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_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_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(&root_task_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_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_USER_SCHED */
 123
 124 #if defined(CONFIG_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 #ifdef CONFIG_FAIR_GROUP_SCHED
 141 static ssize_t cpu_shares_show(struct kobject *kobj,
 142                                struct kobj_attribute *attr,
 143                                char *buf)
 144 {
 145         struct user_struct *up = container_of(kobj, struct user_struct, kobj);
 146
 147         return sprintf(buf, "%lu\n", sched_group_shares(up->tg));
 148 }
 149
 150 static ssize_t cpu_shares_store(struct kobject *kobj,
 151                                 struct kobj_attribute *attr,
 152                                 const char *buf, size_t size)
 153 {
 154         struct user_struct *up = container_of(kobj, struct user_struct, kobj);
 155         unsigned long shares;
 156         int rc;
 157
 158         sscanf(buf, "%lu", &shares);
 159
 160         rc = sched_group_set_shares(up->tg, shares);
 161
 162         return (rc ? rc : size);
 163 }
 164
 165 static struct kobj_attribute cpu_share_attr =
 166         __ATTR(cpu_share, 0644, cpu_shares_show, cpu_shares_store);
 167 #endif
 168
 169 #ifdef CONFIG_RT_GROUP_SCHED
 170 static ssize_t cpu_rt_runtime_show(struct kobject *kobj,
 171                                    struct kobj_attribute *attr,
 172                                    char *buf)
 173 {
 174         struct user_struct *up = container_of(kobj, struct user_struct, kobj);
 175
 176         return sprintf(buf, "%lu\n", sched_group_rt_runtime(up->tg));
 177 }
 178
 179 static ssize_t cpu_rt_runtime_store(struct kobject *kobj,
 180                                     struct kobj_attribute *attr,
 181                                     const char *buf, size_t size)
 182 {
 183         struct user_struct *up = container_of(kobj, struct user_struct, kobj);
 184         unsigned long rt_runtime;
 185         int rc;
 186
 187         sscanf(buf, "%lu", &rt_runtime);
 188
 189         rc = sched_group_set_rt_runtime(up->tg, rt_runtime);
 190
 191         return (rc ? rc : size);
 192 }
 193
 194 static struct kobj_attribute cpu_rt_runtime_attr =
 195         __ATTR(cpu_rt_runtime, 0644, cpu_rt_runtime_show, cpu_rt_runtime_store);
 196
 197 static ssize_t cpu_rt_period_show(struct kobject *kobj,
 198                                    struct kobj_attribute *attr,
 199                                    char *buf)
 200 {
 201         struct user_struct *up = container_of(kobj, struct user_struct, kobj);
 202
 203         return sprintf(buf, "%lu\n", sched_group_rt_period(up->tg));
 204 }
 205
 206 static ssize_t cpu_rt_period_store(struct kobject *kobj,
 207                                     struct kobj_attribute *attr,
 208                                     const char *buf, size_t size)
 209 {
 210         struct user_struct *up = container_of(kobj, struct user_struct, kobj);
 211         unsigned long rt_period;
 212         int rc;
 213
 214         sscanf(buf, "%lu", &rt_period);
 215
 216         rc = sched_group_set_rt_period(up->tg, rt_period);
 217
 218         return (rc ? rc : size);
 219 }
 220
 221 static struct kobj_attribute cpu_rt_period_attr =
 222         __ATTR(cpu_rt_period, 0644, cpu_rt_period_show, cpu_rt_period_store);
 223 #endif
 224
 225 /* default attributes per uid directory */
 226 static struct attribute *uids_attributes[] = {
 227 #ifdef CONFIG_FAIR_GROUP_SCHED
 228         &cpu_share_attr.attr,
 229 #endif
 230 #ifdef CONFIG_RT_GROUP_SCHED
 231         &cpu_rt_runtime_attr.attr,
 232         &cpu_rt_period_attr.attr,
 233 #endif
 234         NULL
 235 };
 236
 237 /* the lifetime of user_struct is not managed by the core (now) */
 238 static void uids_release(struct kobject *kobj)
 239 {
 240         return;
 241 }
 242
 243 static struct kobj_type uids_ktype = {
 244         .sysfs_ops = &kobj_sysfs_ops,
 245         .default_attrs = uids_attributes,
 246         .release = uids_release,
 247 };
 248
 249 /* create /sys/kernel/uids/<uid>/cpu_share file for this user */
 250 static int uids_user_create(struct user_struct *up)
 251 {
 252         struct kobject *kobj = &up->kobj;
 253         int error;
 254
 255         memset(kobj, 0, sizeof(struct kobject));
 256         kobj->kset = uids_kset;
 257         error = kobject_init_and_add(kobj, &uids_ktype, NULL, "%d", up->uid);
 258         if (error) {
 259                 kobject_put(kobj);
 260                 goto done;
 261         }
 262
 263         kobject_uevent(kobj, KOBJ_ADD);
 264 done:
 265         return error;
 266 }
 267
 268 /* create these entries in sysfs:
 269  *      "/sys/kernel/uids" directory
 270  *      "/sys/kernel/uids/0" directory (for root user)
 271  *      "/sys/kernel/uids/0/cpu_share" file (for root user)
 272  */
 273 int __init uids_sysfs_init(void)
 274 {
 275         uids_kset = kset_create_and_add("uids", NULL, kernel_kobj);
 276         if (!uids_kset)
 277                 return -ENOMEM;
 278
 279         return uids_user_create(&root_user);
 280 }
 281
 282 /* work function to remove sysfs directory for a user and free up
 283  * corresponding structures.
