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 struct ucounts init_ucounts = {
13 .uid = GLOBAL_ROOT_UID,
14 .count = ATOMIC_INIT(1),
17 #define UCOUNTS_HASHTABLE_BITS 10
18 static struct hlist_head ucounts_hashtable[(1 << UCOUNTS_HASHTABLE_BITS)];
19 static DEFINE_SPINLOCK(ucounts_lock);
21 #define ucounts_hashfn(ns, uid) \
22 hash_long((unsigned long)__kuid_val(uid) + (unsigned long)(ns), \
23 UCOUNTS_HASHTABLE_BITS)
24 #define ucounts_hashentry(ns, uid) \
25 (ucounts_hashtable + ucounts_hashfn(ns, uid))
29 static struct ctl_table_set *
30 set_lookup(struct ctl_table_root *root)
32 return ¤t_user_ns()->set;
35 static int set_is_seen(struct ctl_table_set *set)
37 return ¤t_user_ns()->set == set;
40 static int set_permissions(struct ctl_table_header *head,
41 struct ctl_table *table)
43 struct user_namespace *user_ns =
44 container_of(head->set, struct user_namespace, set);
47 /* Allow users with CAP_SYS_RESOURCE unrestrained access */
48 if (ns_capable(user_ns, CAP_SYS_RESOURCE))
49 mode = (table->mode & S_IRWXU) >> 6;
51 /* Allow all others at most read-only access */
52 mode = table->mode & S_IROTH;
53 return (mode << 6) | (mode << 3) | mode;
56 static struct ctl_table_root set_root = {
58 .permissions = set_permissions,
61 #define UCOUNT_ENTRY(name) \
64 .maxlen = sizeof(int), \
66 .proc_handler = proc_dointvec_minmax, \
67 .extra1 = SYSCTL_ZERO, \
68 .extra2 = SYSCTL_INT_MAX, \
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 UCOUNT_ENTRY("max_time_namespaces"),
79 #ifdef CONFIG_INOTIFY_USER
80 UCOUNT_ENTRY("max_inotify_instances"),
81 UCOUNT_ENTRY("max_inotify_watches"),
83 #ifdef CONFIG_FANOTIFY
84 UCOUNT_ENTRY("max_fanotify_groups"),
85 UCOUNT_ENTRY("max_fanotify_marks"),
93 #endif /* CONFIG_SYSCTL */
95 bool setup_userns_sysctls(struct user_namespace *ns)
98 struct ctl_table *tbl;
100 BUILD_BUG_ON(ARRAY_SIZE(user_table) != UCOUNT_COUNTS + 1);
101 setup_sysctl_set(&ns->set, &set_root, set_is_seen);
102 tbl = kmemdup(user_table, sizeof(user_table), GFP_KERNEL);
105 for (i = 0; i < UCOUNT_COUNTS; i++) {
106 tbl[i].data = &ns->ucount_max[i];
108 ns->sysctls = __register_sysctl_table(&ns->set, "user", tbl);
112 retire_sysctl_set(&ns->set);
119 void retire_userns_sysctls(struct user_namespace *ns)
122 struct ctl_table *tbl;
124 tbl = ns->sysctls->ctl_table_arg;
125 unregister_sysctl_table(ns->sysctls);
126 retire_sysctl_set(&ns->set);
131 static struct ucounts *find_ucounts(struct user_namespace *ns, kuid_t uid, struct hlist_head *hashent)
133 struct ucounts *ucounts;
135 hlist_for_each_entry(ucounts, hashent, node) {
136 if (uid_eq(ucounts->uid, uid) && (ucounts->ns == ns))
142 static void hlist_add_ucounts(struct ucounts *ucounts)
144 struct hlist_head *hashent = ucounts_hashentry(ucounts->ns, ucounts->uid);
145 spin_lock_irq(&ucounts_lock);
146 hlist_add_head(&ucounts->node, hashent);
147 spin_unlock_irq(&ucounts_lock);
150 struct ucounts *get_ucounts(struct ucounts *ucounts)
152 if (ucounts && atomic_add_negative(1, &ucounts->count)) {
153 put_ucounts(ucounts);
159 struct ucounts *alloc_ucounts(struct user_namespace *ns, kuid_t uid)
161 struct hlist_head *hashent = ucounts_hashentry(ns, uid);
162 struct ucounts *ucounts, *new;
