Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
Merge patch series "ns: rework reference counting"
Christian Brauner <brauner@kernel.org> says:
Stop open accesses to the reference counts and cargo-culting the same
code in all namespace. Use a set of dedicated helpers and make the
actual count private.
* patches from https://lore.kernel.org/20250918-work-namespace-ns_ref-v1-0-1b0a98ee041e@kernel.org:
ns: rename to __ns_ref
nsfs: port to ns_ref_*() helpers
net: port to ns_ref_*() helpers
uts: port to ns_ref_*() helpers
ipv4: use check_net()
net: use check_net()
net-sysfs: use check_net()
user: port to ns_ref_*() helpers
time: port to ns_ref_*() helpers
pid: port to ns_ref_*() helpers
ipc: port to ns_ref_*() helpers
cgroup: port to ns_ref_*() helpers
mnt: port to ns_ref_*() helpers
ns: add reference count helpers
Signed-off-by: Christian Brauner <brauner@kernel.org>
+74
-49
+1
-1
fs/mount.h
+1
-1
fs/mount.h
+3
-3
fs/namespace.c
+3
-3
fs/namespace.c
···
2110
2110
* the mount namespace and it might already be on its
2111
2111
* deathbed.
2112
2112
*/
2113
-
if (!refcount_inc_not_zero(&mntns->ns.count))
2113
+
if (!ns_ref_get(mntns))
2114
2114
continue;
2115
2115
2116
2116
return mntns;
···
6015
6015
.ns.inum = PROC_MNT_INIT_INO,
6016
6016
.ns.ops = &mntns_operations,
6017
6017
.user_ns = &init_user_ns,
6018
-
.ns.count = REFCOUNT_INIT(1),
6018
+
.ns.__ns_ref = REFCOUNT_INIT(1),
6019
6019
.passive = REFCOUNT_INIT(1),
6020
6020
.mounts = RB_ROOT,
6021
6021
.poll = __WAIT_QUEUE_HEAD_INITIALIZER(init_mnt_ns.poll),
···
6084
6084
6085
6085
void put_mnt_ns(struct mnt_namespace *ns)
6086
6086
{
6087
-
if (!refcount_dec_and_test(&ns->ns.count))
6087
+
if (!ns_ref_put(ns))
6088
6088
return;
6089
6089
namespace_lock();
6090
6090
emptied_ns = ns;
+1
-1
fs/nsfs.c
+1
-1
fs/nsfs.c
+2
-2
include/linux/cgroup_namespace.h
+2
-2
include/linux/cgroup_namespace.h
···
29
29
30
30
static inline void get_cgroup_ns(struct cgroup_namespace *ns)
31
31
{
32
-
refcount_inc(&ns->ns.count);
32
+
ns_ref_inc(ns);
33
33
}
34
34
35
35
static inline void put_cgroup_ns(struct cgroup_namespace *ns)
36
36
{
37
-
if (refcount_dec_and_test(&ns->ns.count))
37
+
if (ns_ref_put(ns))
38
38
free_cgroup_ns(ns);
39
39
}
40
40
+2
-2
include/linux/ipc_namespace.h
+2
-2
include/linux/ipc_namespace.h
···
140
140
static inline struct ipc_namespace *get_ipc_ns(struct ipc_namespace *ns)
141
141
{
142
142
if (ns)
143
-
refcount_inc(&ns->ns.count);
143
+
ns_ref_inc(ns);
144
144
return ns;
145
145
}
146
146
147
147
static inline struct ipc_namespace *get_ipc_ns_not_zero(struct ipc_namespace *ns)
148
148
{
149
149
if (ns) {
150
-
if (refcount_inc_not_zero(&ns->ns.count))
150
+
if (ns_ref_get(ns))
151
151
return ns;
152
152
}
153
153
+36
-11
include/linux/ns_common.h
+36
-11
include/linux/ns_common.h
···
29
29
struct dentry *stashed;
30
30
const struct proc_ns_operations *ops;
31
31
unsigned int inum;
32
-
refcount_t count;
32
+
refcount_t __ns_ref; /* do not use directly */
33
33
union {
34
34
struct {
35
35
u64 ns_id;
···
43
43
int __ns_common_init(struct ns_common *ns, const struct proc_ns_operations *ops, int inum);
44
44
void __ns_common_free(struct ns_common *ns);
45
45
46
-
#define to_ns_common(__ns) \
47
-
_Generic((__ns), \
48
-
struct cgroup_namespace *: &(__ns)->ns, \
49
-
struct ipc_namespace *: &(__ns)->ns, \
50
-
struct mnt_namespace *: &(__ns)->ns, \
51
-
struct net *: &(__ns)->ns, \
52
-
struct pid_namespace *: &(__ns)->ns, \
53
-
struct time_namespace *: &(__ns)->ns, \
54
-
struct user_namespace *: &(__ns)->ns, \
55
-
struct uts_namespace *: &(__ns)->ns)
46
+
#define to_ns_common(__ns) \
47
+
_Generic((__ns), \
