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 kobject uids_kobject; /* represents /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 /* return cpu shares held by the user */
132 static ssize_t cpu_shares_show(struct kset *kset, char *buffer)
134 struct user_struct *up = container_of(kset, struct user_struct, kset);
136 return sprintf(buffer, "%lu\n", sched_group_shares(up->tg));
139 /* modify cpu shares held by the user */
140 static ssize_t cpu_shares_store(struct kset *kset, const char *buffer,
143 struct user_struct *up = container_of(kset, struct user_struct, kset);
144 unsigned long shares;
147 sscanf(buffer, "%lu", &shares);
149 rc = sched_group_set_shares(up->tg, shares);
151 return (rc ? rc : size);
154 static void user_attr_init(struct subsys_attribute *sa, char *name, int mode)
156 sa->attr.name = name;
157 sa->attr.mode = mode;
158 sa->show = cpu_shares_show;
159 sa->store = cpu_shares_store;
162 /* Create "/sys/kernel/uids/<uid>" directory and
163 * "/sys/kernel/uids/<uid>/cpu_share" file for this user.
165 static int user_kobject_create(struct user_struct *up)
167 struct kset *kset = &up->kset;
168 struct kobject *kobj = &kset->kobj;
171 memset(kset, 0, sizeof(struct kset));
172 kobj->parent = &uids_kobject; /* create under /sys/kernel/uids dir */
173 kobject_set_name(kobj, "%d", up->uid);
175 user_attr_init(&up->user_attr, "cpu_share", 0644);
177 error = kobject_add(kobj);
181 error = sysfs_create_file(kobj, &up->user_attr.attr);
185 kobject_uevent(kobj, KOBJ_ADD);
191 /* create these in sysfs filesystem:
192 * "/sys/kernel/uids" directory
193 * "/sys/kernel/uids/0" directory (for root user)
194 * "/sys/kernel/uids/0/cpu_share" file (for root user)
196 int __init uids_kobject_init(void)
200 /* create under /sys/kernel dir */
201 uids_kobject.parent = &kernel_subsys.kobj;
202 uids_kobject.kset = &kernel_subsys;
203 kobject_set_name(&uids_kobject, "uids");
204 kobject_init(&uids_kobject);
206 error = kobject_add(&uids_kobject);
208 error = user_kobject_create(&root_user);
213 /* work function to remove sysfs directory for a user and free up
214 * corresponding structures.
216 static void remove_user_sysfs_dir(struct work_struct *w)
218 struct user_struct *up = container_of(w, struct user_struct, work);
219 struct kobject *kobj = &up->kset.kobj;
223 /* Make uid_hash_remove() + sysfs_remove_file() + kobject_del()
228 local_irq_save(flags);
230 if (atomic_dec_and_lock(&up->__count, &uidhash_lock)) {
233 spin_unlock_irqrestore(&uidhash_lock, flags);
235 local_irq_restore(flags);
241 sysfs_remove_file(kobj, &up->user_attr.attr);
242 kobject_uevent(kobj, KOBJ_REMOVE);
245 sched_destroy_user(up);
246 key_put(up->uid_keyring);
247 key_put(up->session_keyring);
248 kmem_cache_free(uid_cachep, up);
254 /* IRQs are disabled and uidhash_lock is held upon function entry.
255 * IRQ state (as stored in flags) is restored and uidhash_lock released
256 * upon function exit.
258 static inline void free_user(struct user_struct *up, unsigned long flags)
260 /* restore back the count */
261 atomic_inc(&up->__count);
262 spin_unlock_irqrestore(&uidhash_lock, flags);
264 INIT_WORK(&up->work, remove_user_sysfs_dir);
265 schedule_work(&up->work);
268 #else /* CONFIG_FAIR_USER_SCHED && CONFIG_SYSFS */
270 static inline int user_kobject_create(struct user_struct *up) { return 0; }
271 static inline void uids_mutex_lock(void) { }
272 static inline void uids_mutex_unlock(void) { }
274 /* IRQs are disabled and uidhash_lock is held upon function entry.
275 * IRQ state (as stored in flags) is restored and uidhash_lock released
276 * upon function exit.
278 static inline void free_user(struct user_struct *up, unsigned long flags)
281 spin_unlock_irqrestore(&uidhash_lock, flags);
282 sched_destroy_user(up);
283 key_put(up->uid_keyring);
284 key_put(up->session_keyring);
285 kmem_cache_free(uid_cachep, up);
291 * Locate the user_struct for the passed UID. If found, take a ref on it. The
292 * caller must undo that ref with free_uid().
294 * If the user_struct could not be found, return NULL.
