~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/kernel/cgroup/cgroup-v1.c

Version: ~ [ linux-5.15-rc1 ] ~ [ linux-5.14.5 ] ~ [ linux-5.13.18 ] ~ [ linux-5.12.19 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.66 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.147 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.206 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.246 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.282 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.283 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.18.140 ] ~ [ linux-3.16.85 ] ~ [ linux-3.14.79 ] ~ [ linux-3.12.74 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 // SPDX-License-Identifier: GPL-2.0-only
  2 #include "cgroup-internal.h"
  3 
  4 #include <linux/ctype.h>
  5 #include <linux/kmod.h>
  6 #include <linux/sort.h>
  7 #include <linux/delay.h>
  8 #include <linux/mm.h>
  9 #include <linux/sched/signal.h>
 10 #include <linux/sched/task.h>
 11 #include <linux/magic.h>
 12 #include <linux/slab.h>
 13 #include <linux/vmalloc.h>
 14 #include <linux/delayacct.h>
 15 #include <linux/pid_namespace.h>
 16 #include <linux/cgroupstats.h>
 17 #include <linux/fs_parser.h>
 18 
 19 #include <trace/events/cgroup.h>
 20 
 21 /*
 22  * pidlists linger the following amount before being destroyed.  The goal
 23  * is avoiding frequent destruction in the middle of consecutive read calls
 24  * Expiring in the middle is a performance problem not a correctness one.
 25  * 1 sec should be enough.
 26  */
 27 #define CGROUP_PIDLIST_DESTROY_DELAY    HZ
 28 
 29 /* Controllers blocked by the commandline in v1 */
 30 static u16 cgroup_no_v1_mask;
 31 
 32 /* disable named v1 mounts */
 33 static bool cgroup_no_v1_named;
 34 
 35 /*
 36  * pidlist destructions need to be flushed on cgroup destruction.  Use a
 37  * separate workqueue as flush domain.
 38  */
 39 static struct workqueue_struct *cgroup_pidlist_destroy_wq;
 40 
 41 /*
 42  * Protects cgroup_subsys->release_agent_path.  Modifying it also requires
 43  * cgroup_mutex.  Reading requires either cgroup_mutex or this spinlock.
 44  */
 45 static DEFINE_SPINLOCK(release_agent_path_lock);
 46 
 47 bool cgroup1_ssid_disabled(int ssid)
 48 {
 49         return cgroup_no_v1_mask & (1 << ssid);
 50 }
 51 
 52 /**
 53  * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from'
 54  * @from: attach to all cgroups of a given task
 55  * @tsk: the task to be attached
 56  */
 57 int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk)
 58 {
 59         struct cgroup_root *root;
 60         int retval = 0;
 61 
 62         mutex_lock(&cgroup_mutex);
 63         percpu_down_write(&cgroup_threadgroup_rwsem);
 64         for_each_root(root) {
 65                 struct cgroup *from_cgrp;
 66 
 67                 if (root == &cgrp_dfl_root)
 68                         continue;
 69 
 70                 spin_lock_irq(&css_set_lock);
 71                 from_cgrp = task_cgroup_from_root(from, root);
 72                 spin_unlock_irq(&css_set_lock);
 73 
 74                 retval = cgroup_attach_task(from_cgrp, tsk, false);
 75                 if (retval)
 76                         break;
 77         }
 78         percpu_up_write(&cgroup_threadgroup_rwsem);
 79         mutex_unlock(&cgroup_mutex);
 80 
 81         return retval;
 82 }
 83 EXPORT_SYMBOL_GPL(cgroup_attach_task_all);
 84 
 85 /**
 86  * cgroup_trasnsfer_tasks - move tasks from one cgroup to another
 87  * @to: cgroup to which the tasks will be moved
 88  * @from: cgroup in which the tasks currently reside
 89  *
 90  * Locking rules between cgroup_post_fork() and the migration path
 91  * guarantee that, if a task is forking while being migrated, the new child
 92  * is guaranteed to be either visible in the source cgroup after the
 93  * parent's migration is complete or put into the target cgroup.  No task
 94  * can slip out of migration through forking.
 95  */
 96 int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from)
 97 {
 98         DEFINE_CGROUP_MGCTX(mgctx);
 99         struct cgrp_cset_link *link;
100         struct css_task_iter it;
101         struct task_struct *task;
102         int ret;
103 
104         if (cgroup_on_dfl(to))
105                 return -EINVAL;
106 
107         ret = cgroup_migrate_vet_dst(to);
108         if (ret)
109                 return ret;
110 
111         mutex_lock(&cgroup_mutex);
112 
113         percpu_down_write(&cgroup_threadgroup_rwsem);
114 
115         /* all tasks in @from are being moved, all csets are source */
116         spin_lock_irq(&css_set_lock);
117         list_for_each_entry(link, &from->cset_links, cset_link)
118                 cgroup_migrate_add_src(link->cset, to, &mgctx);
119         spin_unlock_irq(&css_set_lock);
120 
121         ret = cgroup_migrate_prepare_dst(&mgctx);
122         if (ret)
123                 goto out_err;
124 
125         /*
126          * Migrate tasks one-by-one until @from is empty.  This fails iff
127          * ->can_attach() fails.
128          */
129         do {
130                 css_task_iter_start(&from->self, 0, &it);
131 
132                 do {
133                         task = css_task_iter_next(&it);
134                 } while (task && (task->flags & PF_EXITING));
135 
136                 if (task)
137                         get_task_struct(task);
138                 css_task_iter_end(&it);
139 
140                 if (task) {
141                         ret = cgroup_migrate(task, false, &mgctx);
142                         if (!ret)
143                                 TRACE_CGROUP_PATH(transfer_tasks, to, task, false);
144                         put_task_struct(task);
145                 }
146         } while (task && !ret);
147 out_err:
148         cgroup_migrate_finish(&mgctx);
149         percpu_up_write(&cgroup_threadgroup_rwsem);
150         mutex_unlock(&cgroup_mutex);
151         return ret;
152 }
153 
154 /*
155  * Stuff for reading the 'tasks'/'procs' files.
