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> | |
1da177e4 LT |
19 | |
20 | /* | |
21 | * UID task count cache, to get fast user lookup in "alloc_uid" | |
22 | * when changing user ID's (ie setuid() and friends). | |
23 | */ | |
24 | ||
1da177e4 LT |
25 | #define UIDHASH_MASK (UIDHASH_SZ - 1) |
26 | #define __uidhashfn(uid) (((uid >> UIDHASH_BITS) + uid) & UIDHASH_MASK) | |
acce292c | 27 | #define uidhashentry(ns, uid) ((ns)->uidhash_table + __uidhashfn((uid))) |
1da177e4 | 28 | |
e18b890b | 29 | static struct kmem_cache *uid_cachep; |
4021cb27 IM |
30 | |
31 | /* | |
32 | * The uidhash_lock is mostly taken from process context, but it is | |
33 | * occasionally also taken from softirq/tasklet context, when | |
34 | * task-structs get RCU-freed. Hence all locking must be softirq-safe. | |
3fa97c9d AM |
35 | * But free_uid() is also called with local interrupts disabled, and running |
36 | * local_bh_enable() with local interrupts disabled is an error - we'll run | |
37 | * softirq callbacks, and they can unconditionally enable interrupts, and | |
38 | * the caller of free_uid() didn't expect that.. | |
4021cb27 | 39 | */ |
1da177e4 LT |
40 | static DEFINE_SPINLOCK(uidhash_lock); |
41 | ||
42 | struct user_struct root_user = { | |
43 | .__count = ATOMIC_INIT(1), | |
44 | .processes = ATOMIC_INIT(1), | |
45 | .files = ATOMIC_INIT(0), | |
46 | .sigpending = ATOMIC_INIT(0), | |
1da177e4 LT |
47 | .locked_shm = 0, |
48 | #ifdef CONFIG_KEYS | |
49 | .uid_keyring = &root_user_keyring, | |
50 | .session_keyring = &root_session_keyring, | |
51 | #endif | |
24e377a8 | 52 | #ifdef CONFIG_FAIR_USER_SCHED |
4cf86d77 | 53 | .tg = &init_task_group, |
24e377a8 | 54 | #endif |
1da177e4 LT |
55 | }; |
56 | ||
5cb350ba DG |
57 | /* |
58 | * These routines must be called with the uidhash spinlock held! | |
59 | */ | |
40aeb400 | 60 | static void uid_hash_insert(struct user_struct *up, struct hlist_head *hashent) |
5cb350ba DG |
61 | { |
62 | hlist_add_head(&up->uidhash_node, hashent); | |
63 | } | |
64 | ||
40aeb400 | 65 | static void uid_hash_remove(struct user_struct *up) |
5cb350ba DG |
66 | { |
67 | hlist_del_init(&up->uidhash_node); | |
68 | } | |
69 | ||
40aeb400 | 70 | static struct user_struct *uid_hash_find(uid_t uid, struct hlist_head *hashent) |
5cb350ba DG |
71 | { |
72 | struct user_struct *user; | |
73 | struct hlist_node *h; | |
74 | ||
75 | hlist_for_each_entry(user, h, hashent, uidhash_node) { | |
76 | if (user->uid == uid) { | |
77 | atomic_inc(&user->__count); | |
78 | return user; | |
79 | } | |
80 | } | |
81 | ||
82 | return NULL; | |
83 | } | |
84 | ||
24e377a8 | 85 | #ifdef CONFIG_FAIR_USER_SCHED |
5cb350ba | 86 | |
24e377a8 SV |
87 | static void sched_destroy_user(struct user_struct *up) |
88 | { | |
89 | sched_destroy_group(up->tg); | |
90 | } | |
91 | ||
92 | static int sched_create_user(struct user_struct *up) | |
93 | { | |
94 | int rc = 0; | |
95 | ||
96 | up->tg = sched_create_group(); | |
97 | if (IS_ERR(up->tg)) | |
98 | rc = -ENOMEM; | |
99 | ||
100 | return rc; | |
101 | } | |
102 | ||
103 | static void sched_switch_user(struct task_struct *p) | |
104 | { | |
105 | sched_move_task(p); | |
