Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1 | /* |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 2 | * Generic process-grouping system. |
| 3 | * |
| 4 | * Based originally on the cpuset system, extracted by Paul Menage |
| 5 | * Copyright (C) 2006 Google, Inc |
| 6 | * |
| 7 | * Copyright notices from the original cpuset code: |
| 8 | * -------------------------------------------------- |
| 9 | * Copyright (C) 2003 BULL SA. |
| 10 | * Copyright (C) 2004-2006 Silicon Graphics, Inc. |
| 11 | * |
| 12 | * Portions derived from Patrick Mochel's sysfs code. |
| 13 | * sysfs is Copyright (c) 2001-3 Patrick Mochel |
| 14 | * |
| 15 | * 2003-10-10 Written by Simon Derr. |
| 16 | * 2003-10-22 Updates by Stephen Hemminger. |
| 17 | * 2004 May-July Rework by Paul Jackson. |
| 18 | * --------------------------------------------------- |
| 19 | * |
| 20 | * This file is subject to the terms and conditions of the GNU General Public |
| 21 | * License. See the file COPYING in the main directory of the Linux |
| 22 | * distribution for more details. |
| 23 | */ |
| 24 | |
| 25 | #include <linux/cgroup.h> |
| 26 | #include <linux/errno.h> |
| 27 | #include <linux/fs.h> |
| 28 | #include <linux/kernel.h> |
| 29 | #include <linux/list.h> |
| 30 | #include <linux/mm.h> |
| 31 | #include <linux/mutex.h> |
| 32 | #include <linux/mount.h> |
| 33 | #include <linux/pagemap.h> |
Paul Menage | a424316 | 2007-10-18 23:39:35 -0700 | [diff] [blame] | 34 | #include <linux/proc_fs.h> |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 35 | #include <linux/rcupdate.h> |
| 36 | #include <linux/sched.h> |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 37 | #include <linux/backing-dev.h> |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 38 | #include <linux/seq_file.h> |
| 39 | #include <linux/slab.h> |
| 40 | #include <linux/magic.h> |
| 41 | #include <linux/spinlock.h> |
| 42 | #include <linux/string.h> |
Paul Menage | bbcb81d | 2007-10-18 23:39:32 -0700 | [diff] [blame] | 43 | #include <linux/sort.h> |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 44 | #include <linux/kmod.h> |
Balbir Singh | 846c7bb | 2007-10-18 23:39:44 -0700 | [diff] [blame] | 45 | #include <linux/delayacct.h> |
| 46 | #include <linux/cgroupstats.h> |
Li Zefan | 472b105 | 2008-04-29 01:00:11 -0700 | [diff] [blame] | 47 | #include <linux/hash.h> |
Al Viro | 3f8206d | 2008-07-26 03:46:43 -0400 | [diff] [blame] | 48 | #include <linux/namei.h> |
Alessio Igor Bogani | 337eb00 | 2009-05-12 15:10:54 +0200 | [diff] [blame] | 49 | #include <linux/smp_lock.h> |
Li Zefan | 096b7fe | 2009-07-29 15:04:04 -0700 | [diff] [blame] | 50 | #include <linux/pid_namespace.h> |
Balbir Singh | 846c7bb | 2007-10-18 23:39:44 -0700 | [diff] [blame] | 51 | |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 52 | #include <asm/atomic.h> |
| 53 | |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 54 | static DEFINE_MUTEX(cgroup_mutex); |
| 55 | |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 56 | /* Generate an array of cgroup subsystem pointers */ |
| 57 | #define SUBSYS(_x) &_x ## _subsys, |
| 58 | |
| 59 | static struct cgroup_subsys *subsys[] = { |
| 60 | #include <linux/cgroup_subsys.h> |
| 61 | }; |
| 62 | |
| 63 | /* |
| 64 | * A cgroupfs_root represents the root of a cgroup hierarchy, |
| 65 | * and may be associated with a superblock to form an active |
| 66 | * hierarchy |
| 67 | */ |
| 68 | struct cgroupfs_root { |
| 69 | struct super_block *sb; |
| 70 | |
| 71 | /* |
| 72 | * The bitmask of subsystems intended to be attached to this |
| 73 | * hierarchy |
| 74 | */ |
| 75 | unsigned long subsys_bits; |
| 76 | |
| 77 | /* The bitmask of subsystems currently attached to this hierarchy */ |
| 78 | unsigned long actual_subsys_bits; |
| 79 | |
| 80 | /* A list running through the attached subsystems */ |
| 81 | struct list_head subsys_list; |
| 82 | |
| 83 | /* The root cgroup for this hierarchy */ |
| 84 | struct cgroup top_cgroup; |
| 85 | |
| 86 | /* Tracks how many cgroups are currently defined in hierarchy.*/ |
| 87 | int number_of_cgroups; |
| 88 | |
Li Zefan | e5f6a86 | 2009-01-07 18:07:41 -0800 | [diff] [blame] | 89 | /* A list running through the active hierarchies */ |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 90 | struct list_head root_list; |
| 91 | |
| 92 | /* Hierarchy-specific flags */ |
| 93 | unsigned long flags; |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 94 | |
Paul Menage | e788e06 | 2008-07-25 01:46:59 -0700 | [diff] [blame] | 95 | /* The path to use for release notifications. */ |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 96 | char release_agent_path[PATH_MAX]; |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 97 | }; |
| 98 | |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 99 | /* |
| 100 | * The "rootnode" hierarchy is the "dummy hierarchy", reserved for the |
| 101 | * subsystems that are otherwise unattached - it never has more than a |
| 102 | * single cgroup, and all tasks are part of that cgroup. |
| 103 | */ |
| 104 | static struct cgroupfs_root rootnode; |
| 105 | |
KAMEZAWA Hiroyuki | 38460b4 | 2009-04-02 16:57:25 -0700 | [diff] [blame] | 106 | /* |
| 107 | * CSS ID -- ID per subsys's Cgroup Subsys State(CSS). used only when |
| 108 | * cgroup_subsys->use_id != 0. |
| 109 | */ |
| 110 | #define CSS_ID_MAX (65535) |
| 111 | struct css_id { |
| 112 | /* |
| 113 | * The css to which this ID points. This pointer is set to valid value |
| 114 | * after cgroup is populated. If cgroup is removed, this will be NULL. |
| 115 | * This pointer is expected to be RCU-safe because destroy() |
| 116 | * is called after synchronize_rcu(). But for safe use, css_is_removed() |
| 117 | * css_tryget() should be used for avoiding race. |
| 118 | */ |
| 119 | struct cgroup_subsys_state *css; |
| 120 | /* |
| 121 | * ID of this css. |
| 122 | */ |
| 123 | unsigned short id; |
| 124 | /* |
| 125 | * Depth in hierarchy which this ID belongs to. |
| 126 | */ |
| 127 | unsigned short depth; |
| 128 | /* |
| 129 | * ID is freed by RCU. (and lookup routine is RCU safe.) |
| 130 | */ |
| 131 | struct rcu_head rcu_head; |
| 132 | /* |
| 133 | * Hierarchy of CSS ID belongs to. |
| 134 | */ |
| 135 | unsigned short stack[0]; /* Array of Length (depth+1) */ |
| 136 | }; |
| 137 | |
| 138 | |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 139 | /* The list of hierarchy roots */ |
| 140 | |
| 141 | static LIST_HEAD(roots); |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 142 | static int root_count; |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 143 | |
| 144 | /* dummytop is a shorthand for the dummy hierarchy's top cgroup */ |
| 145 | #define dummytop (&rootnode.top_cgroup) |
| 146 | |
| 147 | /* This flag indicates whether tasks in the fork and exit paths should |
Li Zefan | a043e3b | 2008-02-23 15:24:09 -0800 | [diff] [blame] | 148 | * check for fork/exit handlers to call. This avoids us having to do |
| 149 | * extra work in the fork/exit path if none of the subsystems need to |
| 150 | * be called. |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 151 | */ |
Li Zefan | 8947f9d | 2008-07-25 01:46:56 -0700 | [diff] [blame] | 152 | static int need_forkexit_callback __read_mostly; |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 153 | |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 154 | /* convenient tests for these bits */ |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 155 | inline int cgroup_is_removed(const struct cgroup *cgrp) |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 156 | { |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 157 | return test_bit(CGRP_REMOVED, &cgrp->flags); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 158 | } |
| 159 | |
| 160 | /* bits in struct cgroupfs_root flags field */ |
| 161 | enum { |
| 162 | ROOT_NOPREFIX, /* mounted subsystems have no named prefix */ |
| 163 | }; |
| 164 | |
Adrian Bunk | e9685a0 | 2008-02-07 00:13:46 -0800 | [diff] [blame] | 165 | static int cgroup_is_releasable(const struct cgroup *cgrp) |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 166 | { |
| 167 | const int bits = |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 168 | (1 << CGRP_RELEASABLE) | |
| 169 | (1 << CGRP_NOTIFY_ON_RELEASE); |
| 170 | return (cgrp->flags & bits) == bits; |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 171 | } |
| 172 | |
Adrian Bunk | e9685a0 | 2008-02-07 00:13:46 -0800 | [diff] [blame] | 173 | static int notify_on_release(const struct cgroup *cgrp) |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 174 | { |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 175 | return test_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 176 | } |
| 177 | |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 178 | /* |
| 179 | * for_each_subsys() allows you to iterate on each subsystem attached to |
| 180 | * an active hierarchy |
| 181 | */ |
| 182 | #define for_each_subsys(_root, _ss) \ |
| 183 | list_for_each_entry(_ss, &_root->subsys_list, sibling) |
| 184 | |
Li Zefan | e5f6a86 | 2009-01-07 18:07:41 -0800 | [diff] [blame] | 185 | /* for_each_active_root() allows you to iterate across the active hierarchies */ |
| 186 | #define for_each_active_root(_root) \ |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 187 | list_for_each_entry(_root, &roots, root_list) |
| 188 | |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 189 | /* the list of cgroups eligible for automatic release. Protected by |
| 190 | * release_list_lock */ |
| 191 | static LIST_HEAD(release_list); |
| 192 | static DEFINE_SPINLOCK(release_list_lock); |
| 193 | static void cgroup_release_agent(struct work_struct *work); |
| 194 | static DECLARE_WORK(release_agent_work, cgroup_release_agent); |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 195 | static void check_for_release(struct cgroup *cgrp); |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 196 | |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 197 | /* Link structure for associating css_set objects with cgroups */ |
| 198 | struct cg_cgroup_link { |
| 199 | /* |
| 200 | * List running through cg_cgroup_links associated with a |
| 201 | * cgroup, anchored on cgroup->css_sets |
| 202 | */ |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 203 | struct list_head cgrp_link_list; |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 204 | /* |
| 205 | * List running through cg_cgroup_links pointing at a |
| 206 | * single css_set object, anchored on css_set->cg_links |
| 207 | */ |
| 208 | struct list_head cg_link_list; |
| 209 | struct css_set *cg; |
| 210 | }; |
| 211 | |
| 212 | /* The default css_set - used by init and its children prior to any |
| 213 | * hierarchies being mounted. It contains a pointer to the root state |
| 214 | * for each subsystem. Also used to anchor the list of css_sets. Not |
| 215 | * reference-counted, to improve performance when child cgroups |
| 216 | * haven't been created. |
| 217 | */ |
| 218 | |
| 219 | static struct css_set init_css_set; |
| 220 | static struct cg_cgroup_link init_css_set_link; |
| 221 | |
KAMEZAWA Hiroyuki | 38460b4 | 2009-04-02 16:57:25 -0700 | [diff] [blame] | 222 | static int cgroup_subsys_init_idr(struct cgroup_subsys *ss); |
| 223 | |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 224 | /* css_set_lock protects the list of css_set objects, and the |
| 225 | * chain of tasks off each css_set. Nests outside task->alloc_lock |
| 226 | * due to cgroup_iter_start() */ |
| 227 | static DEFINE_RWLOCK(css_set_lock); |
| 228 | static int css_set_count; |
| 229 | |
Li Zefan | 472b105 | 2008-04-29 01:00:11 -0700 | [diff] [blame] | 230 | /* hash table for cgroup groups. This improves the performance to |
| 231 | * find an existing css_set */ |
| 232 | #define CSS_SET_HASH_BITS 7 |
| 233 | #define CSS_SET_TABLE_SIZE (1 << CSS_SET_HASH_BITS) |
| 234 | static struct hlist_head css_set_table[CSS_SET_TABLE_SIZE]; |
| 235 | |
| 236 | static struct hlist_head *css_set_hash(struct cgroup_subsys_state *css[]) |
| 237 | { |
| 238 | int i; |
| 239 | int index; |
| 240 | unsigned long tmp = 0UL; |
| 241 | |
| 242 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) |
| 243 | tmp += (unsigned long)css[i]; |
| 244 | tmp = (tmp >> 16) ^ tmp; |
| 245 | |
| 246 | index = hash_long(tmp, CSS_SET_HASH_BITS); |
| 247 | |
| 248 | return &css_set_table[index]; |
| 249 | } |
| 250 | |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 251 | /* We don't maintain the lists running through each css_set to its |
| 252 | * task until after the first call to cgroup_iter_start(). This |
| 253 | * reduces the fork()/exit() overhead for people who have cgroups |
| 254 | * compiled into their kernel but not actually in use */ |
Li Zefan | 8947f9d | 2008-07-25 01:46:56 -0700 | [diff] [blame] | 255 | static int use_task_css_set_links __read_mostly; |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 256 | |
| 257 | /* When we create or destroy a css_set, the operation simply |
| 258 | * takes/releases a reference count on all the cgroups referenced |
| 259 | * by subsystems in this css_set. This can end up multiple-counting |
| 260 | * some cgroups, but that's OK - the ref-count is just a |
| 261 | * busy/not-busy indicator; ensuring that we only count each cgroup |
| 262 | * once would require taking a global lock to ensure that no |
Paul Menage | b4f48b6 | 2007-10-18 23:39:33 -0700 | [diff] [blame] | 263 | * subsystems moved between hierarchies while we were doing so. |
| 264 | * |
| 265 | * Possible TODO: decide at boot time based on the number of |
| 266 | * registered subsystems and the number of CPUs or NUMA nodes whether |
| 267 | * it's better for performance to ref-count every subsystem, or to |
| 268 | * take a global lock and only add one ref count to each hierarchy. |
| 269 | */ |
Paul Menage | b4f48b6 | 2007-10-18 23:39:33 -0700 | [diff] [blame] | 270 | |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 271 | /* |
| 272 | * unlink a css_set from the list and free it |
| 273 | */ |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 274 | static void unlink_css_set(struct css_set *cg) |
Paul Menage | b4f48b6 | 2007-10-18 23:39:33 -0700 | [diff] [blame] | 275 | { |
KOSAKI Motohiro | 71cbb94 | 2008-07-25 01:46:55 -0700 | [diff] [blame] | 276 | struct cg_cgroup_link *link; |
| 277 | struct cg_cgroup_link *saved_link; |
| 278 | |
Li Zefan | 472b105 | 2008-04-29 01:00:11 -0700 | [diff] [blame] | 279 | hlist_del(&cg->hlist); |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 280 | css_set_count--; |
KOSAKI Motohiro | 71cbb94 | 2008-07-25 01:46:55 -0700 | [diff] [blame] | 281 | |
| 282 | list_for_each_entry_safe(link, saved_link, &cg->cg_links, |
| 283 | cg_link_list) { |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 284 | list_del(&link->cg_link_list); |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 285 | list_del(&link->cgrp_link_list); |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 286 | kfree(link); |
| 287 | } |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 288 | } |
| 289 | |
Lai Jiangshan | 146aa1b | 2008-10-18 20:28:03 -0700 | [diff] [blame] | 290 | static void __put_css_set(struct css_set *cg, int taskexit) |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 291 | { |
| 292 | int i; |
Lai Jiangshan | 146aa1b | 2008-10-18 20:28:03 -0700 | [diff] [blame] | 293 | /* |
| 294 | * Ensure that the refcount doesn't hit zero while any readers |
| 295 | * can see it. Similar to atomic_dec_and_lock(), but for an |
| 296 | * rwlock |
| 297 | */ |
| 298 | if (atomic_add_unless(&cg->refcount, -1, 1)) |
| 299 | return; |
| 300 | write_lock(&css_set_lock); |
| 301 | if (!atomic_dec_and_test(&cg->refcount)) { |
| 302 | write_unlock(&css_set_lock); |
| 303 | return; |
| 304 | } |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 305 | unlink_css_set(cg); |
Lai Jiangshan | 146aa1b | 2008-10-18 20:28:03 -0700 | [diff] [blame] | 306 | write_unlock(&css_set_lock); |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 307 | |
| 308 | rcu_read_lock(); |
| 309 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { |
Paul Menage | a47295e | 2009-01-07 18:07:44 -0800 | [diff] [blame] | 310 | struct cgroup *cgrp = rcu_dereference(cg->subsys[i]->cgroup); |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 311 | if (atomic_dec_and_test(&cgrp->count) && |
| 312 | notify_on_release(cgrp)) { |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 313 | if (taskexit) |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 314 | set_bit(CGRP_RELEASABLE, &cgrp->flags); |
| 315 | check_for_release(cgrp); |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 316 | } |
| 317 | } |
| 318 | rcu_read_unlock(); |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 319 | kfree(cg); |
| 320 | } |
| 321 | |
| 322 | /* |
| 323 | * refcounted get/put for css_set objects |
| 324 | */ |
| 325 | static inline void get_css_set(struct css_set *cg) |
| 326 | { |
Lai Jiangshan | 146aa1b | 2008-10-18 20:28:03 -0700 | [diff] [blame] | 327 | atomic_inc(&cg->refcount); |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 328 | } |
| 329 | |
| 330 | static inline void put_css_set(struct css_set *cg) |
| 331 | { |
Lai Jiangshan | 146aa1b | 2008-10-18 20:28:03 -0700 | [diff] [blame] | 332 | __put_css_set(cg, 0); |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 333 | } |
| 334 | |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 335 | static inline void put_css_set_taskexit(struct css_set *cg) |
| 336 | { |
Lai Jiangshan | 146aa1b | 2008-10-18 20:28:03 -0700 | [diff] [blame] | 337 | __put_css_set(cg, 1); |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 338 | } |
| 339 | |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 340 | /* |
| 341 | * find_existing_css_set() is a helper for |
| 342 | * find_css_set(), and checks to see whether an existing |
Li Zefan | 472b105 | 2008-04-29 01:00:11 -0700 | [diff] [blame] | 343 | * css_set is suitable. |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 344 | * |
| 345 | * oldcg: the cgroup group that we're using before the cgroup |
| 346 | * transition |
| 347 | * |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 348 | * cgrp: the cgroup that we're moving into |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 349 | * |
| 350 | * template: location in which to build the desired set of subsystem |
| 351 | * state objects for the new cgroup group |
| 352 | */ |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 353 | static struct css_set *find_existing_css_set( |
| 354 | struct css_set *oldcg, |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 355 | struct cgroup *cgrp, |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 356 | struct cgroup_subsys_state *template[]) |
| 357 | { |
| 358 | int i; |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 359 | struct cgroupfs_root *root = cgrp->root; |
Li Zefan | 472b105 | 2008-04-29 01:00:11 -0700 | [diff] [blame] | 360 | struct hlist_head *hhead; |
| 361 | struct hlist_node *node; |
| 362 | struct css_set *cg; |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 363 | |
| 364 | /* Built the set of subsystem state objects that we want to |
| 365 | * see in the new css_set */ |
| 366 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { |
Li Zefan | 8d53d55 | 2008-02-23 15:24:11 -0800 | [diff] [blame] | 367 | if (root->subsys_bits & (1UL << i)) { |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 368 | /* Subsystem is in this hierarchy. So we want |
| 369 | * the subsystem state from the new |
| 370 | * cgroup */ |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 371 | template[i] = cgrp->subsys[i]; |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 372 | } else { |
| 373 | /* Subsystem is not in this hierarchy, so we |
| 374 | * don't want to change the subsystem state */ |
| 375 | template[i] = oldcg->subsys[i]; |
| 376 | } |
| 377 | } |
| 378 | |
Li Zefan | 472b105 | 2008-04-29 01:00:11 -0700 | [diff] [blame] | 379 | hhead = css_set_hash(template); |
| 380 | hlist_for_each_entry(cg, node, hhead, hlist) { |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 381 | if (!memcmp(template, cg->subsys, sizeof(cg->subsys))) { |
| 382 | /* All subsystems matched */ |
| 383 | return cg; |
| 384 | } |
Li Zefan | 472b105 | 2008-04-29 01:00:11 -0700 | [diff] [blame] | 385 | } |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 386 | |
| 387 | /* No existing cgroup group matched */ |
| 388 | return NULL; |
| 389 | } |
| 390 | |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 391 | static void free_cg_links(struct list_head *tmp) |
| 392 | { |
KOSAKI Motohiro | 71cbb94 | 2008-07-25 01:46:55 -0700 | [diff] [blame] | 393 | struct cg_cgroup_link *link; |
| 394 | struct cg_cgroup_link *saved_link; |
| 395 | |
| 396 | list_for_each_entry_safe(link, saved_link, tmp, cgrp_link_list) { |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 397 | list_del(&link->cgrp_link_list); |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 398 | kfree(link); |
| 399 | } |
| 400 | } |
| 401 | |
| 402 | /* |
Li Zefan | 3655343 | 2008-07-29 22:33:19 -0700 | [diff] [blame] | 403 | * allocate_cg_links() allocates "count" cg_cgroup_link structures |
| 404 | * and chains them on tmp through their cgrp_link_list fields. Returns 0 on |
| 405 | * success or a negative error |
| 406 | */ |
| 407 | static int allocate_cg_links(int count, struct list_head *tmp) |
| 408 | { |
| 409 | struct cg_cgroup_link *link; |
| 410 | int i; |
| 411 | INIT_LIST_HEAD(tmp); |
| 412 | for (i = 0; i < count; i++) { |
| 413 | link = kmalloc(sizeof(*link), GFP_KERNEL); |
| 414 | if (!link) { |
| 415 | free_cg_links(tmp); |
| 416 | return -ENOMEM; |
| 417 | } |
| 418 | list_add(&link->cgrp_link_list, tmp); |
| 419 | } |
| 420 | return 0; |
| 421 | } |
| 422 | |
Li Zefan | c12f65d | 2009-01-07 18:07:42 -0800 | [diff] [blame] | 423 | /** |
| 424 | * link_css_set - a helper function to link a css_set to a cgroup |
| 425 | * @tmp_cg_links: cg_cgroup_link objects allocated by allocate_cg_links() |
| 426 | * @cg: the css_set to be linked |
| 427 | * @cgrp: the destination cgroup |
| 428 | */ |
| 429 | static void link_css_set(struct list_head *tmp_cg_links, |
| 430 | struct css_set *cg, struct cgroup *cgrp) |
| 431 | { |
| 432 | struct cg_cgroup_link *link; |
| 433 | |
| 434 | BUG_ON(list_empty(tmp_cg_links)); |
| 435 | link = list_first_entry(tmp_cg_links, struct cg_cgroup_link, |
| 436 | cgrp_link_list); |
| 437 | link->cg = cg; |
| 438 | list_move(&link->cgrp_link_list, &cgrp->css_sets); |
| 439 | list_add(&link->cg_link_list, &cg->cg_links); |
| 440 | } |
| 441 | |
Li Zefan | 3655343 | 2008-07-29 22:33:19 -0700 | [diff] [blame] | 442 | /* |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 443 | * find_css_set() takes an existing cgroup group and a |
| 444 | * cgroup object, and returns a css_set object that's |
| 445 | * equivalent to the old group, but with the given cgroup |
| 446 | * substituted into the appropriate hierarchy. Must be called with |
| 447 | * cgroup_mutex held |
| 448 | */ |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 449 | static struct css_set *find_css_set( |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 450 | struct css_set *oldcg, struct cgroup *cgrp) |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 451 | { |
| 452 | struct css_set *res; |
| 453 | struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT]; |
| 454 | int i; |
| 455 | |
| 456 | struct list_head tmp_cg_links; |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 457 | |
Li Zefan | 472b105 | 2008-04-29 01:00:11 -0700 | [diff] [blame] | 458 | struct hlist_head *hhead; |
| 459 | |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 460 | /* First see if we already have a cgroup group that matches |
