Revision 07b90056cb15ff9877dca0d8f1b6583d1051f724 authored by Vladimir Oltean on 11 January 2021, 23:09:43 UTC, committed by Jakub Kicinski on 13 January 2021, 02:48:40 UTC
Currently the following happens when a DSA master driver unbinds while there are DSA switches attached to it: $ echo 0000:00:00.5 > /sys/bus/pci/drivers/mscc_felix/unbind ------------[ cut here ]------------ WARNING: CPU: 0 PID: 392 at net/core/dev.c:9507 Call trace: rollback_registered_many+0x5fc/0x688 unregister_netdevice_queue+0x98/0x120 dsa_slave_destroy+0x4c/0x88 dsa_port_teardown.part.16+0x78/0xb0 dsa_tree_teardown_switches+0x58/0xc0 dsa_unregister_switch+0x104/0x1b8 felix_pci_remove+0x24/0x48 pci_device_remove+0x48/0xf0 device_release_driver_internal+0x118/0x1e8 device_driver_detach+0x28/0x38 unbind_store+0xd0/0x100 Located at the above location is this WARN_ON: /* Notifier chain MUST detach us all upper devices. */ WARN_ON(netdev_has_any_upper_dev(dev)); Other stacked interfaces, like VLAN, do indeed listen for NETDEV_UNREGISTER on the real_dev and also unregister themselves at that time, which is clearly the behavior that rollback_registered_many expects. But DSA interfaces are not VLAN. They have backing hardware (platform devices, PCI devices, MDIO, SPI etc) which have a life cycle of their own and we can't just trigger an unregister from the DSA framework when we receive a netdev notifier that the master unregisters. Luckily, there is something we can do, and that is to inform the driver core that we have a runtime dependency to the DSA master interface's device, and create a device link where that is the supplier and we are the consumer. Having this device link will make the DSA switch unbind before the DSA master unbinds, which is enough to avoid the WARN_ON from rollback_registered_many. Note that even before the blamed commit, DSA did nothing intelligent when the master interface got unregistered either. See the discussion here: https://lore.kernel.org/netdev/20200505210253.20311-1-f.fainelli@gmail.com/ But this time, at least the WARN_ON is loud enough that the upper_dev_link commit can be blamed. The advantage with this approach vs dev_hold(master) in the attached link is that the latter is not meant for long term reference counting. With dev_hold, the only thing that will happen is that when the user attempts an unbind of the DSA master, netdev_wait_allrefs will keep waiting and waiting, due to DSA keeping the refcount forever. DSA would not access freed memory corresponding to the master interface, but the unbind would still result in a freeze. Whereas with device links, graceful teardown is ensured. It even works with cascaded DSA trees. $ echo 0000:00:00.2 > /sys/bus/pci/drivers/fsl_enetc/unbind [ 1818.797546] device swp0 left promiscuous mode [ 1819.301112] sja1105 spi2.0: Link is Down [ 1819.307981] DSA: tree 1 torn down [ 1819.312408] device eno2 left promiscuous mode [ 1819.656803] mscc_felix 0000:00:00.5: Link is Down [ 1819.667194] DSA: tree 0 torn down [ 1819.711557] fsl_enetc 0000:00:00.2 eno2: Link is Down This approach allows us to keep the DSA framework absolutely unchanged, and the driver core will just know to unbind us first when the master goes away - as opposed to the large (and probably impossible) rework required if attempting to listen for NETDEV_UNREGISTER. As per the documentation at Documentation/driver-api/device_link.rst, specifying the DL_FLAG_AUTOREMOVE_CONSUMER flag causes the device link to be automatically purged when the consumer fails to probe or later unbinds. So we don't need to keep the consumer_link variable in struct dsa_switch. Fixes: 2f1e8ea726e9 ("net: dsa: link interfaces with the DSA master to get rid of lockdep warnings") Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com> Reviewed-by: Florian Fainelli <f.fainelli@gmail.com> Tested-by: Florian Fainelli <f.fainelli@gmail.com> Link: https://lore.kernel.org/r/20210111230943.3701806-1-olteanv@gmail.com Signed-off-by: Jakub Kicinski <kuba@kernel.org>
1 parent a18caa9
pid_namespace.c
// SPDX-License-Identifier: GPL-2.0-only
/*
* Pid namespaces
*
* Authors:
* (C) 2007 Pavel Emelyanov <xemul@openvz.org>, OpenVZ, SWsoft Inc.
