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 = { | |
23 | .refcount = ATOMIC_INIT(2), | |
24 | }, | |
25 | .root_user = &root_user, | |
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 | ||
51 | struct user_struct root_user = { | |
52 | .__count = ATOMIC_INIT(1), | |
53 | .processes = ATOMIC_INIT(1), | |
54 | .files = ATOMIC_INIT(0), | |
55 | .sigpending = ATOMIC_INIT(0), | |
1da177e4 | 56 | .locked_shm = 0, |
052f1dc7 | 57 | #ifdef CONFIG_USER_SCHED |
4cf86d77 | 58 | .tg = &init_task_group, |
24e377a8 | 59 | #endif |
1da177e4 LT |
60 | }; |
61 | ||
5cb350ba DG |
62 | /* |
63 | * These routines must be called with the uidhash spinlock held! | |
64 | */ | |
40aeb400 | 65 | static void uid_hash_insert(struct user_struct *up, struct hlist_head *hashent) |
5cb350ba DG |
66 | { |
67 | hlist_add_head(&up->uidhash_node, hashent); | |
68 | } | |
69 | ||
40aeb400 | 70 | static void uid_hash_remove(struct user_struct *up) |
5cb350ba DG |
71 | { |
72 | hlist_del_init(&up->uidhash_node); | |
73 | } | |
74 | ||
40aeb400 | 75 | static struct user_struct *uid_hash_find(uid_t uid, struct hlist_head *hashent) |
5cb350ba DG |
76 | { |
77 | struct user_struct *user; | |
78 | struct hlist_node *h; | |
79 | ||
80 | hlist_for_each_entry(user, h, hashent, uidhash_node) { | |
81 | if (user->uid == uid) { | |
82 | atomic_inc(&user->__count); | |
83 | return user; | |
84 | } | |
85 | } | |
86 | ||
87 | return NULL; | |
88 | } | |
89 | ||
052f1dc7 | 90 | #ifdef CONFIG_USER_SCHED |
5cb350ba | 91 | |
24e377a8 SV |
92 | static void sched_destroy_user(struct user_struct *up) |
93 | { | |
94 | sched_destroy_group(up->tg); | |
95 | } | |
96 | ||
97 | static int sched_create_user(struct user_struct *up) | |
98 | { | |
99 | int rc = 0; | |
100 | ||
eff766a6 | 101 | up->tg = sched_create_group(&root_task_group); |
24e377a8 SV |
102 | if (IS_ERR(up->tg)) |
103 | rc = -ENOMEM; | |
104 | ||
105 | return rc; | |
106 | } | |
107 | ||
052f1dc7 | 108 | #else /* CONFIG_USER_SCHED */ |
b1a8c172 DG |
109 | |
110 | static void sched_destroy_user(struct user_struct *up) { } | |
111 | static int sched_create_user(struct user_struct *up) { return 0; } | |
b1a8c172 | 112 | |
052f1dc7 | 113 | #endif /* CONFIG_USER_SCHED */ |
b1a8c172 | 114 | |
052f1dc7 | 115 | #if defined(CONFIG_USER_SCHED) && defined(CONFIG_SYSFS) |
b1a8c172 | 116 | |
eb41d946 | 117 | static struct kset *uids_kset; /* represents the /sys/kernel/uids/ directory */ |
b1a8c172 DG |
118 | static DEFINE_MUTEX(uids_mutex); |
119 | ||
5cb350ba DG |
120 | static inline void uids_mutex_lock(void) |
121 | { | |
122 | mutex_lock(&uids_mutex); | |
123 | } | |
24e377a8 | 124 | |
5cb350ba DG |
125 | static inline void uids_mutex_unlock(void) |
126 | { | |
127 | mutex_unlock(&uids_mutex); | |
128 | } | |
24e377a8 | 129 | |
eb41d946 | 130 | /* uid directory attributes */ |
052f1dc7 | 131 | #ifdef CONFIG_FAIR_GROUP_SCHED |
eb41d946 KS |
132 | static ssize_t cpu_shares_show(struct kobject *kobj, |
133 | struct kobj_attribute *attr, | |
134 | char *buf) | |
5cb350ba | 135 | { |
eb41d946 | 136 | struct user_struct *up = container_of(kobj, struct user_struct, kobj); |
24e377a8 | 137 | |
eb41d946 | 138 | return sprintf(buf, "%lu\n", sched_group_shares(up->tg)); |