| 461 | * the desired set */ |
Li Zefan | 7e9abd8 | 2008-07-25 01:46:54 -0700 | [diff] [blame] | 462 | read_lock(&css_set_lock); |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 463 | res = find_existing_css_set(oldcg, cgrp, template); |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 464 | if (res) |
| 465 | get_css_set(res); |
Li Zefan | 7e9abd8 | 2008-07-25 01:46:54 -0700 | [diff] [blame] | 466 | read_unlock(&css_set_lock); |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 467 | |
| 468 | if (res) |
| 469 | return res; |
| 470 | |
| 471 | res = kmalloc(sizeof(*res), GFP_KERNEL); |
| 472 | if (!res) |
| 473 | return NULL; |
| 474 | |
| 475 | /* Allocate all the cg_cgroup_link objects that we'll need */ |
| 476 | if (allocate_cg_links(root_count, &tmp_cg_links) < 0) { |
| 477 | kfree(res); |
| 478 | return NULL; |
| 479 | } |
| 480 | |
Lai Jiangshan | 146aa1b | 2008-10-18 20:28:03 -0700 | [diff] [blame] | 481 | atomic_set(&res->refcount, 1); |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 482 | INIT_LIST_HEAD(&res->cg_links); |
| 483 | INIT_LIST_HEAD(&res->tasks); |
Li Zefan | 472b105 | 2008-04-29 01:00:11 -0700 | [diff] [blame] | 484 | INIT_HLIST_NODE(&res->hlist); |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 485 | |
| 486 | /* Copy the set of subsystem state objects generated in |
| 487 | * find_existing_css_set() */ |
| 488 | memcpy(res->subsys, template, sizeof(res->subsys)); |
| 489 | |
| 490 | write_lock(&css_set_lock); |
| 491 | /* Add reference counts and links from the new css_set. */ |
| 492 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 493 | struct cgroup *cgrp = res->subsys[i]->cgroup; |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 494 | struct cgroup_subsys *ss = subsys[i]; |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 495 | atomic_inc(&cgrp->count); |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 496 | /* |
| 497 | * We want to add a link once per cgroup, so we |
| 498 | * only do it for the first subsystem in each |
| 499 | * hierarchy |
| 500 | */ |
Li Zefan | c12f65d | 2009-01-07 18:07:42 -0800 | [diff] [blame] | 501 | if (ss->root->subsys_list.next == &ss->sibling) |
| 502 | link_css_set(&tmp_cg_links, res, cgrp); |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 503 | } |
Li Zefan | c12f65d | 2009-01-07 18:07:42 -0800 | [diff] [blame] | 504 | if (list_empty(&rootnode.subsys_list)) |
| 505 | link_css_set(&tmp_cg_links, res, dummytop); |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 506 | |
| 507 | BUG_ON(!list_empty(&tmp_cg_links)); |
| 508 | |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 509 | css_set_count++; |
Li Zefan | 472b105 | 2008-04-29 01:00:11 -0700 | [diff] [blame] | 510 | |
| 511 | /* Add this cgroup group to the hash table */ |
| 512 | hhead = css_set_hash(res->subsys); |
| 513 | hlist_add_head(&res->hlist, hhead); |
| 514 | |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 515 | write_unlock(&css_set_lock); |
| 516 | |
| 517 | return res; |
Paul Menage | b4f48b6 | 2007-10-18 23:39:33 -0700 | [diff] [blame] | 518 | } |
| 519 | |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 520 | /* |
| 521 | * There is one global cgroup mutex. We also require taking |
| 522 | * task_lock() when dereferencing a task's cgroup subsys pointers. |
| 523 | * See "The task_lock() exception", at the end of this comment. |
| 524 | * |
| 525 | * A task must hold cgroup_mutex to modify cgroups. |
| 526 | * |
| 527 | * Any task can increment and decrement the count field without lock. |
| 528 | * So in general, code holding cgroup_mutex can't rely on the count |
| 529 | * field not changing. However, if the count goes to zero, then only |
Cliff Wickman | 956db3c | 2008-02-07 00:14:43 -0800 | [diff] [blame] | 530 | * cgroup_attach_task() can increment it again. Because a count of zero |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 531 | * means that no tasks are currently attached, therefore there is no |
| 532 | * way a task attached to that cgroup can fork (the other way to |
| 533 | * increment the count). So code holding cgroup_mutex can safely |
| 534 | * assume that if the count is zero, it will stay zero. Similarly, if |
| 535 | * a task holds cgroup_mutex on a cgroup with zero count, it |
| 536 | * knows that the cgroup won't be removed, as cgroup_rmdir() |
| 537 | * needs that mutex. |
| 538 | * |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 539 | * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't |
| 540 | * (usually) take cgroup_mutex. These are the two most performance |
| 541 | * critical pieces of code here. The exception occurs on cgroup_exit(), |
| 542 | * when a task in a notify_on_release cgroup exits. Then cgroup_mutex |
| 543 | * is taken, and if the cgroup count is zero, a usermode call made |
Li Zefan | a043e3b | 2008-02-23 15:24:09 -0800 | [diff] [blame] | 544 | * to the release agent with the name of the cgroup (path relative to |
| 545 | * the root of cgroup file system) as the argument. |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 546 | * |
| 547 | * A cgroup can only be deleted if both its 'count' of using tasks |
| 548 | * is zero, and its list of 'children' cgroups is empty. Since all |
| 549 | * tasks in the system use _some_ cgroup, and since there is always at |
| 550 | * least one task in the system (init, pid == 1), therefore, top_cgroup |
| 551 | * always has either children cgroups and/or using tasks. So we don't |
| 552 | * need a special hack to ensure that top_cgroup cannot be deleted. |
| 553 | * |
| 554 | * The task_lock() exception |
| 555 | * |
| 556 | * The need for this exception arises from the action of |
Cliff Wickman | 956db3c | 2008-02-07 00:14:43 -0800 | [diff] [blame] | 557 | * cgroup_attach_task(), which overwrites one tasks cgroup pointer with |
Li Zefan | a043e3b | 2008-02-23 15:24:09 -0800 | [diff] [blame] | 558 | * another. It does so using cgroup_mutex, however there are |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 559 | * several performance critical places that need to reference |
| 560 | * task->cgroup without the expense of grabbing a system global |
| 561 | * mutex. Therefore except as noted below, when dereferencing or, as |
Cliff Wickman | 956db3c | 2008-02-07 00:14:43 -0800 | [diff] [blame] | 562 | * in cgroup_attach_task(), modifying a task'ss cgroup pointer we use |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 563 | * task_lock(), which acts on a spinlock (task->alloc_lock) already in |
| 564 | * the task_struct routinely used for such matters. |
| 565 | * |
| 566 | * P.S. One more locking exception. RCU is used to guard the |
Cliff Wickman | 956db3c | 2008-02-07 00:14:43 -0800 | [diff] [blame] | 567 | * update of a tasks cgroup pointer by cgroup_attach_task() |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 568 | */ |
| 569 | |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 570 | /** |
| 571 | * cgroup_lock - lock out any changes to cgroup structures |
| 572 | * |
| 573 | */ |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 574 | void cgroup_lock(void) |
| 575 | { |
| 576 | mutex_lock(&cgroup_mutex); |
| 577 | } |
| 578 | |
| 579 | /** |
| 580 | * cgroup_unlock - release lock on cgroup changes |
| 581 | * |
| 582 | * Undo the lock taken in a previous cgroup_lock() call. |
| 583 | */ |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 584 | void cgroup_unlock(void) |
| 585 | { |
| 586 | mutex_unlock(&cgroup_mutex); |
| 587 | } |
| 588 | |
| 589 | /* |
| 590 | * A couple of forward declarations required, due to cyclic reference loop: |
| 591 | * cgroup_mkdir -> cgroup_create -> cgroup_populate_dir -> |
| 592 | * cgroup_add_file -> cgroup_create_file -> cgroup_dir_inode_operations |
| 593 | * -> cgroup_mkdir. |
| 594 | */ |
| 595 | |
| 596 | static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, int mode); |
| 597 | static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry); |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 598 | static int cgroup_populate_dir(struct cgroup *cgrp); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 599 | static struct inode_operations cgroup_dir_inode_operations; |
Paul Menage | a424316 | 2007-10-18 23:39:35 -0700 | [diff] [blame] | 600 | static struct file_operations proc_cgroupstats_operations; |
| 601 | |
| 602 | static struct backing_dev_info cgroup_backing_dev_info = { |
Jens Axboe | d993831 | 2009-06-12 14:45:52 +0200 | [diff] [blame] | 603 | .name = "cgroup", |
Miklos Szeredi | e4ad08f | 2008-04-30 00:54:37 -0700 | [diff] [blame] | 604 | .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK, |
Paul Menage | a424316 | 2007-10-18 23:39:35 -0700 | [diff] [blame] | 605 | }; |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 606 | |
KAMEZAWA Hiroyuki | 38460b4 | 2009-04-02 16:57:25 -0700 | [diff] [blame] | 607 | static int alloc_css_id(struct cgroup_subsys *ss, |
| 608 | struct cgroup *parent, struct cgroup *child); |
| 609 | |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 610 | static struct inode *cgroup_new_inode(mode_t mode, struct super_block *sb) |
| 611 | { |
| 612 | struct inode *inode = new_inode(sb); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 613 | |
| 614 | if (inode) { |
| 615 | inode->i_mode = mode; |
David Howells | 76aac0e | 2008-11-14 10:39:12 +1100 | [diff] [blame] | 616 | inode->i_uid = current_fsuid(); |
| 617 | inode->i_gid = current_fsgid(); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 618 | inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME; |
| 619 | inode->i_mapping->backing_dev_info = &cgroup_backing_dev_info; |
| 620 | } |
| 621 | return inode; |
| 622 | } |
| 623 | |
KAMEZAWA Hiroyuki | 4fca88c | 2008-02-07 00:14:27 -0800 | [diff] [blame] | 624 | /* |
| 625 | * Call subsys's pre_destroy handler. |
| 626 | * This is called before css refcnt check. |
| 627 | */ |
KAMEZAWA Hiroyuki | ec64f51 | 2009-04-02 16:57:26 -0700 | [diff] [blame] | 628 | static int cgroup_call_pre_destroy(struct cgroup *cgrp) |
KAMEZAWA Hiroyuki | 4fca88c | 2008-02-07 00:14:27 -0800 | [diff] [blame] | 629 | { |
| 630 | struct cgroup_subsys *ss; |
KAMEZAWA Hiroyuki | ec64f51 | 2009-04-02 16:57:26 -0700 | [diff] [blame] | 631 | int ret = 0; |
| 632 | |
KAMEZAWA Hiroyuki | 4fca88c | 2008-02-07 00:14:27 -0800 | [diff] [blame] | 633 | for_each_subsys(cgrp->root, ss) |
KAMEZAWA Hiroyuki | ec64f51 | 2009-04-02 16:57:26 -0700 | [diff] [blame] | 634 | if (ss->pre_destroy) { |
| 635 | ret = ss->pre_destroy(ss, cgrp); |
| 636 | if (ret) |
| 637 | break; |
| 638 | } |
| 639 | return ret; |
KAMEZAWA Hiroyuki | 4fca88c | 2008-02-07 00:14:27 -0800 | [diff] [blame] | 640 | } |
| 641 | |
Paul Menage | a47295e | 2009-01-07 18:07:44 -0800 | [diff] [blame] | 642 | static void free_cgroup_rcu(struct rcu_head *obj) |
| 643 | { |
| 644 | struct cgroup *cgrp = container_of(obj, struct cgroup, rcu_head); |
| 645 | |
| 646 | kfree(cgrp); |
| 647 | } |
| 648 | |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 649 | static void cgroup_diput(struct dentry *dentry, struct inode *inode) |
| 650 | { |
| 651 | /* is dentry a directory ? if so, kfree() associated cgroup */ |
| 652 | if (S_ISDIR(inode->i_mode)) { |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 653 | struct cgroup *cgrp = dentry->d_fsdata; |
Paul Menage | 8dc4f3e | 2008-02-07 00:13:45 -0800 | [diff] [blame] | 654 | struct cgroup_subsys *ss; |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 655 | BUG_ON(!(cgroup_is_removed(cgrp))); |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 656 | /* It's possible for external users to be holding css |
| 657 | * reference counts on a cgroup; css_put() needs to |
| 658 | * be able to access the cgroup after decrementing |
| 659 | * the reference count in order to know if it needs to |
| 660 | * queue the cgroup to be handled by the release |
| 661 | * agent */ |
| 662 | synchronize_rcu(); |
Paul Menage | 8dc4f3e | 2008-02-07 00:13:45 -0800 | [diff] [blame] | 663 | |
| 664 | mutex_lock(&cgroup_mutex); |
| 665 | /* |
| 666 | * Release the subsystem state objects. |
| 667 | */ |
Li Zefan | 75139b8 | 2009-01-07 18:07:33 -0800 | [diff] [blame] | 668 | for_each_subsys(cgrp->root, ss) |
| 669 | ss->destroy(ss, cgrp); |
Paul Menage | 8dc4f3e | 2008-02-07 00:13:45 -0800 | [diff] [blame] | 670 | |
| 671 | cgrp->root->number_of_cgroups--; |
| 672 | mutex_unlock(&cgroup_mutex); |
| 673 | |
Paul Menage | a47295e | 2009-01-07 18:07:44 -0800 | [diff] [blame] | 674 | /* |
| 675 | * Drop the active superblock reference that we took when we |
| 676 | * created the cgroup |
| 677 | */ |
Paul Menage | 8dc4f3e | 2008-02-07 00:13:45 -0800 | [diff] [blame] | 678 | deactivate_super(cgrp->root->sb); |
| 679 | |
Paul Menage | a47295e | 2009-01-07 18:07:44 -0800 | [diff] [blame] | 680 | call_rcu(&cgrp->rcu_head, free_cgroup_rcu); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 681 | } |
| 682 | iput(inode); |
| 683 | } |
| 684 | |
| 685 | static void remove_dir(struct dentry *d) |
| 686 | { |
| 687 | struct dentry *parent = dget(d->d_parent); |
| 688 | |
| 689 | d_delete(d); |
| 690 | simple_rmdir(parent->d_inode, d); |
| 691 | dput(parent); |
| 692 | } |
| 693 | |
| 694 | static void cgroup_clear_directory(struct dentry *dentry) |
| 695 | { |
| 696 | struct list_head *node; |
| 697 | |
| 698 | BUG_ON(!mutex_is_locked(&dentry->d_inode->i_mutex)); |
| 699 | spin_lock(&dcache_lock); |
| 700 | node = dentry->d_subdirs.next; |
| 701 | while (node != &dentry->d_subdirs) { |
| 702 | struct dentry *d = list_entry(node, struct dentry, d_u.d_child); |
| 703 | list_del_init(node); |
| 704 | if (d->d_inode) { |
| 705 | /* This should never be called on a cgroup |
| 706 | * directory with child cgroups */ |
| 707 | BUG_ON(d->d_inode->i_mode & S_IFDIR); |
| 708 | d = dget_locked(d); |
| 709 | spin_unlock(&dcache_lock); |
| 710 | d_delete(d); |
| 711 | simple_unlink(dentry->d_inode, d); |
| 712 | dput(d); |
| 713 | spin_lock(&dcache_lock); |
| 714 | } |
| 715 | node = dentry->d_subdirs.next; |
| 716 | } |
| 717 | spin_unlock(&dcache_lock); |
| 718 | } |
| 719 | |
| 720 | /* |
| 721 | * NOTE : the dentry must have been dget()'ed |
| 722 | */ |
| 723 | static void cgroup_d_remove_dir(struct dentry *dentry) |
| 724 | { |
| 725 | cgroup_clear_directory(dentry); |
| 726 | |
| 727 | spin_lock(&dcache_lock); |
| 728 | list_del_init(&dentry->d_u.d_child); |
| 729 | spin_unlock(&dcache_lock); |
| 730 | remove_dir(dentry); |
| 731 | } |
| 732 | |
KAMEZAWA Hiroyuki | ec64f51 | 2009-04-02 16:57:26 -0700 | [diff] [blame] | 733 | /* |
| 734 | * A queue for waiters to do rmdir() cgroup. A tasks will sleep when |
| 735 | * cgroup->count == 0 && list_empty(&cgroup->children) && subsys has some |
| 736 | * reference to css->refcnt. In general, this refcnt is expected to goes down |
| 737 | * to zero, soon. |
| 738 | * |
KAMEZAWA Hiroyuki | 8870326 | 2009-07-29 15:04:06 -0700 | [diff] [blame] | 739 | * CGRP_WAIT_ON_RMDIR flag is set under cgroup's inode->i_mutex; |
KAMEZAWA Hiroyuki | ec64f51 | 2009-04-02 16:57:26 -0700 | [diff] [blame] | 740 | */ |
| 741 | DECLARE_WAIT_QUEUE_HEAD(cgroup_rmdir_waitq); |
| 742 | |
KAMEZAWA Hiroyuki | 8870326 | 2009-07-29 15:04:06 -0700 | [diff] [blame] | 743 | static void cgroup_wakeup_rmdir_waiter(struct cgroup *cgrp) |
KAMEZAWA Hiroyuki | ec64f51 | 2009-04-02 16:57:26 -0700 | [diff] [blame] | 744 | { |
KAMEZAWA Hiroyuki | 8870326 | 2009-07-29 15:04:06 -0700 | [diff] [blame] | 745 | if (unlikely(test_and_clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags))) |
KAMEZAWA Hiroyuki | ec64f51 | 2009-04-02 16:57:26 -0700 | [diff] [blame] | 746 | wake_up_all(&cgroup_rmdir_waitq); |
| 747 | } |
| 748 | |
KAMEZAWA Hiroyuki | 8870326 | 2009-07-29 15:04:06 -0700 | [diff] [blame] | 749 | void cgroup_exclude_rmdir(struct cgroup_subsys_state *css) |
| 750 | { |
| 751 | css_get(css); |
| 752 | } |
| 753 | |
| 754 | void cgroup_release_and_wakeup_rmdir(struct cgroup_subsys_state *css) |
| 755 | { |
| 756 | cgroup_wakeup_rmdir_waiter(css->cgroup); |
| 757 | css_put(css); |
| 758 | } |
| 759 | |
| 760 | |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 761 | static int rebind_subsystems(struct cgroupfs_root *root, |
| 762 | unsigned long final_bits) |
| 763 | { |
| 764 | unsigned long added_bits, removed_bits; |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 765 | struct cgroup *cgrp = &root->top_cgroup; |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 766 | int i; |
| 767 | |
| 768 | removed_bits = root->actual_subsys_bits & ~final_bits; |
| 769 | added_bits = final_bits & ~root->actual_subsys_bits; |
| 770 | /* Check that any added subsystems are currently free */ |
| 771 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { |
Li Zefan | 8d53d55 | 2008-02-23 15:24:11 -0800 | [diff] [blame] | 772 | unsigned long bit = 1UL << i; |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 773 | struct cgroup_subsys *ss = subsys[i]; |
| 774 | if (!(bit & added_bits)) |
| 775 | continue; |
| 776 | if (ss->root != &rootnode) { |
| 777 | /* Subsystem isn't free */ |
| 778 | return -EBUSY; |
| 779 | } |
| 780 | } |
| 781 | |
| 782 | /* Currently we don't handle adding/removing subsystems when |
| 783 | * any child cgroups exist. This is theoretically supportable |
| 784 | * but involves complex error handling, so it's being left until |
| 785 | * later */ |
Paul Menage | 307257c | 2008-12-15 13:54:22 -0800 | [diff] [blame] | 786 | if (root->number_of_cgroups > 1) |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 787 | return -EBUSY; |
| 788 | |
| 789 | /* Process each subsystem */ |
| 790 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { |
| 791 | struct cgroup_subsys *ss = subsys[i]; |
| 792 | unsigned long bit = 1UL << i; |
| 793 | if (bit & added_bits) { |
| 794 | /* We're binding this subsystem to this hierarchy */ |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 795 | BUG_ON(cgrp->subsys[i]); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 796 | BUG_ON(!dummytop->subsys[i]); |
| 797 | BUG_ON(dummytop->subsys[i]->cgroup != dummytop); |
Paul Menage | 999cd8a | 2009-01-07 18:08:36 -0800 | [diff] [blame] | 798 | mutex_lock(&ss->hierarchy_mutex); |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 799 | cgrp->subsys[i] = dummytop->subsys[i]; |
| 800 | cgrp->subsys[i]->cgroup = cgrp; |
Li Zefan | 33a68ac | 2009-01-07 18:07:42 -0800 | [diff] [blame] | 801 | list_move(&ss->sibling, &root->subsys_list); |
Lai Jiangshan | b2aa30f | 2009-01-07 18:07:37 -0800 | [diff] [blame] | 802 | ss->root = root; |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 803 | if (ss->bind) |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 804 | ss->bind(ss, cgrp); |
Paul Menage | 999cd8a | 2009-01-07 18:08:36 -0800 | [diff] [blame] | 805 | mutex_unlock(&ss->hierarchy_mutex); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 806 | } else if (bit & removed_bits) { |
| 807 | /* We're removing this subsystem */ |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 808 | BUG_ON(cgrp->subsys[i] != dummytop->subsys[i]); |
| 809 | BUG_ON(cgrp->subsys[i]->cgroup != cgrp); |
Paul Menage | 999cd8a | 2009-01-07 18:08:36 -0800 | [diff] [blame] | 810 | mutex_lock(&ss->hierarchy_mutex); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 811 | if (ss->bind) |
| 812 | ss->bind(ss, dummytop); |
| 813 | dummytop->subsys[i]->cgroup = dummytop; |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 814 | cgrp->subsys[i] = NULL; |
Lai Jiangshan | b2aa30f | 2009-01-07 18:07:37 -0800 | [diff] [blame] | 815 | subsys[i]->root = &rootnode; |
Li Zefan | 33a68ac | 2009-01-07 18:07:42 -0800 | [diff] [blame] | 816 | list_move(&ss->sibling, &rootnode.subsys_list); |
Paul Menage | 999cd8a | 2009-01-07 18:08:36 -0800 | [diff] [blame] | 817 | mutex_unlock(&ss->hierarchy_mutex); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 818 | } else if (bit & final_bits) { |
| 819 | /* Subsystem state should already exist */ |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 820 | BUG_ON(!cgrp->subsys[i]); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 821 | } else { |
| 822 | /* Subsystem state shouldn't exist */ |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 823 | BUG_ON(cgrp->subsys[i]); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 824 | } |
| 825 | } |
| 826 | root->subsys_bits = root->actual_subsys_bits = final_bits; |
| 827 | synchronize_rcu(); |
| 828 | |
| 829 | return 0; |
| 830 | } |
| 831 | |
| 832 | static int cgroup_show_options(struct seq_file *seq, struct vfsmount *vfs) |
| 833 | { |
| 834 | struct cgroupfs_root *root = vfs->mnt_sb->s_fs_info; |
| 835 | struct cgroup_subsys *ss; |
| 836 | |
| 837 | mutex_lock(&cgroup_mutex); |
| 838 | for_each_subsys(root, ss) |
| 839 | seq_printf(seq, ",%s", ss->name); |
| 840 | if (test_bit(ROOT_NOPREFIX, &root->flags)) |
| 841 | seq_puts(seq, ",noprefix"); |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 842 | if (strlen(root->release_agent_path)) |
| 843 | seq_printf(seq, ",release_agent=%s", root->release_agent_path); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 844 | mutex_unlock(&cgroup_mutex); |
| 845 | return 0; |
| 846 | } |
| 847 | |
| 848 | struct cgroup_sb_opts { |
| 849 | unsigned long subsys_bits; |
| 850 | unsigned long flags; |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 851 | char *release_agent; |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 852 | }; |
| 853 | |
| 854 | /* Convert a hierarchy specifier into a bitmask of subsystems and |
| 855 | * flags. */ |
| 856 | static int parse_cgroupfs_options(char *data, |
| 857 | struct cgroup_sb_opts *opts) |
| 858 | { |
| 859 | char *token, *o = data ?: "all"; |
Li Zefan | f9ab5b5 | 2009-06-17 16:26:33 -0700 | [diff] [blame] | 860 | unsigned long mask = (unsigned long)-1; |
| 861 | |
| 862 | #ifdef CONFIG_CPUSETS |
| 863 | mask = ~(1UL << cpuset_subsys_id); |
| 864 | #endif |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 865 | |
| 866 | opts->subsys_bits = 0; |
| 867 | opts->flags = 0; |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 868 | opts->release_agent = NULL; |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 869 | |
| 870 | while ((token = strsep(&o, ",")) != NULL) { |
| 871 | if (!*token) |
| 872 | return -EINVAL; |
| 873 | if (!strcmp(token, "all")) { |
Paul Menage | 8bab8dd | 2008-04-04 14:29:57 -0700 | [diff] [blame] | 874 | /* Add all non-disabled subsystems */ |
| 875 | int i; |
| 876 | opts->subsys_bits = 0; |
| 877 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { |
| 878 | struct cgroup_subsys *ss = subsys[i]; |
| 879 | if (!ss->disabled) |
| 880 | opts->subsys_bits |= 1ul << i; |
| 881 | } |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 882 | } else if (!strcmp(token, "noprefix")) { |
| 883 | set_bit(ROOT_NOPREFIX, &opts->flags); |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 884 | } else if (!strncmp(token, "release_agent=", 14)) { |
| 885 | /* Specifying two release agents is forbidden */ |
| 886 | if (opts->release_agent) |
| 887 | return -EINVAL; |
| 888 | opts->release_agent = kzalloc(PATH_MAX, GFP_KERNEL); |
| 889 | if (!opts->release_agent) |
| 890 | return -ENOMEM; |
| 891 | strncpy(opts->release_agent, token + 14, PATH_MAX - 1); |
| 892 | opts->release_agent[PATH_MAX - 1] = 0; |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 893 | } else { |
| 894 | struct cgroup_subsys *ss; |
| 895 | int i; |
| 896 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { |
| 897 | ss = subsys[i]; |
| 898 | if (!strcmp(token, ss->name)) { |
Paul Menage | 8bab8dd | 2008-04-04 14:29:57 -0700 | [diff] [blame] | 899 | if (!ss->disabled) |
| 900 | set_bit(i, &opts->subsys_bits); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 901 | break; |