 284  */
 285 static void remove_user_sysfs_dir(struct work_struct *w)
 286 {
 287         struct user_struct *up = container_of(w, struct user_struct, work);
 288         unsigned long flags;
 289         int remove_user = 0;
 290
 291         /* Make uid_hash_remove() + sysfs_remove_file() + kobject_del()
 292          * atomic.
 293          */
 294         uids_mutex_lock();
 295
 296         local_irq_save(flags);
 297
 298         if (atomic_dec_and_lock(&up->__count, &uidhash_lock)) {
 299                 uid_hash_remove(up);
 300                 remove_user = 1;
 301                 spin_unlock_irqrestore(&uidhash_lock, flags);
 302         } else {
 303                 local_irq_restore(flags);
 304         }
 305
 306         if (!remove_user)
 307                 goto done;
 308
 309         kobject_uevent(&up->kobj, KOBJ_REMOVE);
 310         kobject_del(&up->kobj);
 311         kobject_put(&up->kobj);
 312
 313         sched_destroy_user(up);
 314         key_put(up->uid_keyring);
 315         key_put(up->session_keyring);
 316         kmem_cache_free(uid_cachep, up);
 317
 318 done:
 319         uids_mutex_unlock();
 320 }
 321
 322 /* IRQs are disabled and uidhash_lock is held upon function entry.
 323  * IRQ state (as stored in flags) is restored and uidhash_lock released
 324  * upon function exit.
 325  */
 326 static inline void free_user(struct user_struct *up, unsigned long flags)
 327 {
 328         /* restore back the count */
 329         atomic_inc(&up->__count);
 330         spin_unlock_irqrestore(&uidhash_lock, flags);
 331
 332         INIT_WORK(&up->work, remove_user_sysfs_dir);
 333         schedule_work(&up->work);
 334 }
 335
 336 #else   /* CONFIG_USER_SCHED && CONFIG_SYSFS */
 337
 338 int uids_sysfs_init(void) { return 0; }
 339 static inline int uids_user_create(struct user_struct *up) { return 0; }
 340 static inline void uids_mutex_lock(void) { }
 341 static inline void uids_mutex_unlock(void) { }
 342
 343 /* IRQs are disabled and uidhash_lock is held upon function entry.
 344  * IRQ state (as stored in flags) is restored and uidhash_lock released
 345  * upon function exit.
 346  */
 347 static inline void free_user(struct user_struct *up, unsigned long flags)
 348 {
 349         uid_hash_remove(up);
 350         spin_unlock_irqrestore(&uidhash_lock, flags);
 351         sched_destroy_user(up);
 352         key_put(up->uid_keyring);
 353         key_put(up->session_keyring);
 354         kmem_cache_free(uid_cachep, up);
 355 }
 356
 357 #endif
 358
 359 /*
 360  * Locate the user_struct for the passed UID.  If found, take a ref on it.  The
 361  * caller must undo that ref with free_uid().
 362  *
 363  * If the user_struct could not be found, return NULL.
 364  */
 365 struct user_struct *find_user(uid_t uid)
 366 {
 367         struct user_struct *ret;
 368         unsigned long flags;
 369         struct user_namespace *ns = current->nsproxy->user_ns;
 370
 371         spin_lock_irqsave(&uidhash_lock, flags);
 372         ret = uid_hash_find(uid, uidhashentry(ns, uid));
 373         spin_unlock_irqrestore(&uidhash_lock, flags);
 374         return ret;
 375 }
 376
 377 void free_uid(struct user_struct *up)
 378 {
 379         unsigned long flags;
 380
 381         if (!up)
 382                 return;
 383
 384         local_irq_save(flags);
 385         if (atomic_dec_and_lock(&up->__count, &uidhash_lock))
 386                 free_user(up, flags);
 387         else
 388                 local_irq_restore(flags);
 389 }
 390
 391 struct user_struct * alloc_uid(struct user_namespace *ns, uid_t uid)
 392 {
 393         struct hlist_head *hashent = uidhashentry(ns, uid);
 394         struct user_struct *up, *new;
 395
 396         /* Make uid_hash_find() + uids_user_create() + uid_hash_insert()
 397          * atomic.
