1 // SPDX-License-Identifier: GPL-2.0-only
2 
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>
10 
11 struct ucounts init_ucounts = {
12 	.ns    = &init_user_ns,
13 	.uid   = GLOBAL_ROOT_UID,
14 	.count = RCUREF_INIT(1),
15 };
16 
17 #define UCOUNTS_HASHTABLE_BITS 10
18 #define UCOUNTS_HASHTABLE_ENTRIES (1 << UCOUNTS_HASHTABLE_BITS)
19 static struct hlist_nulls_head ucounts_hashtable[UCOUNTS_HASHTABLE_ENTRIES] = {
20 	[0 ... UCOUNTS_HASHTABLE_ENTRIES - 1] = HLIST_NULLS_HEAD_INIT(0)
21 };
22 static DEFINE_SPINLOCK(ucounts_lock);
23 
24 #define ucounts_hashfn(ns, uid)						\
25 	hash_long((unsigned long)__kuid_val(uid) + (unsigned long)(ns), \
26 		  UCOUNTS_HASHTABLE_BITS)
27 #define ucounts_hashentry(ns, uid)	\
28 	(ucounts_hashtable + ucounts_hashfn(ns, uid))
29 
30 #ifdef CONFIG_SYSCTL
31 static struct ctl_table_set *
32 set_lookup(struct ctl_table_root *root)
33 {
34 	return &current_user_ns()->set;
35 }
36 
37 static int set_is_seen(struct ctl_table_set *set)
38 {
39 	return &current_user_ns()->set == set;
40 }
41 
42 static int set_permissions(struct ctl_table_header *head,
43 			   const struct ctl_table *table)
44 {
45 	struct user_namespace *user_ns =
46 		container_of(head->set, struct user_namespace, set);
47 	int mode;
48 
49 	/* Allow users with CAP_SYS_RESOURCE unrestrained access */
50 	if (ns_capable(user_ns, CAP_SYS_RESOURCE))
51 		mode = (table->mode & S_IRWXU) >> 6;
52 	else
53 	/* Allow all others at most read-only access */
54 		mode = table->mode & S_IROTH;
55 	return (mode << 6) | (mode << 3) | mode;
56 }
57 
58 static struct ctl_table_root set_root = {
59 	.lookup = set_lookup,
60 	.permissions = set_permissions,
61 };
62 
63 static long ue_zero = 0;
64 static long ue_int_max = INT_MAX;
65 
66 #define UCOUNT_ENTRY(name)					\
67 	{							\
68 		.procname	= name,				\
69 		.maxlen		= sizeof(long),			\
70 		.mode		= 0644,				\
71 		.proc_handler	= proc_doulongvec_minmax,	\
72 		.extra1		= &ue_zero,			\
73 		.extra2		= &ue_int_max,			\
74 	}
75 static const struct ctl_table user_table[] = {
76 	UCOUNT_ENTRY("max_user_namespaces"),
77 	UCOUNT_ENTRY("max_pid_namespaces"),
78 	UCOUNT_ENTRY("max_uts_namespaces"),
79 	UCOUNT_ENTRY("max_ipc_namespaces"),
80 	UCOUNT_ENTRY("max_net_namespaces"),
81 	UCOUNT_ENTRY("max_mnt_namespaces"),
82 	UCOUNT_ENTRY("max_cgroup_namespaces"),
83 	UCOUNT_ENTRY("max_time_namespaces"),
84 #ifdef CONFIG_INOTIFY_USER
85 	UCOUNT_ENTRY("max_inotify_instances"),
86 	UCOUNT_ENTRY("max_inotify_watches"),
87 #endif
88 #ifdef CONFIG_FANOTIFY
89 	UCOUNT_ENTRY("max_fanotify_groups"),
90 	UCOUNT_ENTRY("max_fanotify_marks"),
91 #endif
92 };
93 #endif /* CONFIG_SYSCTL */
94 
95 bool setup_userns_sysctls(struct user_namespace *ns)
96 {
97 #ifdef CONFIG_SYSCTL
98 	struct ctl_table *tbl;
99 
100 	BUILD_BUG_ON(ARRAY_SIZE(user_table) != UCOUNT_COUNTS);
101 	setup_sysctl_set(&ns->set, &set_root, set_is_seen);
102 	tbl = kmemdup(user_table, sizeof(user_table), GFP_KERNEL);
103 	if (tbl) {
104 		int i;
105 		for (i = 0; i < UCOUNT_COUNTS; i++) {
106 			tbl[i].data = &ns->ucount_max[i];
107 		}
108 		ns->sysctls = __register_sysctl_table(&ns->set, "user", tbl,
109 						      ARRAY_SIZE(user_table));
110 	}
111 	if (!ns->sysctls) {
112 		kfree(tbl);
113 		retire_sysctl_set(&ns->set);
114 		return false;
115 	}
116 #endif
117 	return true;
118 }
119 
120 void retire_userns_sysctls(struct user_namespace *ns)
121 {
122 #ifdef CONFIG_SYSCTL
123 	const struct ctl_table *tbl;
