1 // SPDX-License-Identifier: GPL-2.0-only
3 #include <linux/stat.h>
4 #include <linux/sysctl.h>
5 #include <linux/slab.h>
6 #include <linux/cred.h>
7 #include <linux/hash.h>
8 #include <linux/kmemleak.h>
9 #include <linux/user_namespace.h>
11 #define UCOUNTS_HASHTABLE_BITS 10
12 static struct hlist_head ucounts_hashtable
[(1 << UCOUNTS_HASHTABLE_BITS
)];
13 static DEFINE_SPINLOCK(ucounts_lock
);
15 #define ucounts_hashfn(ns, uid) \
16 hash_long((unsigned long)__kuid_val(uid) + (unsigned long)(ns), \
17 UCOUNTS_HASHTABLE_BITS)
18 #define ucounts_hashentry(ns, uid) \
19 (ucounts_hashtable + ucounts_hashfn(ns, uid))
23 static struct ctl_table_set
*
24 set_lookup(struct ctl_table_root
*root
)
26 return ¤t_user_ns()->set
;
29 static int set_is_seen(struct ctl_table_set
*set
)
31 return ¤t_user_ns()->set
== set
;
34 static int set_permissions(struct ctl_table_header
*head
,
35 struct ctl_table
*table
)
37 struct user_namespace
*user_ns
=
38 container_of(head
->set
, struct user_namespace
, set
);
41 /* Allow users with CAP_SYS_RESOURCE unrestrained access */
42 if (ns_capable(user_ns
, CAP_SYS_RESOURCE
))
43 mode
= (table
->mode
& S_IRWXU
) >> 6;
45 /* Allow all others at most read-only access */
46 mode
= table
->mode
& S_IROTH
;
47 return (mode
<< 6) | (mode
<< 3) | mode
;
50 static struct ctl_table_root set_root
= {
52 .permissions
= set_permissions
,
56 static int int_max
= INT_MAX
;
57 #define UCOUNT_ENTRY(name) \
60 .maxlen = sizeof(int), \
62 .proc_handler = proc_dointvec_minmax, \
66 static struct ctl_table user_table
[] = {
67 UCOUNT_ENTRY("max_user_namespaces"),
68 UCOUNT_ENTRY("max_pid_namespaces"),
69 UCOUNT_ENTRY("max_uts_namespaces"),
70 UCOUNT_ENTRY("max_ipc_namespaces"),
71 UCOUNT_ENTRY("max_net_namespaces"),
72 UCOUNT_ENTRY("max_mnt_namespaces"),
73 UCOUNT_ENTRY("max_cgroup_namespaces"),
74 #ifdef CONFIG_INOTIFY_USER
75 UCOUNT_ENTRY("max_inotify_instances"),
76 UCOUNT_ENTRY("max_inotify_watches"),
80 #endif /* CONFIG_SYSCTL */
82 bool setup_userns_sysctls(struct user_namespace
*ns
)
85 struct ctl_table
*tbl
;
86 setup_sysctl_set(&ns
->set
, &set_root
, set_is_seen
);
87 tbl
= kmemdup(user_table
, sizeof(user_table
), GFP_KERNEL
);
90 for (i
= 0; i
< UCOUNT_COUNTS
; i
++) {
91 tbl
[i
].data
= &ns
->ucount_max
[i
];
93 ns
->sysctls
= __register_sysctl_table(&ns
->set
, "user", tbl
);
97 retire_sysctl_set(&ns
->set
);
104 void retire_userns_sysctls(struct user_namespace
*ns
)
107 struct ctl_table
*tbl
;
109 tbl
= ns
->sysctls
->ctl_table_arg
;
110 unregister_sysctl_table(ns
->sysctls
);
111 retire_sysctl_set(&ns
->set
);
116 static struct ucounts
*find_ucounts(struct user_namespace
*ns
, kuid_t uid
, struct hlist_head
*hashent
)
118 struct ucounts
*ucounts
;
120 hlist_for_each_entry(ucounts
, hashent
, node
) {
121 if (uid_eq(ucounts
->uid
, uid
) && (ucounts
->ns
== ns
))
127 static struct ucounts
*get_ucounts(struct user_namespace
*ns
, kuid_t uid
)
129 struct hlist_head
*hashent
= ucounts_hashentry(ns
, uid
);
130 struct ucounts
*ucounts
, *new;
132 spin_lock_irq(&ucounts_lock
);
133 ucounts
= find_ucounts(ns
, uid
, hashent
);
135 spin_unlock_irq(&ucounts_lock
);
137 new = kzalloc(sizeof(*new), GFP_KERNEL
);
145 spin_lock_irq(&ucounts_lock
);
146 ucounts
= find_ucounts(ns
, uid
, hashent
);
150 hlist_add_head(&new->node
, hashent
);
154 if (ucounts
->count
== INT_MAX
)
158 spin_unlock_irq(&ucounts_lock
);
162 static void put_ucounts(struct ucounts
*ucounts
)
166 spin_lock_irqsave(&ucounts_lock
, flags
);
169 hlist_del_init(&ucounts
->node
);
172 spin_unlock_irqrestore(&ucounts_lock
, flags
);
177 static inline bool atomic_inc_below(atomic_t
*v
, int u
)
182 if (unlikely(c
>= u
))
184 old
= atomic_cmpxchg(v
, c
, c
+1);
185 if (likely(old
== c
))
191 struct ucounts
*inc_ucount(struct user_namespace
*ns
, kuid_t uid
,
192 enum ucount_type type
)
194 struct ucounts
*ucounts
, *iter
, *bad
;
195 struct user_namespace
*tns
;
196 ucounts
= get_ucounts(ns
, uid
);
197 for (iter
= ucounts
; iter
; iter
= tns
->ucounts
) {
200 max
= READ_ONCE(tns
->ucount_max
[type
]);
201 if (!atomic_inc_below(&iter
->ucount
[type
], max
))
207 for (iter
= ucounts
; iter
!= bad
; iter
= iter
->ns
->ucounts
)
208 atomic_dec(&iter
->ucount
[type
]);
210 put_ucounts(ucounts
);
214 void dec_ucount(struct ucounts
*ucounts
, enum ucount_type type
)
216 struct ucounts
*iter
;
217 for (iter
= ucounts
; iter
; iter
= iter
->ns
->ucounts
) {
218 int dec
= atomic_dec_if_positive(&iter
->ucount
[type
]);
219 WARN_ON_ONCE(dec
< 0);
221 put_ucounts(ucounts
);
224 static __init
int user_namespace_sysctl_init(void)
227 static struct ctl_table_header
*user_header
;
228 static struct ctl_table empty
[1];
230 * It is necessary to register the user directory in the
231 * default set so that registrations in the child sets work
234 user_header
= register_sysctl("user", empty
);
235 kmemleak_ignore(user_header
);
236 BUG_ON(!user_header
);
237 BUG_ON(!setup_userns_sysctls(&init_user_ns
));
241 subsys_initcall(user_namespace_sysctl_init
);