5cb350ba DG |
139 | } |
140 | ||
eb41d946 KS |
141 | static ssize_t cpu_shares_store(struct kobject *kobj, |
142 | struct kobj_attribute *attr, | |
143 | const char *buf, size_t size) | |
5cb350ba | 144 | { |
eb41d946 | 145 | struct user_struct *up = container_of(kobj, struct user_struct, kobj); |
5cb350ba DG |
146 | unsigned long shares; |
147 | int rc; | |
148 | ||
eb41d946 | 149 | sscanf(buf, "%lu", &shares); |
5cb350ba DG |
150 | |
151 | rc = sched_group_set_shares(up->tg, shares); | |
152 | ||
153 | return (rc ? rc : size); | |
154 | } | |
155 | ||
eb41d946 KS |
156 | static struct kobj_attribute cpu_share_attr = |
157 | __ATTR(cpu_share, 0644, cpu_shares_show, cpu_shares_store); | |
052f1dc7 | 158 | #endif |
eb41d946 | 159 | |
052f1dc7 | 160 | #ifdef CONFIG_RT_GROUP_SCHED |
9f0c1e56 PZ |
161 | static ssize_t cpu_rt_runtime_show(struct kobject *kobj, |
162 | struct kobj_attribute *attr, | |
163 | char *buf) | |
164 | { | |
165 | struct user_struct *up = container_of(kobj, struct user_struct, kobj); | |
166 | ||
af4491e5 | 167 | return sprintf(buf, "%ld\n", sched_group_rt_runtime(up->tg)); |
9f0c1e56 PZ |
168 | } |
169 | ||
170 | static ssize_t cpu_rt_runtime_store(struct kobject *kobj, | |
171 | struct kobj_attribute *attr, | |
172 | const char *buf, size_t size) | |
173 | { | |
174 | struct user_struct *up = container_of(kobj, struct user_struct, kobj); | |
175 | unsigned long rt_runtime; | |
176 | int rc; | |
177 | ||
af4491e5 | 178 | sscanf(buf, "%ld", &rt_runtime); |
9f0c1e56 PZ |
179 | |
180 | rc = sched_group_set_rt_runtime(up->tg, rt_runtime); | |
181 | ||
182 | return (rc ? rc : size); | |
183 | } | |
184 | ||
185 | static struct kobj_attribute cpu_rt_runtime_attr = | |
186 | __ATTR(cpu_rt_runtime, 0644, cpu_rt_runtime_show, cpu_rt_runtime_store); | |
d0b27fa7 PZ |
187 | |
188 | static ssize_t cpu_rt_period_show(struct kobject *kobj, | |
189 | struct kobj_attribute *attr, | |
190 | char *buf) | |
191 | { | |
192 | struct user_struct *up = container_of(kobj, struct user_struct, kobj); | |
193 | ||
194 | return sprintf(buf, "%lu\n", sched_group_rt_period(up->tg)); | |
195 | } | |
196 | ||
197 | static ssize_t cpu_rt_period_store(struct kobject *kobj, | |
198 | struct kobj_attribute *attr, | |
199 | const char *buf, size_t size) | |
200 | { | |
201 | struct user_struct *up = container_of(kobj, struct user_struct, kobj); | |
202 | unsigned long rt_period; | |
203 | int rc; | |
204 | ||
205 | sscanf(buf, "%lu", &rt_period); | |
206 | ||
207 | rc = sched_group_set_rt_period(up->tg, rt_period); | |
208 | ||
209 | return (rc ? rc : size); | |
210 | } | |
211 | ||
212 | static struct kobj_attribute cpu_rt_period_attr = | |
213 | __ATTR(cpu_rt_period, 0644, cpu_rt_period_show, cpu_rt_period_store); | |
052f1dc7 | 214 | #endif |
9f0c1e56 | 215 | |
eb41d946 KS |
216 | /* default attributes per uid directory */ |
217 | static struct attribute *uids_attributes[] = { | |
052f1dc7 | 218 | #ifdef CONFIG_FAIR_GROUP_SCHED |
eb41d946 | 219 | &cpu_share_attr.attr, |
052f1dc7 PZ |
220 | #endif |
221 | #ifdef CONFIG_RT_GROUP_SCHED | |
9f0c1e56 | 222 | &cpu_rt_runtime_attr.attr, |
d0b27fa7 | 223 | &cpu_rt_period_attr.attr, |
052f1dc7 | 224 | #endif |
