blob: ee8e57fd6f907847d033d2141e5cdaf6e4c18112 [file] [log] [blame]
Thomas Gleixnerb886d83c2019-06-01 10:08:55 +02001// SPDX-License-Identifier: GPL-2.0-only
Eric W. Biedermandbec2842016-07-30 13:58:49 -05002
3#include <linux/stat.h>
4#include <linux/sysctl.h>
5#include <linux/slab.h>
Ingo Molnar5b825c32017-02-02 17:54:15 +01006#include <linux/cred.h>
Eric W. Biedermanf6b2db1a2016-08-08 13:54:50 -05007#include <linux/hash.h>
Randy Dunlap514c6032018-04-05 16:25:34 -07008#include <linux/kmemleak.h>
Eric W. Biedermandbec2842016-07-30 13:58:49 -05009#include <linux/user_namespace.h>
10
Alexey Gladkov905ae012021-04-22 14:27:09 +020011struct ucounts init_ucounts = {
12 .ns = &init_user_ns,
13 .uid = GLOBAL_ROOT_UID,
Alexey Gladkovb6c33652021-04-22 14:27:10 +020014 .count = ATOMIC_INIT(1),
Alexey Gladkov905ae012021-04-22 14:27:09 +020015};
16
Eric W. Biedermanf6b2db1a2016-08-08 13:54:50 -050017#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
Eric W. Biedermandbec2842016-07-30 13:58:49 -050028#ifdef CONFIG_SYSCTL
29static struct ctl_table_set *
30set_lookup(struct ctl_table_root *root)
31{
32 return &current_user_ns()->set;
33}
34
35static int set_is_seen(struct ctl_table_set *set)
36{
37 return &current_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
Sven Schnellef153c222021-07-30 08:28:54 +020061static 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, \
Eric W. Biederman25f9c082016-08-08 14:41:52 -050072 }
Eric W. Biedermanf6b2db1a2016-08-08 13:54:50 -050073static struct ctl_table user_table[] = {
Eric W. Biederman25f9c082016-08-08 14:41:52 -050074 UCOUNT_ENTRY("max_user_namespaces"),
Eric W. Biedermanf333c702016-08-08 14:08:36 -050075 UCOUNT_ENTRY("max_pid_namespaces"),
Eric W. Biedermanf7af3d12016-08-08 14:11:25 -050076 UCOUNT_ENTRY("max_uts_namespaces"),
Eric W. Biedermanaba35662016-08-08 14:20:23 -050077 UCOUNT_ENTRY("max_ipc_namespaces"),
Eric W. Biederman70328662016-08-08 14:33:23 -050078 UCOUNT_ENTRY("max_net_namespaces"),
Eric W. Biederman537f7cc2016-08-08 14:37:37 -050079 UCOUNT_ENTRY("max_mnt_namespaces"),
Eric W. Biedermand08311d2016-08-08 14:25:30 -050080 UCOUNT_ENTRY("max_cgroup_namespaces"),
Dmitry Safonoveeec26d2020-04-06 18:13:42 +010081 UCOUNT_ENTRY("max_time_namespaces"),
Nikolay Borisov1cce1ee2016-12-14 15:56:33 +020082#ifdef CONFIG_INOTIFY_USER
83 UCOUNT_ENTRY("max_inotify_instances"),
84 UCOUNT_ENTRY("max_inotify_watches"),
85#endif
Amir Goldstein5b8fea62021-03-04 13:29:20 +020086#ifdef CONFIG_FANOTIFY
87 UCOUNT_ENTRY("max_fanotify_groups"),
88 UCOUNT_ENTRY("max_fanotify_marks"),
89#endif
Eric W. Biedermandbec2842016-07-30 13:58:49 -050090 { }
91};
92#endif /* CONFIG_SYSCTL */
93
94bool setup_userns_sysctls(struct user_namespace *ns)
95{
96#ifdef CONFIG_SYSCTL
97 struct ctl_table *tbl;
Jan Kara0f538e32020-04-07 17:46:43 +020098
99 BUILD_BUG_ON(ARRAY_SIZE(user_table) != UCOUNT_COUNTS + 1);
Eric W. Biedermandbec2842016-07-30 13:58:49 -0500100 setup_sysctl_set(&ns->set, &set_root, set_is_seen);
Eric W. Biedermanf6b2db1a2016-08-08 13:54:50 -0500101 tbl = kmemdup(user_table, sizeof(user_table), GFP_KERNEL);