156  *
157  * Reading this file can return large amounts of data if a cgroup has
158  * *lots* of attached tasks. So it may need several calls to read(),
159  * but we cannot guarantee that the information we produce is correct
160  * unless we produce it entirely atomically.
161  *
162  */
163 
164 /* which pidlist file are we talking about? */
165 enum cgroup_filetype {
166         CGROUP_FILE_PROCS,
167         CGROUP_FILE_TASKS,
168 };
169 
170 /*
171  * A pidlist is a list of pids that virtually represents the contents of one
172  * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists,
173  * a pair (one each for procs, tasks) for each pid namespace that's relevant
174  * to the cgroup.
175  */
176 struct cgroup_pidlist {
177         /*
178          * used to find which pidlist is wanted. doesn't change as long as
179          * this particular list stays in the list.
180         */
181         struct { enum cgroup_filetype type; struct pid_namespace *ns; } key;
182         /* array of xids */
183         pid_t *list;
184         /* how many elements the above list has */
185         int length;
186         /* each of these stored in a list by its cgroup */
187         struct list_head links;
188         /* pointer to the cgroup we belong to, for list removal purposes */
189         struct cgroup *owner;
190         /* for delayed destruction */
191         struct delayed_work destroy_dwork;
192 };
193 
194 /*
195  * Used to destroy all pidlists lingering waiting for destroy timer.  None
196  * should be left afterwards.
197  */
198 void cgroup1_pidlist_destroy_all(struct cgroup *cgrp)
199 {
200         struct cgroup_pidlist *l, *tmp_l;
201 
202         mutex_lock(&cgrp->pidlist_mutex);
203         list_for_each_entry_safe(l, tmp_l, &cgrp->pidlists, links)
204                 mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, 0);
205         mutex_unlock(&cgrp->pidlist_mutex);
206 
207         flush_workqueue(cgroup_pidlist_destroy_wq);
208         BUG_ON(!list_empty(&cgrp->pidlists));
209 }
210 
211 static void cgroup_pidlist_destroy_work_fn(struct work_struct *work)
212 {
213         struct delayed_work *dwork = to_delayed_work(work);
214         struct cgroup_pidlist *l = container_of(dwork, struct cgroup_pidlist,
215                                                 destroy_dwork);
216         struct cgroup_pidlist *tofree = NULL;
217 
218         mutex_lock(&l->owner->pidlist_mutex);
219 
220         /*
221          * Destroy iff we didn't get queued again.  The state won't change
222          * as destroy_dwork can only be queued while locked.
223          */
224         if (!delayed_work_pending(dwork)) {
225                 list_del(&l->links);
226                 kvfree(l->list);
227                 put_pid_ns(l->key.ns);
228                 tofree = l;
229         }
230 
231         mutex_unlock(&l->owner->pidlist_mutex);
232         kfree(tofree);
233 }
234 
235 /*
236  * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
237  * Returns the number of unique elements.
238  */
239 static int pidlist_uniq(pid_t *list, int length)
240 {
241         int src, dest = 1;
242 
243         /*
244          * we presume the 0th element is unique, so i starts at 1. trivial
245          * edge cases first; no work needs to be done for either
246          */
247         if (length == 0 || length == 1)
248                 return length;
249         /* src and dest walk down the list; dest counts unique elements */
250         for (src = 1; src < length; src++) {
251                 /* find next unique element */
252                 while (list[src] == list[src-1]) {
253                         src++;
254                         if (src == length)
255                                 goto after;
256                 }
257                 /* dest always points to where the next unique element goes */
258                 list[dest] = list[src];
259                 dest++;
260         }
261 after:
262         return dest;
263 }
264 
265 /*
266  * The two pid files - task and cgroup.procs - guaranteed that the result
267  * is sorted, which forced this whole pidlist fiasco.  As pid order is
268  * different per namespace, each namespace needs differently sorted list,
269  * making it impossible to use, for example, single rbtree of member tasks
270  * sorted by task pointer.  As pidlists can be fairly large, allocating one
271  * per open file is dangerous, so cgroup had to implement shared pool of
272  * pidlists keyed by cgroup and namespace.
273  */
274 static int cmppid(const void *a, const void *b)
275 {
276         return *(pid_t *)a - *(pid_t *)b;
277 }
278 
279 static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp,
280                                                   enum cgroup_filetype type)
281 {
282         struct cgroup_pidlist *l;
283         /* don't need task_nsproxy() if we're looking at ourself */
284         struct pid_namespace *ns = task_active_pid_ns(current);
285 
286         lockdep_assert_held(&cgrp->pidlist_mutex);
287 
288         list_for_each_entry(l, &cgrp->pidlists, links)
289                 if (l->key.type == type && l->key.ns == ns)
290                         return l;
291         return NULL;
292 }
293 
294 /*
295  * find the appropriate pidlist for our purpose (given procs vs tasks)
296  * returns with the lock on that pidlist already held, and takes care
297  * of the use count, or returns NULL with no locks held if we're out of
298  * memory.