| 902 | } |
| 903 | } |
| 904 | if (i == CGROUP_SUBSYS_COUNT) |
| 905 | return -ENOENT; |
| 906 | } |
| 907 | } |
| 908 | |
Li Zefan | f9ab5b5 | 2009-06-17 16:26:33 -0700 | [diff] [blame] | 909 | /* |
| 910 | * Option noprefix was introduced just for backward compatibility |
| 911 | * with the old cpuset, so we allow noprefix only if mounting just |
| 912 | * the cpuset subsystem. |
| 913 | */ |
| 914 | if (test_bit(ROOT_NOPREFIX, &opts->flags) && |
| 915 | (opts->subsys_bits & mask)) |
| 916 | return -EINVAL; |
| 917 | |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 918 | /* We can't have an empty hierarchy */ |
| 919 | if (!opts->subsys_bits) |
| 920 | return -EINVAL; |
| 921 | |
| 922 | return 0; |
| 923 | } |
| 924 | |
| 925 | static int cgroup_remount(struct super_block *sb, int *flags, char *data) |
| 926 | { |
| 927 | int ret = 0; |
| 928 | struct cgroupfs_root *root = sb->s_fs_info; |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 929 | struct cgroup *cgrp = &root->top_cgroup; |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 930 | struct cgroup_sb_opts opts; |
| 931 | |
Alessio Igor Bogani | 337eb00 | 2009-05-12 15:10:54 +0200 | [diff] [blame] | 932 | lock_kernel(); |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 933 | mutex_lock(&cgrp->dentry->d_inode->i_mutex); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 934 | mutex_lock(&cgroup_mutex); |
| 935 | |
| 936 | /* See what subsystems are wanted */ |
| 937 | ret = parse_cgroupfs_options(data, &opts); |
| 938 | if (ret) |
| 939 | goto out_unlock; |
| 940 | |
| 941 | /* Don't allow flags to change at remount */ |
| 942 | if (opts.flags != root->flags) { |
| 943 | ret = -EINVAL; |
| 944 | goto out_unlock; |
| 945 | } |
| 946 | |
| 947 | ret = rebind_subsystems(root, opts.subsys_bits); |
Li Zefan | 0670e08 | 2009-04-02 16:57:30 -0700 | [diff] [blame] | 948 | if (ret) |
| 949 | goto out_unlock; |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 950 | |
| 951 | /* (re)populate subsystem files */ |
Li Zefan | 0670e08 | 2009-04-02 16:57:30 -0700 | [diff] [blame] | 952 | cgroup_populate_dir(cgrp); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 953 | |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 954 | if (opts.release_agent) |
| 955 | strcpy(root->release_agent_path, opts.release_agent); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 956 | out_unlock: |
Jesper Juhl | 66bdc9c | 2009-04-02 16:57:27 -0700 | [diff] [blame] | 957 | kfree(opts.release_agent); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 958 | mutex_unlock(&cgroup_mutex); |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 959 | mutex_unlock(&cgrp->dentry->d_inode->i_mutex); |
Alessio Igor Bogani | 337eb00 | 2009-05-12 15:10:54 +0200 | [diff] [blame] | 960 | unlock_kernel(); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 961 | return ret; |
| 962 | } |
| 963 | |
| 964 | static struct super_operations cgroup_ops = { |
| 965 | .statfs = simple_statfs, |
| 966 | .drop_inode = generic_delete_inode, |
| 967 | .show_options = cgroup_show_options, |
| 968 | .remount_fs = cgroup_remount, |
| 969 | }; |
| 970 | |
Paul Menage | cc31edc | 2008-10-18 20:28:04 -0700 | [diff] [blame] | 971 | static void init_cgroup_housekeeping(struct cgroup *cgrp) |
| 972 | { |
| 973 | INIT_LIST_HEAD(&cgrp->sibling); |
| 974 | INIT_LIST_HEAD(&cgrp->children); |
| 975 | INIT_LIST_HEAD(&cgrp->css_sets); |
| 976 | INIT_LIST_HEAD(&cgrp->release_list); |
Li Zefan | 096b7fe | 2009-07-29 15:04:04 -0700 | [diff] [blame] | 977 | INIT_LIST_HEAD(&cgrp->pids_list); |
Paul Menage | cc31edc | 2008-10-18 20:28:04 -0700 | [diff] [blame] | 978 | init_rwsem(&cgrp->pids_mutex); |
| 979 | } |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 980 | static void init_cgroup_root(struct cgroupfs_root *root) |
| 981 | { |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 982 | struct cgroup *cgrp = &root->top_cgroup; |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 983 | INIT_LIST_HEAD(&root->subsys_list); |
| 984 | INIT_LIST_HEAD(&root->root_list); |
| 985 | root->number_of_cgroups = 1; |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 986 | cgrp->root = root; |
| 987 | cgrp->top_cgroup = cgrp; |
Paul Menage | cc31edc | 2008-10-18 20:28:04 -0700 | [diff] [blame] | 988 | init_cgroup_housekeeping(cgrp); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 989 | } |
| 990 | |
| 991 | static int cgroup_test_super(struct super_block *sb, void *data) |
| 992 | { |
| 993 | struct cgroupfs_root *new = data; |
| 994 | struct cgroupfs_root *root = sb->s_fs_info; |
| 995 | |
| 996 | /* First check subsystems */ |
| 997 | if (new->subsys_bits != root->subsys_bits) |
| 998 | return 0; |
| 999 | |
| 1000 | /* Next check flags */ |
| 1001 | if (new->flags != root->flags) |
| 1002 | return 0; |
| 1003 | |
| 1004 | return 1; |
| 1005 | } |
| 1006 | |
| 1007 | static int cgroup_set_super(struct super_block *sb, void *data) |
| 1008 | { |
| 1009 | int ret; |
| 1010 | struct cgroupfs_root *root = data; |
| 1011 | |
| 1012 | ret = set_anon_super(sb, NULL); |
| 1013 | if (ret) |
| 1014 | return ret; |
| 1015 | |
| 1016 | sb->s_fs_info = root; |
| 1017 | root->sb = sb; |
| 1018 | |
| 1019 | sb->s_blocksize = PAGE_CACHE_SIZE; |
| 1020 | sb->s_blocksize_bits = PAGE_CACHE_SHIFT; |
| 1021 | sb->s_magic = CGROUP_SUPER_MAGIC; |
| 1022 | sb->s_op = &cgroup_ops; |
| 1023 | |
| 1024 | return 0; |
| 1025 | } |
| 1026 | |
| 1027 | static int cgroup_get_rootdir(struct super_block *sb) |
| 1028 | { |
| 1029 | struct inode *inode = |
| 1030 | cgroup_new_inode(S_IFDIR | S_IRUGO | S_IXUGO | S_IWUSR, sb); |
| 1031 | struct dentry *dentry; |
| 1032 | |
| 1033 | if (!inode) |
| 1034 | return -ENOMEM; |
| 1035 | |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1036 | inode->i_fop = &simple_dir_operations; |
| 1037 | inode->i_op = &cgroup_dir_inode_operations; |
| 1038 | /* directories start off with i_nlink == 2 (for "." entry) */ |
| 1039 | inc_nlink(inode); |
| 1040 | dentry = d_alloc_root(inode); |
| 1041 | if (!dentry) { |
| 1042 | iput(inode); |
| 1043 | return -ENOMEM; |
| 1044 | } |
| 1045 | sb->s_root = dentry; |
| 1046 | return 0; |
| 1047 | } |
| 1048 | |
| 1049 | static int cgroup_get_sb(struct file_system_type *fs_type, |
| 1050 | int flags, const char *unused_dev_name, |
| 1051 | void *data, struct vfsmount *mnt) |
| 1052 | { |
| 1053 | struct cgroup_sb_opts opts; |
| 1054 | int ret = 0; |
| 1055 | struct super_block *sb; |
| 1056 | struct cgroupfs_root *root; |
Li Zefan | 28fd5df | 2008-04-29 01:00:13 -0700 | [diff] [blame] | 1057 | struct list_head tmp_cg_links; |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1058 | |
| 1059 | /* First find the desired set of subsystems */ |
| 1060 | ret = parse_cgroupfs_options(data, &opts); |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 1061 | if (ret) { |
Jesper Juhl | 66bdc9c | 2009-04-02 16:57:27 -0700 | [diff] [blame] | 1062 | kfree(opts.release_agent); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1063 | return ret; |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 1064 | } |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1065 | |
| 1066 | root = kzalloc(sizeof(*root), GFP_KERNEL); |
Li Zefan | f777073 | 2008-02-23 15:24:10 -0800 | [diff] [blame] | 1067 | if (!root) { |
Jesper Juhl | 66bdc9c | 2009-04-02 16:57:27 -0700 | [diff] [blame] | 1068 | kfree(opts.release_agent); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1069 | return -ENOMEM; |
Li Zefan | f777073 | 2008-02-23 15:24:10 -0800 | [diff] [blame] | 1070 | } |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1071 | |
| 1072 | init_cgroup_root(root); |
| 1073 | root->subsys_bits = opts.subsys_bits; |
| 1074 | root->flags = opts.flags; |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 1075 | if (opts.release_agent) { |
| 1076 | strcpy(root->release_agent_path, opts.release_agent); |
| 1077 | kfree(opts.release_agent); |
| 1078 | } |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1079 | |
| 1080 | sb = sget(fs_type, cgroup_test_super, cgroup_set_super, root); |
| 1081 | |
| 1082 | if (IS_ERR(sb)) { |
| 1083 | kfree(root); |
| 1084 | return PTR_ERR(sb); |
| 1085 | } |
| 1086 | |
| 1087 | if (sb->s_fs_info != root) { |
| 1088 | /* Reusing an existing superblock */ |
| 1089 | BUG_ON(sb->s_root == NULL); |
| 1090 | kfree(root); |
| 1091 | root = NULL; |
| 1092 | } else { |
| 1093 | /* New superblock */ |
Li Zefan | c12f65d | 2009-01-07 18:07:42 -0800 | [diff] [blame] | 1094 | struct cgroup *root_cgrp = &root->top_cgroup; |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 1095 | struct inode *inode; |
Li Zefan | 28fd5df | 2008-04-29 01:00:13 -0700 | [diff] [blame] | 1096 | int i; |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1097 | |
| 1098 | BUG_ON(sb->s_root != NULL); |
| 1099 | |
| 1100 | ret = cgroup_get_rootdir(sb); |
| 1101 | if (ret) |
| 1102 | goto drop_new_super; |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 1103 | inode = sb->s_root->d_inode; |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1104 | |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 1105 | mutex_lock(&inode->i_mutex); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1106 | mutex_lock(&cgroup_mutex); |
| 1107 | |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 1108 | /* |
| 1109 | * We're accessing css_set_count without locking |
| 1110 | * css_set_lock here, but that's OK - it can only be |
| 1111 | * increased by someone holding cgroup_lock, and |
| 1112 | * that's us. The worst that can happen is that we |
| 1113 | * have some link structures left over |
| 1114 | */ |
| 1115 | ret = allocate_cg_links(css_set_count, &tmp_cg_links); |
| 1116 | if (ret) { |
| 1117 | mutex_unlock(&cgroup_mutex); |
| 1118 | mutex_unlock(&inode->i_mutex); |
| 1119 | goto drop_new_super; |
| 1120 | } |
| 1121 | |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1122 | ret = rebind_subsystems(root, root->subsys_bits); |
| 1123 | if (ret == -EBUSY) { |
| 1124 | mutex_unlock(&cgroup_mutex); |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 1125 | mutex_unlock(&inode->i_mutex); |
Li Zefan | 20ca9b3 | 2008-12-23 13:57:14 -0800 | [diff] [blame] | 1126 | goto free_cg_links; |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1127 | } |
| 1128 | |
| 1129 | /* EBUSY should be the only error here */ |
| 1130 | BUG_ON(ret); |
| 1131 | |
| 1132 | list_add(&root->root_list, &roots); |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 1133 | root_count++; |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1134 | |
Li Zefan | c12f65d | 2009-01-07 18:07:42 -0800 | [diff] [blame] | 1135 | sb->s_root->d_fsdata = root_cgrp; |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1136 | root->top_cgroup.dentry = sb->s_root; |
| 1137 | |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 1138 | /* Link the top cgroup in this hierarchy into all |
| 1139 | * the css_set objects */ |
| 1140 | write_lock(&css_set_lock); |
Li Zefan | 28fd5df | 2008-04-29 01:00:13 -0700 | [diff] [blame] | 1141 | for (i = 0; i < CSS_SET_TABLE_SIZE; i++) { |
| 1142 | struct hlist_head *hhead = &css_set_table[i]; |
| 1143 | struct hlist_node *node; |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 1144 | struct css_set *cg; |
Li Zefan | 28fd5df | 2008-04-29 01:00:13 -0700 | [diff] [blame] | 1145 | |
Li Zefan | c12f65d | 2009-01-07 18:07:42 -0800 | [diff] [blame] | 1146 | hlist_for_each_entry(cg, node, hhead, hlist) |
| 1147 | link_css_set(&tmp_cg_links, cg, root_cgrp); |
Li Zefan | 28fd5df | 2008-04-29 01:00:13 -0700 | [diff] [blame] | 1148 | } |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 1149 | write_unlock(&css_set_lock); |
| 1150 | |
| 1151 | free_cg_links(&tmp_cg_links); |
| 1152 | |
Li Zefan | c12f65d | 2009-01-07 18:07:42 -0800 | [diff] [blame] | 1153 | BUG_ON(!list_empty(&root_cgrp->sibling)); |
| 1154 | BUG_ON(!list_empty(&root_cgrp->children)); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1155 | BUG_ON(root->number_of_cgroups != 1); |
| 1156 | |
Li Zefan | c12f65d | 2009-01-07 18:07:42 -0800 | [diff] [blame] | 1157 | cgroup_populate_dir(root_cgrp); |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 1158 | mutex_unlock(&inode->i_mutex); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1159 | mutex_unlock(&cgroup_mutex); |
| 1160 | } |
| 1161 | |
Sukadev Bhattiprolu | a3ec947 | 2009-03-04 12:06:34 -0800 | [diff] [blame] | 1162 | simple_set_mnt(mnt, sb); |
| 1163 | return 0; |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1164 | |
Li Zefan | 20ca9b3 | 2008-12-23 13:57:14 -0800 | [diff] [blame] | 1165 | free_cg_links: |
| 1166 | free_cg_links(&tmp_cg_links); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1167 | drop_new_super: |
Al Viro | 6f5bbff | 2009-05-06 01:34:22 -0400 | [diff] [blame] | 1168 | deactivate_locked_super(sb); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1169 | return ret; |
| 1170 | } |
| 1171 | |
| 1172 | static void cgroup_kill_sb(struct super_block *sb) { |
| 1173 | struct cgroupfs_root *root = sb->s_fs_info; |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 1174 | struct cgroup *cgrp = &root->top_cgroup; |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1175 | int ret; |
KOSAKI Motohiro | 71cbb94 | 2008-07-25 01:46:55 -0700 | [diff] [blame] | 1176 | struct cg_cgroup_link *link; |
| 1177 | struct cg_cgroup_link *saved_link; |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1178 | |
| 1179 | BUG_ON(!root); |
| 1180 | |
| 1181 | BUG_ON(root->number_of_cgroups != 1); |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 1182 | BUG_ON(!list_empty(&cgrp->children)); |
| 1183 | BUG_ON(!list_empty(&cgrp->sibling)); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1184 | |
| 1185 | mutex_lock(&cgroup_mutex); |
| 1186 | |
| 1187 | /* Rebind all subsystems back to the default hierarchy */ |
| 1188 | ret = rebind_subsystems(root, 0); |
| 1189 | /* Shouldn't be able to fail ... */ |
| 1190 | BUG_ON(ret); |
| 1191 | |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 1192 | /* |
| 1193 | * Release all the links from css_sets to this hierarchy's |
| 1194 | * root cgroup |
| 1195 | */ |
| 1196 | write_lock(&css_set_lock); |
KOSAKI Motohiro | 71cbb94 | 2008-07-25 01:46:55 -0700 | [diff] [blame] | 1197 | |
| 1198 | list_for_each_entry_safe(link, saved_link, &cgrp->css_sets, |
| 1199 | cgrp_link_list) { |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 1200 | list_del(&link->cg_link_list); |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 1201 | list_del(&link->cgrp_link_list); |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 1202 | kfree(link); |
| 1203 | } |
| 1204 | write_unlock(&css_set_lock); |
| 1205 | |
Paul Menage | 839ec54 | 2009-01-29 14:25:22 -0800 | [diff] [blame] | 1206 | if (!list_empty(&root->root_list)) { |
| 1207 | list_del(&root->root_list); |
| 1208 | root_count--; |
| 1209 | } |
Li Zefan | e5f6a86 | 2009-01-07 18:07:41 -0800 | [diff] [blame] | 1210 | |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1211 | mutex_unlock(&cgroup_mutex); |
| 1212 | |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1213 | kill_litter_super(sb); |
Li Zefan | 67e055d | 2009-02-18 14:48:20 -0800 | [diff] [blame] | 1214 | kfree(root); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1215 | } |
| 1216 | |
| 1217 | static struct file_system_type cgroup_fs_type = { |
| 1218 | .name = "cgroup", |
| 1219 | .get_sb = cgroup_get_sb, |
| 1220 | .kill_sb = cgroup_kill_sb, |
| 1221 | }; |
| 1222 | |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 1223 | static inline struct cgroup *__d_cgrp(struct dentry *dentry) |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1224 | { |
| 1225 | return dentry->d_fsdata; |
| 1226 | } |
| 1227 | |
| 1228 | static inline struct cftype *__d_cft(struct dentry *dentry) |
| 1229 | { |
| 1230 | return dentry->d_fsdata; |
| 1231 | } |
| 1232 | |
Li Zefan | a043e3b | 2008-02-23 15:24:09 -0800 | [diff] [blame] | 1233 | /** |
| 1234 | * cgroup_path - generate the path of a cgroup |
| 1235 | * @cgrp: the cgroup in question |
| 1236 | * @buf: the buffer to write the path into |
| 1237 | * @buflen: the length of the buffer |
| 1238 | * |
Paul Menage | a47295e | 2009-01-07 18:07:44 -0800 | [diff] [blame] | 1239 | * Called with cgroup_mutex held or else with an RCU-protected cgroup |
| 1240 | * reference. Writes path of cgroup into buf. Returns 0 on success, |
| 1241 | * -errno on error. |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1242 | */ |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 1243 | int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen) |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1244 | { |
| 1245 | char *start; |
Paul Menage | a47295e | 2009-01-07 18:07:44 -0800 | [diff] [blame] | 1246 | struct dentry *dentry = rcu_dereference(cgrp->dentry); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1247 | |
Paul Menage | a47295e | 2009-01-07 18:07:44 -0800 | [diff] [blame] | 1248 | if (!dentry || cgrp == dummytop) { |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1249 | /* |
| 1250 | * Inactive subsystems have no dentry for their root |
| 1251 | * cgroup |
| 1252 | */ |
| 1253 | strcpy(buf, "/"); |
| 1254 | return 0; |
| 1255 | } |
| 1256 | |
| 1257 | start = buf + buflen; |
| 1258 | |
| 1259 | *--start = '\0'; |
| 1260 | for (;;) { |
Paul Menage | a47295e | 2009-01-07 18:07:44 -0800 | [diff] [blame] | 1261 | int len = dentry->d_name.len; |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1262 | if ((start -= len) < buf) |
| 1263 | return -ENAMETOOLONG; |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 1264 | memcpy(start, cgrp->dentry->d_name.name, len); |
| 1265 | cgrp = cgrp->parent; |
| 1266 | if (!cgrp) |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1267 | break; |
Paul Menage | a47295e | 2009-01-07 18:07:44 -0800 | [diff] [blame] | 1268 | dentry = rcu_dereference(cgrp->dentry); |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 1269 | if (!cgrp->parent) |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1270 | continue; |
| 1271 | if (--start < buf) |
| 1272 | return -ENAMETOOLONG; |
| 1273 | *start = '/'; |
| 1274 | } |
| 1275 | memmove(buf, start, buf + buflen - start); |
| 1276 | return 0; |
| 1277 | } |
| 1278 | |
Paul Menage | bbcb81d | 2007-10-18 23:39:32 -0700 | [diff] [blame] | 1279 | /* |
| 1280 | * Return the first subsystem attached to a cgroup's hierarchy, and |
| 1281 | * its subsystem id. |
| 1282 | */ |
| 1283 | |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 1284 | static void get_first_subsys(const struct cgroup *cgrp, |
Paul Menage | bbcb81d | 2007-10-18 23:39:32 -0700 | [diff] [blame] | 1285 | struct cgroup_subsys_state **css, int *subsys_id) |
| 1286 | { |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 1287 | const struct cgroupfs_root *root = cgrp->root; |
Paul Menage | bbcb81d | 2007-10-18 23:39:32 -0700 | [diff] [blame] | 1288 | const struct cgroup_subsys *test_ss; |
| 1289 | BUG_ON(list_empty(&root->subsys_list)); |
| 1290 | test_ss = list_entry(root->subsys_list.next, |
| 1291 | struct cgroup_subsys, sibling); |
| 1292 | if (css) { |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 1293 | *css = cgrp->subsys[test_ss->subsys_id]; |
Paul Menage | bbcb81d | 2007-10-18 23:39:32 -0700 | [diff] [blame] | 1294 | BUG_ON(!*css); |
| 1295 | } |
| 1296 | if (subsys_id) |
| 1297 | *subsys_id = test_ss->subsys_id; |
| 1298 | } |
| 1299 | |
Li Zefan | a043e3b | 2008-02-23 15:24:09 -0800 | [diff] [blame] | 1300 | /** |
| 1301 | * cgroup_attach_task - attach task 'tsk' to cgroup 'cgrp' |
| 1302 | * @cgrp: the cgroup the task is attaching to |
| 1303 | * @tsk: the task to be attached |
Paul Menage | bbcb81d | 2007-10-18 23:39:32 -0700 | [diff] [blame] | 1304 | * |
Li Zefan | a043e3b | 2008-02-23 15:24:09 -0800 | [diff] [blame] | 1305 | * Call holding cgroup_mutex. May take task_lock of |
| 1306 | * the task 'tsk' during call. |
Paul Menage | bbcb81d | 2007-10-18 23:39:32 -0700 | [diff] [blame] | 1307 | */ |
Cliff Wickman | 956db3c | 2008-02-07 00:14:43 -0800 | [diff] [blame] | 1308 | int cgroup_attach_task(struct cgroup *cgrp, struct task_struct *tsk) |
Paul Menage | bbcb81d | 2007-10-18 23:39:32 -0700 | [diff] [blame] | 1309 | { |
| 1310 | int retval = 0; |
| 1311 | struct cgroup_subsys *ss; |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 1312 | struct cgroup *oldcgrp; |
Lai Jiangshan | 77efecd | 2009-01-07 18:07:39 -0800 | [diff] [blame] | 1313 | struct css_set *cg; |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 1314 | struct css_set *newcg; |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 1315 | struct cgroupfs_root *root = cgrp->root; |
Paul Menage | bbcb81d | 2007-10-18 23:39:32 -0700 | [diff] [blame] | 1316 | int subsys_id; |
| 1317 | |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 1318 | get_first_subsys(cgrp, NULL, &subsys_id); |
Paul Menage | bbcb81d | 2007-10-18 23:39:32 -0700 | [diff] [blame] | 1319 | |
| 1320 | /* Nothing to do if the task is already in that cgroup */ |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 1321 | oldcgrp = task_cgroup(tsk, subsys_id); |
| 1322 | if (cgrp == oldcgrp) |
Paul Menage | bbcb81d | 2007-10-18 23:39:32 -0700 | [diff] [blame] | 1323 | return 0; |
| 1324 | |
| 1325 | for_each_subsys(root, ss) { |
| 1326 | if (ss->can_attach) { |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 1327 | retval = ss->can_attach(ss, cgrp, tsk); |
Paul Jackson | e18f631 | 2008-02-07 00:13:44 -0800 | [diff] [blame] | 1328 | if (retval) |
Paul Menage | bbcb81d | 2007-10-18 23:39:32 -0700 | [diff] [blame] | 1329 | return retval; |
Paul Menage | bbcb81d | 2007-10-18 23:39:32 -0700 | [diff] [blame] | 1330 | } |
| 1331 | } |
| 1332 | |
Lai Jiangshan | 77efecd | 2009-01-07 18:07:39 -0800 | [diff] [blame] | 1333 | task_lock(tsk); |
| 1334 | cg = tsk->cgroups; |
| 1335 | get_css_set(cg); |
| 1336 | task_unlock(tsk); |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 1337 | /* |
| 1338 | * Locate or allocate a new css_set for this task, |
| 1339 | * based on its final set of cgroups |
| 1340 | */ |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 1341 | newcg = find_css_set(cg, cgrp); |
Lai Jiangshan | 77efecd | 2009-01-07 18:07:39 -0800 | [diff] [blame] | 1342 | put_css_set(cg); |
Paul Jackson | e18f631 | 2008-02-07 00:13:44 -0800 | [diff] [blame] | 1343 | if (!newcg) |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 1344 | return -ENOMEM; |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 1345 | |
Paul Menage | bbcb81d | 2007-10-18 23:39:32 -0700 | [diff] [blame] | 1346 | task_lock(tsk); |
| 1347 | if (tsk->flags & PF_EXITING) { |
| 1348 | task_unlock(tsk); |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 1349 | put_css_set(newcg); |
Paul Menage | bbcb81d | 2007-10-18 23:39:32 -0700 | [diff] [blame] | 1350 | return -ESRCH; |
| 1351 | } |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 1352 | rcu_assign_pointer(tsk->cgroups, newcg); |
Paul Menage | bbcb81d | 2007-10-18 23:39:32 -0700 | [diff] [blame] | 1353 | task_unlock(tsk); |
| 1354 | |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 1355 | /* Update the css_set linked lists if we're using them */ |
| 1356 | write_lock(&css_set_lock); |
| 1357 | if (!list_empty(&tsk->cg_list)) { |
| 1358 | list_del(&tsk->cg_list); |
| 1359 | list_add(&tsk->cg_list, &newcg->tasks); |
| 1360 | } |
| 1361 | write_unlock(&css_set_lock); |
| 1362 | |
Paul Menage | bbcb81d | 2007-10-18 23:39:32 -0700 | [diff] [blame] | 1363 | for_each_subsys(root, ss) { |
Paul Jackson | e18f631 | 2008-02-07 00:13:44 -0800 | [diff] [blame] | 1364 | if (ss->attach) |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 1365 | ss->attach(ss, cgrp, oldcgrp, tsk); |
