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 = ATOMIC_INIT(1),
15 };
16 
17 #define UCOUNTS_HASHTABLE_BITS 10
18 static struct hlist_head ucounts_hashtable[(1 << UCOUNTS_HASHTABLE_BITS)];
19 static DEFINE_SPINLOCK(ucounts_lock);
20 
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))
26 
27 
28 #ifdef CONFIG_SYSCTL
29 static struct ctl_table_set *
set_lookup(struct ctl_table_root * root)30 set_lookup(struct ctl_table_root *root)
31 {
32 	return &current_user_ns()->set;
33 }
34 
set_is_seen(struct ctl_table_set * set)35 static int set_is_seen(struct ctl_table_set *set)
36 {
37 	return &current_user_ns()->set == set;
38 }
39 
set_permissions(struct ctl_table_header * head,struct ctl_table * table)40 static int set_permissions(struct ctl_table_header *head,
41 				  struct ctl_table *table)
42 {
43 	struct user_namespace *user_ns =
44 		container_of(head->set, struct user_namespace, set);
45 	int mode;
46 
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;
50 	else
51 	/* Allow all others at most read-only access */
52 		mode = table->mode & S_IROTH;
53 	return (mode << 6) | (mode << 3) | mode;
54 }
55 
56 static struct ctl_table_root set_root = {
57 	.lookup = set_lookup,
58 	.permissions = set_permissions,
59 };
60 
61 static long ue_zero = 0;
62 static long ue_int_max = INT_MAX;
63 
64 #define UCOUNT_ENTRY(name)					\
65 	{							\
66 		.procname	= name,				\
67 		.maxlen		= sizeof(long),			\
68 		.mode		= 0644,				\
69 		.proc_handler	= proc_doulongvec_minmax,	\
70 		.extra1		= &ue_zero,			\
71 		.extra2		= &ue_int_max,			\
72 	}
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"),
85 #endif
86 #ifdef CONFIG_FANOTIFY
87 	UCOUNT_ENTRY("max_fanotify_groups"),
88 	UCOUNT_ENTRY("max_fanotify_marks"),
89 #endif
90 	{ }
91 };
92 #endif /* CONFIG_SYSCTL */
93 
setup_userns_sysctls(struct user_namespace * ns)94 bool setup_userns_sysctls(struct user_namespace *ns)
95 {
96 #ifdef CONFIG_SYSCTL
97 	struct ctl_table *tbl;
98 
99 	BUILD_BUG_ON(ARRAY_SIZE(user_table) != UCOUNT_COUNTS + 1);
100 	setup_sysctl_set(&ns->set, &set_root, set_is_seen);
101 	tbl = kmemdup(user_table, sizeof(user_table), GFP_KERNEL);
102 	if (tbl) {
103 		int i;
104 		for (i = 0; i < UCOUNT_COUNTS; i++) {
105 			tbl[i].data = &ns->ucount_max[i];
106 		}
107 		ns->sysctls = __register_sysctl_table(&ns->set, "user", tbl);
108 	}
109 	if (!ns->sysctls) {
110 		kfree(tbl);
111 		retire_sysctl_set(&ns->set);
112 		return false;
113 	}
114 #endif
115 	return true;
116 }
117 
retire_userns_sysctls(struct user_namespace * ns)118 void retire_userns_sysctls(struct user_namespace *ns)
119 {
120 #ifdef CONFIG_SYSCTL
121 	struct ctl_table *tbl;
122 
123 	tbl = ns->sysctls->ctl_table_arg;
124 	unregister_sysctl_table(ns->sysctls);
125 	retire_sysctl_set(&ns->set);
126 	kfree(tbl);
127 #endif
128 }
129 
find_ucounts(struct user_namespace * ns,kuid_t uid,struct hlist_head * hashent)130 static struct ucounts *find_ucounts(struct user_namespace *ns, kuid_t uid, struct hlist_head *hashent)
131 {
132 	struct ucounts *ucounts;
133 
134 	hlist_for_each_entry(ucounts, hashent, node) {
135 		if (uid_eq(ucounts->uid, uid) && (ucounts->ns == ns))
136 			return ucounts;
137 	}
138 	return NULL;
139 }
140 
hlist_add_ucounts(struct ucounts * ucounts)141 static void hlist_add_ucounts(struct ucounts *ucounts)
142 {
143 	struct hlist_head *hashent = ucounts_hashentry(ucounts->ns, ucounts->uid);
144 	spin_lock_irq(&ucounts_lock);
145 	hlist_add_head(&ucounts->node, hashent);
146 	spin_unlock_irq(&ucounts_lock);
147 }
148 
get_ucounts_or_wrap(struct ucounts * ucounts)149 static inline bool get_ucounts_or_wrap(struct ucounts *ucounts)
