
1135 } 1180 }
1136 if (!strcmp(token, "all")) { 1181 if (!strcmp(token, "all")) {
1137 /* Mutually exclusive option 'all' + subsystem name */ 1182 /* Mutually exclusive option 'all' + subsystem name */
1138 if (one_ss) 1183 if (one_ss)
1139 return -EINVAL; 1184 return -EINVAL;
1140 all_ss = true; 1185 all_ss = true;
1141 continue; 1186 continue;
1142 } 1187 }
1143 if (!strcmp(token, "__DEVEL__sane_behavior")) { 1188 if (!strcmp(token, "__DEVEL__sane_behavior")) {
1144 opts->flags |= CGRP_ROOT_SANE_BEHAVIOR; 1189 opts->flags |= CGRP_ROOT_SANE_BEHAVIOR;
1145 continue; 1190 continue;
1146 } 1191 }
1147 if (!strcmp(token, "noprefix")) { 1192 if (!strcmp(token, "noprefix")) {
1148 opts->flags |= CGRP_ROOT_NOPREFIX; 1193 opts->flags |= CGRP_ROOT_NOPREFIX;
1149 continue; 1194 continue;
1150 } 1195 }
1151 if (!strcmp(token, "clone_children")) { 1196 if (!strcmp(token, "clone_children")) {
1152 opts->cpuset_clone_children = true; 1197 opts->cpuset_clone_children = true;
1153 continue; 1198 continue;
1154 } 1199 }
1155 if (!strcmp(token, "xattr")) { 1200 if (!strcmp(token, "xattr")) {
1156 opts->flags |= CGRP_ROOT_XATTR; 1201 opts->flags |= CGRP_ROOT_XATTR;
1157 continue; 1202 continue;
1158 } 1203 }
1159 if (!strncmp(token, "release_agent=", 14)) { 1204 if (!strncmp(token, "release_agent=", 14)) {
1160 /* Specifying two release agents is forbidden */ 1205 /* Specifying two release agents is forbidden */
1161 if (opts->release_agent) 1206 if (opts->release_agent)
1162 return -EINVAL; 1207 return -EINVAL;
1163 opts->release_agent = 1208 opts->release_agent =
1164 kstrndup(token + 14, PATH_MAX - 1, GFP_KERNEL); 1209 kstrndup(token + 14, PATH_MAX - 1, GFP_KERNEL);
1165 if (!opts->release_agent) 1210 if (!opts->release_agent)
1166 return -ENOMEM; 1211 return -ENOMEM;
1167 continue; 1212 continue;
1168 } 1213 }
1169 if (!strncmp(token, "name=", 5)) { 1214 if (!strncmp(token, "name=", 5)) {
1170 const char *name = token + 5; 1215 const char *name = token + 5;
1171 /* Can't specify an empty name */ 1216 /* Can't specify an empty name */
1172 if (!strlen(name)) 1217 if (!strlen(name))
1173 return -EINVAL; 1218 return -EINVAL;
1174 /* Must match [\w.-]+ */ 1219 /* Must match [\w.-]+ */
1175 for (i = 0; i < strlen(name); i++) { 1220 for (i = 0; i < strlen(name); i++) {
1176 char c = name[i]; 1221 char c = name[i];
1177 if (isalnum(c)) 1222 if (isalnum(c))
1178 continue; 1223 continue;
1179 if ((c == '.') || (c == '-') || (c == '_')) 1224 if ((c == '.') || (c == '-') || (c == '_'))
1180 continue; 1225 continue;
1181 return -EINVAL; 1226 return -EINVAL;
1182 } 1227 }
1183 /* Specifying two names is forbidden */ 1228 /* Specifying two names is forbidden */
1184 if (opts->name) 1229 if (opts->name)
1185 return -EINVAL; 1230 return -EINVAL;
1186 opts->name = kstrndup(name, 1231 opts->name = kstrndup(name,
1187 MAX_CGROUP_ROOT_NAMELEN - 1, 1232 MAX_CGROUP_ROOT_NAMELEN - 1,
1188 GFP_KERNEL); 1233 GFP_KERNEL);
1189 if (!opts->name) 1234 if (!opts->name)
1190 return -ENOMEM; 1235 return -ENOMEM;
1191 1236
1192 continue; 1237 continue;
1193 } 1238 }
1194 1239
1195 for_each_subsys(ss, i) { 1240 for_each_subsys(ss, i) {
1196 if (strcmp(token, ss->name)) 1241 if (strcmp(token, ss->name))
1197 continue; 1242 continue;
1198 if (ss->disabled) 1243 if (ss->disabled)
1199 continue; 1244 continue;
1200 1245
1201 /* Mutually exclusive option 'all' + subsystem name */ 1246 /* Mutually exclusive option 'all' + subsystem name */
1202 if (all_ss) 1247 if (all_ss)
1203 return -EINVAL; 1248 return -EINVAL;
1204 set_bit(i, &opts->subsys_mask); 1249 set_bit(i, &opts->subsys_mask);
1205 one_ss = true; 1250 one_ss = true;
1206 1251
1207 break; 1252 break;
1208 } 1253 }
1209 if (i == CGROUP_SUBSYS_COUNT) 1254 if (i == CGROUP_SUBSYS_COUNT)
1210 return -ENOENT; 1255 return -ENOENT;
1211 } 1256 }
1212 1257
1213 /* Consistency checks */ 1258 /* Consistency checks */
1214 1259
1215 if (opts->flags & CGRP_ROOT_SANE_BEHAVIOR) { 1260 if (opts->flags & CGRP_ROOT_SANE_BEHAVIOR) {
1216 pr_warning("cgroup: sane_behavior: this is still under development and its behaviors will change, proceed at your own risk\n"); 1261 pr_warning("cgroup: sane_behavior: this is still under development and its behaviors will change, proceed at your own risk\n");
1217 1262
1218 if ((opts->flags & (CGRP_ROOT_NOPREFIX | CGRP_ROOT_XATTR)) || 1263 if ((opts->flags & (CGRP_ROOT_NOPREFIX | CGRP_ROOT_XATTR)) ||
1219 opts->cpuset_clone_children || opts->release_agent || 1264 opts->cpuset_clone_children || opts->release_agent ||
1220 opts->name) { 1265 opts->name) {
1221 pr_err("cgroup: sane_behavior: noprefix, xattr, clone_children, release_agent and name are not allowed\n"); 1266 pr_err("cgroup: sane_behavior: noprefix, xattr, clone_children, release_agent and name are not allowed\n");
1222 return -EINVAL; 1267 return -EINVAL;
1223 } 1268 }
1224 } else { 1269 } else {
1225 /* 1270 /*
1226 * If the 'all' option was specified select all the 1271 * If the 'all' option was specified select all the
1227 * subsystems, otherwise if 'none', 'name=' and a subsystem 1272 * subsystems, otherwise if 'none', 'name=' and a subsystem
1228 * name options were not specified, let's default to 'all' 1273 * name options were not specified, let's default to 'all'
1229 */ 1274 */
1230 if (all_ss || (!one_ss && !opts->none && !opts->name)) 1275 if (all_ss || (!one_ss && !opts->none && !opts->name))
1231 for_each_subsys(ss, i) 1276 for_each_subsys(ss, i)
1232 if (!ss->disabled) 1277 if (!ss->disabled)
1233 set_bit(i, &opts->subsys_mask); 1278 set_bit(i, &opts->subsys_mask);
1234 1279
1235 /* 1280 /*
1236 * We either have to specify by name or by subsystems. (So 1281 * We either have to specify by name or by subsystems. (So
1237 * all empty hierarchies must have a name). 1282 * all empty hierarchies must have a name).
1238 */ 1283 */
1239 if (!opts->subsys_mask && !opts->name) 1284 if (!opts->subsys_mask && !opts->name)
1240 return -EINVAL; 1285 return -EINVAL;
1241 } 1286 }
1242 1287
1243 /* 1288 /*
1244 * Option noprefix was introduced just for backward compatibility 1289 * Option noprefix was introduced just for backward compatibility
1245 * with the old cpuset, so we allow noprefix only if mounting just 1290 * with the old cpuset, so we allow noprefix only if mounting just
1246 * the cpuset subsystem. 1291 * the cpuset subsystem.
1247 */ 1292 */
1248 if ((opts->flags & CGRP_ROOT_NOPREFIX) && (opts->subsys_mask & mask)) 1293 if ((opts->flags & CGRP_ROOT_NOPREFIX) && (opts->subsys_mask & mask))
1249 return -EINVAL; 1294 return -EINVAL;
1250 1295
1251 1296
1252 /* Can't specify "none" and some subsystems */ 1297 /* Can't specify "none" and some subsystems */
1253 if (opts->subsys_mask && opts->none) 1298 if (opts->subsys_mask && opts->none)
1254 return -EINVAL; 1299 return -EINVAL;
1255 1300
1256 return 0; 1301 return 0;
1257 } 1302 }
1258 1303
1259 static int cgroup_remount(struct kernfs_root *kf_root, int *flags, char *data) 1304 static int cgroup_remount(struct kernfs_root *kf_root, int *flags, char *data)
1260 { 1305 {
1261 int ret = 0; 1306 int ret = 0;
1262 struct cgroup_root *root = cgroup_root_from_kf(kf_root); 1307 struct cgroup_root *root = cgroup_root_from_kf(kf_root);
1263 struct cgroup_sb_opts opts; 1308 struct cgroup_sb_opts opts;
1264 unsigned long added_mask, removed_mask; 1309 unsigned long added_mask, removed_mask;
1265 1310
1266 if (root->flags & CGRP_ROOT_SANE_BEHAVIOR) { 1311 if (root->flags & CGRP_ROOT_SANE_BEHAVIOR) {
1267 pr_err("cgroup: sane_behavior: remount is not allowed\n"); 1312 pr_err("cgroup: sane_behavior: remount is not allowed\n");
1268 return -EINVAL; 1313 return -EINVAL;
1269 } 1314 }
1270 1315
1271 mutex_lock(&cgroup_tree_mutex); 1316 mutex_lock(&cgroup_tree_mutex);
1272 mutex_lock(&cgroup_mutex); 1317 mutex_lock(&cgroup_mutex);
1273 1318
1274 /* See what subsystems are wanted */ 1319 /* See what subsystems are wanted */
1275 ret = parse_cgroupfs_options(data, &opts); 1320 ret = parse_cgroupfs_options(data, &opts);
1276 if (ret) 1321 if (ret)
1277 goto out_unlock; 1322 goto out_unlock;
1278 1323
1279 if (opts.subsys_mask != root->subsys_mask || opts.release_agent) 1324 if (opts.subsys_mask != root->subsys_mask || opts.release_agent)
1280 pr_warning("cgroup: option changes via remount are deprecated (pid=%d comm=%s)\n", 1325 pr_warning("cgroup: option changes via remount are deprecated (pid=%d comm=%s)\n",
1281 task_tgid_nr(current), current->comm); 1326 task_tgid_nr(current), current->comm);
1282 1327
1283 added_mask = opts.subsys_mask & ~root->subsys_mask; 1328 added_mask = opts.subsys_mask & ~root->subsys_mask;
1284 removed_mask = root->subsys_mask & ~opts.subsys_mask; 1329 removed_mask = root->subsys_mask & ~opts.subsys_mask;
1285 1330
1286 /* Don't allow flags or name to change at remount */ 1331 /* Don't allow flags or name to change at remount */
1287 if (((opts.flags ^ root->flags) & CGRP_ROOT_OPTION_MASK) || 1332 if (((opts.flags ^ root->flags) & CGRP_ROOT_OPTION_MASK) ||
1288 (opts.name && strcmp(opts.name, root->name))) { 1333 (opts.name && strcmp(opts.name, root->name))) {
1289 pr_err("cgroup: option or name mismatch, new: 0x%lx \"%s\", old: 0x%lx \"%s\"\n", 1334 pr_err("cgroup: option or name mismatch, new: 0x%lx \"%s\", old: 0x%lx \"%s\"\n",
1290 opts.flags & CGRP_ROOT_OPTION_MASK, opts.name ?: "", 1335 opts.flags & CGRP_ROOT_OPTION_MASK, opts.name ?: "",
1291 root->flags & CGRP_ROOT_OPTION_MASK, root->name); 1336 root->flags & CGRP_ROOT_OPTION_MASK, root->name);
1292 ret = -EINVAL; 1337 ret = -EINVAL;
1293 goto out_unlock; 1338 goto out_unlock;
1294 } 1339 }
1295 1340
1296 /* remounting is not allowed for populated hierarchies */ 1341 /* remounting is not allowed for populated hierarchies */
1297 if (!list_empty(&root->cgrp.children)) { 1342 if (!list_empty(&root->cgrp.children)) {
1298 ret = -EBUSY; 1343 ret = -EBUSY;
1299 goto out_unlock; 1344 goto out_unlock;
1300 } 1345 }
1301 1346
1302 ret = rebind_subsystems(root, added_mask); 1347 ret = rebind_subsystems(root, added_mask);
1303 if (ret) 1348 if (ret)
1304 goto out_unlock; 1349 goto out_unlock;
1305 1350
1306 rebind_subsystems(&cgrp_dfl_root, removed_mask); 1351 rebind_subsystems(&cgrp_dfl_root, removed_mask);
1307 1352
1308 if (opts.release_agent) { 1353 if (opts.release_agent) {
1309 spin_lock(&release_agent_path_lock); 1354 spin_lock(&release_agent_path_lock);
1310 strcpy(root->release_agent_path, opts.release_agent); 1355 strcpy(root->release_agent_path, opts.release_agent);
1311 spin_unlock(&release_agent_path_lock); 1356 spin_unlock(&release_agent_path_lock);
1312 } 1357 }
1313 out_unlock: 1358 out_unlock:
1314 kfree(opts.release_agent); 1359 kfree(opts.release_agent);
1315 kfree(opts.name); 1360 kfree(opts.name);
1316 mutex_unlock(&cgroup_mutex); 1361 mutex_unlock(&cgroup_mutex);
1317 mutex_unlock(&cgroup_tree_mutex); 1362 mutex_unlock(&cgroup_tree_mutex);
1318 return ret; 1363 return ret;
1319 } 1364 }
1320 1365
1321 /* 1366 /*
1322 * To reduce the fork() overhead for systems that are not actually using 1367 * To reduce the fork() overhead for systems that are not actually using
1323 * their cgroups capability, we don't maintain the lists running through 1368 * their cgroups capability, we don't maintain the lists running through
1324 * each css_set to its tasks until we see the list actually used - in other 1369 * each css_set to its tasks until we see the list actually used - in other
1325 * words after the first mount. 1370 * words after the first mount.
1326 */ 1371 */
1327 static bool use_task_css_set_links __read_mostly; 1372 static bool use_task_css_set_links __read_mostly;
1328 1373
1329 static void cgroup_enable_task_cg_lists(void) 1374 static void cgroup_enable_task_cg_lists(void)
1330 { 1375 {
1331 struct task_struct *p, *g; 1376 struct task_struct *p, *g;
1332 1377
1333 down_write(&css_set_rwsem); 1378 down_write(&css_set_rwsem);
1334 1379
1335 if (use_task_css_set_links) 1380 if (use_task_css_set_links)
1336 goto out_unlock; 1381 goto out_unlock;
1337 1382
1338 use_task_css_set_links = true; 1383 use_task_css_set_links = true;
1339 1384
1340 /* 1385 /*
1341 * We need tasklist_lock because RCU is not safe against 1386 * We need tasklist_lock because RCU is not safe against
1342 * while_each_thread(). Besides, a forking task that has passed 1387 * while_each_thread(). Besides, a forking task that has passed
1343 * cgroup_post_fork() without seeing use_task_css_set_links = 1 1388 * cgroup_post_fork() without seeing use_task_css_set_links = 1
1344 * is not guaranteed to have its child immediately visible in the 1389 * is not guaranteed to have its child immediately visible in the
1345 * tasklist if we walk through it with RCU. 1390 * tasklist if we walk through it with RCU.
1346 */ 1391 */
1347 read_lock(&tasklist_lock); 1392 read_lock(&tasklist_lock);
1348 do_each_thread(g, p) { 1393 do_each_thread(g, p) {
1349 WARN_ON_ONCE(!list_empty(&p->cg_list) || 1394 WARN_ON_ONCE(!list_empty(&p->cg_list) ||
1350 task_css_set(p) != &init_css_set); 1395 task_css_set(p) != &init_css_set);
1351 1396
1352 /* 1397 /*
1353 * We should check if the process is exiting, otherwise 1398 * We should check if the process is exiting, otherwise
1354 * it will race with cgroup_exit() in that the list 1399 * it will race with cgroup_exit() in that the list
1355 * entry won't be deleted though the process has exited. 1400 * entry won't be deleted though the process has exited.
1356 * Do it while holding siglock so that we don't end up 1401 * Do it while holding siglock so that we don't end up
1357 * racing against cgroup_exit(). 1402 * racing against cgroup_exit().
1358 */ 1403 */
1359 spin_lock_irq(&p->sighand->siglock); 1404 spin_lock_irq(&p->sighand->siglock);
1360 if (!(p->flags & PF_EXITING)) { 1405 if (!(p->flags & PF_EXITING)) {
1361 struct css_set *cset = task_css_set(p); 1406 struct css_set *cset = task_css_set(p);
1362 1407
1363 list_add(&p->cg_list, &cset->tasks); 1408 list_add(&p->cg_list, &cset->tasks);
1364 get_css_set(cset); 1409 get_css_set(cset);
1365 } 1410 }
1366 spin_unlock_irq(&p->sighand->siglock); 1411 spin_unlock_irq(&p->sighand->siglock);
1367 } while_each_thread(g, p); 1412 } while_each_thread(g, p);
1368 read_unlock(&tasklist_lock); 1413 read_unlock(&tasklist_lock);
1369 out_unlock: 1414 out_unlock:
1370 up_write(&css_set_rwsem); 1415 up_write(&css_set_rwsem);
1371 } 1416 }
1372 1417
1373 static void init_cgroup_housekeeping(struct cgroup *cgrp) 1418 static void init_cgroup_housekeeping(struct cgroup *cgrp)
1374 { 1419 {
1375 atomic_set(&cgrp->refcnt, 1); 1420 atomic_set(&cgrp->refcnt, 1);
1376 INIT_LIST_HEAD(&cgrp->sibling); 1421 INIT_LIST_HEAD(&cgrp->sibling);
1377 INIT_LIST_HEAD(&cgrp->children); 1422 INIT_LIST_HEAD(&cgrp->children);
1378 INIT_LIST_HEAD(&cgrp->cset_links); 1423 INIT_LIST_HEAD(&cgrp->cset_links);
1379 INIT_LIST_HEAD(&cgrp->release_list); 1424 INIT_LIST_HEAD(&cgrp->release_list);
1380 INIT_LIST_HEAD(&cgrp->pidlists); 1425 INIT_LIST_HEAD(&cgrp->pidlists);
1381 mutex_init(&cgrp->pidlist_mutex); 1426 mutex_init(&cgrp->pidlist_mutex);
1382 cgrp->dummy_css.cgroup = cgrp; 1427 cgrp->dummy_css.cgroup = cgrp;
1383 } 1428 }
1384 1429
1385 static void init_cgroup_root(struct cgroup_root *root, 1430 static void init_cgroup_root(struct cgroup_root *root,
1386 struct cgroup_sb_opts *opts) 1431 struct cgroup_sb_opts *opts)
1387 { 1432 {
1388 struct cgroup *cgrp = &root->cgrp; 1433 struct cgroup *cgrp = &root->cgrp;
1389 1434
1390 INIT_LIST_HEAD(&root->root_list); 1435 INIT_LIST_HEAD(&root->root_list);
1391 atomic_set(&root->nr_cgrps, 1); 1436 atomic_set(&root->nr_cgrps, 1);
1392 cgrp->root = root; 1437 cgrp->root = root;
1393 init_cgroup_housekeeping(cgrp); 1438 init_cgroup_housekeeping(cgrp);
1394 idr_init(&root->cgroup_idr); 1439 idr_init(&root->cgroup_idr);
1395 1440
1396 root->flags = opts->flags; 1441 root->flags = opts->flags;
1397 if (opts->release_agent) 1442 if (opts->release_agent)
1398 strcpy(root->release_agent_path, opts->release_agent); 1443 strcpy(root->release_agent_path, opts->release_agent);
1399 if (opts->name) 1444 if (opts->name)
1400 strcpy(root->name, opts->name); 1445 strcpy(root->name, opts->name);
1401 if (opts->cpuset_clone_children) 1446 if (opts->cpuset_clone_children)
1402 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags); 1447 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags);
1403 } 1448 }
1404 1449
1405 static int cgroup_setup_root(struct cgroup_root *root, unsigned long ss_mask) 1450 static int cgroup_setup_root(struct cgroup_root *root, unsigned long ss_mask)
1406 { 1451 {
1407 LIST_HEAD(tmp_links); 1452 LIST_HEAD(tmp_links);
1408 struct cgroup *root_cgrp = &root->cgrp; 1453 struct cgroup *root_cgrp = &root->cgrp;
1409 struct css_set *cset; 1454 struct css_set *cset;
1410 int i, ret; 1455 int i, ret;
1411 1456
1412 lockdep_assert_held(&cgroup_tree_mutex); 1457 lockdep_assert_held(&cgroup_tree_mutex);
1413 lockdep_assert_held(&cgroup_mutex); 1458 lockdep_assert_held(&cgroup_mutex);
1414 1459
1415 ret = idr_alloc(&root->cgroup_idr, root_cgrp, 0, 1, GFP_KERNEL); 1460 ret = idr_alloc(&root->cgroup_idr, root_cgrp, 0, 1, GFP_KERNEL);
1416 if (ret < 0) 1461 if (ret < 0)
1417 goto out; 1462 goto out;
1418 root_cgrp->id = ret; 1463 root_cgrp->id = ret;
1419 1464
1420 /* 1465 /*
1421 * We're accessing css_set_count without locking css_set_rwsem here, 1466 * We're accessing css_set_count without locking css_set_rwsem here,
1422 * but that's OK - it can only be increased by someone holding 1467 * but that's OK - it can only be increased by someone holding
1423 * cgroup_lock, and that's us. The worst that can happen is that we 1468 * cgroup_lock, and that's us. The worst that can happen is that we
1424 * have some link structures left over 1469 * have some link structures left over
1425 */ 1470 */
1426 ret = allocate_cgrp_cset_links(css_set_count, &tmp_links); 1471 ret = allocate_cgrp_cset_links(css_set_count, &tmp_links);
1427 if (ret) 1472 if (ret)
1428 goto out; 1473 goto out;
1429 1474
1430 ret = cgroup_init_root_id(root); 1475 ret = cgroup_init_root_id(root);
1431 if (ret) 1476 if (ret)
1432 goto out; 1477 goto out;
1433 1478
1434 root->kf_root = kernfs_create_root(&cgroup_kf_syscall_ops, 1479 root->kf_root = kernfs_create_root(&cgroup_kf_syscall_ops,
1435 KERNFS_ROOT_CREATE_DEACTIVATED, 1480 KERNFS_ROOT_CREATE_DEACTIVATED,
1436 root_cgrp); 1481 root_cgrp);
1437 if (IS_ERR(root->kf_root)) { 1482 if (IS_ERR(root->kf_root)) {
1438 ret = PTR_ERR(root->kf_root); 1483 ret = PTR_ERR(root->kf_root);
1439 goto exit_root_id; 1484 goto exit_root_id;
1440 } 1485 }
1441 root_cgrp->kn = root->kf_root->kn; 1486 root_cgrp->kn = root->kf_root->kn;
1442 1487
1443 ret = cgroup_addrm_files(root_cgrp, cgroup_base_files, true); 1488 ret = cgroup_addrm_files(root_cgrp, cgroup_base_files, true);
1444 if (ret) 1489 if (ret)
1445 goto destroy_root; 1490 goto destroy_root;
1446 1491
1447 ret = rebind_subsystems(root, ss_mask); 1492 ret = rebind_subsystems(root, ss_mask);
1448 if (ret) 1493 if (ret)
1449 goto destroy_root; 1494 goto destroy_root;
1450 1495
1451 /* 1496 /*
1452 * There must be no failure case after here, since rebinding takes 1497 * There must be no failure case after here, since rebinding takes
1453 * care of subsystems' refcounts, which are explicitly dropped in 1498 * care of subsystems' refcounts, which are explicitly dropped in
1454 * the failure exit path. 1499 * the failure exit path.
1455 */ 1500 */
1456 list_add(&root->root_list, &cgroup_roots); 1501 list_add(&root->root_list, &cgroup_roots);
1457 cgroup_root_count++; 1502 cgroup_root_count++;
1458 1503
1459 /* 1504 /*
1460 * Link the root cgroup in this hierarchy into all the css_set 1505 * Link the root cgroup in this hierarchy into all the css_set
1461 * objects. 1506 * objects.
1462 */ 1507 */
1463 down_write(&css_set_rwsem); 1508 down_write(&css_set_rwsem);
1464 hash_for_each(css_set_table, i, cset, hlist) 1509 hash_for_each(css_set_table, i, cset, hlist)
1465 link_css_set(&tmp_links, cset, root_cgrp); 1510 link_css_set(&tmp_links, cset, root_cgrp);
1466 up_write(&css_set_rwsem); 1511 up_write(&css_set_rwsem);
1467 1512
1468 BUG_ON(!list_empty(&root_cgrp->children)); 1513 BUG_ON(!list_empty(&root_cgrp->children));
1469 BUG_ON(atomic_read(&root->nr_cgrps) != 1); 1514 BUG_ON(atomic_read(&root->nr_cgrps) != 1);
1470 1515
1471 kernfs_activate(root_cgrp->kn); 1516 kernfs_activate(root_cgrp->kn);
1472 ret = 0; 1517 ret = 0;
1473 goto out; 1518 goto out;
1474 1519
1475 destroy_root: 1520 destroy_root:
1476 kernfs_destroy_root(root->kf_root); 1521 kernfs_destroy_root(root->kf_root);
1477 root->kf_root = NULL; 1522 root->kf_root = NULL;
1478 exit_root_id: 1523 exit_root_id:
1479 cgroup_exit_root_id(root); 1524 cgroup_exit_root_id(root);
1480 out: 1525 out:
1481 free_cgrp_cset_links(&tmp_links); 1526 free_cgrp_cset_links(&tmp_links);
1482 return ret; 1527 return ret;
1483 } 1528 }
1484 1529
1485 static struct dentry *cgroup_mount(struct file_system_type *fs_type, 1530 static struct dentry *cgroup_mount(struct file_system_type *fs_type,
1486 int flags, const char *unused_dev_name, 1531 int flags, const char *unused_dev_name,
1487 void *data) 1532 void *data)
1488 { 1533 {
1489 struct cgroup_root *root; 1534 struct cgroup_root *root;
1490 struct cgroup_sb_opts opts; 1535 struct cgroup_sb_opts opts;
1491 struct dentry *dentry; 1536 struct dentry *dentry;
1492 int ret; 1537 int ret;
1493 bool new_sb; 1538 bool new_sb;
1494 1539
1495 /* 1540 /*
1496 * The first time anyone tries to mount a cgroup, enable the list 1541 * The first time anyone tries to mount a cgroup, enable the list
1497 * linking each css_set to its tasks and fix up all existing tasks. 1542 * linking each css_set to its tasks and fix up all existing tasks.
1498 */ 1543 */
1499 if (!use_task_css_set_links) 1544 if (!use_task_css_set_links)
1500 cgroup_enable_task_cg_lists(); 1545 cgroup_enable_task_cg_lists();
1501 1546
1502 mutex_lock(&cgroup_tree_mutex); 1547 mutex_lock(&cgroup_tree_mutex);
1503 mutex_lock(&cgroup_mutex); 1548 mutex_lock(&cgroup_mutex);
1504 1549
1505 /* First find the desired set of subsystems */ 1550 /* First find the desired set of subsystems */
1506 ret = parse_cgroupfs_options(data, &opts); 1551 ret = parse_cgroupfs_options(data, &opts);
1507 if (ret) 1552 if (ret)
1508 goto out_unlock; 1553 goto out_unlock;
1509 retry: 1554 retry:
1510 /* look for a matching existing root */ 1555 /* look for a matching existing root */
1511 if (!opts.subsys_mask && !opts.none && !opts.name) { 1556 if (!opts.subsys_mask && !opts.none && !opts.name) {
1512 cgrp_dfl_root_visible = true; 1557 cgrp_dfl_root_visible = true;
1513 root = &cgrp_dfl_root; 1558 root = &cgrp_dfl_root;
1514 cgroup_get(&root->cgrp); 1559 cgroup_get(&root->cgrp);
1515 ret = 0; 1560 ret = 0;
1516 goto out_unlock; 1561 goto out_unlock;
1517 } 1562 }
1518 1563
1519 for_each_root(root) { 1564 for_each_root(root) {
1520 bool name_match = false; 1565 bool name_match = false;
1521 1566
1522 if (root == &cgrp_dfl_root) 1567 if (root == &cgrp_dfl_root)
1523 continue; 1568 continue;
1524 1569
1525 /* 1570 /*
1526 * If we asked for a name then it must match. Also, if 1571 * If we asked for a name then it must match. Also, if
1527 * name matches but sybsys_mask doesn't, we should fail. 1572 * name matches but sybsys_mask doesn't, we should fail.
1528 * Remember whether name matched. 1573 * Remember whether name matched.
1529 */ 1574 */
1530 if (opts.name) { 1575 if (opts.name) {
1531 if (strcmp(opts.name, root->name)) 1576 if (strcmp(opts.name, root->name))
1532 continue; 1577 continue;
1533 name_match = true; 1578 name_match = true;
1534 } 1579 }
1535 1580
1536 /* 1581 /*
1537 * If we asked for subsystems (or explicitly for no 1582 * If we asked for subsystems (or explicitly for no
1538 * subsystems) then they must match. 1583 * subsystems) then they must match.
1539 */ 1584 */
1540 if ((opts.subsys_mask || opts.none) && 1585 if ((opts.subsys_mask || opts.none) &&
1541 (opts.subsys_mask != root->subsys_mask)) { 1586 (opts.subsys_mask != root->subsys_mask)) {
1542 if (!name_match) 1587 if (!name_match)
1543 continue; 1588 continue;
1544 ret = -EBUSY; 1589 ret = -EBUSY;
1545 goto out_unlock; 1590 goto out_unlock;
1546 } 1591 }
1547 1592
1548 if ((root->flags ^ opts.flags) & CGRP_ROOT_OPTION_MASK) { 1593 if ((root->flags ^ opts.flags) & CGRP_ROOT_OPTION_MASK) {
1549 if ((root->flags | opts.flags) & CGRP_ROOT_SANE_BEHAVIOR) { 1594 if ((root->flags | opts.flags) & CGRP_ROOT_SANE_BEHAVIOR) {
1550 pr_err("cgroup: sane_behavior: new mount options should match the existing superblock\n"); 1595 pr_err("cgroup: sane_behavior: new mount options should match the existing superblock\n");
1551 ret = -EINVAL; 1596 ret = -EINVAL;
1552 goto out_unlock; 1597 goto out_unlock;
1553 } else { 1598 } else {
1554 pr_warning("cgroup: new mount options do not match the existing superblock, will be ignored\n"); 1599 pr_warning("cgroup: new mount options do not match the existing superblock, will be ignored\n");
1555 } 1600 }
1556 } 1601 }
1557 1602
1558 /* 1603 /*
1559 * A root's lifetime is governed by its root cgroup. Zero 1604 * A root's lifetime is governed by its root cgroup. Zero
1560 * ref indicate that the root is being destroyed. Wait for 1605 * ref indicate that the root is being destroyed. Wait for
1561 * destruction to complete so that the subsystems are free. 1606 * destruction to complete so that the subsystems are free.
1562 * We can use wait_queue for the wait but this path is 1607 * We can use wait_queue for the wait but this path is
1563 * super cold. Let's just sleep for a bit and retry. 1608 * super cold. Let's just sleep for a bit and retry.
1564 */ 1609 */
1565 if (!atomic_inc_not_zero(&root->cgrp.refcnt)) { 1610 if (!atomic_inc_not_zero(&root->cgrp.refcnt)) {
1566 mutex_unlock(&cgroup_mutex); 1611 mutex_unlock(&cgroup_mutex);
1567 mutex_unlock(&cgroup_tree_mutex); 1612 mutex_unlock(&cgroup_tree_mutex);
1568 msleep(10); 1613 msleep(10);
1569 mutex_lock(&cgroup_tree_mutex); 1614 mutex_lock(&cgroup_tree_mutex);
1570 mutex_lock(&cgroup_mutex); 1615 mutex_lock(&cgroup_mutex);
1571 goto retry; 1616 goto retry;
1572 } 1617 }
1573 1618
1574 ret = 0; 1619 ret = 0;
1575 goto out_unlock; 1620 goto out_unlock;
1576 } 1621 }
1577 1622
1578 /* 1623 /*
1579 * No such thing, create a new one. name= matching without subsys 1624 * No such thing, create a new one. name= matching without subsys
1580 * specification is allowed for already existing hierarchies but we 1625 * specification is allowed for already existing hierarchies but we
1581 * can't create new one without subsys specification. 1626 * can't create new one without subsys specification.
