linux-stable/net/core/netprio_cgroup.c
Zefan Li 090e28b229 netprio_cgroup: Fix unlimited memory leak of v2 cgroups
If systemd is configured to use hybrid mode which enables the use of
both cgroup v1 and v2, systemd will create new cgroup on both the default
root (v2) and netprio_cgroup hierarchy (v1) for a new session and attach
task to the two cgroups. If the task does some network thing then the v2
cgroup can never be freed after the session exited.

One of our machines ran into OOM due to this memory leak.

In the scenario described above when sk_alloc() is called
cgroup_sk_alloc() thought it's in v2 mode, so it stores
the cgroup pointer in sk->sk_cgrp_data and increments
the cgroup refcnt, but then sock_update_netprioidx()
thought it's in v1 mode, so it stores netprioidx value
in sk->sk_cgrp_data, so the cgroup refcnt will never be freed.

Currently we do the mode switch when someone writes to the ifpriomap
cgroup control file. The easiest fix is to also do the switch when
a task is attached to a new cgroup.

Fixes: bd1060a1d6 ("sock, cgroup: add sock->sk_cgroup")
Reported-by: Yang Yingliang <yangyingliang@huawei.com>
Tested-by: Yang Yingliang <yangyingliang@huawei.com>
Signed-off-by: Zefan Li <lizefan@huawei.com>
Acked-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2020-05-09 20:59:21 -07:00