299  */
300 static struct cgroup_pidlist *cgroup_pidlist_find_create(struct cgroup *cgrp,
301                                                 enum cgroup_filetype type)
302 {
303         struct cgroup_pidlist *l;
304 
305         lockdep_assert_held(&cgrp->pidlist_mutex);
306 
307         l = cgroup_pidlist_find(cgrp, type);
308         if (l)
309                 return l;
310 
311         /* entry not found; create a new one */
312         l = kzalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL);
313         if (!l)
314                 return l;
315 
316         INIT_DELAYED_WORK(&l->destroy_dwork, cgroup_pidlist_destroy_work_fn);
317         l->key.type = type;
318         /* don't need task_nsproxy() if we're looking at ourself */
319         l->key.ns = get_pid_ns(task_active_pid_ns(current));
320         l->owner = cgrp;
321         list_add(&l->links, &cgrp->pidlists);
322         return l;
323 }
324 
325 /*
326  * Load a cgroup's pidarray with either procs' tgids or tasks' pids
327  */
328 static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type,
329                               struct cgroup_pidlist **lp)
330 {
331         pid_t *array;
332         int length;
333         int pid, n = 0; /* used for populating the array */
334         struct css_task_iter it;
335         struct task_struct *tsk;
336         struct cgroup_pidlist *l;
337 
338         lockdep_assert_held(&cgrp->pidlist_mutex);
339 
340         /*
341          * If cgroup gets more users after we read count, we won't have
342          * enough space - tough.  This race is indistinguishable to the
343          * caller from the case that the additional cgroup users didn't
344          * show up until sometime later on.
345          */
346         length = cgroup_task_count(cgrp);
347         array = kvmalloc_array(length, sizeof(pid_t), GFP_KERNEL);
348         if (!array)
349                 return -ENOMEM;
350         /* now, populate the array */
351         css_task_iter_start(&cgrp->self, 0, &it);
352         while ((tsk = css_task_iter_next(&it))) {
353                 if (unlikely(n == length))
354                         break;
355                 /* get tgid or pid for procs or tasks file respectively */
356                 if (type == CGROUP_FILE_PROCS)
357                         pid = task_tgid_vnr(tsk);
358                 else
359                         pid = task_pid_vnr(tsk);
360                 if (pid > 0) /* make sure to only use valid results */
361                         array[n++] = pid;
362         }
363         css_task_iter_end(&it);
364         length = n;
365         /* now sort & (if procs) strip out duplicates */
366         sort(array, length, sizeof(pid_t), cmppid, NULL);
367         if (type == CGROUP_FILE_PROCS)
368                 length = pidlist_uniq(array, length);
369 
370         l = cgroup_pidlist_find_create(cgrp, type);
371         if (!l) {
372                 kvfree(array);
373                 return -ENOMEM;
374         }
375 
376         /* store array, freeing old if necessary */
377         kvfree(l->list);
378         l->list = array;
379         l->length = length;
380         *lp = l;
381         return 0;
382 }
383 
384 /*
385  * seq_file methods for the tasks/procs files. The seq_file position is the
386  * next pid to display; the seq_file iterator is a pointer to the pid
387  * in the cgroup->l->list array.
388  */
389 
390 static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos)
391 {
392         /*
393          * Initially we receive a position value that corresponds to
394          * one more than the last pid shown (or 0 on the first call or
395          * after a seek to the start). Use a binary-search to find the
396          * next pid to display, if any
397          */
398         struct kernfs_open_file *of = s->private;
399         struct cgroup *cgrp = seq_css(s)->cgroup;
400         struct cgroup_pidlist *l;
401         enum cgroup_filetype type = seq_cft(s)->private;
402         int index = 0, pid = *pos;
403         int *iter, ret;
404 
405         mutex_lock(&cgrp->pidlist_mutex);
406 
407         /*
408          * !NULL @of->priv indicates that this isn't the first start()
409          * after open.  If the matching pidlist is around, we can use that.
410          * Look for it.  Note that @of->priv can't be used directly.  It
411          * could already have been destroyed.
412          */
413         if (of->priv)
414                 of->priv = cgroup_pidlist_find(cgrp, type);
415 
416         /*
417          * Either this is the first start() after open or the matching
418          * pidlist has been destroyed inbetween.  Create a new one.
419          */
420         if (!of->priv) {
421                 ret = pidlist_array_load(cgrp, type,
422                                          (struct cgroup_pidlist **)&of->priv);
423                 if (ret)
424                         return ERR_PTR(ret);
425         }
426         l = of->priv;
427 
428         if (pid) {
429                 int end = l->length;
430 
431                 while (index < end) {
432                         int mid = (index + end) / 2;
433                         if (l->list[mid] == pid) {
434                                 index = mid;
435                                 break;
436                         } else if (l->list[mid] <= pid)
437                                 index = mid + 1;
438                         else
439                                 end = mid;
440                 }
441         }
442         /* If we're off the end of the array, we're done */
443         if (index >= l->length)
444                 return NULL;
445         /* Update the abstract position to be the actual pid that we found */
446         iter = l->list + index;
447         *pos = *iter;
448         return iter;
449 }
450 
451 static void cgroup_pidlist_stop(struct seq_file *s, void *v)
452 {
453         struct kernfs_open_file *of = s->private;
454         struct cgroup_pidlist *l = of->priv;
455 
456         if (l)
457                 mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork,
458                                  CGROUP_PIDLIST_DESTROY_DELAY);
459         mutex_unlock(&seq_css(s)->cgroup->pidlist_mutex);
460 }
461 
462 static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos)
463 {
464         struct kernfs_open_file *of = s->private;
465         struct cgroup_pidlist *l = of->priv;
466         pid_t *p = v;
467         pid_t *end = l->list + l->length;
468         /*
469          * Advance to the next pid in the array. If this goes off the
470          * end, we're done
471          */
472         p++;
473         if (p >= end) {
474                 (*pos)++;
475                 return NULL;
476         } else {
477                 *pos = *p;
478                 return p;
479         }
480 }
481 
482 static int cgroup_pidlist_show(struct seq_file *s, void *v)
483 {
484         seq_printf(s, "%d\n", *(int *)v);
485 
486         return 0;
487 }
488 
489 static ssize_t __cgroup1_procs_write(struct kernfs_open_file *of,
490                                      char *buf, size_t nbytes, loff_t off,
491                                      bool threadgroup)
492 {
493         struct cgroup *cgrp;
494         struct task_struct *task;
495         const struct cred *cred, *tcred;
496         ssize_t ret;
497         bool locked;
498 
499         cgrp = cgroup_kn_lock_live(of->kn, false);
500         if (!cgrp)
501                 return -ENODEV;
502 
503         task = cgroup_procs_write_start(buf, threadgroup, &locked);
504         ret = PTR_ERR_OR_ZERO(task);
505         if (ret)
506                 goto out_unlock;
507 
508         /*
509          * Even if we're attaching all tasks in the thread group, we only
510          * need to check permissions on one of them.
