1 // SPDX-License-Identifier: GPL-2.0-only
6  * after a certain limit is reached.
8  * Since it is trivial to hit the task limit without hitting any kmemcg limits
15  * number of processes currently in the cgroup is given by pids.current.
17  * possible to have pids.current > pids.max. However, it is not possible to
18  * violate a cgroup policy through fork(). fork() will return -EAGAIN if forking
21  * To set a cgroup to have no limit, set pids.max to "max". This is the default
23  * stringent limit in the hierarchy is followed).
25  * pids.current tracks all child cgroup hierarchies, so parent/pids.current is
26  * a superset of parent/child/pids.current.
42 	/* Fork failed in subtree because this pids_cgroup limit was hit. */
44 	/* Fork failed in this pids_cgroup because ancestor limit was hit. */
53 	 * Use 64-bit types so that we can safely represent "max" as
57 	atomic64_t			limit;  member
75 	return css_pids(pids->css.parent);  in parent_pids()
85 		return ERR_PTR(-ENOMEM);  in pids_css_alloc()
87 	atomic64_set(&pids->limit, PIDS_MAX);  in pids_css_alloc()
88 	return &pids->css;  in pids_css_alloc()
102 	if (nr_pids > READ_ONCE(p->watermark))  in pids_update_watermark()
103 		WRITE_ONCE(p->watermark, nr_pids);  in pids_update_watermark()
107  * pids_cancel - uncharge the local pid count
120 	WARN_ON_ONCE(atomic64_add_negative(-num, &pids->counter));  in pids_cancel()
124  * pids_uncharge - hierarchically uncharge the pid count
137  * pids_charge - hierarchically charge the pid count
141  * This function does *not* follow the pid limit set. It cannot fail and the new
142  * pid count may exceed the limit. This is only used for reverting failed
143  * attaches, where there is no other way out than violating the limit.
150 		int64_t new = atomic64_add_return(num, &p->counter);  in pids_charge()
157  * pids_try_charge - hierarchically try to charge the pid count
162  * This function follows the set limit. It will fail if the charge would cause
163  * the new value to exceed the hierarchical limit. Returns 0 if the charge
164  * succeeded, otherwise -EAGAIN.
171 		int64_t new = atomic64_add_return(num, &p->counter);  in pids_try_charge()
172 		int64_t limit = atomic64_read(&p->limit);  in pids_try_charge()  local
176 		 * p->limit is %PIDS_MAX then we know that this test will never  in pids_try_charge()
179 		if (new > limit) {  in pids_try_charge()
184 		 * Not technically accurate if we go over limit somewhere up  in pids_try_charge()
197 	return -EAGAIN;  in pids_try_charge()
248 	/* Only log the first time limit is hit. */  in pids_event()
249 	if (atomic64_inc_return(&p->events_local[PIDCG_FORKFAIL]) == 1) {  in pids_event()
251 		pr_cont_cgroup_path(p->css.cgroup);  in pids_event()
256 		cgroup_file_notify(&p->events_local_file);  in pids_event()
260 	atomic64_inc(&pids_over_limit->events_local[PIDCG_MAX]);  in pids_event()
261 	cgroup_file_notify(&pids_over_limit->events_local_file);  in pids_event()
264 		atomic64_inc(&p->events[PIDCG_MAX]);  in pids_event()
265 		cgroup_file_notify(&p->events_file);  in pids_event()
278 	pids = css_pids(cset->subsys[pids_cgrp_id]);  in pids_can_fork()
290 	pids = css_pids(cset->subsys[pids_cgrp_id]);  in pids_cancel_fork()
306 	int64_t limit;  in pids_max_write()  local
311 		limit = PIDS_MAX;  in pids_max_write()
315 	err = kstrtoll(buf, 0, &limit);  in pids_max_write()
319 	if (limit < 0 || limit >= PIDS_MAX)  in pids_max_write()
320 		return -EINVAL;  in pids_max_write()
324 	 * Limit updates don't need to be mutex'd, since it isn't  in pids_max_write()
325 	 * critical that any racing fork()s follow the new limit.  in pids_max_write()
327 	atomic64_set(&pids->limit, limit);  in pids_max_write()
335 	int64_t limit = atomic64_read(&pids->limit);  in pids_max_show()  local
337 	if (limit >= PIDS_MAX)  in pids_max_show()
340 		seq_printf(sf, "%lld\n", limit);  in pids_max_show()
350 	return atomic64_read(&pids->counter);  in pids_current_read()
358 	return READ_ONCE(pids->watermark);  in pids_peak_read()
372 	events = local ? pids->events_local : pids->events;  in __pids_events_show()
398 		.name = "current",
430 		.name = "current",