2 * This program is free software; you can redistribute it and/or
3 * modify it under the terms of the GNU General Public License as
4 * published by the Free Software Foundation, version 2 of the
8 #include <linux/stat.h>
9 #include <linux/sysctl.h>
10 #include <linux/slab.h>
11 #include <linux/cred.h>
12 #include <linux/hash.h>
13 #include <linux/user_namespace.h>
15 #define UCOUNTS_HASHTABLE_BITS 10
16 static struct hlist_head ucounts_hashtable
[(1 << UCOUNTS_HASHTABLE_BITS
)];
17 static DEFINE_SPINLOCK(ucounts_lock
);
19 #define ucounts_hashfn(ns, uid) \
20 hash_long((unsigned long)__kuid_val(uid) + (unsigned long)(ns), \
21 UCOUNTS_HASHTABLE_BITS)
22 #define ucounts_hashentry(ns, uid) \
23 (ucounts_hashtable + ucounts_hashfn(ns, uid))
27 static struct ctl_table_set
*
28 set_lookup(struct ctl_table_root
*root
)
30 return ¤t_user_ns()->set
;
33 static int set_is_seen(struct ctl_table_set
*set
)
35 return ¤t_user_ns()->set
== set
;
38 static int set_permissions(struct ctl_table_header
*head
,
39 struct ctl_table
*table
)
41 struct user_namespace
*user_ns
=
42 container_of(head
->set
, struct user_namespace
, set
);
45 /* Allow users with CAP_SYS_RESOURCE unrestrained access */
46 if (ns_capable(user_ns
, CAP_SYS_RESOURCE
))
47 mode
= (table
->mode
& S_IRWXU
) >> 6;
49 /* Allow all others at most read-only access */
50 mode
= table
->mode
& S_IROTH
;
51 return (mode
<< 6) | (mode
<< 3) | mode
;
54 static struct ctl_table_root set_root
= {
56 .permissions
= set_permissions
,
60 static int int_max
= INT_MAX
;
61 #define UCOUNT_ENTRY(name) \
64 .maxlen = sizeof(int), \
66 .proc_handler = proc_dointvec_minmax, \
70 static struct ctl_table user_table
[] = {
71 UCOUNT_ENTRY("max_user_namespaces"),
72 UCOUNT_ENTRY("max_pid_namespaces"),
73 UCOUNT_ENTRY("max_uts_namespaces"),
74 UCOUNT_ENTRY("max_ipc_namespaces"),
75 UCOUNT_ENTRY("max_net_namespaces"),
76 UCOUNT_ENTRY("max_mnt_namespaces"),
77 UCOUNT_ENTRY("max_cgroup_namespaces"),
78 #ifdef CONFIG_INOTIFY_USER
79 UCOUNT_ENTRY("max_inotify_instances"),
80 UCOUNT_ENTRY("max_inotify_watches"),
84 #endif /* CONFIG_SYSCTL */
86 bool setup_userns_sysctls(struct user_namespace
*ns
)
89 struct ctl_table
*tbl
;
90 setup_sysctl_set(&ns
->set
, &set_root
, set_is_seen
);
91 tbl
= kmemdup(user_table
, sizeof(user_table
), GFP_KERNEL
);
94 for (i
= 0; i
< UCOUNT_COUNTS
; i
++) {
95 tbl
[i
].data
= &ns
->ucount_max
[i
];
97 ns
->sysctls
= __register_sysctl_table(&ns
->set
, "user", tbl
);
101 retire_sysctl_set(&ns
->set
);
108 void retire_userns_sysctls(struct user_namespace
*ns
)
111 struct ctl_table
*tbl
;
113 tbl
= ns
->sysctls
->ctl_table_arg
;
114 unregister_sysctl_table(ns
->sysctls
);
115 retire_sysctl_set(&ns
->set
);
120 static struct ucounts
*find_ucounts(struct user_namespace
*ns
, kuid_t uid
, struct hlist_head
*hashent
)
122 struct ucounts
*ucounts
;
124 hlist_for_each_entry(ucounts
, hashent
, node
) {
125 if (uid_eq(ucounts
->uid
, uid
) && (ucounts
->ns
== ns
))
131 static struct ucounts
*get_ucounts(struct user_namespace
*ns
, kuid_t uid
)
133 struct hlist_head
*hashent
= ucounts_hashentry(ns
, uid
);
134 struct ucounts
*ucounts
, *new;
136 spin_lock_irq(&ucounts_lock
);
137 ucounts
= find_ucounts(ns
, uid
, hashent
);
139 spin_unlock_irq(&ucounts_lock
);
141 new = kzalloc(sizeof(*new), GFP_KERNEL
);
147 atomic_set(&new->count
, 0);
149 spin_lock_irq(&ucounts_lock
);
150 ucounts
= find_ucounts(ns
, uid
, hashent
);
154 hlist_add_head(&new->node
, hashent
);
158 if (!atomic_add_unless(&ucounts
->count
, 1, INT_MAX
))
160 spin_unlock_irq(&ucounts_lock
);
164 static void put_ucounts(struct ucounts
*ucounts
)
168 if (atomic_dec_and_test(&ucounts
->count
)) {
169 spin_lock_irqsave(&ucounts_lock
, flags
);
170 hlist_del_init(&ucounts
->node
);
171 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
);