296 struct user_struct *find_user(uid_t uid)
298 struct user_struct *ret;
300 struct user_namespace *ns = current->nsproxy->user_ns;
302 spin_lock_irqsave(&uidhash_lock, flags);
303 ret = uid_hash_find(uid, uidhashentry(ns, uid));
304 spin_unlock_irqrestore(&uidhash_lock, flags);
308 void free_uid(struct user_struct *up)
315 local_irq_save(flags);
316 if (atomic_dec_and_lock(&up->__count, &uidhash_lock))
317 free_user(up, flags);
319 local_irq_restore(flags);
322 struct user_struct * alloc_uid(struct user_namespace *ns, uid_t uid)
324 struct hlist_head *hashent = uidhashentry(ns, uid);
325 struct user_struct *up;
327 /* Make uid_hash_find() + user_kobject_create() + uid_hash_insert()
332 spin_lock_irq(&uidhash_lock);
333 up = uid_hash_find(uid, hashent);
334 spin_unlock_irq(&uidhash_lock);
337 struct user_struct *new;
339 new = kmem_cache_alloc(uid_cachep, GFP_KERNEL);
346 atomic_set(&new->__count, 1);
347 atomic_set(&new->processes, 0);
348 atomic_set(&new->files, 0);
349 atomic_set(&new->sigpending, 0);
350 #ifdef CONFIG_INOTIFY_USER
351 atomic_set(&new->inotify_watches, 0);
352 atomic_set(&new->inotify_devs, 0);
354 #ifdef CONFIG_POSIX_MQUEUE
359 if (alloc_uid_keyring(new, current) < 0) {
360 kmem_cache_free(uid_cachep, new);
365 if (sched_create_user(new) < 0) {
366 key_put(new->uid_keyring);
367 key_put(new->session_keyring);
368 kmem_cache_free(uid_cachep, new);
373 if (user_kobject_create(new)) {
374 sched_destroy_user(new);
375 key_put(new->uid_keyring);
376 key_put(new->session_keyring);
377 kmem_cache_free(uid_cachep, new);
383 * Before adding this, check whether we raced
384 * on adding the same user already..
386 spin_lock_irq(&uidhash_lock);
387 up = uid_hash_find(uid, hashent);
389 /* This case is not possible when CONFIG_FAIR_USER_SCHED
390 * is defined, since we serialize alloc_uid() using
391 * uids_mutex. Hence no need to call
392 * sched_destroy_user() or remove_user_sysfs_dir().
394 key_put(new->uid_keyring);
395 key_put(new->session_keyring);
396 kmem_cache_free(uid_cachep, new);
398 uid_hash_insert(new, hashent);
401 spin_unlock_irq(&uidhash_lock);
410 void switch_uid(struct user_struct *new_user)
412 struct user_struct *old_user;
414 /* What if a process setreuid()'s and this brings the
415 * new uid over his NPROC rlimit? We can check this now
416 * cheaply with the new uid cache, so if it matters
417 * we should be checking for it. -DaveM
419 old_user = current->user;
420 atomic_inc(&new_user->processes);
421 atomic_dec(&old_user->processes);
422 switch_uid_keyring(new_user);
423 current->user = new_user;
424 sched_switch_user(current);
427 * We need to synchronize with __sigqueue_alloc()
428 * doing a get_uid(p->user).. If that saw the old
429 * user value, we need to wait until it has exited
430 * its critical region before we can free the old
434 spin_unlock_wait(¤t->sighand->siglock);
440 void release_uids(struct user_namespace *ns)
444 struct hlist_head *head;
445 struct hlist_node *nd;
447 spin_lock_irqsave(&uidhash_lock, flags);
449 * collapse the chains so that the user_struct-s will
450 * be still alive, but not in hashes. subsequent free_uid()
453 for (i = 0; i < UIDHASH_SZ; i++) {
454 head = ns->uidhash_table + i;
455 while (!hlist_empty(head)) {
460 spin_unlock_irqrestore(&uidhash_lock, flags);
462 free_uid(ns->root_user);
465 static int __init uid_cache_init(void)
469 uid_cachep = kmem_cache_create("uid_cache", sizeof(struct user_struct),
470 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
472 for(n = 0; n < UIDHASH_SZ; ++n)
473 INIT_HLIST_HEAD(init_user_ns.uidhash_table + n);
475 /* Insert the root user immediately (init already runs as root) */
476 spin_lock_irq(&uidhash_lock);
477 uid_hash_insert(&root_user, uidhashentry(&init_user_ns, 0));
478 spin_unlock_irq(&uidhash_lock);
483 module_init(uid_cache_init);