511          */
512         cred = current_cred();
513         tcred = get_task_cred(task);
514         if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) &&
515             !uid_eq(cred->euid, tcred->uid) &&
516             !uid_eq(cred->euid, tcred->suid))
517                 ret = -EACCES;
518         put_cred(tcred);
519         if (ret)
520                 goto out_finish;
521 
522         ret = cgroup_attach_task(cgrp, task, threadgroup);
523 
524 out_finish:
525         cgroup_procs_write_finish(task, locked);
526 out_unlock:
527         cgroup_kn_unlock(of->kn);
528 
529         return ret ?: nbytes;
530 }
531 
532 static ssize_t cgroup1_procs_write(struct kernfs_open_file *of,
533                                    char *buf, size_t nbytes, loff_t off)
534 {
535         return __cgroup1_procs_write(of, buf, nbytes, off, true);
536 }
537 
538 static ssize_t cgroup1_tasks_write(struct kernfs_open_file *of,
539                                    char *buf, size_t nbytes, loff_t off)
540 {
541         return __cgroup1_procs_write(of, buf, nbytes, off, false);
542 }
543 
544 static ssize_t cgroup_release_agent_write(struct kernfs_open_file *of,
545                                           char *buf, size_t nbytes, loff_t off)
546 {
547         struct cgroup *cgrp;
548 
549         BUILD_BUG_ON(sizeof(cgrp->root->release_agent_path) < PATH_MAX);
550 
551         cgrp = cgroup_kn_lock_live(of->kn, false);
552         if (!cgrp)
553                 return -ENODEV;
554         spin_lock(&release_agent_path_lock);
555         strlcpy(cgrp->root->release_agent_path, strstrip(buf),
556                 sizeof(cgrp->root->release_agent_path));
557         spin_unlock(&release_agent_path_lock);
558         cgroup_kn_unlock(of->kn);
559         return nbytes;
560 }
561 
562 static int cgroup_release_agent_show(struct seq_file *seq, void *v)
563 {
564         struct cgroup *cgrp = seq_css(seq)->cgroup;
565 
566         spin_lock(&release_agent_path_lock);
567         seq_puts(seq, cgrp->root->release_agent_path);
568         spin_unlock(&release_agent_path_lock);
569         seq_putc(seq, '\n');
570         return 0;
571 }
572 
573 static int cgroup_sane_behavior_show(struct seq_file *seq, void *v)
574 {
575         seq_puts(seq, "\n");
576         return 0;
577 }
578 
579 static u64 cgroup_read_notify_on_release(struct cgroup_subsys_state *css,
580                                          struct cftype *cft)
581 {
582         return notify_on_release(css->cgroup);
583 }
584 
585 static int cgroup_write_notify_on_release(struct cgroup_subsys_state *css,
586                                           struct cftype *cft, u64 val)
587 {
588         if (val)
589                 set_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags);
590         else
591                 clear_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags);
592         return 0;
593 }
594 
595 static u64 cgroup_clone_children_read(struct cgroup_subsys_state *css,
596                                       struct cftype *cft)
597 {
598         return test_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags);
599 }
600 
601 static int cgroup_clone_children_write(struct cgroup_subsys_state *css,
602                                        struct cftype *cft, u64 val)
603 {
604         if (val)
605                 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags);
606         else
607                 clear_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags);
608         return 0;
609 }
610 
611 /* cgroup core interface files for the legacy hierarchies */
612 struct cftype cgroup1_base_files[] = {
613         {
614                 .name = "cgroup.procs",
615                 .seq_start = cgroup_pidlist_start,
616                 .seq_next = cgroup_pidlist_next,
617                 .seq_stop = cgroup_pidlist_stop,
618                 .seq_show = cgroup_pidlist_show,
619                 .private = CGROUP_FILE_PROCS,
620                 .write = cgroup1_procs_write,
621         },
622         {
623                 .name = "cgroup.clone_children",
624                 .read_u64 = cgroup_clone_children_read,
625                 .write_u64 = cgroup_clone_children_write,
626         },
627         {
628                 .name = "cgroup.sane_behavior",
629                 .flags = CFTYPE_ONLY_ON_ROOT,
630                 .seq_show = cgroup_sane_behavior_show,
631         },
632         {
633                 .name = "tasks",
634                 .seq_start = cgroup_pidlist_start,
635                 .seq_next = cgroup_pidlist_next,
636                 .seq_stop = cgroup_pidlist_stop,
637                 .seq_show = cgroup_pidlist_show,
638                 .private = CGROUP_FILE_TASKS,
639                 .write = cgroup1_tasks_write,
640         },
641         {
642                 .name = "notify_on_release",
643                 .read_u64 = cgroup_read_notify_on_release,
644                 .write_u64 = cgroup_write_notify_on_release,
645         },
646         {
647                 .name = "release_agent",
648                 .flags = CFTYPE_ONLY_ON_ROOT,
649                 .seq_show = cgroup_release_agent_show,
650                 .write = cgroup_release_agent_write,
651                 .max_write_len = PATH_MAX - 1,
652         },
653         { }     /* terminate */
654 };
655 
656 /* Display information about each subsystem and each hierarchy */
657 int proc_cgroupstats_show(struct seq_file *m, void *v)
658 {
659         struct cgroup_subsys *ss;
660         int i;
661 
662         seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
663         /*
664          * ideally we don't want subsystems moving around while we do this.