106 | } | |
107 | ||
b1a8c172 DG |
108 | #else /* CONFIG_FAIR_USER_SCHED */ |
109 | ||
110 | static void sched_destroy_user(struct user_struct *up) { } | |
111 | static int sched_create_user(struct user_struct *up) { return 0; } | |
112 | static void sched_switch_user(struct task_struct *p) { } | |
113 | ||
114 | #endif /* CONFIG_FAIR_USER_SCHED */ | |
115 | ||
116 | #if defined(CONFIG_FAIR_USER_SCHED) && defined(CONFIG_SYSFS) | |
117 | ||
eb41d946 | 118 | static struct kset *uids_kset; /* represents the /sys/kernel/uids/ directory */ |
b1a8c172 DG |
119 | static DEFINE_MUTEX(uids_mutex); |
120 | ||
5cb350ba DG |
121 | static inline void uids_mutex_lock(void) |
122 | { | |
123 | mutex_lock(&uids_mutex); | |
124 | } | |
24e377a8 | 125 | |
5cb350ba DG |
126 | static inline void uids_mutex_unlock(void) |
127 | { | |
128 | mutex_unlock(&uids_mutex); | |
129 | } | |
24e377a8 | 130 | |
eb41d946 KS |
131 | /* uid directory attributes */ |
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); | |
158 | ||
159 | /* default attributes per uid directory */ | |
160 | static struct attribute *uids_attributes[] = { | |
161 | &cpu_share_attr.attr, | |
162 | NULL | |
163 | }; | |
164 | ||
165 | /* the lifetime of user_struct is not managed by the core (now) */ | |
166 | static void uids_release(struct kobject *kobj) | |
5cb350ba | 167 | { |
eb41d946 | 168 | return; |
5cb350ba DG |
169 | } |
170 | ||
eb41d946 KS |
171 | static struct kobj_type uids_ktype = { |
172 | .sysfs_ops = &kobj_sysfs_ops, | |
173 | .default_attrs = uids_attributes, | |
174 | .release = uids_release, | |
175 | }; | |
176 | ||
177 | /* create /sys/kernel/uids/<uid>/cpu_share file for this user */ | |
178 | static int uids_user_create(struct user_struct *up) | |
1da177e4 | 179 | { |
eb41d946 | 180 | struct kobject *kobj = &up->kobj; |
5cb350ba DG |
181 | int error; |
182 | ||
eb41d946 KS |
183 | memset(kobj, 0, sizeof(struct kobject)); |
184 | kobj->ktype = &uids_ktype; | |
185 | kobj->kset = uids_kset; | |
186 | kobject_init(kobj); | |
187 | kobject_set_name(&up->kobj, "%d", up->uid); | |
5cb350ba DG |
188 | error = kobject_add(kobj); |
189 | if (error) | |
190 | goto done; | |
191 | ||
fb7dde37 | 192 | kobject_uevent(kobj, KOBJ_ADD); |
5cb350ba DG |
193 | done: |
194 | return error; | |
1da177e4 LT |
195 | } |
196 | ||
eb41d946 | 197 | /* create these entries in sysfs: |
5cb350ba DG |
198 | * "/sys/kernel/uids" directory |
199 | * "/sys/kernel/uids/0" directory (for root user) | |
200 | * "/sys/kernel/uids/0/cpu_share" file (for root user) | |
201 | */ | |
eb41d946 | 202 | int __init uids_sysfs_init(void) |
1da177e4 | 203 | { |
eb41d946 KS |
204 | uids_kset = kset_create_and_add("uids", NULL, &kernel_kset->kobj); |
205 | if (!uids_kset) | |
206 | return -ENOMEM; | |
5cb350ba | 207 | |
eb41d946 | 208 | return uids_user_create(&root_user); |
1da177e4 LT |
209 | } |
210 | ||
5cb350ba DG |
211 | /* work function to remove sysfs directory for a user and free up |
212 | * corresponding structures. | |
213 | */ | |
214 | static void remove_user_sysfs_dir(struct work_struct *w) | |
1da177e4 | 215 | { |