48
+
struct cgroup_namespace *: &(__ns)->ns, \
49
+
const struct cgroup_namespace *: &(__ns)->ns, \
50
+
struct ipc_namespace *: &(__ns)->ns, \
51
+
const struct ipc_namespace *: &(__ns)->ns, \
52
+
struct mnt_namespace *: &(__ns)->ns, \
53
+
const struct mnt_namespace *: &(__ns)->ns, \
54
+
struct net *: &(__ns)->ns, \
55
+
const struct net *: &(__ns)->ns, \
56
+
struct pid_namespace *: &(__ns)->ns, \
57
+
const struct pid_namespace *: &(__ns)->ns, \
58
+
struct time_namespace *: &(__ns)->ns, \
59
+
const struct time_namespace *: &(__ns)->ns, \
60
+
struct user_namespace *: &(__ns)->ns, \
61
+
const struct user_namespace *: &(__ns)->ns, \
62
+
struct uts_namespace *: &(__ns)->ns, \
63
+
const struct uts_namespace *: &(__ns)->ns)
56
64
57
65
#define ns_init_inum(__ns) \
58
66
_Generic((__ns), \
···
90
82
#define ns_common_init_inum(__ns, __ops, __inum) __ns_common_init(to_ns_common(__ns), __ops, __inum)
91
83
92
84
#define ns_common_free(__ns) __ns_common_free(to_ns_common((__ns)))
85
+
86
+
static __always_inline __must_check bool __ns_ref_put(struct ns_common *ns)
87
+
{
88
+
return refcount_dec_and_test(&ns->__ns_ref);
89
+
}
90
+
91
+
static __always_inline __must_check bool __ns_ref_get(struct ns_common *ns)
92
+
{
93
+
return refcount_inc_not_zero(&ns->__ns_ref);
94
+
}
95
+
96
+
#define ns_ref_read(__ns) refcount_read(&to_ns_common((__ns))->__ns_ref)
97
+
#define ns_ref_inc(__ns) refcount_inc(&to_ns_common((__ns))->__ns_ref)
98
+
#define ns_ref_get(__ns) __ns_ref_get(to_ns_common((__ns)))
99
+
#define ns_ref_put(__ns) __ns_ref_put(to_ns_common((__ns)))
100
+
#define ns_ref_put_and_lock(__ns, __lock) \
101
+
refcount_dec_and_lock(&to_ns_common((__ns))->__ns_ref, (__lock))
93
102
94
103
#endif
+1
-1
include/linux/pid_namespace.h
+1
-1
include/linux/pid_namespace.h
+2
-2
include/linux/time_namespace.h
+2
-2
include/linux/time_namespace.h
···
44
44
45
45
static inline struct time_namespace *get_time_ns(struct time_namespace *ns)
46
46
{
47
-
refcount_inc(&ns->ns.count);
47
+
ns_ref_inc(ns);
48
48
return ns;
49
49
}
50
50
···
57
57
58
58
static inline void put_time_ns(struct time_namespace *ns)
59
59
{
60
-
if (refcount_dec_and_test(&ns->ns.count))
60
+
if (ns_ref_put(ns))
61
61
free_time_ns(ns);
62
62
}
63
63
+2
-2
include/linux/user_namespace.h
+2
-2
include/linux/user_namespace.h
···
176
176
static inline struct user_namespace *get_user_ns(struct user_namespace *ns)
177
177
{
178
178
if (ns)
179
-
refcount_inc(&ns->ns.count);
179
+
ns_ref_inc(ns);
180
180
return ns;
181
181
}
182
182
···
186
186
187
187
static inline void put_user_ns(struct user_namespace *ns)
188
188
{
189
-
if (ns && refcount_dec_and_test(&ns->ns.count))
189
+
if (ns && ns_ref_put(ns))
190
190
__put_user_ns(ns);
191
191
}
192
192
+2
-2
include/linux/uts_namespace.h
+2
-2
include/linux/uts_namespace.h
···
25
25
26
26
static inline void get_uts_ns(struct uts_namespace *ns)
27
27
{
28
-
refcount_inc(&ns->ns.count);
28
+
ns_ref_inc(ns);
29
29
}
30
30
31
31
extern struct uts_namespace *copy_utsname(unsigned long flags,
···
34
34
35
35
static inline void put_uts_ns(struct uts_namespace *ns)
36
36
{
37
-
if (refcount_dec_and_test(&ns->ns.count))
37
+
if (ns_ref_put(ns))
38
38
free_uts_ns(ns);
39
39
}
40
40
+4
-4
include/net/net_namespace.h
+4
-4
include/net/net_namespace.h
···
270
270
/* Try using get_net_track() instead */
271
271
static inline struct net *get_net(struct net *net)
272
272
{
273
-
refcount_inc(&net->ns.count);
273
+
ns_ref_inc(net);
274
274
return net;
275
275
}
276
276
···
281
281
* exists. If the reference count is zero this
282
282
* function fails and returns NULL.