* (C) 2007 Sukadev Bhattiprolu <sukadev@us.ibm.com>, IBM
* Many thanks to Oleg Nesterov for comments and help
*
*/
#include <linux/pid.h>
#include <linux/pid_namespace.h>
#include <linux/user_namespace.h>
#include <linux/syscalls.h>
#include <linux/cred.h>
#include <linux/err.h>
#include <linux/acct.h>
#include <linux/slab.h>
#include <linux/proc_ns.h>
#include <linux/reboot.h>
#include <linux/export.h>
#include <linux/sched/task.h>
#include <linux/sched/signal.h>
#include <linux/idr.h>
static DEFINE_MUTEX(pid_caches_mutex);
static struct kmem_cache *pid_ns_cachep;
/* Write once array, filled from the beginning. */
static struct kmem_cache *pid_cache[MAX_PID_NS_LEVEL];
/*
* creates the kmem cache to allocate pids from.
* @level: pid namespace level
*/
static struct kmem_cache *create_pid_cachep(unsigned int level)
{
/* Level 0 is init_pid_ns.pid_cachep */
struct kmem_cache **pkc = &pid_cache[level - 1];
struct kmem_cache *kc;
char name[4 + 10 + 1];
unsigned int len;
kc = READ_ONCE(*pkc);
if (kc)
return kc;
snprintf(name, sizeof(name), "pid_%u", level + 1);
len = sizeof(struct pid) + level * sizeof(struct upid);
mutex_lock(&pid_caches_mutex);
/* Name collision forces to do allocation under mutex. */
if (!*pkc)
*pkc = kmem_cache_create(name, len, 0, SLAB_HWCACHE_ALIGN, 0);
mutex_unlock(&pid_caches_mutex);
/* current can fail, but someone else can succeed. */
return READ_ONCE(*pkc);
}
static struct ucounts *inc_pid_namespaces(struct user_namespace *ns)
{
return inc_ucount(ns, current_euid(), UCOUNT_PID_NAMESPACES);
}
static void dec_pid_namespaces(struct ucounts *ucounts)
{
dec_ucount(ucounts, UCOUNT_PID_NAMESPACES);
}
static struct pid_namespace *create_pid_namespace(struct user_namespace *user_ns,
struct pid_namespace *parent_pid_ns)
{
struct pid_namespace *ns;
unsigned int level = parent_pid_ns->level + 1;
struct ucounts *ucounts;
int err;
err = -EINVAL;
if (!in_userns(parent_pid_ns->user_ns, user_ns))
goto out;
err = -ENOSPC;
if (level > MAX_PID_NS_LEVEL)
goto out;
ucounts = inc_pid_namespaces(user_ns);
if (!ucounts)
goto out;
err = -ENOMEM;
ns = kmem_cache_zalloc(pid_ns_cachep, GFP_KERNEL);
if (ns == NULL)
goto out_dec;
idr_init(&ns->idr);
ns->pid_cachep = create_pid_cachep(level);
if (ns->pid_cachep == NULL)
goto out_free_idr;
err = ns_alloc_inum(&ns->ns);
if (err)
goto out_free_idr;
ns->ns.ops = &pidns_operations;
refcount_set(&ns->ns.count, 1);
ns->level = level;
ns->parent = get_pid_ns(parent_pid_ns);
ns->user_ns = get_user_ns(user_ns);
ns->ucounts = ucounts;
ns->pid_allocated = PIDNS_ADDING;
return ns;
out_free_idr:
idr_destroy(&ns->idr);
kmem_cache_free(pid_ns_cachep, ns);
out_dec:
dec_pid_namespaces(ucounts);
out:
return ERR_PTR(err);
}
static void delayed_free_pidns(struct rcu_head *p)
{
struct pid_namespace *ns = container_of(p, struct pid_namespace, rcu);
dec_pid_namespaces(ns->ucounts);
put_user_ns(ns->user_ns);
kmem_cache_free(pid_ns_cachep, ns);
}
static void destroy_pid_namespace(struct pid_namespace *ns)
{
ns_free_inum(&ns->ns);
idr_destroy(&ns->idr);
call_rcu(&ns->rcu, delayed_free_pidns);
}
struct pid_namespace *copy_pid_ns(unsigned long flags,
struct user_namespace *user_ns, struct pid_namespace *old_ns)
{
if (!(flags & CLONE_NEWPID))
return get_pid_ns(old_ns);
if (task_active_pid_ns(current) != old_ns)
return ERR_PTR(-EINVAL);
return create_pid_namespace(user_ns, old_ns);
}
void put_pid_ns(struct pid_namespace *ns)
{
struct pid_namespace *parent;
while (ns != &init_pid_ns) {
parent = ns->parent;
if (!refcount_dec_and_test(&ns->ns.count))
break;
destroy_pid_namespace(ns);
ns = parent;
}
}
EXPORT_SYMBOL_GPL(put_pid_ns);
void zap_pid_ns_processes(struct pid_namespace *pid_ns)
{
int nr;
int rc;
struct task_struct *task, *me = current;
int init_pids = thread_group_leader(me) ? 1 : 2;
struct pid *pid;
/* Don't allow any more processes into the pid namespace */
disable_pid_allocation(pid_ns);
/*
* Ignore SIGCHLD causing any terminated children to autoreap.
