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->kset = uids_kset;
185 error = kobject_init_and_add(kobj, &uids_ktype, NULL, "%d", up->uid);
191 kobject_uevent(kobj, KOBJ_ADD);
196 /* create these entries in sysfs:
197 * "/sys/kernel/uids" directory
198 * "/sys/kernel/uids/0" directory (for root user)
199 * "/sys/kernel/uids/0/cpu_share" file (for root user)
201 int __init uids_sysfs_init(void)
203 uids_kset = kset_create_and_add("uids", NULL, kernel_kobj);
207 return uids_user_create(&root_user);
210 /* work function to remove sysfs directory for a user and free up
211 * corresponding structures.
213 static void remove_user_sysfs_dir(struct work_struct *w)
215 struct user_struct *up = container_of(w, struct user_struct, work);
219 /* Make uid_hash_remove() + sysfs_remove_file() + kobject_del()
224 local_irq_save(flags);
226 if (atomic_dec_and_lock(&up->__count, &uidhash_lock)) {
229 spin_unlock_irqrestore(&uidhash_lock, flags);
231 local_irq_restore(flags);
237 kobject_uevent(&up->kobj, KOBJ_REMOVE);
238 kobject_del(&up->kobj);
239 kobject_put(&up->kobj);
241 sched_destroy_user(up);
242 key_put(up->uid_keyring);
243 key_put(up->session_keyring);
244 kmem_cache_free(uid_cachep, up);
250 /* IRQs are disabled and uidhash_lock is held upon function entry.
251 * IRQ state (as stored in flags) is restored and uidhash_lock released
252 * upon function exit.
254 static inline void free_user(struct user_struct *up, unsigned long flags)
256 /* restore back the count */
257 atomic_inc(&up->__count);
258 spin_unlock_irqrestore(&uidhash_lock, flags);
260 INIT_WORK(&up->work, remove_user_sysfs_dir);
261 schedule_work(&up->work);
264 #else /* CONFIG_FAIR_USER_SCHED && CONFIG_SYSFS */
266 int uids_sysfs_init(void) { return 0; }
267 static inline int uids_user_create(struct user_struct *up) { return 0; }
268 static inline void uids_mutex_lock(void) { }
269 static inline void uids_mutex_unlock(void) { }
271 /* IRQs are disabled and uidhash_lock is held upon function entry.
272 * IRQ state (as stored in flags) is restored and uidhash_lock released
273 * upon function exit.
275 static inline void free_user(struct user_struct *up, unsigned long flags)
278 spin_unlock_irqrestore(&uidhash_lock, flags);
279 sched_destroy_user(up);
280 key_put(up->uid_keyring);
281 key_put(up->session_keyring);
282 kmem_cache_free(uid_cachep, up);
288 * Locate the user_struct for the passed UID. If found, take a ref on it. The
289 * caller must undo that ref with free_uid().
291 * If the user_struct could not be found, return NULL.
293 struct user_struct *find_user(uid_t uid)
295 struct user_struct *ret;
297 struct user_namespace *ns = current->nsproxy->user_ns;
299 spin_lock_irqsave(&uidhash_lock, flags);
300 ret = uid_hash_find(uid, uidhashentry(ns, uid));
301 spin_unlock_irqrestore(&uidhash_lock, flags);
305 void free_uid(struct user_struct *up)
312 local_irq_save(flags);
313 if (atomic_dec_and_lock(&up->__count, &uidhash_lock))
314 free_user(up, flags);
316 local_irq_restore(flags);
319 struct user_struct * alloc_uid(struct user_namespace *ns, uid_t uid)
321 struct hlist_head *hashent = uidhashentry(ns, uid);
322 struct user_struct *up, *new;
324 /* Make uid_hash_find() + uids_user_create() + uid_hash_insert()
329 spin_lock_irq(&uidhash_lock);
330 up = uid_hash_find(uid, hashent);
331 spin_unlock_irq(&uidhash_lock);
334 new = kmem_cache_alloc(uid_cachep, GFP_KERNEL);
339 atomic_set(&new->__count, 1);
340 atomic_set(&new->processes, 0);
341 atomic_set(&new->files, 0);
342 atomic_set(&new->sigpending, 0);
343 #ifdef CONFIG_INOTIFY_USER
344 atomic_set(&new->inotify_watches, 0);
345 atomic_set(&new->inotify_devs, 0);
347 #ifdef CONFIG_POSIX_MQUEUE
352 if (alloc_uid_keyring(new, current) < 0)
355 if (sched_create_user(new) < 0)
358 if (uids_user_create(new))
359 goto out_destoy_sched;
362 * Before adding this, check whether we raced
363 * on adding the same user already..
365 spin_lock_irq(&uidhash_lock);
366 up = uid_hash_find(uid, hashent);
368 /* This case is not possible when CONFIG_FAIR_USER_SCHED
369 * is defined, since we serialize alloc_uid() using
370 * uids_mutex. Hence no need to call
371 * sched_destroy_user() or remove_user_sysfs_dir().
373 key_put(new->uid_keyring);
374 key_put(new->session_keyring);
375 kmem_cache_free(uid_cachep, new);
377 uid_hash_insert(new, hashent);
380 spin_unlock_irq(&uidhash_lock);
389 sched_destroy_user(new);
391 key_put(new->uid_keyring);
392 key_put(new->session_keyring);
394 kmem_cache_free(uid_cachep, new);
400 void switch_uid(struct user_struct *new_user)
402 struct user_struct *old_user;
404 /* What if a process setreuid()'s and this brings the
405 * new uid over his NPROC rlimit? We can check this now
406 * cheaply with the new uid cache, so if it matters
407 * we should be checking for it. -DaveM
409 old_user = current->user;
410 atomic_inc(&new_user->processes);
411 atomic_dec(&old_user->processes);
412 switch_uid_keyring(new_user);
413 current->user = new_user;
414 sched_switch_user(current);
417 * We need to synchronize with __sigqueue_alloc()
418 * doing a get_uid(p->user).. If that saw the old
419 * user value, we need to wait until it has exited
420 * its critical region before we can free the old
424 spin_unlock_wait(¤t->sighand->siglock);
430 void release_uids(struct user_namespace *ns)
434 struct hlist_head *head;
435 struct hlist_node *nd;
437 spin_lock_irqsave(&uidhash_lock, flags);
439 * collapse the chains so that the user_struct-s will
440 * be still alive, but not in hashes. subsequent free_uid()
443 for (i = 0; i < UIDHASH_SZ; i++) {
444 head = ns->uidhash_table + i;
445 while (!hlist_empty(head)) {
450 spin_unlock_irqrestore(&uidhash_lock, flags);
452 free_uid(ns->root_user);
455 static int __init uid_cache_init(void)
459 uid_cachep = kmem_cache_create("uid_cache", sizeof(struct user_struct),
460 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
462 for(n = 0; n < UIDHASH_SZ; ++n)
463 INIT_HLIST_HEAD(init_user_ns.uidhash_table + n);
465 /* Insert the root user immediately (init already runs as root) */
466 spin_lock_irq(&uidhash_lock);
467 uid_hash_insert(&root_user, uidhashentry(&init_user_ns, 0));
468 spin_unlock_irq(&uidhash_lock);
473 module_init(uid_cache_init);