cgroup: implement the PIDs subsystem

Adds a new single-purpose PIDs subsystem to limit the number of
tasks that can be forked inside a cgroup. Essentially this is an
implementation of RLIMIT_NPROC that applies to a cgroup rather than a
process tree.

However, it should be noted that organisational operations (adding and
removing tasks from a PIDs hierarchy) will *not* be prevented. Rather,
the number of tasks in the hierarchy cannot exceed the limit through
forking. This is due to the fact that, in the unified hierarchy, attach
cannot fail (and it is not possible for a task to overcome its PIDs
cgroup policy limit by attaching to a child cgroup -- even if migrating
mid-fork it must be able to fork in the parent first).

PIDs are fundamentally a global resource, and it is possible to reach
PID exhaustion inside a cgroup without hitting any reasonable kmemcg
policy. Once you've hit PID exhaustion, you're only in a marginally
better state than OOM. This subsystem allows PID exhaustion inside a
cgroup to be prevented.

Signed-off-by: Aleksa Sarai <cyphar@cyphar.com>
Signed-off-by: Tejun Heo <tj@kernel.org>

2 files changed, 367 insertions, 0 deletions

diff --git a/kernel/Makefile b/kernel/Makefile
index 43c4c920f30a..718fb8afab7a 100644
--- a/kernel/Makefile
+++ b/kernel/Makefile
@@ -55,6 +55,7 @@ obj-$(CONFIG_BACKTRACE_SELF_TEST) += backtracetest.o
 obj-$(CONFIG_COMPAT) += compat.o
 obj-$(CONFIG_CGROUPS) += cgroup.o
 obj-$(CONFIG_CGROUP_FREEZER) += cgroup_freezer.o
+obj-$(CONFIG_CGROUP_PIDS) += cgroup_pids.o
 obj-$(CONFIG_CPUSETS) += cpuset.o
 obj-$(CONFIG_UTS_NS) += utsname.o
 obj-$(CONFIG_USER_NS) += user_namespace.o
diff --git a/kernel/cgroup_pids.c b/kernel/cgroup_pids.c
new file mode 100644
index 000000000000..d75488824ae2
--- /dev/null
+++ b/kernel/cgroup_pids.c
@@ -0,0 +1,366 @@
+/*
+ * Process number limiting controller for cgroups.
+ *
+ * Used to allow a cgroup hierarchy to stop any new processes from fork()ing
+ * after a certain limit is reached.
+ *
+ * Since it is trivial to hit the task limit without hitting any kmemcg limits
+ * in place, PIDs are a fundamental resource. As such, PID exhaustion must be
+ * preventable in the scope of a cgroup hierarchy by allowing resource limiting
+ * of the number of tasks in a cgroup.
+ *
+ * In order to use the `pids` controller, set the maximum number of tasks in
+ * pids.max (this is not available in the root cgroup for obvious reasons). The
+ * number of processes currently in the cgroup is given by pids.current.
+ * Organisational operations are not blocked by cgroup policies, so it is
+ * possible to have pids.current > pids.max. However, it is not possible to
+ * violate a cgroup policy through fork(). fork() will return -EAGAIN if forking
+ * would cause a cgroup policy to be violated.
+ *
+ * To set a cgroup to have no limit, set pids.max to "max". This is the default
+ * for all new cgroups (N.B. that PID limits are hierarchical, so the most
+ * stringent limit in the hierarchy is followed).
+ *
+ * pids.current tracks all child cgroup hierarchies, so parent/pids.current is
+ * a superset of parent/child/pids.current.
+ *
+ * Copyright (C) 2015 Aleksa Sarai <cyphar@cyphar.com>
+ *
+ * This file is subject to the terms and conditions of version 2 of the GNU
+ * General Public License.  See the file COPYING in the main directory of the
+ * Linux distribution for more details.
+ */
+
+#include <linux/kernel.h>
+#include <linux/threads.h>
+#include <linux/atomic.h>
+#include <linux/cgroup.h>
+#include <linux/slab.h>
+
+#define PIDS_MAX (PID_MAX_LIMIT + 1ULL)
+#define PIDS_MAX_STR "max"
+
+struct pids_cgroup {
+	struct cgroup_subsys_state	css;
+
+	/*
+	 * Use 64-bit types so that we can safely represent "max" as
+	 * %PIDS_MAX = (%PID_MAX_LIMIT + 1).
+	 */
+	atomic64_t			counter;
+	int64_t				limit;
+};
+
