2 * Process number limiting controller for cgroups.
4 * Used to allow a cgroup hierarchy to stop any new processes from fork()ing
5 * after a certain limit is reached.
7 * Since it is trivial to hit the task limit without hitting any kmemcg limits
8 * in place, PIDs are a fundamental resource. As such, PID exhaustion must be
9 * preventable in the scope of a cgroup hierarchy by allowing resource limiting
10 * of the number of tasks in a cgroup.
12 * In order to use the `pids` controller, set the maximum number of tasks in
13 * pids.max (this is not available in the root cgroup for obvious reasons). The
14 * number of processes currently in the cgroup is given by pids.current.
15 * Organisational operations are not blocked by cgroup policies, so it is
16 * possible to have pids.current > pids.max. However, it is not possible to
17 * violate a cgroup policy through fork(). fork() will return -EAGAIN if forking
18 * would cause a cgroup policy to be violated.
20 * To set a cgroup to have no limit, set pids.max to "max". This is the default
21 * for all new cgroups (N.B. that PID limits are hierarchical, so the most
22 * stringent limit in the hierarchy is followed).
24 * pids.current tracks all child cgroup hierarchies, so parent/pids.current is
25 * a superset of parent/child/pids.current.
27 * Copyright (C) 2015 Aleksa Sarai <cyphar@cyphar.com>
29 * This file is subject to the terms and conditions of version 2 of the GNU
30 * General Public License. See the file COPYING in the main directory of the
31 * Linux distribution for more details.
34 #include <linux/kernel.h>
35 #include <linux/threads.h>
36 #include <linux/atomic.h>
37 #include <linux/cgroup.h>
38 #include <linux/slab.h>
40 #define PIDS_MAX (PID_MAX_LIMIT + 1ULL)
41 #define PIDS_MAX_STR "max"
44 struct cgroup_subsys_state css;
47 * Use 64-bit types so that we can safely represent "max" as
48 * %PIDS_MAX = (%PID_MAX_LIMIT + 1).
54 static struct pids_cgroup *css_pids(struct cgroup_subsys_state *css)
56 return container_of(css, struct pids_cgroup, css);
59 static struct pids_cgroup *parent_pids(struct pids_cgroup *pids)
61 return css_pids(pids->css.parent);
64 static struct cgroup_subsys_state *
65 pids_css_alloc(struct cgroup_subsys_state *parent)
67 struct pids_cgroup *pids;
69 pids = kzalloc(sizeof(struct pids_cgroup), GFP_KERNEL);
71 return ERR_PTR(-ENOMEM);
73 atomic64_set(&pids->counter, 0);
74 atomic64_set(&pids->limit, PIDS_MAX);
78 static void pids_css_free(struct cgroup_subsys_state *css)
84 * pids_cancel - uncharge the local pid count
85 * @pids: the pid cgroup state
86 * @num: the number of pids to cancel
88 * This function will WARN if the pid count goes under 0, because such a case is
89 * a bug in the pids controller proper.
91 static void pids_cancel(struct pids_cgroup *pids, int num)
94 * A negative count (or overflow for that matter) is invalid,
95 * and indicates a bug in the `pids` controller proper.
97 WARN_ON_ONCE(atomic64_add_negative(-num, &pids->counter));
101 * pids_uncharge - hierarchically uncharge the pid count
102 * @pids: the pid cgroup state
103 * @num: the number of pids to uncharge
105 static void pids_uncharge(struct pids_cgroup *pids, int num)
107 struct pids_cgroup *p;
109 for (p = pids; parent_pids(p); p = parent_pids(p))
114 * pids_charge - hierarchically charge the pid count
115 * @pids: the pid cgroup state
116 * @num: the number of pids to charge
118 * This function does *not* follow the pid limit set. It cannot fail and the new
119 * pid count may exceed the limit. This is only used for reverting failed
120 * attaches, where there is no other way out than violating the limit.
122 static void pids_charge(struct pids_cgroup *pids, int num)
124 struct pids_cgroup *p;
126 for (p = pids; parent_pids(p); p = parent_pids(p))
127 atomic64_add(num, &p->counter);
131 * pids_try_charge - hierarchically try to charge the pid count
132 * @pids: the pid cgroup state
133 * @num: the number of pids to charge
135 * This function follows the set limit. It will fail if the charge would cause
136 * the new value to exceed the hierarchical limit. Returns 0 if the charge
137 * succeded, otherwise -EAGAIN.
