Loading...
1/*
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
5 * License.
6 */
7
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/kmemleak.h>
14#include <linux/user_namespace.h>
15
16#define UCOUNTS_HASHTABLE_BITS 10
17static struct hlist_head ucounts_hashtable[(1 << UCOUNTS_HASHTABLE_BITS)];
18static DEFINE_SPINLOCK(ucounts_lock);
19
20#define ucounts_hashfn(ns, uid) \
21 hash_long((unsigned long)__kuid_val(uid) + (unsigned long)(ns), \
22 UCOUNTS_HASHTABLE_BITS)
23#define ucounts_hashentry(ns, uid) \
24 (ucounts_hashtable + ucounts_hashfn(ns, uid))
25
26
27#ifdef CONFIG_SYSCTL
28static struct ctl_table_set *
29set_lookup(struct ctl_table_root *root)
30{
31 return ¤t_user_ns()->set;
32}
33
34static int set_is_seen(struct ctl_table_set *set)
35{
36 return ¤t_user_ns()->set == set;
37}
38
39static int set_permissions(struct ctl_table_header *head,
40 struct ctl_table *table)
41{
42 struct user_namespace *user_ns =
43 container_of(head->set, struct user_namespace, set);
44 int mode;
45
46 /* Allow users with CAP_SYS_RESOURCE unrestrained access */
47 if (ns_capable(user_ns, CAP_SYS_RESOURCE))
48 mode = (table->mode & S_IRWXU) >> 6;
49 else
50 /* Allow all others at most read-only access */
51 mode = table->mode & S_IROTH;
52 return (mode << 6) | (mode << 3) | mode;
53}
54
55static struct ctl_table_root set_root = {
56 .lookup = set_lookup,
57 .permissions = set_permissions,
58};
59
60static int zero = 0;
61static int int_max = INT_MAX;
62#define UCOUNT_ENTRY(name) \
63 { \
64 .procname = name, \
65 .maxlen = sizeof(int), \
66 .mode = 0644, \
67 .proc_handler = proc_dointvec_minmax, \
68 .extra1 = &zero, \
69 .extra2 = &int_max, \
70 }
71static struct ctl_table user_table[] = {
72 UCOUNT_ENTRY("max_user_namespaces"),
73 UCOUNT_ENTRY("max_pid_namespaces"),
74 UCOUNT_ENTRY("max_uts_namespaces"),
75 UCOUNT_ENTRY("max_ipc_namespaces"),
76 UCOUNT_ENTRY("max_net_namespaces"),
77 UCOUNT_ENTRY("max_mnt_namespaces"),
78 UCOUNT_ENTRY("max_cgroup_namespaces"),
79#ifdef CONFIG_INOTIFY_USER
80 UCOUNT_ENTRY("max_inotify_instances"),
81 UCOUNT_ENTRY("max_inotify_watches"),
82#endif
83 { }
84};
85#endif /* CONFIG_SYSCTL */
86
87bool setup_userns_sysctls(struct user_namespace *ns)
88{
89#ifdef CONFIG_SYSCTL
90 struct ctl_table *tbl;
91 setup_sysctl_set(&ns->set, &set_root, set_is_seen);
92 tbl = kmemdup(user_table, sizeof(user_table), GFP_KERNEL);
93 if (tbl) {
94 int i;
95 for (i = 0; i < UCOUNT_COUNTS; i++) {
96 tbl[i].data = &ns->ucount_max[i];
97 }
98 ns->sysctls = __register_sysctl_table(&ns->set, "user", tbl);
99 }
100 if (!ns->sysctls) {
101 kfree(tbl);
102 retire_sysctl_set(&ns->set);
103 return false;
104 }
105#endif
106 return true;
107}
108
109void retire_userns_sysctls(struct user_namespace *ns)
110{
111#ifdef CONFIG_SYSCTL
112 struct ctl_table *tbl;
113
114 tbl = ns->sysctls->ctl_table_arg;
115 unregister_sysctl_table(ns->sysctls);
116 retire_sysctl_set(&ns->set);
117 kfree(tbl);
118#endif
