]>
Commit | Line | Data |
---|---|---|
ddbcc7e8 PM |
1 | /* |
2 | * kernel/cgroup.c | |
3 | * | |
4 | * Generic process-grouping system. | |
5 | * | |
6 | * Based originally on the cpuset system, extracted by Paul Menage | |
7 | * Copyright (C) 2006 Google, Inc | |
8 | * | |
9 | * Copyright notices from the original cpuset code: | |
10 | * -------------------------------------------------- | |
11 | * Copyright (C) 2003 BULL SA. | |
12 | * Copyright (C) 2004-2006 Silicon Graphics, Inc. | |
13 | * | |
14 | * Portions derived from Patrick Mochel's sysfs code. | |
15 | * sysfs is Copyright (c) 2001-3 Patrick Mochel | |
16 | * | |
17 | * 2003-10-10 Written by Simon Derr. | |
18 | * 2003-10-22 Updates by Stephen Hemminger. | |
19 | * 2004 May-July Rework by Paul Jackson. | |
20 | * --------------------------------------------------- | |
21 | * | |
22 | * This file is subject to the terms and conditions of the GNU General Public | |
23 | * License. See the file COPYING in the main directory of the Linux | |
24 | * distribution for more details. | |
25 | */ | |
26 | ||
27 | #include <linux/cgroup.h> | |
28 | #include <linux/errno.h> | |
29 | #include <linux/fs.h> | |
30 | #include <linux/kernel.h> | |
31 | #include <linux/list.h> | |
32 | #include <linux/mm.h> | |
33 | #include <linux/mutex.h> | |
34 | #include <linux/mount.h> | |
35 | #include <linux/pagemap.h> | |
a424316c | 36 | #include <linux/proc_fs.h> |
ddbcc7e8 PM |
37 | #include <linux/rcupdate.h> |
38 | #include <linux/sched.h> | |
39 | #include <linux/seq_file.h> | |
40 | #include <linux/slab.h> | |
41 | #include <linux/magic.h> | |
42 | #include <linux/spinlock.h> | |
43 | #include <linux/string.h> | |
bbcb81d0 | 44 | #include <linux/sort.h> |
ddbcc7e8 PM |
45 | #include <asm/atomic.h> |
46 | ||
47 | /* Generate an array of cgroup subsystem pointers */ | |
48 | #define SUBSYS(_x) &_x ## _subsys, | |
49 | ||
50 | static struct cgroup_subsys *subsys[] = { | |
51 | #include <linux/cgroup_subsys.h> | |
52 | }; | |
53 | ||
54 | /* | |
55 | * A cgroupfs_root represents the root of a cgroup hierarchy, | |
56 | * and may be associated with a superblock to form an active | |
57 | * hierarchy | |
58 | */ | |
59 | struct cgroupfs_root { | |
60 | struct super_block *sb; | |
61 | ||
62 | /* | |
63 | * The bitmask of subsystems intended to be attached to this | |
64 | * hierarchy | |
65 | */ | |
66 | unsigned long subsys_bits; | |
67 | ||
68 | /* The bitmask of subsystems currently attached to this hierarchy */ | |
69 | unsigned long actual_subsys_bits; | |
70 | ||
71 | /* A list running through the attached subsystems */ | |
72 | struct list_head subsys_list; | |
73 | ||
74 | /* The root cgroup for this hierarchy */ | |
75 | struct cgroup top_cgroup; | |
76 | ||
77 | /* Tracks how many cgroups are currently defined in hierarchy.*/ | |
78 | int number_of_cgroups; | |
79 | ||
80 | /* A list running through the mounted hierarchies */ | |
81 | struct list_head root_list; | |
82 | ||
83 | /* Hierarchy-specific flags */ | |
84 | unsigned long flags; | |
85 | }; | |
86 | ||
87 | ||
88 | /* | |
89 | * The "rootnode" hierarchy is the "dummy hierarchy", reserved for the | |
90 | * subsystems that are otherwise unattached - it never has more than a | |
91 | * single cgroup, and all tasks are part of that cgroup. | |
92 | */ | |
93 | static struct cgroupfs_root rootnode; | |
94 | ||
95 | /* The list of hierarchy roots */ | |
96 | ||
97 | static LIST_HEAD(roots); | |
98 | ||
99 | /* dummytop is a shorthand for the dummy hierarchy's top cgroup */ | |
100 | #define dummytop (&rootnode.top_cgroup) | |
101 | ||
102 | /* This flag indicates whether tasks in the fork and exit paths should | |
103 | * take callback_mutex and check for fork/exit handlers to call. This | |
104 | * avoids us having to do extra work in the fork/exit path if none of the | |
105 | * subsystems need to be called. | |
106 | */ | |
107 | static int need_forkexit_callback; | |
108 | ||
109 | /* bits in struct cgroup flags field */ | |
110 | enum { | |
111 | CONT_REMOVED, | |
112 | }; | |
113 | ||
114 | /* convenient tests for these bits */ | |
115 | inline int cgroup_is_removed(const struct cgroup *cont) | |
116 | { | |
117 | return test_bit(CONT_REMOVED, &cont->flags); | |
118 | } | |
119 | ||
120 | /* bits in struct cgroupfs_root flags field */ | |
121 | enum { | |
122 | ROOT_NOPREFIX, /* mounted subsystems have no named prefix */ | |
123 | }; | |
124 | ||
125 | /* | |
126 | * for_each_subsys() allows you to iterate on each subsystem attached to | |
127 | * an active hierarchy | |
128 | */ | |
129 | #define for_each_subsys(_root, _ss) \ | |
130 | list_for_each_entry(_ss, &_root->subsys_list, sibling) | |
131 | ||
132 | /* for_each_root() allows you to iterate across the active hierarchies */ | |
133 | #define for_each_root(_root) \ | |
134 | list_for_each_entry(_root, &roots, root_list) | |
135 | ||
b4f48b63 PM |
136 | /* Each task_struct has an embedded css_set, so the get/put |
137 | * operation simply takes a reference count on all the cgroups | |
138 | * referenced by subsystems in this css_set. This can end up | |
139 | * multiple-counting some cgroups, but that's OK - the ref-count is | |
140 | * just a busy/not-busy indicator; ensuring that we only count each | |
141 | * cgroup once would require taking a global lock to ensure that no | |
142 | * subsystems moved between hierarchies while we were doing so. | |
143 | * | |
144 | * Possible TODO: decide at boot time based on the number of | |
145 | * registered subsystems and the number of CPUs or NUMA nodes whether | |
146 | * it's better for performance to ref-count every subsystem, or to | |
147 | * take a global lock and only add one ref count to each hierarchy. | |
148 | */ | |
149 | static void get_css_set(struct css_set *cg) | |
150 | { | |
151 | int i; | |
152 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) | |
153 | atomic_inc(&cg->subsys[i]->cgroup->count); | |
154 | } | |
155 | ||
156 | static void put_css_set(struct css_set *cg) | |
157 | { | |
158 | int i; | |
159 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) | |
160 | atomic_dec(&cg->subsys[i]->cgroup->count); | |
161 | } | |
162 | ||
ddbcc7e8 PM |
163 | /* |
164 | * There is one global cgroup mutex. We also require taking | |
165 | * task_lock() when dereferencing a task's cgroup subsys pointers. | |
166 | * See "The task_lock() exception", at the end of this comment. | |
167 | * | |
168 | * A task must hold cgroup_mutex to modify cgroups. | |
169 | * | |
170 | * Any task can increment and decrement the count field without lock. | |
171 | * So in general, code holding cgroup_mutex can't rely on the count | |
172 | * field not changing. However, if the count goes to zero, then only | |
173 | * attach_task() can increment it again. Because a count of zero | |
174 | * means that no tasks are currently attached, therefore there is no | |
175 | * way a task attached to that cgroup can fork (the other way to | |
176 | * increment the count). So code holding cgroup_mutex can safely | |
177 | * assume that if the count is zero, it will stay zero. Similarly, if | |
178 | * a task holds cgroup_mutex on a cgroup with zero count, it | |
179 | * knows that the cgroup won't be removed, as cgroup_rmdir() | |
180 | * needs that mutex. | |
181 | * | |
182 | * The cgroup_common_file_write handler for operations that modify | |
183 | * the cgroup hierarchy holds cgroup_mutex across the entire operation, | |
184 | * single threading all such cgroup modifications across the system. | |
185 | * | |
186 | * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't | |
187 | * (usually) take cgroup_mutex. These are the two most performance | |
188 | * critical pieces of code here. The exception occurs on cgroup_exit(), | |
189 | * when a task in a notify_on_release cgroup exits. Then cgroup_mutex | |
190 | * is taken, and if the cgroup count is zero, a usermode call made | |
191 | * to /sbin/cgroup_release_agent with the name of the cgroup (path | |
192 | * relative to the root of cgroup file system) as the argument. | |
193 | * | |
194 | * A cgroup can only be deleted if both its 'count' of using tasks | |
195 | * is zero, and its list of 'children' cgroups is empty. Since all | |
196 | * tasks in the system use _some_ cgroup, and since there is always at | |
197 | * least one task in the system (init, pid == 1), therefore, top_cgroup | |
198 | * always has either children cgroups and/or using tasks. So we don't | |
199 | * need a special hack to ensure that top_cgroup cannot be deleted. | |
200 | * | |
201 | * The task_lock() exception | |
202 | * | |
