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_USER_SCHED
  57         .tg             = &init_task_group,
  58 #endif
  59 };
  60
  61 /*
  62  * These routines must be called with the uidhash spinlock held!
  63  */
  64 static void uid_hash_insert(struct user_struct *up, struct hlist_head *hashent)
  65 {
  66         hlist_add_head(&up->uidhash_node, hashent);
  67 }
  68
  69 static void uid_hash_remove(struct user_struct *up)
  70 {
  71         hlist_del_init(&up->uidhash_node);
  72 }
  73
  74 static struct user_struct *uid_hash_find(uid_t uid, struct hlist_head *hashent)
  75 {
  76         struct user_struct *user;
  77         struct hlist_node *h;
  78
  79         hlist_for_each_entry(user, h, hashent, uidhash_node) {
  80                 if (user->uid == uid) {
  81                         atomic_inc(&user->__count);
  82                         return user;
  83                 }
  84         }
  85
  86         return NULL;
  87 }
  88
  89 #ifdef CONFIG_USER_SCHED
  90
  91 static void sched_destroy_user(struct user_struct *up)
  92 {
  93         sched_destroy_group(up->tg);
  94 }
  95
  96 static int sched_create_user(struct user_struct *up)
  97 {
  98         int rc = 0;
  99
 100         up->tg = sched_create_group(&root_task_group);
 101         if (IS_ERR(up->tg))
 102                 rc = -ENOMEM;
 103
 104         return rc;
 105 }
 106
 107 static void sched_switch_user(struct task_struct *p)
 108 {
 109         sched_move_task(p);
 110 }
 111
 112 #else   /* CONFIG_USER_SCHED */
 113
 114 static void sched_destroy_user(struct user_struct *up) { }
 115 static int sched_create_user(struct user_struct *up) { return 0; }
 116 static void sched_switch_user(struct task_struct *p) { }
 117
 118 #endif  /* CONFIG_USER_SCHED */
 119
 120 #if defined(CONFIG_USER_SCHED) && defined(CONFIG_SYSFS)
 121
 122 static struct kset *uids_kset; /* represents the /sys/kernel/uids/ directory */
 123 static DEFINE_MUTEX(uids_mutex);
 124
 125 static inline void uids_mutex_lock(void)
 126 {
 127         mutex_lock(&uids_mutex);
 128 }
 129
 130 static inline void uids_mutex_unlock(void)
 131 {
 132         mutex_unlock(&uids_mutex);
 133 }
 134
 135 /* uid directory attributes */
 136 #ifdef CONFIG_FAIR_GROUP_SCHED
 137 static ssize_t cpu_shares_show(struct kobject *kobj,
 138                                struct kobj_attribute *attr,
 139                                char *buf)
 140 {
 141         struct user_struct *up = container_of(kobj, struct user_struct, kobj);
 142
 143         return sprintf(buf, "%lu\n", sched_group_shares(up->tg));
 144 }
 145
 146 static ssize_t cpu_shares_store(struct kobject *kobj,
 147                                 struct kobj_attribute *attr,
 148                                 const char *buf, size_t size)
 149 {
 150         struct user_struct *up = container_of(kobj, struct user_struct, kobj);
 151         unsigned long shares;
 152         int rc;
 153
 154         sscanf(buf, "%lu", &shares);
 155
 156         rc = sched_group_set_shares(up->tg, shares);
 157
 158         return (rc ? rc : size);
 159 }
 160
 161 static struct kobj_attribute cpu_share_attr =
 162         __ATTR(cpu_share, 0644, cpu_shares_show, cpu_shares_store);
 163 #endif
 164
 165 #ifdef CONFIG_RT_GROUP_SCHED
 166 static ssize_t cpu_rt_runtime_show(struct kobject *kobj,
 167                                    struct kobj_attribute *attr,
 168                                    char *buf)
 169 {
 170         struct user_struct *up = container_of(kobj, struct user_struct, kobj);
 171
 172         return sprintf(buf, "%lu\n", sched_group_rt_runtime(up->tg));
 173 }
 174
 175 static ssize_t cpu_rt_runtime_store(struct kobject *kobj,
 176                                     struct kobj_attribute *attr,