119}
120
121static struct ucounts *find_ucounts(struct user_namespace *ns, kuid_t uid, struct hlist_head *hashent)
122{
123 struct ucounts *ucounts;
124
125 hlist_for_each_entry(ucounts, hashent, node) {
126 if (uid_eq(ucounts->uid, uid) && (ucounts->ns == ns))
127 return ucounts;
128 }
129 return NULL;
130}
131
132static struct ucounts *get_ucounts(struct user_namespace *ns, kuid_t uid)
133{
134 struct hlist_head *hashent = ucounts_hashentry(ns, uid);
135 struct ucounts *ucounts, *new;
136
137 spin_lock_irq(&ucounts_lock);
138 ucounts = find_ucounts(ns, uid, hashent);
139 if (!ucounts) {
140 spin_unlock_irq(&ucounts_lock);
141
142 new = kzalloc(sizeof(*new), GFP_KERNEL);
143 if (!new)
144 return NULL;
145
146 new->ns = ns;
147 new->uid = uid;
148 new->count = 0;
149
150 spin_lock_irq(&ucounts_lock);
151 ucounts = find_ucounts(ns, uid, hashent);
152 if (ucounts) {
153 kfree(new);
154 } else {
155 hlist_add_head(&new->node, hashent);
156 ucounts = new;
157 }
158 }
159 if (ucounts->count == INT_MAX)
160 ucounts = NULL;
161 else
162 ucounts->count += 1;
163 spin_unlock_irq(&ucounts_lock);
164 return ucounts;
165}
166
167static void put_ucounts(struct ucounts *ucounts)
168{
169 unsigned long flags;
170
171 spin_lock_irqsave(&ucounts_lock, flags);
172 ucounts->count -= 1;
173 if (!ucounts->count)
174 hlist_del_init(&ucounts->node);
175 else
176 ucounts = NULL;
177 spin_unlock_irqrestore(&ucounts_lock, flags);
178
179 kfree(ucounts);
180}
181
182static inline bool atomic_inc_below(atomic_t *v, int u)
183{
184 int c, old;
185 c = atomic_read(v);
186 for (;;) {
187 if (unlikely(c >= u))
188 return false;
189 old = atomic_cmpxchg(v, c, c+1);
190 if (likely(old == c))
191 return true;
192 c = old;
193 }
194}
195
196struct ucounts *inc_ucount(struct user_namespace *ns, kuid_t uid,
197 enum ucount_type type)
198{
199 struct ucounts *ucounts, *iter, *bad;
200 struct user_namespace *tns;
201 ucounts = get_ucounts(ns, uid);
202 for (iter = ucounts; iter; iter = tns->ucounts) {
203 int max;
204 tns = iter->ns;
205 max = READ_ONCE(tns->ucount_max[type]);
206 if (!atomic_inc_below(&iter->ucount[type], max))
207 goto fail;
208 }
209 return ucounts;
210fail:
211 bad = iter;
212 for (iter = ucounts; iter != bad; iter = iter->ns->ucounts)
213 atomic_dec(&iter->ucount[type]);
214
215 put_ucounts(ucounts);
216 return NULL;
217}
218
219void dec_ucount(struct ucounts *ucounts, enum ucount_type type)
220{
221 struct ucounts *iter;
222 for (iter = ucounts; iter; iter = iter->ns->ucounts) {
223 int dec = atomic_dec_if_positive(&iter->ucount[type]);
224 WARN_ON_ONCE(dec < 0);
225 }
226 put_ucounts(ucounts);
227}
228
229static __init int user_namespace_sysctl_init(void)
230{
231#ifdef CONFIG_SYSCTL
232 static struct ctl_table_header *user_header;
233 static struct ctl_table empty[1];
234 /*
235 * It is necessary to register the user directory in the
236 * default set so that registrations in the child sets work
237 * properly.