203 | * The need for this exception arises from the action of | |
204 | * attach_task(), which overwrites one tasks cgroup pointer with | |
205 | * another. It does so using cgroup_mutexe, however there are | |
206 | * several performance critical places that need to reference | |
207 | * task->cgroup without the expense of grabbing a system global | |
208 | * mutex. Therefore except as noted below, when dereferencing or, as | |
209 | * in attach_task(), modifying a task'ss cgroup pointer we use | |
210 | * task_lock(), which acts on a spinlock (task->alloc_lock) already in | |
211 | * the task_struct routinely used for such matters. | |
212 | * | |
213 | * P.S. One more locking exception. RCU is used to guard the | |
214 | * update of a tasks cgroup pointer by attach_task() | |
215 | */ | |
216 | ||
217 | static DEFINE_MUTEX(cgroup_mutex); | |
218 | ||
219 | /** | |
220 | * cgroup_lock - lock out any changes to cgroup structures | |
221 | * | |
222 | */ | |
223 | ||
224 | void cgroup_lock(void) | |
225 | { | |
226 | mutex_lock(&cgroup_mutex); | |
227 | } | |
228 | ||
229 | /** | |
230 | * cgroup_unlock - release lock on cgroup changes | |
231 | * | |
232 | * Undo the lock taken in a previous cgroup_lock() call. | |
233 | */ | |
234 | ||
235 | void cgroup_unlock(void) | |
236 | { | |
237 | mutex_unlock(&cgroup_mutex); | |
238 | } | |
239 | ||
240 | /* | |
241 | * A couple of forward declarations required, due to cyclic reference loop: | |
242 | * cgroup_mkdir -> cgroup_create -> cgroup_populate_dir -> | |
243 | * cgroup_add_file -> cgroup_create_file -> cgroup_dir_inode_operations | |
244 | * -> cgroup_mkdir. | |
245 | */ | |
246 | ||
247 | static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, int mode); | |
248 | static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry); | |
249 | static int cgroup_populate_dir(struct cgroup *cont); | |
250 | static struct inode_operations cgroup_dir_inode_operations; | |
a424316c PM |
251 | static struct file_operations proc_cgroupstats_operations; |
252 | ||
253 | static struct backing_dev_info cgroup_backing_dev_info = { | |
254 | .capabilities = BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_WRITEBACK, | |
255 | }; | |
ddbcc7e8 PM |
256 | |
257 | static struct inode *cgroup_new_inode(mode_t mode, struct super_block *sb) | |
258 | { | |
259 | struct inode *inode = new_inode(sb); | |
ddbcc7e8 PM |
260 | |
261 | if (inode) { | |
262 | inode->i_mode = mode; | |
263 | inode->i_uid = current->fsuid; | |
264 | inode->i_gid = current->fsgid; | |
265 | inode->i_blocks = 0; | |
266 | inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME; | |
267 | inode->i_mapping->backing_dev_info = &cgroup_backing_dev_info; | |
268 | } | |
269 | return inode; | |
270 | } | |
271 | ||
272 | static void cgroup_diput(struct dentry *dentry, struct inode *inode) | |
273 | { | |
274 | /* is dentry a directory ? if so, kfree() associated cgroup */ | |
275 | if (S_ISDIR(inode->i_mode)) { | |
276 | struct cgroup *cont = dentry->d_fsdata; | |
277 | BUG_ON(!(cgroup_is_removed(cont))); | |
278 | kfree(cont); | |
279 | } | |
280 | iput(inode); | |
281 | } | |
282 | ||
283 | static void remove_dir(struct dentry *d) | |
284 | { | |
285 | struct dentry *parent = dget(d->d_parent); | |
286 | ||
287 | d_delete(d); | |
288 | simple_rmdir(parent->d_inode, d); | |
289 | dput(parent); | |
290 | } | |
291 | ||
292 | static void cgroup_clear_directory(struct dentry *dentry) | |
293 | { | |
294 | struct list_head *node; | |
295 | ||
296 | BUG_ON(!mutex_is_locked(&dentry->d_inode->i_mutex)); | |
297 | spin_lock(&dcache_lock); | |
298 | node = dentry->d_subdirs.next; | |
299 | while (node != &dentry->d_subdirs) { | |
300 | struct dentry *d = list_entry(node, struct dentry, d_u.d_child); | |
301 | list_del_init(node); | |
302 | if (d->d_inode) { | |
303 | /* This should never be called on a cgroup | |
304 | * directory with child cgroups */ | |
305 | BUG_ON(d->d_inode->i_mode & S_IFDIR); | |
306 | d = dget_locked(d); | |
307 | spin_unlock(&dcache_lock); | |
308 | d_delete(d); | |
309 | simple_unlink(dentry->d_inode, d); | |
310 | dput(d); | |
311 | spin_lock(&dcache_lock); | |
312 | } | |
313 | node = dentry->d_subdirs.next; | |
314 | } | |
315 | spin_unlock(&dcache_lock); | |
316 | } | |
317 | ||
318 | /* | |
319 | * NOTE : the dentry must have been dget()'ed | |
320 | */ | |
321 | static void cgroup_d_remove_dir(struct dentry *dentry) | |
322 | { | |
323 | cgroup_clear_directory(dentry); | |
324 | ||
325 | spin_lock(&dcache_lock); | |
326 | list_del_init(&dentry->d_u.d_child); | |
327 | spin_unlock(&dcache_lock); | |
328 | remove_dir(dentry); | |
329 | } | |
330 | ||
331 | static int rebind_subsystems(struct cgroupfs_root *root, | |
332 | unsigned long final_bits) | |
333 | { | |
334 | unsigned long added_bits, removed_bits; | |
335 | struct cgroup *cont = &root->top_cgroup; | |
336 | int i; | |
337 | ||
338 | removed_bits = root->actual_subsys_bits & ~final_bits; | |
339 | added_bits = final_bits & ~root->actual_subsys_bits; | |
340 | /* Check that any added subsystems are currently free */ | |
341 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { | |
342 | unsigned long long bit = 1ull << i; | |
343 | struct cgroup_subsys *ss = subsys[i]; | |
344 | if (!(bit & added_bits)) | |
345 | continue; | |
346 | if (ss->root != &rootnode) { | |
347 | /* Subsystem isn't free */ | |
348 | return -EBUSY; | |
349 | } | |
350 | } | |
351 | ||
352 | /* Currently we don't handle adding/removing subsystems when | |
353 | * any child cgroups exist. This is theoretically supportable | |
354 | * but involves complex error handling, so it's being left until | |
355 | * later */ | |
356 | if (!list_empty(&cont->children)) | |
357 | return -EBUSY; | |
358 | ||
359 | /* Process each subsystem */ | |
360 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { | |
361 | struct cgroup_subsys *ss = subsys[i]; | |
362 | unsigned long bit = 1UL << i; | |
363 | if (bit & added_bits) { | |
364 | /* We're binding this subsystem to this hierarchy */ | |
365 | BUG_ON(cont->subsys[i]); | |
366 | BUG_ON(!dummytop->subsys[i]); | |
367 | BUG_ON(dummytop->subsys[i]->cgroup != dummytop); | |
368 | cont->subsys[i] = dummytop->subsys[i]; | |
369 | cont->subsys[i]->cgroup = cont; | |
370 | list_add(&ss->sibling, &root->subsys_list); | |
371 | rcu_assign_pointer(ss->root, root); | |
372 | if (ss->bind) | |
373 | ss->bind(ss, cont); | |
374 | ||
375 | } else if (bit & removed_bits) { | |
376 | /* We're removing this subsystem */ | |
377 | BUG_ON(cont->subsys[i] != dummytop->subsys[i]); | |
378 | BUG_ON(cont->subsys[i]->cgroup != cont); | |
379 | if (ss->bind) | |
380 | ss->bind(ss, dummytop); | |
381 | dummytop->subsys[i]->cgroup = dummytop; | |
382 | cont->subsys[i] = NULL; | |
383 | rcu_assign_pointer(subsys[i]->root, &rootnode); | |
384 | list_del(&ss->sibling); | |
385 | } else if (bit & final_bits) { | |
386 | /* Subsystem state should already exist */ | |
387 | BUG_ON(!cont->subsys[i]); | |
388 | } else { | |
389 | /* Subsystem state shouldn't exist */ | |
390 | BUG_ON(cont->subsys[i]); | |
391 | } | |
392 | } | |
393 | root->subsys_bits = root->actual_subsys_bits = final_bits; | |
394 | synchronize_rcu(); | |
395 | ||
396 | return 0; | |
397 | } | |
398 | ||
399 | static int cgroup_show_options(struct seq_file *seq, struct vfsmount *vfs) | |
400 | { | |
401 | struct cgroupfs_root *root = vfs->mnt_sb->s_fs_info; | |
402 | struct cgroup_subsys *ss; | |
403 | ||
404 | mutex_lock(&cgroup_mutex); | |
405 | for_each_subsys(root, ss) | |
406 | seq_printf(seq, ",%s", ss->name); | |
407 | if (test_bit(ROOT_NOPREFIX, &root->flags)) | |
408 | seq_puts(seq, ",noprefix"); | |
409 | mutex_unlock(&cgroup_mutex); | |
410 | return 0; | |
411 | } | |
412 | ||
413 | struct cgroup_sb_opts { | |
414 | unsigned long subsys_bits; | |
415 | unsigned long flags; | |
416 | }; | |
417 | ||
418 | /* Convert a hierarchy specifier into a bitmask of subsystems and | |
419 | * flags. */ | |
420 | static int parse_cgroupfs_options(char *data, | |
421 | struct cgroup_sb_opts *opts) | |
422 | { | |
423 | char *token, *o = data ?: "all"; | |
424 | ||
425 | opts->subsys_bits = 0; | |
426 | opts->flags = 0; | |
427 | ||
428 | while ((token = strsep(&o, ",")) != NULL) { | |
429 | if (!*token) | |
430 | return -EINVAL; | |
431 | if (!strcmp(token, "all")) { | |
432 | opts->subsys_bits = (1 << CGROUP_SUBSYS_COUNT) - 1; | |
433 | } else if (!strcmp(token, "noprefix")) { | |
434 | set_bit(ROOT_NOPREFIX, &opts->flags); | |
435 | } else { | |
436 | struct cgroup_subsys *ss; | |
437 | int i; | |
438 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { | |
439 | ss = subsys[i]; | |
440 | if (!strcmp(token, ss->name)) { | |
441 | set_bit(i, &opts->subsys_bits); | |
442 | break; | |
443 | } | |
444 | } | |
445 | if (i == CGROUP_SUBSYS_COUNT) | |
446 | return -ENOENT; | |
447 | } | |
448 | } | |
449 | ||
450 | /* We can't have an empty hierarchy */ | |
451 | if (!opts->subsys_bits) | |
452 | return -EINVAL; | |
453 | ||
454 | return 0; | |
455 | } | |
456 | ||
457 | static int cgroup_remount(struct super_block *sb, int *flags, char *data) | |
458 | { | |
459 | int ret = 0; | |
460 | struct cgroupfs_root *root = sb->s_fs_info; | |
461 | struct cgroup *cont = &root->top_cgroup; | |
462 | struct cgroup_sb_opts opts; | |
463 | ||
464 | mutex_lock(&cont->dentry->d_inode->i_mutex); | |
465 | mutex_lock(&cgroup_mutex); | |
466 | ||
467 | /* See what subsystems are wanted */ | |
468 | ret = parse_cgroupfs_options(data, &opts); | |
469 | if (ret) | |
470 | goto out_unlock; | |
471 | ||
472 | /* Don't allow flags to change at remount */ | |
473 | if (opts.flags != root->flags) { | |
474 | ret = -EINVAL; | |
475 | goto out_unlock; | |
476 | } | |
477 | ||
478 | ret = rebind_subsystems(root, opts.subsys_bits); | |
479 | ||
480 | /* (re)populate subsystem files */ | |
481 | if (!ret) | |
482 | cgroup_populate_dir(cont); | |
483 | ||
484 | out_unlock: | |
485 | mutex_unlock(&cgroup_mutex); | |
486 | mutex_unlock(&cont->dentry->d_inode->i_mutex); | |
487 | return ret; | |
488 | } | |
489 | ||
490 | static struct super_operations cgroup_ops = { | |
491 | .statfs = simple_statfs, | |
492 | .drop_inode = generic_delete_inode, | |
493 | .show_options = cgroup_show_options, | |
494 | .remount_fs = cgroup_remount, | |
495 | }; | |
496 | ||
497 | static void init_cgroup_root(struct cgroupfs_root *root) | |
498 | { | |
499 | struct cgroup *cont = &root->top_cgroup; | |
500 | INIT_LIST_HEAD(&root->subsys_list); | |
501 | INIT_LIST_HEAD(&root->root_list); | |
502 | root->number_of_cgroups = 1; | |
503 | cont->root = root; | |
504 | cont->top_cgroup = cont; | |
505 | INIT_LIST_HEAD(&cont->sibling); | |
506 | INIT_LIST_HEAD(&cont->children); | |
507 | } | |
508 | ||
509 | static int cgroup_test_super(struct super_block *sb, void *data) | |
510 | { | |
511 | struct cgroupfs_root *new = data; | |
512 | struct cgroupfs_root *root = sb->s_fs_info; | |
513 | ||
514 | /* First check subsystems */ | |
515 | if (new->subsys_bits != root->subsys_bits) | |
516 | return 0; | |
517 | ||
518 | /* Next check flags */ | |
519 | if (new->flags != root->flags) | |
520 | return 0; | |
521 | ||
522 | return 1; | |
523 | } | |
524 | ||
525 | static int cgroup_set_super(struct super_block *sb, void *data) | |
526 | { | |
527 | int ret; | |
528 | struct cgroupfs_root *root = data; | |
529 | ||
530 | ret = set_anon_super(sb, NULL); | |
531 | if (ret) | |
532 | return ret; | |
533 | ||
534 | sb->s_fs_info = root; | |
535 | root->sb = sb; | |
536 | ||
537 | sb->s_blocksize = PAGE_CACHE_SIZE; | |
538 | sb->s_blocksize_bits = PAGE_CACHE_SHIFT; | |
539 | sb->s_magic = CGROUP_SUPER_MAGIC; | |
540 | sb->s_op = &cgroup_ops; | |
541 | ||
542 | return 0; | |
543 | } | |
544 | ||
545 | static int cgroup_get_rootdir(struct super_block *sb) | |
546 | { | |
547 | struct inode *inode = | |
548 | cgroup_new_inode(S_IFDIR | S_IRUGO | S_IXUGO | S_IWUSR, sb); | |
549 | struct dentry *dentry; | |
550 | ||
551 | if (!inode) | |
552 | return -ENOMEM; | |
553 | ||
554 | inode->i_op = &simple_dir_inode_operations; | |
555 | inode->i_fop = &simple_dir_operations; | |
556 | inode->i_op = &cgroup_dir_inode_operations; | |
557 | /* directories start off with i_nlink == 2 (for "." entry) */ | |
558 | inc_nlink(inode); | |
559 | dentry = d_alloc_root(inode); | |
560 | if (!dentry) { | |
561 | iput(inode); | |
562 | return -ENOMEM; | |
563 | } | |
564 | sb->s_root = dentry; | |
565 | return 0; | |
566 | } | |
567 | ||
568 | static int cgroup_get_sb(struct file_system_type *fs_type, | |
569 | int flags, const char *unused_dev_name, | |
570 | void *data, struct vfsmount *mnt) | |
571 | { | |
572 | struct cgroup_sb_opts opts; | |
573 | int ret = 0; | |
574 | struct super_block *sb; | |
575 | struct cgroupfs_root *root; | |
576 | ||
577 | /* First find the desired set of subsystems */ | |
578 | ret = parse_cgroupfs_options(data, &opts); | |
579 | if (ret) | |
580 | return ret; | |
581 | ||
582 | root = kzalloc(sizeof(*root), GFP_KERNEL); | |
583 | if (!root) | |
584 | return -ENOMEM; | |
585 | ||
586 | init_cgroup_root(root); | |
587 | root->subsys_bits = opts.subsys_bits; | |
588 | root->flags = opts.flags; | |
589 | ||
590 | sb = sget(fs_type, cgroup_test_super, cgroup_set_super, root); | |
591 | ||
592 | if (IS_ERR(sb)) { | |
593 | kfree(root); | |
594 | return PTR_ERR(sb); | |
595 | } | |
596 | ||
597 | if (sb->s_fs_info != root) { | |
598 | /* Reusing an existing superblock */ | |
599 | BUG_ON(sb->s_root == NULL); | |
600 | kfree(root); | |
601 | root = NULL; | |
602 | } else { | |
603 | /* New superblock */ | |
604 | struct cgroup *cont = &root->top_cgroup; | |
605 | ||
606 | BUG_ON(sb->s_root != NULL); | |
607 | ||
608 | ret = cgroup_get_rootdir(sb); | |
609 | if (ret) | |
610 | goto drop_new_super; | |
611 | ||
612 | mutex_lock(&cgroup_mutex); | |
613 | ||
614 | ret = rebind_subsystems(root, root->subsys_bits); | |
615 | if (ret == -EBUSY) { | |
616 | mutex_unlock(&cgroup_mutex); | |
617 | goto drop_new_super; | |
618 | } | |
619 | ||
620 | /* EBUSY should be the only error here */ | |
621 | BUG_ON(ret); | |
622 | ||
623 | list_add(&root->root_list, &roots); | |
624 | ||
625 | sb->s_root->d_fsdata = &root->top_cgroup; | |
626 | root->top_cgroup.dentry = sb->s_root; | |
627 | ||
628 | BUG_ON(!list_empty(&cont->sibling)); | |
629 | BUG_ON(!list_empty(&cont->children)); | |
630 | BUG_ON(root->number_of_cgroups != 1); | |
631 | ||
632 | /* | |
633 | * I believe that it's safe to nest i_mutex inside | |
634 | * cgroup_mutex in this case, since no-one else can | |
635 | * be accessing this directory yet. But we still need | |
636 | * to teach lockdep that this is the case - currently | |
637 | * a cgroupfs remount triggers a lockdep warning | |
638 | */ | |
639 | mutex_lock(&cont->dentry->d_inode->i_mutex); | |
640 | cgroup_populate_dir(cont); | |
641 | mutex_unlock(&cont->dentry->d_inode->i_mutex); | |
642 | mutex_unlock(&cgroup_mutex); | |
643 | } | |
644 | ||
645 | return simple_set_mnt(mnt, sb); | |
646 | ||
647 | drop_new_super: | |
648 | up_write(&sb->s_umount); | |
649 | deactivate_super(sb); | |
650 | return ret; | |
651 | } | |
652 | ||
653 | static void cgroup_kill_sb(struct super_block *sb) { | |
654 | struct cgroupfs_root *root = sb->s_fs_info; | |
655 | struct cgroup *cont = &root->top_cgroup; | |
656 | int ret; | |
657 | ||
658 | BUG_ON(!root); | |
659 | ||
660 | BUG_ON(root->number_of_cgroups != 1); | |
661 | BUG_ON(!list_empty(&cont->children)); | |
662 | BUG_ON(!list_empty(&cont->sibling)); | |
663 | ||
664 | mutex_lock(&cgroup_mutex); | |
665 | ||
666 | /* Rebind all subsystems back to the default hierarchy */ | |
667 | ret = rebind_subsystems(root, 0); | |
668 | /* Shouldn't be able to fail ... */ | |
669 | BUG_ON(ret); | |
670 | ||
671 | if (!list_empty(&root->root_list)) | |
672 | list_del(&root->root_list); | |
673 | mutex_unlock(&cgroup_mutex); | |
674 | ||
675 | kfree(root); | |
676 | kill_litter_super(sb); | |
677 | } | |
678 | ||
679 | static struct file_system_type cgroup_fs_type = { | |
680 | .name = "cgroup", | |
681 | .get_sb = cgroup_get_sb, | |
682 | .kill_sb = cgroup_kill_sb, | |
683 | }; | |
684 | ||
685 | static inline struct cgroup *__d_cont(struct dentry *dentry) | |
686 | { | |
687 | return dentry->d_fsdata; | |
688 | } | |
689 | ||
690 | static inline struct cftype *__d_cft(struct dentry *dentry) | |
691 | { | |
692 | return dentry->d_fsdata; | |
693 | } | |
694 | ||
695 | /* | |
696 | * Called with cgroup_mutex held. Writes path of cgroup into buf. | |
697 | * Returns 0 on success, -errno on error. | |
698 | */ | |
699 | int cgroup_path(const struct cgroup *cont, char *buf, int buflen) | |
700 | { | |
701 | char *start; | |
702 | ||
703 | if (cont == dummytop) { | |
704 | /* | |
705 | * Inactive subsystems have no dentry for their root | |
706 | * cgroup | |
707 | */ | |
708 | strcpy(buf, "/"); | |
709 | return 0; | |
710 | } | |
711 | ||
712 | start = buf + buflen; | |
713 | ||
714 | *--start = '\0'; | |
715 | for (;;) { | |
716 | int len = cont->dentry->d_name.len; | |
717 | if ((start -= len) < buf) | |
718 | return -ENAMETOOLONG; | |
719 | memcpy(start, cont->dentry->d_name.name, len); | |
720 | cont = cont->parent; | |
721 | if (!cont) | |
722 | break; | |
723 | if (!cont->parent) | |