165 spin_lock_irq(&ucounts_lock);
166 ucounts = find_ucounts(ns, uid, hashent);
168 spin_unlock_irq(&ucounts_lock);
170 new = kzalloc(sizeof(*new), GFP_KERNEL);
176 atomic_set(&new->count, 1);
178 spin_lock_irq(&ucounts_lock);
179 ucounts = find_ucounts(ns, uid, hashent);
183 hlist_add_head(&new->node, hashent);
184 spin_unlock_irq(&ucounts_lock);
188 overflow = atomic_add_negative(1, &ucounts->count);
189 spin_unlock_irq(&ucounts_lock);
191 put_ucounts(ucounts);
197 void put_ucounts(struct ucounts *ucounts)
201 if (atomic_dec_and_lock_irqsave(&ucounts->count, &ucounts_lock, flags)) {
202 hlist_del_init(&ucounts->node);
203 spin_unlock_irqrestore(&ucounts_lock, flags);
208 static inline bool atomic_long_inc_below(atomic_long_t *v, int u)
211 c = atomic_long_read(v);
213 if (unlikely(c >= u))
215 old = atomic_long_cmpxchg(v, c, c+1);
216 if (likely(old == c))
222 struct ucounts *inc_ucount(struct user_namespace *ns, kuid_t uid,
223 enum ucount_type type)
225 struct ucounts *ucounts, *iter, *bad;
226 struct user_namespace *tns;
227 ucounts = alloc_ucounts(ns, uid);
228 for (iter = ucounts; iter; iter = tns->ucounts) {
231 max = READ_ONCE(tns->ucount_max[type]);
232 if (!atomic_long_inc_below(&iter->ucount[type], max))
238 for (iter = ucounts; iter != bad; iter = iter->ns->ucounts)
239 atomic_long_dec(&iter->ucount[type]);
241 put_ucounts(ucounts);
245 void dec_ucount(struct ucounts *ucounts, enum ucount_type type)
247 struct ucounts *iter;
248 for (iter = ucounts; iter; iter = iter->ns->ucounts) {
249 long dec = atomic_long_dec_if_positive(&iter->ucount[type]);
250 WARN_ON_ONCE(dec < 0);
252 put_ucounts(ucounts);
255 long inc_rlimit_ucounts(struct ucounts *ucounts, enum ucount_type type, long v)
257 struct ucounts *iter;
260 for (iter = ucounts; iter; iter = iter->ns->ucounts) {
261 long max = READ_ONCE(iter->ns->ucount_max[type]);
262 long new = atomic_long_add_return(v, &iter->ucount[type]);
263 if (new < 0 || new > max)
265 else if (iter == ucounts)
271 bool dec_rlimit_ucounts(struct ucounts *ucounts, enum ucount_type type, long v)
273 struct ucounts *iter;
274 long new = -1; /* Silence compiler warning */
275 for (iter = ucounts; iter; iter = iter->ns->ucounts) {
276 long dec = atomic_long_add_return(-v, &iter->ucount[type]);
277 WARN_ON_ONCE(dec < 0);
284 bool is_ucounts_overlimit(struct ucounts *ucounts, enum ucount_type type, unsigned long max)
286 struct ucounts *iter;
287 if (get_ucounts_value(ucounts, type) > max)
289 for (iter = ucounts; iter; iter = iter->ns->ucounts) {
290 max = READ_ONCE(iter->ns->ucount_max[type]);
291 if (get_ucounts_value(iter, type) > max)
297 static __init int user_namespace_sysctl_init(void)
300 static struct ctl_table_header *user_header;
301 static struct ctl_table empty[1];
303 * It is necessary to register the user directory in the
304 * default set so that registrations in the child sets work
307 user_header = register_sysctl("user", empty);
308 kmemleak_ignore(user_header);
309 BUG_ON(!user_header);
310 BUG_ON(!setup_userns_sysctls(&init_user_ns));
312 hlist_add_ucounts(&init_ucounts);
313 inc_rlimit_ucounts(&init_ucounts, UCOUNT_RLIMIT_NPROC, 1);
316 subsys_initcall(user_namespace_sysctl_init);