5cb350ba | 216 | struct user_struct *up = container_of(w, struct user_struct, work); |
5cb350ba DG |
217 | unsigned long flags; |
218 | int remove_user = 0; | |
1da177e4 | 219 | |
5cb350ba DG |
220 | /* Make uid_hash_remove() + sysfs_remove_file() + kobject_del() |
221 | * atomic. | |
222 | */ | |
223 | uids_mutex_lock(); | |
224 | ||
225 | local_irq_save(flags); | |
226 | ||
227 | if (atomic_dec_and_lock(&up->__count, &uidhash_lock)) { | |
228 | uid_hash_remove(up); | |
229 | remove_user = 1; | |
230 | spin_unlock_irqrestore(&uidhash_lock, flags); | |
231 | } else { | |
232 | local_irq_restore(flags); | |
1da177e4 LT |
233 | } |
234 | ||
5cb350ba DG |
235 | if (!remove_user) |
236 | goto done; | |
237 | ||
eb41d946 KS |
238 | kobject_uevent(&up->kobj, KOBJ_REMOVE); |
239 | kobject_del(&up->kobj); | |
240 | kobject_put(&up->kobj); | |
5cb350ba DG |
241 | |
242 | sched_destroy_user(up); | |
243 | key_put(up->uid_keyring); | |
244 | key_put(up->session_keyring); | |
245 | kmem_cache_free(uid_cachep, up); | |
246 | ||
247 | done: | |
248 | uids_mutex_unlock(); | |
249 | } | |
250 | ||
251 | /* IRQs are disabled and uidhash_lock is held upon function entry. | |
252 | * IRQ state (as stored in flags) is restored and uidhash_lock released | |
253 | * upon function exit. | |
254 | */ | |
255 | static inline void free_user(struct user_struct *up, unsigned long flags) | |
256 | { | |
257 | /* restore back the count */ | |
258 | atomic_inc(&up->__count); | |
259 | spin_unlock_irqrestore(&uidhash_lock, flags); | |
260 | ||
261 | INIT_WORK(&up->work, remove_user_sysfs_dir); | |
262 | schedule_work(&up->work); | |
1da177e4 LT |
263 | } |
264 | ||
b1a8c172 | 265 | #else /* CONFIG_FAIR_USER_SCHED && CONFIG_SYSFS */ |
5cb350ba | 266 | |
eb41d946 KS |
267 | int uids_sysfs_init(void) { return 0; } |
268 | static inline int uids_user_create(struct user_struct *up) { return 0; } | |
5cb350ba DG |
269 | static inline void uids_mutex_lock(void) { } |
270 | static inline void uids_mutex_unlock(void) { } | |
271 | ||
272 | /* IRQs are disabled and uidhash_lock is held upon function entry. | |
273 | * IRQ state (as stored in flags) is restored and uidhash_lock released | |
274 | * upon function exit. | |
275 | */ | |
276 | static inline void free_user(struct user_struct *up, unsigned long flags) | |
277 | { | |
278 | uid_hash_remove(up); | |
279 | spin_unlock_irqrestore(&uidhash_lock, flags); | |
280 | sched_destroy_user(up); | |
281 | key_put(up->uid_keyring); | |
282 | key_put(up->session_keyring); | |
283 | kmem_cache_free(uid_cachep, up); | |
284 | } | |
285 | ||
b1a8c172 | 286 | #endif |
5cb350ba | 287 | |
1da177e4 LT |
288 | /* |
289 | * Locate the user_struct for the passed UID. If found, take a ref on it. The | |
290 | * caller must undo that ref with free_uid(). | |
291 | * | |
292 | * If the user_struct could not be found, return NULL. | |
293 | */ | |
294 | struct user_struct *find_user(uid_t uid) | |
295 | { | |
296 | struct user_struct *ret; | |
3fa97c9d | 297 | unsigned long flags; |
acce292c | 298 | struct user_namespace *ns = current->nsproxy->user_ns; |
1da177e4 | 299 | |
3fa97c9d | 300 | spin_lock_irqsave(&uidhash_lock, flags); |