eb41d946 KS |
225 | NULL |
226 | }; | |
227 | ||
228 | /* the lifetime of user_struct is not managed by the core (now) */ | |
229 | static void uids_release(struct kobject *kobj) | |
5cb350ba | 230 | { |
eb41d946 | 231 | return; |
5cb350ba DG |
232 | } |
233 | ||
eb41d946 KS |
234 | static struct kobj_type uids_ktype = { |
235 | .sysfs_ops = &kobj_sysfs_ops, | |
236 | .default_attrs = uids_attributes, | |
237 | .release = uids_release, | |
238 | }; | |
239 | ||
240 | /* create /sys/kernel/uids/<uid>/cpu_share file for this user */ | |
241 | static int uids_user_create(struct user_struct *up) | |
1da177e4 | 242 | { |
eb41d946 | 243 | struct kobject *kobj = &up->kobj; |
5cb350ba DG |
244 | int error; |
245 | ||
eb41d946 | 246 | memset(kobj, 0, sizeof(struct kobject)); |
eb41d946 | 247 | kobj->kset = uids_kset; |
cf15126b GKH |
248 | error = kobject_init_and_add(kobj, &uids_ktype, NULL, "%d", up->uid); |
249 | if (error) { | |
250 | kobject_put(kobj); | |
5cb350ba | 251 | goto done; |
cf15126b | 252 | } |
5cb350ba | 253 | |
fb7dde37 | 254 | kobject_uevent(kobj, KOBJ_ADD); |
5cb350ba DG |
255 | done: |
256 | return error; | |
1da177e4 LT |
257 | } |
258 | ||
eb41d946 | 259 | /* create these entries in sysfs: |
5cb350ba DG |
260 | * "/sys/kernel/uids" directory |
261 | * "/sys/kernel/uids/0" directory (for root user) | |
262 | * "/sys/kernel/uids/0/cpu_share" file (for root user) | |
263 | */ | |
eb41d946 | 264 | int __init uids_sysfs_init(void) |
1da177e4 | 265 | { |
0ff21e46 | 266 | uids_kset = kset_create_and_add("uids", NULL, kernel_kobj); |
eb41d946 KS |
267 | if (!uids_kset) |
268 | return -ENOMEM; | |
5cb350ba | 269 | |
eb41d946 | 270 | return uids_user_create(&root_user); |
1da177e4 LT |
271 | } |
272 | ||
5cb350ba DG |
273 | /* work function to remove sysfs directory for a user and free up |
274 | * corresponding structures. | |
275 | */ | |
276 | static void remove_user_sysfs_dir(struct work_struct *w) | |
1da177e4 | 277 | { |
5cb350ba | 278 | struct user_struct *up = container_of(w, struct user_struct, work); |
5cb350ba DG |
279 | unsigned long flags; |
280 | int remove_user = 0; | |
1da177e4 | 281 | |
5cb350ba DG |
282 | /* Make uid_hash_remove() + sysfs_remove_file() + kobject_del() |
283 | * atomic. | |
284 | */ | |
285 | uids_mutex_lock(); | |
286 | ||
287 | local_irq_save(flags); | |
288 | ||
289 | if (atomic_dec_and_lock(&up->__count, &uidhash_lock)) { | |
290 | uid_hash_remove(up); | |
291 | remove_user = 1; | |
292 | spin_unlock_irqrestore(&uidhash_lock, flags); | |
293 | } else { | |
294 | local_irq_restore(flags); | |
1da177e4 LT |
295 | } |
296 | ||
5cb350ba DG |
297 | if (!remove_user) |
298 | goto done; | |
299 | ||
eb41d946 KS |
300 | kobject_uevent(&up->kobj, KOBJ_REMOVE); |
301 | kobject_del(&up->kobj); | |
302 | kobject_put(&up->kobj); | |
5cb350ba DG |
303 | |
304 | sched_destroy_user(up); | |
305 | key_put(up->uid_keyring); | |
306 | key_put(up->session_keyring); | |
307 | kmem_cache_free(uid_cachep, up); | |
308 | ||
309 | done: | |
310 | uids_mutex_unlock(); | |
311 | } | |
312 | ||
313 | /* IRQs are disabled and uidhash_lock is held upon function entry. | |
314 | * IRQ state (as stored in flags) is restored and uidhash_lock released | |
