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 static long ue_zero = 0;
62 static long ue_int_max = INT_MAX;
64 #define UCOUNT_ENTRY(name) \
67 .maxlen = sizeof(long), \
69 .proc_handler = proc_doulongvec_minmax, \
71 .extra2 = &ue_int_max, \
73 static struct ctl_table user_table[] = {
74 UCOUNT_ENTRY("max_user_namespaces"),
75 UCOUNT_ENTRY("max_pid_namespaces"),
76 UCOUNT_ENTRY("max_uts_namespaces"),
77 UCOUNT_ENTRY("max_ipc_namespaces"),
78 UCOUNT_ENTRY("max_net_namespaces"),
79 UCOUNT_ENTRY("max_mnt_namespaces"),
80 UCOUNT_ENTRY("max_cgroup_namespaces"),
81 UCOUNT_ENTRY("max_time_namespaces"),
82 #ifdef CONFIG_INOTIFY_USER
83 UCOUNT_ENTRY("max_inotify_instances"),
84 UCOUNT_ENTRY("max_inotify_watches"),
86 #ifdef CONFIG_FANOTIFY
87 UCOUNT_ENTRY("max_fanotify_groups"),
88 UCOUNT_ENTRY("max_fanotify_marks"),
96 #endif /* CONFIG_SYSCTL */
98 bool setup_userns_sysctls(struct user_namespace *ns)
101 struct ctl_table *tbl;
103 BUILD_BUG_ON(ARRAY_SIZE(user_table) != UCOUNT_COUNTS + 1);
104 setup_sysctl_set(&ns->set, &set_root, set_is_seen);
105 tbl = kmemdup(user_table, sizeof(user_table), GFP_KERNEL);
108 for (i = 0; i < UCOUNT_COUNTS; i++) {
109 tbl[i].data = &ns->ucount_max[i];
111 ns->sysctls = __register_sysctl_table(&ns->set, "user", tbl);
115 retire_sysctl_set(&ns->set);
122 void retire_userns_sysctls(struct user_namespace *ns)
125 struct ctl_table *tbl;
127 tbl = ns->sysctls->ctl_table_arg;
128 unregister_sysctl_table(ns->sysctls);
129 retire_sysctl_set(&ns->set);
134 static struct ucounts *find_ucounts(struct user_namespace *ns, kuid_t uid, struct hlist_head *hashent)
136 struct ucounts *ucounts;
138 hlist_for_each_entry(ucounts, hashent, node) {
139 if (uid_eq(ucounts->uid, uid) && (ucounts->ns == ns))
145 static void hlist_add_ucounts(struct ucounts *ucounts)
147 struct hlist_head *hashent = ucounts_hashentry(ucounts->ns, ucounts->uid);
148 spin_lock_irq(&ucounts_lock);
149 hlist_add_head(&ucounts->node, hashent);
150 spin_unlock_irq(&ucounts_lock);
153 struct ucounts *get_ucounts(struct ucounts *ucounts)
155 if (ucounts && atomic_add_negative(1, &ucounts->count)) {
156 put_ucounts(ucounts);
162 struct ucounts *alloc_ucounts(struct user_namespace *ns, kuid_t uid)
164 struct hlist_head *hashent = ucounts_hashentry(ns, uid);
165 struct ucounts *ucounts, *new;
168 spin_lock_irq(&ucounts_lock);
169 ucounts = find_ucounts(ns, uid, hashent);
171 spin_unlock_irq(&ucounts_lock);
173 new = kzalloc(sizeof(*new), GFP_KERNEL);
179 atomic_set(&new->count, 1);
181 spin_lock_irq(&ucounts_lock);
182 ucounts = find_ucounts(ns, uid, hashent);
186 hlist_add_head(&new->node, hashent);
187 spin_unlock_irq(&ucounts_lock);
191 overflow = atomic_add_negative(1, &ucounts->count);
192 spin_unlock_irq(&ucounts_lock);
194 put_ucounts(ucounts);
200 void put_ucounts(struct ucounts *ucounts)
204 if (atomic_dec_and_lock_irqsave(&ucounts->count, &ucounts_lock, flags)) {
205 hlist_del_init(&ucounts->node);
206 spin_unlock_irqrestore(&ucounts_lock, flags);
211 static inline bool atomic_long_inc_below(atomic_long_t *v, int u)
214 c = atomic_long_read(v);
216 if (unlikely(c >= u))
218 old = atomic_long_cmpxchg(v, c, c+1);
219 if (likely(old == c))
225 struct ucounts *inc_ucount(struct user_namespace *ns, kuid_t uid,
226 enum ucount_type type)
