Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * The "user cache". | |
3 | * | |
4 | * (C) Copyright 1991-2000 Linus Torvalds | |
5 | * | |
6 | * We have a per-user structure to keep track of how many | |
7 | * processes, files etc the user has claimed, in order to be | |
8 | * able to have per-user limits for system resources. | |
9 | */ | |
10 | ||
11 | #include <linux/init.h> | |
12 | #include <linux/sched.h> | |
13 | #include <linux/slab.h> | |
14 | #include <linux/bitops.h> | |
15 | #include <linux/key.h> | |
4021cb27 | 16 | #include <linux/interrupt.h> |
acce292c CLG |
17 | #include <linux/module.h> |
18 | #include <linux/user_namespace.h> | |
d84f4f99 | 19 | #include "cred-internals.h" |
1da177e4 | 20 | |
aee16ce7 PE |
21 | struct user_namespace init_user_ns = { |
22 | .kref = { | |
18b6e041 | 23 | .refcount = ATOMIC_INIT(1), |
aee16ce7 | 24 | }, |
18b6e041 | 25 | .creator = &root_user, |
aee16ce7 PE |
26 | }; |
27 | EXPORT_SYMBOL_GPL(init_user_ns); | |
28 | ||
1da177e4 LT |
29 | /* |
30 | * UID task count cache, to get fast user lookup in "alloc_uid" | |
31 | * when changing user ID's (ie setuid() and friends). | |
32 | */ | |
33 | ||
1da177e4 LT |
34 | #define UIDHASH_MASK (UIDHASH_SZ - 1) |
35 | #define __uidhashfn(uid) (((uid >> UIDHASH_BITS) + uid) & UIDHASH_MASK) | |
acce292c | 36 | #define uidhashentry(ns, uid) ((ns)->uidhash_table + __uidhashfn((uid))) |
1da177e4 | 37 | |
e18b890b | 38 | static struct kmem_cache *uid_cachep; |
4021cb27 IM |
39 | |
40 | /* | |
41 | * The uidhash_lock is mostly taken from process context, but it is | |
42 | * occasionally also taken from softirq/tasklet context, when | |
43 | * task-structs get RCU-freed. Hence all locking must be softirq-safe. | |
3fa97c9d AM |
44 | * But free_uid() is also called with local interrupts disabled, and running |
45 | * local_bh_enable() with local interrupts disabled is an error - we'll run | |
46 | * softirq callbacks, and they can unconditionally enable interrupts, and | |
47 | * the caller of free_uid() didn't expect that.. | |
4021cb27 | 48 | */ |
1da177e4 LT |
49 | static DEFINE_SPINLOCK(uidhash_lock); |
50 | ||
18b6e041 | 51 | /* root_user.__count is 2, 1 for init task cred, 1 for init_user_ns->creator */ |
1da177e4 | 52 | struct user_struct root_user = { |
18b6e041 | 53 | .__count = ATOMIC_INIT(2), |
1da177e4 LT |
54 | .processes = ATOMIC_INIT(1), |
55 | .files = ATOMIC_INIT(0), | |
56 | .sigpending = ATOMIC_INIT(0), | |
1da177e4 | 57 | .locked_shm = 0, |
18b6e041 | 58 | .user_ns = &init_user_ns, |
052f1dc7 | 59 | #ifdef CONFIG_USER_SCHED |
4cf86d77 | 60 | .tg = &init_task_group, |
24e377a8 | 61 | #endif |
1da177e4 LT |
62 | }; |
63 | ||
5cb350ba DG |
64 | /* |
65 | * These routines must be called with the uidhash spinlock held! | |
66 | */ | |
40aeb400 | 67 | static void uid_hash_insert(struct user_struct *up, struct hlist_head *hashent) |
5cb350ba DG |
68 | { |