665          * cgroup_mutex is also necessary to guarantee an atomic snapshot of
666          * subsys/hierarchy state.
667          */
668         mutex_lock(&cgroup_mutex);
669 
670         for_each_subsys(ss, i)
671                 seq_printf(m, "%s\t%d\t%d\t%d\n",
672                            ss->legacy_name, ss->root->hierarchy_id,
673                            atomic_read(&ss->root->nr_cgrps),
674                            cgroup_ssid_enabled(i));
675 
676         mutex_unlock(&cgroup_mutex);
677         return 0;
678 }
679 
680 /**
681  * cgroupstats_build - build and fill cgroupstats
682  * @stats: cgroupstats to fill information into
683  * @dentry: A dentry entry belonging to the cgroup for which stats have
684  * been requested.
685  *
686  * Build and fill cgroupstats so that taskstats can export it to user
687  * space.
688  */
689 int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry)
690 {
691         struct kernfs_node *kn = kernfs_node_from_dentry(dentry);
692         struct cgroup *cgrp;
693         struct css_task_iter it;
694         struct task_struct *tsk;
695 
696         /* it should be kernfs_node belonging to cgroupfs and is a directory */
697         if (dentry->d_sb->s_type != &cgroup_fs_type || !kn ||
698             kernfs_type(kn) != KERNFS_DIR)
699                 return -EINVAL;
700 
701         mutex_lock(&cgroup_mutex);
702 
703         /*
704          * We aren't being called from kernfs and there's no guarantee on
705          * @kn->priv's validity.  For this and css_tryget_online_from_dir(),
706          * @kn->priv is RCU safe.  Let's do the RCU dancing.
707          */
708         rcu_read_lock();
709         cgrp = rcu_dereference(*(void __rcu __force **)&kn->priv);
710         if (!cgrp || cgroup_is_dead(cgrp)) {
711                 rcu_read_unlock();
712                 mutex_unlock(&cgroup_mutex);
713                 return -ENOENT;
714         }
715         rcu_read_unlock();
716 
717         css_task_iter_start(&cgrp->self, 0, &it);
718         while ((tsk = css_task_iter_next(&it))) {
719                 switch (tsk->state) {
720                 case TASK_RUNNING:
721                         stats->nr_running++;
722                         break;
723                 case TASK_INTERRUPTIBLE:
724                         stats->nr_sleeping++;
725                         break;
726                 case TASK_UNINTERRUPTIBLE:
727                         stats->nr_uninterruptible++;
728                         break;
729                 case TASK_STOPPED:
730                         stats->nr_stopped++;
731                         break;
732                 default:
733                         if (delayacct_is_task_waiting_on_io(tsk))
734                                 stats->nr_io_wait++;
735                         break;
736                 }
737         }
738         css_task_iter_end(&it);
739 
740         mutex_unlock(&cgroup_mutex);
741         return 0;
742 }
743 
744 void cgroup1_check_for_release(struct cgroup *cgrp)
745 {
746         if (notify_on_release(cgrp) && !cgroup_is_populated(cgrp) &&
747             !css_has_online_children(&cgrp->self) && !cgroup_is_dead(cgrp))
748                 schedule_work(&cgrp->release_agent_work);
749 }
750 
751 /*
752  * Notify userspace when a cgroup is released, by running the
753  * configured release agent with the name of the cgroup (path
754  * relative to the root of cgroup file system) as the argument.
755  *
756  * Most likely, this user command will try to rmdir this cgroup.
757  *
758  * This races with the possibility that some other task will be
759  * attached to this cgroup before it is removed, or that some other
760  * user task will 'mkdir' a child cgroup of this cgroup.  That's ok.
761  * The presumed 'rmdir' will fail quietly if this cgroup is no longer
762  * unused, and this cgroup will be reprieved from its death sentence,
763  * to continue to serve a useful existence.  Next time it's released,
764  * we will get notified again, if it still has 'notify_on_release' set.
765  *
766  * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
767  * means only wait until the task is successfully execve()'d.  The
768  * separate release agent task is forked by call_usermodehelper(),
769  * then control in this thread returns here, without waiting for the
770  * release agent task.  We don't bother to wait because the caller of
771  * this routine has no use for the exit status of the release agent
772  * task, so no sense holding our caller up for that.
773  */
774 void cgroup1_release_agent(struct work_struct *work)
775 {
776         struct cgroup *cgrp =
777                 container_of(work, struct cgroup, release_agent_work);
778         char *pathbuf = NULL, *agentbuf = NULL;
779         char *argv[3], *envp[3];
780         int ret;
781 
782         mutex_lock(&cgroup_mutex);
783 
784         pathbuf = kmalloc(PATH_MAX, GFP_KERNEL);
785         agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL);
786         if (!pathbuf || !agentbuf || !strlen(agentbuf))
787                 goto out;
788 
789         spin_lock_irq(&css_set_lock);
790         ret = cgroup_path_ns_locked(cgrp, pathbuf, PATH_MAX, &init_cgroup_ns);
791         spin_unlock_irq(&css_set_lock);
792         if (ret < 0 || ret >= PATH_MAX)
793                 goto out;
794 
795         argv[0] = agentbuf;
796         argv[1] = pathbuf;
797         argv[2] = NULL;
798 
799         /* minimal command environment */
800         envp[0] = "HOME=/";
801         envp[1] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
802         envp[2] = NULL;
803 
804         mutex_unlock(&cgroup_mutex);
805         call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC);
806         goto out_free;
807 out:
808         mutex_unlock(&cgroup_mutex);
809 out_free:
810         kfree(agentbuf);
811         kfree(pathbuf);
812 }
813 
814 /*
815  * cgroup_rename - Only allow simple rename of directories in place.