228 struct ucounts *ucounts, *iter, *bad;
229 struct user_namespace *tns;
230 ucounts = alloc_ucounts(ns, uid);
231 for (iter = ucounts; iter; iter = tns->ucounts) {
234 max = READ_ONCE(tns->ucount_max[type]);
235 if (!atomic_long_inc_below(&iter->ucount[type], max))
241 for (iter = ucounts; iter != bad; iter = iter->ns->ucounts)
242 atomic_long_dec(&iter->ucount[type]);
244 put_ucounts(ucounts);
248 void dec_ucount(struct ucounts *ucounts, enum ucount_type type)
250 struct ucounts *iter;
251 for (iter = ucounts; iter; iter = iter->ns->ucounts) {
252 long dec = atomic_long_dec_if_positive(&iter->ucount[type]);
253 WARN_ON_ONCE(dec < 0);
255 put_ucounts(ucounts);
258 long inc_rlimit_ucounts(struct ucounts *ucounts, enum ucount_type type, long v)
260 struct ucounts *iter;
263 for (iter = ucounts; iter; iter = iter->ns->ucounts) {
264 long max = READ_ONCE(iter->ns->ucount_max[type]);
265 long new = atomic_long_add_return(v, &iter->ucount[type]);
266 if (new < 0 || new > max)
268 else if (iter == ucounts)
274 bool dec_rlimit_ucounts(struct ucounts *ucounts, enum ucount_type type, long v)
276 struct ucounts *iter;
277 long new = -1; /* Silence compiler warning */
278 for (iter = ucounts; iter; iter = iter->ns->ucounts) {
279 long dec = atomic_long_add_return(-v, &iter->ucount[type]);
280 WARN_ON_ONCE(dec < 0);
287 static void do_dec_rlimit_put_ucounts(struct ucounts *ucounts,
288 struct ucounts *last, enum ucount_type type)
290 struct ucounts *iter, *next;
291 for (iter = ucounts; iter != last; iter = next) {
292 long dec = atomic_long_add_return(-1, &iter->ucount[type]);
293 WARN_ON_ONCE(dec < 0);
294 next = iter->ns->ucounts;
300 void dec_rlimit_put_ucounts(struct ucounts *ucounts, enum ucount_type type)
302 do_dec_rlimit_put_ucounts(ucounts, NULL, type);
305 long inc_rlimit_get_ucounts(struct ucounts *ucounts, enum ucount_type type)
307 /* Caller must hold a reference to ucounts */
308 struct ucounts *iter;
311 for (iter = ucounts; iter; iter = iter->ns->ucounts) {
312 long max = READ_ONCE(iter->ns->ucount_max[type]);
313 long new = atomic_long_add_return(1, &iter->ucount[type]);
314 if (new < 0 || new > max)
319 * Grab an extra ucount reference for the caller when
320 * the rlimit count was previously 0.
324 if (!get_ucounts(iter))
329 dec = atomic_long_add_return(-1, &iter->ucount[type]);
330 WARN_ON_ONCE(dec < 0);
332 do_dec_rlimit_put_ucounts(ucounts, iter, type);
336 bool is_ucounts_overlimit(struct ucounts *ucounts, enum ucount_type type, unsigned long max)
338 struct ucounts *iter;
339 if (get_ucounts_value(ucounts, type) > max)
341 for (iter = ucounts; iter; iter = iter->ns->ucounts) {
342 max = READ_ONCE(iter->ns->ucount_max[type]);
343 if (get_ucounts_value(iter, type) > max)
349 static __init int user_namespace_sysctl_init(void)
352 static struct ctl_table_header *user_header;
353 static struct ctl_table empty[1];
355 * It is necessary to register the user directory in the
356 * default set so that registrations in the child sets work
359 user_header = register_sysctl("user", empty);
360 kmemleak_ignore(user_header);
361 BUG_ON(!user_header);
362 BUG_ON(!setup_userns_sysctls(&init_user_ns));
364 hlist_add_ucounts(&init_ucounts);
365 inc_rlimit_ucounts(&init_ucounts, UCOUNT_RLIMIT_NPROC, 1);
368 subsys_initcall(user_namespace_sysctl_init);