150 {
151 	/* Returns true on a successful get, false if the count wraps. */
152 	return !atomic_add_negative(1, &ucounts->count);
153 }
154 
get_ucounts(struct ucounts * ucounts)155 struct ucounts *get_ucounts(struct ucounts *ucounts)
156 {
157 	if (!get_ucounts_or_wrap(ucounts)) {
158 		put_ucounts(ucounts);
159 		ucounts = NULL;
160 	}
161 	return ucounts;
162 }
163 
alloc_ucounts(struct user_namespace * ns,kuid_t uid)164 struct ucounts *alloc_ucounts(struct user_namespace *ns, kuid_t uid)
165 {
166 	struct hlist_head *hashent = ucounts_hashentry(ns, uid);
167 	struct ucounts *ucounts, *new;
168 	bool wrapped;
169 
170 	spin_lock_irq(&ucounts_lock);
171 	ucounts = find_ucounts(ns, uid, hashent);
172 	if (!ucounts) {
173 		spin_unlock_irq(&ucounts_lock);
174 
175 		new = kzalloc(sizeof(*new), GFP_KERNEL);
176 		if (!new)
177 			return NULL;
178 
179 		new->ns = ns;
180 		new->uid = uid;
181 		atomic_set(&new->count, 1);
182 
183 		spin_lock_irq(&ucounts_lock);
184 		ucounts = find_ucounts(ns, uid, hashent);
185 		if (ucounts) {
186 			kfree(new);
187 		} else {
188 			hlist_add_head(&new->node, hashent);
189 			get_user_ns(new->ns);
190 			spin_unlock_irq(&ucounts_lock);
191 			return new;
192 		}
193 	}
194 	wrapped = !get_ucounts_or_wrap(ucounts);
195 	spin_unlock_irq(&ucounts_lock);
196 	if (wrapped) {
197 		put_ucounts(ucounts);
198 		return NULL;
199 	}
200 	return ucounts;
201 }
202 
put_ucounts(struct ucounts * ucounts)203 void put_ucounts(struct ucounts *ucounts)
204 {
205 	unsigned long flags;
206 
207 	if (atomic_dec_and_lock_irqsave(&ucounts->count, &ucounts_lock, flags)) {
208 		hlist_del_init(&ucounts->node);
209 		spin_unlock_irqrestore(&ucounts_lock, flags);
210 		put_user_ns(ucounts->ns);
211 		kfree(ucounts);
212 	}
213 }
214 
atomic_long_inc_below(atomic_long_t * v,int u)215 static inline bool atomic_long_inc_below(atomic_long_t *v, int u)
216 {
217 	long c, old;
218 	c = atomic_long_read(v);
219 	for (;;) {
220 		if (unlikely(c >= u))
221 			return false;
222 		old = atomic_long_cmpxchg(v, c, c+1);
223 		if (likely(old == c))
224 			return true;
225 		c = old;
226 	}
227 }
228 
inc_ucount(struct user_namespace * ns,kuid_t uid,enum ucount_type type)229 struct ucounts *inc_ucount(struct user_namespace *ns, kuid_t uid,
230 			   enum ucount_type type)
231 {
232 	struct ucounts *ucounts, *iter, *bad;
233 	struct user_namespace *tns;
234 	ucounts = alloc_ucounts(ns, uid);
235 	for (iter = ucounts; iter; iter = tns->ucounts) {
236 		long max;
237 		tns = iter->ns;
238 		max = READ_ONCE(tns->ucount_max[type]);
239 		if (!atomic_long_inc_below(&iter->ucount[type], max))
240 			goto fail;
241 	}
242 	return ucounts;
243 fail:
244 	bad = iter;
245 	for (iter = ucounts; iter != bad; iter = iter->ns->ucounts)
246 		atomic_long_dec(&iter->ucount[type]);
247 
248 	put_ucounts(ucounts);
249 	return NULL;
250 }
251 
dec_ucount(struct ucounts * ucounts,enum ucount_type type)252 void dec_ucount(struct ucounts *ucounts, enum ucount_type type)
253 {
254 	struct ucounts *iter;
255 	for (iter = ucounts; iter; iter = iter->ns->ucounts) {
256 		long dec = atomic_long_dec_if_positive(&iter->ucount[type]);
257 		WARN_ON_ONCE(dec < 0);
258 	}
259 	put_ucounts(ucounts);
260 }
261 
inc_rlimit_ucounts(struct ucounts * ucounts,enum rlimit_type type,long v)262 long inc_rlimit_ucounts(struct ucounts *ucounts, enum rlimit_type type, long v)
263 {
264 	struct ucounts *iter;
265 	long max = LONG_MAX;
266 	long ret = 0;
267 
268 	for (iter = ucounts; iter; iter = iter->ns->ucounts) {
269 		long new = atomic_long_add_return(v, &iter->rlimit[type]);