Eric W. Biedermandbec2842016-07-30 13:58:49 -0500102 if (tbl) {
Eric W. Biederman25f9c082016-08-08 14:41:52 -0500103 int i;
104 for (i = 0; i < UCOUNT_COUNTS; i++) {
105 tbl[i].data = &ns->ucount_max[i];
106 }
Eric W. Biedermanf6b2db1a2016-08-08 13:54:50 -0500107 ns->sysctls = __register_sysctl_table(&ns->set, "user", tbl);
Eric W. Biedermandbec2842016-07-30 13:58:49 -0500108 }
109 if (!ns->sysctls) {
110 kfree(tbl);
111 retire_sysctl_set(&ns->set);
112 return false;
113 }
114#endif
115 return true;
116}
117
118void 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
Eric W. Biedermanf6b2db1a2016-08-08 13:54:50 -0500130static 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
Alexey Gladkov905ae012021-04-22 14:27:09 +0200141static 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
Eric W. Biedermanda70d312021-10-16 14:05:34 -0500149static 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
Alexey Gladkovb6c33652021-04-22 14:27:10 +0200155struct ucounts *get_ucounts(struct ucounts *ucounts)
156{
Eric W. Biedermanda70d312021-10-16 14:05:34 -0500157 if (!get_ucounts_or_wrap(ucounts)) {
Alexey Gladkovb6c33652021-04-22 14:27:10 +0200158 put_ucounts(ucounts);
159 ucounts = NULL;
160 }
161 return ucounts;
162}
163
Alexey Gladkov905ae012021-04-22 14:27:09 +0200164struct ucounts *alloc_ucounts(struct user_namespace *ns, kuid_t uid)
Eric W. Biedermanf6b2db1a2016-08-08 13:54:50 -0500165{
166 struct hlist_head *hashent = ucounts_hashentry(ns, uid);
167 struct ucounts *ucounts, *new;
Eric W. Biedermanda70d312021-10-16 14:05:34 -0500168 bool wrapped;
Eric W. Biedermanf6b2db1a2016-08-08 13:54:50 -0500169
Nikolay Borisov880a3852017-01-20 15:21:35 +0200170 spin_lock_irq(&ucounts_lock);
Eric W. Biedermanf6b2db1a2016-08-08 13:54:50 -0500171 ucounts = find_ucounts(ns, uid, hashent);
172 if (!ucounts) {
Nikolay Borisov880a3852017-01-20 15:21:35 +0200173 spin_unlock_irq(&ucounts_lock);
Eric W. Biedermanf6b2db1a2016-08-08 13:54:50 -0500174
175 new = kzalloc(sizeof(*new), GFP_KERNEL);
176 if (!new)
177 return NULL;
178
179 new->ns = ns;
180 new->uid = uid;
Alexey Gladkovb6c33652021-04-22 14:27:10 +0200181 atomic_set(&new->count, 1);
Eric W. Biedermanf6b2db1a2016-08-08 13:54:50 -0500182
Nikolay Borisov880a3852017-01-20 15:21:35 +0200183 spin_lock_irq(&ucounts_lock);
Eric W. Biedermanf6b2db1a2016-08-08 13:54:50 -0500184 ucounts = find_ucounts(ns, uid, hashent);
185 if (ucounts) {
186 kfree(new);
187 } else {
188 hlist_add_head(&new->node, hashent);
Eric W. Biedermanf9d87922022-01-24 12:46:50 -0600189 get_user_ns(new->ns);
Alexey Gladkovb6c33652021-04-22 14:27:10 +0200190 spin_unlock_irq(&ucounts_lock);
191 return new;
Eric W. Biedermanf6b2db1a2016-08-08 13:54:50 -0500192 }
193 }
Eric W. Biedermanda70d312021-10-16 14:05:34 -0500194 wrapped = !get_ucounts_or_wrap(ucounts);
Nikolay Borisov880a3852017-01-20 15:21:35 +0200195 spin_unlock_irq(&ucounts_lock);
Eric W. Biedermanda70d312021-10-16 14:05:34 -0500196 if (wrapped) {
Alexey Gladkov345daff2021-07-27 17:24:18 +0200197 put_ucounts(ucounts);