acce292c | 301 | ret = uid_hash_find(uid, uidhashentry(ns, uid)); |
3fa97c9d | 302 | spin_unlock_irqrestore(&uidhash_lock, flags); |
1da177e4 LT |
303 | return ret; |
304 | } | |
305 | ||
306 | void free_uid(struct user_struct *up) | |
307 | { | |
3fa97c9d AM |
308 | unsigned long flags; |
309 | ||
36f57413 AM |
310 | if (!up) |
311 | return; | |
312 | ||
3fa97c9d | 313 | local_irq_save(flags); |
5cb350ba DG |
314 | if (atomic_dec_and_lock(&up->__count, &uidhash_lock)) |
315 | free_user(up, flags); | |
316 | else | |
36f57413 | 317 | local_irq_restore(flags); |
1da177e4 LT |
318 | } |
319 | ||
acce292c | 320 | struct user_struct * alloc_uid(struct user_namespace *ns, uid_t uid) |
1da177e4 | 321 | { |
735de223 | 322 | struct hlist_head *hashent = uidhashentry(ns, uid); |
1da177e4 LT |
323 | struct user_struct *up; |
324 | ||
eb41d946 | 325 | /* Make uid_hash_find() + uids_user_create() + uid_hash_insert() |
5cb350ba DG |
326 | * atomic. |
327 | */ | |
328 | uids_mutex_lock(); | |
329 | ||
3fa97c9d | 330 | spin_lock_irq(&uidhash_lock); |
1da177e4 | 331 | up = uid_hash_find(uid, hashent); |
3fa97c9d | 332 | spin_unlock_irq(&uidhash_lock); |
1da177e4 LT |
333 | |
334 | if (!up) { | |
335 | struct user_struct *new; | |
336 | ||
e94b1766 | 337 | new = kmem_cache_alloc(uid_cachep, GFP_KERNEL); |
5e8869bb PE |
338 | if (!new) { |
339 | uids_mutex_unlock(); | |
1da177e4 | 340 | return NULL; |
5e8869bb PE |
341 | } |
342 | ||
1da177e4 LT |
343 | new->uid = uid; |
344 | atomic_set(&new->__count, 1); | |
345 | atomic_set(&new->processes, 0); | |
346 | atomic_set(&new->files, 0); | |
347 | atomic_set(&new->sigpending, 0); | |
2d9048e2 | 348 | #ifdef CONFIG_INOTIFY_USER |
0eeca283 RL |
349 | atomic_set(&new->inotify_watches, 0); |
350 | atomic_set(&new->inotify_devs, 0); | |
351 | #endif | |
970a8645 | 352 | #ifdef CONFIG_POSIX_MQUEUE |
1da177e4 | 353 | new->mq_bytes = 0; |
970a8645 | 354 | #endif |
1da177e4 LT |
355 | new->locked_shm = 0; |
356 | ||
d720024e | 357 | if (alloc_uid_keyring(new, current) < 0) { |
1da177e4 | 358 | kmem_cache_free(uid_cachep, new); |
5e8869bb | 359 | uids_mutex_unlock(); |
1da177e4 LT |
360 | return NULL; |
361 | } | |
362 | ||
24e377a8 SV |
363 | if (sched_create_user(new) < 0) { |
364 | key_put(new->uid_keyring); | |
365 | key_put(new->session_keyring); | |
366 | kmem_cache_free(uid_cachep, new); | |
5e8869bb | 367 | uids_mutex_unlock(); |
24e377a8 SV |
368 | return NULL; |
369 | } | |
370 | ||
eb41d946 | 371 | if (uids_user_create(new)) { |
5cb350ba DG |
372 | sched_destroy_user(new); |
373 | key_put(new->uid_keyring); | |
374 | key_put(new->session_keyring); | |
375 | kmem_cache_free(uid_cachep, new); | |
376 | uids_mutex_unlock(); | |
377 | return NULL; | |
378 | } | |
379 | ||
1da177e4 LT |
380 | /* |
381 | * Before adding this, check whether we raced | |
382 | * on adding the same user already.. | |
383 | */ | |
3fa97c9d | 384 | spin_lock_irq(&uidhash_lock); |
1da177e4 LT |
385 | up = uid_hash_find(uid, hashent); |
386 | if (up) { | |
5cb350ba DG |
387 | /* This case is not possible when CONFIG_FAIR_USER_SCHED |
388 | * is defined, since we serialize alloc_uid() using | |