Paul Menage | bbcb81d | 2007-10-18 23:39:32 -0700 | [diff] [blame] | 1366 | } |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 1367 | set_bit(CGRP_RELEASABLE, &oldcgrp->flags); |
Paul Menage | bbcb81d | 2007-10-18 23:39:32 -0700 | [diff] [blame] | 1368 | synchronize_rcu(); |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 1369 | put_css_set(cg); |
KAMEZAWA Hiroyuki | ec64f51 | 2009-04-02 16:57:26 -0700 | [diff] [blame] | 1370 | |
| 1371 | /* |
| 1372 | * wake up rmdir() waiter. the rmdir should fail since the cgroup |
| 1373 | * is no longer empty. |
| 1374 | */ |
KAMEZAWA Hiroyuki | 8870326 | 2009-07-29 15:04:06 -0700 | [diff] [blame] | 1375 | cgroup_wakeup_rmdir_waiter(cgrp); |
Paul Menage | bbcb81d | 2007-10-18 23:39:32 -0700 | [diff] [blame] | 1376 | return 0; |
| 1377 | } |
| 1378 | |
| 1379 | /* |
Paul Menage | af35102 | 2008-07-25 01:47:01 -0700 | [diff] [blame] | 1380 | * Attach task with pid 'pid' to cgroup 'cgrp'. Call with cgroup_mutex |
| 1381 | * held. May take task_lock of task |
Paul Menage | bbcb81d | 2007-10-18 23:39:32 -0700 | [diff] [blame] | 1382 | */ |
Paul Menage | af35102 | 2008-07-25 01:47:01 -0700 | [diff] [blame] | 1383 | static int attach_task_by_pid(struct cgroup *cgrp, u64 pid) |
Paul Menage | bbcb81d | 2007-10-18 23:39:32 -0700 | [diff] [blame] | 1384 | { |
Paul Menage | bbcb81d | 2007-10-18 23:39:32 -0700 | [diff] [blame] | 1385 | struct task_struct *tsk; |
David Howells | c69e8d9 | 2008-11-14 10:39:19 +1100 | [diff] [blame] | 1386 | const struct cred *cred = current_cred(), *tcred; |
Paul Menage | bbcb81d | 2007-10-18 23:39:32 -0700 | [diff] [blame] | 1387 | int ret; |
| 1388 | |
Paul Menage | bbcb81d | 2007-10-18 23:39:32 -0700 | [diff] [blame] | 1389 | if (pid) { |
| 1390 | rcu_read_lock(); |
Pavel Emelyanov | 73507f3 | 2008-02-07 00:14:47 -0800 | [diff] [blame] | 1391 | tsk = find_task_by_vpid(pid); |
Paul Menage | bbcb81d | 2007-10-18 23:39:32 -0700 | [diff] [blame] | 1392 | if (!tsk || tsk->flags & PF_EXITING) { |
| 1393 | rcu_read_unlock(); |
| 1394 | return -ESRCH; |
| 1395 | } |
Paul Menage | bbcb81d | 2007-10-18 23:39:32 -0700 | [diff] [blame] | 1396 | |
David Howells | c69e8d9 | 2008-11-14 10:39:19 +1100 | [diff] [blame] | 1397 | tcred = __task_cred(tsk); |
| 1398 | if (cred->euid && |
| 1399 | cred->euid != tcred->uid && |
| 1400 | cred->euid != tcred->suid) { |
| 1401 | rcu_read_unlock(); |
Paul Menage | bbcb81d | 2007-10-18 23:39:32 -0700 | [diff] [blame] | 1402 | return -EACCES; |
| 1403 | } |
David Howells | c69e8d9 | 2008-11-14 10:39:19 +1100 | [diff] [blame] | 1404 | get_task_struct(tsk); |
| 1405 | rcu_read_unlock(); |
Paul Menage | bbcb81d | 2007-10-18 23:39:32 -0700 | [diff] [blame] | 1406 | } else { |
| 1407 | tsk = current; |
| 1408 | get_task_struct(tsk); |
| 1409 | } |
| 1410 | |
Cliff Wickman | 956db3c | 2008-02-07 00:14:43 -0800 | [diff] [blame] | 1411 | ret = cgroup_attach_task(cgrp, tsk); |
Paul Menage | bbcb81d | 2007-10-18 23:39:32 -0700 | [diff] [blame] | 1412 | put_task_struct(tsk); |
| 1413 | return ret; |
| 1414 | } |
| 1415 | |
Paul Menage | af35102 | 2008-07-25 01:47:01 -0700 | [diff] [blame] | 1416 | static int cgroup_tasks_write(struct cgroup *cgrp, struct cftype *cft, u64 pid) |
| 1417 | { |
| 1418 | int ret; |
| 1419 | if (!cgroup_lock_live_group(cgrp)) |
| 1420 | return -ENODEV; |
| 1421 | ret = attach_task_by_pid(cgrp, pid); |
| 1422 | cgroup_unlock(); |
| 1423 | return ret; |
| 1424 | } |
| 1425 | |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1426 | /* The various types of files and directories in a cgroup file system */ |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1427 | enum cgroup_filetype { |
| 1428 | FILE_ROOT, |
| 1429 | FILE_DIR, |
| 1430 | FILE_TASKLIST, |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 1431 | FILE_NOTIFY_ON_RELEASE, |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 1432 | FILE_RELEASE_AGENT, |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1433 | }; |
| 1434 | |
Paul Menage | e788e06 | 2008-07-25 01:46:59 -0700 | [diff] [blame] | 1435 | /** |
| 1436 | * cgroup_lock_live_group - take cgroup_mutex and check that cgrp is alive. |
| 1437 | * @cgrp: the cgroup to be checked for liveness |
| 1438 | * |
Paul Menage | 84eea84 | 2008-07-25 01:47:00 -0700 | [diff] [blame] | 1439 | * On success, returns true; the lock should be later released with |
| 1440 | * cgroup_unlock(). On failure returns false with no lock held. |
Paul Menage | e788e06 | 2008-07-25 01:46:59 -0700 | [diff] [blame] | 1441 | */ |
Paul Menage | 84eea84 | 2008-07-25 01:47:00 -0700 | [diff] [blame] | 1442 | bool cgroup_lock_live_group(struct cgroup *cgrp) |
Paul Menage | e788e06 | 2008-07-25 01:46:59 -0700 | [diff] [blame] | 1443 | { |
| 1444 | mutex_lock(&cgroup_mutex); |
| 1445 | if (cgroup_is_removed(cgrp)) { |
| 1446 | mutex_unlock(&cgroup_mutex); |
| 1447 | return false; |
| 1448 | } |
| 1449 | return true; |
| 1450 | } |
| 1451 | |
| 1452 | static int cgroup_release_agent_write(struct cgroup *cgrp, struct cftype *cft, |
| 1453 | const char *buffer) |
| 1454 | { |
| 1455 | BUILD_BUG_ON(sizeof(cgrp->root->release_agent_path) < PATH_MAX); |
| 1456 | if (!cgroup_lock_live_group(cgrp)) |
| 1457 | return -ENODEV; |
| 1458 | strcpy(cgrp->root->release_agent_path, buffer); |
Paul Menage | 84eea84 | 2008-07-25 01:47:00 -0700 | [diff] [blame] | 1459 | cgroup_unlock(); |
Paul Menage | e788e06 | 2008-07-25 01:46:59 -0700 | [diff] [blame] | 1460 | return 0; |
| 1461 | } |
| 1462 | |
| 1463 | static int cgroup_release_agent_show(struct cgroup *cgrp, struct cftype *cft, |
| 1464 | struct seq_file *seq) |
| 1465 | { |
| 1466 | if (!cgroup_lock_live_group(cgrp)) |
| 1467 | return -ENODEV; |
| 1468 | seq_puts(seq, cgrp->root->release_agent_path); |
| 1469 | seq_putc(seq, '\n'); |
Paul Menage | 84eea84 | 2008-07-25 01:47:00 -0700 | [diff] [blame] | 1470 | cgroup_unlock(); |
Paul Menage | e788e06 | 2008-07-25 01:46:59 -0700 | [diff] [blame] | 1471 | return 0; |
| 1472 | } |
| 1473 | |
Paul Menage | 84eea84 | 2008-07-25 01:47:00 -0700 | [diff] [blame] | 1474 | /* A buffer size big enough for numbers or short strings */ |
| 1475 | #define CGROUP_LOCAL_BUFFER_SIZE 64 |
| 1476 | |
Paul Menage | e73d2c6 | 2008-04-29 01:00:06 -0700 | [diff] [blame] | 1477 | static ssize_t cgroup_write_X64(struct cgroup *cgrp, struct cftype *cft, |
Paul Menage | f4c753b | 2008-04-29 00:59:56 -0700 | [diff] [blame] | 1478 | struct file *file, |
| 1479 | const char __user *userbuf, |
| 1480 | size_t nbytes, loff_t *unused_ppos) |
Paul Menage | 355e0c4 | 2007-10-18 23:39:33 -0700 | [diff] [blame] | 1481 | { |
Paul Menage | 84eea84 | 2008-07-25 01:47:00 -0700 | [diff] [blame] | 1482 | char buffer[CGROUP_LOCAL_BUFFER_SIZE]; |
Paul Menage | 355e0c4 | 2007-10-18 23:39:33 -0700 | [diff] [blame] | 1483 | int retval = 0; |
Paul Menage | 355e0c4 | 2007-10-18 23:39:33 -0700 | [diff] [blame] | 1484 | char *end; |
| 1485 | |
| 1486 | if (!nbytes) |
| 1487 | return -EINVAL; |
| 1488 | if (nbytes >= sizeof(buffer)) |
| 1489 | return -E2BIG; |
| 1490 | if (copy_from_user(buffer, userbuf, nbytes)) |
| 1491 | return -EFAULT; |
| 1492 | |
| 1493 | buffer[nbytes] = 0; /* nul-terminate */ |
Paul Menage | b7269df | 2008-04-29 00:59:59 -0700 | [diff] [blame] | 1494 | strstrip(buffer); |
Paul Menage | e73d2c6 | 2008-04-29 01:00:06 -0700 | [diff] [blame] | 1495 | if (cft->write_u64) { |
| 1496 | u64 val = simple_strtoull(buffer, &end, 0); |
| 1497 | if (*end) |
| 1498 | return -EINVAL; |
| 1499 | retval = cft->write_u64(cgrp, cft, val); |
| 1500 | } else { |
| 1501 | s64 val = simple_strtoll(buffer, &end, 0); |
| 1502 | if (*end) |
| 1503 | return -EINVAL; |
| 1504 | retval = cft->write_s64(cgrp, cft, val); |
| 1505 | } |
Paul Menage | 355e0c4 | 2007-10-18 23:39:33 -0700 | [diff] [blame] | 1506 | if (!retval) |
| 1507 | retval = nbytes; |
| 1508 | return retval; |
| 1509 | } |
| 1510 | |
Paul Menage | db3b149 | 2008-07-25 01:46:58 -0700 | [diff] [blame] | 1511 | static ssize_t cgroup_write_string(struct cgroup *cgrp, struct cftype *cft, |
| 1512 | struct file *file, |
| 1513 | const char __user *userbuf, |
| 1514 | size_t nbytes, loff_t *unused_ppos) |
| 1515 | { |
Paul Menage | 84eea84 | 2008-07-25 01:47:00 -0700 | [diff] [blame] | 1516 | char local_buffer[CGROUP_LOCAL_BUFFER_SIZE]; |
Paul Menage | db3b149 | 2008-07-25 01:46:58 -0700 | [diff] [blame] | 1517 | int retval = 0; |
| 1518 | size_t max_bytes = cft->max_write_len; |
| 1519 | char *buffer = local_buffer; |
| 1520 | |
| 1521 | if (!max_bytes) |
| 1522 | max_bytes = sizeof(local_buffer) - 1; |
| 1523 | if (nbytes >= max_bytes) |
| 1524 | return -E2BIG; |
| 1525 | /* Allocate a dynamic buffer if we need one */ |
| 1526 | if (nbytes >= sizeof(local_buffer)) { |
| 1527 | buffer = kmalloc(nbytes + 1, GFP_KERNEL); |
| 1528 | if (buffer == NULL) |
| 1529 | return -ENOMEM; |
| 1530 | } |
Li Zefan | 5a3eb9f | 2008-07-29 22:33:18 -0700 | [diff] [blame] | 1531 | if (nbytes && copy_from_user(buffer, userbuf, nbytes)) { |
| 1532 | retval = -EFAULT; |
| 1533 | goto out; |
| 1534 | } |
Paul Menage | db3b149 | 2008-07-25 01:46:58 -0700 | [diff] [blame] | 1535 | |
| 1536 | buffer[nbytes] = 0; /* nul-terminate */ |
| 1537 | strstrip(buffer); |
| 1538 | retval = cft->write_string(cgrp, cft, buffer); |
| 1539 | if (!retval) |
| 1540 | retval = nbytes; |
Li Zefan | 5a3eb9f | 2008-07-29 22:33:18 -0700 | [diff] [blame] | 1541 | out: |
Paul Menage | db3b149 | 2008-07-25 01:46:58 -0700 | [diff] [blame] | 1542 | if (buffer != local_buffer) |
| 1543 | kfree(buffer); |
| 1544 | return retval; |
| 1545 | } |
| 1546 | |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1547 | static ssize_t cgroup_file_write(struct file *file, const char __user *buf, |
| 1548 | size_t nbytes, loff_t *ppos) |
| 1549 | { |
| 1550 | struct cftype *cft = __d_cft(file->f_dentry); |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 1551 | struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1552 | |
Li Zefan | 75139b8 | 2009-01-07 18:07:33 -0800 | [diff] [blame] | 1553 | if (cgroup_is_removed(cgrp)) |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1554 | return -ENODEV; |
Paul Menage | 355e0c4 | 2007-10-18 23:39:33 -0700 | [diff] [blame] | 1555 | if (cft->write) |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 1556 | return cft->write(cgrp, cft, file, buf, nbytes, ppos); |
Paul Menage | e73d2c6 | 2008-04-29 01:00:06 -0700 | [diff] [blame] | 1557 | if (cft->write_u64 || cft->write_s64) |
| 1558 | return cgroup_write_X64(cgrp, cft, file, buf, nbytes, ppos); |
Paul Menage | db3b149 | 2008-07-25 01:46:58 -0700 | [diff] [blame] | 1559 | if (cft->write_string) |
| 1560 | return cgroup_write_string(cgrp, cft, file, buf, nbytes, ppos); |
Pavel Emelyanov | d447ea2 | 2008-04-29 01:00:08 -0700 | [diff] [blame] | 1561 | if (cft->trigger) { |
| 1562 | int ret = cft->trigger(cgrp, (unsigned int)cft->private); |
| 1563 | return ret ? ret : nbytes; |
| 1564 | } |
Paul Menage | 355e0c4 | 2007-10-18 23:39:33 -0700 | [diff] [blame] | 1565 | return -EINVAL; |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1566 | } |
| 1567 | |
Paul Menage | f4c753b | 2008-04-29 00:59:56 -0700 | [diff] [blame] | 1568 | static ssize_t cgroup_read_u64(struct cgroup *cgrp, struct cftype *cft, |
| 1569 | struct file *file, |
| 1570 | char __user *buf, size_t nbytes, |
| 1571 | loff_t *ppos) |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1572 | { |
Paul Menage | 84eea84 | 2008-07-25 01:47:00 -0700 | [diff] [blame] | 1573 | char tmp[CGROUP_LOCAL_BUFFER_SIZE]; |
Paul Menage | f4c753b | 2008-04-29 00:59:56 -0700 | [diff] [blame] | 1574 | u64 val = cft->read_u64(cgrp, cft); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1575 | int len = sprintf(tmp, "%llu\n", (unsigned long long) val); |
| 1576 | |
| 1577 | return simple_read_from_buffer(buf, nbytes, ppos, tmp, len); |
| 1578 | } |
| 1579 | |
Paul Menage | e73d2c6 | 2008-04-29 01:00:06 -0700 | [diff] [blame] | 1580 | static ssize_t cgroup_read_s64(struct cgroup *cgrp, struct cftype *cft, |
| 1581 | struct file *file, |
| 1582 | char __user *buf, size_t nbytes, |
| 1583 | loff_t *ppos) |
| 1584 | { |
Paul Menage | 84eea84 | 2008-07-25 01:47:00 -0700 | [diff] [blame] | 1585 | char tmp[CGROUP_LOCAL_BUFFER_SIZE]; |
Paul Menage | e73d2c6 | 2008-04-29 01:00:06 -0700 | [diff] [blame] | 1586 | s64 val = cft->read_s64(cgrp, cft); |
| 1587 | int len = sprintf(tmp, "%lld\n", (long long) val); |
| 1588 | |
| 1589 | return simple_read_from_buffer(buf, nbytes, ppos, tmp, len); |
| 1590 | } |
| 1591 | |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1592 | static ssize_t cgroup_file_read(struct file *file, char __user *buf, |
| 1593 | size_t nbytes, loff_t *ppos) |
| 1594 | { |
| 1595 | struct cftype *cft = __d_cft(file->f_dentry); |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 1596 | struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1597 | |
Li Zefan | 75139b8 | 2009-01-07 18:07:33 -0800 | [diff] [blame] | 1598 | if (cgroup_is_removed(cgrp)) |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1599 | return -ENODEV; |
| 1600 | |
| 1601 | if (cft->read) |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 1602 | return cft->read(cgrp, cft, file, buf, nbytes, ppos); |
Paul Menage | f4c753b | 2008-04-29 00:59:56 -0700 | [diff] [blame] | 1603 | if (cft->read_u64) |
| 1604 | return cgroup_read_u64(cgrp, cft, file, buf, nbytes, ppos); |
Paul Menage | e73d2c6 | 2008-04-29 01:00:06 -0700 | [diff] [blame] | 1605 | if (cft->read_s64) |
| 1606 | return cgroup_read_s64(cgrp, cft, file, buf, nbytes, ppos); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1607 | return -EINVAL; |
| 1608 | } |
| 1609 | |
Paul Menage | 9179656 | 2008-04-29 01:00:01 -0700 | [diff] [blame] | 1610 | /* |
| 1611 | * seqfile ops/methods for returning structured data. Currently just |
| 1612 | * supports string->u64 maps, but can be extended in future. |
| 1613 | */ |
| 1614 | |
| 1615 | struct cgroup_seqfile_state { |
| 1616 | struct cftype *cft; |
| 1617 | struct cgroup *cgroup; |
| 1618 | }; |
| 1619 | |
| 1620 | static int cgroup_map_add(struct cgroup_map_cb *cb, const char *key, u64 value) |
| 1621 | { |
| 1622 | struct seq_file *sf = cb->state; |
| 1623 | return seq_printf(sf, "%s %llu\n", key, (unsigned long long)value); |
| 1624 | } |
| 1625 | |
| 1626 | static int cgroup_seqfile_show(struct seq_file *m, void *arg) |
| 1627 | { |
| 1628 | struct cgroup_seqfile_state *state = m->private; |
| 1629 | struct cftype *cft = state->cft; |
Serge E. Hallyn | 29486df | 2008-04-29 01:00:14 -0700 | [diff] [blame] | 1630 | if (cft->read_map) { |
| 1631 | struct cgroup_map_cb cb = { |
| 1632 | .fill = cgroup_map_add, |
| 1633 | .state = m, |
| 1634 | }; |
| 1635 | return cft->read_map(state->cgroup, cft, &cb); |
| 1636 | } |
| 1637 | return cft->read_seq_string(state->cgroup, cft, m); |
Paul Menage | 9179656 | 2008-04-29 01:00:01 -0700 | [diff] [blame] | 1638 | } |
| 1639 | |
Adrian Bunk | 96930a6 | 2008-07-25 19:46:21 -0700 | [diff] [blame] | 1640 | static int cgroup_seqfile_release(struct inode *inode, struct file *file) |
Paul Menage | 9179656 | 2008-04-29 01:00:01 -0700 | [diff] [blame] | 1641 | { |
| 1642 | struct seq_file *seq = file->private_data; |
| 1643 | kfree(seq->private); |
| 1644 | return single_release(inode, file); |
| 1645 | } |
| 1646 | |
| 1647 | static struct file_operations cgroup_seqfile_operations = { |
| 1648 | .read = seq_read, |
Paul Menage | e788e06 | 2008-07-25 01:46:59 -0700 | [diff] [blame] | 1649 | .write = cgroup_file_write, |
Paul Menage | 9179656 | 2008-04-29 01:00:01 -0700 | [diff] [blame] | 1650 | .llseek = seq_lseek, |
| 1651 | .release = cgroup_seqfile_release, |
| 1652 | }; |
| 1653 | |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1654 | static int cgroup_file_open(struct inode *inode, struct file *file) |
| 1655 | { |
| 1656 | int err; |
| 1657 | struct cftype *cft; |
| 1658 | |
| 1659 | err = generic_file_open(inode, file); |
| 1660 | if (err) |
| 1661 | return err; |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1662 | cft = __d_cft(file->f_dentry); |
Li Zefan | 75139b8 | 2009-01-07 18:07:33 -0800 | [diff] [blame] | 1663 | |
Serge E. Hallyn | 29486df | 2008-04-29 01:00:14 -0700 | [diff] [blame] | 1664 | if (cft->read_map || cft->read_seq_string) { |
Paul Menage | 9179656 | 2008-04-29 01:00:01 -0700 | [diff] [blame] | 1665 | struct cgroup_seqfile_state *state = |
| 1666 | kzalloc(sizeof(*state), GFP_USER); |
| 1667 | if (!state) |
| 1668 | return -ENOMEM; |
| 1669 | state->cft = cft; |
| 1670 | state->cgroup = __d_cgrp(file->f_dentry->d_parent); |
| 1671 | file->f_op = &cgroup_seqfile_operations; |
| 1672 | err = single_open(file, cgroup_seqfile_show, state); |
| 1673 | if (err < 0) |
| 1674 | kfree(state); |
| 1675 | } else if (cft->open) |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1676 | err = cft->open(inode, file); |
| 1677 | else |
| 1678 | err = 0; |
| 1679 | |
| 1680 | return err; |
| 1681 | } |
| 1682 | |
| 1683 | static int cgroup_file_release(struct inode *inode, struct file *file) |
| 1684 | { |
| 1685 | struct cftype *cft = __d_cft(file->f_dentry); |
| 1686 | if (cft->release) |
| 1687 | return cft->release(inode, file); |
| 1688 | return 0; |
| 1689 | } |
| 1690 | |
| 1691 | /* |
| 1692 | * cgroup_rename - Only allow simple rename of directories in place. |
| 1693 | */ |
| 1694 | static int cgroup_rename(struct inode *old_dir, struct dentry *old_dentry, |
| 1695 | struct inode *new_dir, struct dentry *new_dentry) |
| 1696 | { |
| 1697 | if (!S_ISDIR(old_dentry->d_inode->i_mode)) |
| 1698 | return -ENOTDIR; |
| 1699 | if (new_dentry->d_inode) |
| 1700 | return -EEXIST; |
| 1701 | if (old_dir != new_dir) |
| 1702 | return -EIO; |
| 1703 | return simple_rename(old_dir, old_dentry, new_dir, new_dentry); |
| 1704 | } |
| 1705 | |
| 1706 | static struct file_operations cgroup_file_operations = { |
| 1707 | .read = cgroup_file_read, |
| 1708 | .write = cgroup_file_write, |
| 1709 | .llseek = generic_file_llseek, |
| 1710 | .open = cgroup_file_open, |
| 1711 | .release = cgroup_file_release, |
| 1712 | }; |
| 1713 | |
| 1714 | static struct inode_operations cgroup_dir_inode_operations = { |
| 1715 | .lookup = simple_lookup, |
| 1716 | .mkdir = cgroup_mkdir, |
| 1717 | .rmdir = cgroup_rmdir, |
| 1718 | .rename = cgroup_rename, |
| 1719 | }; |
| 1720 | |
Li Zefan | 099fca3 | 2009-04-02 16:57:29 -0700 | [diff] [blame] | 1721 | static int cgroup_create_file(struct dentry *dentry, mode_t mode, |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1722 | struct super_block *sb) |
| 1723 | { |
Al Viro | 3ba13d1 | 2009-02-20 06:02:22 +0000 | [diff] [blame] | 1724 | static const struct dentry_operations cgroup_dops = { |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1725 | .d_iput = cgroup_diput, |
| 1726 | }; |
| 1727 | |
| 1728 | struct inode *inode; |
| 1729 | |
| 1730 | if (!dentry) |
| 1731 | return -ENOENT; |
| 1732 | if (dentry->d_inode) |
| 1733 | return -EEXIST; |
| 1734 | |
| 1735 | inode = cgroup_new_inode(mode, sb); |
| 1736 | if (!inode) |
| 1737 | return -ENOMEM; |
| 1738 | |
| 1739 | if (S_ISDIR(mode)) { |
| 1740 | inode->i_op = &cgroup_dir_inode_operations; |
| 1741 | inode->i_fop = &simple_dir_operations; |
| 1742 | |
| 1743 | /* start off with i_nlink == 2 (for "." entry) */ |
| 1744 | inc_nlink(inode); |
| 1745 | |
| 1746 | /* start with the directory inode held, so that we can |
| 1747 | * populate it without racing with another mkdir */ |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 1748 | mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1749 | } else if (S_ISREG(mode)) { |
| 1750 | inode->i_size = 0; |
| 1751 | inode->i_fop = &cgroup_file_operations; |
| 1752 | } |
| 1753 | dentry->d_op = &cgroup_dops; |
| 1754 | d_instantiate(dentry, inode); |
| 1755 | dget(dentry); /* Extra count - pin the dentry in core */ |
| 1756 | return 0; |
| 1757 | } |
| 1758 | |
| 1759 | /* |
Li Zefan | a043e3b | 2008-02-23 15:24:09 -0800 | [diff] [blame] | 1760 | * cgroup_create_dir - create a directory for an object. |
| 1761 | * @cgrp: the cgroup we create the directory for. It must have a valid |
| 1762 | * ->parent field. And we are going to fill its ->dentry field. |
| 1763 | * @dentry: dentry of the new cgroup |
| 1764 | * @mode: mode to set on new directory. |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1765 | */ |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 1766 | static int cgroup_create_dir(struct cgroup *cgrp, struct dentry *dentry, |
Li Zefan | 099fca3 | 2009-04-02 16:57:29 -0700 | [diff] [blame] | 1767 | mode_t mode) |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1768 | { |
| 1769 | struct dentry *parent; |
| 1770 | int error = 0; |
| 1771 | |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 1772 | parent = cgrp->parent->dentry; |
| 1773 | error = cgroup_create_file(dentry, S_IFDIR | mode, cgrp->root->sb); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1774 | if (!error) { |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 1775 | dentry->d_fsdata = cgrp; |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1776 | inc_nlink(parent->d_inode); |
Paul Menage | a47295e | 2009-01-07 18:07:44 -0800 | [diff] [blame] | 1777 | rcu_assign_pointer(cgrp->dentry, dentry); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1778 | dget(dentry); |
| 1779 | } |
| 1780 | dput(dentry); |
| 1781 | |
| 1782 | return error; |
| 1783 | } |
| 1784 | |
Li Zefan | 099fca3 | 2009-04-02 16:57:29 -0700 | [diff] [blame] | 1785 | /** |
| 1786 | * cgroup_file_mode - deduce file mode of a control file |
| 1787 | * @cft: the control file in question |
| 1788 | * |
| 1789 | * returns cft->mode if ->mode is not 0 |
| 1790 | * returns S_IRUGO|S_IWUSR if it has both a read and a write handler |
| 1791 | * returns S_IRUGO if it has only a read handler |
| 1792 | * returns S_IWUSR if it has only a write hander |
| 1793 | */ |
| 1794 | static mode_t cgroup_file_mode(const struct cftype *cft) |
| 1795 | { |
| 1796 | mode_t mode = 0; |
| 1797 | |
| 1798 | if (cft->mode) |
| 1799 | return cft->mode; |
| 1800 | |
| 1801 | if (cft->read || cft->read_u64 || cft->read_s64 || |
| 1802 | cft->read_map || cft->read_seq_string) |
| 1803 | mode |= S_IRUGO; |
| 1804 | |
| 1805 | if (cft->write || cft->write_u64 || cft->write_s64 || |
| 1806 | cft->write_string || cft->trigger) |
| 1807 | mode |= S_IWUSR; |
| 1808 | |
| 1809 | return mode; |
| 1810 | } |
| 1811 | |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 1812 | int cgroup_add_file(struct cgroup *cgrp, |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1813 | struct cgroup_subsys *subsys, |
| 1814 | const struct cftype *cft) |
| 1815 | { |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 1816 | struct dentry *dir = cgrp->dentry; |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1817 | struct dentry *dentry; |
| 1818 | int error; |
Li Zefan | 099fca3 | 2009-04-02 16:57:29 -0700 | [diff] [blame] | 1819 | mode_t mode; |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1820 | |
| 1821 | char name[MAX_CGROUP_TYPE_NAMELEN + MAX_CFTYPE_NAME + 2] = { 0 }; |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 1822 | if (subsys && !test_bit(ROOT_NOPREFIX, &cgrp->root->flags)) { |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1823 | strcpy(name, subsys->name); |
| 1824 | strcat(name, "."); |
| 1825 | } |
| 1826 | strcat(name, cft->name); |
| 1827 | BUG_ON(!mutex_is_locked(&dir->d_inode->i_mutex)); |
| 1828 | dentry = lookup_one_len(name, dir, strlen(name)); |
| 1829 | if (!IS_ERR(dentry)) { |
Li Zefan | 099fca3 | 2009-04-02 16:57:29 -0700 | [diff] [blame] | 1830 | mode = cgroup_file_mode(cft); |
| 1831 | error = cgroup_create_file(dentry, mode | S_IFREG, |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 1832 | cgrp->root->sb); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1833 | if (!error) |