 177                                     const char *buf, size_t size)
 178 {
 179         struct user_struct *up = container_of(kobj, struct user_struct, kobj);
 180         unsigned long rt_runtime;
 181         int rc;
 182
 183         sscanf(buf, "%lu", &rt_runtime);
 184
 185         rc = sched_group_set_rt_runtime(up->tg, rt_runtime);
 186
 187         return (rc ? rc : size);
 188 }
 189
 190 static struct kobj_attribute cpu_rt_runtime_attr =
 191         __ATTR(cpu_rt_runtime, 0644, cpu_rt_runtime_show, cpu_rt_runtime_store);
 192
 193 static ssize_t cpu_rt_period_show(struct kobject *kobj,
 194                                    struct kobj_attribute *attr,
 195                                    char *buf)
 196 {
 197         struct user_struct *up = container_of(kobj, struct user_struct, kobj);
 198
 199         return sprintf(buf, "%lu\n", sched_group_rt_period(up->tg));
 200 }
 201
 202 static ssize_t cpu_rt_period_store(struct kobject *kobj,
 203                                     struct kobj_attribute *attr,
 204                                     const char *buf, size_t size)
 205 {
 206         struct user_struct *up = container_of(kobj, struct user_struct, kobj);
 207         unsigned long rt_period;
 208         int rc;
 209
 210         sscanf(buf, "%lu", &rt_period);
 211
 212         rc = sched_group_set_rt_period(up->tg, rt_period);
 213
 214         return (rc ? rc : size);
 215 }
 216
 217 static struct kobj_attribute cpu_rt_period_attr =
 218         __ATTR(cpu_rt_period, 0644, cpu_rt_period_show, cpu_rt_period_store);
 219 #endif
 220
 221 /* default attributes per uid directory */
 222 static struct attribute *uids_attributes[] = {
 223 #ifdef CONFIG_FAIR_GROUP_SCHED
 224         &cpu_share_attr.attr,
 225 #endif
 226 #ifdef CONFIG_RT_GROUP_SCHED
 227         &cpu_rt_runtime_attr.attr,
 228         &cpu_rt_period_attr.attr,
 229 #endif
 230         NULL
 231 };
 232
 233 /* the lifetime of user_struct is not managed by the core (now) */
 234 static void uids_release(struct kobject *kobj)
 235 {
 236         return;
 237 }
 238
 239 static struct kobj_type uids_ktype = {
 240         .sysfs_ops = &kobj_sysfs_ops,
 241         .default_attrs = uids_attributes,
 242         .release = uids_release,
 243 };
 244
 245 /* create /sys/kernel/uids/<uid>/cpu_share file for this user */
 246 static int uids_user_create(struct user_struct *up)
 247 {
 248         struct kobject *kobj = &up->kobj;
 249         int error;
 250
 251         memset(kobj, 0, sizeof(struct kobject));
 252         kobj->kset = uids_kset;
 253         error = kobject_init_and_add(kobj, &uids_ktype, NULL, "%d", up->uid);
 254         if (error) {
 255                 kobject_put(kobj);
 256                 goto done;
 257         }
 258
 259         kobject_uevent(kobj, KOBJ_ADD);
 260 done:
 261         return error;
 262 }
 263
 264 /* create these entries in sysfs:
 265  *      "/sys/kernel/uids" directory
 266  *      "/sys/kernel/uids/0" directory (for root user)
 267  *      "/sys/kernel/uids/0/cpu_share" file (for root user)
 268  */
 269 int __init uids_sysfs_init(void)
 270 {
 271         uids_kset = kset_create_and_add("uids", NULL, kernel_kobj);
 272         if (!uids_kset)
 273                 return -ENOMEM;
 274
 275         return uids_user_create(&root_user);
 276 }
 277
 278 /* work function to remove sysfs directory for a user and free up
 279  * corresponding structures.
 280  */
 281 static void remove_user_sysfs_dir(struct work_struct *w)
 282 {
 283         struct user_struct *up = container_of(w, struct user_struct, work);
 284         unsigned long flags;
 285         int remove_user = 0;
 286
 287         /* Make uid_hash_remove() + sysfs_remove_file() + kobject_del()
 288          * atomic.