816  */
817 static int cgroup1_rename(struct kernfs_node *kn, struct kernfs_node *new_parent,
818                           const char *new_name_str)
819 {
820         struct cgroup *cgrp = kn->priv;
821         int ret;
822 
823         if (kernfs_type(kn) != KERNFS_DIR)
824                 return -ENOTDIR;
825         if (kn->parent != new_parent)
826                 return -EIO;
827 
828         /*
829          * We're gonna grab cgroup_mutex which nests outside kernfs
830          * active_ref.  kernfs_rename() doesn't require active_ref
831          * protection.  Break them before grabbing cgroup_mutex.
832          */
833         kernfs_break_active_protection(new_parent);
834         kernfs_break_active_protection(kn);
835 
836         mutex_lock(&cgroup_mutex);
837 
838         ret = kernfs_rename(kn, new_parent, new_name_str);
839         if (!ret)
840                 TRACE_CGROUP_PATH(rename, cgrp);
841 
842         mutex_unlock(&cgroup_mutex);
843 
844         kernfs_unbreak_active_protection(kn);
845         kernfs_unbreak_active_protection(new_parent);
846         return ret;
847 }
848 
849 static int cgroup1_show_options(struct seq_file *seq, struct kernfs_root *kf_root)
850 {
851         struct cgroup_root *root = cgroup_root_from_kf(kf_root);
852         struct cgroup_subsys *ss;
853         int ssid;
854 
855         for_each_subsys(ss, ssid)
856                 if (root->subsys_mask & (1 << ssid))
857                         seq_show_option(seq, ss->legacy_name, NULL);
858         if (root->flags & CGRP_ROOT_NOPREFIX)
859                 seq_puts(seq, ",noprefix");
860         if (root->flags & CGRP_ROOT_XATTR)
861                 seq_puts(seq, ",xattr");
862         if (root->flags & CGRP_ROOT_CPUSET_V2_MODE)
863                 seq_puts(seq, ",cpuset_v2_mode");
864 
865         spin_lock(&release_agent_path_lock);
866         if (strlen(root->release_agent_path))
867                 seq_show_option(seq, "release_agent",
868                                 root->release_agent_path);
869         spin_unlock(&release_agent_path_lock);
870 
871         if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags))
872                 seq_puts(seq, ",clone_children");
873         if (strlen(root->name))
874                 seq_show_option(seq, "name", root->name);
875         return 0;
876 }
877 
878 enum cgroup1_param {
879         Opt_all,
880         Opt_clone_children,
881         Opt_cpuset_v2_mode,
882         Opt_name,
883         Opt_none,
884         Opt_noprefix,
885         Opt_release_agent,
886         Opt_xattr,
887 };
888 
889 const struct fs_parameter_spec cgroup1_fs_parameters[] = {
890         fsparam_flag  ("all",           Opt_all),
891         fsparam_flag  ("clone_children", Opt_clone_children),
892         fsparam_flag  ("cpuset_v2_mode", Opt_cpuset_v2_mode),
893         fsparam_string("name",          Opt_name),
894         fsparam_flag  ("none",          Opt_none),
895         fsparam_flag  ("noprefix",      Opt_noprefix),
896         fsparam_string("release_agent", Opt_release_agent),
897         fsparam_flag  ("xattr",         Opt_xattr),
898         {}
899 };
900 
901 int cgroup1_parse_param(struct fs_context *fc, struct fs_parameter *param)
902 {
903         struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
904         struct cgroup_subsys *ss;
905         struct fs_parse_result result;
906         int opt, i;
907 
908         opt = fs_parse(fc, cgroup1_fs_parameters, param, &result);
909         if (opt == -ENOPARAM) {
910                 if (strcmp(param->key, "source") == 0) {
911                         fc->source = param->string;
912                         param->string = NULL;
913                         return 0;
914                 }
915                 for_each_subsys(ss, i) {
916                         if (strcmp(param->key, ss->legacy_name))
917                                 continue;
918                         ctx->subsys_mask |= (1 << i);
919                         return 0;
920                 }
921                 return invalfc(fc, "Unknown subsys name '%s'", param->key);
922         }
923         if (opt < 0)
924                 return opt;
925 
926         switch (opt) {
927         case Opt_none:
928                 /* Explicitly have no subsystems */
929                 ctx->none = true;
930                 break;
931         case Opt_all:
932                 ctx->all_ss = true;
933                 break;
934         case Opt_noprefix:
935                 ctx->flags |= CGRP_ROOT_NOPREFIX;
936                 break;
937         case Opt_clone_children:
938                 ctx->cpuset_clone_children = true;
939                 break;
940         case Opt_cpuset_v2_mode:
941                 ctx->flags |= CGRP_ROOT_CPUSET_V2_MODE;
942                 break;
943         case Opt_xattr:
944                 ctx->flags |= CGRP_ROOT_XATTR;
945                 break;
946         case Opt_release_agent:
947                 /* Specifying two release agents is forbidden */
948                 if (ctx->release_agent)
949                         return invalfc(fc, "release_agent respecified");
950                 ctx->release_agent = param->string;
951                 param->string = NULL;
952                 break;
953         case Opt_name:
954                 /* blocked by boot param? */
955                 if (cgroup_no_v1_named)
956                         return -ENOENT;
957                 /* Can't specify an empty name */
958                 if (!param->size)
959                         return invalfc(fc, "Empty name");
960                 if (param->size > MAX_CGROUP_ROOT_NAMELEN - 1)
961                         return invalfc(fc, "Name too long");
962                 /* Must match [\w.-]+ */
963                 for (i = 0; i < param->size; i++) {
964                         char c = param->string[i];
965                         if (isalnum(c))
966                                 continue;
967                         if ((c == '.') || (c == '-') || (c == '_'))
968                                 continue;
969                         return invalfc(fc, "Invalid name");
970                 }
971                 /* Specifying two names is forbidden */
972                 if (ctx->name)
973                         return invalfc(fc, "name respecified");
974                 ctx->name = param->string;
975                 param->string = NULL;
976                 break;
977         }
978         return 0;
979 }
980 
981 static int check_cgroupfs_options(struct fs_context *fc)
982 {
983         struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
984         u16 mask = U16_MAX;
985         u16 enabled = 0;
986         struct cgroup_subsys *ss;
987         int i;
988 
989 #ifdef CONFIG_CPUSETS
990         mask = ~((u16)1 << cpuset_cgrp_id);
991 #endif
992         for_each_subsys(ss, i)
993                 if (cgroup_ssid_enabled(i) && !cgroup1_ssid_disabled(i))
994                         enabled |= 1 << i;
995 
996         ctx->subsys_mask &= enabled;
997 
998         /*
999          * In absense of 'none', 'name=' or subsystem name options,
1000          * let's default to 'all'.