| 1834 | dentry->d_fsdata = (void *)cft; |
| 1835 | dput(dentry); |
| 1836 | } else |
| 1837 | error = PTR_ERR(dentry); |
| 1838 | return error; |
| 1839 | } |
| 1840 | |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 1841 | int cgroup_add_files(struct cgroup *cgrp, |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1842 | struct cgroup_subsys *subsys, |
| 1843 | const struct cftype cft[], |
| 1844 | int count) |
| 1845 | { |
| 1846 | int i, err; |
| 1847 | for (i = 0; i < count; i++) { |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 1848 | err = cgroup_add_file(cgrp, subsys, &cft[i]); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 1849 | if (err) |
| 1850 | return err; |
| 1851 | } |
| 1852 | return 0; |
| 1853 | } |
| 1854 | |
Li Zefan | a043e3b | 2008-02-23 15:24:09 -0800 | [diff] [blame] | 1855 | /** |
| 1856 | * cgroup_task_count - count the number of tasks in a cgroup. |
| 1857 | * @cgrp: the cgroup in question |
| 1858 | * |
| 1859 | * Return the number of tasks in the cgroup. |
| 1860 | */ |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 1861 | int cgroup_task_count(const struct cgroup *cgrp) |
Paul Menage | bbcb81d | 2007-10-18 23:39:32 -0700 | [diff] [blame] | 1862 | { |
| 1863 | int count = 0; |
KOSAKI Motohiro | 71cbb94 | 2008-07-25 01:46:55 -0700 | [diff] [blame] | 1864 | struct cg_cgroup_link *link; |
Paul Menage | bbcb81d | 2007-10-18 23:39:32 -0700 | [diff] [blame] | 1865 | |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 1866 | read_lock(&css_set_lock); |
KOSAKI Motohiro | 71cbb94 | 2008-07-25 01:46:55 -0700 | [diff] [blame] | 1867 | list_for_each_entry(link, &cgrp->css_sets, cgrp_link_list) { |
Lai Jiangshan | 146aa1b | 2008-10-18 20:28:03 -0700 | [diff] [blame] | 1868 | count += atomic_read(&link->cg->refcount); |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 1869 | } |
| 1870 | read_unlock(&css_set_lock); |
Paul Menage | bbcb81d | 2007-10-18 23:39:32 -0700 | [diff] [blame] | 1871 | return count; |
| 1872 | } |
| 1873 | |
| 1874 | /* |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 1875 | * Advance a list_head iterator. The iterator should be positioned at |
| 1876 | * the start of a css_set |
| 1877 | */ |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 1878 | static void cgroup_advance_iter(struct cgroup *cgrp, |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 1879 | struct cgroup_iter *it) |
| 1880 | { |
| 1881 | struct list_head *l = it->cg_link; |
| 1882 | struct cg_cgroup_link *link; |
| 1883 | struct css_set *cg; |
| 1884 | |
| 1885 | /* Advance to the next non-empty css_set */ |
| 1886 | do { |
| 1887 | l = l->next; |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 1888 | if (l == &cgrp->css_sets) { |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 1889 | it->cg_link = NULL; |
| 1890 | return; |
| 1891 | } |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 1892 | link = list_entry(l, struct cg_cgroup_link, cgrp_link_list); |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 1893 | cg = link->cg; |
| 1894 | } while (list_empty(&cg->tasks)); |
| 1895 | it->cg_link = l; |
| 1896 | it->task = cg->tasks.next; |
| 1897 | } |
| 1898 | |
Cliff Wickman | 31a7df0 | 2008-02-07 00:14:42 -0800 | [diff] [blame] | 1899 | /* |
| 1900 | * To reduce the fork() overhead for systems that are not actually |
| 1901 | * using their cgroups capability, we don't maintain the lists running |
| 1902 | * through each css_set to its tasks until we see the list actually |
| 1903 | * used - in other words after the first call to cgroup_iter_start(). |
| 1904 | * |
| 1905 | * The tasklist_lock is not held here, as do_each_thread() and |
| 1906 | * while_each_thread() are protected by RCU. |
| 1907 | */ |
Adrian Bunk | 3df91fe | 2008-04-29 00:59:54 -0700 | [diff] [blame] | 1908 | static void cgroup_enable_task_cg_lists(void) |
Cliff Wickman | 31a7df0 | 2008-02-07 00:14:42 -0800 | [diff] [blame] | 1909 | { |
| 1910 | struct task_struct *p, *g; |
| 1911 | write_lock(&css_set_lock); |
| 1912 | use_task_css_set_links = 1; |
| 1913 | do_each_thread(g, p) { |
| 1914 | task_lock(p); |
Li Zefan | 0e04388 | 2008-04-17 11:37:15 +0800 | [diff] [blame] | 1915 | /* |
| 1916 | * We should check if the process is exiting, otherwise |
| 1917 | * it will race with cgroup_exit() in that the list |
| 1918 | * entry won't be deleted though the process has exited. |
| 1919 | */ |
| 1920 | if (!(p->flags & PF_EXITING) && list_empty(&p->cg_list)) |
Cliff Wickman | 31a7df0 | 2008-02-07 00:14:42 -0800 | [diff] [blame] | 1921 | list_add(&p->cg_list, &p->cgroups->tasks); |
| 1922 | task_unlock(p); |
| 1923 | } while_each_thread(g, p); |
| 1924 | write_unlock(&css_set_lock); |
| 1925 | } |
| 1926 | |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 1927 | void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it) |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 1928 | { |
| 1929 | /* |
| 1930 | * The first time anyone tries to iterate across a cgroup, |
| 1931 | * we need to enable the list linking each css_set to its |
| 1932 | * tasks, and fix up all existing tasks. |
| 1933 | */ |
Cliff Wickman | 31a7df0 | 2008-02-07 00:14:42 -0800 | [diff] [blame] | 1934 | if (!use_task_css_set_links) |
| 1935 | cgroup_enable_task_cg_lists(); |
| 1936 | |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 1937 | read_lock(&css_set_lock); |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 1938 | it->cg_link = &cgrp->css_sets; |
| 1939 | cgroup_advance_iter(cgrp, it); |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 1940 | } |
| 1941 | |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 1942 | struct task_struct *cgroup_iter_next(struct cgroup *cgrp, |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 1943 | struct cgroup_iter *it) |
| 1944 | { |
| 1945 | struct task_struct *res; |
| 1946 | struct list_head *l = it->task; |
Lai Jiangshan | 2019f63 | 2009-01-07 18:07:36 -0800 | [diff] [blame] | 1947 | struct cg_cgroup_link *link; |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 1948 | |
| 1949 | /* If the iterator cg is NULL, we have no tasks */ |
| 1950 | if (!it->cg_link) |
| 1951 | return NULL; |
| 1952 | res = list_entry(l, struct task_struct, cg_list); |
| 1953 | /* Advance iterator to find next entry */ |
| 1954 | l = l->next; |
Lai Jiangshan | 2019f63 | 2009-01-07 18:07:36 -0800 | [diff] [blame] | 1955 | link = list_entry(it->cg_link, struct cg_cgroup_link, cgrp_link_list); |
| 1956 | if (l == &link->cg->tasks) { |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 1957 | /* We reached the end of this task list - move on to |
| 1958 | * the next cg_cgroup_link */ |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 1959 | cgroup_advance_iter(cgrp, it); |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 1960 | } else { |
| 1961 | it->task = l; |
| 1962 | } |
| 1963 | return res; |
| 1964 | } |
| 1965 | |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 1966 | void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it) |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 1967 | { |
| 1968 | read_unlock(&css_set_lock); |
| 1969 | } |
| 1970 | |
Cliff Wickman | 31a7df0 | 2008-02-07 00:14:42 -0800 | [diff] [blame] | 1971 | static inline int started_after_time(struct task_struct *t1, |
| 1972 | struct timespec *time, |
| 1973 | struct task_struct *t2) |
| 1974 | { |
| 1975 | int start_diff = timespec_compare(&t1->start_time, time); |
| 1976 | if (start_diff > 0) { |
| 1977 | return 1; |
| 1978 | } else if (start_diff < 0) { |
| 1979 | return 0; |
| 1980 | } else { |
| 1981 | /* |
| 1982 | * Arbitrarily, if two processes started at the same |
| 1983 | * time, we'll say that the lower pointer value |
| 1984 | * started first. Note that t2 may have exited by now |
| 1985 | * so this may not be a valid pointer any longer, but |
| 1986 | * that's fine - it still serves to distinguish |
| 1987 | * between two tasks started (effectively) simultaneously. |
| 1988 | */ |
| 1989 | return t1 > t2; |
| 1990 | } |
| 1991 | } |
| 1992 | |
| 1993 | /* |
| 1994 | * This function is a callback from heap_insert() and is used to order |
| 1995 | * the heap. |
| 1996 | * In this case we order the heap in descending task start time. |
| 1997 | */ |
| 1998 | static inline int started_after(void *p1, void *p2) |
| 1999 | { |
| 2000 | struct task_struct *t1 = p1; |
| 2001 | struct task_struct *t2 = p2; |
| 2002 | return started_after_time(t1, &t2->start_time, t2); |
| 2003 | } |
| 2004 | |
| 2005 | /** |
| 2006 | * cgroup_scan_tasks - iterate though all the tasks in a cgroup |
| 2007 | * @scan: struct cgroup_scanner containing arguments for the scan |
| 2008 | * |
| 2009 | * Arguments include pointers to callback functions test_task() and |
| 2010 | * process_task(). |
| 2011 | * Iterate through all the tasks in a cgroup, calling test_task() for each, |
| 2012 | * and if it returns true, call process_task() for it also. |
| 2013 | * The test_task pointer may be NULL, meaning always true (select all tasks). |
| 2014 | * Effectively duplicates cgroup_iter_{start,next,end}() |
| 2015 | * but does not lock css_set_lock for the call to process_task(). |
| 2016 | * The struct cgroup_scanner may be embedded in any structure of the caller's |
| 2017 | * creation. |
| 2018 | * It is guaranteed that process_task() will act on every task that |
| 2019 | * is a member of the cgroup for the duration of this call. This |
| 2020 | * function may or may not call process_task() for tasks that exit |
| 2021 | * or move to a different cgroup during the call, or are forked or |
| 2022 | * move into the cgroup during the call. |
| 2023 | * |
| 2024 | * Note that test_task() may be called with locks held, and may in some |
| 2025 | * situations be called multiple times for the same task, so it should |
| 2026 | * be cheap. |
| 2027 | * If the heap pointer in the struct cgroup_scanner is non-NULL, a heap has been |
| 2028 | * pre-allocated and will be used for heap operations (and its "gt" member will |
| 2029 | * be overwritten), else a temporary heap will be used (allocation of which |
| 2030 | * may cause this function to fail). |
| 2031 | */ |
| 2032 | int cgroup_scan_tasks(struct cgroup_scanner *scan) |
| 2033 | { |
| 2034 | int retval, i; |
| 2035 | struct cgroup_iter it; |
| 2036 | struct task_struct *p, *dropped; |
| 2037 | /* Never dereference latest_task, since it's not refcounted */ |
| 2038 | struct task_struct *latest_task = NULL; |
| 2039 | struct ptr_heap tmp_heap; |
| 2040 | struct ptr_heap *heap; |
| 2041 | struct timespec latest_time = { 0, 0 }; |
| 2042 | |
| 2043 | if (scan->heap) { |
| 2044 | /* The caller supplied our heap and pre-allocated its memory */ |
| 2045 | heap = scan->heap; |
| 2046 | heap->gt = &started_after; |
| 2047 | } else { |
| 2048 | /* We need to allocate our own heap memory */ |
| 2049 | heap = &tmp_heap; |
| 2050 | retval = heap_init(heap, PAGE_SIZE, GFP_KERNEL, &started_after); |
| 2051 | if (retval) |
| 2052 | /* cannot allocate the heap */ |
| 2053 | return retval; |
| 2054 | } |
| 2055 | |
| 2056 | again: |
| 2057 | /* |
| 2058 | * Scan tasks in the cgroup, using the scanner's "test_task" callback |
| 2059 | * to determine which are of interest, and using the scanner's |
| 2060 | * "process_task" callback to process any of them that need an update. |
| 2061 | * Since we don't want to hold any locks during the task updates, |
| 2062 | * gather tasks to be processed in a heap structure. |
| 2063 | * The heap is sorted by descending task start time. |
| 2064 | * If the statically-sized heap fills up, we overflow tasks that |
| 2065 | * started later, and in future iterations only consider tasks that |
| 2066 | * started after the latest task in the previous pass. This |
| 2067 | * guarantees forward progress and that we don't miss any tasks. |
| 2068 | */ |
| 2069 | heap->size = 0; |
| 2070 | cgroup_iter_start(scan->cg, &it); |
| 2071 | while ((p = cgroup_iter_next(scan->cg, &it))) { |
| 2072 | /* |
| 2073 | * Only affect tasks that qualify per the caller's callback, |
| 2074 | * if he provided one |
| 2075 | */ |
| 2076 | if (scan->test_task && !scan->test_task(p, scan)) |
| 2077 | continue; |
| 2078 | /* |
| 2079 | * Only process tasks that started after the last task |
| 2080 | * we processed |
| 2081 | */ |
| 2082 | if (!started_after_time(p, &latest_time, latest_task)) |
| 2083 | continue; |
| 2084 | dropped = heap_insert(heap, p); |
| 2085 | if (dropped == NULL) { |
| 2086 | /* |
| 2087 | * The new task was inserted; the heap wasn't |
| 2088 | * previously full |
| 2089 | */ |
| 2090 | get_task_struct(p); |
| 2091 | } else if (dropped != p) { |
| 2092 | /* |
| 2093 | * The new task was inserted, and pushed out a |
| 2094 | * different task |
| 2095 | */ |
| 2096 | get_task_struct(p); |
| 2097 | put_task_struct(dropped); |
| 2098 | } |
| 2099 | /* |
| 2100 | * Else the new task was newer than anything already in |
| 2101 | * the heap and wasn't inserted |
| 2102 | */ |
| 2103 | } |
| 2104 | cgroup_iter_end(scan->cg, &it); |
| 2105 | |
| 2106 | if (heap->size) { |
| 2107 | for (i = 0; i < heap->size; i++) { |
Paul Jackson | 4fe91d5 | 2008-04-29 00:59:55 -0700 | [diff] [blame] | 2108 | struct task_struct *q = heap->ptrs[i]; |
Cliff Wickman | 31a7df0 | 2008-02-07 00:14:42 -0800 | [diff] [blame] | 2109 | if (i == 0) { |
Paul Jackson | 4fe91d5 | 2008-04-29 00:59:55 -0700 | [diff] [blame] | 2110 | latest_time = q->start_time; |
| 2111 | latest_task = q; |
Cliff Wickman | 31a7df0 | 2008-02-07 00:14:42 -0800 | [diff] [blame] | 2112 | } |
| 2113 | /* Process the task per the caller's callback */ |
Paul Jackson | 4fe91d5 | 2008-04-29 00:59:55 -0700 | [diff] [blame] | 2114 | scan->process_task(q, scan); |
| 2115 | put_task_struct(q); |
Cliff Wickman | 31a7df0 | 2008-02-07 00:14:42 -0800 | [diff] [blame] | 2116 | } |
| 2117 | /* |
| 2118 | * If we had to process any tasks at all, scan again |
| 2119 | * in case some of them were in the middle of forking |
| 2120 | * children that didn't get processed. |
| 2121 | * Not the most efficient way to do it, but it avoids |
| 2122 | * having to take callback_mutex in the fork path |
| 2123 | */ |
| 2124 | goto again; |
| 2125 | } |
| 2126 | if (heap == &tmp_heap) |
| 2127 | heap_free(&tmp_heap); |
| 2128 | return 0; |
| 2129 | } |
| 2130 | |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 2131 | /* |
Paul Menage | bbcb81d | 2007-10-18 23:39:32 -0700 | [diff] [blame] | 2132 | * Stuff for reading the 'tasks' file. |
| 2133 | * |
| 2134 | * Reading this file can return large amounts of data if a cgroup has |
| 2135 | * *lots* of attached tasks. So it may need several calls to read(), |
| 2136 | * but we cannot guarantee that the information we produce is correct |
| 2137 | * unless we produce it entirely atomically. |
| 2138 | * |
Paul Menage | bbcb81d | 2007-10-18 23:39:32 -0700 | [diff] [blame] | 2139 | */ |
Paul Menage | bbcb81d | 2007-10-18 23:39:32 -0700 | [diff] [blame] | 2140 | |
| 2141 | /* |
| 2142 | * Load into 'pidarray' up to 'npids' of the tasks using cgroup |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 2143 | * 'cgrp'. Return actual number of pids loaded. No need to |
Paul Menage | bbcb81d | 2007-10-18 23:39:32 -0700 | [diff] [blame] | 2144 | * task_lock(p) when reading out p->cgroup, since we're in an RCU |
| 2145 | * read section, so the css_set can't go away, and is |
| 2146 | * immutable after creation. |
| 2147 | */ |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 2148 | static int pid_array_load(pid_t *pidarray, int npids, struct cgroup *cgrp) |
Paul Menage | bbcb81d | 2007-10-18 23:39:32 -0700 | [diff] [blame] | 2149 | { |
Gowrishankar M | e7b80bb | 2009-01-07 18:07:43 -0800 | [diff] [blame] | 2150 | int n = 0, pid; |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 2151 | struct cgroup_iter it; |
| 2152 | struct task_struct *tsk; |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 2153 | cgroup_iter_start(cgrp, &it); |
| 2154 | while ((tsk = cgroup_iter_next(cgrp, &it))) { |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 2155 | if (unlikely(n == npids)) |
| 2156 | break; |
Gowrishankar M | e7b80bb | 2009-01-07 18:07:43 -0800 | [diff] [blame] | 2157 | pid = task_pid_vnr(tsk); |
| 2158 | if (pid > 0) |
| 2159 | pidarray[n++] = pid; |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 2160 | } |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 2161 | cgroup_iter_end(cgrp, &it); |
Paul Menage | bbcb81d | 2007-10-18 23:39:32 -0700 | [diff] [blame] | 2162 | return n; |
| 2163 | } |
| 2164 | |
Balbir Singh | 846c7bb | 2007-10-18 23:39:44 -0700 | [diff] [blame] | 2165 | /** |
Li Zefan | a043e3b | 2008-02-23 15:24:09 -0800 | [diff] [blame] | 2166 | * cgroupstats_build - build and fill cgroupstats |
Balbir Singh | 846c7bb | 2007-10-18 23:39:44 -0700 | [diff] [blame] | 2167 | * @stats: cgroupstats to fill information into |
| 2168 | * @dentry: A dentry entry belonging to the cgroup for which stats have |
| 2169 | * been requested. |
Li Zefan | a043e3b | 2008-02-23 15:24:09 -0800 | [diff] [blame] | 2170 | * |
| 2171 | * Build and fill cgroupstats so that taskstats can export it to user |
| 2172 | * space. |
Balbir Singh | 846c7bb | 2007-10-18 23:39:44 -0700 | [diff] [blame] | 2173 | */ |
| 2174 | int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry) |
| 2175 | { |
| 2176 | int ret = -EINVAL; |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 2177 | struct cgroup *cgrp; |
Balbir Singh | 846c7bb | 2007-10-18 23:39:44 -0700 | [diff] [blame] | 2178 | struct cgroup_iter it; |
| 2179 | struct task_struct *tsk; |
Li Zefan | 33d283b | 2008-11-19 15:36:48 -0800 | [diff] [blame] | 2180 | |
Balbir Singh | 846c7bb | 2007-10-18 23:39:44 -0700 | [diff] [blame] | 2181 | /* |
Li Zefan | 33d283b | 2008-11-19 15:36:48 -0800 | [diff] [blame] | 2182 | * Validate dentry by checking the superblock operations, |
| 2183 | * and make sure it's a directory. |
Balbir Singh | 846c7bb | 2007-10-18 23:39:44 -0700 | [diff] [blame] | 2184 | */ |
Li Zefan | 33d283b | 2008-11-19 15:36:48 -0800 | [diff] [blame] | 2185 | if (dentry->d_sb->s_op != &cgroup_ops || |
| 2186 | !S_ISDIR(dentry->d_inode->i_mode)) |
Balbir Singh | 846c7bb | 2007-10-18 23:39:44 -0700 | [diff] [blame] | 2187 | goto err; |
| 2188 | |
| 2189 | ret = 0; |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 2190 | cgrp = dentry->d_fsdata; |
Balbir Singh | 846c7bb | 2007-10-18 23:39:44 -0700 | [diff] [blame] | 2191 | |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 2192 | cgroup_iter_start(cgrp, &it); |
| 2193 | while ((tsk = cgroup_iter_next(cgrp, &it))) { |
Balbir Singh | 846c7bb | 2007-10-18 23:39:44 -0700 | [diff] [blame] | 2194 | switch (tsk->state) { |
| 2195 | case TASK_RUNNING: |
| 2196 | stats->nr_running++; |
| 2197 | break; |
| 2198 | case TASK_INTERRUPTIBLE: |
| 2199 | stats->nr_sleeping++; |
| 2200 | break; |
| 2201 | case TASK_UNINTERRUPTIBLE: |
| 2202 | stats->nr_uninterruptible++; |
| 2203 | break; |
| 2204 | case TASK_STOPPED: |
| 2205 | stats->nr_stopped++; |
| 2206 | break; |
| 2207 | default: |
| 2208 | if (delayacct_is_task_waiting_on_io(tsk)) |
| 2209 | stats->nr_io_wait++; |
| 2210 | break; |
| 2211 | } |
| 2212 | } |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 2213 | cgroup_iter_end(cgrp, &it); |
Balbir Singh | 846c7bb | 2007-10-18 23:39:44 -0700 | [diff] [blame] | 2214 | |
Balbir Singh | 846c7bb | 2007-10-18 23:39:44 -0700 | [diff] [blame] | 2215 | err: |
| 2216 | return ret; |
| 2217 | } |
| 2218 | |
Li Zefan | 096b7fe | 2009-07-29 15:04:04 -0700 | [diff] [blame] | 2219 | /* |
| 2220 | * Cache pids for all threads in the same pid namespace that are |
| 2221 | * opening the same "tasks" file. |
| 2222 | */ |
| 2223 | struct cgroup_pids { |
| 2224 | /* The node in cgrp->pids_list */ |
| 2225 | struct list_head list; |
| 2226 | /* The cgroup those pids belong to */ |
| 2227 | struct cgroup *cgrp; |
| 2228 | /* The namepsace those pids belong to */ |
| 2229 | struct pid_namespace *ns; |
| 2230 | /* Array of process ids in the cgroup */ |
| 2231 | pid_t *tasks_pids; |
| 2232 | /* How many files are using the this tasks_pids array */ |
| 2233 | int use_count; |
| 2234 | /* Length of the current tasks_pids array */ |
| 2235 | int length; |
| 2236 | }; |
| 2237 | |
Paul Menage | bbcb81d | 2007-10-18 23:39:32 -0700 | [diff] [blame] | 2238 | static int cmppid(const void *a, const void *b) |
| 2239 | { |
| 2240 | return *(pid_t *)a - *(pid_t *)b; |
| 2241 | } |
| 2242 | |
Paul Menage | cc31edc | 2008-10-18 20:28:04 -0700 | [diff] [blame] | 2243 | /* |
| 2244 | * seq_file methods for the "tasks" file. The seq_file position is the |
| 2245 | * next pid to display; the seq_file iterator is a pointer to the pid |
| 2246 | * in the cgroup->tasks_pids array. |
| 2247 | */ |
| 2248 | |
| 2249 | static void *cgroup_tasks_start(struct seq_file *s, loff_t *pos) |
| 2250 | { |
| 2251 | /* |
| 2252 | * Initially we receive a position value that corresponds to |
| 2253 | * one more than the last pid shown (or 0 on the first call or |
| 2254 | * after a seek to the start). Use a binary-search to find the |
| 2255 | * next pid to display, if any |
| 2256 | */ |
Li Zefan | 096b7fe | 2009-07-29 15:04:04 -0700 | [diff] [blame] | 2257 | struct cgroup_pids *cp = s->private; |
| 2258 | struct cgroup *cgrp = cp->cgrp; |
Paul Menage | cc31edc | 2008-10-18 20:28:04 -0700 | [diff] [blame] | 2259 | int index = 0, pid = *pos; |
| 2260 | int *iter; |
| 2261 | |
| 2262 | down_read(&cgrp->pids_mutex); |
| 2263 | if (pid) { |
Li Zefan | 096b7fe | 2009-07-29 15:04:04 -0700 | [diff] [blame] | 2264 | int end = cp->length; |
Stephen Rothwell | 2077776 | 2008-10-21 16:11:20 +1100 | [diff] [blame] | 2265 | |
Paul Menage | cc31edc | 2008-10-18 20:28:04 -0700 | [diff] [blame] | 2266 | while (index < end) { |
| 2267 | int mid = (index + end) / 2; |
Li Zefan | 096b7fe | 2009-07-29 15:04:04 -0700 | [diff] [blame] | 2268 | if (cp->tasks_pids[mid] == pid) { |
Paul Menage | cc31edc | 2008-10-18 20:28:04 -0700 | [diff] [blame] | 2269 | index = mid; |
| 2270 | break; |
Li Zefan | 096b7fe | 2009-07-29 15:04:04 -0700 | [diff] [blame] | 2271 | } else if (cp->tasks_pids[mid] <= pid) |
Paul Menage | cc31edc | 2008-10-18 20:28:04 -0700 | [diff] [blame] | 2272 | index = mid + 1; |
| 2273 | else |
| 2274 | end = mid; |
| 2275 | } |
| 2276 | } |
| 2277 | /* If we're off the end of the array, we're done */ |
Li Zefan | 096b7fe | 2009-07-29 15:04:04 -0700 | [diff] [blame] | 2278 | if (index >= cp->length) |
Paul Menage | cc31edc | 2008-10-18 20:28:04 -0700 | [diff] [blame] | 2279 | return NULL; |
| 2280 | /* Update the abstract position to be the actual pid that we found */ |
Li Zefan | 096b7fe | 2009-07-29 15:04:04 -0700 | [diff] [blame] | 2281 | iter = cp->tasks_pids + index; |
Paul Menage | cc31edc | 2008-10-18 20:28:04 -0700 | [diff] [blame] | 2282 | *pos = *iter; |
| 2283 | return iter; |
Paul Menage | bbcb81d | 2007-10-18 23:39:32 -0700 | [diff] [blame] | 2284 | } |
| 2285 | |
Paul Menage | cc31edc | 2008-10-18 20:28:04 -0700 | [diff] [blame] | 2286 | static void cgroup_tasks_stop(struct seq_file *s, void *v) |
| 2287 | { |
Li Zefan | 096b7fe | 2009-07-29 15:04:04 -0700 | [diff] [blame] | 2288 | struct cgroup_pids *cp = s->private; |
| 2289 | struct cgroup *cgrp = cp->cgrp; |
Paul Menage | cc31edc | 2008-10-18 20:28:04 -0700 | [diff] [blame] | 2290 | up_read(&cgrp->pids_mutex); |
| 2291 | } |
| 2292 | |
| 2293 | static void *cgroup_tasks_next(struct seq_file *s, void *v, loff_t *pos) |
| 2294 | { |
Li Zefan | 096b7fe | 2009-07-29 15:04:04 -0700 | [diff] [blame] | 2295 | struct cgroup_pids *cp = s->private; |
Paul Menage | cc31edc | 2008-10-18 20:28:04 -0700 | [diff] [blame] | 2296 | int *p = v; |
Li Zefan | 096b7fe | 2009-07-29 15:04:04 -0700 | [diff] [blame] | 2297 | int *end = cp->tasks_pids + cp->length; |