198 return NULL;
199 }
Eric W. Biedermanf6b2db1a2016-08-08 13:54:50 -0500200 return ucounts;
201}
202
Alexey Gladkov905ae012021-04-22 14:27:09 +0200203void put_ucounts(struct ucounts *ucounts)
Eric W. Biedermanf6b2db1a2016-08-08 13:54:50 -0500204{
Nikolay Borisov880a3852017-01-20 15:21:35 +0200205 unsigned long flags;
206
Alexey Gladkov345daff2021-07-27 17:24:18 +0200207 if (atomic_dec_and_lock_irqsave(&ucounts->count, &ucounts_lock, flags)) {
Eric W. Biedermanf6b2db1a2016-08-08 13:54:50 -0500208 hlist_del_init(&ucounts->node);
Alexey Gladkovb6c33652021-04-22 14:27:10 +0200209 spin_unlock_irqrestore(&ucounts_lock, flags);
Eric W. Biedermanf9d87922022-01-24 12:46:50 -0600210 put_user_ns(ucounts->ns);
Alexey Gladkovb6c33652021-04-22 14:27:10 +0200211 kfree(ucounts);
212 }
Eric W. Biedermanf6b2db1a2016-08-08 13:54:50 -0500213}
214
Alexey Gladkovf9c82a4ea2021-04-22 14:27:08 +0200215static inline bool atomic_long_inc_below(atomic_long_t *v, int u)
Eric W. Biedermanb376c3e2016-08-08 13:41:24 -0500216{
Alexey Gladkovf9c82a4ea2021-04-22 14:27:08 +0200217 long c, old;
218 c = atomic_long_read(v);
Eric W. Biedermanb376c3e2016-08-08 13:41:24 -0500219 for (;;) {
220 if (unlikely(c >= u))
221 return false;
Alexey Gladkovf9c82a4ea2021-04-22 14:27:08 +0200222 old = atomic_long_cmpxchg(v, c, c+1);
Eric W. Biedermanb376c3e2016-08-08 13:41:24 -0500223 if (likely(old == c))
224 return true;
225 c = old;
226 }
227}
228
Eric W. Biederman25f9c082016-08-08 14:41:52 -0500229struct ucounts *inc_ucount(struct user_namespace *ns, kuid_t uid,
230 enum ucount_type type)
Eric W. Biedermanb376c3e2016-08-08 13:41:24 -0500231{
Eric W. Biedermanf6b2db1a2016-08-08 13:54:50 -0500232 struct ucounts *ucounts, *iter, *bad;
233 struct user_namespace *tns;
Alexey Gladkov905ae012021-04-22 14:27:09 +0200234 ucounts = alloc_ucounts(ns, uid);
Eric W. Biedermanf6b2db1a2016-08-08 13:54:50 -0500235 for (iter = ucounts; iter; iter = tns->ucounts) {
Alexey Gladkovf9c82a4ea2021-04-22 14:27:08 +0200236 long max;
Eric W. Biedermanf6b2db1a2016-08-08 13:54:50 -0500237 tns = iter->ns;
Eric W. Biederman25f9c082016-08-08 14:41:52 -0500238 max = READ_ONCE(tns->ucount_max[type]);
Alexey Gladkovf9c82a4ea2021-04-22 14:27:08 +0200239 if (!atomic_long_inc_below(&iter->ucount[type], max))
Eric W. Biedermanb376c3e2016-08-08 13:41:24 -0500240 goto fail;
241 }
Eric W. Biedermanf6b2db1a2016-08-08 13:54:50 -0500242 return ucounts;
Eric W. Biedermanb376c3e2016-08-08 13:41:24 -0500243fail:
Eric W. Biedermanf6b2db1a2016-08-08 13:54:50 -0500244 bad = iter;
245 for (iter = ucounts; iter != bad; iter = iter->ns->ucounts)
Alexey Gladkovf9c82a4ea2021-04-22 14:27:08 +0200246 atomic_long_dec(&iter->ucount[type]);
Eric W. Biedermanb376c3e2016-08-08 13:41:24 -0500247
Eric W. Biedermanf6b2db1a2016-08-08 13:54:50 -0500248 put_ucounts(ucounts);
249 return NULL;
Eric W. Biedermanb376c3e2016-08-08 13:41:24 -0500250}
251
Eric W. Biederman25f9c082016-08-08 14:41:52 -0500252void dec_ucount(struct ucounts *ucounts, enum ucount_type type)
Eric W. Biedermanb376c3e2016-08-08 13:41:24 -0500253{
