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