kernel/user.c

   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>
  16 #include <linux/interrupt.h>
  17 #include <linux/module.h>
  18 #include <linux/user_namespace.h>
  19
  20 struct user_namespace init_user_ns = {
  21         .kref = {
  22                 .refcount       = ATOMIC_INIT(2),
  23         },
  24         .root_user = &root_user,
  25 };
  26 EXPORT_SYMBOL_GPL(init_user_ns);
  27
  28 /*
  29  * UID task count cache, to get fast user lookup in "alloc_uid"
  30  * when changing user ID's (ie setuid() and friends).
  31  */
  32
  33 #define UIDHASH_MASK            (UIDHASH_SZ - 1)
  34 #define __uidhashfn(uid)        (((uid >> UIDHASH_BITS) + uid) & UIDHASH_MASK)
  35 #define uidhashentry(ns, uid)   ((ns)->uidhash_table + __uidhashfn((uid)))
  36
  37 static struct kmem_cache *uid_cachep;
  38
  39 /*
  40  * The uidhash_lock is mostly taken from process context, but it is
  41  * occasionally also taken from softirq/tasklet context, when
  42  * task-structs get RCU-freed. Hence all locking must be softirq-safe.
  43  * But free_uid() is also called with local interrupts disabled, and running
  44  * local_bh_enable() with local interrupts disabled is an error - we'll run
  45  * softirq callbacks, and they can unconditionally enable interrupts, and
  46  * the caller of free_uid() didn't expect that..
  47  */
  48 static DEFINE_SPINLOCK(uidhash_lock);
  49
  50 struct user_struct root_user = {
  51         .__count        = ATOMIC_INIT(1),
  52         .processes      = ATOMIC_INIT(1),
  53         .files          = ATOMIC_INIT(0),
  54         .sigpending     = ATOMIC_INIT(0),
  55         .locked_shm     = 0,
  56 #ifdef CONFIG_KEYS
  57         .uid_keyring    = &root_user_keyring,
  58         .session_keyring = &root_session_keyring,
  59 #endif
  60 #ifdef CONFIG_FAIR_USER_SCHED
  61         .tg             = &init_task_group,
  62 #endif
  63 };
  64
  65 /*
  66  * These routines must be called with the uidhash spinlock held!
  67  */
  68 static void uid_hash_insert(struct user_struct *up, struct hlist_head *hashent)
  69 {
  70         hlist_add_head(&up->uidhash_node, hashent);
  71 }
  72
  73 static void uid_hash_remove(struct user_struct *up)
  74 {
  75         hlist_del_init(&up->uidhash_node);
  76 }
  77
  78 static struct user_struct *uid_hash_find(uid_t uid, struct hlist_head *hashent)
  79 {
  80         struct user_struct *user;
  81         struct hlist_node *h;
  82
  83         hlist_for_each_entry(user, h, hashent, uidhash_node) {
  84                 if (user->uid == uid) {
  85                         atomic_inc(&user->__count);
  86                         return user;
  87                 }
  88         }
  89
  90         return NULL;
  91 }
  92
  93 #ifdef CONFIG_FAIR_USER_SCHED
  94
  95 static void sched_destroy_user(struct user_struct *up)
  96 {
  97         sched_destroy_group(up->tg);
  98 }
  99
 100 static int sched_create_user(struct user_struct *up)
 101 {
 102         int rc = 0;
 103
 104         up->tg = sched_create_group();
 105         if (IS_ERR(up->tg))
 106                 rc = -ENOMEM;
 107
 108         return rc;
 109 }
 110
 111 static void sched_switch_user(struct task_struct *p)
 112 {
 113         sched_move_task(p);
 114 }
 115
 116 #else   /* CONFIG_FAIR_USER_SCHED */
 117
 118 static void sched_destroy_user(struct user_struct *up) { }
 119 static int sched_create_user(struct user_struct *up) { return 0; }
 120 static void sched_switch_user(struct task_struct *p) { }
 121
 122 #endif  /* CONFIG_FAIR_USER_SCHED */
 123
 124 #if defined(CONFIG_FAIR_USER_SCHED) && defined(CONFIG_SYSFS)
 125
 126 static struct kset *uids_kset; /* represents the /sys/kernel/uids/ directory */
 127 static DEFINE_MUTEX(uids_mutex);
 128
 129 static inline void uids_mutex_lock(void)
 130 {
 131         mutex_lock(&uids_mutex);
 132 }
 133
 134 static inline void uids_mutex_unlock(void)
 135 {
 136         mutex_unlock(&uids_mutex);
 137 }
 138
 139 /* uid directory attributes */
 140 static ssize_t cpu_shares_show(struct kobject *kobj,
 141                                struct kobj_attribute *attr,
 142                                char *buf)
 143 {
 144         struct user_struct *up = container_of(kobj, struct user_struct, kobj);
 145
 146         return sprintf(buf, "%lu\n", sched_group_shares(up->tg));
 147 }
 148
 149 static ssize_t cpu_shares_store(struct kobject *kobj,
 150                                 struct kobj_attribute *attr,
 151                                 const char *buf, size_t size)
 152 {
 153         struct user_struct *up = container_of(kobj, struct user_struct, kobj);
 154         unsigned long shares;
 155         int rc;
 156
 157         sscanf(buf, "%lu", &shares);
 158
 159         rc = sched_group_set_shares(up->tg, shares);
 160
 161         return (rc ? rc : size);
 162 }
 163
 164 static struct kobj_attribute cpu_share_attr =
 165         __ATTR(cpu_share, 0644, cpu_shares_show, cpu_shares_store);
 166
 167 /* default attributes per uid directory */
 168 static struct attribute *uids_attributes[] = {
 169         &cpu_share_attr.attr,
 170         NULL
 171 };
 172
 173 /* the lifetime of user_struct is not managed by the core (now) */
 174 static void uids_release(struct kobject *kobj)
 175 {
 176         return;
 177 }
 178
 179 static struct kobj_type uids_ktype = {
 180         .sysfs_ops = &kobj_sysfs_ops,
 181         .default_attrs = uids_attributes,
 182         .release = uids_release,
 183 };
 184
 185 /* create /sys/kernel/uids/<uid>/cpu_share file for this user */
 186 static int uids_user_create(struct user_struct *up)
 187 {
 188         struct kobject *kobj = &up->kobj;
 189         int error;
 190
 191         memset(kobj, 0, sizeof(struct kobject));
 192         kobj->kset = uids_kset;
 193         error = kobject_init_and_add(kobj, &uids_ktype, NULL, "%d", up->uid);
 194         if (error) {
 195                 kobject_put(kobj);
 196                 goto done;
 197         }
 198
 199         kobject_uevent(kobj, KOBJ_ADD);
 200 done:
 201         return error;
 202 }
 203
 204 /* create these entries in sysfs:
 205  *      "/sys/kernel/uids" directory
 206  *      "/sys/kernel/uids/0" directory (for root user)
 207  *      "/sys/kernel/uids/0/cpu_share" file (for root user)
 208  */
 209 int __init uids_sysfs_init(void)
 210 {
 211         uids_kset = kset_create_and_add("uids", NULL, kernel_kobj);
 212         if (!uids_kset)
 213                 return -ENOMEM;
 214
 215         return uids_user_create(&root_user);
 216 }
 217
 218 /* work function to remove sysfs directory for a user and free up
 219  * corresponding structures.
 220  */
 221 static void remove_user_sysfs_dir(struct work_struct *w)
 222 {
 223         struct user_struct *up = container_of(w, struct user_struct, work);
 224         unsigned long flags;
 225         int remove_user = 0;
 226
 227         /* Make uid_hash_remove() + sysfs_remove_file() + kobject_del()
 228          * atomic.
 229          */
 230         uids_mutex_lock();
 231
 232         local_irq_save(flags);
 233
 234         if (atomic_dec_and_lock(&up->__count, &uidhash_lock)) {
 235                 uid_hash_remove(up);
 236                 remove_user = 1;
 237                 spin_unlock_irqrestore(&uidhash_lock, flags);
 238         } else {
 239                 local_irq_restore(flags);
 240         }
 241
 242         if (!remove_user)
 243                 goto done;
 244
 245         kobject_uevent(&up->kobj, KOBJ_REMOVE);
 246         kobject_del(&up->kobj);
 247         kobject_put(&up->kobj);
 248
 249         sched_destroy_user(up);
 250         key_put(up->uid_keyring);
 251         key_put(up->session_keyring);
 252         kmem_cache_free(uid_cachep, up);
 253
 254 done:
 255         uids_mutex_unlock();
 256 }
 257
 258 /* IRQs are disabled and uidhash_lock is held upon function entry.
 259  * IRQ state (as stored in flags) is restored and uidhash_lock released
 260  * upon function exit.
 261  */
 262 static inline void free_user(struct user_struct *up, unsigned long flags)
 263 {
 264         /* restore back the count */
 265         atomic_inc(&up->__count);
 266         spin_unlock_irqrestore(&uidhash_lock, flags);
 267
 268         INIT_WORK(&up->work, remove_user_sysfs_dir);
 269         schedule_work(&up->work);
 270 }
 271
 272 #else   /* CONFIG_FAIR_USER_SCHED && CONFIG_SYSFS */
 273
 274 int uids_sysfs_init(void) { return 0; }
 275 static inline int uids_user_create(struct user_struct *up) { return 0; }
 276 static inline void uids_mutex_lock(void) { }
 277 static inline void uids_mutex_unlock(void) { }
 278
 279 /* IRQs are disabled and uidhash_lock is held upon function entry.
 280  * IRQ state (as stored in flags) is restored and uidhash_lock released
 281  * upon function exit.
 282  */
 283 static inline void free_user(struct user_struct *up, unsigned long flags)
 284 {
 285         uid_hash_remove(up);
 286         spin_unlock_irqrestore(&uidhash_lock, flags);
 287         sched_destroy_user(up);
 288         key_put(up->uid_keyring);
 289         key_put(up->session_keyring);
 290         kmem_cache_free(uid_cachep, up);
 291 }
 292
 293 #endif
 294
 295 /*
 296  * Locate the user_struct for the passed UID.  If found, take a ref on it.  The
 297  * caller must undo that ref with free_uid().
 298  *
 299  * If the user_struct could not be found, return NULL.
 300  */
 301 struct user_struct *find_user(uid_t uid)
 302 {
 303         struct user_struct *ret;
 304         unsigned long flags;
 305         struct user_namespace *ns = current->nsproxy->user_ns;
 306
 307         spin_lock_irqsave(&uidhash_lock, flags);
 308         ret = uid_hash_find(uid, uidhashentry(ns, uid));
 309         spin_unlock_irqrestore(&uidhash_lock, flags);
 310         return ret;
 311 }
 312
 313 void free_uid(struct user_struct *up)
 314 {
 315         unsigned long flags;
 316
 317         if (!up)
 318                 return;
 319
 320         local_irq_save(flags);
 321         if (atomic_dec_and_lock(&up->__count, &uidhash_lock))
 322                 free_user(up, flags);
 323         else
 324                 local_irq_restore(flags);
 325 }
 326
 327 struct user_struct * alloc_uid(struct user_namespace *ns, uid_t uid)
 328 {
 329         struct hlist_head *hashent = uidhashentry(ns, uid);
 330         struct user_struct *up, *new;
 331
 332         /* Make uid_hash_find() + uids_user_create() + uid_hash_insert()
 333          * atomic.