1582 */ 1627 */
1583 if (!opts.subsys_mask && !opts.none) { 1628 if (!opts.subsys_mask && !opts.none) {
1584 ret = -EINVAL; 1629 ret = -EINVAL;
1585 goto out_unlock; 1630 goto out_unlock;
1586 } 1631 }
1587 1632
1588 root = kzalloc(sizeof(*root), GFP_KERNEL); 1633 root = kzalloc(sizeof(*root), GFP_KERNEL);
1589 if (!root) { 1634 if (!root) {
1590 ret = -ENOMEM; 1635 ret = -ENOMEM;
1591 goto out_unlock; 1636 goto out_unlock;
1592 } 1637 }
1593 1638
1594 init_cgroup_root(root, &opts); 1639 init_cgroup_root(root, &opts);
1595 1640
1596 ret = cgroup_setup_root(root, opts.subsys_mask); 1641 ret = cgroup_setup_root(root, opts.subsys_mask);
1597 if (ret) 1642 if (ret)
1598 cgroup_free_root(root); 1643 cgroup_free_root(root);
1599 1644
1600 out_unlock: 1645 out_unlock:
1601 mutex_unlock(&cgroup_mutex); 1646 mutex_unlock(&cgroup_mutex);
1602 mutex_unlock(&cgroup_tree_mutex); 1647 mutex_unlock(&cgroup_tree_mutex);
1603 1648
1604 kfree(opts.release_agent); 1649 kfree(opts.release_agent);
1605 kfree(opts.name); 1650 kfree(opts.name);
1606 1651
1607 if (ret) 1652 if (ret)
1608 return ERR_PTR(ret); 1653 return ERR_PTR(ret);
1609 1654
1610 dentry = kernfs_mount(fs_type, flags, root->kf_root, &new_sb); 1655 dentry = kernfs_mount(fs_type, flags, root->kf_root, &new_sb);
1611 if (IS_ERR(dentry) || !new_sb) 1656 if (IS_ERR(dentry) || !new_sb)
1612 cgroup_put(&root->cgrp); 1657 cgroup_put(&root->cgrp);
1613 return dentry; 1658 return dentry;
1614 } 1659 }
1615 1660
1616 static void cgroup_kill_sb(struct super_block *sb) 1661 static void cgroup_kill_sb(struct super_block *sb)
1617 { 1662 {
1618 struct kernfs_root *kf_root = kernfs_root_from_sb(sb); 1663 struct kernfs_root *kf_root = kernfs_root_from_sb(sb);
1619 struct cgroup_root *root = cgroup_root_from_kf(kf_root); 1664 struct cgroup_root *root = cgroup_root_from_kf(kf_root);
1620 1665
1621 cgroup_put(&root->cgrp); 1666 cgroup_put(&root->cgrp);
1622 kernfs_kill_sb(sb); 1667 kernfs_kill_sb(sb);
1623 } 1668 }
1624 1669
1625 static struct file_system_type cgroup_fs_type = { 1670 static struct file_system_type cgroup_fs_type = {
1626 .name = "cgroup", 1671 .name = "cgroup",
1627 .mount = cgroup_mount, 1672 .mount = cgroup_mount,
1628 .kill_sb = cgroup_kill_sb, 1673 .kill_sb = cgroup_kill_sb,
1629 }; 1674 };
1630 1675
1631 static struct kobject *cgroup_kobj; 1676 static struct kobject *cgroup_kobj;
1632 1677
1633 /** 1678 /**
1634 * task_cgroup_path - cgroup path of a task in the first cgroup hierarchy 1679 * task_cgroup_path - cgroup path of a task in the first cgroup hierarchy
1635 * @task: target task 1680 * @task: target task
1636 * @buf: the buffer to write the path into 1681 * @buf: the buffer to write the path into
1637 * @buflen: the length of the buffer 1682 * @buflen: the length of the buffer
1638 * 1683 *
1639 * Determine @task's cgroup on the first (the one with the lowest non-zero 1684 * Determine @task's cgroup on the first (the one with the lowest non-zero
1640 * hierarchy_id) cgroup hierarchy and copy its path into @buf. This 1685 * hierarchy_id) cgroup hierarchy and copy its path into @buf. This
1641 * function grabs cgroup_mutex and shouldn't be used inside locks used by 1686 * function grabs cgroup_mutex and shouldn't be used inside locks used by
1642 * cgroup controller callbacks. 1687 * cgroup controller callbacks.
1643 * 1688 *
1644 * Return value is the same as kernfs_path(). 1689 * Return value is the same as kernfs_path().
1645 */ 1690 */
1646 char *task_cgroup_path(struct task_struct *task, char *buf, size_t buflen) 1691 char *task_cgroup_path(struct task_struct *task, char *buf, size_t buflen)
1647 { 1692 {
1648 struct cgroup_root *root; 1693 struct cgroup_root *root;
1649 struct cgroup *cgrp; 1694 struct cgroup *cgrp;
1650 int hierarchy_id = 1; 1695 int hierarchy_id = 1;
1651 char *path = NULL; 1696 char *path = NULL;
1652 1697
1653 mutex_lock(&cgroup_mutex); 1698 mutex_lock(&cgroup_mutex);
1654 down_read(&css_set_rwsem); 1699 down_read(&css_set_rwsem);
1655 1700
1656 root = idr_get_next(&cgroup_hierarchy_idr, &hierarchy_id); 1701 root = idr_get_next(&cgroup_hierarchy_idr, &hierarchy_id);
1657 1702
1658 if (root) { 1703 if (root) {
1659 cgrp = task_cgroup_from_root(task, root); 1704 cgrp = task_cgroup_from_root(task, root);
1660 path = cgroup_path(cgrp, buf, buflen); 1705 path = cgroup_path(cgrp, buf, buflen);
1661 } else { 1706 } else {
1662 /* if no hierarchy exists, everyone is in "/" */ 1707 /* if no hierarchy exists, everyone is in "/" */
1663 if (strlcpy(buf, "/", buflen) < buflen) 1708 if (strlcpy(buf, "/", buflen) < buflen)
1664 path = buf; 1709 path = buf;
1665 } 1710 }
1666 1711
1667 up_read(&css_set_rwsem); 1712 up_read(&css_set_rwsem);
1668 mutex_unlock(&cgroup_mutex); 1713 mutex_unlock(&cgroup_mutex);
1669 return path; 1714 return path;
1670 } 1715 }
1671 EXPORT_SYMBOL_GPL(task_cgroup_path); 1716 EXPORT_SYMBOL_GPL(task_cgroup_path);
1672 1717
1673 /* used to track tasks and other necessary states during migration */ 1718 /* used to track tasks and other necessary states during migration */
1674 struct cgroup_taskset { 1719 struct cgroup_taskset {
1675 /* the src and dst cset list running through cset->mg_node */ 1720 /* the src and dst cset list running through cset->mg_node */
1676 struct list_head src_csets; 1721 struct list_head src_csets;
1677 struct list_head dst_csets; 1722 struct list_head dst_csets;
1678 1723
1679 /* 1724 /*
1680 * Fields for cgroup_taskset_*() iteration. 1725 * Fields for cgroup_taskset_*() iteration.
1681 * 1726 *
1682 * Before migration is committed, the target migration tasks are on 1727 * Before migration is committed, the target migration tasks are on
1683 * ->mg_tasks of the csets on ->src_csets. After, on ->mg_tasks of 1728 * ->mg_tasks of the csets on ->src_csets. After, on ->mg_tasks of
1684 * the csets on ->dst_csets. ->csets point to either ->src_csets 1729 * the csets on ->dst_csets. ->csets point to either ->src_csets
1685 * or ->dst_csets depending on whether migration is committed. 1730 * or ->dst_csets depending on whether migration is committed.
1686 * 1731 *
1687 * ->cur_csets and ->cur_task point to the current task position 1732 * ->cur_csets and ->cur_task point to the current task position
1688 * during iteration. 1733 * during iteration.
1689 */ 1734 */
1690 struct list_head *csets; 1735 struct list_head *csets;
1691 struct css_set *cur_cset; 1736 struct css_set *cur_cset;
1692 struct task_struct *cur_task; 1737 struct task_struct *cur_task;
1693 }; 1738 };
1694 1739
1695 /** 1740 /**
1696 * cgroup_taskset_first - reset taskset and return the first task 1741 * cgroup_taskset_first - reset taskset and return the first task
1697 * @tset: taskset of interest 1742 * @tset: taskset of interest
1698 * 1743 *
1699 * @tset iteration is initialized and the first task is returned. 1744 * @tset iteration is initialized and the first task is returned.
1700 */ 1745 */
1701 struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset) 1746 struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset)
1702 { 1747 {
1703 tset->cur_cset = list_first_entry(tset->csets, struct css_set, mg_node); 1748 tset->cur_cset = list_first_entry(tset->csets, struct css_set, mg_node);
1704 tset->cur_task = NULL; 1749 tset->cur_task = NULL;
1705 1750
1706 return cgroup_taskset_next(tset); 1751 return cgroup_taskset_next(tset);
1707 } 1752 }
1708 1753
1709 /** 1754 /**
1710 * cgroup_taskset_next - iterate to the next task in taskset 1755 * cgroup_taskset_next - iterate to the next task in taskset
1711 * @tset: taskset of interest 1756 * @tset: taskset of interest
1712 * 1757 *
1713 * Return the next task in @tset. Iteration must have been initialized 1758 * Return the next task in @tset. Iteration must have been initialized
1714 * with cgroup_taskset_first(). 1759 * with cgroup_taskset_first().
1715 */ 1760 */
1716 struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset) 1761 struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset)
1717 { 1762 {
1718 struct css_set *cset = tset->cur_cset; 1763 struct css_set *cset = tset->cur_cset;
1719 struct task_struct *task = tset->cur_task; 1764 struct task_struct *task = tset->cur_task;
1720 1765
1721 while (&cset->mg_node != tset->csets) { 1766 while (&cset->mg_node != tset->csets) {
1722 if (!task) 1767 if (!task)
1723 task = list_first_entry(&cset->mg_tasks, 1768 task = list_first_entry(&cset->mg_tasks,
1724 struct task_struct, cg_list); 1769 struct task_struct, cg_list);
1725 else 1770 else
1726 task = list_next_entry(task, cg_list); 1771 task = list_next_entry(task, cg_list);
1727 1772
1728 if (&task->cg_list != &cset->mg_tasks) { 1773 if (&task->cg_list != &cset->mg_tasks) {
1729 tset->cur_cset = cset; 1774 tset->cur_cset = cset;
1730 tset->cur_task = task; 1775 tset->cur_task = task;
1731 return task; 1776 return task;
1732 } 1777 }
1733 1778
1734 cset = list_next_entry(cset, mg_node); 1779 cset = list_next_entry(cset, mg_node);
1735 task = NULL; 1780 task = NULL;
1736 } 1781 }
1737 1782
1738 return NULL; 1783 return NULL;
1739 } 1784 }
1740 1785
1741 /** 1786 /**
1742 * cgroup_task_migrate - move a task from one cgroup to another. 1787 * cgroup_task_migrate - move a task from one cgroup to another.
1743 * @old_cgrp; the cgroup @tsk is being migrated from 1788 * @old_cgrp; the cgroup @tsk is being migrated from
1744 * @tsk: the task being migrated 1789 * @tsk: the task being migrated
1745 * @new_cset: the new css_set @tsk is being attached to 1790 * @new_cset: the new css_set @tsk is being attached to
1746 * 1791 *
1747 * Must be called with cgroup_mutex, threadgroup and css_set_rwsem locked. 1792 * Must be called with cgroup_mutex, threadgroup and css_set_rwsem locked.
1748 */ 1793 */
1749 static void cgroup_task_migrate(struct cgroup *old_cgrp, 1794 static void cgroup_task_migrate(struct cgroup *old_cgrp,
1750 struct task_struct *tsk, 1795 struct task_struct *tsk,
1751 struct css_set *new_cset) 1796 struct css_set *new_cset)
1752 { 1797 {
1753 struct css_set *old_cset; 1798 struct css_set *old_cset;
1754 1799
1755 lockdep_assert_held(&cgroup_mutex); 1800 lockdep_assert_held(&cgroup_mutex);
1756 lockdep_assert_held(&css_set_rwsem); 1801 lockdep_assert_held(&css_set_rwsem);
1757 1802
1758 /* 1803 /*
1759 * We are synchronized through threadgroup_lock() against PF_EXITING 1804 * We are synchronized through threadgroup_lock() against PF_EXITING
1760 * setting such that we can't race against cgroup_exit() changing the 1805 * setting such that we can't race against cgroup_exit() changing the
1761 * css_set to init_css_set and dropping the old one. 1806 * css_set to init_css_set and dropping the old one.
1762 */ 1807 */
1763 WARN_ON_ONCE(tsk->flags & PF_EXITING); 1808 WARN_ON_ONCE(tsk->flags & PF_EXITING);
1764 old_cset = task_css_set(tsk); 1809 old_cset = task_css_set(tsk);
1765 1810
1766 get_css_set(new_cset); 1811 get_css_set(new_cset);
1767 rcu_assign_pointer(tsk->cgroups, new_cset); 1812 rcu_assign_pointer(tsk->cgroups, new_cset);
1768 1813
1769 /* 1814 /*
1770 * Use move_tail so that cgroup_taskset_first() still returns the 1815 * Use move_tail so that cgroup_taskset_first() still returns the
1771 * leader after migration. This works because cgroup_migrate() 1816 * leader after migration. This works because cgroup_migrate()
1772 * ensures that the dst_cset of the leader is the first on the 1817 * ensures that the dst_cset of the leader is the first on the
1773 * tset's dst_csets list. 1818 * tset's dst_csets list.
1774 */ 1819 */
1775 list_move_tail(&tsk->cg_list, &new_cset->mg_tasks); 1820 list_move_tail(&tsk->cg_list, &new_cset->mg_tasks);
1776 1821
1777 /* 1822 /*
1778 * We just gained a reference on old_cset by taking it from the 1823 * We just gained a reference on old_cset by taking it from the
1779 * task. As trading it for new_cset is protected by cgroup_mutex, 1824 * task. As trading it for new_cset is protected by cgroup_mutex,
1780 * we're safe to drop it here; it will be freed under RCU. 1825 * we're safe to drop it here; it will be freed under RCU.
1781 */ 1826 */
1782 set_bit(CGRP_RELEASABLE, &old_cgrp->flags); 1827 set_bit(CGRP_RELEASABLE, &old_cgrp->flags);
1783 put_css_set_locked(old_cset, false); 1828 put_css_set_locked(old_cset, false);
1784 } 1829 }
1785 1830
1786 /** 1831 /**
1787 * cgroup_migrate_finish - cleanup after attach 1832 * cgroup_migrate_finish - cleanup after attach
1788 * @preloaded_csets: list of preloaded css_sets 1833 * @preloaded_csets: list of preloaded css_sets
1789 * 1834 *
1790 * Undo cgroup_migrate_add_src() and cgroup_migrate_prepare_dst(). See 1835 * Undo cgroup_migrate_add_src() and cgroup_migrate_prepare_dst(). See
1791 * those functions for details. 1836 * those functions for details.
1792 */ 1837 */
1793 static void cgroup_migrate_finish(struct list_head *preloaded_csets) 1838 static void cgroup_migrate_finish(struct list_head *preloaded_csets)
1794 { 1839 {
1795 struct css_set *cset, *tmp_cset; 1840 struct css_set *cset, *tmp_cset;
1796 1841
1797 lockdep_assert_held(&cgroup_mutex); 1842 lockdep_assert_held(&cgroup_mutex);
1798 1843
1799 down_write(&css_set_rwsem); 1844 down_write(&css_set_rwsem);
1800 list_for_each_entry_safe(cset, tmp_cset, preloaded_csets, mg_preload_node) { 1845 list_for_each_entry_safe(cset, tmp_cset, preloaded_csets, mg_preload_node) {
1801 cset->mg_src_cgrp = NULL; 1846 cset->mg_src_cgrp = NULL;
1802 cset->mg_dst_cset = NULL; 1847 cset->mg_dst_cset = NULL;
1803 list_del_init(&cset->mg_preload_node); 1848 list_del_init(&cset->mg_preload_node);
1804 put_css_set_locked(cset, false); 1849 put_css_set_locked(cset, false);
1805 } 1850 }
1806 up_write(&css_set_rwsem); 1851 up_write(&css_set_rwsem);
1807 } 1852 }
1808 1853
1809 /** 1854 /**
1810 * cgroup_migrate_add_src - add a migration source css_set 1855 * cgroup_migrate_add_src - add a migration source css_set
1811 * @src_cset: the source css_set to add 1856 * @src_cset: the source css_set to add
1812 * @dst_cgrp: the destination cgroup 1857 * @dst_cgrp: the destination cgroup
1813 * @preloaded_csets: list of preloaded css_sets 1858 * @preloaded_csets: list of preloaded css_sets
1814 * 1859 *
1815 * Tasks belonging to @src_cset are about to be migrated to @dst_cgrp. Pin 1860 * Tasks belonging to @src_cset are about to be migrated to @dst_cgrp. Pin
1816 * @src_cset and add it to @preloaded_csets, which should later be cleaned 1861 * @src_cset and add it to @preloaded_csets, which should later be cleaned
1817 * up by cgroup_migrate_finish(). 1862 * up by cgroup_migrate_finish().
1818 * 1863 *
1819 * This function may be called without holding threadgroup_lock even if the 1864 * This function may be called without holding threadgroup_lock even if the
1820 * target is a process. Threads may be created and destroyed but as long 1865 * target is a process. Threads may be created and destroyed but as long
1821 * as cgroup_mutex is not dropped, no new css_set can be put into play and 1866 * as cgroup_mutex is not dropped, no new css_set can be put into play and
1822 * the preloaded css_sets are guaranteed to cover all migrations. 1867 * the preloaded css_sets are guaranteed to cover all migrations.
1823 */ 1868 */
1824 static void cgroup_migrate_add_src(struct css_set *src_cset, 1869 static void cgroup_migrate_add_src(struct css_set *src_cset,
1825 struct cgroup *dst_cgrp, 1870 struct cgroup *dst_cgrp,
1826 struct list_head *preloaded_csets) 1871 struct list_head *preloaded_csets)
1827 { 1872 {
1828 struct cgroup *src_cgrp; 1873 struct cgroup *src_cgrp;
1829 1874
1830 lockdep_assert_held(&cgroup_mutex); 1875 lockdep_assert_held(&cgroup_mutex);
1831 lockdep_assert_held(&css_set_rwsem); 1876 lockdep_assert_held(&css_set_rwsem);
1832 1877
1833 src_cgrp = cset_cgroup_from_root(src_cset, dst_cgrp->root); 1878 src_cgrp = cset_cgroup_from_root(src_cset, dst_cgrp->root);
1834 1879
1835 /* nothing to do if this cset already belongs to the cgroup */ 1880 /* nothing to do if this cset already belongs to the cgroup */
1836 if (src_cgrp == dst_cgrp) 1881 if (src_cgrp == dst_cgrp)
1837 return; 1882 return;
1838 1883
1839 if (!list_empty(&src_cset->mg_preload_node)) 1884 if (!list_empty(&src_cset->mg_preload_node))
1840 return; 1885 return;
1841 1886
1842 WARN_ON(src_cset->mg_src_cgrp); 1887 WARN_ON(src_cset->mg_src_cgrp);
1843 WARN_ON(!list_empty(&src_cset->mg_tasks)); 1888 WARN_ON(!list_empty(&src_cset->mg_tasks));
1844 WARN_ON(!list_empty(&src_cset->mg_node)); 1889 WARN_ON(!list_empty(&src_cset->mg_node));
1845 1890
1846 src_cset->mg_src_cgrp = src_cgrp; 1891 src_cset->mg_src_cgrp = src_cgrp;
1847 get_css_set(src_cset); 1892 get_css_set(src_cset);
1848 list_add(&src_cset->mg_preload_node, preloaded_csets); 1893 list_add(&src_cset->mg_preload_node, preloaded_csets);
1849 } 1894 }
1850 1895
1851 /** 1896 /**
1852 * cgroup_migrate_prepare_dst - prepare destination css_sets for migration 1897 * cgroup_migrate_prepare_dst - prepare destination css_sets for migration
1853 * @dst_cgrp: the destination cgroup 1898 * @dst_cgrp: the destination cgroup
1854 * @preloaded_csets: list of preloaded source css_sets 1899 * @preloaded_csets: list of preloaded source css_sets
1855 * 1900 *
1856 * Tasks are about to be moved to @dst_cgrp and all the source css_sets 1901 * Tasks are about to be moved to @dst_cgrp and all the source css_sets
1857 * have been preloaded to @preloaded_csets. This function looks up and 1902 * have been preloaded to @preloaded_csets. This function looks up and
1858 * pins all destination css_sets, links each to its source, and put them on 1903 * pins all destination css_sets, links each to its source, and put them on
1859 * @preloaded_csets. 1904 * @preloaded_csets.
1860 * 1905 *
1861 * This function must be called after cgroup_migrate_add_src() has been 1906 * This function must be called after cgroup_migrate_add_src() has been
1862 * called on each migration source css_set. After migration is performed 1907 * called on each migration source css_set. After migration is performed
1863 * using cgroup_migrate(), cgroup_migrate_finish() must be called on 1908 * using cgroup_migrate(), cgroup_migrate_finish() must be called on
1864 * @preloaded_csets. 1909 * @preloaded_csets.
1865 */ 1910 */
1866 static int cgroup_migrate_prepare_dst(struct cgroup *dst_cgrp, 1911 static int cgroup_migrate_prepare_dst(struct cgroup *dst_cgrp,
1867 struct list_head *preloaded_csets) 1912 struct list_head *preloaded_csets)
1868 { 1913 {
1869 LIST_HEAD(csets); 1914 LIST_HEAD(csets);
1870 struct css_set *src_cset; 1915 struct css_set *src_cset;
1871 1916
1872 lockdep_assert_held(&cgroup_mutex); 1917 lockdep_assert_held(&cgroup_mutex);
1873 1918
1874 /* look up the dst cset for each src cset and link it to src */ 1919 /* look up the dst cset for each src cset and link it to src */
1875 list_for_each_entry(src_cset, preloaded_csets, mg_preload_node) { 1920 list_for_each_entry(src_cset, preloaded_csets, mg_preload_node) {
1876 struct css_set *dst_cset; 1921 struct css_set *dst_cset;
1877 1922
1878 dst_cset = find_css_set(src_cset, dst_cgrp); 1923 dst_cset = find_css_set(src_cset, dst_cgrp);
1879 if (!dst_cset) 1924 if (!dst_cset)
1880 goto err; 1925 goto err;
1881 1926
1882 WARN_ON_ONCE(src_cset->mg_dst_cset || dst_cset->mg_dst_cset); 1927 WARN_ON_ONCE(src_cset->mg_dst_cset || dst_cset->mg_dst_cset);
1883 src_cset->mg_dst_cset = dst_cset; 1928 src_cset->mg_dst_cset = dst_cset;
1884 1929
1885 if (list_empty(&dst_cset->mg_preload_node)) 1930 if (list_empty(&dst_cset->mg_preload_node))
1886 list_add(&dst_cset->mg_preload_node, &csets); 1931 list_add(&dst_cset->mg_preload_node, &csets);
1887 else 1932 else
1888 put_css_set(dst_cset, false); 1933 put_css_set(dst_cset, false);
1889 } 1934 }
1890 1935
1891 list_splice(&csets, preloaded_csets); 1936 list_splice(&csets, preloaded_csets);
1892 return 0; 1937 return 0;
1893 err: 1938 err:
1894 cgroup_migrate_finish(&csets); 1939 cgroup_migrate_finish(&csets);
1895 return -ENOMEM; 1940 return -ENOMEM;
1896 } 1941 }
1897 1942
1898 /** 1943 /**
1899 * cgroup_migrate - migrate a process or task to a cgroup 1944 * cgroup_migrate - migrate a process or task to a cgroup
1900 * @cgrp: the destination cgroup 1945 * @cgrp: the destination cgroup
1901 * @leader: the leader of the process or the task to migrate 1946 * @leader: the leader of the process or the task to migrate
1902 * @threadgroup: whether @leader points to the whole process or a single task 1947 * @threadgroup: whether @leader points to the whole process or a single task
1903 * 1948 *
1904 * Migrate a process or task denoted by @leader to @cgrp. If migrating a 1949 * Migrate a process or task denoted by @leader to @cgrp. If migrating a
1905 * process, the caller must be holding threadgroup_lock of @leader. The 1950 * process, the caller must be holding threadgroup_lock of @leader. The
1906 * caller is also responsible for invoking cgroup_migrate_add_src() and 1951 * caller is also responsible for invoking cgroup_migrate_add_src() and
1907 * cgroup_migrate_prepare_dst() on the targets before invoking this 1952 * cgroup_migrate_prepare_dst() on the targets before invoking this
1908 * function and following up with cgroup_migrate_finish(). 1953 * function and following up with cgroup_migrate_finish().
1909 * 1954 *
1910 * As long as a controller's ->can_attach() doesn't fail, this function is 1955 * As long as a controller's ->can_attach() doesn't fail, this function is
1911 * guaranteed to succeed. This means that, excluding ->can_attach() 1956 * guaranteed to succeed. This means that, excluding ->can_attach()
1912 * failure, when migrating multiple targets, the success or failure can be 1957 * failure, when migrating multiple targets, the success or failure can be
1913 * decided for all targets by invoking group_migrate_prepare_dst() before 1958 * decided for all targets by invoking group_migrate_prepare_dst() before
1914 * actually starting migrating. 1959 * actually starting migrating.
1915 */ 1960 */
1916 static int cgroup_migrate(struct cgroup *cgrp, struct task_struct *leader, 1961 static int cgroup_migrate(struct cgroup *cgrp, struct task_struct *leader,
1917 bool threadgroup) 1962 bool threadgroup)
1918 { 1963 {
1919 struct cgroup_taskset tset = { 1964 struct cgroup_taskset tset = {
1920 .src_csets = LIST_HEAD_INIT(tset.src_csets), 1965 .src_csets = LIST_HEAD_INIT(tset.src_csets),
1921 .dst_csets = LIST_HEAD_INIT(tset.dst_csets), 1966 .dst_csets = LIST_HEAD_INIT(tset.dst_csets),
1922 .csets = &tset.src_csets, 1967 .csets = &tset.src_csets,
1923 }; 1968 };
1924 struct cgroup_subsys_state *css, *failed_css = NULL; 1969 struct cgroup_subsys_state *css, *failed_css = NULL;
1925 struct css_set *cset, *tmp_cset; 1970 struct css_set *cset, *tmp_cset;
1926 struct task_struct *task, *tmp_task; 1971 struct task_struct *task, *tmp_task;
1927 int i, ret; 1972 int i, ret;
1928 1973
1929 /* 1974 /*
1930 * Prevent freeing of tasks while we take a snapshot. Tasks that are 1975 * Prevent freeing of tasks while we take a snapshot. Tasks that are
1931 * already PF_EXITING could be freed from underneath us unless we 1976 * already PF_EXITING could be freed from underneath us unless we
1932 * take an rcu_read_lock. 1977 * take an rcu_read_lock.
1933 */ 1978 */
1934 down_write(&css_set_rwsem); 1979 down_write(&css_set_rwsem);
1935 rcu_read_lock(); 1980 rcu_read_lock();
1936 task = leader; 1981 task = leader;
1937 do { 1982 do {
1938 /* @task either already exited or can't exit until the end */ 1983 /* @task either already exited or can't exit until the end */
1939 if (task->flags & PF_EXITING) 1984 if (task->flags & PF_EXITING)
1940 goto next; 1985 goto next;
1941 1986
1942 /* leave @task alone if post_fork() hasn't linked it yet */ 1987 /* leave @task alone if post_fork() hasn't linked it yet */
1943 if (list_empty(&task->cg_list)) 1988 if (list_empty(&task->cg_list))
1944 goto next; 1989 goto next;
1945 1990
1946 cset = task_css_set(task); 1991 cset = task_css_set(task);
1947 if (!cset->mg_src_cgrp) 1992 if (!cset->mg_src_cgrp)
1948 goto next; 1993 goto next;
1949 1994
1950 /* 1995 /*
1951 * cgroup_taskset_first() must always return the leader. 1996 * cgroup_taskset_first() must always return the leader.
1952 * Take care to avoid disturbing the ordering. 1997 * Take care to avoid disturbing the ordering.
1953 */ 1998 */
1954 list_move_tail(&task->cg_list, &cset->mg_tasks); 1999 list_move_tail(&task->cg_list, &cset->mg_tasks);
1955 if (list_empty(&cset->mg_node)) 2000 if (list_empty(&cset->mg_node))
1956 list_add_tail(&cset->mg_node, &tset.src_csets); 2001 list_add_tail(&cset->mg_node, &tset.src_csets);
1957 if (list_empty(&cset->mg_dst_cset->mg_node)) 2002 if (list_empty(&cset->mg_dst_cset->mg_node))
1958 list_move_tail(&cset->mg_dst_cset->mg_node, 2003 list_move_tail(&cset->mg_dst_cset->mg_node,
1959 &tset.dst_csets); 2004 &tset.dst_csets);
1960 next: 2005 next:
1961 if (!threadgroup) 2006 if (!threadgroup)
1962 break; 2007 break;
1963 } while_each_thread(leader, task); 2008 } while_each_thread(leader, task);
1964 rcu_read_unlock(); 2009 rcu_read_unlock();
1965 up_write(&css_set_rwsem); 2010 up_write(&css_set_rwsem);
1966 2011
1967 /* methods shouldn't be called if no task is actually migrating */ 2012 /* methods shouldn't be called if no task is actually migrating */
1968 if (list_empty(&tset.src_csets)) 2013 if (list_empty(&tset.src_csets))
1969 return 0; 2014 return 0;
1970 2015
1971 /* check that we can legitimately attach to the cgroup */ 2016 /* check that we can legitimately attach to the cgroup */
1972 for_each_css(css, i, cgrp) { 2017 for_each_e_css(css, i, cgrp) {
1973 if (css->ss->can_attach) { 2018 if (css->ss->can_attach) {
1974 ret = css->ss->can_attach(css, &tset); 2019 ret = css->ss->can_attach(css, &tset);
1975 if (ret) { 2020 if (ret) {
1976 failed_css = css; 2021 failed_css = css;
1977 goto out_cancel_attach; 2022 goto out_cancel_attach;
1978 } 2023 }
1979 } 2024 }
1980 } 2025 }
1981 2026
1982 /* 2027 /*
1983 * Now that we're guaranteed success, proceed to move all tasks to 2028 * Now that we're guaranteed success, proceed to move all tasks to
1984 * the new cgroup. There are no failure cases after here, so this 2029 * the new cgroup. There are no failure cases after here, so this
1985 * is the commit point. 2030 * is the commit point.
1986 */ 2031 */
1987 down_write(&css_set_rwsem); 2032 down_write(&css_set_rwsem);
1988 list_for_each_entry(cset, &tset.src_csets, mg_node) { 2033 list_for_each_entry(cset, &tset.src_csets, mg_node) {
1989 list_for_each_entry_safe(task, tmp_task, &cset->mg_tasks, cg_list) 2034 list_for_each_entry_safe(task, tmp_task, &cset->mg_tasks, cg_list)
1990 cgroup_task_migrate(cset->mg_src_cgrp, task, 2035 cgroup_task_migrate(cset->mg_src_cgrp, task,
1991 cset->mg_dst_cset); 2036 cset->mg_dst_cset);
1992 } 2037 }
1993 up_write(&css_set_rwsem); 2038 up_write(&css_set_rwsem);
1994 2039
1995 /* 2040 /*
1996 * Migration is committed, all target tasks are now on dst_csets. 2041 * Migration is committed, all target tasks are now on dst_csets.
1997 * Nothing is sensitive to fork() after this point. Notify 2042 * Nothing is sensitive to fork() after this point. Notify
1998 * controllers that migration is complete. 2043 * controllers that migration is complete.
1999 */ 2044 */
2000 tset.csets = &tset.dst_csets; 2045 tset.csets = &tset.dst_csets;
2001 2046
2002 for_each_css(css, i, cgrp) 2047 for_each_e_css(css, i, cgrp)
2003 if (css->ss->attach) 2048 if (css->ss->attach)
2004 css->ss->attach(css, &tset); 2049 css->ss->attach(css, &tset);
2005 2050
2006 ret = 0; 2051 ret = 0;
2007 goto out_release_tset; 2052 goto out_release_tset;
2008 2053
2009 out_cancel_attach: 2054 out_cancel_attach:
2010 for_each_css(css, i, cgrp) { 2055 for_each_e_css(css, i, cgrp) {
2011 if (css == failed_css) 2056 if (css == failed_css)
2012 break; 2057 break;
2013 if (css->ss->cancel_attach) 2058 if (css->ss->cancel_attach)
2014 css->ss->cancel_attach(css, &tset); 2059 css->ss->cancel_attach(css, &tset);
2015 } 2060 }
2016 out_release_tset: 2061 out_release_tset:
2017 down_write(&css_set_rwsem); 2062 down_write(&css_set_rwsem);
2018 list_splice_init(&tset.dst_csets, &tset.src_csets); 2063 list_splice_init(&tset.dst_csets, &tset.src_csets);
2019 list_for_each_entry_safe(cset, tmp_cset, &tset.src_csets, mg_node) { 2064 list_for_each_entry_safe(cset, tmp_cset, &tset.src_csets, mg_node) {
2020 list_splice_tail_init(&cset->mg_tasks, &cset->tasks); 2065 list_splice_tail_init(&cset->mg_tasks, &cset->tasks);
2021 list_del_init(&cset->mg_node); 2066 list_del_init(&cset->mg_node);
2022 } 2067 }
2023 up_write(&css_set_rwsem); 2068 up_write(&css_set_rwsem);
2024 return ret; 2069 return ret;
2025 } 2070 }
2026 2071
2027 /** 2072 /**
2028 * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup 2073 * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup
2029 * @dst_cgrp: the cgroup to attach to 2074 * @dst_cgrp: the cgroup to attach to
2030 * @leader: the task or the leader of the threadgroup to be attached 2075 * @leader: the task or the leader of the threadgroup to be attached
2031 * @threadgroup: attach the whole threadgroup? 2076 * @threadgroup: attach the whole threadgroup?