724 | continue; | |
725 | if (--start < buf) | |
726 | return -ENAMETOOLONG; | |
727 | *start = '/'; | |
728 | } | |
729 | memmove(buf, start, buf + buflen - start); | |
730 | return 0; | |
731 | } | |
732 | ||
bbcb81d0 PM |
733 | /* |
734 | * Return the first subsystem attached to a cgroup's hierarchy, and | |
735 | * its subsystem id. | |
736 | */ | |
737 | ||
738 | static void get_first_subsys(const struct cgroup *cont, | |
739 | struct cgroup_subsys_state **css, int *subsys_id) | |
740 | { | |
741 | const struct cgroupfs_root *root = cont->root; | |
742 | const struct cgroup_subsys *test_ss; | |
743 | BUG_ON(list_empty(&root->subsys_list)); | |
744 | test_ss = list_entry(root->subsys_list.next, | |
745 | struct cgroup_subsys, sibling); | |
746 | if (css) { | |
747 | *css = cont->subsys[test_ss->subsys_id]; | |
748 | BUG_ON(!*css); | |
749 | } | |
750 | if (subsys_id) | |
751 | *subsys_id = test_ss->subsys_id; | |
752 | } | |
753 | ||
754 | /* | |
755 | * Attach task 'tsk' to cgroup 'cont' | |
756 | * | |
757 | * Call holding cgroup_mutex. May take task_lock of | |
758 | * the task 'pid' during call. | |
759 | */ | |
760 | static int attach_task(struct cgroup *cont, struct task_struct *tsk) | |
761 | { | |
762 | int retval = 0; | |
763 | struct cgroup_subsys *ss; | |
764 | struct cgroup *oldcont; | |
765 | struct css_set *cg = &tsk->cgroups; | |
766 | struct cgroupfs_root *root = cont->root; | |
767 | int i; | |
768 | int subsys_id; | |
769 | ||
770 | get_first_subsys(cont, NULL, &subsys_id); | |
771 | ||
772 | /* Nothing to do if the task is already in that cgroup */ | |
773 | oldcont = task_cgroup(tsk, subsys_id); | |
774 | if (cont == oldcont) | |
775 | return 0; | |
776 | ||
777 | for_each_subsys(root, ss) { | |
778 | if (ss->can_attach) { | |
779 | retval = ss->can_attach(ss, cont, tsk); | |
780 | if (retval) { | |
781 | return retval; | |
782 | } | |
783 | } | |
784 | } | |
785 | ||
786 | task_lock(tsk); | |
787 | if (tsk->flags & PF_EXITING) { | |
788 | task_unlock(tsk); | |
789 | return -ESRCH; | |
790 | } | |
791 | /* Update the css_set pointers for the subsystems in this | |
792 | * hierarchy */ | |
793 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { | |
794 | if (root->subsys_bits & (1ull << i)) { | |
795 | /* Subsystem is in this hierarchy. So we want | |
796 | * the subsystem state from the new | |
797 | * cgroup. Transfer the refcount from the | |
798 | * old to the new */ | |
799 | atomic_inc(&cont->count); | |
800 | atomic_dec(&cg->subsys[i]->cgroup->count); | |
801 | rcu_assign_pointer(cg->subsys[i], cont->subsys[i]); | |
802 | } | |
803 | } | |
804 | task_unlock(tsk); | |
805 | ||
806 | for_each_subsys(root, ss) { | |
807 | if (ss->attach) { | |
808 | ss->attach(ss, cont, oldcont, tsk); | |
809 | } | |
810 | } | |
811 | ||
812 | synchronize_rcu(); | |
813 | return 0; | |
814 | } | |
815 | ||
816 | /* | |
817 | * Attach task with pid 'pid' to cgroup 'cont'. Call with | |
818 | * cgroup_mutex, may take task_lock of task | |
819 | */ | |
820 | static int attach_task_by_pid(struct cgroup *cont, char *pidbuf) | |
821 | { | |
822 | pid_t pid; | |
823 | struct task_struct *tsk; | |
824 | int ret; | |
825 | ||
826 | if (sscanf(pidbuf, "%d", &pid) != 1) | |
827 | return -EIO; | |
828 | ||
829 | if (pid) { | |
830 | rcu_read_lock(); | |
831 | tsk = find_task_by_pid(pid); | |
832 | if (!tsk || tsk->flags & PF_EXITING) { | |
833 | rcu_read_unlock(); | |
834 | return -ESRCH; | |
835 | } | |
836 | get_task_struct(tsk); | |
837 | rcu_read_unlock(); | |
838 | ||
839 | if ((current->euid) && (current->euid != tsk->uid) | |
840 | && (current->euid != tsk->suid)) { | |
841 | put_task_struct(tsk); | |
842 | return -EACCES; | |
843 | } | |
844 | } else { | |
845 | tsk = current; | |
846 | get_task_struct(tsk); | |
847 | } | |
848 | ||
849 | ret = attach_task(cont, tsk); | |
850 | put_task_struct(tsk); | |
851 | return ret; | |
852 | } | |
853 | ||
ddbcc7e8 PM |
854 | /* The various types of files and directories in a cgroup file system */ |
855 | ||
856 | enum cgroup_filetype { | |
857 | FILE_ROOT, | |
858 | FILE_DIR, | |
859 | FILE_TASKLIST, | |
860 | }; | |
861 | ||
355e0c48 PM |
862 | static ssize_t cgroup_write_uint(struct cgroup *cont, struct cftype *cft, |
863 | struct file *file, | |
864 | const char __user *userbuf, | |
865 | size_t nbytes, loff_t *unused_ppos) | |
866 | { | |
867 | char buffer[64]; | |
868 | int retval = 0; | |
869 | u64 val; | |
870 | char *end; | |
871 | ||
872 | if (!nbytes) | |
873 | return -EINVAL; | |
874 | if (nbytes >= sizeof(buffer)) | |
875 | return -E2BIG; | |
876 | if (copy_from_user(buffer, userbuf, nbytes)) | |
877 | return -EFAULT; | |
878 | ||
879 | buffer[nbytes] = 0; /* nul-terminate */ | |
880 | ||
881 | /* strip newline if necessary */ | |
882 | if (nbytes && (buffer[nbytes-1] == '\n')) | |
883 | buffer[nbytes-1] = 0; | |
884 | val = simple_strtoull(buffer, &end, 0); | |
885 | if (*end) | |
886 | return -EINVAL; | |
887 | ||
888 | /* Pass to subsystem */ | |
889 | retval = cft->write_uint(cont, cft, val); | |
890 | if (!retval) | |
891 | retval = nbytes; | |
892 | return retval; | |
893 | } | |
894 | ||
bbcb81d0 PM |
895 | static ssize_t cgroup_common_file_write(struct cgroup *cont, |
896 | struct cftype *cft, | |
897 | struct file *file, | |
898 | const char __user *userbuf, | |
899 | size_t nbytes, loff_t *unused_ppos) | |
900 | { | |
901 | enum cgroup_filetype type = cft->private; | |
902 | char *buffer; | |
903 | int retval = 0; | |
904 | ||
905 | if (nbytes >= PATH_MAX) | |
906 | return -E2BIG; | |
907 | ||
908 | /* +1 for nul-terminator */ | |
909 | buffer = kmalloc(nbytes + 1, GFP_KERNEL); | |
910 | if (buffer == NULL) | |
911 | return -ENOMEM; | |
912 | ||
913 | if (copy_from_user(buffer, userbuf, nbytes)) { | |
914 | retval = -EFAULT; | |
915 | goto out1; | |
916 | } | |
917 | buffer[nbytes] = 0; /* nul-terminate */ | |
918 | ||
919 | mutex_lock(&cgroup_mutex); | |
920 | ||
921 | if (cgroup_is_removed(cont)) { | |
922 | retval = -ENODEV; | |
923 | goto out2; | |
924 | } | |
925 | ||
926 | switch (type) { | |
927 | case FILE_TASKLIST: | |
928 | retval = attach_task_by_pid(cont, buffer); | |
929 | break; | |
930 | default: | |
931 | retval = -EINVAL; | |
932 | goto out2; | |
933 | } | |
934 | ||
935 | if (retval == 0) | |
936 | retval = nbytes; | |
937 | out2: | |
938 | mutex_unlock(&cgroup_mutex); | |
939 | out1: | |
940 | kfree(buffer); | |
941 | return retval; | |
942 | } | |
943 | ||
ddbcc7e8 PM |
944 | static ssize_t cgroup_file_write(struct file *file, const char __user *buf, |
945 | size_t nbytes, loff_t *ppos) | |
946 | { | |
947 | struct cftype *cft = __d_cft(file->f_dentry); | |
948 | struct cgroup *cont = __d_cont(file->f_dentry->d_parent); | |
949 | ||
950 | if (!cft) | |
951 | return -ENODEV; | |
355e0c48 PM |
952 | if (cft->write) |
953 | return cft->write(cont, cft, file, buf, nbytes, ppos); | |
954 | if (cft->write_uint) | |
955 | return cgroup_write_uint(cont, cft, file, buf, nbytes, ppos); | |
956 | return -EINVAL; | |
ddbcc7e8 PM |
957 | } |
958 | ||
959 | static ssize_t cgroup_read_uint(struct cgroup *cont, struct cftype *cft, | |
960 | struct file *file, | |
961 | char __user *buf, size_t nbytes, | |
962 | loff_t *ppos) | |
963 | { | |
964 | char tmp[64]; | |
965 | u64 val = cft->read_uint(cont, cft); | |
966 | int len = sprintf(tmp, "%llu\n", (unsigned long long) val); | |
967 | ||
968 | return simple_read_from_buffer(buf, nbytes, ppos, tmp, len); | |
969 | } | |
970 | ||
971 | static ssize_t cgroup_file_read(struct file *file, char __user *buf, | |
972 | size_t nbytes, loff_t *ppos) | |
973 | { | |
974 | struct cftype *cft = __d_cft(file->f_dentry); | |
975 | struct cgroup *cont = __d_cont(file->f_dentry->d_parent); | |
976 | ||
977 | if (!cft) | |
978 | return -ENODEV; | |
979 | ||
980 | if (cft->read) | |
981 | return cft->read(cont, cft, file, buf, nbytes, ppos); | |
982 | if (cft->read_uint) | |
983 | return cgroup_read_uint(cont, cft, file, buf, nbytes, ppos); | |
984 | return -EINVAL; | |
985 | } | |
986 | ||
987 | static int cgroup_file_open(struct inode *inode, struct file *file) | |
988 | { | |
989 | int err; | |
990 | struct cftype *cft; | |
991 | ||
992 | err = generic_file_open(inode, file); | |
993 | if (err) | |
994 | return err; | |
995 | ||
996 | cft = __d_cft(file->f_dentry); | |
997 | if (!cft) | |
998 | return -ENODEV; | |
999 | if (cft->open) | |
1000 | err = cft->open(inode, file); | |
1001 | else | |
1002 | err = 0; | |
1003 | ||
1004 | return err; | |
1005 | } | |
1006 | ||
1007 | static int cgroup_file_release(struct inode *inode, struct file *file) | |
1008 | { | |