Eric W. Biedermanf6b2db1a2016-08-08 13:54:50 -0500254 struct ucounts *iter;
255 for (iter = ucounts; iter; iter = iter->ns->ucounts) {
Alexey Gladkovf9c82a4ea2021-04-22 14:27:08 +0200256 long dec = atomic_long_dec_if_positive(&iter->ucount[type]);
Eric W. Biedermanb376c3e2016-08-08 13:41:24 -0500257 WARN_ON_ONCE(dec < 0);
258 }
Eric W. Biedermanf6b2db1a2016-08-08 13:54:50 -0500259 put_ucounts(ucounts);
Eric W. Biedermanb376c3e2016-08-08 13:41:24 -0500260}
261
Alexey Gladkovde399232022-05-18 19:17:30 +0200262long inc_rlimit_ucounts(struct ucounts *ucounts, enum rlimit_type type, long v)
Alexey Gladkov21d1c5e2021-04-22 14:27:11 +0200263{
264 struct ucounts *iter;
Alexey Gladkov59ec7152021-11-29 21:37:25 +0100265 long max = LONG_MAX;
Alexey Gladkov21d1c5e2021-04-22 14:27:11 +0200266 long ret = 0;
267
268 for (iter = ucounts; iter; iter = iter->ns->ucounts) {
Alexey Gladkovde399232022-05-18 19:17:30 +0200269 long new = atomic_long_add_return(v, &iter->rlimit[type]);
Alexey Gladkov21d1c5e2021-04-22 14:27:11 +0200270 if (new < 0 || new > max)
271 ret = LONG_MAX;
272 else if (iter == ucounts)
273 ret = new;
Alexey Gladkovde399232022-05-18 19:17:30 +0200274 max = get_userns_rlimit_max(iter->ns, type);
Alexey Gladkov21d1c5e2021-04-22 14:27:11 +0200275 }
276 return ret;
277}
278
Alexey Gladkovde399232022-05-18 19:17:30 +0200279bool dec_rlimit_ucounts(struct ucounts *ucounts, enum rlimit_type type, long v)
Alexey Gladkov21d1c5e2021-04-22 14:27:11 +0200280{
281 struct ucounts *iter;
Eric W. Biedermanf928ef62021-04-30 13:00:26 -0500282 long new = -1; /* Silence compiler warning */
Alexey Gladkov21d1c5e2021-04-22 14:27:11 +0200283 for (iter = ucounts; iter; iter = iter->ns->ucounts) {
Alexey Gladkovde399232022-05-18 19:17:30 +0200284 long dec = atomic_long_sub_return(v, &iter->rlimit[type]);
Alexey Gladkov21d1c5e2021-04-22 14:27:11 +0200285 WARN_ON_ONCE(dec < 0);
286 if (iter == ucounts)
287 new = dec;
288 }
289 return (new == 0);
290}
291
Eric W. Biederman15bc01e2021-10-16 15:59:49 -0500292static void do_dec_rlimit_put_ucounts(struct ucounts *ucounts,
Alexey Gladkovde399232022-05-18 19:17:30 +0200293 struct ucounts *last, enum rlimit_type type)
Eric W. Biederman15bc01e2021-10-16 15:59:49 -0500294{
295 struct ucounts *iter, *next;
296 for (iter = ucounts; iter != last; iter = next) {
Alexey Gladkovde399232022-05-18 19:17:30 +0200297 long dec = atomic_long_sub_return(1, &iter->rlimit[type]);
Eric W. Biederman15bc01e2021-10-16 15:59:49 -0500298 WARN_ON_ONCE(dec < 0);
299 next = iter->ns->ucounts;
300 if (dec == 0)
301 put_ucounts(iter);
302 }
303}
304
Alexey Gladkovde399232022-05-18 19:17:30 +0200305void dec_rlimit_put_ucounts(struct ucounts *ucounts, enum rlimit_type type)
Eric W. Biederman15bc01e2021-10-16 15:59:49 -0500306{
307 do_dec_rlimit_put_ucounts(ucounts, NULL, type);
308}
309
Alexey Gladkovde399232022-05-18 19:17:30 +0200310long inc_rlimit_get_ucounts(struct ucounts *ucounts, enum rlimit_type type)
Eric W. Biederman15bc01e2021-10-16 15:59:49 -0500311{
312 /* Caller must hold a reference to ucounts */
313 struct ucounts *iter;
Alexey Gladkov59ec7152021-11-29 21:37:25 +0100314 long max = LONG_MAX;
Eric W. Biederman15bc01e2021-10-16 15:59:49 -0500315 long dec, ret = 0;
316
317 for (iter = ucounts; iter; iter = iter->ns->ucounts) {
Alexey Gladkovde399232022-05-18 19:17:30 +0200318 long new = atomic_long_add_return(1, &iter->rlimit[type]);
Eric W. Biederman15bc01e2021-10-16 15:59:49 -0500319 if (new < 0 || new > max)
320 goto unwind;
321 if (iter == ucounts)
322 ret = new;
Alexey Gladkovde399232022-05-18 19:17:30 +0200323 max = get_userns_rlimit_max(iter->ns, type);
Eric W. Biederman15bc01e2021-10-16 15:59:49 -0500324 /*
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;
334dec_unwind:
Alexey Gladkovde399232022-05-18 19:17:30 +0200335 dec = atomic_long_sub_return(1, &iter->rlimit[type]);
Eric W. Biederman15bc01e2021-10-16 15:59:49 -0500336 WARN_ON_ONCE(dec < 0);
337unwind:
338 do_dec_rlimit_put_ucounts(ucounts, iter, type);
339 return 0;
340}
341
Alexey Gladkovde399232022-05-18 19:17:30 +0200342bool is_rlimit_overlimit(struct ucounts *ucounts, enum rlimit_type type, unsigned long rlimit)
Alexey Gladkov21d1c5e2021-04-22 14:27:11 +0200343{
344 struct ucounts *iter;
Alexey Gladkov59ec7152021-11-29 21:37:25 +0100345 long max = rlimit;
346 if (rlimit > LONG_MAX)
347 max = LONG_MAX;
Alexey Gladkov21d1c5e2021-04-22 14:27:11 +0200348 for (iter = ucounts; iter; iter = iter->ns->ucounts) {
Alexey Gladkovde399232022-05-18 19:17:30 +0200349 long val = get_rlimit_value(iter, type);
Eric W. Biederman0cbae9e2022-02-09 18:09:41 -0600350 if (val < 0 || val > max)
Alexey Gladkov21d1c5e2021-04-22 14:27:11 +0200351 return true;
Alexey Gladkovde399232022-05-18 19:17:30 +0200352 max = get_userns_rlimit_max(iter->ns, type);
Alexey Gladkov21d1c5e2021-04-22 14:27:11 +0200353 }
354 return false;
355}
356
Eric W. Biedermandbec2842016-07-30 13:58:49 -0500357static __init int user_namespace_sysctl_init(void)
358{
359#ifdef CONFIG_SYSCTL
Eric W. Biedermanf6b2db1a2016-08-08 13:54:50 -0500360 static struct ctl_table_header *user_header;
Eric W. Biedermandbec2842016-07-30 13:58:49 -0500361 static struct ctl_table empty[1];
362 /*
Eric W. Biedermanf6b2db1a2016-08-08 13:54:50 -0500363 * It is necessary to register the user directory in the
Eric W. Biedermandbec2842016-07-30 13:58:49 -0500364 * default set so that registrations in the child sets work
365 * properly.
366 */
Eric W. Biedermanf6b2db1a2016-08-08 13:54:50 -0500367 user_header = register_sysctl("user", empty);
Luis R. Rodriguezed5bd7d2017-02-08 14:30:50 -0800368 kmemleak_ignore(user_header);
Eric W. Biedermanf6b2db1a2016-08-08 13:54:50 -0500369 BUG_ON(!user_header);
Eric W. Biedermandbec2842016-07-30 13:58:49 -0500370 BUG_ON(!setup_userns_sysctls(&init_user_ns));
371#endif
Alexey Gladkov905ae012021-04-22 14:27:09 +0200372 hlist_add_ucounts(&init_ucounts);
Alexey Gladkov21d1c5e2021-04-22 14:27:11 +0200373 inc_rlimit_ucounts(&init_ucounts, UCOUNT_RLIMIT_NPROC, 1);
Eric W. Biedermandbec2842016-07-30 13:58:49 -0500374 return 0;
375}
376subsys_initcall(user_namespace_sysctl_init);