4 * (C) Copyright 1991-2000 Linus Torvalds
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.
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>
16 #include <linux/interrupt.h>
17 #include <linux/module.h>
18 #include <linux/user_namespace.h>
21 * UID task count cache, to get fast user lookup in "alloc_uid"
22 * when changing user ID's (ie setuid() and friends).
25 #define UIDHASH_MASK (UIDHASH_SZ - 1)
26 #define __uidhashfn(uid) (((uid >> UIDHASH_BITS) + uid) & UIDHASH_MASK)
27 #define uidhashentry(ns, uid) ((ns)->uidhash_table + __uidhashfn((uid)))
29 static struct kmem_cache
*uid_cachep
;
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.
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..
40 static DEFINE_SPINLOCK(uidhash_lock
);
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),
49 .uid_keyring
= &root_user_keyring
,
50 .session_keyring
= &root_session_keyring
,
52 #ifdef CONFIG_FAIR_USER_SCHED
53 .tg
= &init_task_group
,
58 * These routines must be called with the uidhash spinlock held!
60 static void uid_hash_insert(struct user_struct
*up
, struct hlist_head
*hashent
)
62 hlist_add_head(&up
->uidhash_node
, hashent
);
65 static void uid_hash_remove(struct user_struct
*up
)
67 hlist_del_init(&up
->uidhash_node
);
70 static struct user_struct
*uid_hash_find(uid_t uid
, struct hlist_head
*hashent
)
72 struct user_struct
*user
;
75 hlist_for_each_entry(user
, h
, hashent
, uidhash_node
) {
76 if (user
->uid
== uid
) {
77 atomic_inc(&user
->__count
);
85 #ifdef CONFIG_FAIR_USER_SCHED
87 static void sched_destroy_user(struct user_struct
*up
)
89 sched_destroy_group(up
->tg
);
92 static int sched_create_user(struct user_struct
*up
)
96 up
->tg
= sched_create_group();
103 static void sched_switch_user(struct task_struct
*p
)
108 #else /* CONFIG_FAIR_USER_SCHED */
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
) { }
114 #endif /* CONFIG_FAIR_USER_SCHED */
116 #if defined(CONFIG_FAIR_USER_SCHED) && defined(CONFIG_SYSFS)
118 static struct kset
*uids_kset
; /* represents the /sys/kernel/uids/ directory */
119 static DEFINE_MUTEX(uids_mutex
);
121 static inline void uids_mutex_lock(void)
123 mutex_lock(&uids_mutex
);
126 static inline void uids_mutex_unlock(void)
128 mutex_unlock(&uids_mutex
);
131 /* uid directory attributes */
132 static ssize_t
cpu_shares_show(struct kobject
*kobj
,
133 struct kobj_attribute
*attr
,
136 struct user_struct
*up
= container_of(kobj
, struct user_struct
, kobj
);
138 return sprintf(buf
, "%lu\n", sched_group_shares(up
->tg
));
141 static ssize_t
cpu_shares_store(struct kobject
*kobj
,
142 struct kobj_attribute
*attr
,
143 const char *buf
, size_t size
)
145 struct user_struct
*up
= container_of(kobj
, struct user_struct
, kobj
);
146 unsigned long shares
;
149 sscanf(buf
, "%lu", &shares
);
151 rc
= sched_group_set_shares(up
->tg
, shares
);
153 return (rc
? rc
: size
);
156 static struct kobj_attribute cpu_share_attr
=
157 __ATTR(cpu_share
, 0644, cpu_shares_show
, cpu_shares_store
);
159 /* default attributes per uid directory */
160 static struct attribute
*uids_attributes
[] = {
161 &cpu_share_attr
.attr
,
165 /* the lifetime of user_struct is not managed by the core (now) */
166 static void uids_release(struct kobject
*kobj
)
171 static struct kobj_type uids_ktype
= {
172 .sysfs_ops
= &kobj_sysfs_ops
,
173 .default_attrs
= uids_attributes
,
174 .release
= uids_release
,
177 /* create /sys/kernel/uids/<uid>/cpu_share file for this user */
178 static int uids_user_create(struct user_struct
*up
)
180 struct kobject
*kobj
= &up
->kobj
;
183 memset(kobj
, 0, sizeof(struct kobject
));
184 kobj
->ktype
= &uids_ktype
;
185 kobj
->kset
= uids_kset
;
187 kobject_set_name(&up
->kobj
, "%d", up
->uid
);
188 error
= kobject_add(kobj
);
192 kobject_uevent(kobj
, KOBJ_ADD
);
197 /* create these entries in sysfs:
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)
202 int __init
uids_sysfs_init(void)
204 uids_kset
= kset_create_and_add("uids", NULL
, kernel_kobj
);
208 return uids_user_create(&root_user
);
211 /* work function to remove sysfs directory for a user and free up
212 * corresponding structures.