+static struct pids_cgroup *css_pids(struct cgroup_subsys_state *css)
+{
+	return container_of(css, struct pids_cgroup, css);
+}
+
+static struct pids_cgroup *parent_pids(struct pids_cgroup *pids)
+{
+	return css_pids(pids->css.parent);
+}
+
+static struct cgroup_subsys_state *
+pids_css_alloc(struct cgroup_subsys_state *parent)
+{
+	struct pids_cgroup *pids;
+
+	pids = kzalloc(sizeof(struct pids_cgroup), GFP_KERNEL);
+	if (!pids)
+		return ERR_PTR(-ENOMEM);
+
+	pids->limit = PIDS_MAX;
+	atomic64_set(&pids->counter, 0);
+	return &pids->css;
+}
+
+static void pids_css_free(struct cgroup_subsys_state *css)
+{
+	kfree(css_pids(css));
+}
+
+/**
+ * pids_cancel - uncharge the local pid count
+ * @pids: the pid cgroup state
+ * @num: the number of pids to cancel
+ *
+ * This function will WARN if the pid count goes under 0, because such a case is
+ * a bug in the pids controller proper.
+ */
+static void pids_cancel(struct pids_cgroup *pids, int num)
+{
+	/*
+	 * A negative count (or overflow for that matter) is invalid,
+	 * and indicates a bug in the `pids` controller proper.
+	 */
+	WARN_ON_ONCE(atomic64_add_negative(-num, &pids->counter));
+}
+
+/**
+ * pids_uncharge - hierarchically uncharge the pid count
+ * @pids: the pid cgroup state
+ * @num: the number of pids to uncharge
+ */
+static void pids_uncharge(struct pids_cgroup *pids, int num)
+{
+	struct pids_cgroup *p;
+
+	for (p = pids; p; p = parent_pids(p))
+		pids_cancel(p, num);
+}
+
+/**
+ * pids_charge - hierarchically charge the pid count
+ * @pids: the pid cgroup state
+ * @num: the number of pids to charge
+ *
+ * This function does *not* follow the pid limit set. It cannot fail and the new
+ * pid count may exceed the limit. This is only used for reverting failed
+ * attaches, where there is no other way out than violating the limit.
+ */
+static void pids_charge(struct pids_cgroup *pids, int num)
+{
+	struct pids_cgroup *p;
+
+	for (p = pids; p; p = parent_pids(p))
+		atomic64_add(num, &p->counter);
+}
+
+/**
+ * pids_try_charge - hierarchically try to charge the pid count
+ * @pids: the pid cgroup state
+ * @num: the number of pids to charge
+ *
+ * This function follows the set limit. It will fail if the charge would cause
+ * the new value to exceed the hierarchical limit. Returns 0 if the charge
+ * succeded, otherwise -EAGAIN.
+ */
+static int pids_try_charge(struct pids_cgroup *pids, int num)
+{
+	struct pids_cgroup *p, *q;
+
+	for (p = pids; p; p = parent_pids(p)) {
+		int64_t new = atomic64_add_return(num, &p->counter);
+
+		/*
+		 * Since new is capped to the maximum number of pid_t, if
+		 * p->limit is %PIDS_MAX then we know that this test will never
+		 * fail.
+		 */
+		if (new > p->limit)
+			goto revert;
+	}
+
+	return 0;
+
+revert:
+	for (q = pids; q != p; q = parent_pids(q))
+		pids_cancel(q, num);
+	pids_cancel(p, num);
+
+	return -EAGAIN;
+}
+
+static int pids_can_attach(struct cgroup_subsys_state *css,
+			   struct cgroup_taskset *tset)
+{
+	struct pids_cgroup *pids = css_pids(css);
+	struct task_struct *task;
+
+	cgroup_taskset_for_each(task, tset) {
+		struct cgroup_subsys_state *old_css;
+		struct pids_cgroup *old_pids;
+
+		/*
+		 * Grab a ref to each task's css. We don't drop the ref until
+		 * we either fail and hit ->cancel_attach() or succeed and hit
+		 * ->attach().
+		 */
+		old_css = task_get_css(task, pids_cgrp_id);
+		old_pids = css_pids(old_css);
+
+		pids_charge(pids, 1);
+		pids_uncharge(old_pids, 1);
+	}
+
+	return 0;
+}
+
+static void pids_cancel_attach(struct cgroup_subsys_state *css,
+			       struct cgroup_taskset *tset)
+{
+	struct pids_cgroup *pids = css_pids(css);
+	struct task_struct *task;
+
+	cgroup_taskset_for_each(task, tset) {
+		struct cgroup_subsys_state *old_css;
+		struct pids_cgroup *old_pids;
+
+		old_css = task_css(task, pids_cgrp_id);
+		old_pids = css_pids(old_css);