 289          */
 290         uids_mutex_lock();
 291
 292         local_irq_save(flags);
 293
 294         if (atomic_dec_and_lock(&up->__count, &uidhash_lock)) {
 295                 uid_hash_remove(up);
 296                 remove_user = 1;
 297                 spin_unlock_irqrestore(&uidhash_lock, flags);
 298         } else {
 299                 local_irq_restore(flags);
 300         }
 301
 302         if (!remove_user)
 303                 goto done;
 304
 305         kobject_uevent(&up->kobj, KOBJ_REMOVE);
 306         kobject_del(&up->kobj);
 307         kobject_put(&up->kobj);
 308
 309         sched_destroy_user(up);
 310         key_put(up->uid_keyring);
 311         key_put(up->session_keyring);
 312         kmem_cache_free(uid_cachep, up);
 313
 314 done:
 315         uids_mutex_unlock();
 316 }
 317
 318 /* IRQs are disabled and uidhash_lock is held upon function entry.
 319  * IRQ state (as stored in flags) is restored and uidhash_lock released
 320  * upon function exit.
 321  */
 322 static inline void free_user(struct user_struct *up, unsigned long flags)
 323 {
 324         /* restore back the count */
 325         atomic_inc(&up->__count);
 326         spin_unlock_irqrestore(&uidhash_lock, flags);
 327
 328         INIT_WORK(&up->work, remove_user_sysfs_dir);
 329         schedule_work(&up->work);
 330 }
 331
 332 #else   /* CONFIG_USER_SCHED && CONFIG_SYSFS */
 333
 334 int uids_sysfs_init(void) { return 0; }
 335 static inline int uids_user_create(struct user_struct *up) { return 0; }
 336 static inline void uids_mutex_lock(void) { }
 337 static inline void uids_mutex_unlock(void) { }
 338
 339 /* IRQs are disabled and uidhash_lock is held upon function entry.
 340  * IRQ state (as stored in flags) is restored and uidhash_lock released
 341  * upon function exit.
 342  */
 343 static inline void free_user(struct user_struct *up, unsigned long flags)
 344 {
 345         uid_hash_remove(up);
 346         spin_unlock_irqrestore(&uidhash_lock, flags);
 347         sched_destroy_user(up);
 348         key_put(up->uid_keyring);
 349         key_put(up->session_keyring);
 350         kmem_cache_free(uid_cachep, up);
 351 }
 352
 353 #endif
 354
 355 /*
 356  * Locate the user_struct for the passed UID.  If found, take a ref on it.  The
 357  * caller must undo that ref with free_uid().
 358  *
 359  * If the user_struct could not be found, return NULL.
 360  */
 361 struct user_struct *find_user(uid_t uid)
 362 {
 363         struct user_struct *ret;
 364         unsigned long flags;
 365         struct user_namespace *ns = current->nsproxy->user_ns;
 366
 367         spin_lock_irqsave(&uidhash_lock, flags);
 368         ret = uid_hash_find(uid, uidhashentry(ns, uid));
 369         spin_unlock_irqrestore(&uidhash_lock, flags);
 370         return ret;
 371 }
 372
 373 void free_uid(struct user_struct *up)
 374 {
 375         unsigned long flags;
 376
 377         if (!up)
 378                 return;
 379
 380         local_irq_save(flags);
 381         if (atomic_dec_and_lock(&up->__count, &uidhash_lock))
 382                 free_user(up, flags);
 383         else
 384                 local_irq_restore(flags);
 385 }
 386
 387 struct user_struct * alloc_uid(struct user_namespace *ns, uid_t uid)
 388 {
 389         struct hlist_head *hashent = uidhashentry(ns, uid);
 390         struct user_struct *up, *new;
 391
 392         /* Make uid_hash_find() + uids_user_create() + uid_hash_insert()
 393          * atomic.
 394          */
 395         uids_mutex_lock();
 396
 397         spin_lock_irq(&uidhash_lock);
 398         up = uid_hash_find(uid, hashent);
 399         spin_unlock_irq(&uidhash_lock);
 400
 401         if (!up) {
 402                 new = kmem_cache_alloc(uid_cachep, GFP_KERNEL);
 403                 if (!new)
 404                         goto out_unlock;
 405
 406                 new->uid = uid;
 407                 atomic_set(&new->__count, 1);
 408                 atomic_set(&new->processes, 0);
 409                 atomic_set(&new->files, 0);
 410                 atomic_set(&new->sigpending, 0);
 411 #ifdef CONFIG_INOTIFY_USER
 412                 atomic_set(&new->inotify_watches, 0);
 413                 atomic_set(&new->inotify_devs, 0);
 414 #endif
 415 #ifdef CONFIG_POSIX_MQUEUE
 416                 new->mq_bytes = 0;
 417 #endif
 418                 new->locked_shm = 0;
 419 #ifdef CONFIG_KEYS
 420                 new->uid_keyring = new->session_keyring = NULL;
 421 #endif
 422
 423                 if (sched_create_user(new) < 0)
 424                         goto out_free_user;
 425
 426                 if (uids_user_create(new))
 427                         goto out_destoy_sched;
 428
 429                 /*
 430                  * Before adding this, check whether we raced
 431                  * on adding the same user already..