69 | hlist_add_head(&up->uidhash_node, hashent); | |
70 | } | |
71 | ||
40aeb400 | 72 | static void uid_hash_remove(struct user_struct *up) |
5cb350ba DG |
73 | { |
74 | hlist_del_init(&up->uidhash_node); | |
75 | } | |
76 | ||
40aeb400 | 77 | static struct user_struct *uid_hash_find(uid_t uid, struct hlist_head *hashent) |
5cb350ba DG |
78 | { |
79 | struct user_struct *user; | |
80 | struct hlist_node *h; | |
81 | ||
82 | hlist_for_each_entry(user, h, hashent, uidhash_node) { | |
83 | if (user->uid == uid) { | |
84 | atomic_inc(&user->__count); | |
85 | return user; | |
86 | } | |
87 | } | |
88 | ||
89 | return NULL; | |
90 | } | |
91 | ||
052f1dc7 | 92 | #ifdef CONFIG_USER_SCHED |
5cb350ba | 93 | |
24e377a8 SV |
94 | static void sched_destroy_user(struct user_struct *up) |
95 | { | |
96 | sched_destroy_group(up->tg); | |
97 | } | |
98 | ||
99 | static int sched_create_user(struct user_struct *up) | |
100 | { | |
101 | int rc = 0; | |
102 | ||
eff766a6 | 103 | up->tg = sched_create_group(&root_task_group); |
24e377a8 SV |
104 | if (IS_ERR(up->tg)) |
105 | rc = -ENOMEM; | |
106 | ||
107 | return rc; | |
108 | } | |
109 | ||
052f1dc7 | 110 | #else /* CONFIG_USER_SCHED */ |
b1a8c172 DG |
111 | |
112 | static void sched_destroy_user(struct user_struct *up) { } | |
113 | static int sched_create_user(struct user_struct *up) { return 0; } | |
b1a8c172 | 114 | |
052f1dc7 | 115 | #endif /* CONFIG_USER_SCHED */ |
b1a8c172 | 116 | |
052f1dc7 | 117 | #if defined(CONFIG_USER_SCHED) && defined(CONFIG_SYSFS) |
b1a8c172 | 118 | |
eb41d946 | 119 | static struct kset *uids_kset; /* represents the /sys/kernel/uids/ directory */ |
b1a8c172 DG |
120 | static DEFINE_MUTEX(uids_mutex); |
121 | ||
5cb350ba DG |
122 | static inline void uids_mutex_lock(void) |
123 | { | |
124 | mutex_lock(&uids_mutex); | |
125 | } | |
24e377a8 | 126 | |
5cb350ba DG |
127 | static inline void uids_mutex_unlock(void) |
128 | { | |
129 | mutex_unlock(&uids_mutex); | |
130 | } | |
24e377a8 | 131 | |
eb41d946 | 132 | /* uid directory attributes */ |
052f1dc7 | 133 | #ifdef CONFIG_FAIR_GROUP_SCHED |
eb41d946 KS |
134 | static ssize_t cpu_shares_show(struct kobject *kobj, |
135 | struct kobj_attribute *attr, | |
136 | char *buf) | |
5cb350ba | 137 | { |
eb41d946 | 138 | struct user_struct *up = container_of(kobj, struct user_struct, kobj); |
24e377a8 | 139 | |
eb41d946 | 140 | return sprintf(buf, "%lu\n", sched_group_shares(up->tg)); |
5cb350ba DG |
141 | } |
142 | ||
eb41d946 KS |
143 | static ssize_t cpu_shares_store(struct kobject *kobj, |
144 | struct kobj_attribute *attr, | |
145 | const char *buf, size_t size) | |
5cb350ba | 146 | { |
eb41d946 | 147 | struct user_struct *up = container_of(kobj, struct user_struct, kobj); |
5cb350ba DG |
148 | unsigned long shares; |
149 | int rc; | |
150 | ||
eb41d946 | 151 | sscanf(buf, "%lu", &shares); |
5cb350ba DG |
152 | |
153 | rc = sched_group_set_shares(up->tg, shares); | |
154 | ||