315 | * upon function exit. | |
316 | */ | |
317 | static inline void free_user(struct user_struct *up, unsigned long flags) | |
318 | { | |
319 | /* restore back the count */ | |
320 | atomic_inc(&up->__count); | |
321 | spin_unlock_irqrestore(&uidhash_lock, flags); | |
322 | ||
323 | INIT_WORK(&up->work, remove_user_sysfs_dir); | |
324 | schedule_work(&up->work); | |
1da177e4 LT |
325 | } |
326 | ||
052f1dc7 | 327 | #else /* CONFIG_USER_SCHED && CONFIG_SYSFS */ |
5cb350ba | 328 | |
eb41d946 KS |
329 | int uids_sysfs_init(void) { return 0; } |
330 | static inline int uids_user_create(struct user_struct *up) { return 0; } | |
5cb350ba DG |
331 | static inline void uids_mutex_lock(void) { } |
332 | static inline void uids_mutex_unlock(void) { } | |
333 | ||
334 | /* IRQs are disabled and uidhash_lock is held upon function entry. | |
335 | * IRQ state (as stored in flags) is restored and uidhash_lock released | |
336 | * upon function exit. | |
337 | */ | |
338 | static inline void free_user(struct user_struct *up, unsigned long flags) | |
339 | { | |
340 | uid_hash_remove(up); | |
341 | spin_unlock_irqrestore(&uidhash_lock, flags); | |
342 | sched_destroy_user(up); | |
343 | key_put(up->uid_keyring); | |
344 | key_put(up->session_keyring); | |
345 | kmem_cache_free(uid_cachep, up); | |
346 | } | |
347 | ||
b1a8c172 | 348 | #endif |
5cb350ba | 349 | |
1da177e4 LT |
350 | /* |
351 | * Locate the user_struct for the passed UID. If found, take a ref on it. The | |
352 | * caller must undo that ref with free_uid(). | |
353 | * | |
354 | * If the user_struct could not be found, return NULL. | |
355 | */ | |
356 | struct user_struct *find_user(uid_t uid) | |
357 | { | |
358 | struct user_struct *ret; | |
3fa97c9d | 359 | unsigned long flags; |
acce292c | 360 | struct user_namespace *ns = current->nsproxy->user_ns; |
1da177e4 | 361 | |
3fa97c9d | 362 | spin_lock_irqsave(&uidhash_lock, flags); |
acce292c | 363 | ret = uid_hash_find(uid, uidhashentry(ns, uid)); |
3fa97c9d | 364 | spin_unlock_irqrestore(&uidhash_lock, flags); |
1da177e4 LT |
365 | return ret; |
366 | } | |
367 | ||
368 | void free_uid(struct user_struct *up) | |
369 | { | |
3fa97c9d AM |
370 | unsigned long flags; |
371 | ||
36f57413 AM |
372 | if (!up) |
373 | return; | |
374 | ||
3fa97c9d | 375 | local_irq_save(flags); |
5cb350ba DG |
376 | if (atomic_dec_and_lock(&up->__count, &uidhash_lock)) |
377 | free_user(up, flags); | |
378 | else | |
36f57413 | 379 | local_irq_restore(flags); |
1da177e4 LT |
380 | } |
381 | ||
354a1f4d | 382 | struct user_struct *alloc_uid(struct user_namespace *ns, uid_t uid) |
1da177e4 | 383 | { |
735de223 | 384 | struct hlist_head *hashent = uidhashentry(ns, uid); |
8eb703e4 | 385 | struct user_struct *up, *new; |
1da177e4 | 386 | |
eb41d946 | 387 | /* Make uid_hash_find() + uids_user_create() + uid_hash_insert() |
5cb350ba DG |
388 | * atomic. |
389 | */ | |
390 | uids_mutex_lock(); | |
391 | ||
3fa97c9d | 392 | spin_lock_irq(&uidhash_lock); |
1da177e4 | 393 | up = uid_hash_find(uid, hashent); |
3fa97c9d | 394 | spin_unlock_irq(&uidhash_lock); |
1da177e4 LT |
395 | |
396 | if (!up) { | |
354a1f4d | 397 | new = kmem_cache_zalloc(uid_cachep, GFP_KERNEL); |