302 lines
6.6 KiB
C

// SPDX-License-Identifier: GPL-2.0-or-later
/*
* net/core/netprio_cgroup.c Priority Control Group
*
* Authors: Neil Horman <nhorman@tuxdriver.com>
*/
#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
#include <linux/module.h>
#include <linux/slab.h>
#include <linux/types.h>
#include <linux/string.h>
#include <linux/errno.h>
#include <linux/skbuff.h>
#include <linux/cgroup.h>
#include <linux/rcupdate.h>
#include <linux/atomic.h>
#include <linux/sched/task.h>
#include <net/rtnetlink.h>
#include <net/pkt_cls.h>
#include <net/sock.h>
#include <net/netprio_cgroup.h>
#include <linux/fdtable.h>
/*
* netprio allocates per-net_device priomap array which is indexed by
* css->id. Limiting css ID to 16bits doesn't lose anything.
*/
#define NETPRIO_ID_MAX USHRT_MAX
#define PRIOMAP_MIN_SZ 128
/*
* Extend @dev->priomap so that it's large enough to accommodate
* @target_idx. @dev->priomap.priomap_len > @target_idx after successful
* return. Must be called under rtnl lock.
*/
static int extend_netdev_table(struct net_device *dev, u32 target_idx)
{
struct netprio_map *old, *new;
size_t new_sz, new_len;
/* is the existing priomap large enough? */
old = rtnl_dereference(dev->priomap);
if (old && old->priomap_len > target_idx)
return 0;
/*
* Determine the new size. Let's keep it power-of-two. We start
* from PRIOMAP_MIN_SZ and double it until it's large enough to
* accommodate @target_idx.
*/
new_sz = PRIOMAP_MIN_SZ;
while (true) {
new_len = (new_sz - offsetof(struct netprio_map, priomap)) /
sizeof(new->priomap[0]);
if (new_len > target_idx)
break;
new_sz *= 2;
/* overflowed? */
if (WARN_ON(new_sz < PRIOMAP_MIN_SZ))
return -ENOSPC;
}
/* allocate & copy */
new = kzalloc(new_sz, GFP_KERNEL);
if (!new)
return -ENOMEM;
if (old)
memcpy(new->priomap, old->priomap,
old->priomap_len * sizeof(old->priomap[0]));
new->priomap_len = new_len;
/* install the new priomap */
rcu_assign_pointer(dev->priomap, new);
if (old)
kfree_rcu(old, rcu);
return 0;
}
/**
* netprio_prio - return the effective netprio of a cgroup-net_device pair
* @css: css part of the target pair
* @dev: net_device part of the target pair
*
* Should be called under RCU read or rtnl lock.
*/
static u32 netprio_prio(struct cgroup_subsys_state *css, struct net_device *dev)
{
struct netprio_map *map = rcu_dereference_rtnl(dev->priomap);
int id = css->id;
if (map && id < map->priomap_len)
return map->priomap[id];
return 0;
}
/**
* netprio_set_prio - set netprio on a cgroup-net_device pair
* @css: css part of the target pair
* @dev: net_device part of the target pair
* @prio: prio to set
*
* Set netprio to @prio on @css-@dev pair. Should be called under rtnl
* lock and may fail under memory pressure for non-zero @prio.
*/
static int netprio_set_prio(struct cgroup_subsys_state *css,
struct net_device *dev, u32 prio)
{
struct netprio_map *map;
int id = css->id;
int ret;
/* avoid extending priomap for zero writes */
map = rtnl_dereference(dev->priomap);
if (!prio && (!map || map->priomap_len <= id))
return 0;
ret = extend_netdev_table(dev, id);
if (ret)
return ret;
map = rtnl_dereference(dev->priomap);
map->priomap[id] = prio;
return 0;
}
static struct cgroup_subsys_state *
cgrp_css_alloc(struct cgroup_subsys_state *parent_css)
{
struct cgroup_subsys_state *css;
css = kzalloc(sizeof(*css), GFP_KERNEL);
if (!css)
return ERR_PTR(-ENOMEM);
return css;
}
static int cgrp_css_online(struct cgroup_subsys_state *css)
{
struct cgroup_subsys_state *parent_css = css->parent;
struct net_device *dev;
int ret = 0;
if (css->id > NETPRIO_ID_MAX)
return -ENOSPC;
if (!parent_css)
return 0;
rtnl_lock();
/*
* Inherit prios from the parent. As all prios are set during
* onlining, there is no need to clear them on offline.
*/
for_each_netdev(&init_net, dev) {
u32 prio = netprio_prio(parent_css, dev);
ret = netprio_set_prio(css, dev, prio);
if (ret)
break;
}
rtnl_unlock();
return ret;
}
static void cgrp_css_free(struct cgroup_subsys_state *css)
{
kfree(css);
}
static u64 read_prioidx(struct cgroup_subsys_state *css, struct cftype *cft)
{
return css->id;
}
static int read_priomap(struct seq_file *sf, void *v)
{
struct net_device *dev;
rcu_read_lock();
for_each_netdev_rcu(&init_net, dev)
seq_printf(sf, "%s %u\n", dev->name,
netprio_prio(seq_css(sf), dev));
rcu_read_unlock();
return 0;
}
static ssize_t write_priomap(struct kernfs_open_file *of,
char *buf, size_t nbytes, loff_t off)
{
char devname[IFNAMSIZ + 1];
struct net_device *dev;
u32 prio;
int ret;
if (sscanf(buf, "%"__stringify(IFNAMSIZ)"s %u", devname, &prio) != 2)
return -EINVAL;
dev = dev_get_by_name(&init_net, devname);
if (!dev)
return -ENODEV;
cgroup_sk_alloc_disable();
rtnl_lock();
ret = netprio_set_prio(of_css(of), dev, prio);
rtnl_unlock();
dev_put(dev);
return ret ?: nbytes;
}
static int update_netprio(const void *v, struct file *file, unsigned n)
{
int err;
struct socket *sock = sock_from_file(file, &err);
if (sock) {
spin_lock(&cgroup_sk_update_lock);
sock_cgroup_set_prioidx(&sock->sk->sk_cgrp_data,
(unsigned long)v);
spin_unlock(&cgroup_sk_update_lock);
}
return 0;
}
static void net_prio_attach(struct cgroup_taskset *tset)
{
struct task_struct *p;
struct cgroup_subsys_state *css;
cgroup_sk_alloc_disable();
cgroup_taskset_for_each(p, css, tset) {
void *v = (void *)(unsigned long)css->id;
task_lock(p);
iterate_fd(p->files, 0, update_netprio, v);
task_unlock(p);
}
}
static struct cftype ss_files[] = {
{
.name = "prioidx",
.read_u64 = read_prioidx,
},
{
.name = "ifpriomap",
.seq_show = read_priomap,
.write = write_priomap,
},
{ } /* terminate */
};
struct cgroup_subsys net_prio_cgrp_subsys = {
.css_alloc = cgrp_css_alloc,
.css_online = cgrp_css_online,
.css_free = cgrp_css_free,
.attach = net_prio_attach,
.legacy_cftypes = ss_files,
};
static int netprio_device_event(struct notifier_block *unused,
unsigned long event, void *ptr)
{
struct net_device *dev = netdev_notifier_info_to_dev(ptr);
struct netprio_map *old;
/*
* Note this is called with rtnl_lock held so we have update side
* protection on our rcu assignments
*/
switch (event) {
case NETDEV_UNREGISTER:
old = rtnl_dereference(dev->priomap);
RCU_INIT_POINTER(dev->priomap, NULL);
if (old)
kfree_rcu(old, rcu);
break;
}
return NOTIFY_DONE;
}
static struct notifier_block netprio_device_notifier = {
.notifier_call = netprio_device_event
};
static int __init init_cgroup_netprio(void)
{
register_netdevice_notifier(&netprio_device_notifier);
return 0;
}
subsys_initcall(init_cgroup_netprio);