1 // SPDX-License-Identifier: GPL-2.0-only !! 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 */
2 7
3 #include <linux/stat.h> 8 #include <linux/stat.h>
4 #include <linux/sysctl.h> 9 #include <linux/sysctl.h>
5 #include <linux/slab.h> 10 #include <linux/slab.h>
6 #include <linux/cred.h> <<
7 #include <linux/hash.h> 11 #include <linux/hash.h>
8 #include <linux/kmemleak.h> <<
9 #include <linux/user_namespace.h> 12 #include <linux/user_namespace.h>
10 13
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 14 #define UCOUNTS_HASHTABLE_BITS 10
18 static struct hlist_head ucounts_hashtable[(1 15 static struct hlist_head ucounts_hashtable[(1 << UCOUNTS_HASHTABLE_BITS)];
19 static DEFINE_SPINLOCK(ucounts_lock); 16 static DEFINE_SPINLOCK(ucounts_lock);
20 17
21 #define ucounts_hashfn(ns, uid) 18 #define ucounts_hashfn(ns, uid) \
22 hash_long((unsigned long)__kuid_val(ui 19 hash_long((unsigned long)__kuid_val(uid) + (unsigned long)(ns), \
23 UCOUNTS_HASHTABLE_BITS) 20 UCOUNTS_HASHTABLE_BITS)
24 #define ucounts_hashentry(ns, uid) \ 21 #define ucounts_hashentry(ns, uid) \
25 (ucounts_hashtable + ucounts_hashfn(ns 22 (ucounts_hashtable + ucounts_hashfn(ns, uid))
26 23
27 24
28 #ifdef CONFIG_SYSCTL 25 #ifdef CONFIG_SYSCTL
29 static struct ctl_table_set * 26 static struct ctl_table_set *
30 set_lookup(struct ctl_table_root *root) 27 set_lookup(struct ctl_table_root *root)
31 { 28 {
32 return ¤t_user_ns()->set; 29 return ¤t_user_ns()->set;
33 } 30 }
34 31
35 static int set_is_seen(struct ctl_table_set *s 32 static int set_is_seen(struct ctl_table_set *set)
36 { 33 {
37 return ¤t_user_ns()->set == set; 34 return ¤t_user_ns()->set == set;
38 } 35 }
39 36
40 static int set_permissions(struct ctl_table_he 37 static int set_permissions(struct ctl_table_header *head,
41 const struct ctl_ta !! 38 struct ctl_table *table)
42 { 39 {
43 struct user_namespace *user_ns = 40 struct user_namespace *user_ns =
44 container_of(head->set, struct 41 container_of(head->set, struct user_namespace, set);
45 int mode; 42 int mode;
46 43
47 /* Allow users with CAP_SYS_RESOURCE u 44 /* Allow users with CAP_SYS_RESOURCE unrestrained access */
48 if (ns_capable(user_ns, CAP_SYS_RESOUR 45 if (ns_capable(user_ns, CAP_SYS_RESOURCE))
49 mode = (table->mode & S_IRWXU) 46 mode = (table->mode & S_IRWXU) >> 6;
50 else 47 else
51 /* Allow all others at most read-only 48 /* Allow all others at most read-only access */
52 mode = table->mode & S_IROTH; 49 mode = table->mode & S_IROTH;
53 return (mode << 6) | (mode << 3) | mod 50 return (mode << 6) | (mode << 3) | mode;
54 } 51 }
55 52
56 static struct ctl_table_root set_root = { 53 static struct ctl_table_root set_root = {
57 .lookup = set_lookup, 54 .lookup = set_lookup,
58 .permissions = set_permissions, 55 .permissions = set_permissions,
59 }; 56 };
60 57
61 static long ue_zero = 0; !! 58 static int zero = 0;
62 static long ue_int_max = INT_MAX; !! 59 static int int_max = INT_MAX;
63 !! 60 #define UCOUNT_ENTRY(name) \