1009 | struct cftype *cft = __d_cft(file->f_dentry); | |
1010 | if (cft->release) | |
1011 | return cft->release(inode, file); | |
1012 | return 0; | |
1013 | } | |
1014 | ||
1015 | /* | |
1016 | * cgroup_rename - Only allow simple rename of directories in place. | |
1017 | */ | |
1018 | static int cgroup_rename(struct inode *old_dir, struct dentry *old_dentry, | |
1019 | struct inode *new_dir, struct dentry *new_dentry) | |
1020 | { | |
1021 | if (!S_ISDIR(old_dentry->d_inode->i_mode)) | |
1022 | return -ENOTDIR; | |
1023 | if (new_dentry->d_inode) | |
1024 | return -EEXIST; | |
1025 | if (old_dir != new_dir) | |
1026 | return -EIO; | |
1027 | return simple_rename(old_dir, old_dentry, new_dir, new_dentry); | |
1028 | } | |
1029 | ||
1030 | static struct file_operations cgroup_file_operations = { | |
1031 | .read = cgroup_file_read, | |
1032 | .write = cgroup_file_write, | |
1033 | .llseek = generic_file_llseek, | |
1034 | .open = cgroup_file_open, | |
1035 | .release = cgroup_file_release, | |
1036 | }; | |
1037 | ||
1038 | static struct inode_operations cgroup_dir_inode_operations = { | |
1039 | .lookup = simple_lookup, | |
1040 | .mkdir = cgroup_mkdir, | |
1041 | .rmdir = cgroup_rmdir, | |
1042 | .rename = cgroup_rename, | |
1043 | }; | |
1044 | ||
1045 | static int cgroup_create_file(struct dentry *dentry, int mode, | |
1046 | struct super_block *sb) | |
1047 | { | |
1048 | static struct dentry_operations cgroup_dops = { | |
1049 | .d_iput = cgroup_diput, | |
1050 | }; | |
1051 | ||
1052 | struct inode *inode; | |
1053 | ||
1054 | if (!dentry) | |
1055 | return -ENOENT; | |
1056 | if (dentry->d_inode) | |
1057 | return -EEXIST; | |
1058 | ||
1059 | inode = cgroup_new_inode(mode, sb); | |
1060 | if (!inode) | |
1061 | return -ENOMEM; | |
1062 | ||
1063 | if (S_ISDIR(mode)) { | |
1064 | inode->i_op = &cgroup_dir_inode_operations; | |
1065 | inode->i_fop = &simple_dir_operations; | |
1066 | ||
1067 | /* start off with i_nlink == 2 (for "." entry) */ | |
1068 | inc_nlink(inode); | |
1069 | ||
1070 | /* start with the directory inode held, so that we can | |
1071 | * populate it without racing with another mkdir */ | |
1072 | mutex_lock(&inode->i_mutex); | |
1073 | } else if (S_ISREG(mode)) { | |
1074 | inode->i_size = 0; | |
1075 | inode->i_fop = &cgroup_file_operations; | |
1076 | } | |
1077 | dentry->d_op = &cgroup_dops; | |
1078 | d_instantiate(dentry, inode); | |
1079 | dget(dentry); /* Extra count - pin the dentry in core */ | |
1080 | return 0; | |
1081 | } | |
1082 | ||
1083 | /* | |
1084 | * cgroup_create_dir - create a directory for an object. | |
1085 | * cont: the cgroup we create the directory for. | |
1086 | * It must have a valid ->parent field | |
1087 | * And we are going to fill its ->dentry field. | |
1088 | * dentry: dentry of the new container | |
1089 | * mode: mode to set on new directory. | |
1090 | */ | |
1091 | static int cgroup_create_dir(struct cgroup *cont, struct dentry *dentry, | |
1092 | int mode) | |
1093 | { | |
1094 | struct dentry *parent; | |
1095 | int error = 0; | |
1096 | ||
1097 | parent = cont->parent->dentry; | |
1098 | error = cgroup_create_file(dentry, S_IFDIR | mode, cont->root->sb); | |
1099 | if (!error) { | |
1100 | dentry->d_fsdata = cont; | |
1101 | inc_nlink(parent->d_inode); | |
1102 | cont->dentry = dentry; | |
1103 | dget(dentry); | |
1104 | } | |
1105 | dput(dentry); | |
1106 | ||
1107 | return error; | |
1108 | } | |
1109 | ||
1110 | int cgroup_add_file(struct cgroup *cont, | |
1111 | struct cgroup_subsys *subsys, | |
1112 | const struct cftype *cft) | |
1113 | { | |
1114 | struct dentry *dir = cont->dentry; | |
1115 | struct dentry *dentry; | |
1116 | int error; | |
1117 | ||
1118 | char name[MAX_CGROUP_TYPE_NAMELEN + MAX_CFTYPE_NAME + 2] = { 0 }; | |
1119 | if (subsys && !test_bit(ROOT_NOPREFIX, &cont->root->flags)) { | |
1120 | strcpy(name, subsys->name); | |
1121 | strcat(name, "."); | |
1122 | } | |
1123 | strcat(name, cft->name); | |
1124 | BUG_ON(!mutex_is_locked(&dir->d_inode->i_mutex)); | |
1125 | dentry = lookup_one_len(name, dir, strlen(name)); | |
1126 | if (!IS_ERR(dentry)) { | |
1127 | error = cgroup_create_file(dentry, 0644 | S_IFREG, | |
1128 | cont->root->sb); | |
1129 | if (!error) | |
1130 | dentry->d_fsdata = (void *)cft; | |
1131 | dput(dentry); | |
1132 | } else | |
1133 | error = PTR_ERR(dentry); | |
1134 | return error; | |
1135 | } | |
1136 | ||
1137 | int cgroup_add_files(struct cgroup *cont, | |
1138 | struct cgroup_subsys *subsys, | |
1139 | const struct cftype cft[], | |
1140 | int count) | |
1141 | { | |
1142 | int i, err; | |
1143 | for (i = 0; i < count; i++) { | |
1144 | err = cgroup_add_file(cont, subsys, &cft[i]); | |
1145 | if (err) | |
1146 | return err; | |
1147 | } | |
1148 | return 0; | |
1149 | } | |
1150 | ||
bbcb81d0 PM |
1151 | /* Count the number of tasks in a cgroup. Could be made more |
1152 | * time-efficient but less space-efficient with more linked lists | |
1153 | * running through each cgroup and the css_set structures that | |
1154 | * referenced it. Must be called with tasklist_lock held for read or | |
1155 | * write or in an rcu critical section. | |
1156 | */ | |
1157 | int __cgroup_task_count(const struct cgroup *cont) | |
1158 | { | |
1159 | int count = 0; | |
1160 | struct task_struct *g, *p; | |
1161 | struct cgroup_subsys_state *css; | |
1162 | int subsys_id; | |
1163 | ||
1164 | get_first_subsys(cont, &css, &subsys_id); | |
1165 | do_each_thread(g, p) { | |
1166 | if (task_subsys_state(p, subsys_id) == css) | |
1167 | count ++; | |
1168 | } while_each_thread(g, p); | |
1169 | return count; | |
1170 | } | |
1171 | ||
1172 | /* | |
1173 | * Stuff for reading the 'tasks' file. | |
1174 | * | |
1175 | * Reading this file can return large amounts of data if a cgroup has | |
1176 | * *lots* of attached tasks. So it may need several calls to read(), | |
1177 | * but we cannot guarantee that the information we produce is correct | |
1178 | * unless we produce it entirely atomically. | |
1179 | * | |
1180 | * Upon tasks file open(), a struct ctr_struct is allocated, that | |
1181 | * will have a pointer to an array (also allocated here). The struct | |
1182 | * ctr_struct * is stored in file->private_data. Its resources will | |
1183 | * be freed by release() when the file is closed. The array is used | |
1184 | * to sprintf the PIDs and then used by read(). | |
1185 | */ | |
1186 | struct ctr_struct { | |
1187 | char *buf; | |
1188 | int bufsz; | |
1189 | }; | |
1190 | ||
1191 | /* | |
1192 | * Load into 'pidarray' up to 'npids' of the tasks using cgroup | |
1193 | * 'cont'. Return actual number of pids loaded. No need to | |
1194 | * task_lock(p) when reading out p->cgroup, since we're in an RCU | |
1195 | * read section, so the css_set can't go away, and is | |
1196 | * immutable after creation. | |
1197 | */ | |
1198 | static int pid_array_load(pid_t *pidarray, int npids, struct cgroup *cont) | |
1199 | { | |
1200 | int n = 0; | |
1201 | struct task_struct *g, *p; | |
1202 | struct cgroup_subsys_state *css; | |
1203 | int subsys_id; | |
1204 | ||
1205 | get_first_subsys(cont, &css, &subsys_id); | |
1206 | rcu_read_lock(); | |
1207 | do_each_thread(g, p) { | |
1208 | if (task_subsys_state(p, subsys_id) == css) { | |
1209 | pidarray[n++] = pid_nr(task_pid(p)); | |
1210 | if (unlikely(n == npids)) | |
1211 | goto array_full; | |
1212 | } | |
1213 | } while_each_thread(g, p); | |
1214 | ||
1215 | array_full: | |
1216 | rcu_read_unlock(); | |
1217 | return n; | |
1218 | } | |
1219 | ||
1220 | static int cmppid(const void *a, const void *b) | |
1221 | { | |
1222 | return *(pid_t *)a - *(pid_t *)b; | |
1223 | } | |
1224 | ||
1225 | /* | |
1226 | * Convert array 'a' of 'npids' pid_t's to a string of newline separated | |
1227 | * decimal pids in 'buf'. Don't write more than 'sz' chars, but return | |
1228 | * count 'cnt' of how many chars would be written if buf were large enough. | |
1229 | */ | |
1230 | static int pid_array_to_buf(char *buf, int sz, pid_t *a, int npids) | |
1231 | { | |
1232 | int cnt = 0; | |
1233 | int i; | |
1234 | ||
1235 | for (i = 0; i < npids; i++) | |
1236 | cnt += snprintf(buf + cnt, max(sz - cnt, 0), "%d\n", a[i]); | |
1237 | return cnt; | |
1238 | } | |
1239 | ||
1240 | /* | |
1241 | * Handle an open on 'tasks' file. Prepare a buffer listing the | |
1242 | * process id's of tasks currently attached to the cgroup being opened. | |
1243 | * | |
1244 | * Does not require any specific cgroup mutexes, and does not take any. | |
1245 | */ | |
1246 | static int cgroup_tasks_open(struct inode *unused, struct file *file) | |
1247 | { | |