270 		if (new < 0 || new > max)
271 			ret = LONG_MAX;
272 		else if (iter == ucounts)
273 			ret = new;
274 		max = get_userns_rlimit_max(iter->ns, type);
275 	}
276 	return ret;
277 }
278 
dec_rlimit_ucounts(struct ucounts * ucounts,enum rlimit_type type,long v)279 bool dec_rlimit_ucounts(struct ucounts *ucounts, enum rlimit_type type, long v)
280 {
281 	struct ucounts *iter;
282 	long new = -1; /* Silence compiler warning */
283 	for (iter = ucounts; iter; iter = iter->ns->ucounts) {
284 		long dec = atomic_long_sub_return(v, &iter->rlimit[type]);
285 		WARN_ON_ONCE(dec < 0);
286 		if (iter == ucounts)
287 			new = dec;
288 	}
289 	return (new == 0);
290 }
291 
do_dec_rlimit_put_ucounts(struct ucounts * ucounts,struct ucounts * last,enum rlimit_type type)292 static void do_dec_rlimit_put_ucounts(struct ucounts *ucounts,
293 				struct ucounts *last, enum rlimit_type type)
294 {
295 	struct ucounts *iter, *next;
296 	for (iter = ucounts; iter != last; iter = next) {
297 		long dec = atomic_long_sub_return(1, &iter->rlimit[type]);
298 		WARN_ON_ONCE(dec < 0);
299 		next = iter->ns->ucounts;
300 		if (dec == 0)
301 			put_ucounts(iter);
302 	}
303 }
304 
dec_rlimit_put_ucounts(struct ucounts * ucounts,enum rlimit_type type)305 void dec_rlimit_put_ucounts(struct ucounts *ucounts, enum rlimit_type type)
306 {
307 	do_dec_rlimit_put_ucounts(ucounts, NULL, type);
308 }
309 
inc_rlimit_get_ucounts(struct ucounts * ucounts,enum rlimit_type type)310 long inc_rlimit_get_ucounts(struct ucounts *ucounts, enum rlimit_type type)
311 {
312 	/* Caller must hold a reference to ucounts */
313 	struct ucounts *iter;
314 	long max = LONG_MAX;
315 	long dec, ret = 0;
316 
317 	for (iter = ucounts; iter; iter = iter->ns->ucounts) {
318 		long new = atomic_long_add_return(1, &iter->rlimit[type]);
319 		if (new < 0 || new > max)
320 			goto unwind;
321 		if (iter == ucounts)
322 			ret = new;
323 		max = get_userns_rlimit_max(iter->ns, type);
324 		/*
325 		 * Grab an extra ucount reference for the caller when
326 		 * the rlimit count was previously 0.
327 		 */
328 		if (new != 1)
329 			continue;
330 		if (!get_ucounts(iter))
331 			goto dec_unwind;
332 	}
333 	return ret;
334 dec_unwind:
335 	dec = atomic_long_sub_return(1, &iter->rlimit[type]);
336 	WARN_ON_ONCE(dec < 0);
337 unwind:
338 	do_dec_rlimit_put_ucounts(ucounts, iter, type);
339 	return 0;
340 }
341 
is_rlimit_overlimit(struct ucounts * ucounts,enum rlimit_type type,unsigned long rlimit)342 bool is_rlimit_overlimit(struct ucounts *ucounts, enum rlimit_type type, unsigned long rlimit)
343 {
344 	struct ucounts *iter;
345 	long max = rlimit;
346 	if (rlimit > LONG_MAX)
347 		max = LONG_MAX;
348 	for (iter = ucounts; iter; iter = iter->ns->ucounts) {
349 		long val = get_rlimit_value(iter, type);
350 		if (val < 0 || val > max)
351 			return true;
352 		max = get_userns_rlimit_max(iter->ns, type);
353 	}
354 	return false;
355 }
356 
user_namespace_sysctl_init(void)357 static __init int user_namespace_sysctl_init(void)
358 {
359 #ifdef CONFIG_SYSCTL
360 	static struct ctl_table_header *user_header;
361 	static struct ctl_table empty[1];
362 	/*
363 	 * It is necessary to register the user directory in the
364 	 * default set so that registrations in the child sets work
365 	 * properly.
366 	 */
367 	user_header = register_sysctl("user", empty);
368 	kmemleak_ignore(user_header);
369 	BUG_ON(!user_header);
370 	BUG_ON(!setup_userns_sysctls(&init_user_ns));
371 #endif
372 	hlist_add_ucounts(&init_ucounts);
373 	inc_rlimit_ucounts(&init_ucounts, UCOUNT_RLIMIT_NPROC, 1);
374 	return 0;
375 }
376 subsys_initcall(user_namespace_sysctl_init);
377