1001          */
1002         if (!ctx->subsys_mask && !ctx->none && !ctx->name)
1003                 ctx->all_ss = true;
1004 
1005         if (ctx->all_ss) {
1006                 /* Mutually exclusive option 'all' + subsystem name */
1007                 if (ctx->subsys_mask)
1008                         return invalfc(fc, "subsys name conflicts with all");
1009                 /* 'all' => select all the subsystems */
1010                 ctx->subsys_mask = enabled;
1011         }
1012 
1013         /*
1014          * We either have to specify by name or by subsystems. (So all
1015          * empty hierarchies must have a name).
1016          */
1017         if (!ctx->subsys_mask && !ctx->name)
1018                 return invalfc(fc, "Need name or subsystem set");
1019 
1020         /*
1021          * Option noprefix was introduced just for backward compatibility
1022          * with the old cpuset, so we allow noprefix only if mounting just
1023          * the cpuset subsystem.
1024          */
1025         if ((ctx->flags & CGRP_ROOT_NOPREFIX) && (ctx->subsys_mask & mask))
1026                 return invalfc(fc, "noprefix used incorrectly");
1027 
1028         /* Can't specify "none" and some subsystems */
1029         if (ctx->subsys_mask && ctx->none)
1030                 return invalfc(fc, "none used incorrectly");
1031 
1032         return 0;
1033 }
1034 
1035 int cgroup1_reconfigure(struct fs_context *fc)
1036 {
1037         struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
1038         struct kernfs_root *kf_root = kernfs_root_from_sb(fc->root->d_sb);
1039         struct cgroup_root *root = cgroup_root_from_kf(kf_root);
1040         int ret = 0;
1041         u16 added_mask, removed_mask;
1042 
1043         cgroup_lock_and_drain_offline(&cgrp_dfl_root.cgrp);
1044 
1045         /* See what subsystems are wanted */
1046         ret = check_cgroupfs_options(fc);
1047         if (ret)
1048                 goto out_unlock;
1049 
1050         if (ctx->subsys_mask != root->subsys_mask || ctx->release_agent)
1051                 pr_warn("option changes via remount are deprecated (pid=%d comm=%s)\n",
1052                         task_tgid_nr(current), current->comm);
1053 
1054         added_mask = ctx->subsys_mask & ~root->subsys_mask;
1055         removed_mask = root->subsys_mask & ~ctx->subsys_mask;
1056 
1057         /* Don't allow flags or name to change at remount */
1058         if ((ctx->flags ^ root->flags) ||
1059             (ctx->name && strcmp(ctx->name, root->name))) {
1060                 errorfc(fc, "option or name mismatch, new: 0x%x \"%s\", old: 0x%x \"%s\"",
1061                        ctx->flags, ctx->name ?: "", root->flags, root->name);
1062                 ret = -EINVAL;
1063                 goto out_unlock;
1064         }
1065 
1066         /* remounting is not allowed for populated hierarchies */
1067         if (!list_empty(&root->cgrp.self.children)) {
1068                 ret = -EBUSY;
1069                 goto out_unlock;
1070         }
1071 
1072         ret = rebind_subsystems(root, added_mask);
1073         if (ret)
1074                 goto out_unlock;
1075 
1076         WARN_ON(rebind_subsystems(&cgrp_dfl_root, removed_mask));
1077 
1078         if (ctx->release_agent) {
1079                 spin_lock(&release_agent_path_lock);
1080                 strcpy(root->release_agent_path, ctx->release_agent);
1081                 spin_unlock(&release_agent_path_lock);
1082         }
1083 
1084         trace_cgroup_remount(root);
1085 
1086  out_unlock:
1087         mutex_unlock(&cgroup_mutex);
1088         return ret;
1089 }
1090 
1091 struct kernfs_syscall_ops cgroup1_kf_syscall_ops = {
1092         .rename                 = cgroup1_rename,
1093         .show_options           = cgroup1_show_options,
1094         .mkdir                  = cgroup_mkdir,
1095         .rmdir                  = cgroup_rmdir,
1096         .show_path              = cgroup_show_path,
1097 };
1098 
1099 /*
1100  * The guts of cgroup1 mount - find or create cgroup_root to use.
1101  * Called with cgroup_mutex held; returns 0 on success, -E... on
1102  * error and positive - in case when the candidate is busy dying.
1103  * On success it stashes a reference to cgroup_root into given
1104  * cgroup_fs_context; that reference is *NOT* counting towards the
1105  * cgroup_root refcount.