64 #define UCOUNT_ENTRY(name) !! 61 { \
65 { !! 62 .procname = name, \
66 .procname = name, !! 63 .maxlen = sizeof(int), \
67 .maxlen = sizeof(long) !! 64 .mode = 0644, \
68 .mode = 0644, !! 65 .proc_handler = proc_dointvec_minmax, \
69 .proc_handler = proc_doulong !! 66 .extra1 = &zero, \
70 .extra1 = &ue_zero, !! 67 .extra2 = &int_max, \
71 .extra2 = &ue_int_max, <<
72 } 68 }
73 static struct ctl_table user_table[] = { 69 static struct ctl_table user_table[] = {
74 UCOUNT_ENTRY("max_user_namespaces"), 70 UCOUNT_ENTRY("max_user_namespaces"),
75 UCOUNT_ENTRY("max_pid_namespaces"), 71 UCOUNT_ENTRY("max_pid_namespaces"),
76 UCOUNT_ENTRY("max_uts_namespaces"), 72 UCOUNT_ENTRY("max_uts_namespaces"),
77 UCOUNT_ENTRY("max_ipc_namespaces"), 73 UCOUNT_ENTRY("max_ipc_namespaces"),
78 UCOUNT_ENTRY("max_net_namespaces"), 74 UCOUNT_ENTRY("max_net_namespaces"),
79 UCOUNT_ENTRY("max_mnt_namespaces"), 75 UCOUNT_ENTRY("max_mnt_namespaces"),
80 UCOUNT_ENTRY("max_cgroup_namespaces"), 76 UCOUNT_ENTRY("max_cgroup_namespaces"),
81 UCOUNT_ENTRY("max_time_namespaces"), !! 77 { }
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 }; 78 };
91 #endif /* CONFIG_SYSCTL */ 79 #endif /* CONFIG_SYSCTL */
92 80
93 bool setup_userns_sysctls(struct user_namespac 81 bool setup_userns_sysctls(struct user_namespace *ns)
94 { 82 {
95 #ifdef CONFIG_SYSCTL 83 #ifdef CONFIG_SYSCTL
96 struct ctl_table *tbl; 84 struct ctl_table *tbl;
97 <<
98 BUILD_BUG_ON(ARRAY_SIZE(user_table) != <<
99 setup_sysctl_set(&ns->set, &set_root, 85 setup_sysctl_set(&ns->set, &set_root, set_is_seen);
100 tbl = kmemdup(user_table, sizeof(user_ 86 tbl = kmemdup(user_table, sizeof(user_table), GFP_KERNEL);
101 if (tbl) { 87 if (tbl) {
102 int i; 88 int i;
103 for (i = 0; i < UCOUNT_COUNTS; 89 for (i = 0; i < UCOUNT_COUNTS; i++) {
104 tbl[i].data = &ns->uco 90 tbl[i].data = &ns->ucount_max[i];
105 } 91 }
106 ns->sysctls = __register_sysct !! 92 ns->sysctls = __register_sysctl_table(&ns->set, "user", tbl);
107 <<
108 } 93 }
109 if (!ns->sysctls) { 94 if (!ns->sysctls) {
110 kfree(tbl); 95 kfree(tbl);
111 retire_sysctl_set(&ns->set); 96 retire_sysctl_set(&ns->set);
112 return false; 97 return false;
113 } 98 }
114 #endif 99 #endif
115 return true; 100 return true;
116 } 101 }
117 102
118 void retire_userns_sysctls(struct user_namespa 103 void retire_userns_sysctls(struct user_namespace *ns)
119 { 104 {
120 #ifdef CONFIG_SYSCTL 105 #ifdef CONFIG_SYSCTL
121 const struct ctl_table *tbl; !! 106 struct ctl_table *tbl;
122 107
123 tbl = ns->sysctls->ctl_table_arg; 108 tbl = ns->sysctls->ctl_table_arg;
124 unregister_sysctl_table(ns->sysctls); 109 unregister_sysctl_table(ns->sysctls);
125 retire_sysctl_set(&ns->set); 110 retire_sysctl_set(&ns->set);
126 kfree(tbl); 111 kfree(tbl);
127 #endif 112 #endif
128 } 113 }
129 114
130 static struct ucounts *find_ucounts(struct use 115 static struct ucounts *find_ucounts(struct user_namespace *ns, kuid_t uid, struct hlist_head *hashent)
131 { 116 {