2032 * 2077 *
2033 * Call holding cgroup_mutex and threadgroup_lock of @leader. 2078 * Call holding cgroup_mutex and threadgroup_lock of @leader.
2034 */ 2079 */
2035 static int cgroup_attach_task(struct cgroup *dst_cgrp, 2080 static int cgroup_attach_task(struct cgroup *dst_cgrp,
2036 struct task_struct *leader, bool threadgroup) 2081 struct task_struct *leader, bool threadgroup)
2037 { 2082 {
2038 LIST_HEAD(preloaded_csets); 2083 LIST_HEAD(preloaded_csets);
2039 struct task_struct *task; 2084 struct task_struct *task;
2040 int ret; 2085 int ret;
2041 2086
2042 /* look up all src csets */ 2087 /* look up all src csets */
2043 down_read(&css_set_rwsem); 2088 down_read(&css_set_rwsem);
2044 rcu_read_lock(); 2089 rcu_read_lock();
2045 task = leader; 2090 task = leader;
2046 do { 2091 do {
2047 cgroup_migrate_add_src(task_css_set(task), dst_cgrp, 2092 cgroup_migrate_add_src(task_css_set(task), dst_cgrp,
2048 &preloaded_csets); 2093 &preloaded_csets);
2049 if (!threadgroup) 2094 if (!threadgroup)
2050 break; 2095 break;
2051 } while_each_thread(leader, task); 2096 } while_each_thread(leader, task);
2052 rcu_read_unlock(); 2097 rcu_read_unlock();
2053 up_read(&css_set_rwsem); 2098 up_read(&css_set_rwsem);
2054 2099
2055 /* prepare dst csets and commit */ 2100 /* prepare dst csets and commit */
2056 ret = cgroup_migrate_prepare_dst(dst_cgrp, &preloaded_csets); 2101 ret = cgroup_migrate_prepare_dst(dst_cgrp, &preloaded_csets);
2057 if (!ret) 2102 if (!ret)
2058 ret = cgroup_migrate(dst_cgrp, leader, threadgroup); 2103 ret = cgroup_migrate(dst_cgrp, leader, threadgroup);
2059 2104
2060 cgroup_migrate_finish(&preloaded_csets); 2105 cgroup_migrate_finish(&preloaded_csets);
2061 return ret; 2106 return ret;
2062 } 2107 }
2063 2108
2064 /* 2109 /*
2065 * Find the task_struct of the task to attach by vpid and pass it along to the 2110 * Find the task_struct of the task to attach by vpid and pass it along to the
2066 * function to attach either it or all tasks in its threadgroup. Will lock 2111 * function to attach either it or all tasks in its threadgroup. Will lock
2067 * cgroup_mutex and threadgroup. 2112 * cgroup_mutex and threadgroup.
2068 */ 2113 */
2069 static int attach_task_by_pid(struct cgroup *cgrp, u64 pid, bool threadgroup) 2114 static int attach_task_by_pid(struct cgroup *cgrp, u64 pid, bool threadgroup)
2070 { 2115 {
2071 struct task_struct *tsk; 2116 struct task_struct *tsk;
2072 const struct cred *cred = current_cred(), *tcred; 2117 const struct cred *cred = current_cred(), *tcred;
2073 int ret; 2118 int ret;
2074 2119
2075 if (!cgroup_lock_live_group(cgrp)) 2120 if (!cgroup_lock_live_group(cgrp))
2076 return -ENODEV; 2121 return -ENODEV;
2077 2122
2078 retry_find_task: 2123 retry_find_task:
2079 rcu_read_lock(); 2124 rcu_read_lock();
2080 if (pid) { 2125 if (pid) {
2081 tsk = find_task_by_vpid(pid); 2126 tsk = find_task_by_vpid(pid);
2082 if (!tsk) { 2127 if (!tsk) {
2083 rcu_read_unlock(); 2128 rcu_read_unlock();
2084 ret = -ESRCH; 2129 ret = -ESRCH;
2085 goto out_unlock_cgroup; 2130 goto out_unlock_cgroup;
2086 } 2131 }
2087 /* 2132 /*
2088 * even if we're attaching all tasks in the thread group, we 2133 * even if we're attaching all tasks in the thread group, we
2089 * only need to check permissions on one of them. 2134 * only need to check permissions on one of them.
2090 */ 2135 */
2091 tcred = __task_cred(tsk); 2136 tcred = __task_cred(tsk);
2092 if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) && 2137 if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) &&
2093 !uid_eq(cred->euid, tcred->uid) && 2138 !uid_eq(cred->euid, tcred->uid) &&
2094 !uid_eq(cred->euid, tcred->suid)) { 2139 !uid_eq(cred->euid, tcred->suid)) {
2095 rcu_read_unlock(); 2140 rcu_read_unlock();
2096 ret = -EACCES; 2141 ret = -EACCES;
2097 goto out_unlock_cgroup; 2142 goto out_unlock_cgroup;
2098 } 2143 }
2099 } else 2144 } else
2100 tsk = current; 2145 tsk = current;
2101 2146
2102 if (threadgroup) 2147 if (threadgroup)
2103 tsk = tsk->group_leader; 2148 tsk = tsk->group_leader;
2104 2149
2105 /* 2150 /*
2106 * Workqueue threads may acquire PF_NO_SETAFFINITY and become 2151 * Workqueue threads may acquire PF_NO_SETAFFINITY and become
2107 * trapped in a cpuset, or RT worker may be born in a cgroup 2152 * trapped in a cpuset, or RT worker may be born in a cgroup
2108 * with no rt_runtime allocated. Just say no. 2153 * with no rt_runtime allocated. Just say no.
2109 */ 2154 */
2110 if (tsk == kthreadd_task || (tsk->flags & PF_NO_SETAFFINITY)) { 2155 if (tsk == kthreadd_task || (tsk->flags & PF_NO_SETAFFINITY)) {
2111 ret = -EINVAL; 2156 ret = -EINVAL;
2112 rcu_read_unlock(); 2157 rcu_read_unlock();
2113 goto out_unlock_cgroup; 2158 goto out_unlock_cgroup;
2114 } 2159 }
2115 2160
2116 get_task_struct(tsk); 2161 get_task_struct(tsk);
2117 rcu_read_unlock(); 2162 rcu_read_unlock();
2118 2163
2119 threadgroup_lock(tsk); 2164 threadgroup_lock(tsk);
2120 if (threadgroup) { 2165 if (threadgroup) {
2121 if (!thread_group_leader(tsk)) { 2166 if (!thread_group_leader(tsk)) {
2122 /* 2167 /*
2123 * a race with de_thread from another thread's exec() 2168 * a race with de_thread from another thread's exec()
2124 * may strip us of our leadership, if this happens, 2169 * may strip us of our leadership, if this happens,
2125 * there is no choice but to throw this task away and 2170 * there is no choice but to throw this task away and
2126 * try again; this is 2171 * try again; this is
2127 * "double-double-toil-and-trouble-check locking". 2172 * "double-double-toil-and-trouble-check locking".
2128 */ 2173 */
2129 threadgroup_unlock(tsk); 2174 threadgroup_unlock(tsk);
2130 put_task_struct(tsk); 2175 put_task_struct(tsk);
2131 goto retry_find_task; 2176 goto retry_find_task;
2132 } 2177 }
2133 } 2178 }
2134 2179
2135 ret = cgroup_attach_task(cgrp, tsk, threadgroup); 2180 ret = cgroup_attach_task(cgrp, tsk, threadgroup);
2136 2181
2137 threadgroup_unlock(tsk); 2182 threadgroup_unlock(tsk);
2138 2183
2139 put_task_struct(tsk); 2184 put_task_struct(tsk);
2140 out_unlock_cgroup: 2185 out_unlock_cgroup:
2141 mutex_unlock(&cgroup_mutex); 2186 mutex_unlock(&cgroup_mutex);
2142 return ret; 2187 return ret;
2143 } 2188 }
2144 2189
2145 /** 2190 /**
2146 * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from' 2191 * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from'
2147 * @from: attach to all cgroups of a given task 2192 * @from: attach to all cgroups of a given task
2148 * @tsk: the task to be attached 2193 * @tsk: the task to be attached
2149 */ 2194 */
2150 int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk) 2195 int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk)
2151 { 2196 {
2152 struct cgroup_root *root; 2197 struct cgroup_root *root;
2153 int retval = 0; 2198 int retval = 0;
2154 2199
2155 mutex_lock(&cgroup_mutex); 2200 mutex_lock(&cgroup_mutex);
2156 for_each_root(root) { 2201 for_each_root(root) {
2157 struct cgroup *from_cgrp; 2202 struct cgroup *from_cgrp;
2158 2203
2159 if (root == &cgrp_dfl_root) 2204 if (root == &cgrp_dfl_root)
2160 continue; 2205 continue;
2161 2206
2162 down_read(&css_set_rwsem); 2207 down_read(&css_set_rwsem);
2163 from_cgrp = task_cgroup_from_root(from, root); 2208 from_cgrp = task_cgroup_from_root(from, root);
2164 up_read(&css_set_rwsem); 2209 up_read(&css_set_rwsem);
2165 2210
2166 retval = cgroup_attach_task(from_cgrp, tsk, false); 2211 retval = cgroup_attach_task(from_cgrp, tsk, false);
2167 if (retval) 2212 if (retval)
2168 break; 2213 break;
2169 } 2214 }
2170 mutex_unlock(&cgroup_mutex); 2215 mutex_unlock(&cgroup_mutex);
2171 2216
2172 return retval; 2217 return retval;
2173 } 2218 }
2174 EXPORT_SYMBOL_GPL(cgroup_attach_task_all); 2219 EXPORT_SYMBOL_GPL(cgroup_attach_task_all);
2175 2220
2176 static int cgroup_tasks_write(struct cgroup_subsys_state *css, 2221 static int cgroup_tasks_write(struct cgroup_subsys_state *css,
2177 struct cftype *cft, u64 pid) 2222 struct cftype *cft, u64 pid)
2178 { 2223 {
2179 return attach_task_by_pid(css->cgroup, pid, false); 2224 return attach_task_by_pid(css->cgroup, pid, false);
2180 } 2225 }
2181 2226
2182 static int cgroup_procs_write(struct cgroup_subsys_state *css, 2227 static int cgroup_procs_write(struct cgroup_subsys_state *css,
2183 struct cftype *cft, u64 tgid) 2228 struct cftype *cft, u64 tgid)
2184 { 2229 {
2185 return attach_task_by_pid(css->cgroup, tgid, true); 2230 return attach_task_by_pid(css->cgroup, tgid, true);
2186 } 2231 }
2187 2232
2188 static int cgroup_release_agent_write(struct cgroup_subsys_state *css, 2233 static int cgroup_release_agent_write(struct cgroup_subsys_state *css,
2189 struct cftype *cft, char *buffer) 2234 struct cftype *cft, char *buffer)
2190 { 2235 {
2191 struct cgroup_root *root = css->cgroup->root; 2236 struct cgroup_root *root = css->cgroup->root;
2192 2237
2193 BUILD_BUG_ON(sizeof(root->release_agent_path) < PATH_MAX); 2238 BUILD_BUG_ON(sizeof(root->release_agent_path) < PATH_MAX);
2194 if (!cgroup_lock_live_group(css->cgroup)) 2239 if (!cgroup_lock_live_group(css->cgroup))
2195 return -ENODEV; 2240 return -ENODEV;
2196 spin_lock(&release_agent_path_lock); 2241 spin_lock(&release_agent_path_lock);
2197 strlcpy(root->release_agent_path, buffer, 2242 strlcpy(root->release_agent_path, buffer,
2198 sizeof(root->release_agent_path)); 2243 sizeof(root->release_agent_path));
2199 spin_unlock(&release_agent_path_lock); 2244 spin_unlock(&release_agent_path_lock);
2200 mutex_unlock(&cgroup_mutex); 2245 mutex_unlock(&cgroup_mutex);
2201 return 0; 2246 return 0;
2202 } 2247 }
2203 2248
2204 static int cgroup_release_agent_show(struct seq_file *seq, void *v) 2249 static int cgroup_release_agent_show(struct seq_file *seq, void *v)
2205 { 2250 {
2206 struct cgroup *cgrp = seq_css(seq)->cgroup; 2251 struct cgroup *cgrp = seq_css(seq)->cgroup;
2207 2252
2208 if (!cgroup_lock_live_group(cgrp)) 2253 if (!cgroup_lock_live_group(cgrp))
2209 return -ENODEV; 2254 return -ENODEV;
2210 seq_puts(seq, cgrp->root->release_agent_path); 2255 seq_puts(seq, cgrp->root->release_agent_path);
2211 seq_putc(seq, '\n'); 2256 seq_putc(seq, '\n');
2212 mutex_unlock(&cgroup_mutex); 2257 mutex_unlock(&cgroup_mutex);
2213 return 0; 2258 return 0;
2214 } 2259 }
2215 2260
2216 static int cgroup_sane_behavior_show(struct seq_file *seq, void *v) 2261 static int cgroup_sane_behavior_show(struct seq_file *seq, void *v)
2217 { 2262 {
2218 struct cgroup *cgrp = seq_css(seq)->cgroup; 2263 struct cgroup *cgrp = seq_css(seq)->cgroup;
2219 2264
2220 seq_printf(seq, "%d\n", cgroup_sane_behavior(cgrp)); 2265 seq_printf(seq, "%d\n", cgroup_sane_behavior(cgrp));
2221 return 0; 2266 return 0;
2222 } 2267 }
2223 2268
2224 static ssize_t cgroup_file_write(struct kernfs_open_file *of, char *buf, 2269 static ssize_t cgroup_file_write(struct kernfs_open_file *of, char *buf,
2225 size_t nbytes, loff_t off) 2270 size_t nbytes, loff_t off)
2226 { 2271 {
2227 struct cgroup *cgrp = of->kn->parent->priv; 2272 struct cgroup *cgrp = of->kn->parent->priv;
2228 struct cftype *cft = of->kn->priv; 2273 struct cftype *cft = of->kn->priv;
2229 struct cgroup_subsys_state *css; 2274 struct cgroup_subsys_state *css;
2230 int ret; 2275 int ret;
2231 2276
2232 /* 2277 /*
2233 * kernfs guarantees that a file isn't deleted with operations in 2278 * kernfs guarantees that a file isn't deleted with operations in
2234 * flight, which means that the matching css is and stays alive and 2279 * flight, which means that the matching css is and stays alive and
2235 * doesn't need to be pinned. The RCU locking is not necessary 2280 * doesn't need to be pinned. The RCU locking is not necessary
2236 * either. It's just for the convenience of using cgroup_css(). 2281 * either. It's just for the convenience of using cgroup_css().
2237 */ 2282 */
2238 rcu_read_lock(); 2283 rcu_read_lock();
2239 css = cgroup_css(cgrp, cft->ss); 2284 css = cgroup_css(cgrp, cft->ss);
2240 rcu_read_unlock(); 2285 rcu_read_unlock();
2241 2286
2242 if (cft->write_string) { 2287 if (cft->write_string) {
2243 ret = cft->write_string(css, cft, strstrip(buf)); 2288 ret = cft->write_string(css, cft, strstrip(buf));
2244 } else if (cft->write_u64) { 2289 } else if (cft->write_u64) {
2245 unsigned long long v; 2290 unsigned long long v;
2246 ret = kstrtoull(buf, 0, &v); 2291 ret = kstrtoull(buf, 0, &v);
2247 if (!ret) 2292 if (!ret)
2248 ret = cft->write_u64(css, cft, v); 2293 ret = cft->write_u64(css, cft, v);
2249 } else if (cft->write_s64) { 2294 } else if (cft->write_s64) {
2250 long long v; 2295 long long v;
2251 ret = kstrtoll(buf, 0, &v); 2296 ret = kstrtoll(buf, 0, &v);
2252 if (!ret) 2297 if (!ret)
2253 ret = cft->write_s64(css, cft, v); 2298 ret = cft->write_s64(css, cft, v);
2254 } else if (cft->trigger) { 2299 } else if (cft->trigger) {
2255 ret = cft->trigger(css, (unsigned int)cft->private); 2300 ret = cft->trigger(css, (unsigned int)cft->private);
2256 } else { 2301 } else {
2257 ret = -EINVAL; 2302 ret = -EINVAL;
2258 } 2303 }
2259 2304
2260 return ret ?: nbytes; 2305 return ret ?: nbytes;
2261 } 2306 }
2262 2307
2263 static void *cgroup_seqfile_start(struct seq_file *seq, loff_t *ppos) 2308 static void *cgroup_seqfile_start(struct seq_file *seq, loff_t *ppos)
2264 { 2309 {
2265 return seq_cft(seq)->seq_start(seq, ppos); 2310 return seq_cft(seq)->seq_start(seq, ppos);
2266 } 2311 }
2267 2312
2268 static void *cgroup_seqfile_next(struct seq_file *seq, void *v, loff_t *ppos) 2313 static void *cgroup_seqfile_next(struct seq_file *seq, void *v, loff_t *ppos)
2269 { 2314 {
2270 return seq_cft(seq)->seq_next(seq, v, ppos); 2315 return seq_cft(seq)->seq_next(seq, v, ppos);
2271 } 2316 }
2272 2317
2273 static void cgroup_seqfile_stop(struct seq_file *seq, void *v) 2318 static void cgroup_seqfile_stop(struct seq_file *seq, void *v)
2274 { 2319 {
2275 seq_cft(seq)->seq_stop(seq, v); 2320 seq_cft(seq)->seq_stop(seq, v);
2276 } 2321 }
2277 2322
2278 static int cgroup_seqfile_show(struct seq_file *m, void *arg) 2323 static int cgroup_seqfile_show(struct seq_file *m, void *arg)
2279 { 2324 {
2280 struct cftype *cft = seq_cft(m); 2325 struct cftype *cft = seq_cft(m);
2281 struct cgroup_subsys_state *css = seq_css(m); 2326 struct cgroup_subsys_state *css = seq_css(m);
2282 2327
2283 if (cft->seq_show) 2328 if (cft->seq_show)
2284 return cft->seq_show(m, arg); 2329 return cft->seq_show(m, arg);
2285 2330
2286 if (cft->read_u64) 2331 if (cft->read_u64)
2287 seq_printf(m, "%llu\n", cft->read_u64(css, cft)); 2332 seq_printf(m, "%llu\n", cft->read_u64(css, cft));
2288 else if (cft->read_s64) 2333 else if (cft->read_s64)
2289 seq_printf(m, "%lld\n", cft->read_s64(css, cft)); 2334 seq_printf(m, "%lld\n", cft->read_s64(css, cft));
2290 else 2335 else
2291 return -EINVAL; 2336 return -EINVAL;
2292 return 0; 2337 return 0;
2293 } 2338 }
2294 2339
2295 static struct kernfs_ops cgroup_kf_single_ops = { 2340 static struct kernfs_ops cgroup_kf_single_ops = {
2296 .atomic_write_len = PAGE_SIZE, 2341 .atomic_write_len = PAGE_SIZE,
2297 .write = cgroup_file_write, 2342 .write = cgroup_file_write,
2298 .seq_show = cgroup_seqfile_show, 2343 .seq_show = cgroup_seqfile_show,
2299 }; 2344 };
2300 2345
2301 static struct kernfs_ops cgroup_kf_ops = { 2346 static struct kernfs_ops cgroup_kf_ops = {
2302 .atomic_write_len = PAGE_SIZE, 2347 .atomic_write_len = PAGE_SIZE,
2303 .write = cgroup_file_write, 2348 .write = cgroup_file_write,
2304 .seq_start = cgroup_seqfile_start, 2349 .seq_start = cgroup_seqfile_start,
2305 .seq_next = cgroup_seqfile_next, 2350 .seq_next = cgroup_seqfile_next,
2306 .seq_stop = cgroup_seqfile_stop, 2351 .seq_stop = cgroup_seqfile_stop,
2307 .seq_show = cgroup_seqfile_show, 2352 .seq_show = cgroup_seqfile_show,
2308 }; 2353 };
2309 2354
2310 /* 2355 /*
2311 * cgroup_rename - Only allow simple rename of directories in place. 2356 * cgroup_rename - Only allow simple rename of directories in place.
2312 */ 2357 */
2313 static int cgroup_rename(struct kernfs_node *kn, struct kernfs_node *new_parent, 2358 static int cgroup_rename(struct kernfs_node *kn, struct kernfs_node *new_parent,
2314 const char *new_name_str) 2359 const char *new_name_str)
2315 { 2360 {
2316 struct cgroup *cgrp = kn->priv; 2361 struct cgroup *cgrp = kn->priv;
2317 int ret; 2362 int ret;
2318 2363
2319 if (kernfs_type(kn) != KERNFS_DIR) 2364 if (kernfs_type(kn) != KERNFS_DIR)
2320 return -ENOTDIR; 2365 return -ENOTDIR;
2321 if (kn->parent != new_parent) 2366 if (kn->parent != new_parent)
2322 return -EIO; 2367 return -EIO;
2323 2368
2324 /* 2369 /*
2325 * This isn't a proper migration and its usefulness is very 2370 * This isn't a proper migration and its usefulness is very
2326 * limited. Disallow if sane_behavior. 2371 * limited. Disallow if sane_behavior.
2327 */ 2372 */
2328 if (cgroup_sane_behavior(cgrp)) 2373 if (cgroup_sane_behavior(cgrp))
2329 return -EPERM; 2374 return -EPERM;
2330 2375
2331 /* 2376 /*
2332 * We're gonna grab cgroup_tree_mutex which nests outside kernfs 2377 * We're gonna grab cgroup_tree_mutex which nests outside kernfs
2333 * active_ref. kernfs_rename() doesn't require active_ref 2378 * active_ref. kernfs_rename() doesn't require active_ref
2334 * protection. Break them before grabbing cgroup_tree_mutex. 2379 * protection. Break them before grabbing cgroup_tree_mutex.
2335 */ 2380 */
2336 kernfs_break_active_protection(new_parent); 2381 kernfs_break_active_protection(new_parent);
2337 kernfs_break_active_protection(kn); 2382 kernfs_break_active_protection(kn);
2338 2383
2339 mutex_lock(&cgroup_tree_mutex); 2384 mutex_lock(&cgroup_tree_mutex);
2340 mutex_lock(&cgroup_mutex); 2385 mutex_lock(&cgroup_mutex);
2341 2386
2342 ret = kernfs_rename(kn, new_parent, new_name_str); 2387 ret = kernfs_rename(kn, new_parent, new_name_str);
2343 2388
2344 mutex_unlock(&cgroup_mutex); 2389 mutex_unlock(&cgroup_mutex);
2345 mutex_unlock(&cgroup_tree_mutex); 2390 mutex_unlock(&cgroup_tree_mutex);
2346 2391
2347 kernfs_unbreak_active_protection(kn); 2392 kernfs_unbreak_active_protection(kn);
2348 kernfs_unbreak_active_protection(new_parent); 2393 kernfs_unbreak_active_protection(new_parent);
2349 return ret; 2394 return ret;
2350 } 2395 }
2351 2396
2352 /* set uid and gid of cgroup dirs and files to that of the creator */ 2397 /* set uid and gid of cgroup dirs and files to that of the creator */
2353 static int cgroup_kn_set_ugid(struct kernfs_node *kn) 2398 static int cgroup_kn_set_ugid(struct kernfs_node *kn)
2354 { 2399 {
2355 struct iattr iattr = { .ia_valid = ATTR_UID | ATTR_GID, 2400 struct iattr iattr = { .ia_valid = ATTR_UID | ATTR_GID,
2356 .ia_uid = current_fsuid(), 2401 .ia_uid = current_fsuid(),
2357 .ia_gid = current_fsgid(), }; 2402 .ia_gid = current_fsgid(), };
2358 2403
2359 if (uid_eq(iattr.ia_uid, GLOBAL_ROOT_UID) && 2404 if (uid_eq(iattr.ia_uid, GLOBAL_ROOT_UID) &&
2360 gid_eq(iattr.ia_gid, GLOBAL_ROOT_GID)) 2405 gid_eq(iattr.ia_gid, GLOBAL_ROOT_GID))
2361 return 0; 2406 return 0;
2362 2407
2363 return kernfs_setattr(kn, &iattr); 2408 return kernfs_setattr(kn, &iattr);
2364 } 2409 }
2365 2410
2366 static int cgroup_add_file(struct cgroup *cgrp, struct cftype *cft) 2411 static int cgroup_add_file(struct cgroup *cgrp, struct cftype *cft)
2367 { 2412 {
2368 char name[CGROUP_FILE_NAME_MAX]; 2413 char name[CGROUP_FILE_NAME_MAX];
2369 struct kernfs_node *kn; 2414 struct kernfs_node *kn;
2370 struct lock_class_key *key = NULL; 2415 struct lock_class_key *key = NULL;
2371 int ret; 2416 int ret;
2372 2417
2373 #ifdef CONFIG_DEBUG_LOCK_ALLOC 2418 #ifdef CONFIG_DEBUG_LOCK_ALLOC
2374 key = &cft->lockdep_key; 2419 key = &cft->lockdep_key;
2375 #endif 2420 #endif
2376 kn = __kernfs_create_file(cgrp->kn, cgroup_file_name(cgrp, cft, name), 2421 kn = __kernfs_create_file(cgrp->kn, cgroup_file_name(cgrp, cft, name),
2377 cgroup_file_mode(cft), 0, cft->kf_ops, cft, 2422 cgroup_file_mode(cft), 0, cft->kf_ops, cft,
2378 NULL, false, key); 2423 NULL, false, key);
2379 if (IS_ERR(kn)) 2424 if (IS_ERR(kn))
2380 return PTR_ERR(kn); 2425 return PTR_ERR(kn);
2381 2426
2382 ret = cgroup_kn_set_ugid(kn); 2427 ret = cgroup_kn_set_ugid(kn);
2383 if (ret) 2428 if (ret)
2384 kernfs_remove(kn); 2429 kernfs_remove(kn);
2385 return ret; 2430 return ret;
2386 } 2431 }
2387 2432
2388 /** 2433 /**
2389 * cgroup_addrm_files - add or remove files to a cgroup directory 2434 * cgroup_addrm_files - add or remove files to a cgroup directory
2390 * @cgrp: the target cgroup 2435 * @cgrp: the target cgroup
2391 * @cfts: array of cftypes to be added 2436 * @cfts: array of cftypes to be added
2392 * @is_add: whether to add or remove 2437 * @is_add: whether to add or remove
2393 * 2438 *
2394 * Depending on @is_add, add or remove files defined by @cfts on @cgrp. 2439 * Depending on @is_add, add or remove files defined by @cfts on @cgrp.
2395 * For removals, this function never fails. If addition fails, this 2440 * For removals, this function never fails. If addition fails, this
2396 * function doesn't remove files already added. The caller is responsible 2441 * function doesn't remove files already added. The caller is responsible
2397 * for cleaning up. 2442 * for cleaning up.
2398 */ 2443 */
2399 static int cgroup_addrm_files(struct cgroup *cgrp, struct cftype cfts[], 2444 static int cgroup_addrm_files(struct cgroup *cgrp, struct cftype cfts[],
2400 bool is_add) 2445 bool is_add)
2401 { 2446 {
2402 struct cftype *cft; 2447 struct cftype *cft;
2403 int ret; 2448 int ret;
2404 2449
2405 lockdep_assert_held(&cgroup_tree_mutex); 2450 lockdep_assert_held(&cgroup_tree_mutex);
2406 2451
2407 for (cft = cfts; cft->name[0] != '\0'; cft++) { 2452 for (cft = cfts; cft->name[0] != '\0'; cft++) {
2408 /* does cft->flags tell us to skip this file on @cgrp? */ 2453 /* does cft->flags tell us to skip this file on @cgrp? */
2409 if ((cft->flags & CFTYPE_ONLY_ON_DFL) && !cgroup_on_dfl(cgrp)) 2454 if ((cft->flags & CFTYPE_ONLY_ON_DFL) && !cgroup_on_dfl(cgrp))
2410 continue; 2455 continue;
2411 if ((cft->flags & CFTYPE_INSANE) && cgroup_sane_behavior(cgrp)) 2456 if ((cft->flags & CFTYPE_INSANE) && cgroup_sane_behavior(cgrp))
2412 continue; 2457 continue;
2413 if ((cft->flags & CFTYPE_NOT_ON_ROOT) && !cgrp->parent) 2458 if ((cft->flags & CFTYPE_NOT_ON_ROOT) && !cgrp->parent)
2414 continue; 2459 continue;
2415 if ((cft->flags & CFTYPE_ONLY_ON_ROOT) && cgrp->parent) 2460 if ((cft->flags & CFTYPE_ONLY_ON_ROOT) && cgrp->parent)
2416 continue; 2461 continue;
2417 2462
2418 if (is_add) { 2463 if (is_add) {
2419 ret = cgroup_add_file(cgrp, cft); 2464 ret = cgroup_add_file(cgrp, cft);
2420 if (ret) { 2465 if (ret) {
2421 pr_warn("cgroup_addrm_files: failed to add %s, err=%d\n", 2466 pr_warn("cgroup_addrm_files: failed to add %s, err=%d\n",
2422 cft->name, ret); 2467 cft->name, ret);
2423 return ret; 2468 return ret;
2424 } 2469 }
2425 } else { 2470 } else {
2426 cgroup_rm_file(cgrp, cft); 2471 cgroup_rm_file(cgrp, cft);
2427 } 2472 }
2428 } 2473 }
2429 return 0; 2474 return 0;
2430 } 2475 }
2431 2476
2432 static int cgroup_apply_cftypes(struct cftype *cfts, bool is_add) 2477 static int cgroup_apply_cftypes(struct cftype *cfts, bool is_add)
2433 { 2478 {
2434 LIST_HEAD(pending); 2479 LIST_HEAD(pending);
2435 struct cgroup_subsys *ss = cfts[0].ss; 2480 struct cgroup_subsys *ss = cfts[0].ss;
2436 struct cgroup *root = &ss->root->cgrp; 2481 struct cgroup *root = &ss->root->cgrp;
2437 struct cgroup_subsys_state *css; 2482 struct cgroup_subsys_state *css;
2438 int ret = 0; 2483 int ret = 0;
2439 2484
2440 lockdep_assert_held(&cgroup_tree_mutex); 2485 lockdep_assert_held(&cgroup_tree_mutex);
2441 2486
2442 /* add/rm files for all cgroups created before */ 2487 /* add/rm files for all cgroups created before */
2443 css_for_each_descendant_pre(css, cgroup_css(root, ss)) { 2488 css_for_each_descendant_pre(css, cgroup_css(root, ss)) {
2444 struct cgroup *cgrp = css->cgroup; 2489 struct cgroup *cgrp = css->cgroup;
2445 2490
2446 if (cgroup_is_dead(cgrp)) 2491 if (cgroup_is_dead(cgrp))
2447 continue; 2492 continue;
2448 2493
2449 ret = cgroup_addrm_files(cgrp, cfts, is_add); 2494 ret = cgroup_addrm_files(cgrp, cfts, is_add);
2450 if (ret) 2495 if (ret)
2451 break; 2496 break;
2452 } 2497 }
2453 2498
2454 if (is_add && !ret) 2499 if (is_add && !ret)
2455 kernfs_activate(root->kn); 2500 kernfs_activate(root->kn);
2456 return ret; 2501 return ret;
2457 } 2502 }
2458 2503
2459 static void cgroup_exit_cftypes(struct cftype *cfts) 2504 static void cgroup_exit_cftypes(struct cftype *cfts)
2460 { 2505 {
2461 struct cftype *cft; 2506 struct cftype *cft;
2462 2507
2463 for (cft = cfts; cft->name[0] != '\0'; cft++) { 2508 for (cft = cfts; cft->name[0] != '\0'; cft++) {
2464 /* free copy for custom atomic_write_len, see init_cftypes() */ 2509 /* free copy for custom atomic_write_len, see init_cftypes() */
2465 if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) 2510 if (cft->max_write_len && cft->max_write_len != PAGE_SIZE)
2466 kfree(cft->kf_ops); 2511 kfree(cft->kf_ops);
2467 cft->kf_ops = NULL; 2512 cft->kf_ops = NULL;
2468 cft->ss = NULL; 2513 cft->ss = NULL;
2469 } 2514 }
2470 } 2515 }
2471 2516
2472 static int cgroup_init_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) 2517 static int cgroup_init_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
2473 { 2518 {
2474 struct cftype *cft; 2519 struct cftype *cft;
2475 2520
2476 for (cft = cfts; cft->name[0] != '\0'; cft++) { 2521 for (cft = cfts; cft->name[0] != '\0'; cft++) {
2477 struct kernfs_ops *kf_ops; 2522 struct kernfs_ops *kf_ops;
2478 2523
2479 WARN_ON(cft->ss || cft->kf_ops); 2524 WARN_ON(cft->ss || cft->kf_ops);
2480 2525
2481 if (cft->seq_start) 2526 if (cft->seq_start)
2482 kf_ops = &cgroup_kf_ops; 2527 kf_ops = &cgroup_kf_ops;
2483 else 2528 else
2484 kf_ops = &cgroup_kf_single_ops; 2529 kf_ops = &cgroup_kf_single_ops;
2485 2530
2486 /* 2531 /*
2487 * Ugh... if @cft wants a custom max_write_len, we need to 2532 * Ugh... if @cft wants a custom max_write_len, we need to
2488 * make a copy of kf_ops to set its atomic_write_len. 2533 * make a copy of kf_ops to set its atomic_write_len.
2489 */ 2534 */
2490 if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) { 2535 if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) {
2491 kf_ops = kmemdup(kf_ops, sizeof(*kf_ops), GFP_KERNEL); 2536 kf_ops = kmemdup(kf_ops, sizeof(*kf_ops), GFP_KERNEL);
2492 if (!kf_ops) { 2537 if (!kf_ops) {
2493 cgroup_exit_cftypes(cfts); 2538 cgroup_exit_cftypes(cfts);
2494 return -ENOMEM; 2539 return -ENOMEM;
2495 } 2540 }
2496 kf_ops->atomic_write_len = cft->max_write_len; 2541 kf_ops->atomic_write_len = cft->max_write_len;
2497 } 2542 }
2498 2543
2499 cft->kf_ops = kf_ops; 2544 cft->kf_ops = kf_ops;
2500 cft->ss = ss; 2545 cft->ss = ss;
2501 } 2546 }
2502 2547
2503 return 0; 2548 return 0;
2504 } 2549 }
2505 2550
2506 static int cgroup_rm_cftypes_locked(struct cftype *cfts) 2551 static int cgroup_rm_cftypes_locked(struct cftype *cfts)
2507 { 2552 {
2508 lockdep_assert_held(&cgroup_tree_mutex); 2553 lockdep_assert_held(&cgroup_tree_mutex);
2509 2554
2510 if (!cfts || !cfts[0].ss) 2555 if (!cfts || !cfts[0].ss)
2511 return -ENOENT; 2556 return -ENOENT;
2512 2557
2513 list_del(&cfts->node); 2558 list_del(&cfts->node);
2514 cgroup_apply_cftypes(cfts, false); 2559 cgroup_apply_cftypes(cfts, false);
2515 cgroup_exit_cftypes(cfts); 2560 cgroup_exit_cftypes(cfts);
2516 return 0; 2561 return 0;
2517 } 2562 }
2518 2563
2519 /** 2564 /**
2520 * cgroup_rm_cftypes - remove an array of cftypes from a subsystem 2565 * cgroup_rm_cftypes - remove an array of cftypes from a subsystem
2521 * @cfts: zero-length name terminated array of cftypes 2566 * @cfts: zero-length name terminated array of cftypes
2522 * 2567 *
2523 * Unregister @cfts. Files described by @cfts are removed from all 2568 * Unregister @cfts. Files described by @cfts are removed from all
2524 * existing cgroups and all future cgroups won't have them either. This 2569 * existing cgroups and all future cgroups won't have them either. This
2525 * function can be called anytime whether @cfts' subsys is attached or not. 2570 * function can be called anytime whether @cfts' subsys is attached or not.
2526 * 2571 *
2527 * Returns 0 on successful unregistration, -ENOENT if @cfts is not 2572 * Returns 0 on successful unregistration, -ENOENT if @cfts is not
2528 * registered. 2573 * registered.
2529 */ 2574 */
2530 int cgroup_rm_cftypes(struct cftype *cfts) 2575 int cgroup_rm_cftypes(struct cftype *cfts)
2531 { 2576 {
2532 int ret; 2577 int ret;
2533 2578
2534 mutex_lock(&cgroup_tree_mutex); 2579 mutex_lock(&cgroup_tree_mutex);
2535 ret = cgroup_rm_cftypes_locked(cfts); 2580 ret = cgroup_rm_cftypes_locked(cfts);
2536 mutex_unlock(&cgroup_tree_mutex); 2581 mutex_unlock(&cgroup_tree_mutex);
2537 return ret; 2582 return ret;
2538 } 2583 }
2539 2584
2540 /** 2585 /**
2541 * cgroup_add_cftypes - add an array of cftypes to a subsystem 2586 * cgroup_add_cftypes - add an array of cftypes to a subsystem
2542 * @ss: target cgroup subsystem 2587 * @ss: target cgroup subsystem
2543 * @cfts: zero-length name terminated array of cftypes 2588 * @cfts: zero-length name terminated array of cftypes
2544 * 2589 *
2545 * Register @cfts to @ss. Files described by @cfts are created for all 2590 * Register @cfts to @ss. Files described by @cfts are created for all
2546 * existing cgroups to which @ss is attached and all future cgroups will 2591 * existing cgroups to which @ss is attached and all future cgroups will
2547 * have them too. This function can be called anytime whether @ss is 2592 * have them too. This function can be called anytime whether @ss is
2548 * attached or not. 2593 * attached or not.
2549 * 2594 *
2550 * Returns 0 on successful registration, -errno on failure. Note that this 2595 * Returns 0 on successful registration, -errno on failure. Note that this
2551 * function currently returns 0 as long as @cfts registration is successful 2596 * function currently returns 0 as long as @cfts registration is successful
2552 * even if some file creation attempts on existing cgroups fail. 2597 * even if some file creation attempts on existing cgroups fail.
2553 */ 2598 */
2554 int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) 2599 int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
2555 { 2600 {
2556 int ret; 2601 int ret;
2557 2602
2558 if (!cfts || cfts[0].name[0] == '\0') 2603 if (!cfts || cfts[0].name[0] == '\0')
2559 return 0; 2604 return 0;
2560 2605
2561 ret = cgroup_init_cftypes(ss, cfts); 2606 ret = cgroup_init_cftypes(ss, cfts);
2562 if (ret) 2607 if (ret)
2563 return ret; 2608 return ret;
2564 2609
2565 mutex_lock(&cgroup_tree_mutex); 2610 mutex_lock(&cgroup_tree_mutex);
2566 2611
2567 list_add_tail(&cfts->node, &ss->cfts); 2612 list_add_tail(&cfts->node, &ss->cfts);
2568 ret = cgroup_apply_cftypes(cfts, true); 2613 ret = cgroup_apply_cftypes(cfts, true);
2569 if (ret) 2614 if (ret)
2570 cgroup_rm_cftypes_locked(cfts); 2615 cgroup_rm_cftypes_locked(cfts);
2571 2616
2572 mutex_unlock(&cgroup_tree_mutex); 2617 mutex_unlock(&cgroup_tree_mutex);
2573 return ret; 2618 return ret;
2574 } 2619 }
2575 2620
2576 /** 2621 /**
2577 * cgroup_task_count - count the number of tasks in a cgroup. 2622 * cgroup_task_count - count the number of tasks in a cgroup.
2578 * @cgrp: the cgroup in question 2623 * @cgrp: the cgroup in question
2579 * 2624 *
2580 * Return the number of tasks in the cgroup. 2625 * Return the number of tasks in the cgroup.
2581 */ 2626 */
2582 static int cgroup_task_count(const struct cgroup *cgrp) 2627 static int cgroup_task_count(const struct cgroup *cgrp)
2583 { 2628 {
2584 int count = 0; 2629 int count = 0;
2585 struct cgrp_cset_link *link; 2630 struct cgrp_cset_link *link;
2586 2631
2587 down_read(&css_set_rwsem); 2632 down_read(&css_set_rwsem);
2588 list_for_each_entry(link, &cgrp->cset_links, cset_link) 2633 list_for_each_entry(link, &cgrp->cset_links, cset_link)
2589 count += atomic_read(&link->cset->refcount); 2634 count += atomic_read(&link->cset->refcount);
2590 up_read(&css_set_rwsem); 2635 up_read(&css_set_rwsem);
2591 return count; 2636 return count;
2592 } 2637 }
2593 2638
2594 /** 2639 /**
2595 * css_next_child - find the next child of a given css 2640 * css_next_child - find the next child of a given css
2596 * @pos_css: the current position (%NULL to initiate traversal) 2641 * @pos_css: the current position (%NULL to initiate traversal)
2597 * @parent_css: css whose children to walk 2642 * @parent_css: css whose children to walk
2598 * 2643 *
2599 * This function returns the next child of @parent_css and should be called 2644 * This function returns the next child of @parent_css and should be called
2600 * under either cgroup_mutex or RCU read lock. The only requirement is 2645 * under either cgroup_mutex or RCU read lock. The only requirement is
2601 * that @parent_css and @pos_css are accessible. The next sibling is 2646 * that @parent_css and @pos_css are accessible. The next sibling is
2602 * guaranteed to be returned regardless of their states. 2647 * guaranteed to be returned regardless of their states.
2603 */ 2648 */
2604 struct cgroup_subsys_state * 2649 struct cgroup_subsys_state *
2605 css_next_child(struct cgroup_subsys_state *pos_css, 2650 css_next_child(struct cgroup_subsys_state *pos_css,
2606 struct cgroup_subsys_state *parent_css) 2651 struct cgroup_subsys_state *parent_css)
2607 { 2652 {
2608 struct cgroup *pos = pos_css ? pos_css->cgroup : NULL; 2653 struct cgroup *pos = pos_css ? pos_css->cgroup : NULL;
2609 struct cgroup *cgrp = parent_css->cgroup; 2654 struct cgroup *cgrp = parent_css->cgroup;
2610 struct cgroup *next; 2655 struct cgroup *next;
2611 2656
2612 cgroup_assert_mutexes_or_rcu_locked(); 2657 cgroup_assert_mutexes_or_rcu_locked();
2613 2658
2614 /* 2659 /*
2615 * @pos could already have been removed. Once a cgroup is removed, 2660 * @pos could already have been removed. Once a cgroup is removed,
2616 * its ->sibling.next is no longer updated when its next sibling 2661 * its ->sibling.next is no longer updated when its next sibling
2617 * changes. As CGRP_DEAD assertion is serialized and happens 2662 * changes. As CGRP_DEAD assertion is serialized and happens
2618 * before the cgroup is taken off the ->sibling list, if we see it 2663 * before the cgroup is taken off the ->sibling list, if we see it
2619 * unasserted, it's guaranteed that the next sibling hasn't 2664 * unasserted, it's guaranteed that the next sibling hasn't
2620 * finished its grace period even if it's already removed, and thus 2665 * finished its grace period even if it's already removed, and thus
2621 * safe to dereference from this RCU critical section. If 2666 * safe to dereference from this RCU critical section. If
2622 * ->sibling.next is inaccessible, cgroup_is_dead() is guaranteed 2667 * ->sibling.next is inaccessible, cgroup_is_dead() is guaranteed
2623 * to be visible as %true here. 2668 * to be visible as %true here.
2624 * 2669 *
2625 * If @pos is dead, its next pointer can't be dereferenced; 2670 * If @pos is dead, its next pointer can't be dereferenced;
2626 * however, as each cgroup is given a monotonically increasing 2671 * however, as each cgroup is given a monotonically increasing
2627 * unique serial number and always appended to the sibling list, 2672 * unique serial number and always appended to the sibling list,
2628 * the next one can be found by walking the parent's children until 2673 * the next one can be found by walking the parent's children until
2629 * we see a cgroup with higher serial number than @pos's. While 2674 * we see a cgroup with higher serial number than @pos's. While
2630 * this path can be slower, it's taken only when either the current 2675 * this path can be slower, it's taken only when either the current
2631 * cgroup is removed or iteration and removal race. 2676 * cgroup is removed or iteration and removal race.
2632 */ 2677 */
2633 if (!pos) { 2678 if (!pos) {
2634 next = list_entry_rcu(cgrp->children.next, struct cgroup, sibling); 2679 next = list_entry_rcu(cgrp->children.next, struct cgroup, sibling);
2635 } else if (likely(!cgroup_is_dead(pos))) { 2680 } else if (likely(!cgroup_is_dead(pos))) {
2636 next = list_entry_rcu(pos->sibling.next, struct cgroup, sibling); 2681 next = list_entry_rcu(pos->sibling.next, struct cgroup, sibling);
2637 } else { 2682 } else {
2638 list_for_each_entry_rcu(next, &cgrp->children, sibling) 2683 list_for_each_entry_rcu(next, &cgrp->children, sibling)
2639 if (next->serial_nr > pos->serial_nr) 2684 if (next->serial_nr > pos->serial_nr)
2640 break; 2685 break;
2641 } 2686 }
2642 2687
2643 if (&next->sibling == &cgrp->children) 2688 if (&next->sibling == &cgrp->children)
2644 return NULL; 2689 return NULL;
2645 2690
2646 return cgroup_css(next, parent_css->ss); 2691 return cgroup_css(next, parent_css->ss);
2647 } 2692 }
2648 2693
2649 /** 2694 /**
2650 * css_next_descendant_pre - find the next descendant for pre-order walk 2695 * css_next_descendant_pre - find the next descendant for pre-order walk
2651 * @pos: the current position (%NULL to initiate traversal) 2696 * @pos: the current position (%NULL to initiate traversal)
2652 * @root: css whose descendants to walk 2697 * @root: css whose descendants to walk
2653 * 2698 *
2654 * To be used by css_for_each_descendant_pre(). Find the next descendant 2699 * To be used by css_for_each_descendant_pre(). Find the next descendant
2655 * to visit for pre-order traversal of @root's descendants. @root is 2700 * to visit for pre-order traversal of @root's descendants. @root is
2656 * included in the iteration and the first node to be visited. 2701 * included in the iteration and the first node to be visited.
2657 * 2702 *
2658 * While this function requires cgroup_mutex or RCU read locking, it 2703 * While this function requires cgroup_mutex or RCU read locking, it
2659 * doesn't require the whole traversal to be contained in a single critical 2704 * doesn't require the whole traversal to be contained in a single critical
2660 * section. This function will return the correct next descendant as long 2705 * section. This function will return the correct next descendant as long
2661 * as both @pos and @root are accessible and @pos is a descendant of @root. 2706 * as both @pos and @root are accessible and @pos is a descendant of @root.
2662 */ 2707 */
2663 struct cgroup_subsys_state * 2708 struct cgroup_subsys_state *
2664 css_next_descendant_pre(struct cgroup_subsys_state *pos, 2709 css_next_descendant_pre(struct cgroup_subsys_state *pos,
2665 struct cgroup_subsys_state *root) 2710 struct cgroup_subsys_state *root)
2666 { 2711 {
2667 struct cgroup_subsys_state *next; 2712 struct cgroup_subsys_state *next;
2668 2713
2669 cgroup_assert_mutexes_or_rcu_locked(); 2714 cgroup_assert_mutexes_or_rcu_locked();
2670 2715
2671 /* if first iteration, visit @root */ 2716 /* if first iteration, visit @root */
2672 if (!pos) 2717 if (!pos)
2673 return root; 2718 return root;
2674 2719
2675 /* visit the first child if exists */ 2720 /* visit the first child if exists */
2676 next = css_next_child(NULL, pos); 2721 next = css_next_child(NULL, pos);
2677 if (next) 2722 if (next)
2678 return next; 2723 return next;
2679 2724
2680 /* no child, visit my or the closest ancestor's next sibling */ 2725 /* no child, visit my or the closest ancestor's next sibling */
2681 while (pos != root) { 2726 while (pos != root) {
2682 next = css_next_child(pos, css_parent(pos)); 2727 next = css_next_child(pos, css_parent(pos));
2683 if (next) 2728 if (next)
2684 return next; 2729 return next;
2685 pos = css_parent(pos); 2730 pos = css_parent(pos);
2686 } 2731 }
2687 2732
2688 return NULL; 2733 return NULL;
2689 } 2734 }
2690 2735
2691 /** 2736 /**
2692 * css_rightmost_descendant - return the rightmost descendant of a css 2737 * css_rightmost_descendant - return the rightmost descendant of a css
2693 * @pos: css of interest 2738 * @pos: css of interest
2694 * 2739 *
2695 * Return the rightmost descendant of @pos. If there's no descendant, @pos 2740 * Return the rightmost descendant of @pos. If there's no descendant, @pos
2696 * is returned. This can be used during pre-order traversal to skip 2741 * is returned. This can be used during pre-order traversal to skip
2697 * subtree of @pos. 2742 * subtree of @pos.
2698 * 2743 *
2699 * While this function requires cgroup_mutex or RCU read locking, it 2744 * While this function requires cgroup_mutex or RCU read locking, it
2700 * doesn't require the whole traversal to be contained in a single critical 2745 * doesn't require the whole traversal to be contained in a single critical
2701 * section. This function will return the correct rightmost descendant as 2746 * section. This function will return the correct rightmost descendant as
2702 * long as @pos is accessible. 2747 * long as @pos is accessible.
2703 */ 2748 */
2704 struct cgroup_subsys_state * 2749 struct cgroup_subsys_state *
2705 css_rightmost_descendant(struct cgroup_subsys_state *pos) 2750 css_rightmost_descendant(struct cgroup_subsys_state *pos)
2706 { 2751 {
2707 struct cgroup_subsys_state *last, *tmp; 2752 struct cgroup_subsys_state *last, *tmp;
2708 2753
2709 cgroup_assert_mutexes_or_rcu_locked(); 2754 cgroup_assert_mutexes_or_rcu_locked();
2710 2755
2711 do { 2756 do {
2712 last = pos; 2757 last = pos;
2713 /* ->prev isn't RCU safe, walk ->next till the end */ 2758 /* ->prev isn't RCU safe, walk ->next till the end */
2714 pos = NULL; 2759 pos = NULL;
2715 css_for_each_child(tmp, last) 2760 css_for_each_child(tmp, last)
2716 pos = tmp; 2761 pos = tmp;
2717 } while (pos); 2762 } while (pos);
2718 2763
2719 return last; 2764 return last;
2720 } 2765 }
2721 2766
2722 static struct cgroup_subsys_state * 2767 static struct cgroup_subsys_state *
2723 css_leftmost_descendant(struct cgroup_subsys_state *pos) 2768 css_leftmost_descendant(struct cgroup_subsys_state *pos)
2724 { 2769 {
2725 struct cgroup_subsys_state *last; 2770 struct cgroup_subsys_state *last;
2726 2771
2727 do { 2772 do {
2728 last = pos; 2773 last = pos;
2729 pos = css_next_child(NULL, pos); 2774 pos = css_next_child(NULL, pos);
2730 } while (pos); 2775 } while (pos);
2731 2776
2732 return last; 2777 return last;
2733 } 2778 }
2734 2779
2735 /** 2780 /**
2736 * css_next_descendant_post - find the next descendant for post-order walk 2781 * css_next_descendant_post - find the next descendant for post-order walk
2737 * @pos: the current position (%NULL to initiate traversal) 2782 * @pos: the current position (%NULL to initiate traversal)
2738 * @root: css whose descendants to walk 2783 * @root: css whose descendants to walk
2739 * 2784 *
2740 * To be used by css_for_each_descendant_post(). Find the next descendant 2785 * To be used by css_for_each_descendant_post(). Find the next descendant
2741 * to visit for post-order traversal of @root's descendants. @root is 2786 * to visit for post-order traversal of @root's descendants. @root is
2742 * included in the iteration and the last node to be visited. 2787 * included in the iteration and the last node to be visited.
2743 * 2788 *
2744 * While this function requires cgroup_mutex or RCU read locking, it 2789 * While this function requires cgroup_mutex or RCU read locking, it
2745 * doesn't require the whole traversal to be contained in a single critical 2790 * doesn't require the whole traversal to be contained in a single critical
2746 * section. This function will return the correct next descendant as long 2791 * section. This function will return the correct next descendant as long
2747 * as both @pos and @cgroup are accessible and @pos is a descendant of 2792 * as both @pos and @cgroup are accessible and @pos is a descendant of
2748 * @cgroup. 2793 * @cgroup.
2749 */ 2794 */
2750 struct cgroup_subsys_state * 2795 struct cgroup_subsys_state *
2751 css_next_descendant_post(struct cgroup_subsys_state *pos, 2796 css_next_descendant_post(struct cgroup_subsys_state *pos,
2752 struct cgroup_subsys_state *root) 2797 struct cgroup_subsys_state *root)
2753 { 2798 {
2754 struct cgroup_subsys_state *next; 2799 struct cgroup_subsys_state *next;
2755 2800
2756 cgroup_assert_mutexes_or_rcu_locked(); 2801 cgroup_assert_mutexes_or_rcu_locked();
2757 2802
2758 /* if first iteration, visit leftmost descendant which may be @root */ 2803 /* if first iteration, visit leftmost descendant which may be @root */
2759 if (!pos) 2804 if (!pos)
2760 return css_leftmost_descendant(root); 2805 return css_leftmost_descendant(root);
2761 2806
2762 /* if we visited @root, we're done */ 2807 /* if we visited @root, we're done */
2763 if (pos == root) 2808 if (pos == root)
2764 return NULL; 2809 return NULL;
2765 2810
2766 /* if there's an unvisited sibling, visit its leftmost descendant */ 2811 /* if there's an unvisited sibling, visit its leftmost descendant */
2767 next = css_next_child(pos, css_parent(pos)); 2812 next = css_next_child(pos, css_parent(pos));
2768 if (next) 2813 if (next)
2769 return css_leftmost_descendant(next); 2814 return css_leftmost_descendant(next);
2770 2815
2771 /* no sibling left, visit parent */ 2816 /* no sibling left, visit parent */
2772 return css_parent(pos); 2817 return css_parent(pos);
2773 } 2818 }
2774 2819
2775 /** 2820 /**
2776 * css_advance_task_iter - advance a task itererator to the next css_set 2821 * css_advance_task_iter - advance a task itererator to the next css_set
2777 * @it: the iterator to advance 2822 * @it: the iterator to advance
2778 * 2823 *
2779 * Advance @it to the next css_set to walk. 2824 * Advance @it to the next css_set to walk.
2780 */ 2825 */
2781 static void css_advance_task_iter(struct css_task_iter *it) 2826 static void css_advance_task_iter(struct css_task_iter *it)
2782 { 2827 {
2783 struct list_head *l = it->cset_link; 2828 struct list_head *l = it->cset_link;
2784 struct cgrp_cset_link *link; 2829 struct cgrp_cset_link *link;
2785 struct css_set *cset; 2830 struct css_set *cset;
2786 2831
2787 /* Advance to the next non-empty css_set */ 2832 /* Advance to the next non-empty css_set */
2788 do { 2833 do {
2789 l = l->next; 2834 l = l->next;
2790 if (l == &it->origin_css->cgroup->cset_links) { 2835 if (l == &it->origin_css->cgroup->cset_links) {
2791 it->cset_link = NULL; 2836 it->cset_link = NULL;
2792 return; 2837 return;
2793 } 2838 }
2794 link = list_entry(l, struct cgrp_cset_link, cset_link); 2839 link = list_entry(l, struct cgrp_cset_link, cset_link);
2795 cset = link->cset; 2840 cset = link->cset;
2796 } while (list_empty(&cset->tasks) && list_empty(&cset->mg_tasks)); 2841 } while (list_empty(&cset->tasks) && list_empty(&cset->mg_tasks));
2797 2842
2798 it->cset_link = l; 2843 it->cset_link = l;
2799 2844
2800 if (!list_empty(&cset->tasks)) 2845 if (!list_empty(&cset->tasks))
2801 it->task = cset->tasks.next; 2846 it->task = cset->tasks.next;
2802 else 2847 else
2803 it->task = cset->mg_tasks.next; 2848 it->task = cset->mg_tasks.next;
2804 } 2849 }
2805 2850
2806 /** 2851 /**
2807 * css_task_iter_start - initiate task iteration 2852 * css_task_iter_start - initiate task iteration
2808 * @css: the css to walk tasks of 2853 * @css: the css to walk tasks of
2809 * @it: the task iterator to use 2854 * @it: the task iterator to use
2810 * 2855 *
2811 * Initiate iteration through the tasks of @css. The caller can call 2856 * Initiate iteration through the tasks of @css. The caller can call
2812 * css_task_iter_next() to walk through the tasks until the function 2857 * css_task_iter_next() to walk through the tasks until the function
2813 * returns NULL. On completion of iteration, css_task_iter_end() must be 2858 * returns NULL. On completion of iteration, css_task_iter_end() must be
2814 * called. 2859 * called.
2815 * 2860 *
2816 * Note that this function acquires a lock which is released when the 2861 * Note that this function acquires a lock which is released when the
2817 * iteration finishes. The caller can't sleep while iteration is in 2862 * iteration finishes. The caller can't sleep while iteration is in
2818 * progress. 2863 * progress.
2819 */ 2864 */
2820 void css_task_iter_start(struct cgroup_subsys_state *css, 2865 void css_task_iter_start(struct cgroup_subsys_state *css,
2821 struct css_task_iter *it) 2866 struct css_task_iter *it)
2822 __acquires(css_set_rwsem) 2867 __acquires(css_set_rwsem)
2823 { 2868 {
2824 /* no one should try to iterate before mounting cgroups */ 2869 /* no one should try to iterate before mounting cgroups */
2825 WARN_ON_ONCE(!use_task_css_set_links); 2870 WARN_ON_ONCE(!use_task_css_set_links);
2826 2871
2827 down_read(&css_set_rwsem); 2872 down_read(&css_set_rwsem);
2828 2873
2829 it->origin_css = css; 2874 it->origin_css = css;
2830 it->cset_link = &css->cgroup->cset_links; 2875 it->cset_link = &css->cgroup->cset_links;
2831 2876
2832 css_advance_task_iter(it); 2877 css_advance_task_iter(it);
2833 } 2878 }
2834 2879
2835 /** 2880 /**
2836 * css_task_iter_next - return the next task for the iterator 2881 * css_task_iter_next - return the next task for the iterator
2837 * @it: the task iterator being iterated 2882 * @it: the task iterator being iterated
2838 * 2883 *
2839 * The "next" function for task iteration. @it should have been 2884 * The "next" function for task iteration. @it should have been
2840 * initialized via css_task_iter_start(). Returns NULL when the iteration 2885 * initialized via css_task_iter_start(). Returns NULL when the iteration
2841 * reaches the end. 2886 * reaches the end.
2842 */ 2887 */
2843 struct task_struct *css_task_iter_next(struct css_task_iter *it) 2888 struct task_struct *css_task_iter_next(struct css_task_iter *it)
2844 { 2889 {
2845 struct task_struct *res; 2890 struct task_struct *res;
2846 struct list_head *l = it->task; 2891 struct list_head *l = it->task;
2847 struct cgrp_cset_link *link = list_entry(it->cset_link, 2892 struct cgrp_cset_link *link = list_entry(it->cset_link,
2848 struct cgrp_cset_link, cset_link); 2893 struct cgrp_cset_link, cset_link);
2849 2894
2850 /* If the iterator cg is NULL, we have no tasks */ 2895 /* If the iterator cg is NULL, we have no tasks */
2851 if (!it->cset_link) 2896 if (!it->cset_link)
2852 return NULL; 2897 return NULL;
2853 res = list_entry(l, struct task_struct, cg_list); 2898 res = list_entry(l, struct task_struct, cg_list);
2854 2899
2855 /* 2900 /*
2856 * Advance iterator to find next entry. cset->tasks is consumed 2901 * Advance iterator to find next entry. cset->tasks is consumed
2857 * first and then ->mg_tasks. After ->mg_tasks, we move onto the 2902 * first and then ->mg_tasks. After ->mg_tasks, we move onto the
2858 * next cset. 2903 * next cset.
2859 */ 2904 */
2860 l = l->next; 2905 l = l->next;
2861 2906
2862 if (l == &link->cset->tasks) 2907 if (l == &link->cset->tasks)
2863 l = link->cset->mg_tasks.next; 2908 l = link->cset->mg_tasks.next;
2864 2909
2865 if (l == &link->cset->mg_tasks) 2910 if (l == &link->cset->mg_tasks)
2866 css_advance_task_iter(it); 2911 css_advance_task_iter(it);
2867 else 2912 else
2868 it->task = l; 2913 it->task = l;
2869 2914
2870 return res; 2915 return res;
2871 } 2916 }
2872 2917
2873 /** 2918 /**
2874 * css_task_iter_end - finish task iteration 2919 * css_task_iter_end - finish task iteration
2875 * @it: the task iterator to finish 2920 * @it: the task iterator to finish
2876 * 2921 *
2877 * Finish task iteration started by css_task_iter_start(). 2922 * Finish task iteration started by css_task_iter_start().
2878 */ 2923 */
2879 void css_task_iter_end(struct css_task_iter *it) 2924 void css_task_iter_end(struct css_task_iter *it)
2880 __releases(css_set_rwsem) 2925 __releases(css_set_rwsem)
2881 { 2926 {
2882 up_read(&css_set_rwsem); 2927 up_read(&css_set_rwsem);
2883 } 2928 }
2884 2929
2885 /** 2930 /**
2886 * cgroup_trasnsfer_tasks - move tasks from one cgroup to another 2931 * cgroup_trasnsfer_tasks - move tasks from one cgroup to another
2887 * @to: cgroup to which the tasks will be moved 2932 * @to: cgroup to which the tasks will be moved
2888 * @from: cgroup in which the tasks currently reside 2933 * @from: cgroup in which the tasks currently reside
2889 * 2934 *
2890 * Locking rules between cgroup_post_fork() and the migration path 2935 * Locking rules between cgroup_post_fork() and the migration path
2891 * guarantee that, if a task is forking while being migrated, the new child 2936 * guarantee that, if a task is forking while being migrated, the new child
2892 * is guaranteed to be either visible in the source cgroup after the 2937 * is guaranteed to be either visible in the source cgroup after the
2893 * parent's migration is complete or put into the target cgroup. No task 2938 * parent's migration is complete or put into the target cgroup. No task
2894 * can slip out of migration through forking. 2939 * can slip out of migration through forking.
2895 */ 2940 */
2896 int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from) 2941 int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from)
2897 { 2942 {
2898 LIST_HEAD(preloaded_csets); 2943 LIST_HEAD(preloaded_csets);
2899 struct cgrp_cset_link *link; 2944 struct cgrp_cset_link *link;
2900 struct css_task_iter it; 2945 struct css_task_iter it;
2901 struct task_struct *task; 2946 struct task_struct *task;
2902 int ret; 2947 int ret;
2903 2948
2904 mutex_lock(&cgroup_mutex); 2949 mutex_lock(&cgroup_mutex);
2905 2950
2906 /* all tasks in @from are being moved, all csets are source */ 2951 /* all tasks in @from are being moved, all csets are source */
2907 down_read(&css_set_rwsem); 2952 down_read(&css_set_rwsem);
2908 list_for_each_entry(link, &from->cset_links, cset_link) 2953 list_for_each_entry(link, &from->cset_links, cset_link)
2909 cgroup_migrate_add_src(link->cset, to, &preloaded_csets); 2954 cgroup_migrate_add_src(link->cset, to, &preloaded_csets);
2910 up_read(&css_set_rwsem); 2955 up_read(&css_set_rwsem);
2911 2956
2912 ret = cgroup_migrate_prepare_dst(to, &preloaded_csets); 2957 ret = cgroup_migrate_prepare_dst(to, &preloaded_csets);
2913 if (ret) 2958 if (ret)
2914 goto out_err; 2959 goto out_err;
2915 2960
2916 /* 2961 /*
2917 * Migrate tasks one-by-one until @form is empty. This fails iff 2962 * Migrate tasks one-by-one until @form is empty. This fails iff
2918 * ->can_attach() fails. 2963 * ->can_attach() fails.
2919 */ 2964 */
2920 do { 2965 do {
2921 css_task_iter_start(&from->dummy_css, &it); 2966 css_task_iter_start(&from->dummy_css, &it);
2922 task = css_task_iter_next(&it); 2967 task = css_task_iter_next(&it);
2923 if (task) 2968 if (task)
2924 get_task_struct(task); 2969 get_task_struct(task);
2925 css_task_iter_end(&it); 2970 css_task_iter_end(&it);
2926 2971
2927 if (task) { 2972 if (task) {
2928 ret = cgroup_migrate(to, task, false); 2973 ret = cgroup_migrate(to, task, false);
2929 put_task_struct(task); 2974 put_task_struct(task);
2930 } 2975 }
2931 } while (task && !ret); 2976 } while (task && !ret);
2932 out_err: 2977 out_err:
2933 cgroup_migrate_finish(&preloaded_csets); 2978 cgroup_migrate_finish(&preloaded_csets);
2934 mutex_unlock(&cgroup_mutex); 2979 mutex_unlock(&cgroup_mutex);
2935 return ret; 2980 return ret;
2936 } 2981 }
2937 2982
2938 /* 2983 /*
2939 * Stuff for reading the 'tasks'/'procs' files. 2984 * Stuff for reading the 'tasks'/'procs' files.
2940 * 2985 *
2941 * Reading this file can return large amounts of data if a cgroup has 2986 * Reading this file can return large amounts of data if a cgroup has
2942 * *lots* of attached tasks. So it may need several calls to read(), 2987 * *lots* of attached tasks. So it may need several calls to read(),
2943 * but we cannot guarantee that the information we produce is correct 2988 * but we cannot guarantee that the information we produce is correct
2944 * unless we produce it entirely atomically. 2989 * unless we produce it entirely atomically.
2945 * 2990 *
2946 */ 2991 */
2947 2992
2948 /* which pidlist file are we talking about? */ 2993 /* which pidlist file are we talking about? */
2949 enum cgroup_filetype { 2994 enum cgroup_filetype {
2950 CGROUP_FILE_PROCS, 2995 CGROUP_FILE_PROCS,
2951 CGROUP_FILE_TASKS, 2996 CGROUP_FILE_TASKS,
2952 }; 2997 };
2953 2998
2954 /* 2999 /*
2955 * A pidlist is a list of pids that virtually represents the contents of one 3000 * A pidlist is a list of pids that virtually represents the contents of one
2956 * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists, 3001 * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists,
2957 * a pair (one each for procs, tasks) for each pid namespace that's relevant 3002 * a pair (one each for procs, tasks) for each pid namespace that's relevant
2958 * to the cgroup. 3003 * to the cgroup.