214 static void remove_user_sysfs_dir(struct work_struct
*w
)
216 struct user_struct
*up
= container_of(w
, struct user_struct
, work
);
220 /* Make uid_hash_remove() + sysfs_remove_file() + kobject_del()
225 local_irq_save(flags
);
227 if (atomic_dec_and_lock(&up
->__count
, &uidhash_lock
)) {
230 spin_unlock_irqrestore(&uidhash_lock
, flags
);
232 local_irq_restore(flags
);
238 kobject_uevent(&up
->kobj
, KOBJ_REMOVE
);
239 kobject_del(&up
->kobj
);
240 kobject_put(&up
->kobj
);
242 sched_destroy_user(up
);
243 key_put(up
->uid_keyring
);
244 key_put(up
->session_keyring
);
245 kmem_cache_free(uid_cachep
, up
);
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.
255 static inline void free_user(struct user_struct
*up
, unsigned long flags
)
257 /* restore back the count */
258 atomic_inc(&up
->__count
);
259 spin_unlock_irqrestore(&uidhash_lock
, flags
);
261 INIT_WORK(&up
->work
, remove_user_sysfs_dir
);
262 schedule_work(&up
->work
);
265 #else /* CONFIG_FAIR_USER_SCHED && CONFIG_SYSFS */
267 int uids_sysfs_init(void) { return 0; }
268 static inline int uids_user_create(struct user_struct
*up
) { return 0; }
269 static inline void uids_mutex_lock(void) { }
270 static inline void uids_mutex_unlock(void) { }
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.
276 static inline void free_user(struct user_struct
*up
, unsigned long flags
)
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
);
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().
292 * If the user_struct could not be found, return NULL.
294 struct user_struct
*find_user(uid_t uid
)
296 struct user_struct
*ret
;
298 struct user_namespace
*ns
= current
->nsproxy
->user_ns
;
300 spin_lock_irqsave(&uidhash_lock
, flags
);
301 ret
= uid_hash_find(uid
, uidhashentry(ns
, uid
));
302 spin_unlock_irqrestore(&uidhash_lock
, flags
);
306 void free_uid(struct user_struct
*up
)
313 local_irq_save(flags
);
314 if (atomic_dec_and_lock(&up
->__count
, &uidhash_lock
))
315 free_user(up
, flags
);
317 local_irq_restore(flags
);
320 struct user_struct
* alloc_uid(struct user_namespace
*ns
, uid_t uid
)
322 struct hlist_head
*hashent
= uidhashentry(ns
, uid
);
323 struct user_struct
*up
;
325 /* Make uid_hash_find() + uids_user_create() + uid_hash_insert()
330 spin_lock_irq(&uidhash_lock
);
331 up
= uid_hash_find(uid
, hashent
);
332 spin_unlock_irq(&uidhash_lock
);
335 struct user_struct
*new;
337 new = kmem_cache_alloc(uid_cachep
, GFP_KERNEL
);
344 atomic_set(&new->__count
, 1);
345 atomic_set(&new->processes
, 0);
346 atomic_set(&new->files
, 0);
347 atomic_set(&new->sigpending
, 0);
348 #ifdef CONFIG_INOTIFY_USER
349 atomic_set(&new->inotify_watches
, 0);
350 atomic_set(&new->inotify_devs
, 0);
352 #ifdef CONFIG_POSIX_MQUEUE
357 if (alloc_uid_keyring(new, current
) < 0) {
358 kmem_cache_free(uid_cachep
, new);
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);
371 if (uids_user_create(new)) {
372 sched_destroy_user(new);
373 key_put(new->uid_keyring
);
374 key_put(new->session_keyring
);
375 kmem_cache_free(uid_cachep
, new);
381 * Before adding this, check whether we raced
382 * on adding the same user already..
384 spin_lock_irq(&uidhash_lock
);
385 up
= uid_hash_find(uid
, hashent
);
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().
392 key_put(new->uid_keyring
);
393 key_put(new->session_keyring
);
394 kmem_cache_free(uid_cachep
, new);
396 uid_hash_insert(new, hashent
);
399 spin_unlock_irq(&uidhash_lock
);
408 void switch_uid(struct user_struct
*new_user
)
410 struct user_struct
*old_user
;
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
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
;
422 sched_switch_user(current
);
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
432 spin_unlock_wait(¤t
->sighand
->siglock
);
438 void release_uids(struct user_namespace
*ns
)
442 struct hlist_head
*head
;
443 struct hlist_node
*nd
;
445 spin_lock_irqsave(&uidhash_lock
, flags
);
447 * collapse the chains so that the user_struct-s will
448 * be still alive, but not in hashes. subsequent free_uid()
451 for (i
= 0; i
< UIDHASH_SZ
; i
++) {
452 head
= ns
->uidhash_table
+ i
;
453 while (!hlist_empty(head
)) {
458 spin_unlock_irqrestore(&uidhash_lock
, flags
);
460 free_uid(ns
->root_user
);
463 static int __init
uid_cache_init(void)
467 uid_cachep
= kmem_cache_create("uid_cache", sizeof(struct user_struct
),
468 0, SLAB_HWCACHE_ALIGN
|SLAB_PANIC
, NULL
);
470 for(n
= 0; n
< UIDHASH_SZ
; ++n
)
471 INIT_HLIST_HEAD(init_user_ns
.uidhash_table
+ n
);
473 /* Insert the root user immediately (init already runs as root) */
474 spin_lock_irq(&uidhash_lock
);
475 uid_hash_insert(&root_user
, uidhashentry(&init_user_ns
, 0));
476 spin_unlock_irq(&uidhash_lock
);
481 module_init(uid_cache_init
);