Paul Menage | cc31edc | 2008-10-18 20:28:04 -0700 | [diff] [blame] | 2298 | |
| 2299 | /* |
| 2300 | * Advance to the next pid in the array. If this goes off the |
| 2301 | * end, we're done |
| 2302 | */ |
| 2303 | p++; |
| 2304 | if (p >= end) { |
| 2305 | return NULL; |
| 2306 | } else { |
| 2307 | *pos = *p; |
| 2308 | return p; |
| 2309 | } |
| 2310 | } |
| 2311 | |
| 2312 | static int cgroup_tasks_show(struct seq_file *s, void *v) |
| 2313 | { |
| 2314 | return seq_printf(s, "%d\n", *(int *)v); |
| 2315 | } |
| 2316 | |
| 2317 | static struct seq_operations cgroup_tasks_seq_operations = { |
| 2318 | .start = cgroup_tasks_start, |
| 2319 | .stop = cgroup_tasks_stop, |
| 2320 | .next = cgroup_tasks_next, |
| 2321 | .show = cgroup_tasks_show, |
| 2322 | }; |
| 2323 | |
Li Zefan | 096b7fe | 2009-07-29 15:04:04 -0700 | [diff] [blame] | 2324 | static void release_cgroup_pid_array(struct cgroup_pids *cp) |
Paul Menage | cc31edc | 2008-10-18 20:28:04 -0700 | [diff] [blame] | 2325 | { |
Li Zefan | 096b7fe | 2009-07-29 15:04:04 -0700 | [diff] [blame] | 2326 | struct cgroup *cgrp = cp->cgrp; |
| 2327 | |
Paul Menage | cc31edc | 2008-10-18 20:28:04 -0700 | [diff] [blame] | 2328 | down_write(&cgrp->pids_mutex); |
Li Zefan | 096b7fe | 2009-07-29 15:04:04 -0700 | [diff] [blame] | 2329 | BUG_ON(!cp->use_count); |
| 2330 | if (!--cp->use_count) { |
| 2331 | list_del(&cp->list); |
| 2332 | put_pid_ns(cp->ns); |
| 2333 | kfree(cp->tasks_pids); |
| 2334 | kfree(cp); |
Paul Menage | cc31edc | 2008-10-18 20:28:04 -0700 | [diff] [blame] | 2335 | } |
| 2336 | up_write(&cgrp->pids_mutex); |
| 2337 | } |
| 2338 | |
| 2339 | static int cgroup_tasks_release(struct inode *inode, struct file *file) |
Paul Menage | bbcb81d | 2007-10-18 23:39:32 -0700 | [diff] [blame] | 2340 | { |
Li Zefan | 096b7fe | 2009-07-29 15:04:04 -0700 | [diff] [blame] | 2341 | struct seq_file *seq; |
| 2342 | struct cgroup_pids *cp; |
Paul Menage | bbcb81d | 2007-10-18 23:39:32 -0700 | [diff] [blame] | 2343 | |
| 2344 | if (!(file->f_mode & FMODE_READ)) |
| 2345 | return 0; |
| 2346 | |
Li Zefan | 096b7fe | 2009-07-29 15:04:04 -0700 | [diff] [blame] | 2347 | seq = file->private_data; |
| 2348 | cp = seq->private; |
| 2349 | |
| 2350 | release_cgroup_pid_array(cp); |
Paul Menage | cc31edc | 2008-10-18 20:28:04 -0700 | [diff] [blame] | 2351 | return seq_release(inode, file); |
| 2352 | } |
| 2353 | |
| 2354 | static struct file_operations cgroup_tasks_operations = { |
| 2355 | .read = seq_read, |
| 2356 | .llseek = seq_lseek, |
| 2357 | .write = cgroup_file_write, |
| 2358 | .release = cgroup_tasks_release, |
| 2359 | }; |
| 2360 | |
| 2361 | /* |
| 2362 | * Handle an open on 'tasks' file. Prepare an array containing the |
| 2363 | * process id's of tasks currently attached to the cgroup being opened. |
| 2364 | */ |
| 2365 | |
| 2366 | static int cgroup_tasks_open(struct inode *unused, struct file *file) |
| 2367 | { |
| 2368 | struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent); |
Li Zefan | 096b7fe | 2009-07-29 15:04:04 -0700 | [diff] [blame] | 2369 | struct pid_namespace *ns = current->nsproxy->pid_ns; |
| 2370 | struct cgroup_pids *cp; |
Paul Menage | cc31edc | 2008-10-18 20:28:04 -0700 | [diff] [blame] | 2371 | pid_t *pidarray; |
| 2372 | int npids; |
| 2373 | int retval; |
| 2374 | |
| 2375 | /* Nothing to do for write-only files */ |
| 2376 | if (!(file->f_mode & FMODE_READ)) |
| 2377 | return 0; |
Paul Menage | bbcb81d | 2007-10-18 23:39:32 -0700 | [diff] [blame] | 2378 | |
| 2379 | /* |
| 2380 | * If cgroup gets more users after we read count, we won't have |
| 2381 | * enough space - tough. This race is indistinguishable to the |
| 2382 | * caller from the case that the additional cgroup users didn't |
| 2383 | * show up until sometime later on. |
| 2384 | */ |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 2385 | npids = cgroup_task_count(cgrp); |
Paul Menage | cc31edc | 2008-10-18 20:28:04 -0700 | [diff] [blame] | 2386 | pidarray = kmalloc(npids * sizeof(pid_t), GFP_KERNEL); |
| 2387 | if (!pidarray) |
| 2388 | return -ENOMEM; |
| 2389 | npids = pid_array_load(pidarray, npids, cgrp); |
| 2390 | sort(pidarray, npids, sizeof(pid_t), cmppid, NULL); |
Paul Menage | bbcb81d | 2007-10-18 23:39:32 -0700 | [diff] [blame] | 2391 | |
Paul Menage | cc31edc | 2008-10-18 20:28:04 -0700 | [diff] [blame] | 2392 | /* |
| 2393 | * Store the array in the cgroup, freeing the old |
| 2394 | * array if necessary |
| 2395 | */ |
| 2396 | down_write(&cgrp->pids_mutex); |
Li Zefan | 096b7fe | 2009-07-29 15:04:04 -0700 | [diff] [blame] | 2397 | |
| 2398 | list_for_each_entry(cp, &cgrp->pids_list, list) { |
| 2399 | if (ns == cp->ns) |
| 2400 | goto found; |
| 2401 | } |
| 2402 | |
| 2403 | cp = kzalloc(sizeof(*cp), GFP_KERNEL); |
| 2404 | if (!cp) { |
| 2405 | up_write(&cgrp->pids_mutex); |
| 2406 | kfree(pidarray); |
| 2407 | return -ENOMEM; |
| 2408 | } |
| 2409 | cp->cgrp = cgrp; |
| 2410 | cp->ns = ns; |
| 2411 | get_pid_ns(ns); |
| 2412 | list_add(&cp->list, &cgrp->pids_list); |
| 2413 | found: |
| 2414 | kfree(cp->tasks_pids); |
| 2415 | cp->tasks_pids = pidarray; |
| 2416 | cp->length = npids; |
| 2417 | cp->use_count++; |
Paul Menage | cc31edc | 2008-10-18 20:28:04 -0700 | [diff] [blame] | 2418 | up_write(&cgrp->pids_mutex); |
Paul Menage | bbcb81d | 2007-10-18 23:39:32 -0700 | [diff] [blame] | 2419 | |
Paul Menage | cc31edc | 2008-10-18 20:28:04 -0700 | [diff] [blame] | 2420 | file->f_op = &cgroup_tasks_operations; |
Paul Menage | bbcb81d | 2007-10-18 23:39:32 -0700 | [diff] [blame] | 2421 | |
Paul Menage | cc31edc | 2008-10-18 20:28:04 -0700 | [diff] [blame] | 2422 | retval = seq_open(file, &cgroup_tasks_seq_operations); |
| 2423 | if (retval) { |
Li Zefan | 096b7fe | 2009-07-29 15:04:04 -0700 | [diff] [blame] | 2424 | release_cgroup_pid_array(cp); |
Paul Menage | cc31edc | 2008-10-18 20:28:04 -0700 | [diff] [blame] | 2425 | return retval; |
Paul Menage | bbcb81d | 2007-10-18 23:39:32 -0700 | [diff] [blame] | 2426 | } |
Li Zefan | 096b7fe | 2009-07-29 15:04:04 -0700 | [diff] [blame] | 2427 | ((struct seq_file *)file->private_data)->private = cp; |
Paul Menage | bbcb81d | 2007-10-18 23:39:32 -0700 | [diff] [blame] | 2428 | return 0; |
| 2429 | } |
| 2430 | |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 2431 | static u64 cgroup_read_notify_on_release(struct cgroup *cgrp, |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 2432 | struct cftype *cft) |
| 2433 | { |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 2434 | return notify_on_release(cgrp); |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 2435 | } |
| 2436 | |
Paul Menage | 6379c10 | 2008-07-25 01:47:01 -0700 | [diff] [blame] | 2437 | static int cgroup_write_notify_on_release(struct cgroup *cgrp, |
| 2438 | struct cftype *cft, |
| 2439 | u64 val) |
| 2440 | { |
| 2441 | clear_bit(CGRP_RELEASABLE, &cgrp->flags); |
| 2442 | if (val) |
| 2443 | set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); |
| 2444 | else |
| 2445 | clear_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); |
| 2446 | return 0; |
| 2447 | } |
| 2448 | |
Paul Menage | bbcb81d | 2007-10-18 23:39:32 -0700 | [diff] [blame] | 2449 | /* |
| 2450 | * for the common functions, 'private' gives the type of file |
| 2451 | */ |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 2452 | static struct cftype files[] = { |
| 2453 | { |
| 2454 | .name = "tasks", |
| 2455 | .open = cgroup_tasks_open, |
Paul Menage | af35102 | 2008-07-25 01:47:01 -0700 | [diff] [blame] | 2456 | .write_u64 = cgroup_tasks_write, |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 2457 | .release = cgroup_tasks_release, |
| 2458 | .private = FILE_TASKLIST, |
Li Zefan | 099fca3 | 2009-04-02 16:57:29 -0700 | [diff] [blame] | 2459 | .mode = S_IRUGO | S_IWUSR, |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 2460 | }, |
| 2461 | |
| 2462 | { |
| 2463 | .name = "notify_on_release", |
Paul Menage | f4c753b | 2008-04-29 00:59:56 -0700 | [diff] [blame] | 2464 | .read_u64 = cgroup_read_notify_on_release, |
Paul Menage | 6379c10 | 2008-07-25 01:47:01 -0700 | [diff] [blame] | 2465 | .write_u64 = cgroup_write_notify_on_release, |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 2466 | .private = FILE_NOTIFY_ON_RELEASE, |
| 2467 | }, |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 2468 | }; |
| 2469 | |
| 2470 | static struct cftype cft_release_agent = { |
| 2471 | .name = "release_agent", |
Paul Menage | e788e06 | 2008-07-25 01:46:59 -0700 | [diff] [blame] | 2472 | .read_seq_string = cgroup_release_agent_show, |
| 2473 | .write_string = cgroup_release_agent_write, |
| 2474 | .max_write_len = PATH_MAX, |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 2475 | .private = FILE_RELEASE_AGENT, |
Paul Menage | bbcb81d | 2007-10-18 23:39:32 -0700 | [diff] [blame] | 2476 | }; |
| 2477 | |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 2478 | static int cgroup_populate_dir(struct cgroup *cgrp) |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 2479 | { |
| 2480 | int err; |
| 2481 | struct cgroup_subsys *ss; |
| 2482 | |
| 2483 | /* First clear out any existing files */ |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 2484 | cgroup_clear_directory(cgrp->dentry); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 2485 | |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 2486 | err = cgroup_add_files(cgrp, NULL, files, ARRAY_SIZE(files)); |
Paul Menage | bbcb81d | 2007-10-18 23:39:32 -0700 | [diff] [blame] | 2487 | if (err < 0) |
| 2488 | return err; |
| 2489 | |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 2490 | if (cgrp == cgrp->top_cgroup) { |
| 2491 | if ((err = cgroup_add_file(cgrp, NULL, &cft_release_agent)) < 0) |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 2492 | return err; |
| 2493 | } |
| 2494 | |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 2495 | for_each_subsys(cgrp->root, ss) { |
| 2496 | if (ss->populate && (err = ss->populate(ss, cgrp)) < 0) |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 2497 | return err; |
| 2498 | } |
KAMEZAWA Hiroyuki | 38460b4 | 2009-04-02 16:57:25 -0700 | [diff] [blame] | 2499 | /* This cgroup is ready now */ |
| 2500 | for_each_subsys(cgrp->root, ss) { |
| 2501 | struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id]; |
| 2502 | /* |
| 2503 | * Update id->css pointer and make this css visible from |
| 2504 | * CSS ID functions. This pointer will be dereferened |
| 2505 | * from RCU-read-side without locks. |
| 2506 | */ |
| 2507 | if (css->id) |
| 2508 | rcu_assign_pointer(css->id->css, css); |
| 2509 | } |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 2510 | |
| 2511 | return 0; |
| 2512 | } |
| 2513 | |
| 2514 | static void init_cgroup_css(struct cgroup_subsys_state *css, |
| 2515 | struct cgroup_subsys *ss, |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 2516 | struct cgroup *cgrp) |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 2517 | { |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 2518 | css->cgroup = cgrp; |
Paul Menage | e7c5ec9 | 2009-01-07 18:08:38 -0800 | [diff] [blame] | 2519 | atomic_set(&css->refcnt, 1); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 2520 | css->flags = 0; |
KAMEZAWA Hiroyuki | 38460b4 | 2009-04-02 16:57:25 -0700 | [diff] [blame] | 2521 | css->id = NULL; |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 2522 | if (cgrp == dummytop) |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 2523 | set_bit(CSS_ROOT, &css->flags); |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 2524 | BUG_ON(cgrp->subsys[ss->subsys_id]); |
| 2525 | cgrp->subsys[ss->subsys_id] = css; |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 2526 | } |
| 2527 | |
Paul Menage | 999cd8a | 2009-01-07 18:08:36 -0800 | [diff] [blame] | 2528 | static void cgroup_lock_hierarchy(struct cgroupfs_root *root) |
| 2529 | { |
| 2530 | /* We need to take each hierarchy_mutex in a consistent order */ |
| 2531 | int i; |
| 2532 | |
| 2533 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { |
| 2534 | struct cgroup_subsys *ss = subsys[i]; |
| 2535 | if (ss->root == root) |
Li Zefan | cfebe56 | 2009-02-11 13:04:36 -0800 | [diff] [blame] | 2536 | mutex_lock(&ss->hierarchy_mutex); |
Paul Menage | 999cd8a | 2009-01-07 18:08:36 -0800 | [diff] [blame] | 2537 | } |
| 2538 | } |
| 2539 | |
| 2540 | static void cgroup_unlock_hierarchy(struct cgroupfs_root *root) |
| 2541 | { |
| 2542 | int i; |
| 2543 | |
| 2544 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { |
| 2545 | struct cgroup_subsys *ss = subsys[i]; |
| 2546 | if (ss->root == root) |
| 2547 | mutex_unlock(&ss->hierarchy_mutex); |
| 2548 | } |
| 2549 | } |
| 2550 | |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 2551 | /* |
Li Zefan | a043e3b | 2008-02-23 15:24:09 -0800 | [diff] [blame] | 2552 | * cgroup_create - create a cgroup |
| 2553 | * @parent: cgroup that will be parent of the new cgroup |
| 2554 | * @dentry: dentry of the new cgroup |
| 2555 | * @mode: mode to set on new inode |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 2556 | * |
Li Zefan | a043e3b | 2008-02-23 15:24:09 -0800 | [diff] [blame] | 2557 | * Must be called with the mutex on the parent inode held |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 2558 | */ |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 2559 | static long cgroup_create(struct cgroup *parent, struct dentry *dentry, |
Li Zefan | 099fca3 | 2009-04-02 16:57:29 -0700 | [diff] [blame] | 2560 | mode_t mode) |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 2561 | { |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 2562 | struct cgroup *cgrp; |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 2563 | struct cgroupfs_root *root = parent->root; |
| 2564 | int err = 0; |
| 2565 | struct cgroup_subsys *ss; |
| 2566 | struct super_block *sb = root->sb; |
| 2567 | |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 2568 | cgrp = kzalloc(sizeof(*cgrp), GFP_KERNEL); |
| 2569 | if (!cgrp) |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 2570 | return -ENOMEM; |
| 2571 | |
| 2572 | /* Grab a reference on the superblock so the hierarchy doesn't |
| 2573 | * get deleted on unmount if there are child cgroups. This |
| 2574 | * can be done outside cgroup_mutex, since the sb can't |
| 2575 | * disappear while someone has an open control file on the |
| 2576 | * fs */ |
| 2577 | atomic_inc(&sb->s_active); |
| 2578 | |
| 2579 | mutex_lock(&cgroup_mutex); |
| 2580 | |
Paul Menage | cc31edc | 2008-10-18 20:28:04 -0700 | [diff] [blame] | 2581 | init_cgroup_housekeeping(cgrp); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 2582 | |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 2583 | cgrp->parent = parent; |
| 2584 | cgrp->root = parent->root; |
| 2585 | cgrp->top_cgroup = parent->top_cgroup; |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 2586 | |
Li Zefan | b6abdb0 | 2008-03-04 14:28:19 -0800 | [diff] [blame] | 2587 | if (notify_on_release(parent)) |
| 2588 | set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); |
| 2589 | |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 2590 | for_each_subsys(root, ss) { |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 2591 | struct cgroup_subsys_state *css = ss->create(ss, cgrp); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 2592 | if (IS_ERR(css)) { |
| 2593 | err = PTR_ERR(css); |
| 2594 | goto err_destroy; |
| 2595 | } |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 2596 | init_cgroup_css(css, ss, cgrp); |
KAMEZAWA Hiroyuki | 38460b4 | 2009-04-02 16:57:25 -0700 | [diff] [blame] | 2597 | if (ss->use_id) |
| 2598 | if (alloc_css_id(ss, parent, cgrp)) |
| 2599 | goto err_destroy; |
| 2600 | /* At error, ->destroy() callback has to free assigned ID. */ |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 2601 | } |
| 2602 | |
Paul Menage | 999cd8a | 2009-01-07 18:08:36 -0800 | [diff] [blame] | 2603 | cgroup_lock_hierarchy(root); |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 2604 | list_add(&cgrp->sibling, &cgrp->parent->children); |
Paul Menage | 999cd8a | 2009-01-07 18:08:36 -0800 | [diff] [blame] | 2605 | cgroup_unlock_hierarchy(root); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 2606 | root->number_of_cgroups++; |
| 2607 | |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 2608 | err = cgroup_create_dir(cgrp, dentry, mode); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 2609 | if (err < 0) |
| 2610 | goto err_remove; |
| 2611 | |
| 2612 | /* The cgroup directory was pre-locked for us */ |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 2613 | BUG_ON(!mutex_is_locked(&cgrp->dentry->d_inode->i_mutex)); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 2614 | |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 2615 | err = cgroup_populate_dir(cgrp); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 2616 | /* If err < 0, we have a half-filled directory - oh well ;) */ |
| 2617 | |
| 2618 | mutex_unlock(&cgroup_mutex); |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 2619 | mutex_unlock(&cgrp->dentry->d_inode->i_mutex); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 2620 | |
| 2621 | return 0; |
| 2622 | |
| 2623 | err_remove: |
| 2624 | |
KAMEZAWA Hiroyuki | baef99a | 2009-01-29 14:25:10 -0800 | [diff] [blame] | 2625 | cgroup_lock_hierarchy(root); |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 2626 | list_del(&cgrp->sibling); |
KAMEZAWA Hiroyuki | baef99a | 2009-01-29 14:25:10 -0800 | [diff] [blame] | 2627 | cgroup_unlock_hierarchy(root); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 2628 | root->number_of_cgroups--; |
| 2629 | |
| 2630 | err_destroy: |
| 2631 | |
| 2632 | for_each_subsys(root, ss) { |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 2633 | if (cgrp->subsys[ss->subsys_id]) |
| 2634 | ss->destroy(ss, cgrp); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 2635 | } |
| 2636 | |
| 2637 | mutex_unlock(&cgroup_mutex); |
| 2638 | |
| 2639 | /* Release the reference count that we took on the superblock */ |
| 2640 | deactivate_super(sb); |
| 2641 | |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 2642 | kfree(cgrp); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 2643 | return err; |
| 2644 | } |
| 2645 | |
| 2646 | static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, int mode) |
| 2647 | { |
| 2648 | struct cgroup *c_parent = dentry->d_parent->d_fsdata; |
| 2649 | |
| 2650 | /* the vfs holds inode->i_mutex already */ |
| 2651 | return cgroup_create(c_parent, dentry, mode | S_IFDIR); |
| 2652 | } |
| 2653 | |
Li Zefan | 55b6fd0 | 2008-07-29 22:33:20 -0700 | [diff] [blame] | 2654 | static int cgroup_has_css_refs(struct cgroup *cgrp) |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 2655 | { |
| 2656 | /* Check the reference count on each subsystem. Since we |
| 2657 | * already established that there are no tasks in the |
Paul Menage | e7c5ec9 | 2009-01-07 18:08:38 -0800 | [diff] [blame] | 2658 | * cgroup, if the css refcount is also 1, then there should |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 2659 | * be no outstanding references, so the subsystem is safe to |
| 2660 | * destroy. We scan across all subsystems rather than using |
| 2661 | * the per-hierarchy linked list of mounted subsystems since |
| 2662 | * we can be called via check_for_release() with no |
| 2663 | * synchronization other than RCU, and the subsystem linked |
| 2664 | * list isn't RCU-safe */ |
| 2665 | int i; |
| 2666 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { |
| 2667 | struct cgroup_subsys *ss = subsys[i]; |
| 2668 | struct cgroup_subsys_state *css; |
| 2669 | /* Skip subsystems not in this hierarchy */ |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 2670 | if (ss->root != cgrp->root) |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 2671 | continue; |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 2672 | css = cgrp->subsys[ss->subsys_id]; |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 2673 | /* When called from check_for_release() it's possible |
| 2674 | * that by this point the cgroup has been removed |
| 2675 | * and the css deleted. But a false-positive doesn't |
| 2676 | * matter, since it can only happen if the cgroup |
| 2677 | * has been deleted and hence no longer needs the |
| 2678 | * release agent to be called anyway. */ |
Paul Menage | e7c5ec9 | 2009-01-07 18:08:38 -0800 | [diff] [blame] | 2679 | if (css && (atomic_read(&css->refcnt) > 1)) |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 2680 | return 1; |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 2681 | } |
| 2682 | return 0; |
| 2683 | } |
| 2684 | |
Paul Menage | e7c5ec9 | 2009-01-07 18:08:38 -0800 | [diff] [blame] | 2685 | /* |
| 2686 | * Atomically mark all (or else none) of the cgroup's CSS objects as |
| 2687 | * CSS_REMOVED. Return true on success, or false if the cgroup has |
| 2688 | * busy subsystems. Call with cgroup_mutex held |
| 2689 | */ |
| 2690 | |
| 2691 | static int cgroup_clear_css_refs(struct cgroup *cgrp) |
| 2692 | { |
| 2693 | struct cgroup_subsys *ss; |
| 2694 | unsigned long flags; |
| 2695 | bool failed = false; |
| 2696 | local_irq_save(flags); |
| 2697 | for_each_subsys(cgrp->root, ss) { |
| 2698 | struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id]; |
| 2699 | int refcnt; |
Paul Menage | 804b3c2 | 2009-01-29 14:25:21 -0800 | [diff] [blame] | 2700 | while (1) { |
Paul Menage | e7c5ec9 | 2009-01-07 18:08:38 -0800 | [diff] [blame] | 2701 | /* We can only remove a CSS with a refcnt==1 */ |
| 2702 | refcnt = atomic_read(&css->refcnt); |
| 2703 | if (refcnt > 1) { |
| 2704 | failed = true; |
| 2705 | goto done; |
| 2706 | } |
| 2707 | BUG_ON(!refcnt); |
| 2708 | /* |
| 2709 | * Drop the refcnt to 0 while we check other |
| 2710 | * subsystems. This will cause any racing |
| 2711 | * css_tryget() to spin until we set the |
| 2712 | * CSS_REMOVED bits or abort |
| 2713 | */ |
Paul Menage | 804b3c2 | 2009-01-29 14:25:21 -0800 | [diff] [blame] | 2714 | if (atomic_cmpxchg(&css->refcnt, refcnt, 0) == refcnt) |
| 2715 | break; |
| 2716 | cpu_relax(); |
| 2717 | } |
Paul Menage | e7c5ec9 | 2009-01-07 18:08:38 -0800 | [diff] [blame] | 2718 | } |
| 2719 | done: |
| 2720 | for_each_subsys(cgrp->root, ss) { |
| 2721 | struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id]; |
| 2722 | if (failed) { |
| 2723 | /* |
| 2724 | * Restore old refcnt if we previously managed |
| 2725 | * to clear it from 1 to 0 |
| 2726 | */ |
| 2727 | if (!atomic_read(&css->refcnt)) |
| 2728 | atomic_set(&css->refcnt, 1); |
| 2729 | } else { |
| 2730 | /* Commit the fact that the CSS is removed */ |
| 2731 | set_bit(CSS_REMOVED, &css->flags); |
| 2732 | } |
| 2733 | } |
| 2734 | local_irq_restore(flags); |
| 2735 | return !failed; |
| 2736 | } |
| 2737 | |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 2738 | static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry) |
| 2739 | { |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 2740 | struct cgroup *cgrp = dentry->d_fsdata; |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 2741 | struct dentry *d; |
| 2742 | struct cgroup *parent; |
KAMEZAWA Hiroyuki | ec64f51 | 2009-04-02 16:57:26 -0700 | [diff] [blame] | 2743 | DEFINE_WAIT(wait); |
| 2744 | int ret; |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 2745 | |
| 2746 | /* the vfs holds both inode->i_mutex already */ |
KAMEZAWA Hiroyuki | ec64f51 | 2009-04-02 16:57:26 -0700 | [diff] [blame] | 2747 | again: |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 2748 | mutex_lock(&cgroup_mutex); |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 2749 | if (atomic_read(&cgrp->count) != 0) { |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 2750 | mutex_unlock(&cgroup_mutex); |