2959 */ 3004 */
2960 struct cgroup_pidlist { 3005 struct cgroup_pidlist {
2961 /* 3006 /*
2962 * used to find which pidlist is wanted. doesn't change as long as 3007 * used to find which pidlist is wanted. doesn't change as long as
2963 * this particular list stays in the list. 3008 * this particular list stays in the list.
2964 */ 3009 */
2965 struct { enum cgroup_filetype type; struct pid_namespace *ns; } key; 3010 struct { enum cgroup_filetype type; struct pid_namespace *ns; } key;
2966 /* array of xids */ 3011 /* array of xids */
2967 pid_t *list; 3012 pid_t *list;
2968 /* how many elements the above list has */ 3013 /* how many elements the above list has */
2969 int length; 3014 int length;
2970 /* each of these stored in a list by its cgroup */ 3015 /* each of these stored in a list by its cgroup */
2971 struct list_head links; 3016 struct list_head links;
2972 /* pointer to the cgroup we belong to, for list removal purposes */ 3017 /* pointer to the cgroup we belong to, for list removal purposes */
2973 struct cgroup *owner; 3018 struct cgroup *owner;
2974 /* for delayed destruction */ 3019 /* for delayed destruction */
2975 struct delayed_work destroy_dwork; 3020 struct delayed_work destroy_dwork;
2976 }; 3021 };
2977 3022
2978 /* 3023 /*
2979 * The following two functions "fix" the issue where there are more pids 3024 * The following two functions "fix" the issue where there are more pids
2980 * than kmalloc will give memory for; in such cases, we use vmalloc/vfree. 3025 * than kmalloc will give memory for; in such cases, we use vmalloc/vfree.
2981 * TODO: replace with a kernel-wide solution to this problem 3026 * TODO: replace with a kernel-wide solution to this problem
2982 */ 3027 */
2983 #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2)) 3028 #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2))
2984 static void *pidlist_allocate(int count) 3029 static void *pidlist_allocate(int count)
2985 { 3030 {
2986 if (PIDLIST_TOO_LARGE(count)) 3031 if (PIDLIST_TOO_LARGE(count))
2987 return vmalloc(count * sizeof(pid_t)); 3032 return vmalloc(count * sizeof(pid_t));
2988 else 3033 else
2989 return kmalloc(count * sizeof(pid_t), GFP_KERNEL); 3034 return kmalloc(count * sizeof(pid_t), GFP_KERNEL);
2990 } 3035 }
2991 3036
2992 static void pidlist_free(void *p) 3037 static void pidlist_free(void *p)
2993 { 3038 {
2994 if (is_vmalloc_addr(p)) 3039 if (is_vmalloc_addr(p))
2995 vfree(p); 3040 vfree(p);
2996 else 3041 else
2997 kfree(p); 3042 kfree(p);
2998 } 3043 }
2999 3044
3000 /* 3045 /*
3001 * Used to destroy all pidlists lingering waiting for destroy timer. None 3046 * Used to destroy all pidlists lingering waiting for destroy timer. None
3002 * should be left afterwards. 3047 * should be left afterwards.
3003 */ 3048 */
3004 static void cgroup_pidlist_destroy_all(struct cgroup *cgrp) 3049 static void cgroup_pidlist_destroy_all(struct cgroup *cgrp)
3005 { 3050 {
3006 struct cgroup_pidlist *l, *tmp_l; 3051 struct cgroup_pidlist *l, *tmp_l;
3007 3052
3008 mutex_lock(&cgrp->pidlist_mutex); 3053 mutex_lock(&cgrp->pidlist_mutex);
3009 list_for_each_entry_safe(l, tmp_l, &cgrp->pidlists, links) 3054 list_for_each_entry_safe(l, tmp_l, &cgrp->pidlists, links)
3010 mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, 0); 3055 mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, 0);
3011 mutex_unlock(&cgrp->pidlist_mutex); 3056 mutex_unlock(&cgrp->pidlist_mutex);
3012 3057
3013 flush_workqueue(cgroup_pidlist_destroy_wq); 3058 flush_workqueue(cgroup_pidlist_destroy_wq);
3014 BUG_ON(!list_empty(&cgrp->pidlists)); 3059 BUG_ON(!list_empty(&cgrp->pidlists));
3015 } 3060 }
3016 3061
3017 static void cgroup_pidlist_destroy_work_fn(struct work_struct *work) 3062 static void cgroup_pidlist_destroy_work_fn(struct work_struct *work)
3018 { 3063 {
3019 struct delayed_work *dwork = to_delayed_work(work); 3064 struct delayed_work *dwork = to_delayed_work(work);
3020 struct cgroup_pidlist *l = container_of(dwork, struct cgroup_pidlist, 3065 struct cgroup_pidlist *l = container_of(dwork, struct cgroup_pidlist,
3021 destroy_dwork); 3066 destroy_dwork);
3022 struct cgroup_pidlist *tofree = NULL; 3067 struct cgroup_pidlist *tofree = NULL;
3023 3068
3024 mutex_lock(&l->owner->pidlist_mutex); 3069 mutex_lock(&l->owner->pidlist_mutex);
3025 3070
3026 /* 3071 /*
3027 * Destroy iff we didn't get queued again. The state won't change 3072 * Destroy iff we didn't get queued again. The state won't change
3028 * as destroy_dwork can only be queued while locked. 3073 * as destroy_dwork can only be queued while locked.
3029 */ 3074 */
3030 if (!delayed_work_pending(dwork)) { 3075 if (!delayed_work_pending(dwork)) {
3031 list_del(&l->links); 3076 list_del(&l->links);
3032 pidlist_free(l->list); 3077 pidlist_free(l->list);
3033 put_pid_ns(l->key.ns); 3078 put_pid_ns(l->key.ns);
3034 tofree = l; 3079 tofree = l;
3035 } 3080 }
3036 3081
3037 mutex_unlock(&l->owner->pidlist_mutex); 3082 mutex_unlock(&l->owner->pidlist_mutex);
3038 kfree(tofree); 3083 kfree(tofree);
3039 } 3084 }
3040 3085
3041 /* 3086 /*
3042 * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries 3087 * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
3043 * Returns the number of unique elements. 3088 * Returns the number of unique elements.
3044 */ 3089 */
3045 static int pidlist_uniq(pid_t *list, int length) 3090 static int pidlist_uniq(pid_t *list, int length)
3046 { 3091 {
3047 int src, dest = 1; 3092 int src, dest = 1;
3048 3093
3049 /* 3094 /*
3050 * we presume the 0th element is unique, so i starts at 1. trivial 3095 * we presume the 0th element is unique, so i starts at 1. trivial
3051 * edge cases first; no work needs to be done for either 3096 * edge cases first; no work needs to be done for either
3052 */ 3097 */
3053 if (length == 0 || length == 1) 3098 if (length == 0 || length == 1)
3054 return length; 3099 return length;
3055 /* src and dest walk down the list; dest counts unique elements */ 3100 /* src and dest walk down the list; dest counts unique elements */
3056 for (src = 1; src < length; src++) { 3101 for (src = 1; src < length; src++) {
3057 /* find next unique element */ 3102 /* find next unique element */
3058 while (list[src] == list[src-1]) { 3103 while (list[src] == list[src-1]) {
3059 src++; 3104 src++;
3060 if (src == length) 3105 if (src == length)
3061 goto after; 3106 goto after;
3062 } 3107 }
3063 /* dest always points to where the next unique element goes */ 3108 /* dest always points to where the next unique element goes */
3064 list[dest] = list[src]; 3109 list[dest] = list[src];
3065 dest++; 3110 dest++;
3066 } 3111 }
3067 after: 3112 after:
3068 return dest; 3113 return dest;
3069 } 3114 }
3070 3115
3071 /* 3116 /*
3072 * The two pid files - task and cgroup.procs - guaranteed that the result 3117 * The two pid files - task and cgroup.procs - guaranteed that the result
3073 * is sorted, which forced this whole pidlist fiasco. As pid order is 3118 * is sorted, which forced this whole pidlist fiasco. As pid order is
3074 * different per namespace, each namespace needs differently sorted list, 3119 * different per namespace, each namespace needs differently sorted list,
3075 * making it impossible to use, for example, single rbtree of member tasks 3120 * making it impossible to use, for example, single rbtree of member tasks
3076 * sorted by task pointer. As pidlists can be fairly large, allocating one 3121 * sorted by task pointer. As pidlists can be fairly large, allocating one
3077 * per open file is dangerous, so cgroup had to implement shared pool of 3122 * per open file is dangerous, so cgroup had to implement shared pool of
3078 * pidlists keyed by cgroup and namespace. 3123 * pidlists keyed by cgroup and namespace.
3079 * 3124 *
3080 * All this extra complexity was caused by the original implementation 3125 * All this extra complexity was caused by the original implementation
3081 * committing to an entirely unnecessary property. In the long term, we 3126 * committing to an entirely unnecessary property. In the long term, we
3082 * want to do away with it. Explicitly scramble sort order if 3127 * want to do away with it. Explicitly scramble sort order if
3083 * sane_behavior so that no such expectation exists in the new interface. 3128 * sane_behavior so that no such expectation exists in the new interface.
3084 * 3129 *
3085 * Scrambling is done by swapping every two consecutive bits, which is 3130 * Scrambling is done by swapping every two consecutive bits, which is
3086 * non-identity one-to-one mapping which disturbs sort order sufficiently. 3131 * non-identity one-to-one mapping which disturbs sort order sufficiently.
3087 */ 3132 */
3088 static pid_t pid_fry(pid_t pid) 3133 static pid_t pid_fry(pid_t pid)
3089 { 3134 {
3090 unsigned a = pid & 0x55555555; 3135 unsigned a = pid & 0x55555555;
3091 unsigned b = pid & 0xAAAAAAAA; 3136 unsigned b = pid & 0xAAAAAAAA;
3092 3137
3093 return (a << 1) | (b >> 1); 3138 return (a << 1) | (b >> 1);
3094 } 3139 }
3095 3140
3096 static pid_t cgroup_pid_fry(struct cgroup *cgrp, pid_t pid) 3141 static pid_t cgroup_pid_fry(struct cgroup *cgrp, pid_t pid)
3097 { 3142 {
3098 if (cgroup_sane_behavior(cgrp)) 3143 if (cgroup_sane_behavior(cgrp))
3099 return pid_fry(pid); 3144 return pid_fry(pid);
3100 else 3145 else
3101 return pid; 3146 return pid;
3102 } 3147 }
3103 3148
3104 static int cmppid(const void *a, const void *b) 3149 static int cmppid(const void *a, const void *b)
3105 { 3150 {
3106 return *(pid_t *)a - *(pid_t *)b; 3151 return *(pid_t *)a - *(pid_t *)b;
3107 } 3152 }
3108 3153
3109 static int fried_cmppid(const void *a, const void *b) 3154 static int fried_cmppid(const void *a, const void *b)
3110 { 3155 {
3111 return pid_fry(*(pid_t *)a) - pid_fry(*(pid_t *)b); 3156 return pid_fry(*(pid_t *)a) - pid_fry(*(pid_t *)b);
3112 } 3157 }
3113 3158
3114 static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp, 3159 static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp,
3115 enum cgroup_filetype type) 3160 enum cgroup_filetype type)
3116 { 3161 {
3117 struct cgroup_pidlist *l; 3162 struct cgroup_pidlist *l;
3118 /* don't need task_nsproxy() if we're looking at ourself */ 3163 /* don't need task_nsproxy() if we're looking at ourself */
3119 struct pid_namespace *ns = task_active_pid_ns(current); 3164 struct pid_namespace *ns = task_active_pid_ns(current);
3120 3165
3121 lockdep_assert_held(&cgrp->pidlist_mutex); 3166 lockdep_assert_held(&cgrp->pidlist_mutex);
3122 3167
3123 list_for_each_entry(l, &cgrp->pidlists, links) 3168 list_for_each_entry(l, &cgrp->pidlists, links)
3124 if (l->key.type == type && l->key.ns == ns) 3169 if (l->key.type == type && l->key.ns == ns)
3125 return l; 3170 return l;
3126 return NULL; 3171 return NULL;
3127 } 3172 }
3128 3173
3129 /* 3174 /*
3130 * find the appropriate pidlist for our purpose (given procs vs tasks) 3175 * find the appropriate pidlist for our purpose (given procs vs tasks)
3131 * returns with the lock on that pidlist already held, and takes care 3176 * returns with the lock on that pidlist already held, and takes care
3132 * of the use count, or returns NULL with no locks held if we're out of 3177 * of the use count, or returns NULL with no locks held if we're out of
3133 * memory. 3178 * memory.
3134 */ 3179 */
3135 static struct cgroup_pidlist *cgroup_pidlist_find_create(struct cgroup *cgrp, 3180 static struct cgroup_pidlist *cgroup_pidlist_find_create(struct cgroup *cgrp,
3136 enum cgroup_filetype type) 3181 enum cgroup_filetype type)
3137 { 3182 {
3138 struct cgroup_pidlist *l; 3183 struct cgroup_pidlist *l;
3139 3184
3140 lockdep_assert_held(&cgrp->pidlist_mutex); 3185 lockdep_assert_held(&cgrp->pidlist_mutex);
3141 3186
3142 l = cgroup_pidlist_find(cgrp, type); 3187 l = cgroup_pidlist_find(cgrp, type);
3143 if (l) 3188 if (l)
3144 return l; 3189 return l;
3145 3190
3146 /* entry not found; create a new one */ 3191 /* entry not found; create a new one */
3147 l = kzalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL); 3192 l = kzalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL);
3148 if (!l) 3193 if (!l)
3149 return l; 3194 return l;
3150 3195
3151 INIT_DELAYED_WORK(&l->destroy_dwork, cgroup_pidlist_destroy_work_fn); 3196 INIT_DELAYED_WORK(&l->destroy_dwork, cgroup_pidlist_destroy_work_fn);
3152 l->key.type = type; 3197 l->key.type = type;
3153 /* don't need task_nsproxy() if we're looking at ourself */ 3198 /* don't need task_nsproxy() if we're looking at ourself */
3154 l->key.ns = get_pid_ns(task_active_pid_ns(current)); 3199 l->key.ns = get_pid_ns(task_active_pid_ns(current));
3155 l->owner = cgrp; 3200 l->owner = cgrp;
3156 list_add(&l->links, &cgrp->pidlists); 3201 list_add(&l->links, &cgrp->pidlists);
3157 return l; 3202 return l;
3158 } 3203 }
3159 3204
3160 /* 3205 /*
3161 * Load a cgroup's pidarray with either procs' tgids or tasks' pids 3206 * Load a cgroup's pidarray with either procs' tgids or tasks' pids
3162 */ 3207 */
3163 static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type, 3208 static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type,
3164 struct cgroup_pidlist **lp) 3209 struct cgroup_pidlist **lp)
3165 { 3210 {
3166 pid_t *array; 3211 pid_t *array;
3167 int length; 3212 int length;
3168 int pid, n = 0; /* used for populating the array */ 3213 int pid, n = 0; /* used for populating the array */
3169 struct css_task_iter it; 3214 struct css_task_iter it;
3170 struct task_struct *tsk; 3215 struct task_struct *tsk;
3171 struct cgroup_pidlist *l; 3216 struct cgroup_pidlist *l;
3172 3217
3173 lockdep_assert_held(&cgrp->pidlist_mutex); 3218 lockdep_assert_held(&cgrp->pidlist_mutex);
3174 3219
3175 /* 3220 /*
3176 * If cgroup gets more users after we read count, we won't have 3221 * If cgroup gets more users after we read count, we won't have
3177 * enough space - tough. This race is indistinguishable to the 3222 * enough space - tough. This race is indistinguishable to the
3178 * caller from the case that the additional cgroup users didn't 3223 * caller from the case that the additional cgroup users didn't
3179 * show up until sometime later on. 3224 * show up until sometime later on.
3180 */ 3225 */
3181 length = cgroup_task_count(cgrp); 3226 length = cgroup_task_count(cgrp);
3182 array = pidlist_allocate(length); 3227 array = pidlist_allocate(length);
3183 if (!array) 3228 if (!array)
3184 return -ENOMEM; 3229 return -ENOMEM;
3185 /* now, populate the array */ 3230 /* now, populate the array */
3186 css_task_iter_start(&cgrp->dummy_css, &it); 3231 css_task_iter_start(&cgrp->dummy_css, &it);
3187 while ((tsk = css_task_iter_next(&it))) { 3232 while ((tsk = css_task_iter_next(&it))) {
3188 if (unlikely(n == length)) 3233 if (unlikely(n == length))
3189 break; 3234 break;
3190 /* get tgid or pid for procs or tasks file respectively */ 3235 /* get tgid or pid for procs or tasks file respectively */
3191 if (type == CGROUP_FILE_PROCS) 3236 if (type == CGROUP_FILE_PROCS)
3192 pid = task_tgid_vnr(tsk); 3237 pid = task_tgid_vnr(tsk);
3193 else 3238 else
3194 pid = task_pid_vnr(tsk); 3239 pid = task_pid_vnr(tsk);
3195 if (pid > 0) /* make sure to only use valid results */ 3240 if (pid > 0) /* make sure to only use valid results */
3196 array[n++] = pid; 3241 array[n++] = pid;
3197 } 3242 }
3198 css_task_iter_end(&it); 3243 css_task_iter_end(&it);
3199 length = n; 3244 length = n;
3200 /* now sort & (if procs) strip out duplicates */ 3245 /* now sort & (if procs) strip out duplicates */
3201 if (cgroup_sane_behavior(cgrp)) 3246 if (cgroup_sane_behavior(cgrp))
3202 sort(array, length, sizeof(pid_t), fried_cmppid, NULL); 3247 sort(array, length, sizeof(pid_t), fried_cmppid, NULL);
3203 else 3248 else
3204 sort(array, length, sizeof(pid_t), cmppid, NULL); 3249 sort(array, length, sizeof(pid_t), cmppid, NULL);
3205 if (type == CGROUP_FILE_PROCS) 3250 if (type == CGROUP_FILE_PROCS)
3206 length = pidlist_uniq(array, length); 3251 length = pidlist_uniq(array, length);
3207 3252
3208 l = cgroup_pidlist_find_create(cgrp, type); 3253 l = cgroup_pidlist_find_create(cgrp, type);
3209 if (!l) { 3254 if (!l) {
3210 mutex_unlock(&cgrp->pidlist_mutex); 3255 mutex_unlock(&cgrp->pidlist_mutex);
3211 pidlist_free(array); 3256 pidlist_free(array);
3212 return -ENOMEM; 3257 return -ENOMEM;
3213 } 3258 }
3214 3259
3215 /* store array, freeing old if necessary */ 3260 /* store array, freeing old if necessary */
3216 pidlist_free(l->list); 3261 pidlist_free(l->list);
3217 l->list = array; 3262 l->list = array;
3218 l->length = length; 3263 l->length = length;
3219 *lp = l; 3264 *lp = l;
3220 return 0; 3265 return 0;
3221 } 3266 }
3222 3267
3223 /** 3268 /**
3224 * cgroupstats_build - build and fill cgroupstats 3269 * cgroupstats_build - build and fill cgroupstats
3225 * @stats: cgroupstats to fill information into 3270 * @stats: cgroupstats to fill information into
3226 * @dentry: A dentry entry belonging to the cgroup for which stats have 3271 * @dentry: A dentry entry belonging to the cgroup for which stats have
3227 * been requested. 3272 * been requested.
3228 * 3273 *
3229 * Build and fill cgroupstats so that taskstats can export it to user 3274 * Build and fill cgroupstats so that taskstats can export it to user
3230 * space. 3275 * space.
3231 */ 3276 */
3232 int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry) 3277 int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry)
3233 { 3278 {
3234 struct kernfs_node *kn = kernfs_node_from_dentry(dentry); 3279 struct kernfs_node *kn = kernfs_node_from_dentry(dentry);
3235 struct cgroup *cgrp; 3280 struct cgroup *cgrp;
3236 struct css_task_iter it; 3281 struct css_task_iter it;
3237 struct task_struct *tsk; 3282 struct task_struct *tsk;
3238 3283
3239 /* it should be kernfs_node belonging to cgroupfs and is a directory */ 3284 /* it should be kernfs_node belonging to cgroupfs and is a directory */
3240 if (dentry->d_sb->s_type != &cgroup_fs_type || !kn || 3285 if (dentry->d_sb->s_type != &cgroup_fs_type || !kn ||
3241 kernfs_type(kn) != KERNFS_DIR) 3286 kernfs_type(kn) != KERNFS_DIR)
3242 return -EINVAL; 3287 return -EINVAL;
3243 3288
3244 mutex_lock(&cgroup_mutex); 3289 mutex_lock(&cgroup_mutex);
3245 3290
3246 /* 3291 /*
3247 * We aren't being called from kernfs and there's no guarantee on 3292 * We aren't being called from kernfs and there's no guarantee on
3248 * @kn->priv's validity. For this and css_tryget_from_dir(), 3293 * @kn->priv's validity. For this and css_tryget_from_dir(),
3249 * @kn->priv is RCU safe. Let's do the RCU dancing. 3294 * @kn->priv is RCU safe. Let's do the RCU dancing.
3250 */ 3295 */
3251 rcu_read_lock(); 3296 rcu_read_lock();
3252 cgrp = rcu_dereference(kn->priv); 3297 cgrp = rcu_dereference(kn->priv);
3253 if (!cgrp || cgroup_is_dead(cgrp)) { 3298 if (!cgrp || cgroup_is_dead(cgrp)) {
3254 rcu_read_unlock(); 3299 rcu_read_unlock();
3255 mutex_unlock(&cgroup_mutex); 3300 mutex_unlock(&cgroup_mutex);
3256 return -ENOENT; 3301 return -ENOENT;
3257 } 3302 }
3258 rcu_read_unlock(); 3303 rcu_read_unlock();
3259 3304
3260 css_task_iter_start(&cgrp->dummy_css, &it); 3305 css_task_iter_start(&cgrp->dummy_css, &it);
3261 while ((tsk = css_task_iter_next(&it))) { 3306 while ((tsk = css_task_iter_next(&it))) {
3262 switch (tsk->state) { 3307 switch (tsk->state) {
3263 case TASK_RUNNING: 3308 case TASK_RUNNING:
3264 stats->nr_running++; 3309 stats->nr_running++;
3265 break; 3310 break;
3266 case TASK_INTERRUPTIBLE: 3311 case TASK_INTERRUPTIBLE:
3267 stats->nr_sleeping++; 3312 stats->nr_sleeping++;
3268 break; 3313 break;
3269 case TASK_UNINTERRUPTIBLE: 3314 case TASK_UNINTERRUPTIBLE:
3270 stats->nr_uninterruptible++; 3315 stats->nr_uninterruptible++;
3271 break; 3316 break;
3272 case TASK_STOPPED: 3317 case TASK_STOPPED:
3273 stats->nr_stopped++; 3318 stats->nr_stopped++;
3274 break; 3319 break;
3275 default: 3320 default:
3276 if (delayacct_is_task_waiting_on_io(tsk)) 3321 if (delayacct_is_task_waiting_on_io(tsk))
3277 stats->nr_io_wait++; 3322 stats->nr_io_wait++;
3278 break; 3323 break;
3279 } 3324 }
3280 } 3325 }
3281 css_task_iter_end(&it); 3326 css_task_iter_end(&it);
3282 3327
3283 mutex_unlock(&cgroup_mutex); 3328 mutex_unlock(&cgroup_mutex);
3284 return 0; 3329 return 0;
3285 } 3330 }
3286 3331
3287 3332
3288 /* 3333 /*
3289 * seq_file methods for the tasks/procs files. The seq_file position is the 3334 * seq_file methods for the tasks/procs files. The seq_file position is the
3290 * next pid to display; the seq_file iterator is a pointer to the pid 3335 * next pid to display; the seq_file iterator is a pointer to the pid
3291 * in the cgroup->l->list array. 3336 * in the cgroup->l->list array.
3292 */ 3337 */
3293 3338
3294 static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos) 3339 static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos)
3295 { 3340 {
3296 /* 3341 /*
3297 * Initially we receive a position value that corresponds to 3342 * Initially we receive a position value that corresponds to
3298 * one more than the last pid shown (or 0 on the first call or 3343 * one more than the last pid shown (or 0 on the first call or
3299 * after a seek to the start). Use a binary-search to find the 3344 * after a seek to the start). Use a binary-search to find the
3300 * next pid to display, if any 3345 * next pid to display, if any
3301 */ 3346 */
3302 struct kernfs_open_file *of = s->private; 3347 struct kernfs_open_file *of = s->private;
3303 struct cgroup *cgrp = seq_css(s)->cgroup; 3348 struct cgroup *cgrp = seq_css(s)->cgroup;
3304 struct cgroup_pidlist *l; 3349 struct cgroup_pidlist *l;
3305 enum cgroup_filetype type = seq_cft(s)->private; 3350 enum cgroup_filetype type = seq_cft(s)->private;
3306 int index = 0, pid = *pos; 3351 int index = 0, pid = *pos;
3307 int *iter, ret; 3352 int *iter, ret;
3308 3353
3309 mutex_lock(&cgrp->pidlist_mutex); 3354 mutex_lock(&cgrp->pidlist_mutex);
3310 3355
3311 /* 3356 /*
3312 * !NULL @of->priv indicates that this isn't the first start() 3357 * !NULL @of->priv indicates that this isn't the first start()
3313 * after open. If the matching pidlist is around, we can use that. 3358 * after open. If the matching pidlist is around, we can use that.
3314 * Look for it. Note that @of->priv can't be used directly. It 3359 * Look for it. Note that @of->priv can't be used directly. It
3315 * could already have been destroyed. 3360 * could already have been destroyed.
3316 */ 3361 */
3317 if (of->priv) 3362 if (of->priv)
3318 of->priv = cgroup_pidlist_find(cgrp, type); 3363 of->priv = cgroup_pidlist_find(cgrp, type);
3319 3364
3320 /* 3365 /*
3321 * Either this is the first start() after open or the matching 3366 * Either this is the first start() after open or the matching
3322 * pidlist has been destroyed inbetween. Create a new one. 3367 * pidlist has been destroyed inbetween. Create a new one.
3323 */ 3368 */
3324 if (!of->priv) { 3369 if (!of->priv) {
3325 ret = pidlist_array_load(cgrp, type, 3370 ret = pidlist_array_load(cgrp, type,
3326 (struct cgroup_pidlist **)&of->priv); 3371 (struct cgroup_pidlist **)&of->priv);
3327 if (ret) 3372 if (ret)
3328 return ERR_PTR(ret); 3373 return ERR_PTR(ret);
3329 } 3374 }
3330 l = of->priv; 3375 l = of->priv;
3331 3376
3332 if (pid) { 3377 if (pid) {
3333 int end = l->length; 3378 int end = l->length;
3334 3379
3335 while (index < end) { 3380 while (index < end) {
3336 int mid = (index + end) / 2; 3381 int mid = (index + end) / 2;
3337 if (cgroup_pid_fry(cgrp, l->list[mid]) == pid) { 3382 if (cgroup_pid_fry(cgrp, l->list[mid]) == pid) {
3338 index = mid; 3383 index = mid;
3339 break; 3384 break;
3340 } else if (cgroup_pid_fry(cgrp, l->list[mid]) <= pid) 3385 } else if (cgroup_pid_fry(cgrp, l->list[mid]) <= pid)
3341 index = mid + 1; 3386 index = mid + 1;
3342 else 3387 else
3343 end = mid; 3388 end = mid;
3344 } 3389 }
3345 } 3390 }
3346 /* If we're off the end of the array, we're done */ 3391 /* If we're off the end of the array, we're done */
3347 if (index >= l->length) 3392 if (index >= l->length)
3348 return NULL; 3393 return NULL;
3349 /* Update the abstract position to be the actual pid that we found */ 3394 /* Update the abstract position to be the actual pid that we found */
3350 iter = l->list + index; 3395 iter = l->list + index;
3351 *pos = cgroup_pid_fry(cgrp, *iter); 3396 *pos = cgroup_pid_fry(cgrp, *iter);
3352 return iter; 3397 return iter;
3353 } 3398 }
3354 3399
3355 static void cgroup_pidlist_stop(struct seq_file *s, void *v) 3400 static void cgroup_pidlist_stop(struct seq_file *s, void *v)
3356 { 3401 {
3357 struct kernfs_open_file *of = s->private; 3402 struct kernfs_open_file *of = s->private;
3358 struct cgroup_pidlist *l = of->priv; 3403 struct cgroup_pidlist *l = of->priv;
3359 3404
3360 if (l) 3405 if (l)
3361 mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, 3406 mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork,
3362 CGROUP_PIDLIST_DESTROY_DELAY); 3407 CGROUP_PIDLIST_DESTROY_DELAY);
3363 mutex_unlock(&seq_css(s)->cgroup->pidlist_mutex); 3408 mutex_unlock(&seq_css(s)->cgroup->pidlist_mutex);
3364 } 3409 }
3365 3410
3366 static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos) 3411 static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos)
3367 { 3412 {
3368 struct kernfs_open_file *of = s->private; 3413 struct kernfs_open_file *of = s->private;
3369 struct cgroup_pidlist *l = of->priv; 3414 struct cgroup_pidlist *l = of->priv;
3370 pid_t *p = v; 3415 pid_t *p = v;
3371 pid_t *end = l->list + l->length; 3416 pid_t *end = l->list + l->length;
3372 /* 3417 /*
3373 * Advance to the next pid in the array. If this goes off the 3418 * Advance to the next pid in the array. If this goes off the
3374 * end, we're done 3419 * end, we're done
3375 */ 3420 */
3376 p++; 3421 p++;
3377 if (p >= end) { 3422 if (p >= end) {
3378 return NULL; 3423 return NULL;
3379 } else { 3424 } else {
3380 *pos = cgroup_pid_fry(seq_css(s)->cgroup, *p); 3425 *pos = cgroup_pid_fry(seq_css(s)->cgroup, *p);
3381 return p; 3426 return p;
3382 } 3427 }
3383 } 3428 }
3384 3429
3385 static int cgroup_pidlist_show(struct seq_file *s, void *v) 3430 static int cgroup_pidlist_show(struct seq_file *s, void *v)
3386 { 3431 {
3387 return seq_printf(s, "%d\n", *(int *)v); 3432 return seq_printf(s, "%d\n", *(int *)v);
3388 } 3433 }
3389 3434
3390 /* 3435 /*
3391 * seq_operations functions for iterating on pidlists through seq_file - 3436 * seq_operations functions for iterating on pidlists through seq_file -
3392 * independent of whether it's tasks or procs 3437 * independent of whether it's tasks or procs
3393 */ 3438 */
3394 static const struct seq_operations cgroup_pidlist_seq_operations = { 3439 static const struct seq_operations cgroup_pidlist_seq_operations = {
3395 .start = cgroup_pidlist_start, 3440 .start = cgroup_pidlist_start,
3396 .stop = cgroup_pidlist_stop, 3441 .stop = cgroup_pidlist_stop,
3397 .next = cgroup_pidlist_next, 3442 .next = cgroup_pidlist_next,
3398 .show = cgroup_pidlist_show, 3443 .show = cgroup_pidlist_show,
3399 }; 3444 };
3400 3445
3401 static u64 cgroup_read_notify_on_release(struct cgroup_subsys_state *css, 3446 static u64 cgroup_read_notify_on_release(struct cgroup_subsys_state *css,
3402 struct cftype *cft) 3447 struct cftype *cft)
3403 { 3448 {
3404 return notify_on_release(css->cgroup); 3449 return notify_on_release(css->cgroup);
3405 } 3450 }
3406 3451
3407 static int cgroup_write_notify_on_release(struct cgroup_subsys_state *css, 3452 static int cgroup_write_notify_on_release(struct cgroup_subsys_state *css,
3408 struct cftype *cft, u64 val) 3453 struct cftype *cft, u64 val)
3409 { 3454 {
3410 clear_bit(CGRP_RELEASABLE, &css->cgroup->flags); 3455 clear_bit(CGRP_RELEASABLE, &css->cgroup->flags);
3411 if (val) 3456 if (val)
3412 set_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags); 3457 set_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags);
3413 else 3458 else
3414 clear_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags); 3459 clear_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags);
3415 return 0; 3460 return 0;
3416 } 3461 }
3417 3462
3418 static u64 cgroup_clone_children_read(struct cgroup_subsys_state *css, 3463 static u64 cgroup_clone_children_read(struct cgroup_subsys_state *css,
3419 struct cftype *cft) 3464 struct cftype *cft)
3420 { 3465 {
3421 return test_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); 3466 return test_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags);
3422 } 3467 }
3423 3468
3424 static int cgroup_clone_children_write(struct cgroup_subsys_state *css, 3469 static int cgroup_clone_children_write(struct cgroup_subsys_state *css,
3425 struct cftype *cft, u64 val) 3470 struct cftype *cft, u64 val)
3426 { 3471 {
3427 if (val) 3472 if (val)
3428 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); 3473 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags);
3429 else 3474 else
3430 clear_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); 3475 clear_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags);
3431 return 0; 3476 return 0;
3432 } 3477 }
3433 3478
3434 static struct cftype cgroup_base_files[] = { 3479 static struct cftype cgroup_base_files[] = {
3435 { 3480 {
3436 .name = "cgroup.procs", 3481 .name = "cgroup.procs",
3437 .seq_start = cgroup_pidlist_start, 3482 .seq_start = cgroup_pidlist_start,
3438 .seq_next = cgroup_pidlist_next, 3483 .seq_next = cgroup_pidlist_next,
3439 .seq_stop = cgroup_pidlist_stop, 3484 .seq_stop = cgroup_pidlist_stop,
3440 .seq_show = cgroup_pidlist_show, 3485 .seq_show = cgroup_pidlist_show,
3441 .private = CGROUP_FILE_PROCS, 3486 .private = CGROUP_FILE_PROCS,
3442 .write_u64 = cgroup_procs_write, 3487 .write_u64 = cgroup_procs_write,
3443 .mode = S_IRUGO | S_IWUSR, 3488 .mode = S_IRUGO | S_IWUSR,
3444 }, 3489 },
3445 { 3490 {
3446 .name = "cgroup.clone_children", 3491 .name = "cgroup.clone_children",
3447 .flags = CFTYPE_INSANE, 3492 .flags = CFTYPE_INSANE,
3448 .read_u64 = cgroup_clone_children_read, 3493 .read_u64 = cgroup_clone_children_read,
3449 .write_u64 = cgroup_clone_children_write, 3494 .write_u64 = cgroup_clone_children_write,
3450 }, 3495 },
3451 { 3496 {
3452 .name = "cgroup.sane_behavior", 3497 .name = "cgroup.sane_behavior",
3453 .flags = CFTYPE_ONLY_ON_ROOT, 3498 .flags = CFTYPE_ONLY_ON_ROOT,
3454 .seq_show = cgroup_sane_behavior_show, 3499 .seq_show = cgroup_sane_behavior_show,
3455 }, 3500 },
3456 3501
3457 /* 3502 /*
3458 * Historical crazy stuff. These don't have "cgroup." prefix and 3503 * Historical crazy stuff. These don't have "cgroup." prefix and
3459 * don't exist if sane_behavior. If you're depending on these, be 3504 * don't exist if sane_behavior. If you're depending on these, be
3460 * prepared to be burned. 3505 * prepared to be burned.