1106  */
1107 static int cgroup1_root_to_use(struct fs_context *fc)
1108 {
1109         struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
1110         struct cgroup_root *root;
1111         struct cgroup_subsys *ss;
1112         int i, ret;
1113 
1114         /* First find the desired set of subsystems */
1115         ret = check_cgroupfs_options(fc);
1116         if (ret)
1117                 return ret;
1118 
1119         /*
1120          * Destruction of cgroup root is asynchronous, so subsystems may
1121          * still be dying after the previous unmount.  Let's drain the
1122          * dying subsystems.  We just need to ensure that the ones
1123          * unmounted previously finish dying and don't care about new ones
1124          * starting.  Testing ref liveliness is good enough.
1125          */
1126         for_each_subsys(ss, i) {
1127                 if (!(ctx->subsys_mask & (1 << i)) ||
1128                     ss->root == &cgrp_dfl_root)
1129                         continue;
1130 
1131                 if (!percpu_ref_tryget_live(&ss->root->cgrp.self.refcnt))
1132                         return 1;       /* restart */
1133                 cgroup_put(&ss->root->cgrp);
1134         }
1135 
1136         for_each_root(root) {
1137                 bool name_match = false;
1138 
1139                 if (root == &cgrp_dfl_root)
1140                         continue;
1141 
1142                 /*
1143                  * If we asked for a name then it must match.  Also, if
1144                  * name matches but sybsys_mask doesn't, we should fail.
1145                  * Remember whether name matched.
1146                  */
1147                 if (ctx->name) {
1148                         if (strcmp(ctx->name, root->name))
1149                                 continue;
1150                         name_match = true;
1151                 }
1152 
1153                 /*
1154                  * If we asked for subsystems (or explicitly for no
1155                  * subsystems) then they must match.
1156                  */
1157                 if ((ctx->subsys_mask || ctx->none) &&
1158                     (ctx->subsys_mask != root->subsys_mask)) {
1159                         if (!name_match)
1160                                 continue;
1161                         return -EBUSY;
1162                 }
1163 
1164                 if (root->flags ^ ctx->flags)
1165                         pr_warn("new mount options do not match the existing superblock, will be ignored\n");
1166 
1167                 ctx->root = root;
1168                 return 0;
1169         }
1170 
1171         /*
1172          * No such thing, create a new one.  name= matching without subsys
1173          * specification is allowed for already existing hierarchies but we
1174          * can't create new one without subsys specification.
1175          */
1176         if (!ctx->subsys_mask && !ctx->none)
1177                 return invalfc(fc, "No subsys list or none specified");
1178 
1179         /* Hierarchies may only be created in the initial cgroup namespace. */
1180         if (ctx->ns != &init_cgroup_ns)
1181                 return -EPERM;
1182 
1183         root = kzalloc(sizeof(*root), GFP_KERNEL);
1184         if (!root)
1185                 return -ENOMEM;
1186 
1187         ctx->root = root;
1188         init_cgroup_root(ctx);
1189 
1190         ret = cgroup_setup_root(root, ctx->subsys_mask);
1191         if (ret)
1192                 cgroup_free_root(root);
1193         return ret;
1194 }
1195 
1196 int cgroup1_get_tree(struct fs_context *fc)
1197 {
1198         struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
1199         int ret;
1200 
1201         /* Check if the caller has permission to mount. */
1202         if (!ns_capable(ctx->ns->user_ns, CAP_SYS_ADMIN))
1203                 return -EPERM;
1204 
1205         cgroup_lock_and_drain_offline(&cgrp_dfl_root.cgrp);
1206 
1207         ret = cgroup1_root_to_use(fc);
1208         if (!ret && !percpu_ref_tryget_live(&ctx->root->cgrp.self.refcnt))
1209                 ret = 1;        /* restart */
1210 
1211         mutex_unlock(&cgroup_mutex);
1212 
1213         if (!ret)
1214                 ret = cgroup_do_get_tree(fc);
1215 
1216         if (!ret && percpu_ref_is_dying(&ctx->root->cgrp.self.refcnt)) {
1217                 struct super_block *sb = fc->root->d_sb;
1218                 dput(fc->root);
1219                 deactivate_locked_super(sb);
1220                 ret = 1;
1221         }
1222 
1223         if (unlikely(ret > 0)) {
1224                 msleep(10);
1225                 return restart_syscall();
1226         }
1227         return ret;
1228 }
1229 
1230 static int __init cgroup1_wq_init(void)
1231 {
1232         /*
1233          * Used to destroy pidlists and separate to serve as flush domain.
1234          * Cap @max_active to 1 too.
1235          */
1236         cgroup_pidlist_destroy_wq = alloc_workqueue("cgroup_pidlist_destroy",
1237                                                     0, 1);
1238         BUG_ON(!cgroup_pidlist_destroy_wq);
1239         return 0;
1240 }
1241 core_initcall(cgroup1_wq_init);
1242 
1243 static int __init cgroup_no_v1(char *str)
1244 {
1245         struct cgroup_subsys *ss;
1246         char *token;
1247         int i;
1248 
1249         while ((token = strsep(&str, ",")) != NULL) {
1250                 if (!*token)
1251                         continue;
1252 
1253                 if (!strcmp(token, "all")) {
1254                         cgroup_no_v1_mask = U16_MAX;
1255                         continue;
1256                 }
1257 
1258                 if (!strcmp(token, "named")) {
1259                         cgroup_no_v1_named = true;
1260                         continue;
1261                 }
1262 
1263                 for_each_subsys(ss, i) {
1264                         if (strcmp(token, ss->name) &&
1265                             strcmp(token, ss->legacy_name))
1266                                 continue;
1267 
1268                         cgroup_no_v1_mask |= 1 << i;
1269                 }
1270         }
1271         return 1;
1272 }
1273 __setup("cgroup_no_v1=", cgroup_no_v1);
1274 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp