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),
50 .uid_keyring = &root_user_keyring,
51 .session_keyring = &root_session_keyring,
53 #ifdef CONFIG_FAIR_USER_SCHED
54 .tg = &init_task_group,
59 * These routines must be called with the uidhash spinlock held!
61 static inline void uid_hash_insert(struct user_struct *up,
62 struct hlist_head *hashent)
64 hlist_add_head(&up->uidhash_node, hashent);
67 static inline void uid_hash_remove(struct user_struct *up)
69 hlist_del_init(&up->uidhash_node);
72 static inline struct user_struct *uid_hash_find(uid_t uid,
73 struct hlist_head *hashent)
75 struct user_struct *user;
78 hlist_for_each_entry(user, h, hashent, uidhash_node) {
79 if (user->uid == uid) {
80 atomic_inc(&user->__count);
88 #ifdef CONFIG_FAIR_USER_SCHED
90 static struct kobject uids_kobject; /* represents /sys/kernel/uids directory */
91 static DEFINE_MUTEX(uids_mutex);
93 static void sched_destroy_user(struct user_struct *up)
95 sched_destroy_group(up->tg);
98 static int sched_create_user(struct user_struct *up)
102 up->tg = sched_create_group();
109 static void sched_switch_user(struct task_struct *p)
114 static inline void uids_mutex_lock(void)
116 mutex_lock(&uids_mutex);
119 static inline void uids_mutex_unlock(void)
121 mutex_unlock(&uids_mutex);
124 /* return cpu shares held by the user */
125 ssize_t cpu_shares_show(struct kset *kset, char *buffer)
127 struct user_struct *up = container_of(kset, struct user_struct, kset);
129 return sprintf(buffer, "%lu\n", sched_group_shares(up->tg));
132 /* modify cpu shares held by the user */
133 ssize_t cpu_shares_store(struct kset *kset, const char *buffer, size_t size)
135 struct user_struct *up = container_of(kset, struct user_struct, kset);
136 unsigned long shares;
139 sscanf(buffer, "%lu", &shares);
141 rc = sched_group_set_shares(up->tg, shares);
143 return (rc ? rc : size);
146 static void user_attr_init(struct subsys_attribute *sa, char *name, int mode)
148 sa->attr.name = name;
149 sa->attr.mode = mode;
150 sa->show = cpu_shares_show;
151 sa->store = cpu_shares_store;
154 /* Create "/sys/kernel/uids/<uid>" directory and
155 * "/sys/kernel/uids/<uid>/cpu_share" file for this user.
157 static int user_kobject_create(struct user_struct *up)
159 struct kset *kset = &up->kset;
160 struct kobject *kobj = &kset->kobj;
163 memset(kset, 0, sizeof(struct kset));
164 kobj->parent = &uids_kobject; /* create under /sys/kernel/uids dir */
165 kobject_set_name(kobj, "%d", up->uid);
167 user_attr_init(&up->user_attr, "cpu_share", 0644);
169 error = kobject_add(kobj);
173 error = sysfs_create_file(kobj, &up->user_attr.attr);
181 /* create these in sysfs filesystem:
182 * "/sys/kernel/uids" directory
183 * "/sys/kernel/uids/0" directory (for root user)
184 * "/sys/kernel/uids/0/cpu_share" file (for root user)
186 int __init uids_kobject_init(void)
190 /* create under /sys/kernel dir */
191 uids_kobject.parent = &kernel_subsys.kobj;
192 kobject_set_name(&uids_kobject, "uids");
193 kobject_init(&uids_kobject);
195 error = kobject_add(&uids_kobject);
197 error = user_kobject_create(&root_user);
202 /* work function to remove sysfs directory for a user and free up
203 * corresponding structures.
205 static void remove_user_sysfs_dir(struct work_struct *w)
207 struct user_struct *up = container_of(w, struct user_struct, work);
208 struct kobject *kobj = &up->kset.kobj;
212 /* Make uid_hash_remove() + sysfs_remove_file() + kobject_del()
217 local_irq_save(flags);
219 if (atomic_dec_and_lock(&up->__count, &uidhash_lock)) {
222 spin_unlock_irqrestore(&uidhash_lock, flags);
224 local_irq_restore(flags);
230 sysfs_remove_file(kobj, &up->user_attr.attr);
233 sched_destroy_user(up);
234 key_put(up->uid_keyring);
235 key_put(up->session_keyring);
236 kmem_cache_free(uid_cachep, up);
242 /* IRQs are disabled and uidhash_lock is held upon function entry.
243 * IRQ state (as stored in flags) is restored and uidhash_lock released
244 * upon function exit.
246 static inline void free_user(struct user_struct *up, unsigned long flags)
248 /* restore back the count */
249 atomic_inc(&up->__count);
250 spin_unlock_irqrestore(&uidhash_lock, flags);
252 INIT_WORK(&up->work, remove_user_sysfs_dir);
253 schedule_work(&up->work);
256 #else /* CONFIG_FAIR_USER_SCHED */
258 static void sched_destroy_user(struct user_struct *up) { }
259 static int sched_create_user(struct user_struct *up) { return 0; }
260 static void sched_switch_user(struct task_struct *p) { }
261 static inline int user_kobject_create(struct user_struct *up) { return 0; }
262 static inline void uids_mutex_lock(void) { }
263 static inline void uids_mutex_unlock(void) { }
265 /* IRQs are disabled and uidhash_lock is held upon function entry.
266 * IRQ state (as stored in flags) is restored and uidhash_lock released
267 * upon function exit.
269 static inline void free_user(struct user_struct *up, unsigned long flags)
272 spin_unlock_irqrestore(&uidhash_lock, flags);
273 sched_destroy_user(up);
274 key_put(up->uid_keyring);
275 key_put(up->session_keyring);
276 kmem_cache_free(uid_cachep, up);
279 #endif /* CONFIG_FAIR_USER_SCHED */
282 * Locate the user_struct for the passed UID. If found, take a ref on it. The
283 * caller must undo that ref with free_uid().
285 * If the user_struct could not be found, return NULL.
287 struct user_struct *find_user(uid_t uid)
289 struct user_struct *ret;
291 struct user_namespace *ns = current->nsproxy->user_ns;
293 spin_lock_irqsave(&uidhash_lock, flags);
294 ret = uid_hash_find(uid, uidhashentry(ns, uid));
295 spin_unlock_irqrestore(&uidhash_lock, flags);
299 void free_uid(struct user_struct *up)
306 local_irq_save(flags);
307 if (atomic_dec_and_lock(&up->__count, &uidhash_lock))
308 free_user(up, flags);
310 local_irq_restore(flags);
313 struct user_struct * alloc_uid(struct user_namespace *ns, uid_t uid)
315 struct hlist_head *hashent = uidhashentry(ns, uid);
316 struct user_struct *up;
318 /* Make uid_hash_find() + user_kobject_create() + uid_hash_insert()
323 spin_lock_irq(&uidhash_lock);
324 up = uid_hash_find(uid, hashent);
325 spin_unlock_irq(&uidhash_lock);
328 struct user_struct *new;
330 new = kmem_cache_alloc(uid_cachep, GFP_KERNEL);
334 atomic_set(&new->__count, 1);
335 atomic_set(&new->processes, 0);
336 atomic_set(&new->files, 0);
337 atomic_set(&new->sigpending, 0);
338 #ifdef CONFIG_INOTIFY_USER
339 atomic_set(&new->inotify_watches, 0);
340 atomic_set(&new->inotify_devs, 0);
346 if (alloc_uid_keyring(new, current) < 0) {
347 kmem_cache_free(uid_cachep, new);
351 if (sched_create_user(new) < 0) {
352 key_put(new->uid_keyring);
353 key_put(new->session_keyring);
354 kmem_cache_free(uid_cachep, new);
358 if (user_kobject_create(new)) {
359 sched_destroy_user(new);
360 key_put(new->uid_keyring);
361 key_put(new->session_keyring);
362 kmem_cache_free(uid_cachep, new);
368 * Before adding this, check whether we raced
369 * on adding the same user already..
371 spin_lock_irq(&uidhash_lock);
372 up = uid_hash_find(uid, hashent);
374 /* This case is not possible when CONFIG_FAIR_USER_SCHED
375 * is defined, since we serialize alloc_uid() using
376 * uids_mutex. Hence no need to call
377 * sched_destroy_user() or remove_user_sysfs_dir().
379 key_put(new->uid_keyring);
380 key_put(new->session_keyring);
381 kmem_cache_free(uid_cachep, new);
383 uid_hash_insert(new, hashent);
386 spin_unlock_irq(&uidhash_lock);
395 void switch_uid(struct user_struct *new_user)
397 struct user_struct *old_user;
399 /* What if a process setreuid()'s and this brings the
400 * new uid over his NPROC rlimit? We can check this now
401 * cheaply with the new uid cache, so if it matters
402 * we should be checking for it. -DaveM
404 old_user = current->user;
405 atomic_inc(&new_user->processes);
406 atomic_dec(&old_user->processes);
407 switch_uid_keyring(new_user);
408 current->user = new_user;
409 sched_switch_user(current);
412 * We need to synchronize with __sigqueue_alloc()
413 * doing a get_uid(p->user).. If that saw the old
414 * user value, we need to wait until it has exited
415 * its critical region before we can free the old
419 spin_unlock_wait(¤t->sighand->siglock);
425 void release_uids(struct user_namespace *ns)
429 struct hlist_head *head;
430 struct hlist_node *nd;
432 spin_lock_irqsave(&uidhash_lock, flags);
434 * collapse the chains so that the user_struct-s will
435 * be still alive, but not in hashes. subsequent free_uid()
438 for (i = 0; i < UIDHASH_SZ; i++) {
439 head = ns->uidhash_table + i;
440 while (!hlist_empty(head)) {
445 spin_unlock_irqrestore(&uidhash_lock, flags);
447 free_uid(ns->root_user);
450 static int __init uid_cache_init(void)
454 uid_cachep = kmem_cache_create("uid_cache", sizeof(struct user_struct),
455 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
457 for(n = 0; n < UIDHASH_SZ; ++n)
458 INIT_HLIST_HEAD(init_user_ns.uidhash_table + n);
460 /* Insert the root user immediately (init already runs as root) */
461 spin_lock_irq(&uidhash_lock);
462 uid_hash_insert(&root_user, uidhashentry(&init_user_ns, 0));
463 spin_unlock_irq(&uidhash_lock);
468 module_init(uid_cache_init);