3461 */ 3506 */
3462 { 3507 {
3463 .name = "tasks", 3508 .name = "tasks",
3464 .flags = CFTYPE_INSANE, /* use "procs" instead */ 3509 .flags = CFTYPE_INSANE, /* use "procs" instead */
3465 .seq_start = cgroup_pidlist_start, 3510 .seq_start = cgroup_pidlist_start,
3466 .seq_next = cgroup_pidlist_next, 3511 .seq_next = cgroup_pidlist_next,
3467 .seq_stop = cgroup_pidlist_stop, 3512 .seq_stop = cgroup_pidlist_stop,
3468 .seq_show = cgroup_pidlist_show, 3513 .seq_show = cgroup_pidlist_show,
3469 .private = CGROUP_FILE_TASKS, 3514 .private = CGROUP_FILE_TASKS,
3470 .write_u64 = cgroup_tasks_write, 3515 .write_u64 = cgroup_tasks_write,
3471 .mode = S_IRUGO | S_IWUSR, 3516 .mode = S_IRUGO | S_IWUSR,
3472 }, 3517 },
3473 { 3518 {
3474 .name = "notify_on_release", 3519 .name = "notify_on_release",
3475 .flags = CFTYPE_INSANE, 3520 .flags = CFTYPE_INSANE,
3476 .read_u64 = cgroup_read_notify_on_release, 3521 .read_u64 = cgroup_read_notify_on_release,
3477 .write_u64 = cgroup_write_notify_on_release, 3522 .write_u64 = cgroup_write_notify_on_release,
3478 }, 3523 },
3479 { 3524 {
3480 .name = "release_agent", 3525 .name = "release_agent",
3481 .flags = CFTYPE_INSANE | CFTYPE_ONLY_ON_ROOT, 3526 .flags = CFTYPE_INSANE | CFTYPE_ONLY_ON_ROOT,
3482 .seq_show = cgroup_release_agent_show, 3527 .seq_show = cgroup_release_agent_show,
3483 .write_string = cgroup_release_agent_write, 3528 .write_string = cgroup_release_agent_write,
3484 .max_write_len = PATH_MAX - 1, 3529 .max_write_len = PATH_MAX - 1,
3485 }, 3530 },
3486 { } /* terminate */ 3531 { } /* terminate */
3487 }; 3532 };
3488 3533
3489 /** 3534 /**
3490 * cgroup_populate_dir - create subsys files in a cgroup directory 3535 * cgroup_populate_dir - create subsys files in a cgroup directory
3491 * @cgrp: target cgroup 3536 * @cgrp: target cgroup
3492 * @subsys_mask: mask of the subsystem ids whose files should be added 3537 * @subsys_mask: mask of the subsystem ids whose files should be added
3493 * 3538 *
3494 * On failure, no file is added. 3539 * On failure, no file is added.
3495 */ 3540 */
3496 static int cgroup_populate_dir(struct cgroup *cgrp, unsigned long subsys_mask) 3541 static int cgroup_populate_dir(struct cgroup *cgrp, unsigned long subsys_mask)
3497 { 3542 {
3498 struct cgroup_subsys *ss; 3543 struct cgroup_subsys *ss;
3499 int i, ret = 0; 3544 int i, ret = 0;
3500 3545
3501 /* process cftsets of each subsystem */ 3546 /* process cftsets of each subsystem */
3502 for_each_subsys(ss, i) { 3547 for_each_subsys(ss, i) {
3503 struct cftype *cfts; 3548 struct cftype *cfts;
3504 3549
3505 if (!test_bit(i, &subsys_mask)) 3550 if (!test_bit(i, &subsys_mask))
3506 continue; 3551 continue;
3507 3552
3508 list_for_each_entry(cfts, &ss->cfts, node) { 3553 list_for_each_entry(cfts, &ss->cfts, node) {
3509 ret = cgroup_addrm_files(cgrp, cfts, true); 3554 ret = cgroup_addrm_files(cgrp, cfts, true);
3510 if (ret < 0) 3555 if (ret < 0)
3511 goto err; 3556 goto err;
3512 } 3557 }
3513 } 3558 }
3514 return 0; 3559 return 0;
3515 err: 3560 err:
3516 cgroup_clear_dir(cgrp, subsys_mask); 3561 cgroup_clear_dir(cgrp, subsys_mask);
3517 return ret; 3562 return ret;
3518 } 3563 }
3519 3564
3520 /* 3565 /*
3521 * css destruction is four-stage process. 3566 * css destruction is four-stage process.
3522 * 3567 *
3523 * 1. Destruction starts. Killing of the percpu_ref is initiated. 3568 * 1. Destruction starts. Killing of the percpu_ref is initiated.
3524 * Implemented in kill_css(). 3569 * Implemented in kill_css().
3525 * 3570 *
3526 * 2. When the percpu_ref is confirmed to be visible as killed on all CPUs 3571 * 2. When the percpu_ref is confirmed to be visible as killed on all CPUs
3527 * and thus css_tryget() is guaranteed to fail, the css can be offlined 3572 * and thus css_tryget() is guaranteed to fail, the css can be offlined
3528 * by invoking offline_css(). After offlining, the base ref is put. 3573 * by invoking offline_css(). After offlining, the base ref is put.
3529 * Implemented in css_killed_work_fn(). 3574 * Implemented in css_killed_work_fn().
3530 * 3575 *
3531 * 3. When the percpu_ref reaches zero, the only possible remaining 3576 * 3. When the percpu_ref reaches zero, the only possible remaining
3532 * accessors are inside RCU read sections. css_release() schedules the 3577 * accessors are inside RCU read sections. css_release() schedules the
3533 * RCU callback. 3578 * RCU callback.
3534 * 3579 *
3535 * 4. After the grace period, the css can be freed. Implemented in 3580 * 4. After the grace period, the css can be freed. Implemented in
3536 * css_free_work_fn(). 3581 * css_free_work_fn().
3537 * 3582 *
3538 * It is actually hairier because both step 2 and 4 require process context 3583 * It is actually hairier because both step 2 and 4 require process context
3539 * and thus involve punting to css->destroy_work adding two additional 3584 * and thus involve punting to css->destroy_work adding two additional
3540 * steps to the already complex sequence. 3585 * steps to the already complex sequence.
3541 */ 3586 */
3542 static void css_free_work_fn(struct work_struct *work) 3587 static void css_free_work_fn(struct work_struct *work)
3543 { 3588 {
3544 struct cgroup_subsys_state *css = 3589 struct cgroup_subsys_state *css =
3545 container_of(work, struct cgroup_subsys_state, destroy_work); 3590 container_of(work, struct cgroup_subsys_state, destroy_work);
3546 struct cgroup *cgrp = css->cgroup; 3591 struct cgroup *cgrp = css->cgroup;
3547 3592
3548 if (css->parent) 3593 if (css->parent)
3549 css_put(css->parent); 3594 css_put(css->parent);
3550 3595
3551 css->ss->css_free(css); 3596 css->ss->css_free(css);
3552 cgroup_put(cgrp); 3597 cgroup_put(cgrp);
3553 } 3598 }
3554 3599
3555 static void css_free_rcu_fn(struct rcu_head *rcu_head) 3600 static void css_free_rcu_fn(struct rcu_head *rcu_head)
3556 { 3601 {
3557 struct cgroup_subsys_state *css = 3602 struct cgroup_subsys_state *css =
3558 container_of(rcu_head, struct cgroup_subsys_state, rcu_head); 3603 container_of(rcu_head, struct cgroup_subsys_state, rcu_head);
3559 3604
3560 INIT_WORK(&css->destroy_work, css_free_work_fn); 3605 INIT_WORK(&css->destroy_work, css_free_work_fn);
3561 queue_work(cgroup_destroy_wq, &css->destroy_work); 3606 queue_work(cgroup_destroy_wq, &css->destroy_work);
3562 } 3607 }
3563 3608
3564 static void css_release(struct percpu_ref *ref) 3609 static void css_release(struct percpu_ref *ref)
3565 { 3610 {
3566 struct cgroup_subsys_state *css = 3611 struct cgroup_subsys_state *css =
3567 container_of(ref, struct cgroup_subsys_state, refcnt); 3612 container_of(ref, struct cgroup_subsys_state, refcnt);
3568 3613
3569 RCU_INIT_POINTER(css->cgroup->subsys[css->ss->id], NULL); 3614 RCU_INIT_POINTER(css->cgroup->subsys[css->ss->id], NULL);
3570 call_rcu(&css->rcu_head, css_free_rcu_fn); 3615 call_rcu(&css->rcu_head, css_free_rcu_fn);
3571 } 3616 }
3572 3617
3573 static void init_css(struct cgroup_subsys_state *css, struct cgroup_subsys *ss, 3618 static void init_css(struct cgroup_subsys_state *css, struct cgroup_subsys *ss,
3574 struct cgroup *cgrp) 3619 struct cgroup *cgrp)
3575 { 3620 {
3576 css->cgroup = cgrp; 3621 css->cgroup = cgrp;
3577 css->ss = ss; 3622 css->ss = ss;
3578 css->flags = 0; 3623 css->flags = 0;
3579 3624
3580 if (cgrp->parent) 3625 if (cgrp->parent)
3581 css->parent = cgroup_css(cgrp->parent, ss); 3626 css->parent = cgroup_css(cgrp->parent, ss);
3582 else 3627 else
3583 css->flags |= CSS_ROOT; 3628 css->flags |= CSS_ROOT;
3584 3629
3585 BUG_ON(cgroup_css(cgrp, ss)); 3630 BUG_ON(cgroup_css(cgrp, ss));
3586 } 3631 }
3587 3632
3588 /* invoke ->css_online() on a new CSS and mark it online if successful */ 3633 /* invoke ->css_online() on a new CSS and mark it online if successful */
3589 static int online_css(struct cgroup_subsys_state *css) 3634 static int online_css(struct cgroup_subsys_state *css)
3590 { 3635 {
3591 struct cgroup_subsys *ss = css->ss; 3636 struct cgroup_subsys *ss = css->ss;
3592 int ret = 0; 3637 int ret = 0;
3593 3638
3594 lockdep_assert_held(&cgroup_tree_mutex); 3639 lockdep_assert_held(&cgroup_tree_mutex);
3595 lockdep_assert_held(&cgroup_mutex); 3640 lockdep_assert_held(&cgroup_mutex);
3596 3641
3597 if (ss->css_online) 3642 if (ss->css_online)
3598 ret = ss->css_online(css); 3643 ret = ss->css_online(css);
3599 if (!ret) { 3644 if (!ret) {
3600 css->flags |= CSS_ONLINE; 3645 css->flags |= CSS_ONLINE;
3601 css->cgroup->nr_css++; 3646 css->cgroup->nr_css++;
3602 rcu_assign_pointer(css->cgroup->subsys[ss->id], css); 3647 rcu_assign_pointer(css->cgroup->subsys[ss->id], css);
3603 } 3648 }
3604 return ret; 3649 return ret;
3605 } 3650 }
3606 3651
3607 /* if the CSS is online, invoke ->css_offline() on it and mark it offline */ 3652 /* if the CSS is online, invoke ->css_offline() on it and mark it offline */
3608 static void offline_css(struct cgroup_subsys_state *css) 3653 static void offline_css(struct cgroup_subsys_state *css)
3609 { 3654 {
3610 struct cgroup_subsys *ss = css->ss; 3655 struct cgroup_subsys *ss = css->ss;
3611 3656
3612 lockdep_assert_held(&cgroup_tree_mutex); 3657 lockdep_assert_held(&cgroup_tree_mutex);
3613 lockdep_assert_held(&cgroup_mutex); 3658 lockdep_assert_held(&cgroup_mutex);
3614 3659
3615 if (!(css->flags & CSS_ONLINE)) 3660 if (!(css->flags & CSS_ONLINE))
3616 return; 3661 return;
3617 3662
3618 if (ss->css_offline) 3663 if (ss->css_offline)
3619 ss->css_offline(css); 3664 ss->css_offline(css);
3620 3665
3621 css->flags &= ~CSS_ONLINE; 3666 css->flags &= ~CSS_ONLINE;
3622 css->cgroup->nr_css--; 3667 css->cgroup->nr_css--;
3623 RCU_INIT_POINTER(css->cgroup->subsys[ss->id], css); 3668 RCU_INIT_POINTER(css->cgroup->subsys[ss->id], css);
3624 } 3669 }
3625 3670
3626 /** 3671 /**
3627 * create_css - create a cgroup_subsys_state 3672 * create_css - create a cgroup_subsys_state
3628 * @cgrp: the cgroup new css will be associated with 3673 * @cgrp: the cgroup new css will be associated with
3629 * @ss: the subsys of new css 3674 * @ss: the subsys of new css
3630 * 3675 *
3631 * Create a new css associated with @cgrp - @ss pair. On success, the new 3676 * Create a new css associated with @cgrp - @ss pair. On success, the new
3632 * css is online and installed in @cgrp with all interface files created. 3677 * css is online and installed in @cgrp with all interface files created.
3633 * Returns 0 on success, -errno on failure. 3678 * Returns 0 on success, -errno on failure.
3634 */ 3679 */
3635 static int create_css(struct cgroup *cgrp, struct cgroup_subsys *ss) 3680 static int create_css(struct cgroup *cgrp, struct cgroup_subsys *ss)
3636 { 3681 {
3637 struct cgroup *parent = cgrp->parent; 3682 struct cgroup *parent = cgrp->parent;
3638 struct cgroup_subsys_state *css; 3683 struct cgroup_subsys_state *css;
3639 int err; 3684 int err;
3640 3685
3641 lockdep_assert_held(&cgroup_mutex); 3686 lockdep_assert_held(&cgroup_mutex);
3642 3687
3643 css = ss->css_alloc(cgroup_css(parent, ss)); 3688 css = ss->css_alloc(cgroup_css(parent, ss));
3644 if (IS_ERR(css)) 3689 if (IS_ERR(css))
3645 return PTR_ERR(css); 3690 return PTR_ERR(css);
3646 3691
3647 err = percpu_ref_init(&css->refcnt, css_release); 3692 err = percpu_ref_init(&css->refcnt, css_release);
3648 if (err) 3693 if (err)
3649 goto err_free_css; 3694 goto err_free_css;
3650 3695
3651 init_css(css, ss, cgrp); 3696 init_css(css, ss, cgrp);
3652 3697
3653 err = cgroup_populate_dir(cgrp, 1 << ss->id); 3698 err = cgroup_populate_dir(cgrp, 1 << ss->id);
3654 if (err) 3699 if (err)
3655 goto err_free_percpu_ref; 3700 goto err_free_percpu_ref;
3656 3701
3657 err = online_css(css); 3702 err = online_css(css);
3658 if (err) 3703 if (err)
3659 goto err_clear_dir; 3704 goto err_clear_dir;
3660 3705
3661 cgroup_get(cgrp); 3706 cgroup_get(cgrp);
3662 css_get(css->parent); 3707 css_get(css->parent);
3663 3708
3664 if (ss->broken_hierarchy && !ss->warned_broken_hierarchy && 3709 if (ss->broken_hierarchy && !ss->warned_broken_hierarchy &&
3665 parent->parent) { 3710 parent->parent) {
3666 pr_warning("cgroup: %s (%d) created nested cgroup for controller \"%s\" which has incomplete hierarchy support. Nested cgroups may change behavior in the future.\n", 3711 pr_warning("cgroup: %s (%d) created nested cgroup for controller \"%s\" which has incomplete hierarchy support. Nested cgroups may change behavior in the future.\n",
3667 current->comm, current->pid, ss->name); 3712 current->comm, current->pid, ss->name);
3668 if (!strcmp(ss->name, "memory")) 3713 if (!strcmp(ss->name, "memory"))
3669 pr_warning("cgroup: \"memory\" requires setting use_hierarchy to 1 on the root.\n"); 3714 pr_warning("cgroup: \"memory\" requires setting use_hierarchy to 1 on the root.\n");
3670 ss->warned_broken_hierarchy = true; 3715 ss->warned_broken_hierarchy = true;
3671 } 3716 }
3672 3717
3673 return 0; 3718 return 0;
3674 3719
3675 err_clear_dir: 3720 err_clear_dir:
3676 cgroup_clear_dir(css->cgroup, 1 << css->ss->id); 3721 cgroup_clear_dir(css->cgroup, 1 << css->ss->id);
3677 err_free_percpu_ref: 3722 err_free_percpu_ref:
3678 percpu_ref_cancel_init(&css->refcnt); 3723 percpu_ref_cancel_init(&css->refcnt);
3679 err_free_css: 3724 err_free_css:
3680 ss->css_free(css); 3725 ss->css_free(css);
3681 return err; 3726 return err;
3682 } 3727 }
3683 3728
3684 /** 3729 /**
3685 * cgroup_create - create a cgroup 3730 * cgroup_create - create a cgroup
3686 * @parent: cgroup that will be parent of the new cgroup 3731 * @parent: cgroup that will be parent of the new cgroup
3687 * @name: name of the new cgroup 3732 * @name: name of the new cgroup
3688 * @mode: mode to set on new cgroup 3733 * @mode: mode to set on new cgroup
3689 */ 3734 */
3690 static long cgroup_create(struct cgroup *parent, const char *name, 3735 static long cgroup_create(struct cgroup *parent, const char *name,
3691 umode_t mode) 3736 umode_t mode)
3692 { 3737 {
3693 struct cgroup *cgrp; 3738 struct cgroup *cgrp;
3694 struct cgroup_root *root = parent->root; 3739 struct cgroup_root *root = parent->root;
3695 int ssid, err; 3740 int ssid, err;
3696 struct cgroup_subsys *ss; 3741 struct cgroup_subsys *ss;
3697 struct kernfs_node *kn; 3742 struct kernfs_node *kn;
3698 3743
3699 /* 3744 /*
3700 * XXX: The default hierarchy isn't fully implemented yet. Block 3745 * XXX: The default hierarchy isn't fully implemented yet. Block
3701 * !root cgroup creation on it for now. 3746 * !root cgroup creation on it for now.
3702 */ 3747 */
3703 if (root == &cgrp_dfl_root) 3748 if (root == &cgrp_dfl_root)
3704 return -EINVAL; 3749 return -EINVAL;
3705 3750
3706 /* allocate the cgroup and its ID, 0 is reserved for the root */ 3751 /* allocate the cgroup and its ID, 0 is reserved for the root */
3707 cgrp = kzalloc(sizeof(*cgrp), GFP_KERNEL); 3752 cgrp = kzalloc(sizeof(*cgrp), GFP_KERNEL);
3708 if (!cgrp) 3753 if (!cgrp)
3709 return -ENOMEM; 3754 return -ENOMEM;
3710 3755
3711 mutex_lock(&cgroup_tree_mutex); 3756 mutex_lock(&cgroup_tree_mutex);
3712 3757
3713 /* 3758 /*
3714 * Only live parents can have children. Note that the liveliness 3759 * Only live parents can have children. Note that the liveliness
3715 * check isn't strictly necessary because cgroup_mkdir() and 3760 * check isn't strictly necessary because cgroup_mkdir() and
3716 * cgroup_rmdir() are fully synchronized by i_mutex; however, do it 3761 * cgroup_rmdir() are fully synchronized by i_mutex; however, do it
3717 * anyway so that locking is contained inside cgroup proper and we 3762 * anyway so that locking is contained inside cgroup proper and we
3718 * don't get nasty surprises if we ever grow another caller. 3763 * don't get nasty surprises if we ever grow another caller.
3719 */ 3764 */
3720 if (!cgroup_lock_live_group(parent)) { 3765 if (!cgroup_lock_live_group(parent)) {
3721 err = -ENODEV; 3766 err = -ENODEV;
3722 goto err_unlock_tree; 3767 goto err_unlock_tree;
3723 } 3768 }
3724 3769
3725 /* 3770 /*
3726 * Temporarily set the pointer to NULL, so idr_find() won't return 3771 * Temporarily set the pointer to NULL, so idr_find() won't return
3727 * a half-baked cgroup. 3772 * a half-baked cgroup.
3728 */ 3773 */
3729 cgrp->id = idr_alloc(&root->cgroup_idr, NULL, 1, 0, GFP_KERNEL); 3774 cgrp->id = idr_alloc(&root->cgroup_idr, NULL, 1, 0, GFP_KERNEL);
3730 if (cgrp->id < 0) { 3775 if (cgrp->id < 0) {
3731 err = -ENOMEM; 3776 err = -ENOMEM;
3732 goto err_unlock; 3777 goto err_unlock;
3733 } 3778 }
3734 3779
3735 init_cgroup_housekeeping(cgrp); 3780 init_cgroup_housekeeping(cgrp);
3736 3781
3737 cgrp->parent = parent; 3782 cgrp->parent = parent;
3738 cgrp->dummy_css.parent = &parent->dummy_css; 3783 cgrp->dummy_css.parent = &parent->dummy_css;
3739 cgrp->root = parent->root; 3784 cgrp->root = parent->root;
3740 3785
3741 if (notify_on_release(parent)) 3786 if (notify_on_release(parent))
3742 set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); 3787 set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
3743 3788
3744 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &parent->flags)) 3789 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &parent->flags))
3745 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags); 3790 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags);
3746 3791
3747 /* create the directory */ 3792 /* create the directory */
3748 kn = kernfs_create_dir(parent->kn, name, mode, cgrp); 3793 kn = kernfs_create_dir(parent->kn, name, mode, cgrp);
3749 if (IS_ERR(kn)) { 3794 if (IS_ERR(kn)) {
3750 err = PTR_ERR(kn); 3795 err = PTR_ERR(kn);
3751 goto err_free_id; 3796 goto err_free_id;
3752 } 3797 }
3753 cgrp->kn = kn; 3798 cgrp->kn = kn;
3754 3799
3755 /* 3800 /*
3756 * This extra ref will be put in cgroup_free_fn() and guarantees 3801 * This extra ref will be put in cgroup_free_fn() and guarantees
3757 * that @cgrp->kn is always accessible. 3802 * that @cgrp->kn is always accessible.
3758 */ 3803 */
3759 kernfs_get(kn); 3804 kernfs_get(kn);
3760 3805
3761 cgrp->serial_nr = cgroup_serial_nr_next++; 3806 cgrp->serial_nr = cgroup_serial_nr_next++;
3762 3807
3763 /* allocation complete, commit to creation */ 3808 /* allocation complete, commit to creation */
3764 list_add_tail_rcu(&cgrp->sibling, &cgrp->parent->children); 3809 list_add_tail_rcu(&cgrp->sibling, &cgrp->parent->children);
3765 atomic_inc(&root->nr_cgrps); 3810 atomic_inc(&root->nr_cgrps);
3766 cgroup_get(parent); 3811 cgroup_get(parent);
3767 3812
3768 /* 3813 /*
3769 * @cgrp is now fully operational. If something fails after this 3814 * @cgrp is now fully operational. If something fails after this
3770 * point, it'll be released via the normal destruction path. 3815 * point, it'll be released via the normal destruction path.
3771 */ 3816 */
3772 idr_replace(&root->cgroup_idr, cgrp, cgrp->id); 3817 idr_replace(&root->cgroup_idr, cgrp, cgrp->id);
3773 3818
3774 err = cgroup_kn_set_ugid(kn); 3819 err = cgroup_kn_set_ugid(kn);
3775 if (err) 3820 if (err)
3776 goto err_destroy; 3821 goto err_destroy;
3777 3822
3778 err = cgroup_addrm_files(cgrp, cgroup_base_files, true); 3823 err = cgroup_addrm_files(cgrp, cgroup_base_files, true);
3779 if (err) 3824 if (err)
3780 goto err_destroy; 3825 goto err_destroy;
3781 3826
3782 /* let's create and online css's */ 3827 /* let's create and online css's */
3783 for_each_subsys(ss, ssid) { 3828 for_each_subsys(ss, ssid) {
3784 if (parent->child_subsys_mask & (1 << ssid)) { 3829 if (parent->child_subsys_mask & (1 << ssid)) {
3785 err = create_css(cgrp, ss); 3830 err = create_css(cgrp, ss);
3786 if (err) 3831 if (err)
3787 goto err_destroy; 3832 goto err_destroy;
3788 } 3833 }
3789 } 3834 }
3790 3835
3791 cgrp->child_subsys_mask = parent->child_subsys_mask; 3836 cgrp->child_subsys_mask = parent->child_subsys_mask;
3792 3837
3793 kernfs_activate(kn); 3838 kernfs_activate(kn);
3794 3839
3795 mutex_unlock(&cgroup_mutex); 3840 mutex_unlock(&cgroup_mutex);
3796 mutex_unlock(&cgroup_tree_mutex); 3841 mutex_unlock(&cgroup_tree_mutex);
3797 3842
3798 return 0; 3843 return 0;
3799 3844
3800 err_free_id: 3845 err_free_id:
3801 idr_remove(&root->cgroup_idr, cgrp->id); 3846 idr_remove(&root->cgroup_idr, cgrp->id);
3802 err_unlock: 3847 err_unlock:
3803 mutex_unlock(&cgroup_mutex); 3848 mutex_unlock(&cgroup_mutex);
3804 err_unlock_tree: 3849 err_unlock_tree:
3805 mutex_unlock(&cgroup_tree_mutex); 3850 mutex_unlock(&cgroup_tree_mutex);
3806 kfree(cgrp); 3851 kfree(cgrp);
3807 return err; 3852 return err;
3808 3853
3809 err_destroy: 3854 err_destroy:
3810 cgroup_destroy_locked(cgrp); 3855 cgroup_destroy_locked(cgrp);
3811 mutex_unlock(&cgroup_mutex); 3856 mutex_unlock(&cgroup_mutex);
3812 mutex_unlock(&cgroup_tree_mutex); 3857 mutex_unlock(&cgroup_tree_mutex);
3813 return err; 3858 return err;
3814 } 3859 }
3815 3860
3816 static int cgroup_mkdir(struct kernfs_node *parent_kn, const char *name, 3861 static int cgroup_mkdir(struct kernfs_node *parent_kn, const char *name,
3817 umode_t mode) 3862 umode_t mode)
3818 { 3863 {
3819 struct cgroup *parent = parent_kn->priv; 3864 struct cgroup *parent = parent_kn->priv;
3820 int ret; 3865 int ret;
3821 3866
3822 /* 3867 /*
3823 * cgroup_create() grabs cgroup_tree_mutex which nests outside 3868 * cgroup_create() grabs cgroup_tree_mutex which nests outside
3824 * kernfs active_ref and cgroup_create() already synchronizes 3869 * kernfs active_ref and cgroup_create() already synchronizes
3825 * properly against removal through cgroup_lock_live_group(). 3870 * properly against removal through cgroup_lock_live_group().
3826 * Break it before calling cgroup_create(). 3871 * Break it before calling cgroup_create().
3827 */ 3872 */
3828 cgroup_get(parent); 3873 cgroup_get(parent);
3829 kernfs_break_active_protection(parent_kn); 3874 kernfs_break_active_protection(parent_kn);
3830 3875
3831 ret = cgroup_create(parent, name, mode); 3876 ret = cgroup_create(parent, name, mode);
3832 3877
3833 kernfs_unbreak_active_protection(parent_kn); 3878 kernfs_unbreak_active_protection(parent_kn);
3834 cgroup_put(parent); 3879 cgroup_put(parent);
3835 return ret; 3880 return ret;
3836 } 3881 }
3837 3882
3838 /* 3883 /*
3839 * This is called when the refcnt of a css is confirmed to be killed. 3884 * This is called when the refcnt of a css is confirmed to be killed.
3840 * css_tryget() is now guaranteed to fail. 3885 * css_tryget() is now guaranteed to fail.
3841 */ 3886 */
3842 static void css_killed_work_fn(struct work_struct *work) 3887 static void css_killed_work_fn(struct work_struct *work)
3843 { 3888 {
3844 struct cgroup_subsys_state *css = 3889 struct cgroup_subsys_state *css =
3845 container_of(work, struct cgroup_subsys_state, destroy_work); 3890 container_of(work, struct cgroup_subsys_state, destroy_work);
3846 struct cgroup *cgrp = css->cgroup; 3891 struct cgroup *cgrp = css->cgroup;
3847 3892
3848 mutex_lock(&cgroup_tree_mutex); 3893 mutex_lock(&cgroup_tree_mutex);
3849 mutex_lock(&cgroup_mutex); 3894 mutex_lock(&cgroup_mutex);
3850 3895
3851 /* 3896 /*
3852 * css_tryget() is guaranteed to fail now. Tell subsystems to 3897 * css_tryget() is guaranteed to fail now. Tell subsystems to
3853 * initate destruction. 3898 * initate destruction.
3854 */ 3899 */
3855 offline_css(css); 3900 offline_css(css);
3856 3901
3857 /* 3902 /*
3858 * If @cgrp is marked dead, it's waiting for refs of all css's to 3903 * If @cgrp is marked dead, it's waiting for refs of all css's to
3859 * be disabled before proceeding to the second phase of cgroup 3904 * be disabled before proceeding to the second phase of cgroup
3860 * destruction. If we are the last one, kick it off. 3905 * destruction. If we are the last one, kick it off.
3861 */ 3906 */
3862 if (!cgrp->nr_css && cgroup_is_dead(cgrp)) 3907 if (!cgrp->nr_css && cgroup_is_dead(cgrp))
3863 cgroup_destroy_css_killed(cgrp); 3908 cgroup_destroy_css_killed(cgrp);
3864 3909
3865 mutex_unlock(&cgroup_mutex); 3910 mutex_unlock(&cgroup_mutex);
3866 mutex_unlock(&cgroup_tree_mutex); 3911 mutex_unlock(&cgroup_tree_mutex);
3867 3912
3868 /* 3913 /*
3869 * Put the css refs from kill_css(). Each css holds an extra 3914 * Put the css refs from kill_css(). Each css holds an extra
3870 * reference to the cgroup's dentry and cgroup removal proceeds 3915 * reference to the cgroup's dentry and cgroup removal proceeds
3871 * regardless of css refs. On the last put of each css, whenever 3916 * regardless of css refs. On the last put of each css, whenever
3872 * that may be, the extra dentry ref is put so that dentry 3917 * that may be, the extra dentry ref is put so that dentry
3873 * destruction happens only after all css's are released. 3918 * destruction happens only after all css's are released.
3874 */ 3919 */
3875 css_put(css); 3920 css_put(css);
3876 } 3921 }
3877 3922
3878 /* css kill confirmation processing requires process context, bounce */ 3923 /* css kill confirmation processing requires process context, bounce */
3879 static void css_killed_ref_fn(struct percpu_ref *ref) 3924 static void css_killed_ref_fn(struct percpu_ref *ref)
3880 { 3925 {
3881 struct cgroup_subsys_state *css = 3926 struct cgroup_subsys_state *css =
3882 container_of(ref, struct cgroup_subsys_state, refcnt); 3927 container_of(ref, struct cgroup_subsys_state, refcnt);
3883 3928
3884 INIT_WORK(&css->destroy_work, css_killed_work_fn); 3929 INIT_WORK(&css->destroy_work, css_killed_work_fn);
3885 queue_work(cgroup_destroy_wq, &css->destroy_work); 3930 queue_work(cgroup_destroy_wq, &css->destroy_work);
3886 } 3931 }
3887 3932
3888 /** 3933 /**
3889 * kill_css - destroy a css 3934 * kill_css - destroy a css
3890 * @css: css to destroy 3935 * @css: css to destroy
3891 * 3936 *
3892 * This function initiates destruction of @css by removing cgroup interface 3937 * This function initiates destruction of @css by removing cgroup interface
3893 * files and putting its base reference. ->css_offline() will be invoked 3938 * files and putting its base reference. ->css_offline() will be invoked
3894 * asynchronously once css_tryget() is guaranteed to fail and when the 3939 * asynchronously once css_tryget() is guaranteed to fail and when the
3895 * reference count reaches zero, @css will be released. 3940 * reference count reaches zero, @css will be released.