 334          */
 335         uids_mutex_lock();
 336
 337         spin_lock_irq(&uidhash_lock);
 338         up = uid_hash_find(uid, hashent);
 339         spin_unlock_irq(&uidhash_lock);
 340
 341         if (!up) {
 342                 new = kmem_cache_alloc(uid_cachep, GFP_KERNEL);
 343                 if (!new)
 344                         goto out_unlock;
 345
 346                 new->uid = uid;
 347                 atomic_set(&new->__count, 1);
 348                 atomic_set(&new->processes, 0);
 349                 atomic_set(&new->files, 0);
 350                 atomic_set(&new->sigpending, 0);
 351 #ifdef CONFIG_INOTIFY_USER
 352                 atomic_set(&new->inotify_watches, 0);
 353                 atomic_set(&new->inotify_devs, 0);
 354 #endif
 355 #ifdef CONFIG_POSIX_MQUEUE
 356                 new->mq_bytes = 0;
 357 #endif
 358                 new->locked_shm = 0;
 359
 360                 if (alloc_uid_keyring(new, current) < 0)
 361                         goto out_free_user;
 362
 363                 if (sched_create_user(new) < 0)
 364                         goto out_put_keys;
 365
 366                 if (uids_user_create(new))
 367                         goto out_destoy_sched;
 368
 369                 /*
 370                  * Before adding this, check whether we raced
 371                  * on adding the same user already..
 372                  */
 373                 spin_lock_irq(&uidhash_lock);
 374                 up = uid_hash_find(uid, hashent);
 375                 if (up) {
 376                         /* This case is not possible when CONFIG_FAIR_USER_SCHED
 377                          * is defined, since we serialize alloc_uid() using
 378                          * uids_mutex. Hence no need to call
 379                          * sched_destroy_user() or remove_user_sysfs_dir().
 380                          */
 381                         key_put(new->uid_keyring);
 382                         key_put(new->session_keyring);
 383                         kmem_cache_free(uid_cachep, new);
 384                 } else {
 385                         uid_hash_insert(new, hashent);
 386                         up = new;
 387                 }
 388                 spin_unlock_irq(&uidhash_lock);
 389
 390         }
 391
 392         uids_mutex_unlock();
 393
 394         return up;
 395
 396 out_destoy_sched:
 397         sched_destroy_user(new);
 398 out_put_keys:
 399         key_put(new->uid_keyring);
 400         key_put(new->session_keyring);
 401 out_free_user:
 402         kmem_cache_free(uid_cachep, new);
 403 out_unlock:
 404         uids_mutex_unlock();
 405         return NULL;
 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;
 422         sched_switch_user(current);
 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(&current->sighand->siglock);
 433
 434         free_uid(old_user);
 435         suid_keys(current);
 436 }
 437
 438 #ifdef CONFIG_USER_NS
 439 void release_uids(struct user_namespace *ns)
 440 {
 441         int i;
 442         unsigned long flags;
 443         struct hlist_head *head;
 444         struct hlist_node *nd;
 445
 446         spin_lock_irqsave(&uidhash_lock, flags);
 447         /*
 448          * collapse the chains so that the user_struct-s will
 449          * be still alive, but not in hashes. subsequent free_uid()
 450          * will free them.
 451          */
 452         for (i = 0; i < UIDHASH_SZ; i++) {
 453                 head = ns->uidhash_table + i;
 454                 while (!hlist_empty(head)) {
 455                         nd = head->first;
 456                         hlist_del_init(nd);
 457                 }
 458         }
 459         spin_unlock_irqrestore(&uidhash_lock, flags);
 460
 461         free_uid(ns->root_user);
 462 }
 463 #endif
 464
 465 static int __init uid_cache_init(void)
 466 {
 467         int n;
 468
 469         uid_cachep = kmem_cache_create("uid_cache", sizeof(struct user_struct),
 470                         0, SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
 471
 472         for(n = 0; n < UIDHASH_SZ; ++n)
 473                 INIT_HLIST_HEAD(init_user_ns.uidhash_table + n);
 474
 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);
 479
 480         return 0;
 481 }
 482
 483 module_init(uid_cache_init);