1248 | struct cgroup *cont = __d_cont(file->f_dentry->d_parent); | |
1249 | struct ctr_struct *ctr; | |
1250 | pid_t *pidarray; | |
1251 | int npids; | |
1252 | char c; | |
1253 | ||
1254 | if (!(file->f_mode & FMODE_READ)) | |
1255 | return 0; | |
1256 | ||
1257 | ctr = kmalloc(sizeof(*ctr), GFP_KERNEL); | |
1258 | if (!ctr) | |
1259 | goto err0; | |
1260 | ||
1261 | /* | |
1262 | * If cgroup gets more users after we read count, we won't have | |
1263 | * enough space - tough. This race is indistinguishable to the | |
1264 | * caller from the case that the additional cgroup users didn't | |
1265 | * show up until sometime later on. | |
1266 | */ | |
1267 | npids = cgroup_task_count(cont); | |
1268 | if (npids) { | |
1269 | pidarray = kmalloc(npids * sizeof(pid_t), GFP_KERNEL); | |
1270 | if (!pidarray) | |
1271 | goto err1; | |
1272 | ||
1273 | npids = pid_array_load(pidarray, npids, cont); | |
1274 | sort(pidarray, npids, sizeof(pid_t), cmppid, NULL); | |
1275 | ||
1276 | /* Call pid_array_to_buf() twice, first just to get bufsz */ | |
1277 | ctr->bufsz = pid_array_to_buf(&c, sizeof(c), pidarray, npids) + 1; | |
1278 | ctr->buf = kmalloc(ctr->bufsz, GFP_KERNEL); | |
1279 | if (!ctr->buf) | |
1280 | goto err2; | |
1281 | ctr->bufsz = pid_array_to_buf(ctr->buf, ctr->bufsz, pidarray, npids); | |
1282 | ||
1283 | kfree(pidarray); | |
1284 | } else { | |
1285 | ctr->buf = 0; | |
1286 | ctr->bufsz = 0; | |
1287 | } | |
1288 | file->private_data = ctr; | |
1289 | return 0; | |
1290 | ||
1291 | err2: | |
1292 | kfree(pidarray); | |
1293 | err1: | |
1294 | kfree(ctr); | |
1295 | err0: | |
1296 | return -ENOMEM; | |
1297 | } | |
1298 | ||
1299 | static ssize_t cgroup_tasks_read(struct cgroup *cont, | |
1300 | struct cftype *cft, | |
1301 | struct file *file, char __user *buf, | |
1302 | size_t nbytes, loff_t *ppos) | |
1303 | { | |
1304 | struct ctr_struct *ctr = file->private_data; | |
1305 | ||
1306 | return simple_read_from_buffer(buf, nbytes, ppos, ctr->buf, ctr->bufsz); | |
1307 | } | |
1308 | ||
1309 | static int cgroup_tasks_release(struct inode *unused_inode, | |
1310 | struct file *file) | |
1311 | { | |
1312 | struct ctr_struct *ctr; | |
1313 | ||
1314 | if (file->f_mode & FMODE_READ) { | |
1315 | ctr = file->private_data; | |
1316 | kfree(ctr->buf); | |
1317 | kfree(ctr); | |
1318 | } | |
1319 | return 0; | |
1320 | } | |
1321 | ||
1322 | /* | |
1323 | * for the common functions, 'private' gives the type of file | |
1324 | */ | |
1325 | static struct cftype cft_tasks = { | |
1326 | .name = "tasks", | |
1327 | .open = cgroup_tasks_open, | |
1328 | .read = cgroup_tasks_read, | |
1329 | .write = cgroup_common_file_write, | |
1330 | .release = cgroup_tasks_release, | |
1331 | .private = FILE_TASKLIST, | |
1332 | }; | |
1333 | ||
ddbcc7e8 PM |
1334 | static int cgroup_populate_dir(struct cgroup *cont) |
1335 | { | |
1336 | int err; | |
1337 | struct cgroup_subsys *ss; | |
1338 | ||
1339 | /* First clear out any existing files */ | |
1340 | cgroup_clear_directory(cont->dentry); | |
1341 | ||
bbcb81d0 PM |
1342 | err = cgroup_add_file(cont, NULL, &cft_tasks); |
1343 | if (err < 0) | |
1344 | return err; | |
1345 | ||
ddbcc7e8 PM |
1346 | for_each_subsys(cont->root, ss) { |
1347 | if (ss->populate && (err = ss->populate(ss, cont)) < 0) | |
1348 | return err; | |
1349 | } | |
1350 | ||
1351 | return 0; | |
1352 | } | |
1353 | ||
1354 | static void init_cgroup_css(struct cgroup_subsys_state *css, | |
1355 | struct cgroup_subsys *ss, | |
1356 | struct cgroup *cont) | |
1357 | { | |
1358 | css->cgroup = cont; | |
1359 | atomic_set(&css->refcnt, 0); | |
1360 | css->flags = 0; | |
1361 | if (cont == dummytop) | |
1362 | set_bit(CSS_ROOT, &css->flags); | |
1363 | BUG_ON(cont->subsys[ss->subsys_id]); | |
1364 | cont->subsys[ss->subsys_id] = css; | |
1365 | } | |
1366 | ||
1367 | /* | |
1368 | * cgroup_create - create a cgroup | |
1369 | * parent: cgroup that will be parent of the new cgroup. | |
1370 | * name: name of the new cgroup. Will be strcpy'ed. | |
1371 | * mode: mode to set on new inode | |
1372 | * | |
1373 | * Must be called with the mutex on the parent inode held | |
1374 | */ | |
1375 | ||
1376 | static long cgroup_create(struct cgroup *parent, struct dentry *dentry, | |
1377 | int mode) | |
1378 | { | |
1379 | struct cgroup *cont; | |
1380 | struct cgroupfs_root *root = parent->root; | |
1381 | int err = 0; | |
1382 | struct cgroup_subsys *ss; | |
1383 | struct super_block *sb = root->sb; | |
1384 | ||
1385 | cont = kzalloc(sizeof(*cont), GFP_KERNEL); | |
1386 | if (!cont) | |
1387 | return -ENOMEM; | |
1388 | ||
1389 | /* Grab a reference on the superblock so the hierarchy doesn't | |
1390 | * get deleted on unmount if there are child cgroups. This | |
1391 | * can be done outside cgroup_mutex, since the sb can't | |
1392 | * disappear while someone has an open control file on the | |
1393 | * fs */ | |
1394 | atomic_inc(&sb->s_active); | |
1395 | ||
1396 | mutex_lock(&cgroup_mutex); | |
1397 | ||
1398 | cont->flags = 0; | |
1399 | INIT_LIST_HEAD(&cont->sibling); | |
1400 | INIT_LIST_HEAD(&cont->children); | |
1401 | ||
1402 | cont->parent = parent; | |
1403 | cont->root = parent->root; | |
1404 | cont->top_cgroup = parent->top_cgroup; | |
1405 | ||
1406 | for_each_subsys(root, ss) { | |
1407 | struct cgroup_subsys_state *css = ss->create(ss, cont); | |
1408 | if (IS_ERR(css)) { | |
1409 | err = PTR_ERR(css); | |
1410 | goto err_destroy; | |
1411 | } | |
1412 | init_cgroup_css(css, ss, cont); | |
1413 | } | |
1414 | ||
1415 | list_add(&cont->sibling, &cont->parent->children); | |
1416 | root->number_of_cgroups++; | |
1417 | ||
1418 | err = cgroup_create_dir(cont, dentry, mode); | |
1419 | if (err < 0) | |
1420 | goto err_remove; | |
1421 | ||
1422 | /* The cgroup directory was pre-locked for us */ | |
1423 | BUG_ON(!mutex_is_locked(&cont->dentry->d_inode->i_mutex)); | |
1424 | ||
1425 | err = cgroup_populate_dir(cont); | |
1426 | /* If err < 0, we have a half-filled directory - oh well ;) */ | |
1427 | ||
1428 | mutex_unlock(&cgroup_mutex); | |
1429 | mutex_unlock(&cont->dentry->d_inode->i_mutex); | |
1430 | ||
1431 | return 0; | |
1432 | ||
1433 | err_remove: | |
1434 | ||
1435 | list_del(&cont->sibling); | |
1436 | root->number_of_cgroups--; | |
1437 | ||
1438 | err_destroy: | |
1439 | ||
1440 | for_each_subsys(root, ss) { | |
1441 | if (cont->subsys[ss->subsys_id]) | |
1442 | ss->destroy(ss, cont); | |
1443 | } | |
1444 | ||
1445 | mutex_unlock(&cgroup_mutex); | |
1446 | ||
1447 | /* Release the reference count that we took on the superblock */ | |
1448 | deactivate_super(sb); | |
1449 | ||
1450 | kfree(cont); | |
1451 | return err; | |
1452 | } | |
1453 | ||
1454 | static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, int mode) | |
1455 | { | |
1456 | struct cgroup *c_parent = dentry->d_parent->d_fsdata; | |
1457 | ||
1458 | /* the vfs holds inode->i_mutex already */ | |
1459 | return cgroup_create(c_parent, dentry, mode | S_IFDIR); | |
1460 | } | |
1461 | ||
1462 | static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry) | |
1463 | { | |
1464 | struct cgroup *cont = dentry->d_fsdata; | |
1465 | struct dentry *d; | |
1466 | struct cgroup *parent; | |
1467 | struct cgroup_subsys *ss; | |
1468 | struct super_block *sb; | |
1469 | struct cgroupfs_root *root; | |
1470 | int css_busy = 0; | |
1471 | ||
1472 | /* the vfs holds both inode->i_mutex already */ | |
1473 | ||
1474 | mutex_lock(&cgroup_mutex); | |
1475 | if (atomic_read(&cont->count) != 0) { | |
1476 | mutex_unlock(&cgroup_mutex); | |
1477 | return -EBUSY; | |
1478 | } | |
1479 | if (!list_empty(&cont->children)) { | |
1480 | mutex_unlock(&cgroup_mutex); | |
1481 | return -EBUSY; | |
1482 | } | |
1483 | ||
1484 | parent = cont->parent; | |
1485 | root = cont->root; | |
1486 | sb = root->sb; | |
1487 | ||
1488 | /* Check the reference count on each subsystem. Since we | |
1489 | * already established that there are no tasks in the | |
1490 | * cgroup, if the css refcount is also 0, then there should | |
1491 | * be no outstanding references, so the subsystem is safe to | |
1492 | * destroy */ | |
1493 | for_each_subsys(root, ss) { | |
1494 | struct cgroup_subsys_state *css; | |
1495 | css = cont->subsys[ss->subsys_id]; | |
1496 | if (atomic_read(&css->refcnt)) { | |
1497 | css_busy = 1; | |
1498 | break; | |
1499 | } | |
1500 | } | |
1501 | if (css_busy) { | |
1502 | mutex_unlock(&cgroup_mutex); | |
1503 | return -EBUSY; | |
1504 | } | |
1505 | ||
1506 | for_each_subsys(root, ss) { | |
1507 | if (cont->subsys[ss->subsys_id]) | |
1508 | ss->destroy(ss, cont); | |
1509 | } | |
1510 | ||
1511 | set_bit(CONT_REMOVED, &cont->flags); | |