238 */
239 user_header = register_sysctl("user", empty);
240 kmemleak_ignore(user_header);
241 BUG_ON(!user_header);
242 BUG_ON(!setup_userns_sysctls(&init_user_ns));
243#endif
244 return 0;
245}
246subsys_initcall(user_namespace_sysctl_init);
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
11struct 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
18static struct hlist_head ucounts_hashtable[(1 << UCOUNTS_HASHTABLE_BITS)];
19static 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
29static struct ctl_table_set *
30set_lookup(struct ctl_table_root *root)
31{
32 return ¤t_user_ns()->set;
33}
34
35static int set_is_seen(struct ctl_table_set *set)
36{
37 return ¤t_user_ns()->set == set;
38}
39
40static 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
56static struct ctl_table_root set_root = {
57 .lookup = set_lookup,
58 .permissions = set_permissions,
59};
60
61static long ue_zero = 0;
62static 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 }
73static 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 { },
93 { },
94 { }
95};
96#endif /* CONFIG_SYSCTL */
97
98bool setup_userns_sysctls(struct user_namespace *ns)
99{
100#ifdef CONFIG_SYSCTL
101 struct ctl_table *tbl;
102
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);
106 if (tbl) {
107 int i;
108 for (i = 0; i < UCOUNT_COUNTS; i++) {
109 tbl[i].data = &ns->ucount_max[i];
110 }
111 ns->sysctls = __register_sysctl_table(&ns->set, "user", tbl);
112 }
113 if (!ns->sysctls) {
114 kfree(tbl);
115 retire_sysctl_set(&ns->set);
116 return false;
117 }
118#endif
119 return true;
120}
121
122void retire_userns_sysctls(struct user_namespace *ns)
123{
124#ifdef CONFIG_SYSCTL
125 struct ctl_table *tbl;
126
127 tbl = ns->sysctls->ctl_table_arg;
128 unregister_sysctl_table(ns->sysctls);
129 retire_sysctl_set(&ns->set);
130 kfree(tbl);
131#endif
132}
133
134static struct ucounts *find_ucounts(struct user_namespace *ns, kuid_t uid, struct hlist_head *hashent)
135{
136 struct ucounts *ucounts;
137
138 hlist_for_each_entry(ucounts, hashent, node) {
139 if (uid_eq(ucounts->uid, uid) && (ucounts->ns == ns))
140 return ucounts;
141 }
142 return NULL;
143}
144
145static void hlist_add_ucounts(struct ucounts *ucounts)
146{
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);
151}
152
153struct ucounts *get_ucounts(struct ucounts *ucounts)
154{
155 if (ucounts && atomic_add_negative(1, &ucounts->count)) {
156 put_ucounts(ucounts);
157 ucounts = NULL;
158 }
159 return ucounts;
160}
161
162struct ucounts *alloc_ucounts(struct user_namespace *ns, kuid_t uid)
163{
164 struct hlist_head *hashent = ucounts_hashentry(ns, uid);
165 struct ucounts *ucounts, *new;
166 long overflow;
167
168 spin_lock_irq(&ucounts_lock);
169 ucounts = find_ucounts(ns, uid, hashent);
170 if (!ucounts) {
171 spin_unlock_irq(&ucounts_lock);
172
173 new = kzalloc(sizeof(*new), GFP_KERNEL);
174 if (!new)
175 return NULL;
176
177 new->ns = ns;
178 new->uid = uid;
179 atomic_set(&new->count, 1);
180
181 spin_lock_irq(&ucounts_lock);
182 ucounts = find_ucounts(ns, uid, hashent);
183 if (ucounts) {
184 kfree(new);
185 } else {
186 hlist_add_head(&new->node, hashent);
187 spin_unlock_irq(&ucounts_lock);
188 return new;
189 }
190 }
191 overflow = atomic_add_negative(1, &ucounts->count);
192 spin_unlock_irq(&ucounts_lock);
193 if (overflow) {
194 put_ucounts(ucounts);
195 return NULL;
196 }
197 return ucounts;
198}
199
200void put_ucounts(struct ucounts *ucounts)
201{
202 unsigned long flags;
203