 398          */
 399         uids_mutex_lock();
 400
 401         spin_lock_irq(&uidhash_lock);
 402         up = uid_hash_find(uid, hashent);
 403         spin_unlock_irq(&uidhash_lock);
 404
 405         if (!up) {
 406                 new = kmem_cache_alloc(uid_cachep, GFP_KERNEL);
 407                 if (!new)
 408                         goto out_unlock;
 409
 410                 new->uid = uid;
 411                 atomic_set(&new->__count, 1);
 412                 atomic_set(&new->processes, 0);
 413                 atomic_set(&new->files, 0);
 414                 atomic_set(&new->sigpending, 0);
 415 #ifdef CONFIG_INOTIFY_USER
 416                 atomic_set(&new->inotify_watches, 0);
 417                 atomic_set(&new->inotify_devs, 0);
 418 #endif
 419 #ifdef CONFIG_POSIX_MQUEUE
 420                 new->mq_bytes = 0;
 421 #endif
 422                 new->locked_shm = 0;
 423
 424                 if (alloc_uid_keyring(new, current) < 0)
 425                         goto out_free_user;
 426
 427                 if (sched_create_user(new) < 0)
 428                         goto out_put_keys;
 429
 430                 if (uids_user_create(new))
 431                         goto out_destoy_sched;
 432
 433                 /*
 434                  * Before adding this, check whether we raced
 435                  * on adding the same user already..
 436                  */
 437                 spin_lock_irq(&uidhash_lock);
 438                 up = uid_hash_find(uid, hashent);
 439                 if (up) {
 440                         /* This case is not possible when CONFIG_USER_SCHED
 441                          * is defined, since we serialize alloc_uid() using
 442                          * uids_mutex. Hence no need to call
 443                          * sched_destroy_user() or remove_user_sysfs_dir().
 444                          */
 445                         key_put(new->uid_keyring);
 446                         key_put(new->session_keyring);
 447                         kmem_cache_free(uid_cachep, new);
 448                 } else {
 449                         uid_hash_insert(new, hashent);
 450                         up = new;
 451                 }
 452                 spin_unlock_irq(&uidhash_lock);
 453
 454         }
 455
 456         uids_mutex_unlock();
 457
 458         return up;
 459
 460 out_destoy_sched:
 461         sched_destroy_user(new);
 462 out_put_keys:
 463         key_put(new->uid_keyring);
 464         key_put(new->session_keyring);
 465 out_free_user:
 466         kmem_cache_free(uid_cachep, new);
 467 out_unlock:
 468         uids_mutex_unlock();
 469         return NULL;
 470 }
 471
 472 void switch_uid(struct user_struct *new_user)
 473 {
 474         struct user_struct *old_user;
 475
 476         /* What if a process setreuid()'s and this brings the
 477          * new uid over his NPROC rlimit?  We can check this now
 478          * cheaply with the new uid cache, so if it matters
 479          * we should be checking for it.  -DaveM
 480          */
 481         old_user = current->user;
 482         atomic_inc(&new_user->processes);
 483         atomic_dec(&old_user->processes);
 484         switch_uid_keyring(new_user);
 485         current->user = new_user;
 486         sched_switch_user(current);
 487
 488         /*
 489          * We need to synchronize with __sigqueue_alloc()
 490          * doing a get_uid(p->user).. If that saw the old
 491          * user value, we need to wait until it has exited
 492          * its critical region before we can free the old
 493          * structure.
 494          */
 495         smp_mb();
 496         spin_unlock_wait(&current->sighand->siglock);
 497
 498         free_uid(old_user);
 499         suid_keys(current);
 500 }
 501
 502 #ifdef CONFIG_USER_NS
 503 void release_uids(struct user_namespace *ns)
 504 {
 505         int i;
 506         unsigned long flags;
 507         struct hlist_head *head;
 508         struct hlist_node *nd;
 509
 510         spin_lock_irqsave(&uidhash_lock, flags);
 511         /*
 512          * collapse the chains so that the user_struct-s will
 513          * be still alive, but not in hashes. subsequent free_uid()
 514          * will free them.
 515          */
 516         for (i = 0; i < UIDHASH_SZ; i++) {
 517                 head = ns->uidhash_table + i;
 518                 while (!hlist_empty(head)) {
 519                         nd = head->first;
 520                         hlist_del_init(nd);
 521                 }
 522         }
 523         spin_unlock_irqrestore(&uidhash_lock, flags);
 524
 525         free_uid(ns->root_user);
 526 }
 527 #endif
 528
 529 static int __init uid_cache_init(void)
 530 {
 531         int n;
 532
 533         uid_cachep = kmem_cache_create("uid_cache", sizeof(struct user_struct),
 534                         0, SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
 535
 536         for(n = 0; n < UIDHASH_SZ; ++n)
 537                 INIT_HLIST_HEAD(init_user_ns.uidhash_table + n);
 538
 539         /* Insert the root user immediately (init already runs as root) */
 540         spin_lock_irq(&uidhash_lock);
 541         uid_hash_insert(&root_user, uidhashentry(&init_user_ns, 0));
 542         spin_unlock_irq(&uidhash_lock);
 543
 544         return 0;
 545 }
 546
 547 module_init(uid_cache_init);