| 2751 | return -EBUSY; |
| 2752 | } |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 2753 | if (!list_empty(&cgrp->children)) { |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 2754 | mutex_unlock(&cgroup_mutex); |
| 2755 | return -EBUSY; |
| 2756 | } |
KAMEZAWA Hiroyuki | 3fa59df | 2008-11-19 15:36:34 -0800 | [diff] [blame] | 2757 | mutex_unlock(&cgroup_mutex); |
Li Zefan | a043e3b | 2008-02-23 15:24:09 -0800 | [diff] [blame] | 2758 | |
KAMEZAWA Hiroyuki | 4fca88c | 2008-02-07 00:14:27 -0800 | [diff] [blame] | 2759 | /* |
KAMEZAWA Hiroyuki | 8870326 | 2009-07-29 15:04:06 -0700 | [diff] [blame] | 2760 | * In general, subsystem has no css->refcnt after pre_destroy(). But |
| 2761 | * in racy cases, subsystem may have to get css->refcnt after |
| 2762 | * pre_destroy() and it makes rmdir return with -EBUSY. This sometimes |
| 2763 | * make rmdir return -EBUSY too often. To avoid that, we use waitqueue |
| 2764 | * for cgroup's rmdir. CGRP_WAIT_ON_RMDIR is for synchronizing rmdir |
| 2765 | * and subsystem's reference count handling. Please see css_get/put |
| 2766 | * and css_tryget() and cgroup_wakeup_rmdir_waiter() implementation. |
| 2767 | */ |
| 2768 | set_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags); |
| 2769 | |
| 2770 | /* |
Li Zefan | a043e3b | 2008-02-23 15:24:09 -0800 | [diff] [blame] | 2771 | * Call pre_destroy handlers of subsys. Notify subsystems |
| 2772 | * that rmdir() request comes. |
KAMEZAWA Hiroyuki | 4fca88c | 2008-02-07 00:14:27 -0800 | [diff] [blame] | 2773 | */ |
KAMEZAWA Hiroyuki | ec64f51 | 2009-04-02 16:57:26 -0700 | [diff] [blame] | 2774 | ret = cgroup_call_pre_destroy(cgrp); |
KAMEZAWA Hiroyuki | 8870326 | 2009-07-29 15:04:06 -0700 | [diff] [blame] | 2775 | if (ret) { |
| 2776 | clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags); |
KAMEZAWA Hiroyuki | ec64f51 | 2009-04-02 16:57:26 -0700 | [diff] [blame] | 2777 | return ret; |
KAMEZAWA Hiroyuki | 8870326 | 2009-07-29 15:04:06 -0700 | [diff] [blame] | 2778 | } |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 2779 | |
KAMEZAWA Hiroyuki | 3fa59df | 2008-11-19 15:36:34 -0800 | [diff] [blame] | 2780 | mutex_lock(&cgroup_mutex); |
| 2781 | parent = cgrp->parent; |
KAMEZAWA Hiroyuki | ec64f51 | 2009-04-02 16:57:26 -0700 | [diff] [blame] | 2782 | if (atomic_read(&cgrp->count) || !list_empty(&cgrp->children)) { |
KAMEZAWA Hiroyuki | 8870326 | 2009-07-29 15:04:06 -0700 | [diff] [blame] | 2783 | clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 2784 | mutex_unlock(&cgroup_mutex); |
| 2785 | return -EBUSY; |
| 2786 | } |
KAMEZAWA Hiroyuki | ec64f51 | 2009-04-02 16:57:26 -0700 | [diff] [blame] | 2787 | prepare_to_wait(&cgroup_rmdir_waitq, &wait, TASK_INTERRUPTIBLE); |
KAMEZAWA Hiroyuki | ec64f51 | 2009-04-02 16:57:26 -0700 | [diff] [blame] | 2788 | if (!cgroup_clear_css_refs(cgrp)) { |
| 2789 | mutex_unlock(&cgroup_mutex); |
KAMEZAWA Hiroyuki | 8870326 | 2009-07-29 15:04:06 -0700 | [diff] [blame] | 2790 | /* |
| 2791 | * Because someone may call cgroup_wakeup_rmdir_waiter() before |
| 2792 | * prepare_to_wait(), we need to check this flag. |
| 2793 | */ |
| 2794 | if (test_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags)) |
| 2795 | schedule(); |
KAMEZAWA Hiroyuki | ec64f51 | 2009-04-02 16:57:26 -0700 | [diff] [blame] | 2796 | finish_wait(&cgroup_rmdir_waitq, &wait); |
| 2797 | clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags); |
| 2798 | if (signal_pending(current)) |
| 2799 | return -EINTR; |
| 2800 | goto again; |
| 2801 | } |
| 2802 | /* NO css_tryget() can success after here. */ |
| 2803 | finish_wait(&cgroup_rmdir_waitq, &wait); |
| 2804 | clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 2805 | |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 2806 | spin_lock(&release_list_lock); |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 2807 | set_bit(CGRP_REMOVED, &cgrp->flags); |
| 2808 | if (!list_empty(&cgrp->release_list)) |
| 2809 | list_del(&cgrp->release_list); |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 2810 | spin_unlock(&release_list_lock); |
Paul Menage | 999cd8a | 2009-01-07 18:08:36 -0800 | [diff] [blame] | 2811 | |
| 2812 | cgroup_lock_hierarchy(cgrp->root); |
| 2813 | /* delete this cgroup from parent->children */ |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 2814 | list_del(&cgrp->sibling); |
Paul Menage | 999cd8a | 2009-01-07 18:08:36 -0800 | [diff] [blame] | 2815 | cgroup_unlock_hierarchy(cgrp->root); |
| 2816 | |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 2817 | spin_lock(&cgrp->dentry->d_lock); |
| 2818 | d = dget(cgrp->dentry); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 2819 | spin_unlock(&d->d_lock); |
| 2820 | |
| 2821 | cgroup_d_remove_dir(d); |
| 2822 | dput(d); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 2823 | |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 2824 | set_bit(CGRP_RELEASABLE, &parent->flags); |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 2825 | check_for_release(parent); |
| 2826 | |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 2827 | mutex_unlock(&cgroup_mutex); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 2828 | return 0; |
| 2829 | } |
| 2830 | |
Li Zefan | 06a1192 | 2008-04-29 01:00:07 -0700 | [diff] [blame] | 2831 | static void __init cgroup_init_subsys(struct cgroup_subsys *ss) |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 2832 | { |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 2833 | struct cgroup_subsys_state *css; |
Diego Calleja | cfe36bd | 2007-11-14 16:58:54 -0800 | [diff] [blame] | 2834 | |
| 2835 | printk(KERN_INFO "Initializing cgroup subsys %s\n", ss->name); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 2836 | |
| 2837 | /* Create the top cgroup state for this subsystem */ |
Li Zefan | 33a68ac | 2009-01-07 18:07:42 -0800 | [diff] [blame] | 2838 | list_add(&ss->sibling, &rootnode.subsys_list); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 2839 | ss->root = &rootnode; |
| 2840 | css = ss->create(ss, dummytop); |
| 2841 | /* We don't handle early failures gracefully */ |
| 2842 | BUG_ON(IS_ERR(css)); |
| 2843 | init_cgroup_css(css, ss, dummytop); |
| 2844 | |
Li Zefan | e8d55fd | 2008-04-29 01:00:13 -0700 | [diff] [blame] | 2845 | /* Update the init_css_set to contain a subsys |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 2846 | * pointer to this state - since the subsystem is |
Li Zefan | e8d55fd | 2008-04-29 01:00:13 -0700 | [diff] [blame] | 2847 | * newly registered, all tasks and hence the |
| 2848 | * init_css_set is in the subsystem's top cgroup. */ |
| 2849 | init_css_set.subsys[ss->subsys_id] = dummytop->subsys[ss->subsys_id]; |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 2850 | |
| 2851 | need_forkexit_callback |= ss->fork || ss->exit; |
| 2852 | |
Li Zefan | e8d55fd | 2008-04-29 01:00:13 -0700 | [diff] [blame] | 2853 | /* At system boot, before all subsystems have been |
| 2854 | * registered, no tasks have been forked, so we don't |
| 2855 | * need to invoke fork callbacks here. */ |
| 2856 | BUG_ON(!list_empty(&init_task.tasks)); |
| 2857 | |
Paul Menage | 999cd8a | 2009-01-07 18:08:36 -0800 | [diff] [blame] | 2858 | mutex_init(&ss->hierarchy_mutex); |
Li Zefan | cfebe56 | 2009-02-11 13:04:36 -0800 | [diff] [blame] | 2859 | lockdep_set_class(&ss->hierarchy_mutex, &ss->subsys_key); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 2860 | ss->active = 1; |
| 2861 | } |
| 2862 | |
| 2863 | /** |
Li Zefan | a043e3b | 2008-02-23 15:24:09 -0800 | [diff] [blame] | 2864 | * cgroup_init_early - cgroup initialization at system boot |
| 2865 | * |
| 2866 | * Initialize cgroups at system boot, and initialize any |
| 2867 | * subsystems that request early init. |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 2868 | */ |
| 2869 | int __init cgroup_init_early(void) |
| 2870 | { |
| 2871 | int i; |
Lai Jiangshan | 146aa1b | 2008-10-18 20:28:03 -0700 | [diff] [blame] | 2872 | atomic_set(&init_css_set.refcount, 1); |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 2873 | INIT_LIST_HEAD(&init_css_set.cg_links); |
| 2874 | INIT_LIST_HEAD(&init_css_set.tasks); |
Li Zefan | 472b105 | 2008-04-29 01:00:11 -0700 | [diff] [blame] | 2875 | INIT_HLIST_NODE(&init_css_set.hlist); |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 2876 | css_set_count = 1; |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 2877 | init_cgroup_root(&rootnode); |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 2878 | root_count = 1; |
| 2879 | init_task.cgroups = &init_css_set; |
| 2880 | |
| 2881 | init_css_set_link.cg = &init_css_set; |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 2882 | list_add(&init_css_set_link.cgrp_link_list, |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 2883 | &rootnode.top_cgroup.css_sets); |
| 2884 | list_add(&init_css_set_link.cg_link_list, |
| 2885 | &init_css_set.cg_links); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 2886 | |
Li Zefan | 472b105 | 2008-04-29 01:00:11 -0700 | [diff] [blame] | 2887 | for (i = 0; i < CSS_SET_TABLE_SIZE; i++) |
| 2888 | INIT_HLIST_HEAD(&css_set_table[i]); |
| 2889 | |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 2890 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { |
| 2891 | struct cgroup_subsys *ss = subsys[i]; |
| 2892 | |
| 2893 | BUG_ON(!ss->name); |
| 2894 | BUG_ON(strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN); |
| 2895 | BUG_ON(!ss->create); |
| 2896 | BUG_ON(!ss->destroy); |
| 2897 | if (ss->subsys_id != i) { |
Diego Calleja | cfe36bd | 2007-11-14 16:58:54 -0800 | [diff] [blame] | 2898 | printk(KERN_ERR "cgroup: Subsys %s id == %d\n", |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 2899 | ss->name, ss->subsys_id); |
| 2900 | BUG(); |
| 2901 | } |
| 2902 | |
| 2903 | if (ss->early_init) |
| 2904 | cgroup_init_subsys(ss); |
| 2905 | } |
| 2906 | return 0; |
| 2907 | } |
| 2908 | |
| 2909 | /** |
Li Zefan | a043e3b | 2008-02-23 15:24:09 -0800 | [diff] [blame] | 2910 | * cgroup_init - cgroup initialization |
| 2911 | * |
| 2912 | * Register cgroup filesystem and /proc file, and initialize |
| 2913 | * any subsystems that didn't request early init. |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 2914 | */ |
| 2915 | int __init cgroup_init(void) |
| 2916 | { |
| 2917 | int err; |
| 2918 | int i; |
Li Zefan | 472b105 | 2008-04-29 01:00:11 -0700 | [diff] [blame] | 2919 | struct hlist_head *hhead; |
Paul Menage | a424316 | 2007-10-18 23:39:35 -0700 | [diff] [blame] | 2920 | |
| 2921 | err = bdi_init(&cgroup_backing_dev_info); |
| 2922 | if (err) |
| 2923 | return err; |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 2924 | |
| 2925 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { |
| 2926 | struct cgroup_subsys *ss = subsys[i]; |
| 2927 | if (!ss->early_init) |
| 2928 | cgroup_init_subsys(ss); |
KAMEZAWA Hiroyuki | 38460b4 | 2009-04-02 16:57:25 -0700 | [diff] [blame] | 2929 | if (ss->use_id) |
| 2930 | cgroup_subsys_init_idr(ss); |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 2931 | } |
| 2932 | |
Li Zefan | 472b105 | 2008-04-29 01:00:11 -0700 | [diff] [blame] | 2933 | /* Add init_css_set to the hash table */ |
| 2934 | hhead = css_set_hash(init_css_set.subsys); |
| 2935 | hlist_add_head(&init_css_set.hlist, hhead); |
| 2936 | |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 2937 | err = register_filesystem(&cgroup_fs_type); |
| 2938 | if (err < 0) |
| 2939 | goto out; |
| 2940 | |
Li Zefan | 46ae220 | 2008-04-29 01:00:08 -0700 | [diff] [blame] | 2941 | proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations); |
Paul Menage | a424316 | 2007-10-18 23:39:35 -0700 | [diff] [blame] | 2942 | |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 2943 | out: |
Paul Menage | a424316 | 2007-10-18 23:39:35 -0700 | [diff] [blame] | 2944 | if (err) |
| 2945 | bdi_destroy(&cgroup_backing_dev_info); |
| 2946 | |
Paul Menage | ddbcc7e | 2007-10-18 23:39:30 -0700 | [diff] [blame] | 2947 | return err; |
| 2948 | } |
Paul Menage | b4f48b6 | 2007-10-18 23:39:33 -0700 | [diff] [blame] | 2949 | |
Paul Menage | a424316 | 2007-10-18 23:39:35 -0700 | [diff] [blame] | 2950 | /* |
| 2951 | * proc_cgroup_show() |
| 2952 | * - Print task's cgroup paths into seq_file, one line for each hierarchy |
| 2953 | * - Used for /proc/<pid>/cgroup. |
| 2954 | * - No need to task_lock(tsk) on this tsk->cgroup reference, as it |
| 2955 | * doesn't really matter if tsk->cgroup changes after we read it, |
Cliff Wickman | 956db3c | 2008-02-07 00:14:43 -0800 | [diff] [blame] | 2956 | * and we take cgroup_mutex, keeping cgroup_attach_task() from changing it |
Paul Menage | a424316 | 2007-10-18 23:39:35 -0700 | [diff] [blame] | 2957 | * anyway. No need to check that tsk->cgroup != NULL, thanks to |
| 2958 | * the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks |
| 2959 | * cgroup to top_cgroup. |
| 2960 | */ |
| 2961 | |
| 2962 | /* TODO: Use a proper seq_file iterator */ |
| 2963 | static int proc_cgroup_show(struct seq_file *m, void *v) |
| 2964 | { |
| 2965 | struct pid *pid; |
| 2966 | struct task_struct *tsk; |
| 2967 | char *buf; |
| 2968 | int retval; |
| 2969 | struct cgroupfs_root *root; |
| 2970 | |
| 2971 | retval = -ENOMEM; |
| 2972 | buf = kmalloc(PAGE_SIZE, GFP_KERNEL); |
| 2973 | if (!buf) |
| 2974 | goto out; |
| 2975 | |
| 2976 | retval = -ESRCH; |
| 2977 | pid = m->private; |
| 2978 | tsk = get_pid_task(pid, PIDTYPE_PID); |
| 2979 | if (!tsk) |
| 2980 | goto out_free; |
| 2981 | |
| 2982 | retval = 0; |
| 2983 | |
| 2984 | mutex_lock(&cgroup_mutex); |
| 2985 | |
Li Zefan | e5f6a86 | 2009-01-07 18:07:41 -0800 | [diff] [blame] | 2986 | for_each_active_root(root) { |
Paul Menage | a424316 | 2007-10-18 23:39:35 -0700 | [diff] [blame] | 2987 | struct cgroup_subsys *ss; |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 2988 | struct cgroup *cgrp; |
Paul Menage | a424316 | 2007-10-18 23:39:35 -0700 | [diff] [blame] | 2989 | int subsys_id; |
| 2990 | int count = 0; |
| 2991 | |
Paul Menage | b6c3006 | 2008-04-10 21:29:16 -0700 | [diff] [blame] | 2992 | seq_printf(m, "%lu:", root->subsys_bits); |
Paul Menage | a424316 | 2007-10-18 23:39:35 -0700 | [diff] [blame] | 2993 | for_each_subsys(root, ss) |
| 2994 | seq_printf(m, "%s%s", count++ ? "," : "", ss->name); |
| 2995 | seq_putc(m, ':'); |
| 2996 | get_first_subsys(&root->top_cgroup, NULL, &subsys_id); |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 2997 | cgrp = task_cgroup(tsk, subsys_id); |
| 2998 | retval = cgroup_path(cgrp, buf, PAGE_SIZE); |
Paul Menage | a424316 | 2007-10-18 23:39:35 -0700 | [diff] [blame] | 2999 | if (retval < 0) |
| 3000 | goto out_unlock; |
| 3001 | seq_puts(m, buf); |
| 3002 | seq_putc(m, '\n'); |
| 3003 | } |
| 3004 | |
| 3005 | out_unlock: |
| 3006 | mutex_unlock(&cgroup_mutex); |
| 3007 | put_task_struct(tsk); |
| 3008 | out_free: |
| 3009 | kfree(buf); |
| 3010 | out: |
| 3011 | return retval; |
| 3012 | } |
| 3013 | |
| 3014 | static int cgroup_open(struct inode *inode, struct file *file) |
| 3015 | { |
| 3016 | struct pid *pid = PROC_I(inode)->pid; |
| 3017 | return single_open(file, proc_cgroup_show, pid); |
| 3018 | } |
| 3019 | |
| 3020 | struct file_operations proc_cgroup_operations = { |
| 3021 | .open = cgroup_open, |
| 3022 | .read = seq_read, |
| 3023 | .llseek = seq_lseek, |
| 3024 | .release = single_release, |
| 3025 | }; |
| 3026 | |
| 3027 | /* Display information about each subsystem and each hierarchy */ |
| 3028 | static int proc_cgroupstats_show(struct seq_file *m, void *v) |
| 3029 | { |
| 3030 | int i; |
Paul Menage | a424316 | 2007-10-18 23:39:35 -0700 | [diff] [blame] | 3031 | |
Paul Menage | 8bab8dd | 2008-04-04 14:29:57 -0700 | [diff] [blame] | 3032 | seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n"); |
Paul Menage | a424316 | 2007-10-18 23:39:35 -0700 | [diff] [blame] | 3033 | mutex_lock(&cgroup_mutex); |
Paul Menage | a424316 | 2007-10-18 23:39:35 -0700 | [diff] [blame] | 3034 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { |
| 3035 | struct cgroup_subsys *ss = subsys[i]; |
Paul Menage | 8bab8dd | 2008-04-04 14:29:57 -0700 | [diff] [blame] | 3036 | seq_printf(m, "%s\t%lu\t%d\t%d\n", |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 3037 | ss->name, ss->root->subsys_bits, |
Paul Menage | 8bab8dd | 2008-04-04 14:29:57 -0700 | [diff] [blame] | 3038 | ss->root->number_of_cgroups, !ss->disabled); |
Paul Menage | a424316 | 2007-10-18 23:39:35 -0700 | [diff] [blame] | 3039 | } |
| 3040 | mutex_unlock(&cgroup_mutex); |
| 3041 | return 0; |
| 3042 | } |
| 3043 | |
| 3044 | static int cgroupstats_open(struct inode *inode, struct file *file) |
| 3045 | { |
Al Viro | 9dce07f | 2008-03-29 03:07:28 +0000 | [diff] [blame] | 3046 | return single_open(file, proc_cgroupstats_show, NULL); |
Paul Menage | a424316 | 2007-10-18 23:39:35 -0700 | [diff] [blame] | 3047 | } |
| 3048 | |
| 3049 | static struct file_operations proc_cgroupstats_operations = { |
| 3050 | .open = cgroupstats_open, |
| 3051 | .read = seq_read, |
| 3052 | .llseek = seq_lseek, |
| 3053 | .release = single_release, |
| 3054 | }; |
| 3055 | |
Paul Menage | b4f48b6 | 2007-10-18 23:39:33 -0700 | [diff] [blame] | 3056 | /** |
| 3057 | * cgroup_fork - attach newly forked task to its parents cgroup. |
Li Zefan | a043e3b | 2008-02-23 15:24:09 -0800 | [diff] [blame] | 3058 | * @child: pointer to task_struct of forking parent process. |
Paul Menage | b4f48b6 | 2007-10-18 23:39:33 -0700 | [diff] [blame] | 3059 | * |
| 3060 | * Description: A task inherits its parent's cgroup at fork(). |
| 3061 | * |
| 3062 | * A pointer to the shared css_set was automatically copied in |
| 3063 | * fork.c by dup_task_struct(). However, we ignore that copy, since |
| 3064 | * it was not made under the protection of RCU or cgroup_mutex, so |
Cliff Wickman | 956db3c | 2008-02-07 00:14:43 -0800 | [diff] [blame] | 3065 | * might no longer be a valid cgroup pointer. cgroup_attach_task() might |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 3066 | * have already changed current->cgroups, allowing the previously |
| 3067 | * referenced cgroup group to be removed and freed. |
Paul Menage | b4f48b6 | 2007-10-18 23:39:33 -0700 | [diff] [blame] | 3068 | * |
| 3069 | * At the point that cgroup_fork() is called, 'current' is the parent |
| 3070 | * task, and the passed argument 'child' points to the child task. |
| 3071 | */ |
| 3072 | void cgroup_fork(struct task_struct *child) |
| 3073 | { |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 3074 | task_lock(current); |
| 3075 | child->cgroups = current->cgroups; |
| 3076 | get_css_set(child->cgroups); |
| 3077 | task_unlock(current); |
| 3078 | INIT_LIST_HEAD(&child->cg_list); |
Paul Menage | b4f48b6 | 2007-10-18 23:39:33 -0700 | [diff] [blame] | 3079 | } |
| 3080 | |
| 3081 | /** |
Li Zefan | a043e3b | 2008-02-23 15:24:09 -0800 | [diff] [blame] | 3082 | * cgroup_fork_callbacks - run fork callbacks |
| 3083 | * @child: the new task |
| 3084 | * |
| 3085 | * Called on a new task very soon before adding it to the |
| 3086 | * tasklist. No need to take any locks since no-one can |
| 3087 | * be operating on this task. |
Paul Menage | b4f48b6 | 2007-10-18 23:39:33 -0700 | [diff] [blame] | 3088 | */ |
| 3089 | void cgroup_fork_callbacks(struct task_struct *child) |
| 3090 | { |
| 3091 | if (need_forkexit_callback) { |
| 3092 | int i; |
| 3093 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { |
| 3094 | struct cgroup_subsys *ss = subsys[i]; |
| 3095 | if (ss->fork) |
| 3096 | ss->fork(ss, child); |
| 3097 | } |
| 3098 | } |
| 3099 | } |
| 3100 | |
| 3101 | /** |
Li Zefan | a043e3b | 2008-02-23 15:24:09 -0800 | [diff] [blame] | 3102 | * cgroup_post_fork - called on a new task after adding it to the task list |
| 3103 | * @child: the task in question |
| 3104 | * |
| 3105 | * Adds the task to the list running through its css_set if necessary. |
| 3106 | * Has to be after the task is visible on the task list in case we race |
| 3107 | * with the first call to cgroup_iter_start() - to guarantee that the |
| 3108 | * new task ends up on its list. |
| 3109 | */ |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 3110 | void cgroup_post_fork(struct task_struct *child) |
| 3111 | { |
| 3112 | if (use_task_css_set_links) { |
| 3113 | write_lock(&css_set_lock); |
Lai Jiangshan | b12b533 | 2009-01-07 18:07:36 -0800 | [diff] [blame] | 3114 | task_lock(child); |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 3115 | if (list_empty(&child->cg_list)) |
| 3116 | list_add(&child->cg_list, &child->cgroups->tasks); |
Lai Jiangshan | b12b533 | 2009-01-07 18:07:36 -0800 | [diff] [blame] | 3117 | task_unlock(child); |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 3118 | write_unlock(&css_set_lock); |
| 3119 | } |
| 3120 | } |
| 3121 | /** |
Paul Menage | b4f48b6 | 2007-10-18 23:39:33 -0700 | [diff] [blame] | 3122 | * cgroup_exit - detach cgroup from exiting task |
| 3123 | * @tsk: pointer to task_struct of exiting process |
Li Zefan | a043e3b | 2008-02-23 15:24:09 -0800 | [diff] [blame] | 3124 | * @run_callback: run exit callbacks? |
Paul Menage | b4f48b6 | 2007-10-18 23:39:33 -0700 | [diff] [blame] | 3125 | * |
| 3126 | * Description: Detach cgroup from @tsk and release it. |
| 3127 | * |
| 3128 | * Note that cgroups marked notify_on_release force every task in |
| 3129 | * them to take the global cgroup_mutex mutex when exiting. |
| 3130 | * This could impact scaling on very large systems. Be reluctant to |
| 3131 | * use notify_on_release cgroups where very high task exit scaling |
| 3132 | * is required on large systems. |
| 3133 | * |
| 3134 | * the_top_cgroup_hack: |
| 3135 | * |
| 3136 | * Set the exiting tasks cgroup to the root cgroup (top_cgroup). |
| 3137 | * |
| 3138 | * We call cgroup_exit() while the task is still competent to |
| 3139 | * handle notify_on_release(), then leave the task attached to the |
| 3140 | * root cgroup in each hierarchy for the remainder of its exit. |
| 3141 | * |
| 3142 | * To do this properly, we would increment the reference count on |
| 3143 | * top_cgroup, and near the very end of the kernel/exit.c do_exit() |
| 3144 | * code we would add a second cgroup function call, to drop that |
| 3145 | * reference. This would just create an unnecessary hot spot on |
| 3146 | * the top_cgroup reference count, to no avail. |
| 3147 | * |
| 3148 | * Normally, holding a reference to a cgroup without bumping its |
| 3149 | * count is unsafe. The cgroup could go away, or someone could |
| 3150 | * attach us to a different cgroup, decrementing the count on |
| 3151 | * the first cgroup that we never incremented. But in this case, |
| 3152 | * top_cgroup isn't going away, and either task has PF_EXITING set, |
Cliff Wickman | 956db3c | 2008-02-07 00:14:43 -0800 | [diff] [blame] | 3153 | * which wards off any cgroup_attach_task() attempts, or task is a failed |
| 3154 | * fork, never visible to cgroup_attach_task. |
Paul Menage | b4f48b6 | 2007-10-18 23:39:33 -0700 | [diff] [blame] | 3155 | */ |
| 3156 | void cgroup_exit(struct task_struct *tsk, int run_callbacks) |
| 3157 | { |
| 3158 | int i; |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 3159 | struct css_set *cg; |
Paul Menage | b4f48b6 | 2007-10-18 23:39:33 -0700 | [diff] [blame] | 3160 | |
| 3161 | if (run_callbacks && need_forkexit_callback) { |
| 3162 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { |
| 3163 | struct cgroup_subsys *ss = subsys[i]; |
| 3164 | if (ss->exit) |
| 3165 | ss->exit(ss, tsk); |
| 3166 | } |
| 3167 | } |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 3168 | |
| 3169 | /* |
| 3170 | * Unlink from the css_set task list if necessary. |
| 3171 | * Optimistically check cg_list before taking |
| 3172 | * css_set_lock |
| 3173 | */ |
| 3174 | if (!list_empty(&tsk->cg_list)) { |