283
283
*/
284
-
if (!refcount_inc_not_zero(&net->ns.count))
284
+
if (!ns_ref_get(net))
285
285
net = NULL;
286
286
return net;
287
287
}
···
289
289
/* Try using put_net_track() instead */
290
290
static inline void put_net(struct net *net)
291
291
{
292
-
if (refcount_dec_and_test(&net->ns.count))
292
+
if (ns_ref_put(net))
293
293
__put_net(net);
294
294
}
295
295
···
301
301
302
302
static inline int check_net(const struct net *net)
303
303
{
304
-
return refcount_read(&net->ns.count) != 0;
304
+
return ns_ref_read(net) != 0;
305
305
}
306
306
307
307
void net_drop_ns(void *);
+1
-1
init/version-timestamp.c
+1
-1
init/version-timestamp.c
+1
-1
ipc/msgutil.c
+1
-1
ipc/msgutil.c
+1
-1
ipc/namespace.c
+1
-1
ipc/namespace.c
+1
-1
kernel/cgroup/cgroup.c
+1
-1
kernel/cgroup/cgroup.c
+1
-1
kernel/nscommon.c
+1
-1
kernel/nscommon.c
+1
-1
kernel/pid.c
+1
-1
kernel/pid.c
+2
-2
kernel/pid_namespace.c
+2
-2
kernel/pid_namespace.c
···
169
169
parent = ns->parent;
170
170
destroy_pid_namespace(ns);
171
171
ns = parent;
172
-
} while (ns != &init_pid_ns && refcount_dec_and_test(&ns->ns.count));
172
+
} while (ns != &init_pid_ns && ns_ref_put(ns));
173
173
}
174
174
175
175
struct pid_namespace *copy_pid_ns(unsigned long flags,
···
184
184
185
185
void put_pid_ns(struct pid_namespace *ns)
186
186
{
187
-
if (ns && ns != &init_pid_ns && refcount_dec_and_test(&ns->ns.count))
187
+
if (ns && ns != &init_pid_ns && ns_ref_put(ns))
188
188
schedule_work(&ns->work);
189
189
}
190
190
EXPORT_SYMBOL_GPL(put_pid_ns);
+1
-1
kernel/time/namespace.c
+1
-1
kernel/time/namespace.c
+1
-1
kernel/user.c
+1
-1
kernel/user.c
+1
-1
kernel/user_namespace.c
+1
-1
kernel/user_namespace.c
+3
-3
net/core/net-sysfs.c
+3
-3
net/core/net-sysfs.c
···
1328
1328
struct netdev_rx_queue *queue = &dev->_rx[i];
1329
1329
struct kobject *kobj = &queue->kobj;
1330
1330
1331
-
if (!refcount_read(&dev_net(dev)->ns.count))
1331
+
if (!check_net(dev_net(dev)))
1332
1332
kobj->uevent_suppress = 1;
1333
1333
if (dev->sysfs_rx_queue_group)
1334
1334
sysfs_remove_group(kobj, dev->sysfs_rx_queue_group);
···
2061
2061
while (--i >= new_num) {
2062
2062
struct netdev_queue *queue = dev->_tx + i;
2063
2063
2064
-
if (!refcount_read(&dev_net(dev)->ns.count))
2064
+
if (!check_net(dev_net(dev)))
2065
2065
queue->kobj.uevent_suppress = 1;
2066
2066
2067
2067
if (netdev_uses_bql(dev))
···
2315
2315
{
2316
2316
struct device *dev = &ndev->dev;
2317
2317
2318
-
if (!refcount_read(&dev_net(ndev)->ns.count))
2318
+
if (!check_net(dev_net(ndev)))
2319
2319
dev_set_uevent_suppress(dev, 1);
2320
2320
2321
2321
kobject_get(&dev->kobj);
+1
-1
net/core/net_namespace.c
+1
-1
net/core/net_namespace.c
+2
-2
net/ipv4/inet_timewait_sock.c
+2
-2
net/ipv4/inet_timewait_sock.c
···
329
329
TCPF_NEW_SYN_RECV))
330
330
continue;
331
331
332
-
if (refcount_read(&sock_net(sk)->ns.count))
332
+
if (check_net(sock_net(sk)))
333
333
continue;
334
334
335
335
if (unlikely(!refcount_inc_not_zero(&sk->sk_refcnt)))
336
336
continue;
337
337
338
-
if (refcount_read(&sock_net(sk)->ns.count)) {
338
+
if (check_net(sock_net(sk))) {
339
339
sock_gen_put(sk);
340
340
goto restart;
341
341
}
+1
-1
net/ipv4/tcp_metrics.c
+1
-1
net/ipv4/tcp_metrics.c
···
912
912
spin_lock_bh(&tcp_metrics_lock);
913
913
for (tm = deref_locked(*pp); tm; tm = deref_locked(*pp)) {
914
914
match = net ? net_eq(tm_net(tm), net) :
915
-
!refcount_read(&tm_net(tm)->ns.count);
915
+
!check_net(tm_net(tm));
916
916
if (match) {
917
917
rcu_assign_pointer(*pp, tm->tcpm_next);
918
918
kfree_rcu(tm, rcu_head);