132 struct ucounts *ucounts; 117 struct ucounts *ucounts;
133 118
134 hlist_for_each_entry(ucounts, hashent, 119 hlist_for_each_entry(ucounts, hashent, node) {
135 if (uid_eq(ucounts->uid, uid) 120 if (uid_eq(ucounts->uid, uid) && (ucounts->ns == ns))
136 return ucounts; 121 return ucounts;
137 } 122 }
138 return NULL; 123 return NULL;
139 } 124 }
140 125
141 static void hlist_add_ucounts(struct ucounts * !! 126 static struct ucounts *get_ucounts(struct user_namespace *ns, kuid_t uid)
142 { <<
143 struct hlist_head *hashent = ucounts_h <<
144 spin_lock_irq(&ucounts_lock); <<
145 hlist_add_head(&ucounts->node, hashent <<
146 spin_unlock_irq(&ucounts_lock); <<
147 } <<
148 <<
149 static inline bool get_ucounts_or_wrap(struct <<
150 { <<
151 /* Returns true on a successful get, f <<
152 return !atomic_add_negative(1, &ucount <<
153 } <<
154 <<
155 struct ucounts *get_ucounts(struct ucounts *uc <<
156 { <<
157 if (!get_ucounts_or_wrap(ucounts)) { <<
158 put_ucounts(ucounts); <<
159 ucounts = NULL; <<
160 } <<
161 return ucounts; <<
162 } <<
163 <<
164 struct ucounts *alloc_ucounts(struct user_name <<
165 { 127 {
166 struct hlist_head *hashent = ucounts_h 128 struct hlist_head *hashent = ucounts_hashentry(ns, uid);
167 struct ucounts *ucounts, *new; 129 struct ucounts *ucounts, *new;
168 bool wrapped; <<
169 130
170 spin_lock_irq(&ucounts_lock); 131 spin_lock_irq(&ucounts_lock);
171 ucounts = find_ucounts(ns, uid, hashen 132 ucounts = find_ucounts(ns, uid, hashent);
172 if (!ucounts) { 133 if (!ucounts) {
173 spin_unlock_irq(&ucounts_lock) 134 spin_unlock_irq(&ucounts_lock);
174 135
175 new = kzalloc(sizeof(*new), GF 136 new = kzalloc(sizeof(*new), GFP_KERNEL);
176 if (!new) 137 if (!new)
177 return NULL; 138 return NULL;
178 139
179 new->ns = ns; 140 new->ns = ns;
180 new->uid = uid; 141 new->uid = uid;
181 atomic_set(&new->count, 1); !! 142 new->count = 0;
182 143
183 spin_lock_irq(&ucounts_lock); 144 spin_lock_irq(&ucounts_lock);
184 ucounts = find_ucounts(ns, uid 145 ucounts = find_ucounts(ns, uid, hashent);
185 if (ucounts) { 146 if (ucounts) {
186 kfree(new); 147 kfree(new);
187 } else { 148 } else {
188 hlist_add_head(&new->n 149 hlist_add_head(&new->node, hashent);
189 get_user_ns(new->ns); !! 150 ucounts = new;
190 spin_unlock_irq(&ucoun <<
191 return new; <<
192 } 151 }
193 } 152 }
194 wrapped = !get_ucounts_or_wrap(ucounts !! 153 if (ucounts->count == INT_MAX)
>> 154 ucounts = NULL;
>> 155 else
>> 156 ucounts->count += 1;
195 spin_unlock_irq(&ucounts_lock); 157 spin_unlock_irq(&ucounts_lock);
196 if (wrapped) { <<
197 put_ucounts(ucounts); <<
198 return NULL; <<
199 } <<
200 return ucounts; 158 return ucounts;
201 } 159 }
202 160
203 void put_ucounts(struct ucounts *ucounts) !! 161 static void put_ucounts(struct ucounts *ucounts)
204 { 162 {
205 unsigned long flags; 163 unsigned long flags;
206 164
207 if (atomic_dec_and_lock_irqsave(&ucoun !! 165 spin_lock_irqsave(&ucounts_lock, flags);
>> 166 ucounts->count -= 1;
>> 167 if (!ucounts->count)