124 
125 	tbl = ns->sysctls->ctl_table_arg;
126 	unregister_sysctl_table(ns->sysctls);
127 	retire_sysctl_set(&ns->set);
128 	kfree(tbl);
129 #endif
130 }
131 
132 static struct ucounts *find_ucounts(struct user_namespace *ns, kuid_t uid,
133 				    struct hlist_nulls_head *hashent)
134 {
135 	struct ucounts *ucounts;
136 	struct hlist_nulls_node *pos;
137 
138 	guard(rcu)();
139 	hlist_nulls_for_each_entry_rcu(ucounts, pos, hashent, node) {
140 		if (uid_eq(ucounts->uid, uid) && (ucounts->ns == ns)) {
141 			if (rcuref_get(&ucounts->count))
142 				return ucounts;
143 		}
144 	}
145 	return NULL;
146 }
147 
148 static void hlist_add_ucounts(struct ucounts *ucounts)
149 {
150 	struct hlist_nulls_head *hashent = ucounts_hashentry(ucounts->ns, ucounts->uid);
151 
152 	spin_lock_irq(&ucounts_lock);
153 	hlist_nulls_add_head_rcu(&ucounts->node, hashent);
154 	spin_unlock_irq(&ucounts_lock);
155 }
156 
157 struct ucounts *alloc_ucounts(struct user_namespace *ns, kuid_t uid)
158 {
159 	struct hlist_nulls_head *hashent = ucounts_hashentry(ns, uid);
160 	struct ucounts *ucounts, *new;
161 
162 	ucounts = find_ucounts(ns, uid, hashent);
163 	if (ucounts)
164 		return ucounts;
165 
166 	new = kzalloc(sizeof(*new), GFP_KERNEL);
167 	if (!new)
168 		return NULL;
169 
170 	new->ns = ns;
171 	new->uid = uid;
172 	rcuref_init(&new->count, 1);
173 
174 	spin_lock_irq(&ucounts_lock);
175 	ucounts = find_ucounts(ns, uid, hashent);
176 	if (ucounts) {
177 		spin_unlock_irq(&ucounts_lock);
178 		kfree(new);
179 		return ucounts;
180 	}
181 
182 	hlist_nulls_add_head_rcu(&new->node, hashent);
183 	get_user_ns(new->ns);
184 	spin_unlock_irq(&ucounts_lock);
185 	return new;
186 }
187 
188 void put_ucounts(struct ucounts *ucounts)
189 {
190 	unsigned long flags;
191 
192 	if (rcuref_put(&ucounts->count)) {
193 		spin_lock_irqsave(&ucounts_lock, flags);
194 		hlist_nulls_del_rcu(&ucounts->node);
195 		spin_unlock_irqrestore(&ucounts_lock, flags);
196 
197 		put_user_ns(ucounts->ns);
198 		kfree_rcu(ucounts, rcu);
199 	}
200 }
201 
202 static inline bool atomic_long_inc_below(atomic_long_t *v, int u)
203 {
204 	long c, old;
205 	c = atomic_long_read(v);
206 	for (;;) {
207 		if (unlikely(c >= u))
208 			return false;
209 		old = atomic_long_cmpxchg(v, c, c+1);
210 		if (likely(old == c))
211 			return true;
212 		c = old;
213 	}
214 }
215 
216 struct ucounts *inc_ucount(struct user_namespace *ns, kuid_t uid,
217 			   enum ucount_type type)
218 {
219 	struct ucounts *ucounts, *iter, *bad;
220 	struct user_namespace *tns;
221 	ucounts = alloc_ucounts(ns, uid);
222 	for (iter = ucounts; iter; iter = tns->ucounts) {
223 		long max;
224 		tns = iter->ns;
225 		max = READ_ONCE(tns->ucount_max[type]);
226 		if (!atomic_long_inc_below(&iter->ucount[type], max))
227 			goto fail;
228 	}
229 	return ucounts;
230 fail:
231 	bad = iter;
232 	for (iter = ucounts; iter != bad; iter = iter->ns->ucounts)
233 		atomic_long_dec(&iter->ucount[type]);
234 
235 	put_ucounts(ucounts);
236 	return NULL;
237 }
238 
239 void dec_ucount(struct ucounts *ucounts, enum ucount_type type)
240 {
241 	struct ucounts *iter;
242 	for (iter = ucounts; iter; iter = iter->ns->ucounts) {
243 		long dec = atomic_long_dec_if_positive(&iter->ucount[type]);
244 		WARN_ON_ONCE(dec < 0);
245 	}
246 	put_ucounts(ucounts);
247 }
248 
249 long inc_rlimit_ucounts(struct ucounts *ucounts, enum rlimit_type type, long v)
250 {
251 	struct ucounts *iter;