155 | return (rc ? rc : size); | |
156 | } | |
157 | ||
eb41d946 KS |
158 | static struct kobj_attribute cpu_share_attr = |
159 | __ATTR(cpu_share, 0644, cpu_shares_show, cpu_shares_store); | |
052f1dc7 | 160 | #endif |
eb41d946 | 161 | |
052f1dc7 | 162 | #ifdef CONFIG_RT_GROUP_SCHED |
9f0c1e56 PZ |
163 | static ssize_t cpu_rt_runtime_show(struct kobject *kobj, |
164 | struct kobj_attribute *attr, | |
165 | char *buf) | |
166 | { | |
167 | struct user_struct *up = container_of(kobj, struct user_struct, kobj); | |
168 | ||
af4491e5 | 169 | return sprintf(buf, "%ld\n", sched_group_rt_runtime(up->tg)); |
9f0c1e56 PZ |
170 | } |
171 | ||
172 | static ssize_t cpu_rt_runtime_store(struct kobject *kobj, | |
173 | struct kobj_attribute *attr, | |
174 | const char *buf, size_t size) | |
175 | { | |
176 | struct user_struct *up = container_of(kobj, struct user_struct, kobj); | |
177 | unsigned long rt_runtime; | |
178 | int rc; | |
179 | ||
af4491e5 | 180 | sscanf(buf, "%ld", &rt_runtime); |
9f0c1e56 PZ |
181 | |
182 | rc = sched_group_set_rt_runtime(up->tg, rt_runtime); | |
183 | ||
184 | return (rc ? rc : size); | |
185 | } | |
186 | ||
187 | static struct kobj_attribute cpu_rt_runtime_attr = | |
188 | __ATTR(cpu_rt_runtime, 0644, cpu_rt_runtime_show, cpu_rt_runtime_store); | |
d0b27fa7 PZ |
189 | |
190 | static ssize_t cpu_rt_period_show(struct kobject *kobj, | |
191 | struct kobj_attribute *attr, | |
192 | char *buf) | |
193 | { | |
194 | struct user_struct *up = container_of(kobj, struct user_struct, kobj); | |
195 | ||
196 | return sprintf(buf, "%lu\n", sched_group_rt_period(up->tg)); | |
197 | } | |
198 | ||
199 | static ssize_t cpu_rt_period_store(struct kobject *kobj, | |
200 | struct kobj_attribute *attr, | |
201 | const char *buf, size_t size) | |
202 | { | |
203 | struct user_struct *up = container_of(kobj, struct user_struct, kobj); | |
204 | unsigned long rt_period; | |
205 | int rc; | |
206 | ||
207 | sscanf(buf, "%lu", &rt_period); | |
208 | ||
209 | rc = sched_group_set_rt_period(up->tg, rt_period); | |
210 | ||
211 | return (rc ? rc : size); | |
212 | } | |
213 | ||
214 | static struct kobj_attribute cpu_rt_period_attr = | |
215 | __ATTR(cpu_rt_period, 0644, cpu_rt_period_show, cpu_rt_period_store); | |
052f1dc7 | 216 | #endif |
9f0c1e56 | 217 | |
eb41d946 KS |
218 | /* default attributes per uid directory */ |
219 | static struct attribute *uids_attributes[] = { | |
052f1dc7 | 220 | #ifdef CONFIG_FAIR_GROUP_SCHED |
eb41d946 | 221 | &cpu_share_attr.attr, |
052f1dc7 PZ |
222 | #endif |
223 | #ifdef CONFIG_RT_GROUP_SCHED | |
9f0c1e56 | 224 | &cpu_rt_runtime_attr.attr, |
d0b27fa7 | 225 | &cpu_rt_period_attr.attr, |
052f1dc7 | 226 | #endif |
eb41d946 KS |
227 | NULL |
228 | }; | |
229 | ||
230 | /* the lifetime of user_struct is not managed by the core (now) */ | |
231 | static void uids_release(struct kobject *kobj) | |
5cb350ba | 232 | { |
eb41d946 | 233 | return; |