208 hlist_del_init(&ucounts->node) 168 hlist_del_init(&ucounts->node);
209 spin_unlock_irqrestore(&ucount !! 169 else
210 put_user_ns(ucounts->ns); !! 170 ucounts = NULL;
211 kfree(ucounts); !! 171 spin_unlock_irqrestore(&ucounts_lock, flags);
212 } !! 172
>> 173 kfree(ucounts);
213 } 174 }
214 175
215 static inline bool atomic_long_inc_below(atomi !! 176 static inline bool atomic_inc_below(atomic_t *v, int u)
216 { 177 {
217 long c, old; !! 178 int c, old;
218 c = atomic_long_read(v); !! 179 c = atomic_read(v);
219 for (;;) { 180 for (;;) {
220 if (unlikely(c >= u)) 181 if (unlikely(c >= u))
221 return false; 182 return false;
222 old = atomic_long_cmpxchg(v, c !! 183 old = atomic_cmpxchg(v, c, c+1);
223 if (likely(old == c)) 184 if (likely(old == c))
224 return true; 185 return true;
225 c = old; 186 c = old;
226 } 187 }
227 } 188 }
228 189
229 struct ucounts *inc_ucount(struct user_namespa 190 struct ucounts *inc_ucount(struct user_namespace *ns, kuid_t uid,
230 enum ucount_type ty 191 enum ucount_type type)
231 { 192 {
232 struct ucounts *ucounts, *iter, *bad; 193 struct ucounts *ucounts, *iter, *bad;
233 struct user_namespace *tns; 194 struct user_namespace *tns;
234 ucounts = alloc_ucounts(ns, uid); !! 195 ucounts = get_ucounts(ns, uid);
235 for (iter = ucounts; iter; iter = tns- 196 for (iter = ucounts; iter; iter = tns->ucounts) {
236 long max; !! 197 int max;
237 tns = iter->ns; 198 tns = iter->ns;
238 max = READ_ONCE(tns->ucount_ma 199 max = READ_ONCE(tns->ucount_max[type]);
239 if (!atomic_long_inc_below(&it !! 200 if (!atomic_inc_below(&iter->ucount[type], max))
240 goto fail; 201 goto fail;
241 } 202 }
242 return ucounts; 203 return ucounts;
243 fail: 204 fail:
244 bad = iter; 205 bad = iter;
245 for (iter = ucounts; iter != bad; iter 206 for (iter = ucounts; iter != bad; iter = iter->ns->ucounts)
246 atomic_long_dec(&iter->ucount[ !! 207 atomic_dec(&iter->ucount[type]);
247 208
248 put_ucounts(ucounts); 209 put_ucounts(ucounts);
249 return NULL; 210 return NULL;
250 } 211 }
251 212
252 void dec_ucount(struct ucounts *ucounts, enum 213 void dec_ucount(struct ucounts *ucounts, enum ucount_type type)
253 { 214 {
254 struct ucounts *iter; 215 struct ucounts *iter;
255 for (iter = ucounts; iter; iter = iter 216 for (iter = ucounts; iter; iter = iter->ns->ucounts) {
256 long dec = atomic_long_dec_if_ !! 217 int dec = atomic_dec_if_positive(&iter->ucount[type]);
257 WARN_ON_ONCE(dec < 0); 218 WARN_ON_ONCE(dec < 0);
258 } 219 }
259 put_ucounts(ucounts); 220 put_ucounts(ucounts);
260 } 221 }
261 222
262 long inc_rlimit_ucounts(struct ucounts *ucount <<
263 { <<
264 struct ucounts *iter; <<
265 long max = LONG_MAX; <<
266 long ret = 0; <<
267 <<
268 for (iter = ucounts; iter; iter = iter <<
269 long new = atomic_long_add_ret <<
270 if (new < 0 || new > max) <<
271 ret = LONG_MAX; <<
272 else if (iter == ucounts) <<
273 ret = new; <<
274 max = get_userns_rlimit_max(it <<
275 } <<
276 return ret; <<
277 } <<
278 <<