* This speeds up the namespace shutdown, plus see the comment
* below.
*/
spin_lock_irq(&me->sighand->siglock);
me->sighand->action[SIGCHLD - 1].sa.sa_handler = SIG_IGN;
spin_unlock_irq(&me->sighand->siglock);
/*
* The last thread in the cgroup-init thread group is terminating.
* Find remaining pid_ts in the namespace, signal and wait for them
* to exit.
*
* Note: This signals each threads in the namespace - even those that
* belong to the same thread group, To avoid this, we would have
* to walk the entire tasklist looking a processes in this
* namespace, but that could be unnecessarily expensive if the
* pid namespace has just a few processes. Or we need to
* maintain a tasklist for each pid namespace.
*
*/
rcu_read_lock();
read_lock(&tasklist_lock);
nr = 2;
idr_for_each_entry_continue(&pid_ns->idr, pid, nr) {
task = pid_task(pid, PIDTYPE_PID);
if (task && !__fatal_signal_pending(task))
group_send_sig_info(SIGKILL, SEND_SIG_PRIV, task, PIDTYPE_MAX);
}
read_unlock(&tasklist_lock);
rcu_read_unlock();
/*
* Reap the EXIT_ZOMBIE children we had before we ignored SIGCHLD.
* kernel_wait4() will also block until our children traced from the
* parent namespace are detached and become EXIT_DEAD.
*/
do {
clear_thread_flag(TIF_SIGPENDING);
rc = kernel_wait4(-1, NULL, __WALL, NULL);
} while (rc != -ECHILD);
/*
* kernel_wait4() misses EXIT_DEAD children, and EXIT_ZOMBIE
* process whose parents processes are outside of the pid
* namespace. Such processes are created with setns()+fork().
*
* If those EXIT_ZOMBIE processes are not reaped by their
* parents before their parents exit, they will be reparented
* to pid_ns->child_reaper. Thus pidns->child_reaper needs to
* stay valid until they all go away.
*
* The code relies on the pid_ns->child_reaper ignoring
* SIGCHILD to cause those EXIT_ZOMBIE processes to be
* autoreaped if reparented.
*
* Semantically it is also desirable to wait for EXIT_ZOMBIE
* processes before allowing the child_reaper to be reaped, as
* that gives the invariant that when the init process of a
* pid namespace is reaped all of the processes in the pid
* namespace are gone.
*
* Once all of the other tasks are gone from the pid_namespace
* free_pid() will awaken this task.
*/
for (;;) {
set_current_state(TASK_INTERRUPTIBLE);
if (pid_ns->pid_allocated == init_pids)
break;
schedule();
}
__set_current_state(TASK_RUNNING);
if (pid_ns->reboot)
current->signal->group_exit_code = pid_ns->reboot;
acct_exit_ns(pid_ns);
return;
}
#ifdef CONFIG_CHECKPOINT_RESTORE
static int pid_ns_ctl_handler(struct ctl_table *table, int write,
void *buffer, size_t *lenp, loff_t *ppos)
{
struct pid_namespace *pid_ns = task_active_pid_ns(current);
struct ctl_table tmp = *table;
int ret, next;
if (write && !checkpoint_restore_ns_capable(pid_ns->user_ns))
return -EPERM;
/*
* Writing directly to ns' last_pid field is OK, since this field
* is volatile in a living namespace anyway and a code writing to
* it should synchronize its usage with external means.