252 	long max = LONG_MAX;
253 	long ret = 0;
254 
255 	for (iter = ucounts; iter; iter = iter->ns->ucounts) {
256 		long new = atomic_long_add_return(v, &iter->rlimit[type]);
257 		if (new < 0 || new > max)
258 			ret = LONG_MAX;
259 		else if (iter == ucounts)
260 			ret = new;
261 		max = get_userns_rlimit_max(iter->ns, type);
262 	}
263 	return ret;
264 }
265 
266 bool dec_rlimit_ucounts(struct ucounts *ucounts, enum rlimit_type type, long v)
267 {
268 	struct ucounts *iter;
269 	long new = -1; /* Silence compiler warning */
270 	for (iter = ucounts; iter; iter = iter->ns->ucounts) {
271 		long dec = atomic_long_sub_return(v, &iter->rlimit[type]);
272 		WARN_ON_ONCE(dec < 0);
273 		if (iter == ucounts)
274 			new = dec;
275 	}
276 	return (new == 0);
277 }
278 
279 static void do_dec_rlimit_put_ucounts(struct ucounts *ucounts,
280 				struct ucounts *last, enum rlimit_type type)
281 {
282 	struct ucounts *iter, *next;
283 	for (iter = ucounts; iter != last; iter = next) {
284 		long dec = atomic_long_sub_return(1, &iter->rlimit[type]);
285 		WARN_ON_ONCE(dec < 0);
286 		next = iter->ns->ucounts;
287 		if (dec == 0)
288 			put_ucounts(iter);
289 	}
290 }
291 
292 void dec_rlimit_put_ucounts(struct ucounts *ucounts, enum rlimit_type type)
293 {
294 	do_dec_rlimit_put_ucounts(ucounts, NULL, type);
295 }
296 
297 long inc_rlimit_get_ucounts(struct ucounts *ucounts, enum rlimit_type type,
298 			    bool override_rlimit)
299 {
300 	/* Caller must hold a reference to ucounts */
301 	struct ucounts *iter;
302 	long max = LONG_MAX;
303 	long dec, ret = 0;
304 
305 	for (iter = ucounts; iter; iter = iter->ns->ucounts) {
306 		long new = atomic_long_add_return(1, &iter->rlimit[type]);
307 		if (new < 0 || new > max)
308 			goto dec_unwind;
309 		if (iter == ucounts)
310 			ret = new;
311 		if (!override_rlimit)
312 			max = get_userns_rlimit_max(iter->ns, type);
313 		/*
314 		 * Grab an extra ucount reference for the caller when
315 		 * the rlimit count was previously 0.
316 		 */
317 		if (new != 1)
318 			continue;
319 		if (!get_ucounts(iter))
320 			goto dec_unwind;
321 	}
322 	return ret;
323 dec_unwind:
324 	dec = atomic_long_sub_return(1, &iter->rlimit[type]);
325 	WARN_ON_ONCE(dec < 0);
326 	do_dec_rlimit_put_ucounts(ucounts, iter, type);
327 	return 0;
328 }
329 
330 bool is_rlimit_overlimit(struct ucounts *ucounts, enum rlimit_type type, unsigned long rlimit)
331 {
332 	struct ucounts *iter;
333 	long max = rlimit;
334 	if (rlimit > LONG_MAX)
335 		max = LONG_MAX;
336 	for (iter = ucounts; iter; iter = iter->ns->ucounts) {
337 		long val = get_rlimit_value(iter, type);
338 		if (val < 0 || val > max)
339 			return true;
340 		max = get_userns_rlimit_max(iter->ns, type);
341 	}
342 	return false;
343 }
344 
345 static __init int user_namespace_sysctl_init(void)
346 {
347 #ifdef CONFIG_SYSCTL
348 	static struct ctl_table_header *user_header;
349 	static struct ctl_table empty[1];
350 	/*
351 	 * It is necessary to register the user directory in the
352 	 * default set so that registrations in the child sets work
353 	 * properly.
354 	 */
355 	user_header = register_sysctl_sz("user", empty, 0);
356 	kmemleak_ignore(user_header);
357 	BUG_ON(!user_header);
358 	BUG_ON(!setup_userns_sysctls(&init_user_ns));
359 #endif
360 	hlist_add_ucounts(&init_ucounts);
361 	inc_rlimit_ucounts(&init_ucounts, UCOUNT_RLIMIT_NPROC, 1);
362 	return 0;
363 }
364 subsys_initcall(user_namespace_sysctl_init);
365