3896 */ 3941 */
3897 static void kill_css(struct cgroup_subsys_state *css) 3942 static void kill_css(struct cgroup_subsys_state *css)
3898 { 3943 {
3899 lockdep_assert_held(&cgroup_tree_mutex); 3944 lockdep_assert_held(&cgroup_tree_mutex);
3900 3945
3901 /* 3946 /*
3902 * This must happen before css is disassociated with its cgroup. 3947 * This must happen before css is disassociated with its cgroup.
3903 * See seq_css() for details. 3948 * See seq_css() for details.
3904 */ 3949 */
3905 cgroup_clear_dir(css->cgroup, 1 << css->ss->id); 3950 cgroup_clear_dir(css->cgroup, 1 << css->ss->id);
3906 3951
3907 /* 3952 /*
3908 * Killing would put the base ref, but we need to keep it alive 3953 * Killing would put the base ref, but we need to keep it alive
3909 * until after ->css_offline(). 3954 * until after ->css_offline().
3910 */ 3955 */
3911 css_get(css); 3956 css_get(css);
3912 3957
3913 /* 3958 /*
3914 * cgroup core guarantees that, by the time ->css_offline() is 3959 * cgroup core guarantees that, by the time ->css_offline() is
3915 * invoked, no new css reference will be given out via 3960 * invoked, no new css reference will be given out via
3916 * css_tryget(). We can't simply call percpu_ref_kill() and 3961 * css_tryget(). We can't simply call percpu_ref_kill() and
3917 * proceed to offlining css's because percpu_ref_kill() doesn't 3962 * proceed to offlining css's because percpu_ref_kill() doesn't
3918 * guarantee that the ref is seen as killed on all CPUs on return. 3963 * guarantee that the ref is seen as killed on all CPUs on return.
3919 * 3964 *
3920 * Use percpu_ref_kill_and_confirm() to get notifications as each 3965 * Use percpu_ref_kill_and_confirm() to get notifications as each
3921 * css is confirmed to be seen as killed on all CPUs. 3966 * css is confirmed to be seen as killed on all CPUs.
3922 */ 3967 */
3923 percpu_ref_kill_and_confirm(&css->refcnt, css_killed_ref_fn); 3968 percpu_ref_kill_and_confirm(&css->refcnt, css_killed_ref_fn);
3924 } 3969 }
3925 3970
3926 /** 3971 /**
3927 * cgroup_destroy_locked - the first stage of cgroup destruction 3972 * cgroup_destroy_locked - the first stage of cgroup destruction
3928 * @cgrp: cgroup to be destroyed 3973 * @cgrp: cgroup to be destroyed
3929 * 3974 *
3930 * css's make use of percpu refcnts whose killing latency shouldn't be 3975 * css's make use of percpu refcnts whose killing latency shouldn't be
3931 * exposed to userland and are RCU protected. Also, cgroup core needs to 3976 * exposed to userland and are RCU protected. Also, cgroup core needs to
3932 * guarantee that css_tryget() won't succeed by the time ->css_offline() is 3977 * guarantee that css_tryget() won't succeed by the time ->css_offline() is
3933 * invoked. To satisfy all the requirements, destruction is implemented in 3978 * invoked. To satisfy all the requirements, destruction is implemented in
3934 * the following two steps. 3979 * the following two steps.
3935 * 3980 *
3936 * s1. Verify @cgrp can be destroyed and mark it dying. Remove all 3981 * s1. Verify @cgrp can be destroyed and mark it dying. Remove all
3937 * userland visible parts and start killing the percpu refcnts of 3982 * userland visible parts and start killing the percpu refcnts of
3938 * css's. Set up so that the next stage will be kicked off once all 3983 * css's. Set up so that the next stage will be kicked off once all
3939 * the percpu refcnts are confirmed to be killed. 3984 * the percpu refcnts are confirmed to be killed.
3940 * 3985 *
3941 * s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the 3986 * s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the
3942 * rest of destruction. Once all cgroup references are gone, the 3987 * rest of destruction. Once all cgroup references are gone, the
3943 * cgroup is RCU-freed. 3988 * cgroup is RCU-freed.
3944 * 3989 *
3945 * This function implements s1. After this step, @cgrp is gone as far as 3990 * This function implements s1. After this step, @cgrp is gone as far as
3946 * the userland is concerned and a new cgroup with the same name may be 3991 * the userland is concerned and a new cgroup with the same name may be
3947 * created. As cgroup doesn't care about the names internally, this 3992 * created. As cgroup doesn't care about the names internally, this
3948 * doesn't cause any problem. 3993 * doesn't cause any problem.
3949 */ 3994 */
3950 static int cgroup_destroy_locked(struct cgroup *cgrp) 3995 static int cgroup_destroy_locked(struct cgroup *cgrp)
3951 __releases(&cgroup_mutex) __acquires(&cgroup_mutex) 3996 __releases(&cgroup_mutex) __acquires(&cgroup_mutex)
3952 { 3997 {
3953 struct cgroup *child; 3998 struct cgroup *child;
3954 struct cgroup_subsys_state *css; 3999 struct cgroup_subsys_state *css;
3955 bool empty; 4000 bool empty;
3956 int ssid; 4001 int ssid;
3957 4002
3958 lockdep_assert_held(&cgroup_tree_mutex); 4003 lockdep_assert_held(&cgroup_tree_mutex);
3959 lockdep_assert_held(&cgroup_mutex); 4004 lockdep_assert_held(&cgroup_mutex);
3960 4005
3961 /* 4006 /*
3962 * css_set_rwsem synchronizes access to ->cset_links and prevents 4007 * css_set_rwsem synchronizes access to ->cset_links and prevents
3963 * @cgrp from being removed while put_css_set() is in progress. 4008 * @cgrp from being removed while put_css_set() is in progress.
3964 */ 4009 */
3965 down_read(&css_set_rwsem); 4010 down_read(&css_set_rwsem);
3966 empty = list_empty(&cgrp->cset_links); 4011 empty = list_empty(&cgrp->cset_links);
3967 up_read(&css_set_rwsem); 4012 up_read(&css_set_rwsem);
3968 if (!empty) 4013 if (!empty)
3969 return -EBUSY; 4014 return -EBUSY;
3970 4015
3971 /* 4016 /*
3972 * Make sure there's no live children. We can't test ->children 4017 * Make sure there's no live children. We can't test ->children
3973 * emptiness as dead children linger on it while being destroyed; 4018 * emptiness as dead children linger on it while being destroyed;
3974 * otherwise, "rmdir parent/child parent" may fail with -EBUSY. 4019 * otherwise, "rmdir parent/child parent" may fail with -EBUSY.
3975 */ 4020 */
3976 empty = true; 4021 empty = true;
3977 rcu_read_lock(); 4022 rcu_read_lock();
3978 list_for_each_entry_rcu(child, &cgrp->children, sibling) { 4023 list_for_each_entry_rcu(child, &cgrp->children, sibling) {
3979 empty = cgroup_is_dead(child); 4024 empty = cgroup_is_dead(child);
3980 if (!empty) 4025 if (!empty)
3981 break; 4026 break;
3982 } 4027 }
3983 rcu_read_unlock(); 4028 rcu_read_unlock();
3984 if (!empty) 4029 if (!empty)
3985 return -EBUSY; 4030 return -EBUSY;
3986 4031
3987 /* 4032 /*
3988 * Mark @cgrp dead. This prevents further task migration and child 4033 * Mark @cgrp dead. This prevents further task migration and child
3989 * creation by disabling cgroup_lock_live_group(). Note that 4034 * creation by disabling cgroup_lock_live_group(). Note that
3990 * CGRP_DEAD assertion is depended upon by css_next_child() to 4035 * CGRP_DEAD assertion is depended upon by css_next_child() to
3991 * resume iteration after dropping RCU read lock. See 4036 * resume iteration after dropping RCU read lock. See
3992 * css_next_child() for details. 4037 * css_next_child() for details.
3993 */ 4038 */
3994 set_bit(CGRP_DEAD, &cgrp->flags); 4039 set_bit(CGRP_DEAD, &cgrp->flags);
3995 4040
3996 /* 4041 /*
3997 * Initiate massacre of all css's. cgroup_destroy_css_killed() 4042 * Initiate massacre of all css's. cgroup_destroy_css_killed()
3998 * will be invoked to perform the rest of destruction once the 4043 * will be invoked to perform the rest of destruction once the
3999 * percpu refs of all css's are confirmed to be killed. This 4044 * percpu refs of all css's are confirmed to be killed. This
4000 * involves removing the subsystem's files, drop cgroup_mutex. 4045 * involves removing the subsystem's files, drop cgroup_mutex.
4001 */ 4046 */
4002 mutex_unlock(&cgroup_mutex); 4047 mutex_unlock(&cgroup_mutex);
4003 for_each_css(css, ssid, cgrp) 4048 for_each_css(css, ssid, cgrp)
4004 kill_css(css); 4049 kill_css(css);
4005 mutex_lock(&cgroup_mutex); 4050 mutex_lock(&cgroup_mutex);
4006 4051
4007 /* CGRP_DEAD is set, remove from ->release_list for the last time */ 4052 /* CGRP_DEAD is set, remove from ->release_list for the last time */
4008 raw_spin_lock(&release_list_lock); 4053 raw_spin_lock(&release_list_lock);
4009 if (!list_empty(&cgrp->release_list)) 4054 if (!list_empty(&cgrp->release_list))
4010 list_del_init(&cgrp->release_list); 4055 list_del_init(&cgrp->release_list);
4011 raw_spin_unlock(&release_list_lock); 4056 raw_spin_unlock(&release_list_lock);
4012 4057
4013 /* 4058 /*
4014 * If @cgrp has css's attached, the second stage of cgroup 4059 * If @cgrp has css's attached, the second stage of cgroup
4015 * destruction is kicked off from css_killed_work_fn() after the 4060 * destruction is kicked off from css_killed_work_fn() after the
4016 * refs of all attached css's are killed. If @cgrp doesn't have 4061 * refs of all attached css's are killed. If @cgrp doesn't have
4017 * any css, we kick it off here. 4062 * any css, we kick it off here.
4018 */ 4063 */
4019 if (!cgrp->nr_css) 4064 if (!cgrp->nr_css)
4020 cgroup_destroy_css_killed(cgrp); 4065 cgroup_destroy_css_killed(cgrp);
4021 4066
4022 /* remove @cgrp directory along with the base files */ 4067 /* remove @cgrp directory along with the base files */
4023 mutex_unlock(&cgroup_mutex); 4068 mutex_unlock(&cgroup_mutex);
4024 4069
4025 /* 4070 /*
4026 * There are two control paths which try to determine cgroup from 4071 * There are two control paths which try to determine cgroup from
4027 * dentry without going through kernfs - cgroupstats_build() and 4072 * dentry without going through kernfs - cgroupstats_build() and
4028 * css_tryget_from_dir(). Those are supported by RCU protecting 4073 * css_tryget_from_dir(). Those are supported by RCU protecting
4029 * clearing of cgrp->kn->priv backpointer, which should happen 4074 * clearing of cgrp->kn->priv backpointer, which should happen
4030 * after all files under it have been removed. 4075 * after all files under it have been removed.
4031 */ 4076 */
4032 kernfs_remove(cgrp->kn); /* @cgrp has an extra ref on its kn */ 4077 kernfs_remove(cgrp->kn); /* @cgrp has an extra ref on its kn */
4033 RCU_INIT_POINTER(*(void __rcu __force **)&cgrp->kn->priv, NULL); 4078 RCU_INIT_POINTER(*(void __rcu __force **)&cgrp->kn->priv, NULL);
4034 4079
4035 mutex_lock(&cgroup_mutex); 4080 mutex_lock(&cgroup_mutex);
4036 4081
4037 return 0; 4082 return 0;
4038 }; 4083 };
4039 4084
4040 /** 4085 /**
4041 * cgroup_destroy_css_killed - the second step of cgroup destruction 4086 * cgroup_destroy_css_killed - the second step of cgroup destruction
4042 * @work: cgroup->destroy_free_work 4087 * @work: cgroup->destroy_free_work
4043 * 4088 *
4044 * This function is invoked from a work item for a cgroup which is being 4089 * This function is invoked from a work item for a cgroup which is being
4045 * destroyed after all css's are offlined and performs the rest of 4090 * destroyed after all css's are offlined and performs the rest of
4046 * destruction. This is the second step of destruction described in the 4091 * destruction. This is the second step of destruction described in the
4047 * comment above cgroup_destroy_locked(). 4092 * comment above cgroup_destroy_locked().
4048 */ 4093 */
4049 static void cgroup_destroy_css_killed(struct cgroup *cgrp) 4094 static void cgroup_destroy_css_killed(struct cgroup *cgrp)
4050 { 4095 {
4051 struct cgroup *parent = cgrp->parent; 4096 struct cgroup *parent = cgrp->parent;
4052 4097
4053 lockdep_assert_held(&cgroup_tree_mutex); 4098 lockdep_assert_held(&cgroup_tree_mutex);
4054 lockdep_assert_held(&cgroup_mutex); 4099 lockdep_assert_held(&cgroup_mutex);
4055 4100
4056 /* delete this cgroup from parent->children */ 4101 /* delete this cgroup from parent->children */
4057 list_del_rcu(&cgrp->sibling); 4102 list_del_rcu(&cgrp->sibling);
4058 4103
4059 cgroup_put(cgrp); 4104 cgroup_put(cgrp);
4060 4105
4061 set_bit(CGRP_RELEASABLE, &parent->flags); 4106 set_bit(CGRP_RELEASABLE, &parent->flags);
4062 check_for_release(parent); 4107 check_for_release(parent);
4063 } 4108 }
4064 4109
4065 static int cgroup_rmdir(struct kernfs_node *kn) 4110 static int cgroup_rmdir(struct kernfs_node *kn)
4066 { 4111 {
4067 struct cgroup *cgrp = kn->priv; 4112 struct cgroup *cgrp = kn->priv;
4068 int ret = 0; 4113 int ret = 0;
4069 4114
4070 /* 4115 /*
4071 * This is self-destruction but @kn can't be removed while this 4116 * This is self-destruction but @kn can't be removed while this
4072 * callback is in progress. Let's break active protection. Once 4117 * callback is in progress. Let's break active protection. Once
4073 * the protection is broken, @cgrp can be destroyed at any point. 4118 * the protection is broken, @cgrp can be destroyed at any point.
4074 * Pin it so that it stays accessible. 4119 * Pin it so that it stays accessible.
4075 */ 4120 */
4076 cgroup_get(cgrp); 4121 cgroup_get(cgrp);
4077 kernfs_break_active_protection(kn); 4122 kernfs_break_active_protection(kn);
4078 4123
4079 mutex_lock(&cgroup_tree_mutex); 4124 mutex_lock(&cgroup_tree_mutex);
4080 mutex_lock(&cgroup_mutex); 4125 mutex_lock(&cgroup_mutex);
4081 4126
4082 /* 4127 /*
4083 * @cgrp might already have been destroyed while we're trying to 4128 * @cgrp might already have been destroyed while we're trying to
4084 * grab the mutexes. 4129 * grab the mutexes.
4085 */ 4130 */
4086 if (!cgroup_is_dead(cgrp)) 4131 if (!cgroup_is_dead(cgrp))
4087 ret = cgroup_destroy_locked(cgrp); 4132 ret = cgroup_destroy_locked(cgrp);
4088 4133
4089 mutex_unlock(&cgroup_mutex); 4134 mutex_unlock(&cgroup_mutex);
4090 mutex_unlock(&cgroup_tree_mutex); 4135 mutex_unlock(&cgroup_tree_mutex);
4091 4136
4092 kernfs_unbreak_active_protection(kn); 4137 kernfs_unbreak_active_protection(kn);
4093 cgroup_put(cgrp); 4138 cgroup_put(cgrp);
4094 return ret; 4139 return ret;
4095 } 4140 }
4096 4141
4097 static struct kernfs_syscall_ops cgroup_kf_syscall_ops = { 4142 static struct kernfs_syscall_ops cgroup_kf_syscall_ops = {
4098 .remount_fs = cgroup_remount, 4143 .remount_fs = cgroup_remount,
4099 .show_options = cgroup_show_options, 4144 .show_options = cgroup_show_options,
4100 .mkdir = cgroup_mkdir, 4145 .mkdir = cgroup_mkdir,
4101 .rmdir = cgroup_rmdir, 4146 .rmdir = cgroup_rmdir,
4102 .rename = cgroup_rename, 4147 .rename = cgroup_rename,
4103 }; 4148 };
4104 4149
4105 static void __init cgroup_init_subsys(struct cgroup_subsys *ss) 4150 static void __init cgroup_init_subsys(struct cgroup_subsys *ss)
4106 { 4151 {
4107 struct cgroup_subsys_state *css; 4152 struct cgroup_subsys_state *css;
4108 4153
4109 printk(KERN_INFO "Initializing cgroup subsys %s\n", ss->name); 4154 printk(KERN_INFO "Initializing cgroup subsys %s\n", ss->name);
4110 4155
4111 mutex_lock(&cgroup_tree_mutex); 4156 mutex_lock(&cgroup_tree_mutex);
4112 mutex_lock(&cgroup_mutex); 4157 mutex_lock(&cgroup_mutex);
4113 4158
4114 INIT_LIST_HEAD(&ss->cfts); 4159 INIT_LIST_HEAD(&ss->cfts);
4115 4160
4116 /* Create the root cgroup state for this subsystem */ 4161 /* Create the root cgroup state for this subsystem */
4117 ss->root = &cgrp_dfl_root; 4162 ss->root = &cgrp_dfl_root;
4118 css = ss->css_alloc(cgroup_css(&cgrp_dfl_root.cgrp, ss)); 4163 css = ss->css_alloc(cgroup_css(&cgrp_dfl_root.cgrp, ss));
4119 /* We don't handle early failures gracefully */ 4164 /* We don't handle early failures gracefully */
4120 BUG_ON(IS_ERR(css)); 4165 BUG_ON(IS_ERR(css));
4121 init_css(css, ss, &cgrp_dfl_root.cgrp); 4166 init_css(css, ss, &cgrp_dfl_root.cgrp);
4122 4167
4123 /* Update the init_css_set to contain a subsys 4168 /* Update the init_css_set to contain a subsys
4124 * pointer to this state - since the subsystem is 4169 * pointer to this state - since the subsystem is
4125 * newly registered, all tasks and hence the 4170 * newly registered, all tasks and hence the
4126 * init_css_set is in the subsystem's root cgroup. */ 4171 * init_css_set is in the subsystem's root cgroup. */
4127 init_css_set.subsys[ss->id] = css; 4172 init_css_set.subsys[ss->id] = css;
4128 4173
4129 need_forkexit_callback |= ss->fork || ss->exit; 4174 need_forkexit_callback |= ss->fork || ss->exit;
4130 4175
4131 /* At system boot, before all subsystems have been 4176 /* At system boot, before all subsystems have been
4132 * registered, no tasks have been forked, so we don't 4177 * registered, no tasks have been forked, so we don't
4133 * need to invoke fork callbacks here. */ 4178 * need to invoke fork callbacks here. */
4134 BUG_ON(!list_empty(&init_task.tasks)); 4179 BUG_ON(!list_empty(&init_task.tasks));
4135 4180
4136 BUG_ON(online_css(css)); 4181 BUG_ON(online_css(css));
4137 4182
4138 cgrp_dfl_root.subsys_mask |= 1 << ss->id; 4183 cgrp_dfl_root.subsys_mask |= 1 << ss->id;
4139 4184
4140 mutex_unlock(&cgroup_mutex); 4185 mutex_unlock(&cgroup_mutex);
4141 mutex_unlock(&cgroup_tree_mutex); 4186 mutex_unlock(&cgroup_tree_mutex);
4142 } 4187 }
4143 4188
4144 /** 4189 /**
4145 * cgroup_init_early - cgroup initialization at system boot 4190 * cgroup_init_early - cgroup initialization at system boot
4146 * 4191 *
4147 * Initialize cgroups at system boot, and initialize any 4192 * Initialize cgroups at system boot, and initialize any
4148 * subsystems that request early init. 4193 * subsystems that request early init.
4149 */ 4194 */
4150 int __init cgroup_init_early(void) 4195 int __init cgroup_init_early(void)
4151 { 4196 {
4152 static struct cgroup_sb_opts __initdata opts = 4197 static struct cgroup_sb_opts __initdata opts =
4153 { .flags = CGRP_ROOT_SANE_BEHAVIOR }; 4198 { .flags = CGRP_ROOT_SANE_BEHAVIOR };
4154 struct cgroup_subsys *ss; 4199 struct cgroup_subsys *ss;
4155 int i; 4200 int i;
4156 4201
4157 init_cgroup_root(&cgrp_dfl_root, &opts); 4202 init_cgroup_root(&cgrp_dfl_root, &opts);
4158 RCU_INIT_POINTER(init_task.cgroups, &init_css_set); 4203 RCU_INIT_POINTER(init_task.cgroups, &init_css_set);
4159 4204
4160 for_each_subsys(ss, i) { 4205 for_each_subsys(ss, i) {
4161 WARN(!ss->css_alloc || !ss->css_free || ss->name || ss->id, 4206 WARN(!ss->css_alloc || !ss->css_free || ss->name || ss->id,
4162 "invalid cgroup_subsys %d:%s css_alloc=%p css_free=%p name:id=%d:%s\n", 4207 "invalid cgroup_subsys %d:%s css_alloc=%p css_free=%p name:id=%d:%s\n",
4163 i, cgroup_subsys_name[i], ss->css_alloc, ss->css_free, 4208 i, cgroup_subsys_name[i], ss->css_alloc, ss->css_free,
4164 ss->id, ss->name); 4209 ss->id, ss->name);
4165 WARN(strlen(cgroup_subsys_name[i]) > MAX_CGROUP_TYPE_NAMELEN, 4210 WARN(strlen(cgroup_subsys_name[i]) > MAX_CGROUP_TYPE_NAMELEN,
4166 "cgroup_subsys_name %s too long\n", cgroup_subsys_name[i]); 4211 "cgroup_subsys_name %s too long\n", cgroup_subsys_name[i]);
4167 4212
4168 ss->id = i; 4213 ss->id = i;
4169 ss->name = cgroup_subsys_name[i]; 4214 ss->name = cgroup_subsys_name[i];
4170 4215
4171 if (ss->early_init) 4216 if (ss->early_init)
4172 cgroup_init_subsys(ss); 4217 cgroup_init_subsys(ss);
4173 } 4218 }
4174 return 0; 4219 return 0;
4175 } 4220 }
4176 4221
4177 /** 4222 /**
4178 * cgroup_init - cgroup initialization 4223 * cgroup_init - cgroup initialization
4179 * 4224 *
4180 * Register cgroup filesystem and /proc file, and initialize 4225 * Register cgroup filesystem and /proc file, and initialize
4181 * any subsystems that didn't request early init. 4226 * any subsystems that didn't request early init.
4182 */ 4227 */
4183 int __init cgroup_init(void) 4228 int __init cgroup_init(void)
4184 { 4229 {
4185 struct cgroup_subsys *ss; 4230 struct cgroup_subsys *ss;
4186 unsigned long key; 4231 unsigned long key;
4187 int ssid, err; 4232 int ssid, err;
4188 4233
4189 BUG_ON(cgroup_init_cftypes(NULL, cgroup_base_files)); 4234 BUG_ON(cgroup_init_cftypes(NULL, cgroup_base_files));
4190 4235
4191 mutex_lock(&cgroup_tree_mutex); 4236 mutex_lock(&cgroup_tree_mutex);
4192 mutex_lock(&cgroup_mutex); 4237 mutex_lock(&cgroup_mutex);
4193 4238
4194 /* Add init_css_set to the hash table */ 4239 /* Add init_css_set to the hash table */
4195 key = css_set_hash(init_css_set.subsys); 4240 key = css_set_hash(init_css_set.subsys);
4196 hash_add(css_set_table, &init_css_set.hlist, key); 4241 hash_add(css_set_table, &init_css_set.hlist, key);
4197 4242
4198 BUG_ON(cgroup_setup_root(&cgrp_dfl_root, 0)); 4243 BUG_ON(cgroup_setup_root(&cgrp_dfl_root, 0));
4199 4244
4200 mutex_unlock(&cgroup_mutex); 4245 mutex_unlock(&cgroup_mutex);
4201 mutex_unlock(&cgroup_tree_mutex); 4246 mutex_unlock(&cgroup_tree_mutex);
4202 4247
4203 for_each_subsys(ss, ssid) { 4248 for_each_subsys(ss, ssid) {
4204 if (!ss->early_init) 4249 if (!ss->early_init)
4205 cgroup_init_subsys(ss); 4250 cgroup_init_subsys(ss);
4206 4251
4207 /* 4252 /*
4208 * cftype registration needs kmalloc and can't be done 4253 * cftype registration needs kmalloc and can't be done
4209 * during early_init. Register base cftypes separately. 4254 * during early_init. Register base cftypes separately.
4210 */ 4255 */
4211 if (ss->base_cftypes) 4256 if (ss->base_cftypes)
4212 WARN_ON(cgroup_add_cftypes(ss, ss->base_cftypes)); 4257 WARN_ON(cgroup_add_cftypes(ss, ss->base_cftypes));
4213 } 4258 }
4214 4259
4215 cgroup_kobj = kobject_create_and_add("cgroup", fs_kobj); 4260 cgroup_kobj = kobject_create_and_add("cgroup", fs_kobj);
4216 if (!cgroup_kobj) 4261 if (!cgroup_kobj)
4217 return -ENOMEM; 4262 return -ENOMEM;
4218 4263
4219 err = register_filesystem(&cgroup_fs_type); 4264 err = register_filesystem(&cgroup_fs_type);
4220 if (err < 0) { 4265 if (err < 0) {
4221 kobject_put(cgroup_kobj); 4266 kobject_put(cgroup_kobj);
4222 return err; 4267 return err;
4223 } 4268 }
4224 4269
4225 proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations); 4270 proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations);
4226 return 0; 4271 return 0;
4227 } 4272 }
4228 4273
4229 static int __init cgroup_wq_init(void) 4274 static int __init cgroup_wq_init(void)
4230 { 4275 {
4231 /* 4276 /*
4232 * There isn't much point in executing destruction path in 4277 * There isn't much point in executing destruction path in
4233 * parallel. Good chunk is serialized with cgroup_mutex anyway. 4278 * parallel. Good chunk is serialized with cgroup_mutex anyway.
4234 * Use 1 for @max_active. 4279 * Use 1 for @max_active.
4235 * 4280 *
4236 * We would prefer to do this in cgroup_init() above, but that 4281 * We would prefer to do this in cgroup_init() above, but that
4237 * is called before init_workqueues(): so leave this until after. 4282 * is called before init_workqueues(): so leave this until after.
4238 */ 4283 */
4239 cgroup_destroy_wq = alloc_workqueue("cgroup_destroy", 0, 1); 4284 cgroup_destroy_wq = alloc_workqueue("cgroup_destroy", 0, 1);
4240 BUG_ON(!cgroup_destroy_wq); 4285 BUG_ON(!cgroup_destroy_wq);
4241 4286
4242 /* 4287 /*
4243 * Used to destroy pidlists and separate to serve as flush domain. 4288 * Used to destroy pidlists and separate to serve as flush domain.
4244 * Cap @max_active to 1 too. 4289 * Cap @max_active to 1 too.
4245 */ 4290 */
4246 cgroup_pidlist_destroy_wq = alloc_workqueue("cgroup_pidlist_destroy", 4291 cgroup_pidlist_destroy_wq = alloc_workqueue("cgroup_pidlist_destroy",
4247 0, 1); 4292 0, 1);
4248 BUG_ON(!cgroup_pidlist_destroy_wq); 4293 BUG_ON(!cgroup_pidlist_destroy_wq);
4249 4294
4250 return 0; 4295 return 0;
4251 } 4296 }
4252 core_initcall(cgroup_wq_init); 4297 core_initcall(cgroup_wq_init);
4253 4298
4254 /* 4299 /*
4255 * proc_cgroup_show() 4300 * proc_cgroup_show()
4256 * - Print task's cgroup paths into seq_file, one line for each hierarchy 4301 * - Print task's cgroup paths into seq_file, one line for each hierarchy
4257 * - Used for /proc/<pid>/cgroup. 4302 * - Used for /proc/<pid>/cgroup.
4258 */ 4303 */
4259 4304
4260 /* TODO: Use a proper seq_file iterator */ 4305 /* TODO: Use a proper seq_file iterator */
4261 int proc_cgroup_show(struct seq_file *m, void *v) 4306 int proc_cgroup_show(struct seq_file *m, void *v)
4262 { 4307 {
4263 struct pid *pid; 4308 struct pid *pid;
4264 struct task_struct *tsk; 4309 struct task_struct *tsk;
4265 char *buf, *path; 4310 char *buf, *path;
4266 int retval; 4311 int retval;
4267 struct cgroup_root *root; 4312 struct cgroup_root *root;
4268 4313
4269 retval = -ENOMEM; 4314 retval = -ENOMEM;
4270 buf = kmalloc(PATH_MAX, GFP_KERNEL); 4315 buf = kmalloc(PATH_MAX, GFP_KERNEL);
4271 if (!buf) 4316 if (!buf)
4272 goto out; 4317 goto out;
4273 4318
4274 retval = -ESRCH; 4319 retval = -ESRCH;
4275 pid = m->private; 4320 pid = m->private;
4276 tsk = get_pid_task(pid, PIDTYPE_PID); 4321 tsk = get_pid_task(pid, PIDTYPE_PID);
4277 if (!tsk) 4322 if (!tsk)
4278 goto out_free; 4323 goto out_free;
4279 4324
4280 retval = 0; 4325 retval = 0;
4281 4326
4282 mutex_lock(&cgroup_mutex); 4327 mutex_lock(&cgroup_mutex);
4283 down_read(&css_set_rwsem); 4328 down_read(&css_set_rwsem);
4284 4329
4285 for_each_root(root) { 4330 for_each_root(root) {
4286 struct cgroup_subsys *ss; 4331 struct cgroup_subsys *ss;
4287 struct cgroup *cgrp; 4332 struct cgroup *cgrp;
4288 int ssid, count = 0; 4333 int ssid, count = 0;
4289 4334
4290 if (root == &cgrp_dfl_root && !cgrp_dfl_root_visible) 4335 if (root == &cgrp_dfl_root && !cgrp_dfl_root_visible)
4291 continue; 4336 continue;
4292 4337
4293 seq_printf(m, "%d:", root->hierarchy_id); 4338 seq_printf(m, "%d:", root->hierarchy_id);
4294 for_each_subsys(ss, ssid) 4339 for_each_subsys(ss, ssid)
4295 if (root->subsys_mask & (1 << ssid)) 4340 if (root->subsys_mask & (1 << ssid))
4296 seq_printf(m, "%s%s", count++ ? "," : "", ss->name); 4341 seq_printf(m, "%s%s", count++ ? "," : "", ss->name);
4297 if (strlen(root->name)) 4342 if (strlen(root->name))
4298 seq_printf(m, "%sname=%s", count ? "," : "", 4343 seq_printf(m, "%sname=%s", count ? "," : "",
4299 root->name); 4344 root->name);
4300 seq_putc(m, ':'); 4345 seq_putc(m, ':');
4301 cgrp = task_cgroup_from_root(tsk, root); 4346 cgrp = task_cgroup_from_root(tsk, root);
4302 path = cgroup_path(cgrp, buf, PATH_MAX); 4347 path = cgroup_path(cgrp, buf, PATH_MAX);
4303 if (!path) { 4348 if (!path) {
4304 retval = -ENAMETOOLONG; 4349 retval = -ENAMETOOLONG;
4305 goto out_unlock; 4350 goto out_unlock;
4306 } 4351 }
4307 seq_puts(m, path); 4352 seq_puts(m, path);
4308 seq_putc(m, '\n'); 4353 seq_putc(m, '\n');
4309 } 4354 }
4310 4355
4311 out_unlock: 4356 out_unlock:
4312 up_read(&css_set_rwsem); 4357 up_read(&css_set_rwsem);
4313 mutex_unlock(&cgroup_mutex); 4358 mutex_unlock(&cgroup_mutex);
4314 put_task_struct(tsk); 4359 put_task_struct(tsk);
4315 out_free: 4360 out_free:
4316 kfree(buf); 4361 kfree(buf);
4317 out: 4362 out:
4318 return retval; 4363 return retval;
4319 } 4364 }
4320 4365
4321 /* Display information about each subsystem and each hierarchy */ 4366 /* Display information about each subsystem and each hierarchy */
4322 static int proc_cgroupstats_show(struct seq_file *m, void *v) 4367 static int proc_cgroupstats_show(struct seq_file *m, void *v)
4323 { 4368 {
4324 struct cgroup_subsys *ss; 4369 struct cgroup_subsys *ss;
4325 int i; 4370 int i;
4326 4371
4327 seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n"); 4372 seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
4328 /* 4373 /*
4329 * ideally we don't want subsystems moving around while we do this. 4374 * ideally we don't want subsystems moving around while we do this.
4330 * cgroup_mutex is also necessary to guarantee an atomic snapshot of 4375 * cgroup_mutex is also necessary to guarantee an atomic snapshot of
4331 * subsys/hierarchy state. 4376 * subsys/hierarchy state.