*/
next = idr_get_cursor(&pid_ns->idr) - 1;
tmp.data = &next;
ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
if (!ret && write)
idr_set_cursor(&pid_ns->idr, next + 1);
return ret;
}
extern int pid_max;
static struct ctl_table pid_ns_ctl_table[] = {
{
.procname = "ns_last_pid",
.maxlen = sizeof(int),
.mode = 0666, /* permissions are checked in the handler */
.proc_handler = pid_ns_ctl_handler,
.extra1 = SYSCTL_ZERO,
.extra2 = &pid_max,
},
{ }
};
static struct ctl_path kern_path[] = { { .procname = "kernel", }, { } };
#endif /* CONFIG_CHECKPOINT_RESTORE */
int reboot_pid_ns(struct pid_namespace *pid_ns, int cmd)
{
if (pid_ns == &init_pid_ns)
return 0;
switch (cmd) {
case LINUX_REBOOT_CMD_RESTART2:
case LINUX_REBOOT_CMD_RESTART:
pid_ns->reboot = SIGHUP;
break;
case LINUX_REBOOT_CMD_POWER_OFF:
case LINUX_REBOOT_CMD_HALT:
pid_ns->reboot = SIGINT;
break;
default:
return -EINVAL;
}
read_lock(&tasklist_lock);
send_sig(SIGKILL, pid_ns->child_reaper, 1);
read_unlock(&tasklist_lock);
do_exit(0);
/* Not reached */
return 0;
}
static inline struct pid_namespace *to_pid_ns(struct ns_common *ns)
{
return container_of(ns, struct pid_namespace, ns);
}
static struct ns_common *pidns_get(struct task_struct *task)
{
struct pid_namespace *ns;
rcu_read_lock();
ns = task_active_pid_ns(task);
if (ns)
get_pid_ns(ns);
rcu_read_unlock();
return ns ? &ns->ns : NULL;
}
static struct ns_common *pidns_for_children_get(struct task_struct *task)
{
struct pid_namespace *ns = NULL;
task_lock(task);
if (task->nsproxy) {
ns = task->nsproxy->pid_ns_for_children;
get_pid_ns(ns);
}
task_unlock(task);
if (ns) {
read_lock(&tasklist_lock);
if (!ns->child_reaper) {
put_pid_ns(ns);
ns = NULL;
}
read_unlock(&tasklist_lock);
}
return ns ? &ns->ns : NULL;
}
static void pidns_put(struct ns_common *ns)
{
put_pid_ns(to_pid_ns(ns));
}
static int pidns_install(struct nsset *nsset, struct ns_common *ns)
{
struct nsproxy *nsproxy = nsset->nsproxy;
struct pid_namespace *active = task_active_pid_ns(current);
struct pid_namespace *ancestor, *new = to_pid_ns(ns);
if (!ns_capable(new->user_ns, CAP_SYS_ADMIN) ||
!ns_capable(nsset->cred->user_ns, CAP_SYS_ADMIN))
return -EPERM;
/*
* Only allow entering the current active pid namespace
* or a child of the current active pid namespace.
*
* This is required for fork to return a usable pid value and
* this maintains the property that processes and their
* children can not escape their current pid namespace.
*/
if (new->level < active->level)
return -EINVAL;
ancestor = new;
while (ancestor->level > active->level)
ancestor = ancestor->parent;
if (ancestor != active)
return -EINVAL;
put_pid_ns(nsproxy->pid_ns_for_children);
nsproxy->pid_ns_for_children = get_pid_ns(new);
return 0;
}
static struct ns_common *pidns_get_parent(struct ns_common *ns)
{
struct pid_namespace *active = task_active_pid_ns(current);
struct pid_namespace *pid_ns, *p;
/* See if the parent is in the current namespace */
pid_ns = p = to_pid_ns(ns)->parent;
for (;;) {
if (!p)
return ERR_PTR(-EPERM);
if (p == active)
break;
p = p->parent;
}
return &get_pid_ns(pid_ns)->ns;
}
static struct user_namespace *pidns_owner(struct ns_common *ns)
{
return to_pid_ns(ns)->user_ns;
}
const struct proc_ns_operations pidns_operations = {
.name = "pid",
.type = CLONE_NEWPID,
.get = pidns_get,
.put = pidns_put,
.install = pidns_install,
.owner = pidns_owner,
.get_parent = pidns_get_parent,
};
const struct proc_ns_operations pidns_for_children_operations = {
.name = "pid_for_children",
.real_ns_name = "pid",
.type = CLONE_NEWPID,
.get = pidns_for_children_get,
.put = pidns_put,
.install = pidns_install,
.owner = pidns_owner,
.get_parent = pidns_get_parent,
};
static __init int pid_namespaces_init(void)
{
pid_ns_cachep = KMEM_CACHE(pid_namespace, SLAB_PANIC);
#ifdef CONFIG_CHECKPOINT_RESTORE
register_sysctl_paths(kern_path, pid_ns_ctl_table);
#endif
return 0;
}
__initcall(pid_namespaces_init);
Computing file changes ...