1512 | /* delete my sibling from parent->children */ | |
1513 | list_del(&cont->sibling); | |
1514 | spin_lock(&cont->dentry->d_lock); | |
1515 | d = dget(cont->dentry); | |
1516 | cont->dentry = NULL; | |
1517 | spin_unlock(&d->d_lock); | |
1518 | ||
1519 | cgroup_d_remove_dir(d); | |
1520 | dput(d); | |
1521 | root->number_of_cgroups--; | |
1522 | ||
1523 | mutex_unlock(&cgroup_mutex); | |
1524 | /* Drop the active superblock reference that we took when we | |
1525 | * created the cgroup */ | |
1526 | deactivate_super(sb); | |
1527 | return 0; | |
1528 | } | |
1529 | ||
1530 | static void cgroup_init_subsys(struct cgroup_subsys *ss) | |
1531 | { | |
1532 | struct task_struct *g, *p; | |
1533 | struct cgroup_subsys_state *css; | |
1534 | printk(KERN_ERR "Initializing cgroup subsys %s\n", ss->name); | |
1535 | ||
1536 | /* Create the top cgroup state for this subsystem */ | |
1537 | ss->root = &rootnode; | |
1538 | css = ss->create(ss, dummytop); | |
1539 | /* We don't handle early failures gracefully */ | |
1540 | BUG_ON(IS_ERR(css)); | |
1541 | init_cgroup_css(css, ss, dummytop); | |
1542 | ||
1543 | /* Update all tasks to contain a subsys pointer to this state | |
1544 | * - since the subsystem is newly registered, all tasks are in | |
1545 | * the subsystem's top cgroup. */ | |
1546 | ||
1547 | /* If this subsystem requested that it be notified with fork | |
1548 | * events, we should send it one now for every process in the | |
1549 | * system */ | |
1550 | ||
1551 | read_lock(&tasklist_lock); | |
1552 | init_task.cgroups.subsys[ss->subsys_id] = css; | |
1553 | if (ss->fork) | |
1554 | ss->fork(ss, &init_task); | |
1555 | ||
1556 | do_each_thread(g, p) { | |
1557 | printk(KERN_INFO "Setting task %p css to %p (%d)\n", css, p, p->pid); | |
1558 | p->cgroups.subsys[ss->subsys_id] = css; | |
1559 | if (ss->fork) | |
1560 | ss->fork(ss, p); | |
1561 | } while_each_thread(g, p); | |
1562 | read_unlock(&tasklist_lock); | |
1563 | ||
1564 | need_forkexit_callback |= ss->fork || ss->exit; | |
1565 | ||
1566 | ss->active = 1; | |
1567 | } | |
1568 | ||
1569 | /** | |
1570 | * cgroup_init_early - initialize cgroups at system boot, and | |
1571 | * initialize any subsystems that request early init. | |
1572 | */ | |
1573 | int __init cgroup_init_early(void) | |
1574 | { | |
1575 | int i; | |
1576 | init_cgroup_root(&rootnode); | |
1577 | list_add(&rootnode.root_list, &roots); | |
1578 | ||
1579 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { | |
1580 | struct cgroup_subsys *ss = subsys[i]; | |
1581 | ||
1582 | BUG_ON(!ss->name); | |
1583 | BUG_ON(strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN); | |
1584 | BUG_ON(!ss->create); | |
1585 | BUG_ON(!ss->destroy); | |
1586 | if (ss->subsys_id != i) { | |
1587 | printk(KERN_ERR "Subsys %s id == %d\n", | |
1588 | ss->name, ss->subsys_id); | |
1589 | BUG(); | |
1590 | } | |
1591 | ||
1592 | if (ss->early_init) | |
1593 | cgroup_init_subsys(ss); | |
1594 | } | |
1595 | return 0; | |
1596 | } | |
1597 | ||
1598 | /** | |
1599 | * cgroup_init - register cgroup filesystem and /proc file, and | |
1600 | * initialize any subsystems that didn't request early init. | |
1601 | */ | |
1602 | int __init cgroup_init(void) | |
1603 | { | |
1604 | int err; | |
1605 | int i; | |
a424316c PM |
1606 | struct proc_dir_entry *entry; |
1607 | ||
1608 | err = bdi_init(&cgroup_backing_dev_info); | |
1609 | if (err) | |
1610 | return err; | |
ddbcc7e8 PM |
1611 | |
1612 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { | |
1613 | struct cgroup_subsys *ss = subsys[i]; | |
1614 | if (!ss->early_init) | |
1615 | cgroup_init_subsys(ss); | |
1616 | } | |
1617 | ||
1618 | err = register_filesystem(&cgroup_fs_type); | |
1619 | if (err < 0) | |
1620 | goto out; | |
1621 | ||
a424316c PM |
1622 | entry = create_proc_entry("cgroups", 0, NULL); |
1623 | if (entry) | |
1624 | entry->proc_fops = &proc_cgroupstats_operations; | |
1625 | ||
ddbcc7e8 | 1626 | out: |
a424316c PM |
1627 | if (err) |
1628 | bdi_destroy(&cgroup_backing_dev_info); | |
1629 | ||
ddbcc7e8 PM |
1630 | return err; |
1631 | } | |
b4f48b63 | 1632 | |
a424316c PM |
1633 | /* |
1634 | * proc_cgroup_show() | |
1635 | * - Print task's cgroup paths into seq_file, one line for each hierarchy | |
1636 | * - Used for /proc/<pid>/cgroup. | |
1637 | * - No need to task_lock(tsk) on this tsk->cgroup reference, as it | |
1638 | * doesn't really matter if tsk->cgroup changes after we read it, | |
1639 | * and we take cgroup_mutex, keeping attach_task() from changing it | |
1640 | * anyway. No need to check that tsk->cgroup != NULL, thanks to | |
1641 | * the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks | |
1642 | * cgroup to top_cgroup. | |
1643 | */ | |
1644 | ||
1645 | /* TODO: Use a proper seq_file iterator */ | |
1646 | static int proc_cgroup_show(struct seq_file *m, void *v) | |
1647 | { | |
1648 | struct pid *pid; | |
1649 | struct task_struct *tsk; | |
1650 | char *buf; | |
1651 | int retval; | |
1652 | struct cgroupfs_root *root; | |
1653 | ||
1654 | retval = -ENOMEM; | |
1655 | buf = kmalloc(PAGE_SIZE, GFP_KERNEL); | |
1656 | if (!buf) | |
1657 | goto out; | |
1658 | ||
1659 | retval = -ESRCH; | |
1660 | pid = m->private; | |
1661 | tsk = get_pid_task(pid, PIDTYPE_PID); | |
1662 | if (!tsk) | |
1663 | goto out_free; | |
1664 | ||
1665 | retval = 0; | |
1666 | ||
1667 | mutex_lock(&cgroup_mutex); | |
1668 | ||
1669 | for_each_root(root) { | |
1670 | struct cgroup_subsys *ss; | |
1671 | struct cgroup *cont; | |
1672 | int subsys_id; | |
1673 | int count = 0; | |
1674 | ||
1675 | /* Skip this hierarchy if it has no active subsystems */ | |
1676 | if (!root->actual_subsys_bits) | |
1677 | continue; | |
1678 | for_each_subsys(root, ss) | |
1679 | seq_printf(m, "%s%s", count++ ? "," : "", ss->name); | |
1680 | seq_putc(m, ':'); | |
1681 | get_first_subsys(&root->top_cgroup, NULL, &subsys_id); | |
1682 | cont = task_cgroup(tsk, subsys_id); | |
1683 | retval = cgroup_path(cont, buf, PAGE_SIZE); | |
1684 | if (retval < 0) | |
1685 | goto out_unlock; | |
1686 | seq_puts(m, buf); | |
1687 | seq_putc(m, '\n'); | |
1688 | } | |
1689 | ||
1690 | out_unlock: | |
1691 | mutex_unlock(&cgroup_mutex); | |
1692 | put_task_struct(tsk); | |
1693 | out_free: | |
1694 | kfree(buf); | |
1695 | out: | |
1696 | return retval; | |
1697 | } | |
1698 | ||
1699 | static int cgroup_open(struct inode *inode, struct file *file) | |
1700 | { | |
1701 | struct pid *pid = PROC_I(inode)->pid; | |
1702 | return single_open(file, proc_cgroup_show, pid); | |
1703 | } | |
1704 | ||
1705 | struct file_operations proc_cgroup_operations = { | |
1706 | .open = cgroup_open, | |
1707 | .read = seq_read, | |
1708 | .llseek = seq_lseek, | |
1709 | .release = single_release, | |
1710 | }; | |
1711 | ||
1712 | /* Display information about each subsystem and each hierarchy */ | |
1713 | static int proc_cgroupstats_show(struct seq_file *m, void *v) | |
1714 | { | |
1715 | int i; | |
1716 | struct cgroupfs_root *root; | |
1717 | ||
1718 | mutex_lock(&cgroup_mutex); | |
1719 | seq_puts(m, "Hierarchies:\n"); | |
1720 | for_each_root(root) { | |
1721 | struct cgroup_subsys *ss; | |
1722 | int first = 1; | |
1723 | seq_printf(m, "%p: bits=%lx cgroups=%d (", root, | |
1724 | root->subsys_bits, root->number_of_cgroups); | |
1725 | for_each_subsys(root, ss) { | |
1726 | seq_printf(m, "%s%s", first ? "" : ", ", ss->name); | |
1727 | first = false; | |
1728 | } | |
1729 | seq_putc(m, ')'); | |
1730 | if (root->sb) { | |
1731 | seq_printf(m, " s_active=%d", | |
1732 | atomic_read(&root->sb->s_active)); | |
1733 | } | |
1734 | seq_putc(m, '\n'); | |
1735 | } | |
1736 | seq_puts(m, "Subsystems:\n"); | |
1737 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { | |
1738 | struct cgroup_subsys *ss = subsys[i]; | |
1739 | seq_printf(m, "%d: name=%s hierarchy=%p\n", | |
1740 | i, ss->name, ss->root); | |
1741 | } | |
1742 | mutex_unlock(&cgroup_mutex); | |
1743 | return 0; | |
1744 | } | |
1745 | ||
1746 | static int cgroupstats_open(struct inode *inode, struct file *file) | |
1747 | { | |
1748 | return single_open(file, proc_cgroupstats_show, 0); | |
1749 | } | |
1750 | ||
1751 | static struct file_operations proc_cgroupstats_operations = { | |
1752 | .open = cgroupstats_open, | |
1753 | .read = seq_read, | |
1754 | .llseek = seq_lseek, | |
1755 | .release = single_release, | |
1756 | }; | |
1757 | ||
b4f48b63 PM |
1758 | /** |
1759 | * cgroup_fork - attach newly forked task to its parents cgroup. | |
1760 | * @tsk: pointer to task_struct of forking parent process. | |
1761 | * | |
1762 | * Description: A task inherits its parent's cgroup at fork(). | |