139 static int pids_try_charge(struct pids_cgroup *pids, int num)
141 struct pids_cgroup *p, *q;
143 for (p = pids; parent_pids(p); p = parent_pids(p)) {
144 int64_t new = atomic64_add_return(num, &p->counter);
145 int64_t limit = atomic64_read(&p->limit);
148 * Since new is capped to the maximum number of pid_t, if
149 * p->limit is %PIDS_MAX then we know that this test will never
159 for (q = pids; q != p; q = parent_pids(q))
166 static int pids_can_attach(struct cgroup_taskset *tset)
168 struct task_struct *task;
169 struct cgroup_subsys_state *dst_css;
171 cgroup_taskset_for_each(task, dst_css, tset) {
172 struct pids_cgroup *pids = css_pids(dst_css);
173 struct cgroup_subsys_state *old_css;
174 struct pids_cgroup *old_pids;
177 * No need to pin @old_css between here and cancel_attach()
178 * because cgroup core protects it from being freed before
179 * the migration completes or fails.
181 old_css = task_css(task, pids_cgrp_id);
182 old_pids = css_pids(old_css);
184 pids_charge(pids, 1);
185 pids_uncharge(old_pids, 1);
191 static void pids_cancel_attach(struct cgroup_taskset *tset)
193 struct task_struct *task;
194 struct cgroup_subsys_state *dst_css;
196 cgroup_taskset_for_each(task, dst_css, tset) {
197 struct pids_cgroup *pids = css_pids(dst_css);
198 struct cgroup_subsys_state *old_css;
199 struct pids_cgroup *old_pids;
201 old_css = task_css(task, pids_cgrp_id);
202 old_pids = css_pids(old_css);
204 pids_charge(old_pids, 1);
205 pids_uncharge(pids, 1);
210 * task_css_check(true) in pids_can_fork() and pids_cancel_fork() relies
211 * on threadgroup_change_begin() held by the copy_process().
213 static int pids_can_fork(struct task_struct *task, void **priv_p)
215 struct cgroup_subsys_state *css;
216 struct pids_cgroup *pids;
218 css = task_css_check(current, pids_cgrp_id, true);
219 pids = css_pids(css);
220 return pids_try_charge(pids, 1);
223 static void pids_cancel_fork(struct task_struct *task, void *priv)
225 struct cgroup_subsys_state *css;
226 struct pids_cgroup *pids;
228 css = task_css_check(current, pids_cgrp_id, true);
229 pids = css_pids(css);
230 pids_uncharge(pids, 1);
233 static void pids_free(struct task_struct *task)
235 struct pids_cgroup *pids = css_pids(task_css(task, pids_cgrp_id));
237 pids_uncharge(pids, 1);
240 static ssize_t pids_max_write(struct kernfs_open_file *of, char *buf,
241 size_t nbytes, loff_t off)
243 struct cgroup_subsys_state *css = of_css(of);
244 struct pids_cgroup *pids = css_pids(css);
249 if (!strcmp(buf, PIDS_MAX_STR)) {
254 err = kstrtoll(buf, 0, &limit);
258 if (limit < 0 || limit >= PIDS_MAX)
263 * Limit updates don't need to be mutex'd, since it isn't
264 * critical that any racing fork()s follow the new limit.
266 atomic64_set(&pids->limit, limit);
270 static int pids_max_show(struct seq_file *sf, void *v)
272 struct cgroup_subsys_state *css = seq_css(sf);
273 struct pids_cgroup *pids = css_pids(css);
274 int64_t limit = atomic64_read(&pids->limit);
276 if (limit >= PIDS_MAX)
277 seq_printf(sf, "%s\n", PIDS_MAX_STR);
279 seq_printf(sf, "%lld\n", limit);
284 static s64 pids_current_read(struct cgroup_subsys_state *css,
287 struct pids_cgroup *pids = css_pids(css);
289 return atomic64_read(&pids->counter);
292 static struct cftype pids_files[] = {
295 .write = pids_max_write,
296 .seq_show = pids_max_show,
297 .flags = CFTYPE_NOT_ON_ROOT,
301 .read_s64 = pids_current_read,
302 .flags = CFTYPE_NOT_ON_ROOT,
307 struct cgroup_subsys pids_cgrp_subsys = {
308 .css_alloc = pids_css_alloc,
309 .css_free = pids_css_free,
310 .can_attach = pids_can_attach,
311 .cancel_attach = pids_cancel_attach,
312 .can_fork = pids_can_fork,
313 .cancel_fork = pids_cancel_fork,
315 .legacy_cftypes = pids_files,
316 .dfl_cftypes = pids_files,