279 bool dec_rlimit_ucounts(struct ucounts *ucount <<
280 { <<
281 struct ucounts *iter; <<
282 long new = -1; /* Silence compiler war <<
283 for (iter = ucounts; iter; iter = iter <<
284 long dec = atomic_long_sub_ret <<
285 WARN_ON_ONCE(dec < 0); <<
286 if (iter == ucounts) <<
287 new = dec; <<
288 } <<
289 return (new == 0); <<
290 } <<
291 <<
292 static void do_dec_rlimit_put_ucounts(struct u <<
293 struct ucounts <<
294 { <<
295 struct ucounts *iter, *next; <<
296 for (iter = ucounts; iter != last; ite <<
297 long dec = atomic_long_sub_ret <<
298 WARN_ON_ONCE(dec < 0); <<
299 next = iter->ns->ucounts; <<
300 if (dec == 0) <<
301 put_ucounts(iter); <<
302 } <<
303 } <<
304 <<
305 void dec_rlimit_put_ucounts(struct ucounts *uc <<
306 { <<
307 do_dec_rlimit_put_ucounts(ucounts, NUL <<
308 } <<
309 <<
310 long inc_rlimit_get_ucounts(struct ucounts *uc <<
311 { <<
312 /* Caller must hold a reference to uco <<
313 struct ucounts *iter; <<
314 long max = LONG_MAX; <<
315 long dec, ret = 0; <<
316 <<
317 for (iter = ucounts; iter; iter = iter <<
318 long new = atomic_long_add_ret <<
319 if (new < 0 || new > max) <<
320 goto unwind; <<
321 if (iter == ucounts) <<
322 ret = new; <<
323 max = get_userns_rlimit_max(it <<
324 /* <<
325 * Grab an extra ucount refere <<
326 * the rlimit count was previo <<
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- <<
336 WARN_ON_ONCE(dec < 0); <<
337 unwind: <<
338 do_dec_rlimit_put_ucounts(ucounts, ite <<
339 return 0; <<
340 } <<
341 <<
342 bool is_rlimit_overlimit(struct ucounts *ucoun <<
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 <<
349 long val = get_rlimit_value(it <<
350 if (val < 0 || val > max) <<
351 return true; <<
352 max = get_userns_rlimit_max(it <<
353 } <<
354 return false; <<
355 } <<
356 <<
357 static __init int user_namespace_sysctl_init(v 223 static __init int user_namespace_sysctl_init(void)
358 { 224 {
359 #ifdef CONFIG_SYSCTL 225 #ifdef CONFIG_SYSCTL
360 static struct ctl_table_header *user_h 226 static struct ctl_table_header *user_header;
361 static struct ctl_table empty[1]; 227 static struct ctl_table empty[1];
362 /* 228 /*
363 * It is necessary to register the use 229 * It is necessary to register the user directory in the
364 * default set so that registrations i 230 * default set so that registrations in the child sets work
365 * properly. 231 * properly.
366 */ 232 */
367 user_header = register_sysctl_sz("user !! 233 user_header = register_sysctl("user", empty);
368 kmemleak_ignore(user_header); 234 kmemleak_ignore(user_header);
369 BUG_ON(!user_header); 235 BUG_ON(!user_header);
370 BUG_ON(!setup_userns_sysctls(&init_use 236 BUG_ON(!setup_userns_sysctls(&init_user_ns));
371 #endif 237 #endif
372 hlist_add_ucounts(&init_ucounts); <<
373 inc_rlimit_ucounts(&init_ucounts, UCOU <<
374 return 0; 238 return 0;
375 } 239 }
376 subsys_initcall(user_namespace_sysctl_init); 240 subsys_initcall(user_namespace_sysctl_init);
377 241
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.