+
+		pids_charge(old_pids, 1);
+		pids_uncharge(pids, 1);
+		css_put(old_css);
+	}
+}
+
+static void pids_attach(struct cgroup_subsys_state *css,
+			struct cgroup_taskset *tset)
+{
+	struct task_struct *task;
+
+	cgroup_taskset_for_each(task, tset)
+		css_put(task_css(task, pids_cgrp_id));
+}
+
+static int pids_can_fork(struct task_struct *task, void **priv_p)
+{
+	struct cgroup_subsys_state *css;
+	struct pids_cgroup *pids;
+	int err;
+
+	/*
+	 * Use the "current" task_css for the pids subsystem as the tentative
+	 * css. It is possible we will charge the wrong hierarchy, in which
+	 * case we will forcefully revert/reapply the charge on the right
+	 * hierarchy after it is committed to the task proper.
+	 */
+	css = task_get_css(current, pids_cgrp_id);
+	pids = css_pids(css);
+
+	err = pids_try_charge(pids, 1);
+	if (err)
+		goto err_css_put;
+
+	*priv_p = css;
+	return 0;
+
+err_css_put:
+	css_put(css);
+	return err;
+}
+
+static void pids_cancel_fork(struct task_struct *task, void *priv)
+{
+	struct cgroup_subsys_state *css = priv;
+	struct pids_cgroup *pids = css_pids(css);
+
+	pids_uncharge(pids, 1);
+	css_put(css);
+}
+
+static void pids_fork(struct task_struct *task, void *priv)
+{
+	struct cgroup_subsys_state *css;
+	struct cgroup_subsys_state *old_css = priv;
+	struct pids_cgroup *pids;
+	struct pids_cgroup *old_pids = css_pids(old_css);
+
+	css = task_get_css(task, pids_cgrp_id);
+	pids = css_pids(css);
+
+	/*
+	 * If the association has changed, we have to revert and reapply the
+	 * charge/uncharge on the wrong hierarchy to the current one. Since
+	 * the association can only change due to an organisation event, its
+	 * okay for us to ignore the limit in this case.
+	 */
+	if (pids != old_pids) {
+		pids_uncharge(old_pids, 1);
+		pids_charge(pids, 1);
+	}
+
+	css_put(css);
+	css_put(old_css);
+}
+
+static void pids_exit(struct cgroup_subsys_state *css,
+		      struct cgroup_subsys_state *old_css,
+		      struct task_struct *task)
+{
+	struct pids_cgroup *pids = css_pids(old_css);
+
+	pids_uncharge(pids, 1);
+}
+
+static ssize_t pids_max_write(struct kernfs_open_file *of, char *buf,
+			      size_t nbytes, loff_t off)
+{
+	struct cgroup_subsys_state *css = of_css(of);
+	struct pids_cgroup *pids = css_pids(css);
+	int64_t limit;
+	int err;
+
+	buf = strstrip(buf);
+	if (!strcmp(buf, PIDS_MAX_STR)) {
+		limit = PIDS_MAX;
+		goto set_limit;
+	}
+
+	err = kstrtoll(buf, 0, &limit);
+	if (err)
+		return err;
+
+	if (limit < 0 || limit >= PIDS_MAX)
+		return -EINVAL;
+
+set_limit:
+	/*
+	 * Limit updates don't need to be mutex'd, since it isn't
+	 * critical that any racing fork()s follow the new limit.
+	 */
+	pids->limit = limit;
+	return nbytes;
+}
+
+static int pids_max_show(struct seq_file *sf, void *v)
+{
+	struct cgroup_subsys_state *css = seq_css(sf);
+	struct pids_cgroup *pids = css_pids(css);
+	int64_t limit = pids->limit;
+
+	if (limit >= PIDS_MAX)
+		seq_printf(sf, "%s\n", PIDS_MAX_STR);
+	else
+		seq_printf(sf, "%lld\n", limit);
+
+	return 0;
+}
+
+static s64 pids_current_read(struct cgroup_subsys_state *css,
+			     struct cftype *cft)
+{
+	struct pids_cgroup *pids = css_pids(css);
+
+	return atomic64_read(&pids->counter);
+}
+
+static struct cftype pids_files[] = {
+	{
+		.name = "max",
+		.write = pids_max_write,
+		.seq_show = pids_max_show,
+		.flags = CFTYPE_NOT_ON_ROOT,
+	},
+	{
+		.name = "current",
+		.read_s64 = pids_current_read,
+	},
+	{ }	/* terminate */
+};
+
+struct cgroup_subsys pids_cgrp_subsys = {
+	.css_alloc	= pids_css_alloc,
+	.css_free	= pids_css_free,
+	.attach		= pids_attach,
+	.can_attach 	= pids_can_attach,
+	.cancel_attach 	= pids_cancel_attach,
+	.can_fork	= pids_can_fork,
+	.cancel_fork	= pids_cancel_fork,
+	.fork		= pids_fork,
+	.exit		= pids_exit,
+	.legacy_cftypes	= pids_files,
+	.dfl_cftypes	= pids_files,
+};