5cb350ba DG |
234 | } |
235 | ||
eb41d946 KS |
236 | static struct kobj_type uids_ktype = { |
237 | .sysfs_ops = &kobj_sysfs_ops, | |
238 | .default_attrs = uids_attributes, | |
239 | .release = uids_release, | |
240 | }; | |
241 | ||
242 | /* create /sys/kernel/uids/<uid>/cpu_share file for this user */ | |
243 | static int uids_user_create(struct user_struct *up) | |
1da177e4 | 244 | { |
eb41d946 | 245 | struct kobject *kobj = &up->kobj; |
5cb350ba DG |
246 | int error; |
247 | ||
eb41d946 | 248 | memset(kobj, 0, sizeof(struct kobject)); |
eb41d946 | 249 | kobj->kset = uids_kset; |
cf15126b GKH |
250 | error = kobject_init_and_add(kobj, &uids_ktype, NULL, "%d", up->uid); |
251 | if (error) { | |
252 | kobject_put(kobj); | |
5cb350ba | 253 | goto done; |
cf15126b | 254 | } |
5cb350ba | 255 | |
fb7dde37 | 256 | kobject_uevent(kobj, KOBJ_ADD); |
5cb350ba DG |
257 | done: |
258 | return error; | |
1da177e4 LT |
259 | } |
260 | ||
eb41d946 | 261 | /* create these entries in sysfs: |
5cb350ba DG |
262 | * "/sys/kernel/uids" directory |
263 | * "/sys/kernel/uids/0" directory (for root user) | |
264 | * "/sys/kernel/uids/0/cpu_share" file (for root user) | |
265 | */ | |
eb41d946 | 266 | int __init uids_sysfs_init(void) |
1da177e4 | 267 | { |
0ff21e46 | 268 | uids_kset = kset_create_and_add("uids", NULL, kernel_kobj); |
eb41d946 KS |
269 | if (!uids_kset) |
270 | return -ENOMEM; | |
5cb350ba | 271 | |
eb41d946 | 272 | return uids_user_create(&root_user); |
1da177e4 LT |
273 | } |
274 | ||
5cb350ba DG |
275 | /* work function to remove sysfs directory for a user and free up |
276 | * corresponding structures. | |
277 | */ | |
278 | static void remove_user_sysfs_dir(struct work_struct *w) | |
1da177e4 | 279 | { |
5cb350ba | 280 | struct user_struct *up = container_of(w, struct user_struct, work); |
5cb350ba DG |
281 | unsigned long flags; |
282 | int remove_user = 0; | |
1da177e4 | 283 | |
5cb350ba DG |
284 | /* Make uid_hash_remove() + sysfs_remove_file() + kobject_del() |
285 | * atomic. | |
286 | */ | |
287 | uids_mutex_lock(); | |
288 | ||
289 | local_irq_save(flags); | |
290 | ||
291 | if (atomic_dec_and_lock(&up->__count, &uidhash_lock)) { | |
292 | uid_hash_remove(up); | |
293 | remove_user = 1; | |
294 | spin_unlock_irqrestore(&uidhash_lock, flags); | |
295 | } else { | |
296 | local_irq_restore(flags); | |
1da177e4 LT |
297 | } |
298 | ||
5cb350ba DG |
299 | if (!remove_user) |
300 | goto done; | |
301 | ||
eb41d946 KS |
302 | kobject_uevent(&up->kobj, KOBJ_REMOVE); |
303 | kobject_del(&up->kobj); | |
304 | kobject_put(&up->kobj); | |
5cb350ba DG |
305 | |
306 | sched_destroy_user(up); | |
307 | key_put(up->uid_keyring); | |
308 | key_put(up->session_keyring); | |
309 | kmem_cache_free(uid_cachep, up); | |
310 | ||
311 | done: | |
312 | uids_mutex_unlock(); | |
313 | } | |
314 | ||
315 | /* IRQs are disabled and uidhash_lock is held upon function entry. | |