| 3175 | write_lock(&css_set_lock); |
| 3176 | if (!list_empty(&tsk->cg_list)) |
| 3177 | list_del(&tsk->cg_list); |
| 3178 | write_unlock(&css_set_lock); |
| 3179 | } |
| 3180 | |
Paul Menage | b4f48b6 | 2007-10-18 23:39:33 -0700 | [diff] [blame] | 3181 | /* Reassign the task to the init_css_set. */ |
| 3182 | task_lock(tsk); |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 3183 | cg = tsk->cgroups; |
| 3184 | tsk->cgroups = &init_css_set; |
Paul Menage | b4f48b6 | 2007-10-18 23:39:33 -0700 | [diff] [blame] | 3185 | task_unlock(tsk); |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 3186 | if (cg) |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 3187 | put_css_set_taskexit(cg); |
Paul Menage | b4f48b6 | 2007-10-18 23:39:33 -0700 | [diff] [blame] | 3188 | } |
Paul Menage | 697f416 | 2007-10-18 23:39:34 -0700 | [diff] [blame] | 3189 | |
| 3190 | /** |
Li Zefan | a043e3b | 2008-02-23 15:24:09 -0800 | [diff] [blame] | 3191 | * cgroup_clone - clone the cgroup the given subsystem is attached to |
| 3192 | * @tsk: the task to be moved |
| 3193 | * @subsys: the given subsystem |
Serge E. Hallyn | e885dcd | 2008-07-25 01:47:06 -0700 | [diff] [blame] | 3194 | * @nodename: the name for the new cgroup |
Li Zefan | a043e3b | 2008-02-23 15:24:09 -0800 | [diff] [blame] | 3195 | * |
| 3196 | * Duplicate the current cgroup in the hierarchy that the given |
| 3197 | * subsystem is attached to, and move this task into the new |
| 3198 | * child. |
Paul Menage | 697f416 | 2007-10-18 23:39:34 -0700 | [diff] [blame] | 3199 | */ |
Serge E. Hallyn | e885dcd | 2008-07-25 01:47:06 -0700 | [diff] [blame] | 3200 | int cgroup_clone(struct task_struct *tsk, struct cgroup_subsys *subsys, |
| 3201 | char *nodename) |
Paul Menage | 697f416 | 2007-10-18 23:39:34 -0700 | [diff] [blame] | 3202 | { |
| 3203 | struct dentry *dentry; |
| 3204 | int ret = 0; |
Paul Menage | 697f416 | 2007-10-18 23:39:34 -0700 | [diff] [blame] | 3205 | struct cgroup *parent, *child; |
| 3206 | struct inode *inode; |
| 3207 | struct css_set *cg; |
| 3208 | struct cgroupfs_root *root; |
| 3209 | struct cgroup_subsys *ss; |
| 3210 | |
| 3211 | /* We shouldn't be called by an unregistered subsystem */ |
| 3212 | BUG_ON(!subsys->active); |
| 3213 | |
| 3214 | /* First figure out what hierarchy and cgroup we're dealing |
| 3215 | * with, and pin them so we can drop cgroup_mutex */ |
| 3216 | mutex_lock(&cgroup_mutex); |
| 3217 | again: |
| 3218 | root = subsys->root; |
| 3219 | if (root == &rootnode) { |
Paul Menage | 697f416 | 2007-10-18 23:39:34 -0700 | [diff] [blame] | 3220 | mutex_unlock(&cgroup_mutex); |
| 3221 | return 0; |
| 3222 | } |
Paul Menage | 697f416 | 2007-10-18 23:39:34 -0700 | [diff] [blame] | 3223 | |
Paul Menage | 697f416 | 2007-10-18 23:39:34 -0700 | [diff] [blame] | 3224 | /* Pin the hierarchy */ |
Li Zefan | 1404f06 | 2009-01-29 14:25:21 -0800 | [diff] [blame] | 3225 | if (!atomic_inc_not_zero(&root->sb->s_active)) { |
Li Zefan | 7b574b7 | 2009-01-04 12:00:45 -0800 | [diff] [blame] | 3226 | /* We race with the final deactivate_super() */ |
| 3227 | mutex_unlock(&cgroup_mutex); |
| 3228 | return 0; |
| 3229 | } |
Paul Menage | 697f416 | 2007-10-18 23:39:34 -0700 | [diff] [blame] | 3230 | |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 3231 | /* Keep the cgroup alive */ |
Li Zefan | 1404f06 | 2009-01-29 14:25:21 -0800 | [diff] [blame] | 3232 | task_lock(tsk); |
| 3233 | parent = task_cgroup(tsk, subsys->subsys_id); |
| 3234 | cg = tsk->cgroups; |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 3235 | get_css_set(cg); |
Lai Jiangshan | 104cbd5 | 2009-01-07 18:07:38 -0800 | [diff] [blame] | 3236 | task_unlock(tsk); |
Li Zefan | 1404f06 | 2009-01-29 14:25:21 -0800 | [diff] [blame] | 3237 | |
Paul Menage | 697f416 | 2007-10-18 23:39:34 -0700 | [diff] [blame] | 3238 | mutex_unlock(&cgroup_mutex); |
| 3239 | |
| 3240 | /* Now do the VFS work to create a cgroup */ |
| 3241 | inode = parent->dentry->d_inode; |
| 3242 | |
| 3243 | /* Hold the parent directory mutex across this operation to |
| 3244 | * stop anyone else deleting the new cgroup */ |
| 3245 | mutex_lock(&inode->i_mutex); |
| 3246 | dentry = lookup_one_len(nodename, parent->dentry, strlen(nodename)); |
| 3247 | if (IS_ERR(dentry)) { |
| 3248 | printk(KERN_INFO |
Diego Calleja | cfe36bd | 2007-11-14 16:58:54 -0800 | [diff] [blame] | 3249 | "cgroup: Couldn't allocate dentry for %s: %ld\n", nodename, |
Paul Menage | 697f416 | 2007-10-18 23:39:34 -0700 | [diff] [blame] | 3250 | PTR_ERR(dentry)); |
| 3251 | ret = PTR_ERR(dentry); |
| 3252 | goto out_release; |
| 3253 | } |
| 3254 | |
| 3255 | /* Create the cgroup directory, which also creates the cgroup */ |
Li Zefan | 75139b8 | 2009-01-07 18:07:33 -0800 | [diff] [blame] | 3256 | ret = vfs_mkdir(inode, dentry, 0755); |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 3257 | child = __d_cgrp(dentry); |
Paul Menage | 697f416 | 2007-10-18 23:39:34 -0700 | [diff] [blame] | 3258 | dput(dentry); |
| 3259 | if (ret) { |
| 3260 | printk(KERN_INFO |
| 3261 | "Failed to create cgroup %s: %d\n", nodename, |
| 3262 | ret); |
| 3263 | goto out_release; |
| 3264 | } |
| 3265 | |
Paul Menage | 697f416 | 2007-10-18 23:39:34 -0700 | [diff] [blame] | 3266 | /* The cgroup now exists. Retake cgroup_mutex and check |
| 3267 | * that we're still in the same state that we thought we |
| 3268 | * were. */ |
| 3269 | mutex_lock(&cgroup_mutex); |
| 3270 | if ((root != subsys->root) || |
| 3271 | (parent != task_cgroup(tsk, subsys->subsys_id))) { |
| 3272 | /* Aargh, we raced ... */ |
| 3273 | mutex_unlock(&inode->i_mutex); |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 3274 | put_css_set(cg); |
Paul Menage | 697f416 | 2007-10-18 23:39:34 -0700 | [diff] [blame] | 3275 | |
Li Zefan | 1404f06 | 2009-01-29 14:25:21 -0800 | [diff] [blame] | 3276 | deactivate_super(root->sb); |
Paul Menage | 697f416 | 2007-10-18 23:39:34 -0700 | [diff] [blame] | 3277 | /* The cgroup is still accessible in the VFS, but |
| 3278 | * we're not going to try to rmdir() it at this |
| 3279 | * point. */ |
| 3280 | printk(KERN_INFO |
| 3281 | "Race in cgroup_clone() - leaking cgroup %s\n", |
| 3282 | nodename); |
| 3283 | goto again; |
| 3284 | } |
| 3285 | |
| 3286 | /* do any required auto-setup */ |
| 3287 | for_each_subsys(root, ss) { |
| 3288 | if (ss->post_clone) |
| 3289 | ss->post_clone(ss, child); |
| 3290 | } |
| 3291 | |
| 3292 | /* All seems fine. Finish by moving the task into the new cgroup */ |
Cliff Wickman | 956db3c | 2008-02-07 00:14:43 -0800 | [diff] [blame] | 3293 | ret = cgroup_attach_task(child, tsk); |
Paul Menage | 697f416 | 2007-10-18 23:39:34 -0700 | [diff] [blame] | 3294 | mutex_unlock(&cgroup_mutex); |
| 3295 | |
| 3296 | out_release: |
| 3297 | mutex_unlock(&inode->i_mutex); |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 3298 | |
| 3299 | mutex_lock(&cgroup_mutex); |
Paul Menage | 817929e | 2007-10-18 23:39:36 -0700 | [diff] [blame] | 3300 | put_css_set(cg); |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 3301 | mutex_unlock(&cgroup_mutex); |
Li Zefan | 1404f06 | 2009-01-29 14:25:21 -0800 | [diff] [blame] | 3302 | deactivate_super(root->sb); |
Paul Menage | 697f416 | 2007-10-18 23:39:34 -0700 | [diff] [blame] | 3303 | return ret; |
| 3304 | } |
| 3305 | |
Li Zefan | a043e3b | 2008-02-23 15:24:09 -0800 | [diff] [blame] | 3306 | /** |
Grzegorz Nosek | 313e924 | 2009-04-02 16:57:23 -0700 | [diff] [blame] | 3307 | * cgroup_is_descendant - see if @cgrp is a descendant of @task's cgrp |
Li Zefan | a043e3b | 2008-02-23 15:24:09 -0800 | [diff] [blame] | 3308 | * @cgrp: the cgroup in question |
Grzegorz Nosek | 313e924 | 2009-04-02 16:57:23 -0700 | [diff] [blame] | 3309 | * @task: the task in question |
Li Zefan | a043e3b | 2008-02-23 15:24:09 -0800 | [diff] [blame] | 3310 | * |
Grzegorz Nosek | 313e924 | 2009-04-02 16:57:23 -0700 | [diff] [blame] | 3311 | * See if @cgrp is a descendant of @task's cgroup in the appropriate |
| 3312 | * hierarchy. |
Paul Menage | 697f416 | 2007-10-18 23:39:34 -0700 | [diff] [blame] | 3313 | * |
| 3314 | * If we are sending in dummytop, then presumably we are creating |
| 3315 | * the top cgroup in the subsystem. |
| 3316 | * |
| 3317 | * Called only by the ns (nsproxy) cgroup. |
| 3318 | */ |
Grzegorz Nosek | 313e924 | 2009-04-02 16:57:23 -0700 | [diff] [blame] | 3319 | int cgroup_is_descendant(const struct cgroup *cgrp, struct task_struct *task) |
Paul Menage | 697f416 | 2007-10-18 23:39:34 -0700 | [diff] [blame] | 3320 | { |
| 3321 | int ret; |
| 3322 | struct cgroup *target; |
| 3323 | int subsys_id; |
| 3324 | |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 3325 | if (cgrp == dummytop) |
Paul Menage | 697f416 | 2007-10-18 23:39:34 -0700 | [diff] [blame] | 3326 | return 1; |
| 3327 | |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 3328 | get_first_subsys(cgrp, NULL, &subsys_id); |
Grzegorz Nosek | 313e924 | 2009-04-02 16:57:23 -0700 | [diff] [blame] | 3329 | target = task_cgroup(task, subsys_id); |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 3330 | while (cgrp != target && cgrp!= cgrp->top_cgroup) |
| 3331 | cgrp = cgrp->parent; |
| 3332 | ret = (cgrp == target); |
Paul Menage | 697f416 | 2007-10-18 23:39:34 -0700 | [diff] [blame] | 3333 | return ret; |
| 3334 | } |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 3335 | |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 3336 | static void check_for_release(struct cgroup *cgrp) |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 3337 | { |
| 3338 | /* All of these checks rely on RCU to keep the cgroup |
| 3339 | * structure alive */ |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 3340 | if (cgroup_is_releasable(cgrp) && !atomic_read(&cgrp->count) |
| 3341 | && list_empty(&cgrp->children) && !cgroup_has_css_refs(cgrp)) { |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 3342 | /* Control Group is currently removeable. If it's not |
| 3343 | * already queued for a userspace notification, queue |
| 3344 | * it now */ |
| 3345 | int need_schedule_work = 0; |
| 3346 | spin_lock(&release_list_lock); |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 3347 | if (!cgroup_is_removed(cgrp) && |
| 3348 | list_empty(&cgrp->release_list)) { |
| 3349 | list_add(&cgrp->release_list, &release_list); |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 3350 | need_schedule_work = 1; |
| 3351 | } |
| 3352 | spin_unlock(&release_list_lock); |
| 3353 | if (need_schedule_work) |
| 3354 | schedule_work(&release_agent_work); |
| 3355 | } |
| 3356 | } |
| 3357 | |
| 3358 | void __css_put(struct cgroup_subsys_state *css) |
| 3359 | { |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 3360 | struct cgroup *cgrp = css->cgroup; |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 3361 | rcu_read_lock(); |
KAMEZAWA Hiroyuki | ec64f51 | 2009-04-02 16:57:26 -0700 | [diff] [blame] | 3362 | if (atomic_dec_return(&css->refcnt) == 1) { |
| 3363 | if (notify_on_release(cgrp)) { |
| 3364 | set_bit(CGRP_RELEASABLE, &cgrp->flags); |
| 3365 | check_for_release(cgrp); |
| 3366 | } |
KAMEZAWA Hiroyuki | 8870326 | 2009-07-29 15:04:06 -0700 | [diff] [blame] | 3367 | cgroup_wakeup_rmdir_waiter(cgrp); |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 3368 | } |
| 3369 | rcu_read_unlock(); |
| 3370 | } |
| 3371 | |
| 3372 | /* |
| 3373 | * Notify userspace when a cgroup is released, by running the |
| 3374 | * configured release agent with the name of the cgroup (path |
| 3375 | * relative to the root of cgroup file system) as the argument. |
| 3376 | * |
| 3377 | * Most likely, this user command will try to rmdir this cgroup. |
| 3378 | * |
| 3379 | * This races with the possibility that some other task will be |
| 3380 | * attached to this cgroup before it is removed, or that some other |
| 3381 | * user task will 'mkdir' a child cgroup of this cgroup. That's ok. |
| 3382 | * The presumed 'rmdir' will fail quietly if this cgroup is no longer |
| 3383 | * unused, and this cgroup will be reprieved from its death sentence, |
| 3384 | * to continue to serve a useful existence. Next time it's released, |
| 3385 | * we will get notified again, if it still has 'notify_on_release' set. |
| 3386 | * |
| 3387 | * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which |
| 3388 | * means only wait until the task is successfully execve()'d. The |
| 3389 | * separate release agent task is forked by call_usermodehelper(), |
| 3390 | * then control in this thread returns here, without waiting for the |
| 3391 | * release agent task. We don't bother to wait because the caller of |
| 3392 | * this routine has no use for the exit status of the release agent |
| 3393 | * task, so no sense holding our caller up for that. |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 3394 | */ |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 3395 | static void cgroup_release_agent(struct work_struct *work) |
| 3396 | { |
| 3397 | BUG_ON(work != &release_agent_work); |
| 3398 | mutex_lock(&cgroup_mutex); |
| 3399 | spin_lock(&release_list_lock); |
| 3400 | while (!list_empty(&release_list)) { |
| 3401 | char *argv[3], *envp[3]; |
| 3402 | int i; |
Paul Menage | e788e06 | 2008-07-25 01:46:59 -0700 | [diff] [blame] | 3403 | char *pathbuf = NULL, *agentbuf = NULL; |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 3404 | struct cgroup *cgrp = list_entry(release_list.next, |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 3405 | struct cgroup, |
| 3406 | release_list); |
Paul Menage | bd89aab | 2007-10-18 23:40:44 -0700 | [diff] [blame] | 3407 | list_del_init(&cgrp->release_list); |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 3408 | spin_unlock(&release_list_lock); |
| 3409 | pathbuf = kmalloc(PAGE_SIZE, GFP_KERNEL); |
Paul Menage | e788e06 | 2008-07-25 01:46:59 -0700 | [diff] [blame] | 3410 | if (!pathbuf) |
| 3411 | goto continue_free; |
| 3412 | if (cgroup_path(cgrp, pathbuf, PAGE_SIZE) < 0) |
| 3413 | goto continue_free; |
| 3414 | agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL); |
| 3415 | if (!agentbuf) |
| 3416 | goto continue_free; |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 3417 | |
| 3418 | i = 0; |
Paul Menage | e788e06 | 2008-07-25 01:46:59 -0700 | [diff] [blame] | 3419 | argv[i++] = agentbuf; |
| 3420 | argv[i++] = pathbuf; |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 3421 | argv[i] = NULL; |
| 3422 | |
| 3423 | i = 0; |
| 3424 | /* minimal command environment */ |
| 3425 | envp[i++] = "HOME=/"; |
| 3426 | envp[i++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin"; |
| 3427 | envp[i] = NULL; |
| 3428 | |
| 3429 | /* Drop the lock while we invoke the usermode helper, |
| 3430 | * since the exec could involve hitting disk and hence |
| 3431 | * be a slow process */ |
| 3432 | mutex_unlock(&cgroup_mutex); |
| 3433 | call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC); |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 3434 | mutex_lock(&cgroup_mutex); |
Paul Menage | e788e06 | 2008-07-25 01:46:59 -0700 | [diff] [blame] | 3435 | continue_free: |
| 3436 | kfree(pathbuf); |
| 3437 | kfree(agentbuf); |
Paul Menage | 81a6a5c | 2007-10-18 23:39:38 -0700 | [diff] [blame] | 3438 | spin_lock(&release_list_lock); |
| 3439 | } |
| 3440 | spin_unlock(&release_list_lock); |
| 3441 | mutex_unlock(&cgroup_mutex); |
| 3442 | } |
Paul Menage | 8bab8dd | 2008-04-04 14:29:57 -0700 | [diff] [blame] | 3443 | |
| 3444 | static int __init cgroup_disable(char *str) |
| 3445 | { |
| 3446 | int i; |
| 3447 | char *token; |
| 3448 | |
| 3449 | while ((token = strsep(&str, ",")) != NULL) { |
| 3450 | if (!*token) |
| 3451 | continue; |
| 3452 | |
| 3453 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { |
| 3454 | struct cgroup_subsys *ss = subsys[i]; |
| 3455 | |
| 3456 | if (!strcmp(token, ss->name)) { |
| 3457 | ss->disabled = 1; |
| 3458 | printk(KERN_INFO "Disabling %s control group" |
| 3459 | " subsystem\n", ss->name); |
| 3460 | break; |
| 3461 | } |
| 3462 | } |
| 3463 | } |
| 3464 | return 1; |
| 3465 | } |
| 3466 | __setup("cgroup_disable=", cgroup_disable); |
KAMEZAWA Hiroyuki | 38460b4 | 2009-04-02 16:57:25 -0700 | [diff] [blame] | 3467 | |
| 3468 | /* |
| 3469 | * Functons for CSS ID. |
| 3470 | */ |
| 3471 | |
| 3472 | /* |
| 3473 | *To get ID other than 0, this should be called when !cgroup_is_removed(). |
| 3474 | */ |
| 3475 | unsigned short css_id(struct cgroup_subsys_state *css) |
| 3476 | { |
| 3477 | struct css_id *cssid = rcu_dereference(css->id); |
| 3478 | |
| 3479 | if (cssid) |
| 3480 | return cssid->id; |
| 3481 | return 0; |
| 3482 | } |
| 3483 | |
| 3484 | unsigned short css_depth(struct cgroup_subsys_state *css) |
| 3485 | { |
| 3486 | struct css_id *cssid = rcu_dereference(css->id); |
| 3487 | |
| 3488 | if (cssid) |
| 3489 | return cssid->depth; |
| 3490 | return 0; |
| 3491 | } |
| 3492 | |
| 3493 | bool css_is_ancestor(struct cgroup_subsys_state *child, |
KAMEZAWA Hiroyuki | 0b7f569 | 2009-04-02 16:57:38 -0700 | [diff] [blame] | 3494 | const struct cgroup_subsys_state *root) |
KAMEZAWA Hiroyuki | 38460b4 | 2009-04-02 16:57:25 -0700 | [diff] [blame] | 3495 | { |
| 3496 | struct css_id *child_id = rcu_dereference(child->id); |
| 3497 | struct css_id *root_id = rcu_dereference(root->id); |
| 3498 | |
| 3499 | if (!child_id || !root_id || (child_id->depth < root_id->depth)) |
| 3500 | return false; |
| 3501 | return child_id->stack[root_id->depth] == root_id->id; |
| 3502 | } |
| 3503 | |
| 3504 | static void __free_css_id_cb(struct rcu_head *head) |
| 3505 | { |
| 3506 | struct css_id *id; |
| 3507 | |
| 3508 | id = container_of(head, struct css_id, rcu_head); |
| 3509 | kfree(id); |
| 3510 | } |
| 3511 | |
| 3512 | void free_css_id(struct cgroup_subsys *ss, struct cgroup_subsys_state *css) |
| 3513 | { |
| 3514 | struct css_id *id = css->id; |
| 3515 | /* When this is called before css_id initialization, id can be NULL */ |
| 3516 | if (!id) |
| 3517 | return; |
| 3518 | |
| 3519 | BUG_ON(!ss->use_id); |
| 3520 | |
| 3521 | rcu_assign_pointer(id->css, NULL); |
| 3522 | rcu_assign_pointer(css->id, NULL); |
| 3523 | spin_lock(&ss->id_lock); |
| 3524 | idr_remove(&ss->idr, id->id); |
| 3525 | spin_unlock(&ss->id_lock); |
| 3526 | call_rcu(&id->rcu_head, __free_css_id_cb); |
| 3527 | } |
| 3528 | |
| 3529 | /* |
| 3530 | * This is called by init or create(). Then, calls to this function are |
| 3531 | * always serialized (By cgroup_mutex() at create()). |
| 3532 | */ |
| 3533 | |
| 3534 | static struct css_id *get_new_cssid(struct cgroup_subsys *ss, int depth) |
| 3535 | { |
| 3536 | struct css_id *newid; |
| 3537 | int myid, error, size; |
| 3538 | |
| 3539 | BUG_ON(!ss->use_id); |
| 3540 | |
| 3541 | size = sizeof(*newid) + sizeof(unsigned short) * (depth + 1); |
| 3542 | newid = kzalloc(size, GFP_KERNEL); |
| 3543 | if (!newid) |
| 3544 | return ERR_PTR(-ENOMEM); |
| 3545 | /* get id */ |
| 3546 | if (unlikely(!idr_pre_get(&ss->idr, GFP_KERNEL))) { |
| 3547 | error = -ENOMEM; |
| 3548 | goto err_out; |
| 3549 | } |
| 3550 | spin_lock(&ss->id_lock); |
| 3551 | /* Don't use 0. allocates an ID of 1-65535 */ |
| 3552 | error = idr_get_new_above(&ss->idr, newid, 1, &myid); |
| 3553 | spin_unlock(&ss->id_lock); |
| 3554 | |
| 3555 | /* Returns error when there are no free spaces for new ID.*/ |
| 3556 | if (error) { |
| 3557 | error = -ENOSPC; |
| 3558 | goto err_out; |
| 3559 | } |
| 3560 | if (myid > CSS_ID_MAX) |
| 3561 | goto remove_idr; |
| 3562 | |
| 3563 | newid->id = myid; |
| 3564 | newid->depth = depth; |
| 3565 | return newid; |
| 3566 | remove_idr: |
| 3567 | error = -ENOSPC; |
| 3568 | spin_lock(&ss->id_lock); |
| 3569 | idr_remove(&ss->idr, myid); |
| 3570 | spin_unlock(&ss->id_lock); |
| 3571 | err_out: |
| 3572 | kfree(newid); |
| 3573 | return ERR_PTR(error); |
| 3574 | |
| 3575 | } |
| 3576 | |
| 3577 | static int __init cgroup_subsys_init_idr(struct cgroup_subsys *ss) |
| 3578 | { |
| 3579 | struct css_id *newid; |
| 3580 | struct cgroup_subsys_state *rootcss; |
| 3581 | |
| 3582 | spin_lock_init(&ss->id_lock); |
| 3583 | idr_init(&ss->idr); |
| 3584 | |
| 3585 | rootcss = init_css_set.subsys[ss->subsys_id]; |
| 3586 | newid = get_new_cssid(ss, 0); |
| 3587 | if (IS_ERR(newid)) |
| 3588 | return PTR_ERR(newid); |
| 3589 | |
| 3590 | newid->stack[0] = newid->id; |
| 3591 | newid->css = rootcss; |
| 3592 | rootcss->id = newid; |
| 3593 | return 0; |
| 3594 | } |
| 3595 | |
| 3596 | static int alloc_css_id(struct cgroup_subsys *ss, struct cgroup *parent, |
| 3597 | struct cgroup *child) |
| 3598 | { |
| 3599 | int subsys_id, i, depth = 0; |
| 3600 | struct cgroup_subsys_state *parent_css, *child_css; |
| 3601 | struct css_id *child_id, *parent_id = NULL; |
| 3602 | |
| 3603 | subsys_id = ss->subsys_id; |
| 3604 | parent_css = parent->subsys[subsys_id]; |
| 3605 | child_css = child->subsys[subsys_id]; |
| 3606 | depth = css_depth(parent_css) + 1; |
| 3607 | parent_id = parent_css->id; |
| 3608 | |
| 3609 | child_id = get_new_cssid(ss, depth); |
| 3610 | if (IS_ERR(child_id)) |
| 3611 | return PTR_ERR(child_id); |
| 3612 | |
| 3613 | for (i = 0; i < depth; i++) |
| 3614 | child_id->stack[i] = parent_id->stack[i]; |
| 3615 | child_id->stack[depth] = child_id->id; |
| 3616 | /* |
| 3617 | * child_id->css pointer will be set after this cgroup is available |
| 3618 | * see cgroup_populate_dir() |
| 3619 | */ |
| 3620 | rcu_assign_pointer(child_css->id, child_id); |
| 3621 | |
| 3622 | return 0; |
| 3623 | } |
| 3624 | |
| 3625 | /** |
| 3626 | * css_lookup - lookup css by id |
| 3627 | * @ss: cgroup subsys to be looked into. |
| 3628 | * @id: the id |
| 3629 | * |
| 3630 | * Returns pointer to cgroup_subsys_state if there is valid one with id. |
| 3631 | * NULL if not. Should be called under rcu_read_lock() |
| 3632 | */ |
| 3633 | struct cgroup_subsys_state *css_lookup(struct cgroup_subsys *ss, int id) |
| 3634 | { |
| 3635 | struct css_id *cssid = NULL; |
| 3636 | |
| 3637 | BUG_ON(!ss->use_id); |
| 3638 | cssid = idr_find(&ss->idr, id); |
| 3639 | |
| 3640 | if (unlikely(!cssid)) |
| 3641 | return NULL; |
| 3642 | |
| 3643 | return rcu_dereference(cssid->css); |
| 3644 | } |
| 3645 | |
| 3646 | /** |
| 3647 | * css_get_next - lookup next cgroup under specified hierarchy. |
| 3648 | * @ss: pointer to subsystem |
| 3649 | * @id: current position of iteration. |
| 3650 | * @root: pointer to css. search tree under this. |
| 3651 | * @foundid: position of found object. |
| 3652 | * |
| 3653 | * Search next css under the specified hierarchy of rootid. Calling under |
| 3654 | * rcu_read_lock() is necessary. Returns NULL if it reaches the end. |
| 3655 | */ |
| 3656 | struct cgroup_subsys_state * |
| 3657 | css_get_next(struct cgroup_subsys *ss, int id, |
| 3658 | struct cgroup_subsys_state *root, int *foundid) |
| 3659 | { |
| 3660 | struct cgroup_subsys_state *ret = NULL; |
| 3661 | struct css_id *tmp; |
| 3662 | int tmpid; |
| 3663 | int rootid = css_id(root); |
| 3664 | int depth = css_depth(root); |
| 3665 | |
| 3666 | if (!rootid) |
| 3667 | return NULL; |
| 3668 | |
| 3669 | BUG_ON(!ss->use_id); |
| 3670 | /* fill start point for scan */ |
| 3671 | tmpid = id; |
| 3672 | while (1) { |
| 3673 | /* |
| 3674 | * scan next entry from bitmap(tree), tmpid is updated after |
| 3675 | * idr_get_next(). |
| 3676 | */ |
| 3677 | spin_lock(&ss->id_lock); |
| 3678 | tmp = idr_get_next(&ss->idr, &tmpid); |
| 3679 | spin_unlock(&ss->id_lock); |
| 3680 | |
| 3681 | if (!tmp) |
| 3682 | break; |
| 3683 | if (tmp->depth >= depth && tmp->stack[depth] == rootid) { |
| 3684 | ret = rcu_dereference(tmp->css); |
| 3685 | if (ret) { |
| 3686 | *foundid = tmpid; |
| 3687 | break; |
| 3688 | } |
| 3689 | } |
| 3690 | /* continue to scan from next id */ |
| 3691 | tmpid = tmpid + 1; |
| 3692 | } |
| 3693 | return ret; |
| 3694 | } |
| 3695 | |