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);
207 kfree(ucounts);
208 }
209}
210
211static inline bool atomic_long_inc_below(atomic_long_t *v, int u)
212{
213 long c, old;
214 c = atomic_long_read(v);
215 for (;;) {
216 if (unlikely(c >= u))
217 return false;
218 old = atomic_long_cmpxchg(v, c, c+1);
219 if (likely(old == c))
220 return true;
221 c = old;
222 }
223}
224
225struct ucounts *inc_ucount(struct user_namespace *ns, kuid_t uid,
226 enum ucount_type type)
227{
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) {
232 long max;
233 tns = iter->ns;
234 max = READ_ONCE(tns->ucount_max[type]);
235 if (!atomic_long_inc_below(&iter->ucount[type], max))
236 goto fail;
237 }
238 return ucounts;
239fail:
240 bad = iter;
241 for (iter = ucounts; iter != bad; iter = iter->ns->ucounts)
242 atomic_long_dec(&iter->ucount[type]);
243
244 put_ucounts(ucounts);
245 return NULL;
246}
247
248void dec_ucount(struct ucounts *ucounts, enum ucount_type type)
249{
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);
254 }
255 put_ucounts(ucounts);
256}
257
258long inc_rlimit_ucounts(struct ucounts *ucounts, enum ucount_type type, long v)
259{
260 struct ucounts *iter;
261 long ret = 0;
262
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)
267 ret = LONG_MAX;
268 else if (iter == ucounts)
269 ret = new;
270 }
271 return ret;
272}
273
274bool dec_rlimit_ucounts(struct ucounts *ucounts, enum ucount_type type, long v)
275{
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);
281 if (iter == ucounts)
282 new = dec;
283 }
284 return (new == 0);
285}
286
287static void do_dec_rlimit_put_ucounts(struct ucounts *ucounts,
288 struct ucounts *last, enum ucount_type type)
289{
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;
295 if (dec == 0)
296 put_ucounts(iter);
297 }
298}
299
300void dec_rlimit_put_ucounts(struct ucounts *ucounts, enum ucount_type type)
301{
302 do_dec_rlimit_put_ucounts(ucounts, NULL, type);
303}
304
305long inc_rlimit_get_ucounts(struct ucounts *ucounts, enum ucount_type type)
306{
307 /* Caller must hold a reference to ucounts */
308 struct ucounts *iter;
309 long dec, ret = 0;
310
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)
315 goto unwind;
316 if (iter == ucounts)
317 ret = new;
318 /*
319 * Grab an extra ucount reference for the caller when
320 * the rlimit count was previously 0.
321 */
322 if (new != 1)
323 continue;
324 if (!get_ucounts(iter))
325 goto dec_unwind;
326 }
327 return ret;
328dec_unwind:
329 dec = atomic_long_add_return(-1, &iter->ucount[type]);
330 WARN_ON_ONCE(dec < 0);
331unwind:
332 do_dec_rlimit_put_ucounts(ucounts, iter, type);
333 return 0;
334}
335
336bool is_ucounts_overlimit(struct ucounts *ucounts, enum ucount_type type, unsigned long max)
337{
338 struct ucounts *iter;
339 if (get_ucounts_value(ucounts, type) > max)
340 return true;
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)
344 return true;
345 }
346 return false;
347}
348
349static __init int user_namespace_sysctl_init(void)
350{
351#ifdef CONFIG_SYSCTL
352 static struct ctl_table_header *user_header;
353 static struct ctl_table empty[1];
354 /*
355 * It is necessary to register the user directory in the
356 * default set so that registrations in the child sets work
357 * properly.
358 */
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));
363#endif
364 hlist_add_ucounts(&init_ucounts);
365 inc_rlimit_ucounts(&init_ucounts, UCOUNT_RLIMIT_NPROC, 1);
366 return 0;
367}
368subsys_initcall(user_namespace_sysctl_init);