316 | * IRQ state (as stored in flags) is restored and uidhash_lock released | |
317 | * upon function exit. | |
318 | */ | |
18b6e041 | 319 | static void free_user(struct user_struct *up, unsigned long flags) |
5cb350ba DG |
320 | { |
321 | /* restore back the count */ | |
322 | atomic_inc(&up->__count); | |
323 | spin_unlock_irqrestore(&uidhash_lock, flags); | |
324 | ||
18b6e041 | 325 | put_user_ns(up->user_ns); |
5cb350ba DG |
326 | INIT_WORK(&up->work, remove_user_sysfs_dir); |
327 | schedule_work(&up->work); | |
1da177e4 LT |
328 | } |
329 | ||
052f1dc7 | 330 | #else /* CONFIG_USER_SCHED && CONFIG_SYSFS */ |
5cb350ba | 331 | |
eb41d946 KS |
332 | int uids_sysfs_init(void) { return 0; } |
333 | static inline int uids_user_create(struct user_struct *up) { return 0; } | |
5cb350ba DG |
334 | static inline void uids_mutex_lock(void) { } |
335 | static inline void uids_mutex_unlock(void) { } | |
336 | ||
337 | /* IRQs are disabled and uidhash_lock is held upon function entry. | |
338 | * IRQ state (as stored in flags) is restored and uidhash_lock released | |
339 | * upon function exit. | |
340 | */ | |
18b6e041 | 341 | static void free_user(struct user_struct *up, unsigned long flags) |
5cb350ba DG |
342 | { |
343 | uid_hash_remove(up); | |
344 | spin_unlock_irqrestore(&uidhash_lock, flags); | |
345 | sched_destroy_user(up); | |
346 | key_put(up->uid_keyring); | |
347 | key_put(up->session_keyring); | |
18b6e041 | 348 | put_user_ns(up->user_ns); |
5cb350ba DG |
349 | kmem_cache_free(uid_cachep, up); |
350 | } | |
351 | ||
b1a8c172 | 352 | #endif |
5cb350ba | 353 | |
1da177e4 LT |
354 | /* |
355 | * Locate the user_struct for the passed UID. If found, take a ref on it. The | |
356 | * caller must undo that ref with free_uid(). | |
357 | * | |
358 | * If the user_struct could not be found, return NULL. | |
359 | */ | |
360 | struct user_struct *find_user(uid_t uid) | |
361 | { | |
362 | struct user_struct *ret; | |
3fa97c9d | 363 | unsigned long flags; |
6ded6ab9 | 364 | struct user_namespace *ns = current_user_ns(); |
1da177e4 | 365 | |
3fa97c9d | 366 | spin_lock_irqsave(&uidhash_lock, flags); |
acce292c | 367 | ret = uid_hash_find(uid, uidhashentry(ns, uid)); |
3fa97c9d | 368 | spin_unlock_irqrestore(&uidhash_lock, flags); |
1da177e4 LT |
369 | return ret; |
370 | } | |
371 | ||
372 | void free_uid(struct user_struct *up) | |
373 | { | |
3fa97c9d AM |
374 | unsigned long flags; |
375 | ||
36f57413 AM |
376 | if (!up) |
377 | return; | |
378 | ||
3fa97c9d | 379 | local_irq_save(flags); |
5cb350ba DG |
380 | if (atomic_dec_and_lock(&up->__count, &uidhash_lock)) |
381 | free_user(up, flags); | |
382 | else | |
36f57413 | 383 | local_irq_restore(flags); |
1da177e4 LT |
384 | } |
385 | ||
354a1f4d | 386 | struct user_struct *alloc_uid(struct user_namespace *ns, uid_t uid) |
1da177e4 | 387 | { |
735de223 | 388 | struct hlist_head *hashent = uidhashentry(ns, uid); |
8eb703e4 | 389 | struct user_struct *up, *new; |