 432                  */
 433                 spin_lock_irq(&uidhash_lock);
 434                 up = uid_hash_find(uid, hashent);
 435                 if (up) {
 436                         /* This case is not possible when CONFIG_USER_SCHED
 437                          * is defined, since we serialize alloc_uid() using
 438                          * uids_mutex. Hence no need to call
 439                          * sched_destroy_user() or remove_user_sysfs_dir().
 440                          */
 441                         key_put(new->uid_keyring);
 442                         key_put(new->session_keyring);
 443                         kmem_cache_free(uid_cachep, new);
 444                 } else {
 445                         uid_hash_insert(new, hashent);
 446                         up = new;
 447                 }
 448                 spin_unlock_irq(&uidhash_lock);
 449
 450         }
 451
 452         uids_mutex_unlock();
 453
 454         return up;
 455
 456 out_destoy_sched:
 457         sched_destroy_user(new);
 458 out_free_user:
 459         kmem_cache_free(uid_cachep, new);
 460 out_unlock:
 461         uids_mutex_unlock();
 462         return NULL;
 463 }
 464
 465 void switch_uid(struct user_struct *new_user)
 466 {
 467         struct user_struct *old_user;
 468
 469         /* What if a process setreuid()'s and this brings the
 470          * new uid over his NPROC rlimit?  We can check this now
 471          * cheaply with the new uid cache, so if it matters
 472          * we should be checking for it.  -DaveM
 473          */
 474         old_user = current->user;
 475         atomic_inc(&new_user->processes);
 476         atomic_dec(&old_user->processes);
 477         switch_uid_keyring(new_user);
 478         current->user = new_user;
 479         sched_switch_user(current);
 480
 481         /*
 482          * We need to synchronize with __sigqueue_alloc()
 483          * doing a get_uid(p->user).. If that saw the old
 484          * user value, we need to wait until it has exited
 485          * its critical region before we can free the old
 486          * structure.
 487          */
 488         smp_mb();
 489         spin_unlock_wait(&current->sighand->siglock);
 490
 491         free_uid(old_user);
 492         suid_keys(current);
 493 }
 494
 495 #ifdef CONFIG_USER_NS
 496 void release_uids(struct user_namespace *ns)
 497 {
 498         int i;
 499         unsigned long flags;
 500         struct hlist_head *head;
 501         struct hlist_node *nd;
 502
 503         spin_lock_irqsave(&uidhash_lock, flags);
 504         /*
 505          * collapse the chains so that the user_struct-s will
 506          * be still alive, but not in hashes. subsequent free_uid()
 507          * will free them.
 508          */
 509         for (i = 0; i < UIDHASH_SZ; i++) {
 510                 head = ns->uidhash_table + i;
 511                 while (!hlist_empty(head)) {
 512                         nd = head->first;
 513                         hlist_del_init(nd);
 514                 }
 515         }
 516         spin_unlock_irqrestore(&uidhash_lock, flags);
 517
 518         free_uid(ns->root_user);
 519 }
 520 #endif
 521
 522 static int __init uid_cache_init(void)
 523 {
 524         int n;
 525
 526         uid_cachep = kmem_cache_create("uid_cache", sizeof(struct user_struct),
 527                         0, SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
 528
 529         for(n = 0; n < UIDHASH_SZ; ++n)
 530                 INIT_HLIST_HEAD(init_user_ns.uidhash_table + n);
 531
 532         /* Insert the root user immediately (init already runs as root) */
 533         spin_lock_irq(&uidhash_lock);
 534         uid_hash_insert(&root_user, uidhashentry(&init_user_ns, 0));
 535         spin_unlock_irq(&uidhash_lock);
 536
 537         return 0;
 538 }
 539
 540 module_init(uid_cache_init);