4332 */ 4377 */
4333 mutex_lock(&cgroup_mutex); 4378 mutex_lock(&cgroup_mutex);
4334 4379
4335 for_each_subsys(ss, i) 4380 for_each_subsys(ss, i)
4336 seq_printf(m, "%s\t%d\t%d\t%d\n", 4381 seq_printf(m, "%s\t%d\t%d\t%d\n",
4337 ss->name, ss->root->hierarchy_id, 4382 ss->name, ss->root->hierarchy_id,
4338 atomic_read(&ss->root->nr_cgrps), !ss->disabled); 4383 atomic_read(&ss->root->nr_cgrps), !ss->disabled);
4339 4384
4340 mutex_unlock(&cgroup_mutex); 4385 mutex_unlock(&cgroup_mutex);
4341 return 0; 4386 return 0;
4342 } 4387 }
4343 4388
4344 static int cgroupstats_open(struct inode *inode, struct file *file) 4389 static int cgroupstats_open(struct inode *inode, struct file *file)
4345 { 4390 {
4346 return single_open(file, proc_cgroupstats_show, NULL); 4391 return single_open(file, proc_cgroupstats_show, NULL);
4347 } 4392 }
4348 4393
4349 static const struct file_operations proc_cgroupstats_operations = { 4394 static const struct file_operations proc_cgroupstats_operations = {
4350 .open = cgroupstats_open, 4395 .open = cgroupstats_open,
4351 .read = seq_read, 4396 .read = seq_read,
4352 .llseek = seq_lseek, 4397 .llseek = seq_lseek,
4353 .release = single_release, 4398 .release = single_release,
4354 }; 4399 };
4355 4400
4356 /** 4401 /**
4357 * cgroup_fork - initialize cgroup related fields during copy_process() 4402 * cgroup_fork - initialize cgroup related fields during copy_process()
4358 * @child: pointer to task_struct of forking parent process. 4403 * @child: pointer to task_struct of forking parent process.
4359 * 4404 *
4360 * A task is associated with the init_css_set until cgroup_post_fork() 4405 * A task is associated with the init_css_set until cgroup_post_fork()
4361 * attaches it to the parent's css_set. Empty cg_list indicates that 4406 * attaches it to the parent's css_set. Empty cg_list indicates that
4362 * @child isn't holding reference to its css_set. 4407 * @child isn't holding reference to its css_set.
4363 */ 4408 */
4364 void cgroup_fork(struct task_struct *child) 4409 void cgroup_fork(struct task_struct *child)
4365 { 4410 {
4366 RCU_INIT_POINTER(child->cgroups, &init_css_set); 4411 RCU_INIT_POINTER(child->cgroups, &init_css_set);
4367 INIT_LIST_HEAD(&child->cg_list); 4412 INIT_LIST_HEAD(&child->cg_list);
4368 } 4413 }
4369 4414
4370 /** 4415 /**
4371 * cgroup_post_fork - called on a new task after adding it to the task list 4416 * cgroup_post_fork - called on a new task after adding it to the task list
4372 * @child: the task in question 4417 * @child: the task in question
4373 * 4418 *
4374 * Adds the task to the list running through its css_set if necessary and 4419 * Adds the task to the list running through its css_set if necessary and
4375 * call the subsystem fork() callbacks. Has to be after the task is 4420 * call the subsystem fork() callbacks. Has to be after the task is
4376 * visible on the task list in case we race with the first call to 4421 * visible on the task list in case we race with the first call to
4377 * cgroup_task_iter_start() - to guarantee that the new task ends up on its 4422 * cgroup_task_iter_start() - to guarantee that the new task ends up on its
4378 * list. 4423 * list.
4379 */ 4424 */
4380 void cgroup_post_fork(struct task_struct *child) 4425 void cgroup_post_fork(struct task_struct *child)
4381 { 4426 {
4382 struct cgroup_subsys *ss; 4427 struct cgroup_subsys *ss;
4383 int i; 4428 int i;
4384 4429
4385 /* 4430 /*
4386 * This may race against cgroup_enable_task_cg_links(). As that 4431 * This may race against cgroup_enable_task_cg_links(). As that
4387 * function sets use_task_css_set_links before grabbing 4432 * function sets use_task_css_set_links before grabbing
4388 * tasklist_lock and we just went through tasklist_lock to add 4433 * tasklist_lock and we just went through tasklist_lock to add
4389 * @child, it's guaranteed that either we see the set 4434 * @child, it's guaranteed that either we see the set
4390 * use_task_css_set_links or cgroup_enable_task_cg_lists() sees 4435 * use_task_css_set_links or cgroup_enable_task_cg_lists() sees
4391 * @child during its iteration. 4436 * @child during its iteration.
4392 * 4437 *
4393 * If we won the race, @child is associated with %current's 4438 * If we won the race, @child is associated with %current's
4394 * css_set. Grabbing css_set_rwsem guarantees both that the 4439 * css_set. Grabbing css_set_rwsem guarantees both that the
4395 * association is stable, and, on completion of the parent's 4440 * association is stable, and, on completion of the parent's
4396 * migration, @child is visible in the source of migration or 4441 * migration, @child is visible in the source of migration or
4397 * already in the destination cgroup. This guarantee is necessary 4442 * already in the destination cgroup. This guarantee is necessary
4398 * when implementing operations which need to migrate all tasks of 4443 * when implementing operations which need to migrate all tasks of
4399 * a cgroup to another. 4444 * a cgroup to another.
4400 * 4445 *
4401 * Note that if we lose to cgroup_enable_task_cg_links(), @child 4446 * Note that if we lose to cgroup_enable_task_cg_links(), @child
4402 * will remain in init_css_set. This is safe because all tasks are 4447 * will remain in init_css_set. This is safe because all tasks are
4403 * in the init_css_set before cg_links is enabled and there's no 4448 * in the init_css_set before cg_links is enabled and there's no
4404 * operation which transfers all tasks out of init_css_set. 4449 * operation which transfers all tasks out of init_css_set.
4405 */ 4450 */
4406 if (use_task_css_set_links) { 4451 if (use_task_css_set_links) {
4407 struct css_set *cset; 4452 struct css_set *cset;
4408 4453
4409 down_write(&css_set_rwsem); 4454 down_write(&css_set_rwsem);
4410 cset = task_css_set(current); 4455 cset = task_css_set(current);
4411 if (list_empty(&child->cg_list)) { 4456 if (list_empty(&child->cg_list)) {
4412 rcu_assign_pointer(child->cgroups, cset); 4457 rcu_assign_pointer(child->cgroups, cset);
4413 list_add(&child->cg_list, &cset->tasks); 4458 list_add(&child->cg_list, &cset->tasks);
4414 get_css_set(cset); 4459 get_css_set(cset);
4415 } 4460 }
4416 up_write(&css_set_rwsem); 4461 up_write(&css_set_rwsem);
4417 } 4462 }
4418 4463
4419 /* 4464 /*
4420 * Call ss->fork(). This must happen after @child is linked on 4465 * Call ss->fork(). This must happen after @child is linked on
4421 * css_set; otherwise, @child might change state between ->fork() 4466 * css_set; otherwise, @child might change state between ->fork()
4422 * and addition to css_set. 4467 * and addition to css_set.
4423 */ 4468 */
4424 if (need_forkexit_callback) { 4469 if (need_forkexit_callback) {
4425 for_each_subsys(ss, i) 4470 for_each_subsys(ss, i)
4426 if (ss->fork) 4471 if (ss->fork)
4427 ss->fork(child); 4472 ss->fork(child);
4428 } 4473 }
4429 } 4474 }
4430 4475
4431 /** 4476 /**
4432 * cgroup_exit - detach cgroup from exiting task 4477 * cgroup_exit - detach cgroup from exiting task
4433 * @tsk: pointer to task_struct of exiting process 4478 * @tsk: pointer to task_struct of exiting process
4434 * 4479 *
4435 * Description: Detach cgroup from @tsk and release it. 4480 * Description: Detach cgroup from @tsk and release it.
4436 * 4481 *
4437 * Note that cgroups marked notify_on_release force every task in 4482 * Note that cgroups marked notify_on_release force every task in
4438 * them to take the global cgroup_mutex mutex when exiting. 4483 * them to take the global cgroup_mutex mutex when exiting.
4439 * This could impact scaling on very large systems. Be reluctant to 4484 * This could impact scaling on very large systems. Be reluctant to
4440 * use notify_on_release cgroups where very high task exit scaling 4485 * use notify_on_release cgroups where very high task exit scaling
4441 * is required on large systems. 4486 * is required on large systems.
4442 * 4487 *
4443 * We set the exiting tasks cgroup to the root cgroup (top_cgroup). We 4488 * We set the exiting tasks cgroup to the root cgroup (top_cgroup). We
4444 * call cgroup_exit() while the task is still competent to handle 4489 * call cgroup_exit() while the task is still competent to handle
4445 * notify_on_release(), then leave the task attached to the root cgroup in 4490 * notify_on_release(), then leave the task attached to the root cgroup in
4446 * each hierarchy for the remainder of its exit. No need to bother with 4491 * each hierarchy for the remainder of its exit. No need to bother with
4447 * init_css_set refcnting. init_css_set never goes away and we can't race 4492 * init_css_set refcnting. init_css_set never goes away and we can't race
4448 * with migration path - PF_EXITING is visible to migration path. 4493 * with migration path - PF_EXITING is visible to migration path.
4449 */ 4494 */
4450 void cgroup_exit(struct task_struct *tsk) 4495 void cgroup_exit(struct task_struct *tsk)
4451 { 4496 {
4452 struct cgroup_subsys *ss; 4497 struct cgroup_subsys *ss;
4453 struct css_set *cset; 4498 struct css_set *cset;
4454 bool put_cset = false; 4499 bool put_cset = false;
4455 int i; 4500 int i;
4456 4501
4457 /* 4502 /*
4458 * Unlink from @tsk from its css_set. As migration path can't race 4503 * Unlink from @tsk from its css_set. As migration path can't race
4459 * with us, we can check cg_list without grabbing css_set_rwsem. 4504 * with us, we can check cg_list without grabbing css_set_rwsem.
4460 */ 4505 */
4461 if (!list_empty(&tsk->cg_list)) { 4506 if (!list_empty(&tsk->cg_list)) {
4462 down_write(&css_set_rwsem); 4507 down_write(&css_set_rwsem);
4463 list_del_init(&tsk->cg_list); 4508 list_del_init(&tsk->cg_list);
4464 up_write(&css_set_rwsem); 4509 up_write(&css_set_rwsem);
4465 put_cset = true; 4510 put_cset = true;
4466 } 4511 }
4467 4512
4468 /* Reassign the task to the init_css_set. */ 4513 /* Reassign the task to the init_css_set. */
4469 cset = task_css_set(tsk); 4514 cset = task_css_set(tsk);
4470 RCU_INIT_POINTER(tsk->cgroups, &init_css_set); 4515 RCU_INIT_POINTER(tsk->cgroups, &init_css_set);
4471 4516
4472 if (need_forkexit_callback) { 4517 if (need_forkexit_callback) {
4473 /* see cgroup_post_fork() for details */ 4518 /* see cgroup_post_fork() for details */
4474 for_each_subsys(ss, i) { 4519 for_each_subsys(ss, i) {
4475 if (ss->exit) { 4520 if (ss->exit) {
4476 struct cgroup_subsys_state *old_css = cset->subsys[i]; 4521 struct cgroup_subsys_state *old_css = cset->subsys[i];
4477 struct cgroup_subsys_state *css = task_css(tsk, i); 4522 struct cgroup_subsys_state *css = task_css(tsk, i);
4478 4523
4479 ss->exit(css, old_css, tsk); 4524 ss->exit(css, old_css, tsk);
4480 } 4525 }
4481 } 4526 }
4482 } 4527 }
4483 4528
4484 if (put_cset) 4529 if (put_cset)
4485 put_css_set(cset, true); 4530 put_css_set(cset, true);
4486 } 4531 }
4487 4532
4488 static void check_for_release(struct cgroup *cgrp) 4533 static void check_for_release(struct cgroup *cgrp)
4489 { 4534 {
4490 if (cgroup_is_releasable(cgrp) && 4535 if (cgroup_is_releasable(cgrp) &&
4491 list_empty(&cgrp->cset_links) && list_empty(&cgrp->children)) { 4536 list_empty(&cgrp->cset_links) && list_empty(&cgrp->children)) {
4492 /* 4537 /*
4493 * Control Group is currently removeable. If it's not 4538 * Control Group is currently removeable. If it's not
4494 * already queued for a userspace notification, queue 4539 * already queued for a userspace notification, queue
4495 * it now 4540 * it now
4496 */ 4541 */
4497 int need_schedule_work = 0; 4542 int need_schedule_work = 0;
4498 4543
4499 raw_spin_lock(&release_list_lock); 4544 raw_spin_lock(&release_list_lock);
4500 if (!cgroup_is_dead(cgrp) && 4545 if (!cgroup_is_dead(cgrp) &&
4501 list_empty(&cgrp->release_list)) { 4546 list_empty(&cgrp->release_list)) {
4502 list_add(&cgrp->release_list, &release_list); 4547 list_add(&cgrp->release_list, &release_list);
4503 need_schedule_work = 1; 4548 need_schedule_work = 1;
4504 } 4549 }
4505 raw_spin_unlock(&release_list_lock); 4550 raw_spin_unlock(&release_list_lock);
4506 if (need_schedule_work) 4551 if (need_schedule_work)
4507 schedule_work(&release_agent_work); 4552 schedule_work(&release_agent_work);
4508 } 4553 }
4509 } 4554 }
4510 4555
4511 /* 4556 /*
4512 * Notify userspace when a cgroup is released, by running the 4557 * Notify userspace when a cgroup is released, by running the
4513 * configured release agent with the name of the cgroup (path 4558 * configured release agent with the name of the cgroup (path
4514 * relative to the root of cgroup file system) as the argument. 4559 * relative to the root of cgroup file system) as the argument.
4515 * 4560 *
4516 * Most likely, this user command will try to rmdir this cgroup. 4561 * Most likely, this user command will try to rmdir this cgroup.
4517 * 4562 *
4518 * This races with the possibility that some other task will be 4563 * This races with the possibility that some other task will be
4519 * attached to this cgroup before it is removed, or that some other 4564 * attached to this cgroup before it is removed, or that some other
4520 * user task will 'mkdir' a child cgroup of this cgroup. That's ok. 4565 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
4521 * The presumed 'rmdir' will fail quietly if this cgroup is no longer 4566 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
4522 * unused, and this cgroup will be reprieved from its death sentence, 4567 * unused, and this cgroup will be reprieved from its death sentence,
4523 * to continue to serve a useful existence. Next time it's released, 4568 * to continue to serve a useful existence. Next time it's released,
4524 * we will get notified again, if it still has 'notify_on_release' set. 4569 * we will get notified again, if it still has 'notify_on_release' set.
4525 * 4570 *
4526 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which 4571 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
4527 * means only wait until the task is successfully execve()'d. The 4572 * means only wait until the task is successfully execve()'d. The
4528 * separate release agent task is forked by call_usermodehelper(), 4573 * separate release agent task is forked by call_usermodehelper(),
4529 * then control in this thread returns here, without waiting for the 4574 * then control in this thread returns here, without waiting for the
4530 * release agent task. We don't bother to wait because the caller of 4575 * release agent task. We don't bother to wait because the caller of
4531 * this routine has no use for the exit status of the release agent 4576 * this routine has no use for the exit status of the release agent
4532 * task, so no sense holding our caller up for that. 4577 * task, so no sense holding our caller up for that.
4533 */ 4578 */
4534 static void cgroup_release_agent(struct work_struct *work) 4579 static void cgroup_release_agent(struct work_struct *work)
4535 { 4580 {
4536 BUG_ON(work != &release_agent_work); 4581 BUG_ON(work != &release_agent_work);
4537 mutex_lock(&cgroup_mutex); 4582 mutex_lock(&cgroup_mutex);
4538 raw_spin_lock(&release_list_lock); 4583 raw_spin_lock(&release_list_lock);
4539 while (!list_empty(&release_list)) { 4584 while (!list_empty(&release_list)) {
4540 char *argv[3], *envp[3]; 4585 char *argv[3], *envp[3];
4541 int i; 4586 int i;
4542 char *pathbuf = NULL, *agentbuf = NULL, *path; 4587 char *pathbuf = NULL, *agentbuf = NULL, *path;
4543 struct cgroup *cgrp = list_entry(release_list.next, 4588 struct cgroup *cgrp = list_entry(release_list.next,
4544 struct cgroup, 4589 struct cgroup,
4545 release_list); 4590 release_list);
4546 list_del_init(&cgrp->release_list); 4591 list_del_init(&cgrp->release_list);
4547 raw_spin_unlock(&release_list_lock); 4592 raw_spin_unlock(&release_list_lock);
4548 pathbuf = kmalloc(PATH_MAX, GFP_KERNEL); 4593 pathbuf = kmalloc(PATH_MAX, GFP_KERNEL);
4549 if (!pathbuf) 4594 if (!pathbuf)
4550 goto continue_free; 4595 goto continue_free;
4551 path = cgroup_path(cgrp, pathbuf, PATH_MAX); 4596 path = cgroup_path(cgrp, pathbuf, PATH_MAX);
4552 if (!path) 4597 if (!path)
4553 goto continue_free; 4598 goto continue_free;
4554 agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL); 4599 agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL);
4555 if (!agentbuf) 4600 if (!agentbuf)
4556 goto continue_free; 4601 goto continue_free;
4557 4602
4558 i = 0; 4603 i = 0;
4559 argv[i++] = agentbuf; 4604 argv[i++] = agentbuf;
4560 argv[i++] = path; 4605 argv[i++] = path;
4561 argv[i] = NULL; 4606 argv[i] = NULL;
4562 4607
4563 i = 0; 4608 i = 0;
4564 /* minimal command environment */ 4609 /* minimal command environment */
4565 envp[i++] = "HOME=/"; 4610 envp[i++] = "HOME=/";
4566 envp[i++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin"; 4611 envp[i++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
4567 envp[i] = NULL; 4612 envp[i] = NULL;
4568 4613
4569 /* Drop the lock while we invoke the usermode helper, 4614 /* Drop the lock while we invoke the usermode helper,
4570 * since the exec could involve hitting disk and hence 4615 * since the exec could involve hitting disk and hence
4571 * be a slow process */ 4616 * be a slow process */
4572 mutex_unlock(&cgroup_mutex); 4617 mutex_unlock(&cgroup_mutex);
4573 call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC); 4618 call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC);
4574 mutex_lock(&cgroup_mutex); 4619 mutex_lock(&cgroup_mutex);
4575 continue_free: 4620 continue_free:
4576 kfree(pathbuf); 4621 kfree(pathbuf);
4577 kfree(agentbuf); 4622 kfree(agentbuf);
4578 raw_spin_lock(&release_list_lock); 4623 raw_spin_lock(&release_list_lock);
4579 } 4624 }
4580 raw_spin_unlock(&release_list_lock); 4625 raw_spin_unlock(&release_list_lock);
4581 mutex_unlock(&cgroup_mutex); 4626 mutex_unlock(&cgroup_mutex);
4582 } 4627 }
4583 4628
4584 static int __init cgroup_disable(char *str) 4629 static int __init cgroup_disable(char *str)
4585 { 4630 {
4586 struct cgroup_subsys *ss; 4631 struct cgroup_subsys *ss;
4587 char *token; 4632 char *token;
4588 int i; 4633 int i;
4589 4634
4590 while ((token = strsep(&str, ",")) != NULL) { 4635 while ((token = strsep(&str, ",")) != NULL) {
4591 if (!*token) 4636 if (!*token)
4592 continue; 4637 continue;
4593 4638
4594 for_each_subsys(ss, i) { 4639 for_each_subsys(ss, i) {
4595 if (!strcmp(token, ss->name)) { 4640 if (!strcmp(token, ss->name)) {
4596 ss->disabled = 1; 4641 ss->disabled = 1;
4597 printk(KERN_INFO "Disabling %s control group" 4642 printk(KERN_INFO "Disabling %s control group"
4598 " subsystem\n", ss->name); 4643 " subsystem\n", ss->name);
4599 break; 4644 break;
4600 } 4645 }
4601 } 4646 }
4602 } 4647 }
4603 return 1; 4648 return 1;
4604 } 4649 }
4605 __setup("cgroup_disable=", cgroup_disable); 4650 __setup("cgroup_disable=", cgroup_disable);
4606 4651
4607 /** 4652 /**
4608 * css_tryget_from_dir - get corresponding css from the dentry of a cgroup dir 4653 * css_tryget_from_dir - get corresponding css from the dentry of a cgroup dir
4609 * @dentry: directory dentry of interest 4654 * @dentry: directory dentry of interest
4610 * @ss: subsystem of interest 4655 * @ss: subsystem of interest
4611 * 4656 *
4612 * If @dentry is a directory for a cgroup which has @ss enabled on it, try 4657 * If @dentry is a directory for a cgroup which has @ss enabled on it, try
4613 * to get the corresponding css and return it. If such css doesn't exist 4658 * to get the corresponding css and return it. If such css doesn't exist
4614 * or can't be pinned, an ERR_PTR value is returned. 4659 * or can't be pinned, an ERR_PTR value is returned.
4615 */ 4660 */
4616 struct cgroup_subsys_state *css_tryget_from_dir(struct dentry *dentry, 4661 struct cgroup_subsys_state *css_tryget_from_dir(struct dentry *dentry,
4617 struct cgroup_subsys *ss) 4662 struct cgroup_subsys *ss)
4618 { 4663 {
4619 struct kernfs_node *kn = kernfs_node_from_dentry(dentry); 4664 struct kernfs_node *kn = kernfs_node_from_dentry(dentry);
4620 struct cgroup_subsys_state *css = NULL; 4665 struct cgroup_subsys_state *css = NULL;
4621 struct cgroup *cgrp; 4666 struct cgroup *cgrp;
4622 4667
4623 /* is @dentry a cgroup dir? */ 4668 /* is @dentry a cgroup dir? */
4624 if (dentry->d_sb->s_type != &cgroup_fs_type || !kn || 4669 if (dentry->d_sb->s_type != &cgroup_fs_type || !kn ||
4625 kernfs_type(kn) != KERNFS_DIR) 4670 kernfs_type(kn) != KERNFS_DIR)
4626 return ERR_PTR(-EBADF); 4671 return ERR_PTR(-EBADF);
4627 4672
4628 rcu_read_lock(); 4673 rcu_read_lock();
4629 4674
4630 /* 4675 /*
4631 * This path doesn't originate from kernfs and @kn could already 4676 * This path doesn't originate from kernfs and @kn could already
4632 * have been or be removed at any point. @kn->priv is RCU 4677 * have been or be removed at any point. @kn->priv is RCU
4633 * protected for this access. See destroy_locked() for details. 4678 * protected for this access. See destroy_locked() for details.
4634 */ 4679 */
4635 cgrp = rcu_dereference(kn->priv); 4680 cgrp = rcu_dereference(kn->priv);
4636 if (cgrp) 4681 if (cgrp)
4637 css = cgroup_css(cgrp, ss); 4682 css = cgroup_css(cgrp, ss);
4638 4683
4639 if (!css || !css_tryget(css)) 4684 if (!css || !css_tryget(css))
4640 css = ERR_PTR(-ENOENT); 4685 css = ERR_PTR(-ENOENT);
4641 4686
4642 rcu_read_unlock(); 4687 rcu_read_unlock();
4643 return css; 4688 return css;
4644 } 4689 }
4645 4690
4646 /** 4691 /**
4647 * css_from_id - lookup css by id 4692 * css_from_id - lookup css by id
4648 * @id: the cgroup id 4693 * @id: the cgroup id
4649 * @ss: cgroup subsys to be looked into 4694 * @ss: cgroup subsys to be looked into
4650 * 4695 *
4651 * Returns the css if there's valid one with @id, otherwise returns NULL. 4696 * Returns the css if there's valid one with @id, otherwise returns NULL.
4652 * Should be called under rcu_read_lock(). 4697 * Should be called under rcu_read_lock().
4653 */ 4698 */
4654 struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss) 4699 struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss)
4655 { 4700 {
4656 struct cgroup *cgrp; 4701 struct cgroup *cgrp;
4657 4702
4658 cgroup_assert_mutexes_or_rcu_locked(); 4703 cgroup_assert_mutexes_or_rcu_locked();
4659 4704
4660 cgrp = idr_find(&ss->root->cgroup_idr, id); 4705 cgrp = idr_find(&ss->root->cgroup_idr, id);
4661 if (cgrp) 4706 if (cgrp)
4662 return cgroup_css(cgrp, ss); 4707 return cgroup_css(cgrp, ss);
4663 return NULL; 4708 return NULL;
4664 } 4709 }
4665 4710
4666 #ifdef CONFIG_CGROUP_DEBUG 4711 #ifdef CONFIG_CGROUP_DEBUG
4667 static struct cgroup_subsys_state * 4712 static struct cgroup_subsys_state *
4668 debug_css_alloc(struct cgroup_subsys_state *parent_css) 4713 debug_css_alloc(struct cgroup_subsys_state *parent_css)
4669 { 4714 {
4670 struct cgroup_subsys_state *css = kzalloc(sizeof(*css), GFP_KERNEL); 4715 struct cgroup_subsys_state *css = kzalloc(sizeof(*css), GFP_KERNEL);
4671 4716
4672 if (!css) 4717 if (!css)
4673 return ERR_PTR(-ENOMEM); 4718 return ERR_PTR(-ENOMEM);
4674 4719
4675 return css; 4720 return css;
4676 } 4721 }
4677 4722
4678 static void debug_css_free(struct cgroup_subsys_state *css) 4723 static void debug_css_free(struct cgroup_subsys_state *css)
4679 { 4724 {
4680 kfree(css); 4725 kfree(css);
4681 } 4726 }
4682 4727
4683 static u64 debug_taskcount_read(struct cgroup_subsys_state *css, 4728 static u64 debug_taskcount_read(struct cgroup_subsys_state *css,
4684 struct cftype *cft) 4729 struct cftype *cft)
4685 { 4730 {
4686 return cgroup_task_count(css->cgroup); 4731 return cgroup_task_count(css->cgroup);
4687 } 4732 }
4688 4733
4689 static u64 current_css_set_read(struct cgroup_subsys_state *css, 4734 static u64 current_css_set_read(struct cgroup_subsys_state *css,
4690 struct cftype *cft) 4735 struct cftype *cft)
4691 { 4736 {
4692 return (u64)(unsigned long)current->cgroups; 4737 return (u64)(unsigned long)current->cgroups;
4693 } 4738 }
4694 4739
4695 static u64 current_css_set_refcount_read(struct cgroup_subsys_state *css, 4740 static u64 current_css_set_refcount_read(struct cgroup_subsys_state *css,
4696 struct cftype *cft) 4741 struct cftype *cft)
4697 { 4742 {
4698 u64 count; 4743 u64 count;
4699 4744
4700 rcu_read_lock(); 4745 rcu_read_lock();
4701 count = atomic_read(&task_css_set(current)->refcount); 4746 count = atomic_read(&task_css_set(current)->refcount);
4702 rcu_read_unlock(); 4747 rcu_read_unlock();
4703 return count; 4748 return count;
4704 } 4749 }
4705 4750
4706 static int current_css_set_cg_links_read(struct seq_file *seq, void *v) 4751 static int current_css_set_cg_links_read(struct seq_file *seq, void *v)
4707 { 4752 {
4708 struct cgrp_cset_link *link; 4753 struct cgrp_cset_link *link;
4709 struct css_set *cset; 4754 struct css_set *cset;
4710 char *name_buf; 4755 char *name_buf;
4711 4756
4712 name_buf = kmalloc(NAME_MAX + 1, GFP_KERNEL); 4757 name_buf = kmalloc(NAME_MAX + 1, GFP_KERNEL);
4713 if (!name_buf) 4758 if (!name_buf)
4714 return -ENOMEM; 4759 return -ENOMEM;
4715 4760
4716 down_read(&css_set_rwsem); 4761 down_read(&css_set_rwsem);
4717 rcu_read_lock(); 4762 rcu_read_lock();
4718 cset = rcu_dereference(current->cgroups); 4763 cset = rcu_dereference(current->cgroups);
4719 list_for_each_entry(link, &cset->cgrp_links, cgrp_link) { 4764 list_for_each_entry(link, &cset->cgrp_links, cgrp_link) {
4720 struct cgroup *c = link->cgrp; 4765 struct cgroup *c = link->cgrp;
4721 4766
4722 cgroup_name(c, name_buf, NAME_MAX + 1); 4767 cgroup_name(c, name_buf, NAME_MAX + 1);
4723 seq_printf(seq, "Root %d group %s\n", 4768 seq_printf(seq, "Root %d group %s\n",
4724 c->root->hierarchy_id, name_buf); 4769 c->root->hierarchy_id, name_buf);
4725 } 4770 }
4726 rcu_read_unlock(); 4771 rcu_read_unlock();
4727 up_read(&css_set_rwsem); 4772 up_read(&css_set_rwsem);
4728 kfree(name_buf); 4773 kfree(name_buf);
4729 return 0; 4774 return 0;
4730 } 4775 }
4731 4776
4732 #define MAX_TASKS_SHOWN_PER_CSS 25 4777 #define MAX_TASKS_SHOWN_PER_CSS 25
4733 static int cgroup_css_links_read(struct seq_file *seq, void *v) 4778 static int cgroup_css_links_read(struct seq_file *seq, void *v)
4734 { 4779 {
4735 struct cgroup_subsys_state *css = seq_css(seq); 4780 struct cgroup_subsys_state *css = seq_css(seq);
4736 struct cgrp_cset_link *link; 4781 struct cgrp_cset_link *link;
4737 4782
4738 down_read(&css_set_rwsem); 4783 down_read(&css_set_rwsem);
4739 list_for_each_entry(link, &css->cgroup->cset_links, cset_link) { 4784 list_for_each_entry(link, &css->cgroup->cset_links, cset_link) {
4740 struct css_set *cset = link->cset; 4785 struct css_set *cset = link->cset;
4741 struct task_struct *task; 4786 struct task_struct *task;
4742 int count = 0; 4787 int count = 0;
4743 4788
4744 seq_printf(seq, "css_set %p\n", cset); 4789 seq_printf(seq, "css_set %p\n", cset);
4745 4790
4746 list_for_each_entry(task, &cset->tasks, cg_list) { 4791 list_for_each_entry(task, &cset->tasks, cg_list) {
4747 if (count++ > MAX_TASKS_SHOWN_PER_CSS) 4792 if (count++ > MAX_TASKS_SHOWN_PER_CSS)
4748 goto overflow; 4793 goto overflow;
4749 seq_printf(seq, " task %d\n", task_pid_vnr(task)); 4794 seq_printf(seq, " task %d\n", task_pid_vnr(task));
4750 } 4795 }
4751 4796
4752 list_for_each_entry(task, &cset->mg_tasks, cg_list) { 4797 list_for_each_entry(task, &cset->mg_tasks, cg_list) {
4753 if (count++ > MAX_TASKS_SHOWN_PER_CSS) 4798 if (count++ > MAX_TASKS_SHOWN_PER_CSS)
4754 goto overflow; 4799 goto overflow;
4755 seq_printf(seq, " task %d\n", task_pid_vnr(task)); 4800 seq_printf(seq, " task %d\n", task_pid_vnr(task));
4756 } 4801 }
4757 continue; 4802 continue;
4758 overflow: 4803 overflow:
4759 seq_puts(seq, " ...\n"); 4804 seq_puts(seq, " ...\n");
4760 } 4805 }
4761 up_read(&css_set_rwsem); 4806 up_read(&css_set_rwsem);
4762 return 0; 4807 return 0;
4763 } 4808 }
4764 4809
4765 static u64 releasable_read(struct cgroup_subsys_state *css, struct cftype *cft) 4810 static u64 releasable_read(struct cgroup_subsys_state *css, struct cftype *cft)
4766 { 4811 {
4767 return test_bit(CGRP_RELEASABLE, &css->cgroup->flags); 4812 return test_bit(CGRP_RELEASABLE, &css->cgroup->flags);
4768 } 4813 }
4769 4814
4770 static struct cftype debug_files[] = { 4815 static struct cftype debug_files[] = {
4771 { 4816 {
4772 .name = "taskcount", 4817 .name = "taskcount",
4773 .read_u64 = debug_taskcount_read, 4818 .read_u64 = debug_taskcount_read,
4774 }, 4819 },
4775 4820
4776 { 4821 {
4777 .name = "current_css_set", 4822 .name = "current_css_set",
4778 .read_u64 = current_css_set_read, 4823 .read_u64 = current_css_set_read,
4779 }, 4824 },
4780 4825
4781 { 4826 {
4782 .name = "current_css_set_refcount", 4827 .name = "current_css_set_refcount",
4783 .read_u64 = current_css_set_refcount_read, 4828 .read_u64 = current_css_set_refcount_read,
4784 }, 4829 },
4785 4830
4786 { 4831 {
4787 .name = "current_css_set_cg_links", 4832 .name = "current_css_set_cg_links",
4788 .seq_show = current_css_set_cg_links_read, 4833 .seq_show = current_css_set_cg_links_read,
4789 }, 4834 },
4790 4835
4791 { 4836 {
4792 .name = "cgroup_css_links", 4837 .name = "cgroup_css_links",
4793 .seq_show = cgroup_css_links_read, 4838 .seq_show = cgroup_css_links_read,
4794 }, 4839 },
4795 4840
4796 { 4841 {
4797 .name = "releasable", 4842 .name = "releasable",
4798 .read_u64 = releasable_read, 4843 .read_u64 = releasable_read,
4799 }, 4844 },
4800 4845
4801 { } /* terminate */ 4846 { } /* terminate */
4802 }; 4847 };
4803 4848
4804 struct cgroup_subsys debug_cgrp_subsys = { 4849 struct cgroup_subsys debug_cgrp_subsys = {
4805 .css_alloc = debug_css_alloc, 4850 .css_alloc = debug_css_alloc,
4806 .css_free = debug_css_free, 4851 .css_free = debug_css_free,