1da177e4 | 390 | |
eb41d946 | 391 | /* Make uid_hash_find() + uids_user_create() + uid_hash_insert() |
5cb350ba DG |
392 | * atomic. |
393 | */ | |
394 | uids_mutex_lock(); | |
395 | ||
3fa97c9d | 396 | spin_lock_irq(&uidhash_lock); |
1da177e4 | 397 | up = uid_hash_find(uid, hashent); |
3fa97c9d | 398 | spin_unlock_irq(&uidhash_lock); |
1da177e4 LT |
399 | |
400 | if (!up) { | |
354a1f4d | 401 | new = kmem_cache_zalloc(uid_cachep, GFP_KERNEL); |
8eb703e4 PE |
402 | if (!new) |
403 | goto out_unlock; | |
5e8869bb | 404 | |
1da177e4 LT |
405 | new->uid = uid; |
406 | atomic_set(&new->__count, 1); | |
1da177e4 | 407 | |
8eb703e4 | 408 | if (sched_create_user(new) < 0) |
69664cf1 | 409 | goto out_free_user; |
24e377a8 | 410 | |
18b6e041 SH |
411 | new->user_ns = get_user_ns(ns); |
412 | ||
8eb703e4 PE |
413 | if (uids_user_create(new)) |
414 | goto out_destoy_sched; | |
5cb350ba | 415 | |
1da177e4 LT |
416 | /* |
417 | * Before adding this, check whether we raced | |
418 | * on adding the same user already.. | |
419 | */ | |
3fa97c9d | 420 | spin_lock_irq(&uidhash_lock); |
1da177e4 LT |
421 | up = uid_hash_find(uid, hashent); |
422 | if (up) { | |
052f1dc7 | 423 | /* This case is not possible when CONFIG_USER_SCHED |
5cb350ba DG |
424 | * is defined, since we serialize alloc_uid() using |
425 | * uids_mutex. Hence no need to call | |
426 | * sched_destroy_user() or remove_user_sysfs_dir(). | |
427 | */ | |
1da177e4 LT |
428 | key_put(new->uid_keyring); |
429 | key_put(new->session_keyring); | |
430 | kmem_cache_free(uid_cachep, new); | |
431 | } else { | |
432 | uid_hash_insert(new, hashent); | |
433 | up = new; | |
434 | } | |
3fa97c9d | 435 | spin_unlock_irq(&uidhash_lock); |
1da177e4 | 436 | } |
5cb350ba DG |
437 | |
438 | uids_mutex_unlock(); | |
439 | ||
1da177e4 | 440 | return up; |
8eb703e4 PE |
441 | |
442 | out_destoy_sched: | |
443 | sched_destroy_user(new); | |
18b6e041 | 444 | put_user_ns(new->user_ns); |
8eb703e4 PE |
445 | out_free_user: |
446 | kmem_cache_free(uid_cachep, new); | |
447 | out_unlock: | |
448 | uids_mutex_unlock(); | |
449 | return NULL; | |
1da177e4 LT |
450 | } |
451 | ||
1da177e4 LT |
452 | static int __init uid_cache_init(void) |
453 | { | |
454 | int n; | |
455 | ||
456 | uid_cachep = kmem_cache_create("uid_cache", sizeof(struct user_struct), | |
20c2df83 | 457 | 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); |
1da177e4 LT |
458 | |
459 | for(n = 0; n < UIDHASH_SZ; ++n) | |
735de223 | 460 | INIT_HLIST_HEAD(init_user_ns.uidhash_table + n); |
1da177e4 LT |
461 | |
462 | /* Insert the root user immediately (init already runs as root) */ | |
3fa97c9d | 463 | spin_lock_irq(&uidhash_lock); |
acce292c | 464 | uid_hash_insert(&root_user, uidhashentry(&init_user_ns, 0)); |
3fa97c9d | 465 | spin_unlock_irq(&uidhash_lock); |
1da177e4 LT |
466 | |
467 | return 0; | |
468 | } | |
469 | ||
470 | module_init(uid_cache_init); |