8eb703e4 PE |
398 | if (!new) |
399 | goto out_unlock; | |
5e8869bb | 400 | |
1da177e4 LT |
401 | new->uid = uid; |
402 | atomic_set(&new->__count, 1); | |
1da177e4 | 403 | |
8eb703e4 | 404 | if (sched_create_user(new) < 0) |
69664cf1 | 405 | goto out_free_user; |
24e377a8 | 406 | |
8eb703e4 PE |
407 | if (uids_user_create(new)) |
408 | goto out_destoy_sched; | |
5cb350ba | 409 | |
1da177e4 LT |
410 | /* |
411 | * Before adding this, check whether we raced | |
412 | * on adding the same user already.. | |
413 | */ | |
3fa97c9d | 414 | spin_lock_irq(&uidhash_lock); |
1da177e4 LT |
415 | up = uid_hash_find(uid, hashent); |
416 | if (up) { | |
052f1dc7 | 417 | /* This case is not possible when CONFIG_USER_SCHED |
5cb350ba DG |
418 | * is defined, since we serialize alloc_uid() using |
419 | * uids_mutex. Hence no need to call | |
420 | * sched_destroy_user() or remove_user_sysfs_dir(). | |
421 | */ | |
1da177e4 LT |
422 | key_put(new->uid_keyring); |
423 | key_put(new->session_keyring); | |
424 | kmem_cache_free(uid_cachep, new); | |
425 | } else { | |
426 | uid_hash_insert(new, hashent); | |
427 | up = new; | |
428 | } | |
3fa97c9d | 429 | spin_unlock_irq(&uidhash_lock); |
1da177e4 LT |
430 | |
431 | } | |
5cb350ba DG |
432 | |
433 | uids_mutex_unlock(); | |
434 | ||
1da177e4 | 435 | return up; |
8eb703e4 PE |
436 | |
437 | out_destoy_sched: | |
438 | sched_destroy_user(new); | |
8eb703e4 PE |
439 | out_free_user: |
440 | kmem_cache_free(uid_cachep, new); | |
441 | out_unlock: | |
442 | uids_mutex_unlock(); | |
443 | return NULL; | |
1da177e4 LT |
444 | } |
445 | ||
aee16ce7 | 446 | #ifdef CONFIG_USER_NS |
28f300d2 PE |
447 | void release_uids(struct user_namespace *ns) |
448 | { | |
449 | int i; | |
450 | unsigned long flags; | |
451 | struct hlist_head *head; | |
452 | struct hlist_node *nd; | |
453 | ||
454 | spin_lock_irqsave(&uidhash_lock, flags); | |
455 | /* | |
456 | * collapse the chains so that the user_struct-s will | |
457 | * be still alive, but not in hashes. subsequent free_uid() | |
458 | * will free them. | |
459 | */ | |
460 | for (i = 0; i < UIDHASH_SZ; i++) { | |
461 | head = ns->uidhash_table + i; | |
462 | while (!hlist_empty(head)) { | |
463 | nd = head->first; | |
464 | hlist_del_init(nd); | |
465 | } | |
466 | } | |
467 | spin_unlock_irqrestore(&uidhash_lock, flags); | |
468 | ||
469 | free_uid(ns->root_user); | |
470 | } | |
aee16ce7 | 471 | #endif |
1da177e4 LT |
472 | |
473 | static int __init uid_cache_init(void) | |
474 | { | |
475 | int n; | |
476 | ||
477 | uid_cachep = kmem_cache_create("uid_cache", sizeof(struct user_struct), | |
20c2df83 | 478 | 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); |
1da177e4 LT |
479 | |
480 | for(n = 0; n < UIDHASH_SZ; ++n) | |
735de223 | 481 | INIT_HLIST_HEAD(init_user_ns.uidhash_table + n); |
1da177e4 LT |
482 | |
483 | /* Insert the root user immediately (init already runs as root) */ | |
3fa97c9d | 484 | spin_lock_irq(&uidhash_lock); |
acce292c | 485 | uid_hash_insert(&root_user, uidhashentry(&init_user_ns, 0)); |
3fa97c9d | 486 | spin_unlock_irq(&uidhash_lock); |
1da177e4 LT |
487 | |
488 | return 0; | |
489 | } | |
490 | ||
491 | module_init(uid_cache_init); |