1763 | * | |
1764 | * A pointer to the shared css_set was automatically copied in | |
1765 | * fork.c by dup_task_struct(). However, we ignore that copy, since | |
1766 | * it was not made under the protection of RCU or cgroup_mutex, so | |
1767 | * might no longer be a valid cgroup pointer. attach_task() might | |
1768 | * have already changed current->cgroup, allowing the previously | |
1769 | * referenced cgroup to be removed and freed. | |
1770 | * | |
1771 | * At the point that cgroup_fork() is called, 'current' is the parent | |
1772 | * task, and the passed argument 'child' points to the child task. | |
1773 | */ | |
1774 | void cgroup_fork(struct task_struct *child) | |
1775 | { | |
1776 | rcu_read_lock(); | |
1777 | child->cgroups = rcu_dereference(current->cgroups); | |
1778 | get_css_set(&child->cgroups); | |
1779 | rcu_read_unlock(); | |
1780 | } | |
1781 | ||
1782 | /** | |
1783 | * cgroup_fork_callbacks - called on a new task very soon before | |
1784 | * adding it to the tasklist. No need to take any locks since no-one | |
1785 | * can be operating on this task | |
1786 | */ | |
1787 | void cgroup_fork_callbacks(struct task_struct *child) | |
1788 | { | |
1789 | if (need_forkexit_callback) { | |
1790 | int i; | |
1791 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { | |
1792 | struct cgroup_subsys *ss = subsys[i]; | |
1793 | if (ss->fork) | |
1794 | ss->fork(ss, child); | |
1795 | } | |
1796 | } | |
1797 | } | |
1798 | ||
1799 | /** | |
1800 | * cgroup_exit - detach cgroup from exiting task | |
1801 | * @tsk: pointer to task_struct of exiting process | |
1802 | * | |
1803 | * Description: Detach cgroup from @tsk and release it. | |
1804 | * | |
1805 | * Note that cgroups marked notify_on_release force every task in | |
1806 | * them to take the global cgroup_mutex mutex when exiting. | |
1807 | * This could impact scaling on very large systems. Be reluctant to | |
1808 | * use notify_on_release cgroups where very high task exit scaling | |
1809 | * is required on large systems. | |
1810 | * | |
1811 | * the_top_cgroup_hack: | |
1812 | * | |
1813 | * Set the exiting tasks cgroup to the root cgroup (top_cgroup). | |
1814 | * | |
1815 | * We call cgroup_exit() while the task is still competent to | |
1816 | * handle notify_on_release(), then leave the task attached to the | |
1817 | * root cgroup in each hierarchy for the remainder of its exit. | |
1818 | * | |
1819 | * To do this properly, we would increment the reference count on | |
1820 | * top_cgroup, and near the very end of the kernel/exit.c do_exit() | |
1821 | * code we would add a second cgroup function call, to drop that | |
1822 | * reference. This would just create an unnecessary hot spot on | |
1823 | * the top_cgroup reference count, to no avail. | |
1824 | * | |
1825 | * Normally, holding a reference to a cgroup without bumping its | |
1826 | * count is unsafe. The cgroup could go away, or someone could | |
1827 | * attach us to a different cgroup, decrementing the count on | |
1828 | * the first cgroup that we never incremented. But in this case, | |
1829 | * top_cgroup isn't going away, and either task has PF_EXITING set, | |
1830 | * which wards off any attach_task() attempts, or task is a failed | |
1831 | * fork, never visible to attach_task. | |
1832 | * | |
1833 | */ | |
1834 | void cgroup_exit(struct task_struct *tsk, int run_callbacks) | |
1835 | { | |
1836 | int i; | |
1837 | ||
1838 | if (run_callbacks && need_forkexit_callback) { | |
1839 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { | |
1840 | struct cgroup_subsys *ss = subsys[i]; | |
1841 | if (ss->exit) | |
1842 | ss->exit(ss, tsk); | |
1843 | } | |
1844 | } | |
1845 | /* Reassign the task to the init_css_set. */ | |
1846 | task_lock(tsk); | |
1847 | put_css_set(&tsk->cgroups); | |
1848 | tsk->cgroups = init_task.cgroups; | |
1849 | task_unlock(tsk); | |
1850 | } | |
697f4161 PM |
1851 | |
1852 | /** | |
1853 | * cgroup_clone - duplicate the current cgroup in the hierarchy | |
1854 | * that the given subsystem is attached to, and move this task into | |
1855 | * the new child | |
1856 | */ | |
1857 | int cgroup_clone(struct task_struct *tsk, struct cgroup_subsys *subsys) | |
1858 | { | |
1859 | struct dentry *dentry; | |
1860 | int ret = 0; | |
1861 | char nodename[MAX_CGROUP_TYPE_NAMELEN]; | |
1862 | struct cgroup *parent, *child; | |
1863 | struct inode *inode; | |
1864 | struct css_set *cg; | |
1865 | struct cgroupfs_root *root; | |
1866 | struct cgroup_subsys *ss; | |
1867 | ||
1868 | /* We shouldn't be called by an unregistered subsystem */ | |
1869 | BUG_ON(!subsys->active); | |
1870 | ||
1871 | /* First figure out what hierarchy and cgroup we're dealing | |
1872 | * with, and pin them so we can drop cgroup_mutex */ | |
1873 | mutex_lock(&cgroup_mutex); | |
1874 | again: | |
1875 | root = subsys->root; | |
1876 | if (root == &rootnode) { | |
1877 | printk(KERN_INFO | |
1878 | "Not cloning cgroup for unused subsystem %s\n", | |
1879 | subsys->name); | |
1880 | mutex_unlock(&cgroup_mutex); | |
1881 | return 0; | |
1882 | } | |
1883 | cg = &tsk->cgroups; | |
1884 | parent = task_cgroup(tsk, subsys->subsys_id); | |
1885 | ||
1886 | snprintf(nodename, MAX_CGROUP_TYPE_NAMELEN, "node_%d", tsk->pid); | |
1887 | ||
1888 | /* Pin the hierarchy */ | |
1889 | atomic_inc(&parent->root->sb->s_active); | |
1890 | ||
1891 | mutex_unlock(&cgroup_mutex); | |
1892 | ||
1893 | /* Now do the VFS work to create a cgroup */ | |
1894 | inode = parent->dentry->d_inode; | |
1895 | ||
1896 | /* Hold the parent directory mutex across this operation to | |
1897 | * stop anyone else deleting the new cgroup */ | |
1898 | mutex_lock(&inode->i_mutex); | |
1899 | dentry = lookup_one_len(nodename, parent->dentry, strlen(nodename)); | |
1900 | if (IS_ERR(dentry)) { | |
1901 | printk(KERN_INFO | |
1902 | "Couldn't allocate dentry for %s: %ld\n", nodename, | |
1903 | PTR_ERR(dentry)); | |
1904 | ret = PTR_ERR(dentry); | |
1905 | goto out_release; | |
1906 | } | |
1907 | ||
1908 | /* Create the cgroup directory, which also creates the cgroup */ | |
1909 | ret = vfs_mkdir(inode, dentry, S_IFDIR | 0755); | |
1910 | child = __d_cont(dentry); | |
1911 | dput(dentry); | |
1912 | if (ret) { | |
1913 | printk(KERN_INFO | |
1914 | "Failed to create cgroup %s: %d\n", nodename, | |
1915 | ret); | |
1916 | goto out_release; | |
1917 | } | |
1918 | ||
1919 | if (!child) { | |
1920 | printk(KERN_INFO | |
1921 | "Couldn't find new cgroup %s\n", nodename); | |
1922 | ret = -ENOMEM; | |
1923 | goto out_release; | |
1924 | } | |
1925 | ||
1926 | /* The cgroup now exists. Retake cgroup_mutex and check | |
1927 | * that we're still in the same state that we thought we | |
1928 | * were. */ | |
1929 | mutex_lock(&cgroup_mutex); | |
1930 | if ((root != subsys->root) || | |
1931 | (parent != task_cgroup(tsk, subsys->subsys_id))) { | |
1932 | /* Aargh, we raced ... */ | |
1933 | mutex_unlock(&inode->i_mutex); | |
1934 | ||
1935 | deactivate_super(parent->root->sb); | |
1936 | /* The cgroup is still accessible in the VFS, but | |
1937 | * we're not going to try to rmdir() it at this | |
1938 | * point. */ | |
1939 | printk(KERN_INFO | |
1940 | "Race in cgroup_clone() - leaking cgroup %s\n", | |
1941 | nodename); | |
1942 | goto again; | |
1943 | } | |
1944 | ||
1945 | /* do any required auto-setup */ | |
1946 | for_each_subsys(root, ss) { | |
1947 | if (ss->post_clone) | |
1948 | ss->post_clone(ss, child); | |
1949 | } | |
1950 | ||
1951 | /* All seems fine. Finish by moving the task into the new cgroup */ | |
1952 | ret = attach_task(child, tsk); | |
1953 | mutex_unlock(&cgroup_mutex); | |
1954 | ||
1955 | out_release: | |
1956 | mutex_unlock(&inode->i_mutex); | |
1957 | deactivate_super(parent->root->sb); | |
1958 | return ret; | |
1959 | } | |
1960 | ||
1961 | /* | |
1962 | * See if "cont" is a descendant of the current task's cgroup in | |
1963 | * the appropriate hierarchy | |
1964 | * | |
1965 | * If we are sending in dummytop, then presumably we are creating | |
1966 | * the top cgroup in the subsystem. | |
1967 | * | |
1968 | * Called only by the ns (nsproxy) cgroup. | |
1969 | */ | |
1970 | int cgroup_is_descendant(const struct cgroup *cont) | |
1971 | { | |
1972 | int ret; | |
1973 | struct cgroup *target; | |
1974 | int subsys_id; | |
1975 | ||
1976 | if (cont == dummytop) | |
1977 | return 1; | |
1978 | ||
1979 | get_first_subsys(cont, NULL, &subsys_id); | |
1980 | target = task_cgroup(current, subsys_id); | |
1981 | while (cont != target && cont!= cont->top_cgroup) | |
1982 | cont = cont->parent; | |
1983 | ret = (cont == target); | |
1984 | return ret; | |
1985 | } |