389 | * uids_mutex. Hence no need to call | |
390 | * sched_destroy_user() or remove_user_sysfs_dir(). | |
391 | */ | |
1da177e4 LT |
392 | key_put(new->uid_keyring); |
393 | key_put(new->session_keyring); | |
394 | kmem_cache_free(uid_cachep, new); | |
395 | } else { | |
396 | uid_hash_insert(new, hashent); | |
397 | up = new; | |
398 | } | |
3fa97c9d | 399 | spin_unlock_irq(&uidhash_lock); |
1da177e4 LT |
400 | |
401 | } | |
5cb350ba DG |
402 | |
403 | uids_mutex_unlock(); | |
404 | ||
1da177e4 LT |
405 | return up; |
406 | } | |
407 | ||
408 | void switch_uid(struct user_struct *new_user) | |
409 | { | |
410 | struct user_struct *old_user; | |
411 | ||
412 | /* What if a process setreuid()'s and this brings the | |
413 | * new uid over his NPROC rlimit? We can check this now | |
414 | * cheaply with the new uid cache, so if it matters | |
415 | * we should be checking for it. -DaveM | |
416 | */ | |
417 | old_user = current->user; | |
418 | atomic_inc(&new_user->processes); | |
419 | atomic_dec(&old_user->processes); | |
420 | switch_uid_keyring(new_user); | |
421 | current->user = new_user; | |
24e377a8 | 422 | sched_switch_user(current); |
45c18b0b LT |
423 | |
424 | /* | |
425 | * We need to synchronize with __sigqueue_alloc() | |
426 | * doing a get_uid(p->user).. If that saw the old | |
427 | * user value, we need to wait until it has exited | |
428 | * its critical region before we can free the old | |
429 | * structure. | |
430 | */ | |
431 | smp_mb(); | |
432 | spin_unlock_wait(¤t->sighand->siglock); | |
433 | ||
1da177e4 LT |
434 | free_uid(old_user); |
435 | suid_keys(current); | |
436 | } | |
437 | ||
28f300d2 PE |
438 | void release_uids(struct user_namespace *ns) |
439 | { | |
440 | int i; | |
441 | unsigned long flags; | |
442 | struct hlist_head *head; | |
443 | struct hlist_node *nd; | |
444 | ||
445 | spin_lock_irqsave(&uidhash_lock, flags); | |
446 | /* | |
447 | * collapse the chains so that the user_struct-s will | |
448 | * be still alive, but not in hashes. subsequent free_uid() | |
449 | * will free them. | |
450 | */ | |
451 | for (i = 0; i < UIDHASH_SZ; i++) { | |
452 | head = ns->uidhash_table + i; | |
453 | while (!hlist_empty(head)) { | |
454 | nd = head->first; | |
455 | hlist_del_init(nd); | |
456 | } | |
457 | } | |
458 | spin_unlock_irqrestore(&uidhash_lock, flags); | |
459 | ||
460 | free_uid(ns->root_user); | |
461 | } | |
1da177e4 LT |
462 | |
463 | static int __init uid_cache_init(void) | |
464 | { | |
465 | int n; | |
466 | ||
467 | uid_cachep = kmem_cache_create("uid_cache", sizeof(struct user_struct), | |
20c2df83 | 468 | 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); |
1da177e4 LT |
469 | |
470 | for(n = 0; n < UIDHASH_SZ; ++n) | |
735de223 | 471 | INIT_HLIST_HEAD(init_user_ns.uidhash_table + n); |
1da177e4 LT |
472 | |
473 | /* Insert the root user immediately (init already runs as root) */ | |
3fa97c9d | 474 | spin_lock_irq(&uidhash_lock); |
acce292c | 475 | uid_hash_insert(&root_user, uidhashentry(&init_user_ns, 0)); |
3fa97c9d | 476 | spin_unlock_irq(&uidhash_lock); |
1da177e4 LT |
477 | |
478 | return 0; | |
479 | } | |
480 | ||
481 | module_init(uid_cache_init); |